From b9192d31aabb73ffd7d70eb6dd383aa9f924dc0a Mon Sep 17 00:00:00 2001 From: Martin Matuska Date: Fri, 1 Mar 2013 20:49:56 +0000 Subject: [PATCH 1/4] Update vendor-sys/illumos/dist to illumos-gate 13971:d1648d817bd6 Illumos ZFS issues: 3581 spa_zio_taskq[ZIO_TYPE_FREE][ZIO_TASKQ_ISSUE]->tq_lock is piping hot --- uts/common/fs/zfs/spa.c | 208 +++++++++++++++++++++---------- uts/common/fs/zfs/sys/spa_impl.h | 16 ++- uts/common/fs/zfs/vdev_file.c | 2 +- uts/common/fs/zfs/zio.c | 31 +++-- 4 files changed, 173 insertions(+), 84 deletions(-) diff --git a/uts/common/fs/zfs/spa.c b/uts/common/fs/zfs/spa.c index 151b78091a8e..fdc28700c820 100644 --- a/uts/common/fs/zfs/spa.c +++ b/uts/common/fs/zfs/spa.c @@ -77,23 +77,25 @@ #include "zfs_comutil.h" typedef enum zti_modes { - zti_mode_fixed, /* value is # of threads (min 1) */ - zti_mode_online_percent, /* value is % of online CPUs */ - zti_mode_batch, /* cpu-intensive; value is ignored */ - zti_mode_null, /* don't create a taskq */ - zti_nmodes + ZTI_MODE_FIXED, /* value is # of threads (min 1) */ + ZTI_MODE_ONLINE_PERCENT, /* value is % of online CPUs */ + ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ + ZTI_MODE_NULL, /* don't create a taskq */ + ZTI_NMODES } zti_modes_t; -#define ZTI_FIX(n) { zti_mode_fixed, (n) } -#define ZTI_PCT(n) { zti_mode_online_percent, (n) } -#define ZTI_BATCH { zti_mode_batch, 0 } -#define ZTI_NULL { zti_mode_null, 0 } +#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } +#define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 } +#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } +#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } -#define ZTI_ONE ZTI_FIX(1) +#define ZTI_N(n) ZTI_P(n, 1) +#define ZTI_ONE ZTI_N(1) typedef struct zio_taskq_info { - enum zti_modes zti_mode; + zti_modes_t zti_mode; uint_t zti_value; + uint_t zti_count; } zio_taskq_info_t; static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { @@ -101,17 +103,30 @@ static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { }; /* - * Define the taskq threads for the following I/O types: - * NULL, READ, WRITE, FREE, CLAIM, and IOCTL + * This table defines the taskq settings for each ZFS I/O type. When + * initializing a pool, we use this table to create an appropriately sized + * taskq. Some operations are low volume and therefore have a small, static + * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE + * macros. Other operations process a large amount of data; the ZTI_BATCH + * macro causes us to create a taskq oriented for throughput. Some operations + * are so high frequency and short-lived that the taskq itself can become a a + * point of lock contention. The ZTI_P(#, #) macro indicates that we need an + * additional degree of parallelism specified by the number of threads per- + * taskq and the number of taskqs; when dispatching an event in this case, the + * particular taskq is chosen at random. + * + * The different taskq priorities are to handle the different contexts (issue + * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that + * need to be handled with minimum delay. */ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ - { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, - { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, - { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, - { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, - { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, - { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, + { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ + { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */ + { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ + { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ + { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ + { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ }; static dsl_syncfunc_t spa_sync_version; @@ -794,48 +809,120 @@ spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) offsetof(spa_error_entry_t, se_avl)); } -static taskq_t * -spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode, - uint_t value) +static void +spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) { + const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; + enum zti_modes mode = ztip->zti_mode; + uint_t value = ztip->zti_value; + uint_t count = ztip->zti_count; + spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; + char name[32]; uint_t flags = 0; boolean_t batch = B_FALSE; - switch (mode) { - case zti_mode_null: - return (NULL); /* no taskq needed */ - - case zti_mode_fixed: - ASSERT3U(value, >=, 1); - value = MAX(value, 1); - break; - - case zti_mode_batch: - batch = B_TRUE; - flags |= TASKQ_THREADS_CPU_PCT; - value = zio_taskq_batch_pct; - break; - - case zti_mode_online_percent: - flags |= TASKQ_THREADS_CPU_PCT; - break; - - default: - panic("unrecognized mode for %s taskq (%u:%u) in " - "spa_activate()", - name, mode, value); - break; + if (mode == ZTI_MODE_NULL) { + tqs->stqs_count = 0; + tqs->stqs_taskq = NULL; + return; } - if (zio_taskq_sysdc && spa->spa_proc != &p0) { - if (batch) - flags |= TASKQ_DC_BATCH; + ASSERT3U(count, >, 0); - return (taskq_create_sysdc(name, value, 50, INT_MAX, - spa->spa_proc, zio_taskq_basedc, flags)); + tqs->stqs_count = count; + tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); + + for (uint_t i = 0; i < count; i++) { + taskq_t *tq; + + switch (mode) { + case ZTI_MODE_FIXED: + ASSERT3U(value, >=, 1); + value = MAX(value, 1); + break; + + case ZTI_MODE_BATCH: + batch = B_TRUE; + flags |= TASKQ_THREADS_CPU_PCT; + value = zio_taskq_batch_pct; + break; + + case ZTI_MODE_ONLINE_PERCENT: + flags |= TASKQ_THREADS_CPU_PCT; + break; + + default: + panic("unrecognized mode for %s_%s taskq (%u:%u) in " + "spa_activate()", + zio_type_name[t], zio_taskq_types[q], mode, value); + break; + } + + if (count > 1) { + (void) snprintf(name, sizeof (name), "%s_%s_%u", + zio_type_name[t], zio_taskq_types[q], i); + } else { + (void) snprintf(name, sizeof (name), "%s_%s", + zio_type_name[t], zio_taskq_types[q]); + } + + if (zio_taskq_sysdc && spa->spa_proc != &p0) { + if (batch) + flags |= TASKQ_DC_BATCH; + + tq = taskq_create_sysdc(name, value, 50, INT_MAX, + spa->spa_proc, zio_taskq_basedc, flags); + } else { + tq = taskq_create_proc(name, value, maxclsyspri, 50, + INT_MAX, spa->spa_proc, flags); + } + + tqs->stqs_taskq[i] = tq; } - return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX, - spa->spa_proc, flags)); +} + +static void +spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) +{ + spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; + + if (tqs->stqs_taskq == NULL) { + ASSERT0(tqs->stqs_count); + return; + } + + for (uint_t i = 0; i < tqs->stqs_count; i++) { + ASSERT3P(tqs->stqs_taskq[i], !=, NULL); + taskq_destroy(tqs->stqs_taskq[i]); + } + + kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); + tqs->stqs_taskq = NULL; +} + +/* + * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. + * Note that a type may have multiple discrete taskqs to avoid lock contention + * on the taskq itself. In that case we choose which taskq at random by using + * the low bits of gethrtime(). + */ +void +spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, + task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) +{ + spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; + taskq_t *tq; + + ASSERT3P(tqs->stqs_taskq, !=, NULL); + ASSERT3U(tqs->stqs_count, !=, 0); + + if (tqs->stqs_count == 1) { + tq = tqs->stqs_taskq[0]; + } else { + tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; + } + + taskq_dispatch_ent(tq, func, arg, flags, ent); } static void @@ -843,16 +930,7 @@ spa_create_zio_taskqs(spa_t *spa) { for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { - const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; - enum zti_modes mode = ztip->zti_mode; - uint_t value = ztip->zti_value; - char name[32]; - - (void) snprintf(name, sizeof (name), - "%s_%s", zio_type_name[t], zio_taskq_types[q]); - - spa->spa_zio_taskq[t][q] = - spa_taskq_create(spa, name, mode, value); + spa_taskqs_init(spa, t, q); } } } @@ -1009,9 +1087,7 @@ spa_deactivate(spa_t *spa) for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { - if (spa->spa_zio_taskq[t][q] != NULL) - taskq_destroy(spa->spa_zio_taskq[t][q]); - spa->spa_zio_taskq[t][q] = NULL; + spa_taskqs_fini(spa, t, q); } } diff --git a/uts/common/fs/zfs/sys/spa_impl.h b/uts/common/fs/zfs/sys/spa_impl.h index ffd676e2ed77..983103e3862f 100644 --- a/uts/common/fs/zfs/sys/spa_impl.h +++ b/uts/common/fs/zfs/sys/spa_impl.h @@ -80,16 +80,16 @@ typedef struct spa_config_dirent { char *scd_path; } spa_config_dirent_t; -enum zio_taskq_type { +typedef enum zio_taskq_type { ZIO_TASKQ_ISSUE = 0, ZIO_TASKQ_ISSUE_HIGH, ZIO_TASKQ_INTERRUPT, ZIO_TASKQ_INTERRUPT_HIGH, ZIO_TASKQ_TYPES -}; +} zio_taskq_type_t; /* - * State machine for the zpool-pooname process. The states transitions + * State machine for the zpool-poolname process. The states transitions * are done as follows: * * From To Routine @@ -107,6 +107,11 @@ typedef enum spa_proc_state { SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ } spa_proc_state_t; +typedef struct spa_taskqs { + uint_t stqs_count; + taskq_t **stqs_taskq; +} spa_taskqs_t; + struct spa { /* * Fields protected by spa_namespace_lock. @@ -125,7 +130,7 @@ struct spa { uint8_t spa_sync_on; /* sync threads are running */ spa_load_state_t spa_load_state; /* current load operation */ uint64_t spa_import_flags; /* import specific flags */ - taskq_t *spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; + spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; dsl_pool_t *spa_dsl_pool; boolean_t spa_is_initializing; /* true while opening pool */ metaslab_class_t *spa_normal_class; /* normal data class */ @@ -245,6 +250,9 @@ struct spa { extern const char *spa_config_path; +extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, + task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); + #ifdef __cplusplus } #endif diff --git a/uts/common/fs/zfs/vdev_file.c b/uts/common/fs/zfs/vdev_file.c index 1fbce5e54227..f7cfa9b095a3 100644 --- a/uts/common/fs/zfs/vdev_file.c +++ b/uts/common/fs/zfs/vdev_file.c @@ -224,7 +224,7 @@ vdev_file_io_start(zio_t *zio) bp->b_private = vf->vf_vnode; bp->b_iodone = (int (*)())vdev_file_io_intr; - taskq_dispatch_ent(spa->spa_zio_taskq[ZIO_TYPE_FREE][ZIO_TASKQ_ISSUE], + spa_taskq_dispatch_ent(spa, ZIO_TYPE_FREE, ZIO_TASKQ_ISSUE, vdev_file_io_strategy, bp, 0, &zio->io_tqent); return (ZIO_PIPELINE_STOP); diff --git a/uts/common/fs/zfs/zio.c b/uts/common/fs/zfs/zio.c index 04b8ddfaf71b..432a992b26f4 100644 --- a/uts/common/fs/zfs/zio.c +++ b/uts/common/fs/zfs/zio.c @@ -1107,7 +1107,7 @@ zio_free_bp_init(zio_t *zio) */ static void -zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) +zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) { spa_t *spa = zio->io_spa; zio_type_t t = zio->io_type; @@ -1128,10 +1128,11 @@ zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) t = ZIO_TYPE_NULL; /* - * If this is a high priority I/O, then use the high priority taskq. + * If this is a high priority I/O, then use the high priority taskq if + * available. */ if (zio->io_priority == ZIO_PRIORITY_NOW && - spa->spa_zio_taskq[t][q + 1] != NULL) + spa->spa_zio_taskq[t][q + 1].stqs_count != 0) q++; ASSERT3U(q, <, ZIO_TASKQ_TYPES); @@ -1142,19 +1143,24 @@ zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) * to dispatch the zio to another taskq at the same time. */ ASSERT(zio->io_tqent.tqent_next == NULL); - taskq_dispatch_ent(spa->spa_zio_taskq[t][q], - (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); + spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, + flags, &zio->io_tqent); } static boolean_t -zio_taskq_member(zio_t *zio, enum zio_taskq_type q) +zio_taskq_member(zio_t *zio, zio_taskq_type_t q) { kthread_t *executor = zio->io_executor; spa_t *spa = zio->io_spa; - for (zio_type_t t = 0; t < ZIO_TYPES; t++) - if (taskq_member(spa->spa_zio_taskq[t][q], executor)) - return (B_TRUE); + for (zio_type_t t = 0; t < ZIO_TYPES; t++) { + spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; + uint_t i; + for (i = 0; i < tqs->stqs_count; i++) { + if (taskq_member(tqs->stqs_taskq[i], executor)) + return (B_TRUE); + } + } return (B_FALSE); } @@ -3017,10 +3023,9 @@ zio_done(zio_t *zio) * Hand it off to the otherwise-unused claim taskq. */ ASSERT(zio->io_tqent.tqent_next == NULL); - taskq_dispatch_ent( - spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE], - (task_func_t *)zio_reexecute, zio, 0, - &zio->io_tqent); + spa_taskq_dispatch_ent(spa, ZIO_TYPE_CLAIM, + ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio, + 0, &zio->io_tqent); } return (ZIO_PIPELINE_STOP); } From 577227be9d7e7e691365c6df7c2c70a0cd5c0d96 Mon Sep 17 00:00:00 2001 From: Martin Matuska Date: Fri, 1 Mar 2013 21:01:45 +0000 Subject: [PATCH 2/4] Update vendor/illumos/dist and vendor-sys/illumos/dist to illumos-gate 13973:4972ab336f54 Illumos ZFS issues: 3464 zfs synctask code needs restructuring --- cmd/zdb/zdb.c | 9 +- cmd/zfs/zfs_main.c | 116 +- cmd/zhack/zhack.c | 27 +- cmd/ztest/ztest.c | 190 +- common/nvpair/fnvpair.c | 13 + lib/libzfs/common/libzfs.h | 6 +- lib/libzfs/common/libzfs_dataset.c | 318 ++- lib/libzfs/common/libzfs_sendrecv.c | 12 +- lib/libzfs_core/common/libzfs_core.c | 102 +- lib/libzfs_core/common/libzfs_core.h | 4 + lib/libzpool/common/kernel.c | 5 + lib/libzpool/common/sys/zfs_context.h | 15 +- man/man1m/zfs.1m | 6 +- uts/common/Makefile.files | 4 +- uts/common/fs/zfs/arc.c | 10 +- uts/common/fs/zfs/bplist.c | 8 + uts/common/fs/zfs/bpobj.c | 4 + uts/common/fs/zfs/dbuf.c | 87 +- uts/common/fs/zfs/dmu.c | 2 +- uts/common/fs/zfs/dmu_diff.c | 82 +- uts/common/fs/zfs/dmu_objset.c | 889 +++--- uts/common/fs/zfs/dmu_send.c | 917 +++--- uts/common/fs/zfs/dmu_traverse.c | 42 +- uts/common/fs/zfs/dmu_tx.c | 31 +- uts/common/fs/zfs/dnode.c | 6 +- uts/common/fs/zfs/dnode_sync.c | 1 + uts/common/fs/zfs/dsl_dataset.c | 3718 ++++++++----------------- uts/common/fs/zfs/dsl_deleg.c | 150 +- uts/common/fs/zfs/dsl_destroy.c | 926 ++++++ uts/common/fs/zfs/dsl_dir.c | 587 ++-- uts/common/fs/zfs/dsl_pool.c | 216 +- uts/common/fs/zfs/dsl_prop.c | 483 ++-- uts/common/fs/zfs/dsl_scan.c | 116 +- uts/common/fs/zfs/dsl_synctask.c | 239 +- uts/common/fs/zfs/dsl_userhold.c | 536 ++++ uts/common/fs/zfs/metaslab.c | 38 + uts/common/fs/zfs/refcount.c | 21 +- uts/common/fs/zfs/rrwlock.c | 41 +- uts/common/fs/zfs/sa.c | 6 +- uts/common/fs/zfs/spa.c | 81 +- uts/common/fs/zfs/spa_history.c | 26 +- uts/common/fs/zfs/spa_misc.c | 23 +- uts/common/fs/zfs/space_map.c | 43 +- uts/common/fs/zfs/sys/arc.h | 2 +- uts/common/fs/zfs/sys/dbuf.h | 9 +- uts/common/fs/zfs/sys/dmu.h | 52 +- uts/common/fs/zfs/sys/dmu_objset.h | 11 +- uts/common/fs/zfs/sys/dmu_send.h | 66 + uts/common/fs/zfs/sys/dmu_tx.h | 6 +- uts/common/fs/zfs/sys/dsl_dataset.h | 112 +- uts/common/fs/zfs/sys/dsl_destroy.h | 52 + uts/common/fs/zfs/sys/dsl_dir.h | 17 +- uts/common/fs/zfs/sys/dsl_pool.h | 16 +- uts/common/fs/zfs/sys/dsl_prop.h | 53 +- uts/common/fs/zfs/sys/dsl_synctask.h | 46 +- uts/common/fs/zfs/sys/dsl_userhold.h | 57 + uts/common/fs/zfs/sys/metaslab.h | 3 +- uts/common/fs/zfs/sys/refcount.h | 5 +- uts/common/fs/zfs/sys/rrwlock.h | 7 +- uts/common/fs/zfs/sys/space_map.h | 2 + uts/common/fs/zfs/sys/txg.h | 9 +- uts/common/fs/zfs/sys/zfeature.h | 17 +- uts/common/fs/zfs/sys/zfs_debug.h | 12 +- uts/common/fs/zfs/sys/zfs_ioctl.h | 4 +- uts/common/fs/zfs/sys/zfs_znode.h | 3 +- uts/common/fs/zfs/sys/zil.h | 4 +- uts/common/fs/zfs/txg.c | 37 +- uts/common/fs/zfs/zfs_ctldir.c | 33 +- uts/common/fs/zfs/zfs_ioctl.c | 1139 ++++---- uts/common/fs/zfs/zfs_vfsops.c | 74 +- uts/common/fs/zfs/zil.c | 137 +- uts/common/fs/zfs/zio.c | 8 +- uts/common/fs/zfs/zvol.c | 81 +- uts/common/sys/nvpair.h | 1 + 74 files changed, 6341 insertions(+), 5890 deletions(-) create mode 100644 uts/common/fs/zfs/dsl_destroy.c create mode 100644 uts/common/fs/zfs/dsl_userhold.c create mode 100644 uts/common/fs/zfs/sys/dmu_send.h create mode 100644 uts/common/fs/zfs/sys/dsl_destroy.h create mode 100644 uts/common/fs/zfs/sys/dsl_userhold.h diff --git a/cmd/zdb/zdb.c b/cmd/zdb/zdb.c index 2234a71528bf..46a630fb4230 100644 --- a/cmd/zdb/zdb.c +++ b/cmd/zdb/zdb.c @@ -1658,7 +1658,9 @@ dump_dir(objset_t *os) int print_header = 1; int i, error; + dsl_pool_config_enter(dmu_objset_pool(os), FTAG); dmu_objset_fast_stat(os, &dds); + dsl_pool_config_exit(dmu_objset_pool(os), FTAG); if (dds.dds_type < DMU_OST_NUMTYPES) type = objset_types[dds.dds_type]; @@ -2109,7 +2111,6 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, zio_nowait(zio_read(NULL, spa, bp, data, size, zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb)); - } zcb->zcb_readfails = 0; @@ -2297,8 +2298,10 @@ dump_block_stats(spa_t *spa) */ (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj, count_block_cb, &zcb, NULL); - (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj, - count_block_cb, &zcb, NULL); + if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { + (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj, + count_block_cb, &zcb, NULL); + } if (spa_feature_is_active(spa, &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset, diff --git a/cmd/zfs/zfs_main.c b/cmd/zfs/zfs_main.c index e5e35e6f166c..5e95aed88b4f 100644 --- a/cmd/zfs/zfs_main.c +++ b/cmd/zfs/zfs_main.c @@ -898,6 +898,7 @@ typedef struct destroy_cbdata { boolean_t cb_parsable; boolean_t cb_dryrun; nvlist_t *cb_nvl; + nvlist_t *cb_batchedsnaps; /* first snap in contiguous run */ char *cb_firstsnap; @@ -994,9 +995,27 @@ destroy_callback(zfs_handle_t *zhp, void *data) zfs_close(zhp); return (0); } + if (cb->cb_dryrun) { + zfs_close(zhp); + return (0); + } - if (!cb->cb_dryrun) { - if (zfs_unmount(zhp, NULL, cb->cb_force ? MS_FORCE : 0) != 0 || + /* + * We batch up all contiguous snapshots (even of different + * filesystems) and destroy them with one ioctl. We can't + * simply do all snap deletions and then all fs deletions, + * because we must delete a clone before its origin. + */ + if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT) { + fnvlist_add_boolean(cb->cb_batchedsnaps, name); + } else { + int error = zfs_destroy_snaps_nvl(g_zfs, + cb->cb_batchedsnaps, B_FALSE); + fnvlist_free(cb->cb_batchedsnaps); + cb->cb_batchedsnaps = fnvlist_alloc(); + + if (error != 0 || + zfs_unmount(zhp, NULL, cb->cb_force ? MS_FORCE : 0) != 0 || zfs_destroy(zhp, cb->cb_defer_destroy) != 0) { zfs_close(zhp); return (-1); @@ -1152,8 +1171,10 @@ static int zfs_do_destroy(int argc, char **argv) { destroy_cbdata_t cb = { 0 }; + int rv = 0; + int err = 0; int c; - zfs_handle_t *zhp; + zfs_handle_t *zhp = NULL; char *at; zfs_type_t type = ZFS_TYPE_DATASET; @@ -1207,11 +1228,9 @@ zfs_do_destroy(int argc, char **argv) at = strchr(argv[0], '@'); if (at != NULL) { - int err = 0; /* Build the list of snaps to destroy in cb_nvl. */ - if (nvlist_alloc(&cb.cb_nvl, NV_UNIQUE_NAME, 0) != 0) - nomem(); + cb.cb_nvl = fnvlist_alloc(); *at = '\0'; zhp = zfs_open(g_zfs, argv[0], @@ -1222,17 +1241,15 @@ zfs_do_destroy(int argc, char **argv) cb.cb_snapspec = at + 1; if (gather_snapshots(zfs_handle_dup(zhp), &cb) != 0 || cb.cb_error) { - zfs_close(zhp); - nvlist_free(cb.cb_nvl); - return (1); + rv = 1; + goto out; } if (nvlist_empty(cb.cb_nvl)) { (void) fprintf(stderr, gettext("could not find any " "snapshots to destroy; check snapshot names.\n")); - zfs_close(zhp); - nvlist_free(cb.cb_nvl); - return (1); + rv = 1; + goto out; } if (cb.cb_verbose) { @@ -1251,18 +1268,26 @@ zfs_do_destroy(int argc, char **argv) } if (!cb.cb_dryrun) { - if (cb.cb_doclones) + if (cb.cb_doclones) { + cb.cb_batchedsnaps = fnvlist_alloc(); err = destroy_clones(&cb); + if (err == 0) { + err = zfs_destroy_snaps_nvl(g_zfs, + cb.cb_batchedsnaps, B_FALSE); + } + if (err != 0) { + rv = 1; + goto out; + } + } if (err == 0) { - err = zfs_destroy_snaps_nvl(zhp, cb.cb_nvl, + err = zfs_destroy_snaps_nvl(g_zfs, cb.cb_nvl, cb.cb_defer_destroy); } } - zfs_close(zhp); - nvlist_free(cb.cb_nvl); if (err != 0) - return (1); + rv = 1; } else { /* Open the given dataset */ if ((zhp = zfs_open(g_zfs, argv[0], type)) == NULL) @@ -1283,8 +1308,8 @@ zfs_do_destroy(int argc, char **argv) zfs_get_name(zhp)); (void) fprintf(stderr, gettext("use 'zpool destroy %s' " "to destroy the pool itself\n"), zfs_get_name(zhp)); - zfs_close(zhp); - return (1); + rv = 1; + goto out; } /* @@ -1294,30 +1319,42 @@ zfs_do_destroy(int argc, char **argv) if (!cb.cb_doclones && zfs_iter_dependents(zhp, B_TRUE, destroy_check_dependent, &cb) != 0) { - zfs_close(zhp); - return (1); + rv = 1; + goto out; } if (cb.cb_error) { - zfs_close(zhp); - return (1); + rv = 1; + goto out; } + cb.cb_batchedsnaps = fnvlist_alloc(); if (zfs_iter_dependents(zhp, B_FALSE, destroy_callback, &cb) != 0) { - zfs_close(zhp); - return (1); + rv = 1; + goto out; } /* * Do the real thing. The callback will close the * handle regardless of whether it succeeds or not. */ - if (destroy_callback(zhp, &cb) != 0) - return (1); + err = destroy_callback(zhp, &cb); + zhp = NULL; + if (err == 0) { + err = zfs_destroy_snaps_nvl(g_zfs, + cb.cb_batchedsnaps, cb.cb_defer_destroy); + } + if (err != 0) + rv = 1; } - return (0); +out: + fnvlist_free(cb.cb_batchedsnaps); + fnvlist_free(cb.cb_nvl); + if (zhp != NULL) + zfs_close(zhp); + return (rv); } static boolean_t @@ -5052,28 +5089,12 @@ zfs_do_allow_unallow_impl(int argc, char **argv, boolean_t un) return (error); } -/* - * zfs allow [-r] [-t] ... - * - * -r Recursively hold - * -t Temporary hold (hidden option) - * - * Apply a user-hold with the given tag to the list of snapshots. - */ static int zfs_do_allow(int argc, char **argv) { return (zfs_do_allow_unallow_impl(argc, argv, B_FALSE)); } -/* - * zfs unallow [-r] [-t] ... - * - * -r Recursively hold - * -t Temporary hold (hidden option) - * - * Apply a user-hold with the given tag to the list of snapshots. - */ static int zfs_do_unallow(int argc, char **argv) { @@ -5087,7 +5108,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding) int i; const char *tag; boolean_t recursive = B_FALSE; - boolean_t temphold = B_FALSE; const char *opts = holding ? "rt" : "r"; int c; @@ -5097,9 +5117,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding) case 'r': recursive = B_TRUE; break; - case 't': - temphold = B_TRUE; - break; case '?': (void) fprintf(stderr, gettext("invalid option '%c'\n"), optopt); @@ -5148,7 +5165,7 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding) } if (holding) { if (zfs_hold(zhp, delim+1, tag, recursive, - temphold, B_FALSE, -1, 0, 0) != 0) + B_FALSE, -1) != 0) ++errors; } else { if (zfs_release(zhp, delim+1, tag, recursive) != 0) @@ -5164,7 +5181,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding) * zfs hold [-r] [-t] ... * * -r Recursively hold - * -t Temporary hold (hidden option) * * Apply a user-hold with the given tag to the list of snapshots. */ diff --git a/cmd/zhack/zhack.c b/cmd/zhack/zhack.c index 7cc83d2dc0b3..d9ad36cccb3b 100644 --- a/cmd/zhack/zhack.c +++ b/cmd/zhack/zhack.c @@ -46,6 +46,7 @@ #include #include #include +#include #undef ZFS_MAXNAMELEN #undef verify #include @@ -273,10 +274,10 @@ zhack_do_feature_stat(int argc, char **argv) } static void -feature_enable_sync(void *arg1, void *arg2, dmu_tx_t *tx) +feature_enable_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - zfeature_info_t *feature = arg2; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + zfeature_info_t *feature = arg; spa_feature_enable(spa, feature, tx); spa_history_log_internal(spa, "zhack enable feature", tx, @@ -344,8 +345,8 @@ zhack_do_feature_enable(int argc, char **argv) if (0 == zap_contains(mos, spa->spa_feat_desc_obj, feature.fi_guid)) fatal("feature already enabled: %s", feature.fi_guid); - VERIFY3U(0, ==, dsl_sync_task_do(spa->spa_dsl_pool, NULL, - feature_enable_sync, spa, &feature, 5)); + VERIFY0(dsl_sync_task(spa_name(spa), NULL, + feature_enable_sync, &feature, 5)); spa_close(spa, FTAG); @@ -353,10 +354,10 @@ zhack_do_feature_enable(int argc, char **argv) } static void -feature_incr_sync(void *arg1, void *arg2, dmu_tx_t *tx) +feature_incr_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - zfeature_info_t *feature = arg2; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + zfeature_info_t *feature = arg; spa_feature_incr(spa, feature, tx); spa_history_log_internal(spa, "zhack feature incr", tx, @@ -364,10 +365,10 @@ feature_incr_sync(void *arg1, void *arg2, dmu_tx_t *tx) } static void -feature_decr_sync(void *arg1, void *arg2, dmu_tx_t *tx) +feature_decr_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - zfeature_info_t *feature = arg2; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + zfeature_info_t *feature = arg; spa_feature_decr(spa, feature, tx); spa_history_log_internal(spa, "zhack feature decr", tx, @@ -442,8 +443,8 @@ zhack_do_feature_ref(int argc, char **argv) if (decr && !spa_feature_is_active(spa, &feature)) fatal("feature refcount already 0: %s", feature.fi_guid); - VERIFY3U(0, ==, dsl_sync_task_do(spa->spa_dsl_pool, NULL, - decr ? feature_decr_sync : feature_incr_sync, spa, &feature, 5)); + VERIFY0(dsl_sync_task(spa_name(spa), NULL, + decr ? feature_decr_sync : feature_incr_sync, &feature, 5)); spa_close(spa, FTAG); } diff --git a/cmd/ztest/ztest.c b/cmd/ztest/ztest.c index b51ab84bfc38..ed460551c69c 100644 --- a/cmd/ztest/ztest.c +++ b/cmd/ztest/ztest.c @@ -103,10 +103,12 @@ #include #include #include +#include #include #include #include #include +#include #include #include #include @@ -365,7 +367,7 @@ ztest_info_t ztest_info[] = { { ztest_scrub, 1, &zopt_rarely }, { ztest_spa_upgrade, 1, &zopt_rarely }, { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, - { ztest_vdev_attach_detach, 1, &zopt_rarely }, + { ztest_vdev_attach_detach, 1, &zopt_sometimes }, { ztest_vdev_LUN_growth, 1, &zopt_rarely }, { ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime }, @@ -1006,9 +1008,8 @@ ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, uint64_t curval; int error; - error = dsl_prop_set(osname, propname, - (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), - sizeof (value), 1, &value); + error = dsl_prop_set_int(osname, propname, + (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); if (error == ENOSPC) { ztest_record_enospc(FTAG); @@ -1016,8 +1017,7 @@ ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, } ASSERT0(error); - VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval), - 1, &curval, setpoint), ==, 0); + VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); if (ztest_opts.zo_verbose >= 6) { VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); @@ -2479,8 +2479,7 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) int error; VERIFY(mutex_lock(&ztest_vdev_lock) == 0); - leaves = - MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; + leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); @@ -3180,7 +3179,7 @@ ztest_objset_destroy_cb(const char *name, void *arg) /* * Verify that the dataset contains a directory object. */ - VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os)); + VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); error = dmu_object_info(os, ZTEST_DIROBJ, &doi); if (error != ENOENT) { /* We could have crashed in the middle of destroying it */ @@ -3188,12 +3187,16 @@ ztest_objset_destroy_cb(const char *name, void *arg) ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); ASSERT3S(doi.doi_physical_blocks_512, >=, 0); } - dmu_objset_rele(os, FTAG); + dmu_objset_disown(os, FTAG); /* * Destroy the dataset. */ - VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE)); + if (strchr(name, '@') != NULL) { + VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); + } else { + VERIFY0(dsl_destroy_head(name)); + } return (0); } @@ -3203,16 +3206,17 @@ ztest_snapshot_create(char *osname, uint64_t id) char snapname[MAXNAMELEN]; int error; - (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, - (u_longlong_t)id); + (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); - error = dmu_objset_snapshot_one(osname, strchr(snapname, '@') + 1); + error = dmu_objset_snapshot_one(osname, snapname); if (error == ENOSPC) { ztest_record_enospc(FTAG); return (B_FALSE); } - if (error != 0 && error != EEXIST) - fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error); + if (error != 0 && error != EEXIST) { + fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, + snapname, error); + } return (B_TRUE); } @@ -3225,7 +3229,7 @@ ztest_snapshot_destroy(char *osname, uint64_t id) (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, (u_longlong_t)id); - error = dmu_objset_destroy(snapname, B_FALSE); + error = dsl_destroy_snapshot(snapname, B_FALSE); if (error != 0 && error != ENOENT) fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); return (B_TRUE); @@ -3271,7 +3275,8 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) /* * Verify that the destroyed dataset is no longer in the namespace. */ - VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os)); + VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, + FTAG, &os)); /* * Verify that we can create a new dataset. @@ -3286,8 +3291,7 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_objset_create(%s) = %d", name, error); } - VERIFY3U(0, ==, - dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); + VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); ztest_zd_init(&zdtmp, NULL, os); @@ -3363,21 +3367,21 @@ ztest_dsl_dataset_cleanup(char *osname, uint64_t id) (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); - error = dmu_objset_destroy(clone2name, B_FALSE); + error = dsl_destroy_head(clone2name); if (error && error != ENOENT) - fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error); - error = dmu_objset_destroy(snap3name, B_FALSE); + fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); + error = dsl_destroy_snapshot(snap3name, B_FALSE); if (error && error != ENOENT) - fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error); - error = dmu_objset_destroy(snap2name, B_FALSE); + fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); + error = dsl_destroy_snapshot(snap2name, B_FALSE); if (error && error != ENOENT) - fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error); - error = dmu_objset_destroy(clone1name, B_FALSE); + fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); + error = dsl_destroy_head(clone1name); if (error && error != ENOENT) - fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error); - error = dmu_objset_destroy(snap1name, B_FALSE); + fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); + error = dsl_destroy_snapshot(snap1name, B_FALSE); if (error && error != ENOENT) - fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error); + fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); } /* @@ -3386,8 +3390,7 @@ ztest_dsl_dataset_cleanup(char *osname, uint64_t id) void ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) { - objset_t *clone; - dsl_dataset_t *ds; + objset_t *os; char snap1name[MAXNAMELEN]; char clone1name[MAXNAMELEN]; char snap2name[MAXNAMELEN]; @@ -3415,12 +3418,7 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); } - error = dmu_objset_hold(snap1name, FTAG, &clone); - if (error) - fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error); - - error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0); - dmu_objset_rele(clone, FTAG); + error = dmu_objset_clone(clone1name, snap1name); if (error) { if (error == ENOSPC) { ztest_record_enospc(FTAG); @@ -3447,12 +3445,7 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); } - error = dmu_objset_hold(snap3name, FTAG, &clone); - if (error) - fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); - - error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0); - dmu_objset_rele(clone, FTAG); + error = dmu_objset_clone(clone2name, snap3name); if (error) { if (error == ENOSPC) { ztest_record_enospc(FTAG); @@ -3461,14 +3454,14 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); } - error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds); + error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); if (error) - fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error); + fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); error = dsl_dataset_promote(clone2name, NULL); if (error != EBUSY) fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, error); - dsl_dataset_disown(ds, FTAG); + dmu_objset_disown(os, FTAG); out: ztest_dsl_dataset_cleanup(osname, id); @@ -4280,7 +4273,7 @@ ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) } count = -1ULL; - VERIFY(zap_count(os, object, &count) == 0); + VERIFY0(zap_count(os, object, &count)); ASSERT(count != -1ULL); /* @@ -4591,6 +4584,22 @@ ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) (void) rw_unlock(&ztest_name_lock); } +static int +user_release_one(const char *snapname, const char *holdname) +{ + nvlist_t *snaps, *holds; + int error; + + snaps = fnvlist_alloc(); + holds = fnvlist_alloc(); + fnvlist_add_boolean(holds, holdname); + fnvlist_add_nvlist(snaps, snapname, holds); + fnvlist_free(holds); + error = dsl_dataset_user_release(snaps, NULL); + fnvlist_free(snaps); + return (error); +} + /* * Test snapshot hold/release and deferred destroy. */ @@ -4605,22 +4614,30 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) char clonename[100]; char tag[100]; char osname[MAXNAMELEN]; + nvlist_t *holds; (void) rw_rdlock(&ztest_name_lock); dmu_objset_name(os, osname); - (void) snprintf(snapname, 100, "sh1_%llu", id); - (void) snprintf(fullname, 100, "%s@%s", osname, snapname); - (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id); - (void) snprintf(tag, 100, "%tag_%llu", id); + (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); + (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); + (void) snprintf(clonename, sizeof (clonename), + "%s/ch1_%llu", osname, id); + (void) snprintf(tag, sizeof (tag), "tag_%llu", id); /* * Clean up from any previous run. */ - (void) dmu_objset_destroy(clonename, B_FALSE); - (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE); - (void) dmu_objset_destroy(fullname, B_FALSE); + error = dsl_destroy_head(clonename); + if (error != ENOENT) + ASSERT0(error); + error = user_release_one(fullname, tag); + if (error != ESRCH && error != ENOENT) + ASSERT0(error); + error = dsl_destroy_snapshot(fullname, B_FALSE); + if (error != ENOENT) + ASSERT0(error); /* * Create snapshot, clone it, mark snap for deferred destroy, @@ -4635,12 +4652,7 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); } - error = dmu_objset_hold(fullname, FTAG, &origin); - if (error) - fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); - - error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0); - dmu_objset_rele(origin, FTAG); + error = dmu_objset_clone(clonename, fullname); if (error) { if (error == ENOSPC) { ztest_record_enospc("dmu_objset_clone"); @@ -4649,15 +4661,15 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); } - error = dmu_objset_destroy(fullname, B_TRUE); + error = dsl_destroy_snapshot(fullname, B_TRUE); if (error) { - fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", + fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", fullname, error); } - error = dmu_objset_destroy(clonename, B_FALSE); + error = dsl_destroy_head(clonename); if (error) - fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error); + fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); error = dmu_objset_hold(fullname, FTAG, &origin); if (error != ENOENT) @@ -4677,28 +4689,31 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); } - error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE, - B_TRUE, -1); + holds = fnvlist_alloc(); + fnvlist_add_string(holds, fullname, tag); + error = dsl_dataset_user_hold(holds, 0, NULL); + fnvlist_free(holds); + if (error) fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag); - error = dmu_objset_destroy(fullname, B_FALSE); + error = dsl_destroy_snapshot(fullname, B_FALSE); if (error != EBUSY) { - fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d", + fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", fullname, error); } - error = dmu_objset_destroy(fullname, B_TRUE); + error = dsl_destroy_snapshot(fullname, B_TRUE); if (error) { - fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d", + fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", fullname, error); } - error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE); + error = user_release_one(fullname, tag); if (error) - fatal(0, "dsl_dataset_user_release(%s)", fullname, tag); + fatal(0, "user_release_one(%s)", fullname, tag); - VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT); + VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); out: (void) rw_unlock(&ztest_name_lock); @@ -4952,8 +4967,12 @@ ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) */ for (int i = 0; i < copies; i++) { uint64_t offset = i * blocksize; - VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &db, - DMU_READ_NO_PREFETCH)); + int error = dmu_buf_hold(os, object, offset, FTAG, &db, + DMU_READ_NO_PREFETCH); + if (error != 0) { + fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", + os, (long long)object, (long long) offset, error); + } ASSERT(db->db_offset == offset); ASSERT(db->db_size == blocksize); ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || @@ -5163,6 +5182,7 @@ ztest_spa_import_export(char *oldname, char *newname) nvlist_t *config, *newconfig; uint64_t pool_guid; spa_t *spa; + int error; if (ztest_opts.zo_verbose >= 4) { (void) printf("import/export: old = %s, new = %s\n", @@ -5207,7 +5227,12 @@ ztest_spa_import_export(char *oldname, char *newname) /* * Import it under the new name. */ - VERIFY3U(0, ==, spa_import(newname, config, NULL, 0)); + error = spa_import(newname, config, NULL, 0); + if (error != 0) { + dump_nvlist(config, 0); + fatal(B_FALSE, "couldn't import pool %s as %s: error %u", + oldname, newname, error); + } ztest_walk_pool_directory("pools after import"); @@ -5414,7 +5439,7 @@ ztest_dataset_open(int d) } ASSERT(error == 0 || error == EEXIST); - VERIFY0(dmu_objset_hold(name, zd, &os)); + VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); (void) rw_unlock(&ztest_name_lock); ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); @@ -5455,7 +5480,7 @@ ztest_dataset_close(int d) ztest_ds_t *zd = &ztest_ds[d]; zil_close(zd->zd_zilog); - dmu_objset_rele(zd->zd_os, zd); + dmu_objset_disown(zd->zd_os, zd); ztest_zd_fini(zd); } @@ -5499,13 +5524,14 @@ ztest_run(ztest_shared_t *zs) * Open our pool. */ kernel_init(FREAD | FWRITE); - VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0); + VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); spa->spa_debug = B_TRUE; ztest_spa = spa; - VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os)); + VERIFY0(dmu_objset_own(ztest_opts.zo_pool, + DMU_OST_ANY, B_TRUE, FTAG, &os)); zs->zs_guid = dmu_objset_fsid_guid(os); - dmu_objset_rele(os, FTAG); + dmu_objset_disown(os, FTAG); spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; diff --git a/common/nvpair/fnvpair.c b/common/nvpair/fnvpair.c index 8c5591c0b2f7..b3cf173f6499 100644 --- a/common/nvpair/fnvpair.c +++ b/common/nvpair/fnvpair.c @@ -26,6 +26,7 @@ #include #include #include +#include #ifndef _KERNEL #include #endif @@ -114,6 +115,18 @@ fnvlist_merge(nvlist_t *dst, nvlist_t *src) VERIFY0(nvlist_merge(dst, src, KM_SLEEP)); } +size_t +fnvlist_num_pairs(nvlist_t *nvl) +{ + size_t count = 0; + nvpair_t *pair; + + for (pair = nvlist_next_nvpair(nvl, 0); pair != NULL; + pair = nvlist_next_nvpair(nvl, pair)) + count++; + return (count); +} + void fnvlist_add_boolean(nvlist_t *nvl, const char *name) { diff --git a/lib/libzfs/common/libzfs.h b/lib/libzfs/common/libzfs.h index 49c86dd18f7c..4d1e8186d286 100644 --- a/lib/libzfs/common/libzfs.h +++ b/lib/libzfs/common/libzfs.h @@ -550,7 +550,7 @@ extern int zfs_create(libzfs_handle_t *, const char *, zfs_type_t, extern int zfs_create_ancestors(libzfs_handle_t *, const char *); extern int zfs_destroy(zfs_handle_t *, boolean_t); extern int zfs_destroy_snaps(zfs_handle_t *, char *, boolean_t); -extern int zfs_destroy_snaps_nvl(zfs_handle_t *, nvlist_t *, boolean_t); +extern int zfs_destroy_snaps_nvl(libzfs_handle_t *, nvlist_t *, boolean_t); extern int zfs_clone(zfs_handle_t *, const char *, nvlist_t *); extern int zfs_snapshot(libzfs_handle_t *, const char *, boolean_t, nvlist_t *); extern int zfs_snapshot_nvl(libzfs_handle_t *hdl, nvlist_t *snaps, @@ -593,8 +593,8 @@ extern int zfs_send(zfs_handle_t *, const char *, const char *, sendflags_t *, int, snapfilter_cb_t, void *, nvlist_t **); extern int zfs_promote(zfs_handle_t *); -extern int zfs_hold(zfs_handle_t *, const char *, const char *, boolean_t, - boolean_t, boolean_t, int, uint64_t, uint64_t); +extern int zfs_hold(zfs_handle_t *, const char *, const char *, + boolean_t, boolean_t, int); extern int zfs_release(zfs_handle_t *, const char *, const char *, boolean_t); extern int zfs_get_holds(zfs_handle_t *, nvlist_t **); extern uint64_t zvol_volsize_to_reservation(uint64_t, nvlist_t *); diff --git a/lib/libzfs/common/libzfs_dataset.c b/lib/libzfs/common/libzfs_dataset.c index 3ad1ed69d57b..6121a0f16188 100644 --- a/lib/libzfs/common/libzfs_dataset.c +++ b/lib/libzfs/common/libzfs_dataset.c @@ -1973,10 +1973,7 @@ get_clones_cb(zfs_handle_t *zhp, void *arg) NULL, NULL, 0, B_TRUE) != 0) goto out; if (strcmp(gca->buf, gca->origin) == 0) { - if (nvlist_add_boolean(gca->value, zfs_get_name(zhp)) != 0) { - zfs_close(zhp); - return (no_memory(zhp->zfs_hdl)); - } + fnvlist_add_boolean(gca->value, zfs_get_name(zhp)); gca->numclones--; } @@ -3142,45 +3139,49 @@ zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname, boolean_t defer) dgettext(TEXT_DOMAIN, "cannot destroy '%s@%s'"), zhp->zfs_name, snapname); } else { - ret = zfs_destroy_snaps_nvl(zhp, dd.nvl, defer); + ret = zfs_destroy_snaps_nvl(zhp->zfs_hdl, dd.nvl, defer); } nvlist_free(dd.nvl); return (ret); } /* - * Destroys all the snapshots named in the nvlist. They must be underneath - * the zhp (either snapshots of it, or snapshots of its descendants). + * Destroys all the snapshots named in the nvlist. */ int -zfs_destroy_snaps_nvl(zfs_handle_t *zhp, nvlist_t *snaps, boolean_t defer) +zfs_destroy_snaps_nvl(libzfs_handle_t *hdl, nvlist_t *snaps, boolean_t defer) { int ret; nvlist_t *errlist; ret = lzc_destroy_snaps(snaps, defer, &errlist); - if (ret != 0) { - for (nvpair_t *pair = nvlist_next_nvpair(errlist, NULL); - pair != NULL; pair = nvlist_next_nvpair(errlist, pair)) { - char errbuf[1024]; - (void) snprintf(errbuf, sizeof (errbuf), - dgettext(TEXT_DOMAIN, "cannot destroy snapshot %s"), - nvpair_name(pair)); + if (ret == 0) + return (0); - switch (fnvpair_value_int32(pair)) { - case EEXIST: - zfs_error_aux(zhp->zfs_hdl, - dgettext(TEXT_DOMAIN, - "snapshot is cloned")); - ret = zfs_error(zhp->zfs_hdl, EZFS_EXISTS, - errbuf); - break; - default: - ret = zfs_standard_error(zhp->zfs_hdl, errno, - errbuf); - break; - } + if (nvlist_next_nvpair(errlist, NULL) == NULL) { + char errbuf[1024]; + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, "cannot destroy snapshots")); + + ret = zfs_standard_error(hdl, ret, errbuf); + } + for (nvpair_t *pair = nvlist_next_nvpair(errlist, NULL); + pair != NULL; pair = nvlist_next_nvpair(errlist, pair)) { + char errbuf[1024]; + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, "cannot destroy snapshot %s"), + nvpair_name(pair)); + + switch (fnvpair_value_int32(pair)) { + case EEXIST: + zfs_error_aux(hdl, + dgettext(TEXT_DOMAIN, "snapshot is cloned")); + ret = zfs_error(hdl, EZFS_EXISTS, errbuf); + break; + default: + ret = zfs_standard_error(hdl, errno, errbuf); + break; } } @@ -4047,7 +4048,7 @@ zfs_userspace(zfs_handle_t *zhp, zfs_userquota_prop_t type, zc.zc_nvlist_dst_size = sizeof (buf); if (zfs_ioctl(hdl, ZFS_IOC_USERSPACE_MANY, &zc) != 0) { - char errbuf[ZFS_MAXNAMELEN + 32]; + char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, @@ -4069,37 +4070,83 @@ zfs_userspace(zfs_handle_t *zhp, zfs_userquota_prop_t type, return (0); } +struct holdarg { + nvlist_t *nvl; + const char *snapname; + const char *tag; + boolean_t recursive; +}; + +static int +zfs_hold_one(zfs_handle_t *zhp, void *arg) +{ + struct holdarg *ha = arg; + zfs_handle_t *szhp; + char name[ZFS_MAXNAMELEN]; + int rv = 0; + + (void) snprintf(name, sizeof (name), + "%s@%s", zhp->zfs_name, ha->snapname); + + szhp = make_dataset_handle(zhp->zfs_hdl, name); + if (szhp) { + fnvlist_add_string(ha->nvl, name, ha->tag); + zfs_close(szhp); + } + + if (ha->recursive) + rv = zfs_iter_filesystems(zhp, zfs_hold_one, ha); + zfs_close(zhp); + return (rv); +} + int zfs_hold(zfs_handle_t *zhp, const char *snapname, const char *tag, - boolean_t recursive, boolean_t temphold, boolean_t enoent_ok, - int cleanup_fd, uint64_t dsobj, uint64_t createtxg) + boolean_t recursive, boolean_t enoent_ok, int cleanup_fd) { - zfs_cmd_t zc = { 0 }; + int ret; + struct holdarg ha; + nvlist_t *errors; libzfs_handle_t *hdl = zhp->zfs_hdl; + char errbuf[1024]; + nvpair_t *elem; - ASSERT(!recursive || dsobj == 0); + ha.nvl = fnvlist_alloc(); + ha.snapname = snapname; + ha.tag = tag; + ha.recursive = recursive; + (void) zfs_hold_one(zfs_handle_dup(zhp), &ha); + ret = lzc_hold(ha.nvl, cleanup_fd, &errors); + fnvlist_free(ha.nvl); - (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); - (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); - if (strlcpy(zc.zc_string, tag, sizeof (zc.zc_string)) - >= sizeof (zc.zc_string)) - return (zfs_error(hdl, EZFS_TAGTOOLONG, tag)); - zc.zc_cookie = recursive; - zc.zc_temphold = temphold; - zc.zc_cleanup_fd = cleanup_fd; - zc.zc_sendobj = dsobj; - zc.zc_createtxg = createtxg; + if (ret == 0) + return (0); - if (zfs_ioctl(hdl, ZFS_IOC_HOLD, &zc) != 0) { - char errbuf[ZFS_MAXNAMELEN+32]; + if (nvlist_next_nvpair(errors, NULL) == NULL) { + /* no hold-specific errors */ + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, "cannot hold")); + switch (ret) { + case ENOTSUP: + zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, + "pool must be upgraded")); + (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); + break; + case EINVAL: + (void) zfs_error(hdl, EZFS_BADTYPE, errbuf); + break; + default: + (void) zfs_standard_error(hdl, ret, errbuf); + } + } - /* - * if it was recursive, the one that actually failed will be in - * zc.zc_name. - */ - (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, - "cannot hold '%s@%s'"), zc.zc_name, snapname); - switch (errno) { + for (elem = nvlist_next_nvpair(errors, NULL); + elem != NULL; + elem = nvlist_next_nvpair(errors, elem)) { + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, + "cannot hold snapshot '%s'"), nvpair_name(elem)); + switch (fnvpair_value_int32(elem)) { case E2BIG: /* * Temporary tags wind up having the ds object id @@ -4107,66 +4154,122 @@ zfs_hold(zfs_handle_t *zhp, const char *snapname, const char *tag, * above, it's still possible for the tag to wind * up being slightly too long. */ - return (zfs_error(hdl, EZFS_TAGTOOLONG, errbuf)); - case ENOTSUP: - zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, - "pool must be upgraded")); - return (zfs_error(hdl, EZFS_BADVERSION, errbuf)); + (void) zfs_error(hdl, EZFS_TAGTOOLONG, errbuf); + break; case EINVAL: - return (zfs_error(hdl, EZFS_BADTYPE, errbuf)); + (void) zfs_error(hdl, EZFS_BADTYPE, errbuf); + break; case EEXIST: - return (zfs_error(hdl, EZFS_REFTAG_HOLD, errbuf)); + (void) zfs_error(hdl, EZFS_REFTAG_HOLD, errbuf); + break; case ENOENT: if (enoent_ok) return (ENOENT); /* FALLTHROUGH */ default: - return (zfs_standard_error_fmt(hdl, errno, errbuf)); + (void) zfs_standard_error(hdl, + fnvpair_value_int32(elem), errbuf); } } - return (0); + fnvlist_free(errors); + return (ret); +} + +struct releasearg { + nvlist_t *nvl; + const char *snapname; + const char *tag; + boolean_t recursive; +}; + +static int +zfs_release_one(zfs_handle_t *zhp, void *arg) +{ + struct holdarg *ha = arg; + zfs_handle_t *szhp; + char name[ZFS_MAXNAMELEN]; + int rv = 0; + + (void) snprintf(name, sizeof (name), + "%s@%s", zhp->zfs_name, ha->snapname); + + szhp = make_dataset_handle(zhp->zfs_hdl, name); + if (szhp) { + nvlist_t *holds = fnvlist_alloc(); + fnvlist_add_boolean(holds, ha->tag); + fnvlist_add_nvlist(ha->nvl, name, holds); + zfs_close(szhp); + } + + if (ha->recursive) + rv = zfs_iter_filesystems(zhp, zfs_release_one, ha); + zfs_close(zhp); + return (rv); } int zfs_release(zfs_handle_t *zhp, const char *snapname, const char *tag, boolean_t recursive) { - zfs_cmd_t zc = { 0 }; + int ret; + struct holdarg ha; + nvlist_t *errors; + nvpair_t *elem; libzfs_handle_t *hdl = zhp->zfs_hdl; - (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); - (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); - if (strlcpy(zc.zc_string, tag, sizeof (zc.zc_string)) - >= sizeof (zc.zc_string)) - return (zfs_error(hdl, EZFS_TAGTOOLONG, tag)); - zc.zc_cookie = recursive; + ha.nvl = fnvlist_alloc(); + ha.snapname = snapname; + ha.tag = tag; + ha.recursive = recursive; + (void) zfs_release_one(zfs_handle_dup(zhp), &ha); + ret = lzc_release(ha.nvl, &errors); + fnvlist_free(ha.nvl); - if (zfs_ioctl(hdl, ZFS_IOC_RELEASE, &zc) != 0) { - char errbuf[ZFS_MAXNAMELEN+32]; + if (ret == 0) + return (0); + + if (nvlist_next_nvpair(errors, NULL) == NULL) { + /* no hold-specific errors */ + char errbuf[1024]; - /* - * if it was recursive, the one that actually failed will be in - * zc.zc_name. - */ (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, - "cannot release '%s' from '%s@%s'"), tag, zc.zc_name, - snapname); + "cannot release")); switch (errno) { - case ESRCH: - return (zfs_error(hdl, EZFS_REFTAG_RELE, errbuf)); case ENOTSUP: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be upgraded")); - return (zfs_error(hdl, EZFS_BADVERSION, errbuf)); - case EINVAL: - return (zfs_error(hdl, EZFS_BADTYPE, errbuf)); + (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); + break; default: - return (zfs_standard_error_fmt(hdl, errno, errbuf)); + (void) zfs_standard_error_fmt(hdl, errno, errbuf); } } - return (0); + for (elem = nvlist_next_nvpair(errors, NULL); + elem != NULL; + elem = nvlist_next_nvpair(errors, elem)) { + char errbuf[1024]; + + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, + "cannot release hold from snapshot '%s'"), + nvpair_name(elem)); + switch (fnvpair_value_int32(elem)) { + case ESRCH: + (void) zfs_error(hdl, EZFS_REFTAG_RELE, errbuf); + break; + case EINVAL: + (void) zfs_error(hdl, EZFS_BADTYPE, errbuf); + break; + default: + (void) zfs_standard_error_fmt(hdl, + fnvpair_value_int32(elem), errbuf); + } + } + + fnvlist_free(errors); + return (ret); } int @@ -4177,7 +4280,7 @@ zfs_get_fsacl(zfs_handle_t *zhp, nvlist_t **nvl) int nvsz = 2048; void *nvbuf; int err = 0; - char errbuf[ZFS_MAXNAMELEN+32]; + char errbuf[1024]; assert(zhp->zfs_type == ZFS_TYPE_VOLUME || zhp->zfs_type == ZFS_TYPE_FILESYSTEM); @@ -4242,7 +4345,7 @@ zfs_set_fsacl(zfs_handle_t *zhp, boolean_t un, nvlist_t *nvl) zfs_cmd_t zc = { 0 }; libzfs_handle_t *hdl = zhp->zfs_hdl; char *nvbuf; - char errbuf[ZFS_MAXNAMELEN+32]; + char errbuf[1024]; size_t nvsz; int err; @@ -4293,38 +4396,18 @@ zfs_set_fsacl(zfs_handle_t *zhp, boolean_t un, nvlist_t *nvl) int zfs_get_holds(zfs_handle_t *zhp, nvlist_t **nvl) { - zfs_cmd_t zc = { 0 }; - libzfs_handle_t *hdl = zhp->zfs_hdl; - int nvsz = 2048; - void *nvbuf; - int err = 0; - char errbuf[ZFS_MAXNAMELEN+32]; + int err; + char errbuf[1024]; - assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); + err = lzc_get_holds(zhp->zfs_name, nvl); -tryagain: + if (err != 0) { + libzfs_handle_t *hdl = zhp->zfs_hdl; - nvbuf = malloc(nvsz); - if (nvbuf == NULL) { - err = (zfs_error(hdl, EZFS_NOMEM, strerror(errno))); - goto out; - } - - zc.zc_nvlist_dst_size = nvsz; - zc.zc_nvlist_dst = (uintptr_t)nvbuf; - - (void) strlcpy(zc.zc_name, zhp->zfs_name, ZFS_MAXNAMELEN); - - if (zfs_ioctl(hdl, ZFS_IOC_GET_HOLDS, &zc) != 0) { (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot get holds for '%s'"), - zc.zc_name); - switch (errno) { - case ENOMEM: - free(nvbuf); - nvsz = zc.zc_nvlist_dst_size; - goto tryagain; - + zhp->zfs_name); + switch (err) { case ENOTSUP: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be upgraded")); @@ -4340,19 +4423,8 @@ zfs_get_holds(zfs_handle_t *zhp, nvlist_t **nvl) err = zfs_standard_error_fmt(hdl, errno, errbuf); break; } - } else { - /* success */ - int rc = nvlist_unpack(nvbuf, zc.zc_nvlist_dst_size, nvl, 0); - if (rc) { - (void) snprintf(errbuf, sizeof (errbuf), - dgettext(TEXT_DOMAIN, "cannot get holds for '%s'"), - zc.zc_name); - err = zfs_standard_error_fmt(hdl, rc, errbuf); - } } - free(nvbuf); -out: return (err); } diff --git a/lib/libzfs/common/libzfs_sendrecv.c b/lib/libzfs/common/libzfs_sendrecv.c index ee6e64319e40..ea0634ee70c4 100644 --- a/lib/libzfs/common/libzfs_sendrecv.c +++ b/lib/libzfs/common/libzfs_sendrecv.c @@ -972,9 +972,7 @@ hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd) */ if (pzhp) { error = zfs_hold(pzhp, thissnap, sdd->holdtag, - B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd, - zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID), - zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG)); + B_FALSE, B_TRUE, sdd->cleanup_fd); zfs_close(pzhp); } @@ -1713,12 +1711,11 @@ recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, err = ENOENT; } - if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) { + if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { seq++; - (void) strncpy(newname, name, baselen); - (void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen, - "recv-%u-%u", getpid(), seq); + (void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u", + baselen, name, getpid(), seq); (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value)); if (flags->verbose) { @@ -2643,7 +2640,6 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, /* * Determine name of destination snapshot, store in zc_value. */ - (void) strcpy(zc.zc_top_ds, tosnap); (void) strcpy(zc.zc_value, tosnap); (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); free(cp); diff --git a/lib/libzfs_core/common/libzfs_core.c b/lib/libzfs_core/common/libzfs_core.c index 73afd50b8de2..5e8ee0779673 100644 --- a/lib/libzfs_core/common/libzfs_core.c +++ b/lib/libzfs_core/common/libzfs_core.c @@ -132,6 +132,7 @@ lzc_ioctl(zfs_ioc_t ioc, const char *name, zc.zc_nvlist_src_size = size; if (resultp != NULL) { + *resultp = NULL; zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); zc.zc_nvlist_dst = (uint64_t)(uintptr_t) malloc(zc.zc_nvlist_dst_size); @@ -159,8 +160,6 @@ lzc_ioctl(zfs_ioc_t ioc, const char *name, if (zc.zc_nvlist_dst_filled) { *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, zc.zc_nvlist_dst_size); - } else if (resultp != NULL) { - *resultp = NULL; } out: @@ -209,7 +208,7 @@ lzc_clone(const char *fsname, const char *origin, * The value will be the (int32) error code. * * The return value will be 0 if all snapshots were created, otherwise it will - * be the errno of a (undetermined) snapshot that failed. + * be the errno of a (unspecified) snapshot that failed. */ int lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) @@ -258,7 +257,7 @@ lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) * The return value will be 0 if all snapshots were destroyed (or marked for * later destruction if 'defer' is set) or didn't exist to begin with. * - * Otherwise the return value will be the errno of a (undetermined) snapshot + * Otherwise the return value will be the errno of a (unspecified) snapshot * that failed, no snapshots will be destroyed, and the errlist will have an * entry for each snapshot that failed. The value in the errlist will be * the (int32) error code. @@ -332,6 +331,101 @@ lzc_exists(const char *dataset) return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); } +/* + * Create "user holds" on snapshots. If there is a hold on a snapshot, + * the snapshot can not be destroyed. (However, it can be marked for deletion + * by lzc_destroy_snaps(defer=B_TRUE).) + * + * The keys in the nvlist are snapshot names. + * The snapshots must all be in the same pool. + * The value is the name of the hold (string type). + * + * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). + * In this case, when the cleanup_fd is closed (including on process + * termination), the holds will be released. If the system is shut down + * uncleanly, the holds will be released when the pool is next opened + * or imported. + * + * The return value will be 0 if all holds were created. Otherwise the return + * value will be the errno of a (unspecified) hold that failed, no holds will + * be created, and the errlist will have an entry for each hold that + * failed (name = snapshot). The value in the errlist will be the error + * code (int32). + */ +int +lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) +{ + char pool[MAXNAMELEN]; + nvlist_t *args; + nvpair_t *elem; + int error; + + /* determine the pool name */ + elem = nvlist_next_nvpair(holds, NULL); + if (elem == NULL) + return (0); + (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); + pool[strcspn(pool, "/@")] = '\0'; + + args = fnvlist_alloc(); + fnvlist_add_nvlist(args, "holds", holds); + if (cleanup_fd != -1) + fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); + + error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); + nvlist_free(args); + return (error); +} + +/* + * Release "user holds" on snapshots. If the snapshot has been marked for + * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have + * any clones, and all the user holds are removed, then the snapshot will be + * destroyed. + * + * The keys in the nvlist are snapshot names. + * The snapshots must all be in the same pool. + * The value is a nvlist whose keys are the holds to remove. + * + * The return value will be 0 if all holds were removed. + * Otherwise the return value will be the errno of a (unspecified) release + * that failed, no holds will be released, and the errlist will have an + * entry for each snapshot that has failed releases (name = snapshot). + * The value in the errlist will be the error code (int32) of a failed release. + */ +int +lzc_release(nvlist_t *holds, nvlist_t **errlist) +{ + char pool[MAXNAMELEN]; + nvpair_t *elem; + + /* determine the pool name */ + elem = nvlist_next_nvpair(holds, NULL); + if (elem == NULL) + return (0); + (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); + pool[strcspn(pool, "/@")] = '\0'; + + return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); +} + +/* + * Retrieve list of user holds on the specified snapshot. + * + * On success, *holdsp will be set to a nvlist which the caller must free. + * The keys are the names of the holds, and the value is the creation time + * of the hold (uint64) in seconds since the epoch. + */ +int +lzc_get_holds(const char *snapname, nvlist_t **holdsp) +{ + int error; + nvlist_t *innvl = fnvlist_alloc(); + error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); + fnvlist_free(innvl); + return (error); +} + /* * If fromsnap is NULL, a full (non-incremental) stream will be sent. */ diff --git a/lib/libzfs_core/common/libzfs_core.h b/lib/libzfs_core/common/libzfs_core.h index 9edc884a14d1..f5fd6cda9f0d 100644 --- a/lib/libzfs_core/common/libzfs_core.h +++ b/lib/libzfs_core/common/libzfs_core.h @@ -46,6 +46,10 @@ int lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist); int lzc_snaprange_space(const char *firstsnap, const char *lastsnap, uint64_t *usedp); +int lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist); +int lzc_release(nvlist_t *holds, nvlist_t **errlist); +int lzc_get_holds(const char *snapname, nvlist_t **holdsp); + int lzc_send(const char *snapname, const char *fromsnap, int fd); int lzc_receive(const char *snapname, nvlist_t *props, const char *origin, boolean_t force, int fd); diff --git a/lib/libzpool/common/kernel.c b/lib/libzpool/common/kernel.c index cc0d5428beeb..96280941a646 100644 --- a/lib/libzpool/common/kernel.c +++ b/lib/libzpool/common/kernel.c @@ -34,6 +34,7 @@ #include #include #include +#include #include #include #include @@ -859,6 +860,8 @@ umem_out_of_memory(void) void kernel_init(int mode) { + extern uint_t rrw_tsd_key; + umem_nofail_callback(umem_out_of_memory); physmem = sysconf(_SC_PHYS_PAGES); @@ -877,6 +880,8 @@ kernel_init(int mode) mutex_init(&cpu_lock, NULL, MUTEX_DEFAULT, NULL); spa_init(mode); + + tsd_create(&rrw_tsd_key, rrw_tsd_destroy); } void diff --git a/lib/libzpool/common/sys/zfs_context.h b/lib/libzpool/common/sys/zfs_context.h index bcb27cf1a5cd..7802da4e2b00 100644 --- a/lib/libzpool/common/sys/zfs_context.h +++ b/lib/libzpool/common/sys/zfs_context.h @@ -61,6 +61,8 @@ extern "C" { #include #include #include +#include +#include #include #include #include @@ -224,6 +226,9 @@ typedef int krw_t; #undef RW_WRITE_HELD #define RW_WRITE_HELD(x) _rw_write_held(&(x)->rw_lock) +#undef RW_LOCK_HELD +#define RW_LOCK_HELD(x) (RW_READ_HELD(x) || RW_WRITE_HELD(x)) + extern void rw_init(krwlock_t *rwlp, char *name, int type, void *arg); extern void rw_destroy(krwlock_t *rwlp); extern void rw_enter(krwlock_t *rwlp, krw_t rw); @@ -252,6 +257,14 @@ extern clock_t cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime); extern void cv_signal(kcondvar_t *cv); extern void cv_broadcast(kcondvar_t *cv); +/* + * Thread-specific data + */ +#define tsd_get(k) pthread_getspecific(k) +#define tsd_set(k, v) pthread_setspecific(k, v) +#define tsd_create(kp, d) pthread_key_create(kp, d) +#define tsd_destroy(kp) /* nothing */ + /* * kstat creation, installation and deletion */ @@ -519,7 +532,7 @@ typedef struct callb_cpr { #define INGLOBALZONE(z) (1) extern char *kmem_asprintf(const char *fmt, ...); -#define strfree(str) kmem_free((str), strlen(str)+1) +#define strfree(str) kmem_free((str), strlen(str) + 1) /* * Hostname information diff --git a/man/man1m/zfs.1m b/man/man1m/zfs.1m index 18441dad6ef8..c4d579d7cbe4 100644 --- a/man/man1m/zfs.1m +++ b/man/man1m/zfs.1m @@ -1868,7 +1868,9 @@ descendent file systems. .ad .sp .6 .RS 4n -Recursively destroy all dependents. +Recursively destroy all clones of these snapshots, including the clones, +snapshots, and children. If this flag is specified, the \fB-d\fR flag will +have no effect. .RE .sp @@ -1904,7 +1906,7 @@ Print verbose information about the deleted data. .RE .sp -Extreme care should be taken when applying either the \fB-r\fR or the \fB-f\fR +Extreme care should be taken when applying either the \fB-r\fR or the \fB-R\fR options, as they can destroy large portions of a pool and cause unexpected behavior for mounted file systems in use. .RE diff --git a/uts/common/Makefile.files b/uts/common/Makefile.files index 1eb0e3158a78..e0fa959279aa 100644 --- a/uts/common/Makefile.files +++ b/uts/common/Makefile.files @@ -1346,8 +1346,10 @@ ZFS_COMMON_OBJS += \ dsl_dir.o \ dsl_dataset.o \ dsl_deadlist.o \ + dsl_destroy.o \ dsl_pool.o \ dsl_synctask.o \ + dsl_userhold.o \ dmu_zfetch.o \ dsl_deleg.o \ dsl_prop.o \ @@ -1358,6 +1360,7 @@ ZFS_COMMON_OBJS += \ lzjb.o \ metaslab.o \ refcount.o \ + rrwlock.o \ sa.o \ sha256.o \ spa.o \ @@ -1417,7 +1420,6 @@ ZFS_OBJS += \ zfs_onexit.o \ zfs_replay.o \ zfs_rlock.o \ - rrwlock.o \ zfs_vfsops.o \ zfs_vnops.o \ zvol.o diff --git a/uts/common/fs/zfs/arc.c b/uts/common/fs/zfs/arc.c index ca3baeaeebb9..9588e404332f 100644 --- a/uts/common/fs/zfs/arc.c +++ b/uts/common/fs/zfs/arc.c @@ -1633,12 +1633,12 @@ arc_buf_free(arc_buf_t *buf, void *tag) } } -int +boolean_t arc_buf_remove_ref(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock = HDR_LOCK(hdr); - int no_callback = (buf->b_efunc == NULL); + boolean_t no_callback = (buf->b_efunc == NULL); if (hdr->b_state == arc_anon) { ASSERT(hdr->b_datacnt == 1); @@ -1843,7 +1843,7 @@ arc_evict(arc_state_t *state, uint64_t spa, int64_t bytes, boolean_t recycle, ARCSTAT_INCR(arcstat_mutex_miss, missed); /* - * We have just evicted some date into the ghost state, make + * We have just evicted some data into the ghost state, make * sure we also adjust the ghost state size if necessary. */ if (arc_no_grow && @@ -2622,7 +2622,7 @@ arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) { if (zio == NULL || zio->io_error == 0) bcopy(buf->b_data, arg, buf->b_hdr->b_size); - VERIFY(arc_buf_remove_ref(buf, arg) == 1); + VERIFY(arc_buf_remove_ref(buf, arg)); } /* a generic arc_done_func_t */ @@ -2631,7 +2631,7 @@ arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) { arc_buf_t **bufp = arg; if (zio && zio->io_error) { - VERIFY(arc_buf_remove_ref(buf, arg) == 1); + VERIFY(arc_buf_remove_ref(buf, arg)); *bufp = NULL; } else { *bufp = buf; diff --git a/uts/common/fs/zfs/bplist.c b/uts/common/fs/zfs/bplist.c index 066ccc6b1e05..ee12db3a266d 100644 --- a/uts/common/fs/zfs/bplist.c +++ b/uts/common/fs/zfs/bplist.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include @@ -52,6 +53,12 @@ bplist_append(bplist_t *bpl, const blkptr_t *bp) mutex_exit(&bpl->bpl_lock); } +/* + * To aid debugging, we keep the most recently removed entry. This way if + * we are in the callback, we can easily locate the entry. + */ +static bplist_entry_t *bplist_iterate_last_removed; + void bplist_iterate(bplist_t *bpl, bplist_itor_t *func, void *arg, dmu_tx_t *tx) { @@ -59,6 +66,7 @@ bplist_iterate(bplist_t *bpl, bplist_itor_t *func, void *arg, dmu_tx_t *tx) mutex_enter(&bpl->bpl_lock); while (bpe = list_head(&bpl->bpl_list)) { + bplist_iterate_last_removed = bpe; list_remove(&bpl->bpl_list, bpe); mutex_exit(&bpl->bpl_lock); func(arg, &bpe->bpe_blk, tx); diff --git a/uts/common/fs/zfs/bpobj.c b/uts/common/fs/zfs/bpobj.c index 1920da4408c8..bcb5f331f0b3 100644 --- a/uts/common/fs/zfs/bpobj.c +++ b/uts/common/fs/zfs/bpobj.c @@ -392,6 +392,10 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx) DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx); } + dmu_object_info_t doi; + ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi)); + ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ); + mutex_enter(&bpo->bpo_lock); dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj), diff --git a/uts/common/fs/zfs/dbuf.c b/uts/common/fs/zfs/dbuf.c index 8bf3d099ef0c..57abfa100340 100644 --- a/uts/common/fs/zfs/dbuf.c +++ b/uts/common/fs/zfs/dbuf.c @@ -39,7 +39,7 @@ #include static void dbuf_destroy(dmu_buf_impl_t *db); -static int dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); +static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx); /* @@ -499,7 +499,7 @@ dbuf_read_done(zio_t *zio, arc_buf_t *buf, void *vdb) } else { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); - VERIFY(arc_buf_remove_ref(buf, db) == 1); + VERIFY(arc_buf_remove_ref(buf, db)); db->db_state = DB_UNCACHED; } cv_broadcast(&db->db_changed); @@ -828,10 +828,12 @@ dbuf_free_range(dnode_t *dn, uint64_t start, uint64_t end, dmu_tx_t *tx) continue; /* found a level 0 buffer in the range */ - if (dbuf_undirty(db, tx)) - continue; - mutex_enter(&db->db_mtx); + if (dbuf_undirty(db, tx)) { + /* mutex has been dropped and dbuf destroyed */ + continue; + } + if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { @@ -958,7 +960,7 @@ dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); - VERIFY(arc_buf_remove_ref(obuf, db) == 1); + VERIFY(arc_buf_remove_ref(obuf, db)); db->db.db_size = size; if (db->db_level == 0) { @@ -1258,7 +1260,10 @@ dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) return (dr); } -static int +/* + * Return TRUE if this evicted the dbuf. + */ +static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; @@ -1267,18 +1272,17 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) ASSERT(txg != 0); ASSERT(db->db_blkid != DMU_BONUS_BLKID); + ASSERT0(db->db_level); + ASSERT(MUTEX_HELD(&db->db_mtx)); - mutex_enter(&db->db_mtx); /* * If this buffer is not dirty, we're done. */ for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) if (dr->dr_txg <= txg) break; - if (dr == NULL || dr->dr_txg < txg) { - mutex_exit(&db->db_mtx); - return (0); - } + if (dr == NULL || dr->dr_txg < txg) + return (B_FALSE); ASSERT(dr->dr_txg == txg); ASSERT(dr->dr_dbuf == db); @@ -1286,24 +1290,12 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) dn = DB_DNODE(db); /* - * If this buffer is currently held, we cannot undirty - * it, since one of the current holders may be in the - * middle of an update. Note that users of dbuf_undirty() - * should not place a hold on the dbuf before the call. - * Also note: we can get here with a spill block, so - * test for that similar to how dbuf_dirty does. + * Note: This code will probably work even if there are concurrent + * holders, but it is untested in that scenerio, as the ZPL and + * ztest have additional locking (the range locks) that prevents + * that type of concurrent access. */ - if (refcount_count(&db->db_holds) > db->db_dirtycnt) { - mutex_exit(&db->db_mtx); - /* Make sure we don't toss this buffer at sync phase */ - if (db->db_blkid != DMU_SPILL_BLKID) { - mutex_enter(&dn->dn_mtx); - dnode_clear_range(dn, db->db_blkid, 1, tx); - mutex_exit(&dn->dn_mtx); - } - DB_DNODE_EXIT(db); - return (0); - } + ASSERT3U(refcount_count(&db->db_holds), ==, db->db_dirtycnt); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); @@ -1332,21 +1324,13 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) } DB_DNODE_EXIT(db); - if (db->db_level == 0) { - if (db->db_state != DB_NOFILL) { - dbuf_unoverride(dr); + if (db->db_state != DB_NOFILL) { + dbuf_unoverride(dr); - ASSERT(db->db_buf != NULL); - ASSERT(dr->dt.dl.dr_data != NULL); - if (dr->dt.dl.dr_data != db->db_buf) - VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, - db) == 1); - } - } else { ASSERT(db->db_buf != NULL); - ASSERT(list_head(&dr->dt.di.dr_children) == NULL); - mutex_destroy(&dr->dt.di.dr_mtx); - list_destroy(&dr->dt.di.dr_children); + ASSERT(dr->dt.dl.dr_data != NULL); + if (dr->dt.dl.dr_data != db->db_buf) + VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db)); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); @@ -1358,13 +1342,12 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) ASSERT(db->db_state == DB_NOFILL || arc_released(buf)); dbuf_set_data(db, NULL); - VERIFY(arc_buf_remove_ref(buf, db) == 1); + VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(db); - return (1); + return (B_TRUE); } - mutex_exit(&db->db_mtx); - return (0); + return (B_FALSE); } #pragma weak dmu_buf_will_dirty = dbuf_will_dirty @@ -1463,7 +1446,7 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); bcopy(buf->b_data, db->db.db_data, db->db.db_size); - VERIFY(arc_buf_remove_ref(buf, db) == 1); + VERIFY(arc_buf_remove_ref(buf, db)); xuio_stat_wbuf_copied(); return; } @@ -1481,10 +1464,10 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; - VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1); + VERIFY(arc_buf_remove_ref(db->db_buf, db)); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); - VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1); + VERIFY(arc_buf_remove_ref(db->db_buf, db)); } db->db_buf = NULL; } @@ -2067,10 +2050,10 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag) * This dbuf has anonymous data associated with it. */ dbuf_set_data(db, NULL); - VERIFY(arc_buf_remove_ref(buf, db) == 1); + VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(db); } else { - VERIFY(arc_buf_remove_ref(db->db_buf, db) == 0); + VERIFY(!arc_buf_remove_ref(db->db_buf, db)); /* * A dbuf will be eligible for eviction if either the @@ -2567,7 +2550,7 @@ dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb) if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, - db) == 1); + db)); else if (!arc_released(db->db_buf)) arc_set_callback(db->db_buf, dbuf_do_evict, db); } diff --git a/uts/common/fs/zfs/dmu.c b/uts/common/fs/zfs/dmu.c index 21cdd7c1eda4..6ee37ac8b774 100644 --- a/uts/common/fs/zfs/dmu.c +++ b/uts/common/fs/zfs/dmu.c @@ -1194,7 +1194,7 @@ void dmu_return_arcbuf(arc_buf_t *buf) { arc_return_buf(buf, FTAG); - VERIFY(arc_buf_remove_ref(buf, FTAG) == 1); + VERIFY(arc_buf_remove_ref(buf, FTAG)); } /* diff --git a/uts/common/fs/zfs/dmu_diff.c b/uts/common/fs/zfs/dmu_diff.c index dc237780c0b0..2d1aaa4c449d 100644 --- a/uts/common/fs/zfs/dmu_diff.c +++ b/uts/common/fs/zfs/dmu_diff.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include @@ -155,51 +156,49 @@ diff_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, } int -dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp, offset_t *offp) +dmu_diff(const char *tosnap_name, const char *fromsnap_name, + struct vnode *vp, offset_t *offp) { struct diffarg da; - dsl_dataset_t *ds = tosnap->os_dsl_dataset; - dsl_dataset_t *fromds = fromsnap->os_dsl_dataset; - dsl_dataset_t *findds; - dsl_dataset_t *relds; - int err = 0; + dsl_dataset_t *fromsnap; + dsl_dataset_t *tosnap; + dsl_pool_t *dp; + int error; + uint64_t fromtxg; - /* make certain we are looking at snapshots */ - if (!dsl_dataset_is_snapshot(ds) || !dsl_dataset_is_snapshot(fromds)) + if (strchr(tosnap_name, '@') == NULL || + strchr(fromsnap_name, '@') == NULL) return (EINVAL); - /* fromsnap must be earlier and from the same lineage as tosnap */ - if (fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg) - return (EXDEV); + error = dsl_pool_hold(tosnap_name, FTAG, &dp); + if (error != 0) + return (error); - relds = NULL; - findds = ds; - - while (fromds->ds_dir != findds->ds_dir) { - dsl_pool_t *dp = ds->ds_dir->dd_pool; - - if (!dsl_dir_is_clone(findds->ds_dir)) { - if (relds) - dsl_dataset_rele(relds, FTAG); - return (EXDEV); - } - - rw_enter(&dp->dp_config_rwlock, RW_READER); - err = dsl_dataset_hold_obj(dp, - findds->ds_dir->dd_phys->dd_origin_obj, FTAG, &findds); - rw_exit(&dp->dp_config_rwlock); - - if (relds) - dsl_dataset_rele(relds, FTAG); - - if (err) - return (EXDEV); - - relds = findds; + error = dsl_dataset_hold(dp, tosnap_name, FTAG, &tosnap); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); } - if (relds) - dsl_dataset_rele(relds, FTAG); + error = dsl_dataset_hold(dp, fromsnap_name, FTAG, &fromsnap); + if (error != 0) { + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); + return (error); + } + + if (!dsl_dataset_is_before(tosnap, fromsnap)) { + dsl_dataset_rele(fromsnap, FTAG); + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); + return (EXDEV); + } + + fromtxg = fromsnap->ds_phys->ds_creation_txg; + dsl_dataset_rele(fromsnap, FTAG); + + dsl_dataset_long_hold(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); da.da_vp = vp; da.da_offp = offp; @@ -207,15 +206,18 @@ dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp, offset_t *offp) da.da_ddr.ddr_first = da.da_ddr.ddr_last = 0; da.da_err = 0; - err = traverse_dataset(ds, fromds->ds_phys->ds_creation_txg, + error = traverse_dataset(tosnap, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, diff_cb, &da); - if (err) { - da.da_err = err; + if (error != 0) { + da.da_err = error; } else { /* we set the da.da_err we return as side-effect */ (void) write_record(&da); } + dsl_dataset_long_rele(tosnap, FTAG); + dsl_dataset_rele(tosnap, FTAG); + return (da.da_err); } diff --git a/uts/common/fs/zfs/dmu_objset.c b/uts/common/fs/zfs/dmu_objset.c index 74c1192017bc..a646f4045064 100644 --- a/uts/common/fs/zfs/dmu_objset.c +++ b/uts/common/fs/zfs/dmu_objset.c @@ -44,6 +44,7 @@ #include #include #include +#include /* * Needed to close a window in dnode_move() that allows the objset to be freed @@ -280,7 +281,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, err = arc_read(NULL, spa, os->os_rootbp, arc_getbuf_func, &os->os_phys_buf, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); - if (err) { + if (err != 0) { kmem_free(os, sizeof (objset_t)); /* convert checksum errors into IO errors */ if (err == ECKSUM) @@ -320,34 +321,49 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, * checksum/compression/copies. */ if (ds) { - err = dsl_prop_register(ds, "primarycache", + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), primary_cache_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "secondarycache", + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), secondary_cache_changed_cb, os); - if (!dsl_dataset_is_snapshot(ds)) { - if (err == 0) - err = dsl_prop_register(ds, "checksum", - checksum_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "compression", - compression_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "copies", - copies_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "dedup", - dedup_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "logbias", - logbias_changed_cb, os); - if (err == 0) - err = dsl_prop_register(ds, "sync", - sync_changed_cb, os); } - if (err) { + if (!dsl_dataset_is_snapshot(ds)) { + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_CHECKSUM), + checksum_changed_cb, os); + } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_COMPRESSION), + compression_changed_cb, os); + } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_COPIES), + copies_changed_cb, os); + } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_DEDUP), + dedup_changed_cb, os); + } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_LOGBIAS), + logbias_changed_cb, os); + } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_SYNC), + sync_changed_cb, os); + } + } + if (err != 0) { VERIFY(arc_buf_remove_ref(os->os_phys_buf, - &os->os_phys_buf) == 1); + &os->os_phys_buf)); kmem_free(os, sizeof (objset_t)); return (err); } @@ -425,44 +441,66 @@ dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) return (err); } -/* called from zpl */ +/* + * Holds the pool while the objset is held. Therefore only one objset + * can be held at a time. + */ int dmu_objset_hold(const char *name, void *tag, objset_t **osp) { + dsl_pool_t *dp; dsl_dataset_t *ds; int err; - err = dsl_dataset_hold(name, tag, &ds); - if (err) + err = dsl_pool_hold(name, tag, &dp); + if (err != 0) return (err); + err = dsl_dataset_hold(dp, name, tag, &ds); + if (err != 0) { + dsl_pool_rele(dp, tag); + return (err); + } err = dmu_objset_from_ds(ds, osp); - if (err) + if (err != 0) { dsl_dataset_rele(ds, tag); + dsl_pool_rele(dp, tag); + } return (err); } -/* called from zpl */ +/* + * dsl_pool must not be held when this is called. + * Upon successful return, there will be a longhold on the dataset, + * and the dsl_pool will not be held. + */ int dmu_objset_own(const char *name, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp) { + dsl_pool_t *dp; dsl_dataset_t *ds; int err; - err = dsl_dataset_own(name, B_FALSE, tag, &ds); - if (err) + err = dsl_pool_hold(name, FTAG, &dp); + if (err != 0) return (err); + err = dsl_dataset_own(dp, name, tag, &ds); + if (err != 0) { + dsl_pool_rele(dp, FTAG); + return (err); + } err = dmu_objset_from_ds(ds, osp); - if (err) { + dsl_pool_rele(dp, FTAG); + if (err != 0) { dsl_dataset_disown(ds, tag); } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { - dmu_objset_disown(*osp, tag); + dsl_dataset_disown(ds, tag); return (EINVAL); } else if (!readonly && dsl_dataset_is_snapshot(ds)) { - dmu_objset_disown(*osp, tag); + dsl_dataset_disown(ds, tag); return (EROFS); } return (err); @@ -471,7 +509,9 @@ dmu_objset_own(const char *name, dmu_objset_type_t type, void dmu_objset_rele(objset_t *os, void *tag) { + dsl_pool_t *dp = dmu_objset_pool(os); dsl_dataset_rele(os->os_dsl_dataset, tag); + dsl_pool_rele(dp, tag); } void @@ -480,7 +520,7 @@ dmu_objset_disown(objset_t *os, void *tag) dsl_dataset_disown(os->os_dsl_dataset, tag); } -int +void dmu_objset_evict_dbufs(objset_t *os) { dnode_t *dn; @@ -515,9 +555,7 @@ dmu_objset_evict_dbufs(objset_t *os) mutex_enter(&os->os_lock); dn = next_dn; } - dn = list_head(&os->os_dnodes); mutex_exit(&os->os_lock); - return (dn != DMU_META_DNODE(os)); } void @@ -530,33 +568,37 @@ dmu_objset_evict(objset_t *os) if (ds) { if (!dsl_dataset_is_snapshot(ds)) { - VERIFY(0 == dsl_prop_unregister(ds, "checksum", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "compression", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_COMPRESSION), compression_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "copies", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_COPIES), copies_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "dedup", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_DEDUP), dedup_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "logbias", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_LOGBIAS), logbias_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "sync", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_SYNC), sync_changed_cb, os)); } - VERIFY(0 == dsl_prop_unregister(ds, "primarycache", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), primary_cache_changed_cb, os)); - VERIFY(0 == dsl_prop_unregister(ds, "secondarycache", + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), secondary_cache_changed_cb, os)); } if (os->os_sa) sa_tear_down(os); - /* - * We should need only a single pass over the dnode list, since - * nothing can be added to the list at this point. - */ - (void) dmu_objset_evict_dbufs(os); + dmu_objset_evict_dbufs(os); dnode_special_close(&os->os_meta_dnode); if (DMU_USERUSED_DNODE(os)) { @@ -567,7 +609,7 @@ dmu_objset_evict(objset_t *os) ASSERT3P(list_head(&os->os_dnodes), ==, NULL); - VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1); + VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf)); /* * This is a barrier to prevent the objset from going away in @@ -599,10 +641,11 @@ dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, dnode_t *mdn; ASSERT(dmu_tx_is_syncing(tx)); + if (ds != NULL) - VERIFY(0 == dmu_objset_from_ds(ds, &os)); + VERIFY0(dmu_objset_from_ds(ds, &os)); else - VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os)); + VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); mdn = DMU_META_DNODE(os); @@ -650,359 +693,181 @@ dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, return (os); } -struct oscarg { - void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx); - void *userarg; - dsl_dataset_t *clone_origin; - const char *lastname; - dmu_objset_type_t type; - uint64_t flags; - cred_t *cr; -}; +typedef struct dmu_objset_create_arg { + const char *doca_name; + cred_t *doca_cred; + void (*doca_userfunc)(objset_t *os, void *arg, + cred_t *cr, dmu_tx_t *tx); + void *doca_userarg; + dmu_objset_type_t doca_type; + uint64_t doca_flags; +} dmu_objset_create_arg_t; /*ARGSUSED*/ static int -dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx) +dmu_objset_create_check(void *arg, dmu_tx_t *tx) { - dsl_dir_t *dd = arg1; - struct oscarg *oa = arg2; - objset_t *mos = dd->dd_pool->dp_meta_objset; - int err; - uint64_t ddobj; + dmu_objset_create_arg_t *doca = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dir_t *pdd; + const char *tail; + int error; - err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj, - oa->lastname, sizeof (uint64_t), 1, &ddobj); - if (err != ENOENT) - return (err ? err : EEXIST); + if (strchr(doca->doca_name, '@') != NULL) + return (EINVAL); - if (oa->clone_origin != NULL) { - /* You can't clone across pools. */ - if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool) - return (EXDEV); - - /* You can only clone snapshots, not the head datasets. */ - if (!dsl_dataset_is_snapshot(oa->clone_origin)) - return (EINVAL); + error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); + if (error != 0) + return (error); + if (tail == NULL) { + dsl_dir_rele(pdd, FTAG); + return (EEXIST); } + dsl_dir_rele(pdd, FTAG); return (0); } static void -dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dmu_objset_create_sync(void *arg, dmu_tx_t *tx) { - dsl_dir_t *dd = arg1; - spa_t *spa = dd->dd_pool->dp_spa; - struct oscarg *oa = arg2; - uint64_t obj; + dmu_objset_create_arg_t *doca = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dir_t *pdd; + const char *tail; dsl_dataset_t *ds; + uint64_t obj; blkptr_t *bp; + objset_t *os; - ASSERT(dmu_tx_is_syncing(tx)); + VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); - obj = dsl_dataset_create_sync(dd, oa->lastname, - oa->clone_origin, oa->flags, oa->cr, tx); + obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, + doca->doca_cred, tx); - VERIFY3U(0, ==, dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds)); + VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); bp = dsl_dataset_get_blkptr(ds); - if (BP_IS_HOLE(bp)) { - objset_t *os = - dmu_objset_create_impl(spa, ds, bp, oa->type, tx); + os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, + ds, bp, doca->doca_type, tx); - if (oa->userfunc) - oa->userfunc(os, oa->userarg, oa->cr, tx); + if (doca->doca_userfunc != NULL) { + doca->doca_userfunc(os, doca->doca_userarg, + doca->doca_cred, tx); } - if (oa->clone_origin == NULL) { - spa_history_log_internal_ds(ds, "create", tx, ""); - } else { - char namebuf[MAXNAMELEN]; - dsl_dataset_name(oa->clone_origin, namebuf); - spa_history_log_internal_ds(ds, "clone", tx, - "origin=%s (%llu)", namebuf, oa->clone_origin->ds_object); - } + spa_history_log_internal_ds(ds, "create", tx, ""); dsl_dataset_rele(ds, FTAG); + dsl_dir_rele(pdd, FTAG); } int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) { - dsl_dir_t *pdd; - const char *tail; - int err = 0; - struct oscarg oa = { 0 }; + dmu_objset_create_arg_t doca; - ASSERT(strchr(name, '@') == NULL); - err = dsl_dir_open(name, FTAG, &pdd, &tail); - if (err) - return (err); - if (tail == NULL) { - dsl_dir_close(pdd, FTAG); - return (EEXIST); - } + doca.doca_name = name; + doca.doca_cred = CRED(); + doca.doca_flags = flags; + doca.doca_userfunc = func; + doca.doca_userarg = arg; + doca.doca_type = type; - oa.userfunc = func; - oa.userarg = arg; - oa.lastname = tail; - oa.type = type; - oa.flags = flags; - oa.cr = CRED(); - - err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check, - dmu_objset_create_sync, pdd, &oa, 5); - dsl_dir_close(pdd, FTAG); - return (err); + return (dsl_sync_task(name, + dmu_objset_create_check, dmu_objset_create_sync, &doca, 5)); } -int -dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags) -{ - dsl_dir_t *pdd; - const char *tail; - int err = 0; - struct oscarg oa = { 0 }; - - ASSERT(strchr(name, '@') == NULL); - err = dsl_dir_open(name, FTAG, &pdd, &tail); - if (err) - return (err); - if (tail == NULL) { - dsl_dir_close(pdd, FTAG); - return (EEXIST); - } - - oa.lastname = tail; - oa.clone_origin = clone_origin; - oa.flags = flags; - oa.cr = CRED(); - - err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check, - dmu_objset_create_sync, pdd, &oa, 5); - dsl_dir_close(pdd, FTAG); - return (err); -} - -int -dmu_objset_destroy(const char *name, boolean_t defer) -{ - dsl_dataset_t *ds; - int error; - - error = dsl_dataset_own(name, B_TRUE, FTAG, &ds); - if (error == 0) { - error = dsl_dataset_destroy(ds, FTAG, defer); - /* dsl_dataset_destroy() closes the ds. */ - } - - return (error); -} - -typedef struct snapallarg { - dsl_sync_task_group_t *saa_dstg; - boolean_t saa_needsuspend; - nvlist_t *saa_props; - - /* the following are used only if 'temporary' is set: */ - boolean_t saa_temporary; - const char *saa_htag; - struct dsl_ds_holdarg *saa_ha; - dsl_dataset_t *saa_newds; -} snapallarg_t; - -typedef struct snaponearg { - const char *soa_longname; /* long snap name */ - const char *soa_snapname; /* short snap name */ - snapallarg_t *soa_saa; -} snaponearg_t; +typedef struct dmu_objset_clone_arg { + const char *doca_clone; + const char *doca_origin; + cred_t *doca_cred; +} dmu_objset_clone_arg_t; +/*ARGSUSED*/ static int -snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) +dmu_objset_clone_check(void *arg, dmu_tx_t *tx) { - objset_t *os = arg1; - snaponearg_t *soa = arg2; - snapallarg_t *saa = soa->soa_saa; + dmu_objset_clone_arg_t *doca = arg; + dsl_dir_t *pdd; + const char *tail; int error; + dsl_dataset_t *origin; + dsl_pool_t *dp = dmu_tx_pool(tx); - /* The props have already been checked by zfs_check_userprops(). */ + if (strchr(doca->doca_clone, '@') != NULL) + return (EINVAL); - error = dsl_dataset_snapshot_check(os->os_dsl_dataset, - soa->soa_snapname, tx); - if (error) + error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); + if (error != 0) + return (error); + if (tail == NULL) { + dsl_dir_rele(pdd, FTAG); + return (EEXIST); + } + /* You can't clone across pools. */ + if (pdd->dd_pool != dp) { + dsl_dir_rele(pdd, FTAG); + return (EXDEV); + } + dsl_dir_rele(pdd, FTAG); + + error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); + if (error != 0) return (error); - if (saa->saa_temporary) { - /* - * Ideally we would just call - * dsl_dataset_user_hold_check() and - * dsl_dataset_destroy_check() here. However the - * dataset we want to hold and destroy is the snapshot - * that we just confirmed we can create, but it won't - * exist until after these checks are run. Do any - * checks we can here and if more checks are added to - * those routines in the future, similar checks may be - * necessary here. - */ - if (spa_version(os->os_spa) < SPA_VERSION_USERREFS) - return (ENOTSUP); - /* - * Not checking number of tags because the tag will be - * unique, as it will be the only tag. - */ - if (strlen(saa->saa_htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) - return (E2BIG); - - saa->saa_ha = kmem_alloc(sizeof (struct dsl_ds_holdarg), - KM_SLEEP); - saa->saa_ha->temphold = B_TRUE; - saa->saa_ha->htag = saa->saa_htag; - } - return (error); -} - -static void -snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) -{ - objset_t *os = arg1; - dsl_dataset_t *ds = os->os_dsl_dataset; - snaponearg_t *soa = arg2; - snapallarg_t *saa = soa->soa_saa; - - dsl_dataset_snapshot_sync(ds, soa->soa_snapname, tx); - - if (saa->saa_props != NULL) { - dsl_props_arg_t pa; - pa.pa_props = saa->saa_props; - pa.pa_source = ZPROP_SRC_LOCAL; - dsl_props_set_sync(ds->ds_prev, &pa, tx); + /* You can't clone across pools. */ + if (origin->ds_dir->dd_pool != dp) { + dsl_dataset_rele(origin, FTAG); + return (EXDEV); } - if (saa->saa_temporary) { - struct dsl_ds_destroyarg da; - - dsl_dataset_user_hold_sync(ds->ds_prev, saa->saa_ha, tx); - kmem_free(saa->saa_ha, sizeof (struct dsl_ds_holdarg)); - saa->saa_ha = NULL; - saa->saa_newds = ds->ds_prev; - - da.ds = ds->ds_prev; - da.defer = B_TRUE; - dsl_dataset_destroy_sync(&da, FTAG, tx); + /* You can only clone snapshots, not the head datasets. */ + if (!dsl_dataset_is_snapshot(origin)) { + dsl_dataset_rele(origin, FTAG); + return (EINVAL); } -} - -static int -snapshot_one_impl(const char *snapname, void *arg) -{ - char fsname[MAXPATHLEN]; - snapallarg_t *saa = arg; - snaponearg_t *soa; - objset_t *os; - int err; - - (void) strlcpy(fsname, snapname, sizeof (fsname)); - strchr(fsname, '@')[0] = '\0'; - - err = dmu_objset_hold(fsname, saa, &os); - if (err != 0) - return (err); - - /* - * If the objset is in an inconsistent state (eg, in the process - * of being destroyed), don't snapshot it. - */ - if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) { - dmu_objset_rele(os, saa); - return (EBUSY); - } - - if (saa->saa_needsuspend) { - err = zil_suspend(dmu_objset_zil(os)); - if (err) { - dmu_objset_rele(os, saa); - return (err); - } - } - - soa = kmem_zalloc(sizeof (*soa), KM_SLEEP); - soa->soa_saa = saa; - soa->soa_longname = snapname; - soa->soa_snapname = strchr(snapname, '@') + 1; - - dsl_sync_task_create(saa->saa_dstg, snapshot_check, snapshot_sync, - os, soa, 3); + dsl_dataset_rele(origin, FTAG); return (0); } -/* - * The snapshots must all be in the same pool. - */ -int -dmu_objset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) +static void +dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) { - dsl_sync_task_t *dst; - snapallarg_t saa = { 0 }; - spa_t *spa; - int rv = 0; - int err; - nvpair_t *pair; + dmu_objset_clone_arg_t *doca = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dir_t *pdd; + const char *tail; + dsl_dataset_t *origin, *ds; + uint64_t obj; + char namebuf[MAXNAMELEN]; - pair = nvlist_next_nvpair(snaps, NULL); - if (pair == NULL) - return (0); + VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); + VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); - err = spa_open(nvpair_name(pair), &spa, FTAG); - if (err) - return (err); - saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - saa.saa_props = props; - saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); + obj = dsl_dataset_create_sync(pdd, tail, origin, 0, + doca->doca_cred, tx); - for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; - pair = nvlist_next_nvpair(snaps, pair)) { - err = snapshot_one_impl(nvpair_name(pair), &saa); - if (err != 0) { - if (errors != NULL) { - fnvlist_add_int32(errors, - nvpair_name(pair), err); - } - rv = err; - } - } + VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); + dsl_dataset_name(origin, namebuf); + spa_history_log_internal_ds(ds, "clone", tx, + "origin=%s (%llu)", namebuf, origin->ds_object); + dsl_dataset_rele(ds, FTAG); + dsl_dataset_rele(origin, FTAG); + dsl_dir_rele(pdd, FTAG); +} - /* - * If any call to snapshot_one_impl() failed, don't execute the - * sync task. The error handling code below will clean up the - * snaponearg_t from any successful calls to - * snapshot_one_impl(). - */ - if (rv == 0) - err = dsl_sync_task_group_wait(saa.saa_dstg); - if (err != 0) - rv = err; +int +dmu_objset_clone(const char *clone, const char *origin) +{ + dmu_objset_clone_arg_t doca; - for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst; - dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) { - objset_t *os = dst->dst_arg1; - snaponearg_t *soa = dst->dst_arg2; - if (dst->dst_err != 0) { - if (errors != NULL) { - fnvlist_add_int32(errors, - soa->soa_longname, dst->dst_err); - } - rv = dst->dst_err; - } + doca.doca_clone = clone; + doca.doca_origin = origin; + doca.doca_cred = CRED(); - if (saa.saa_needsuspend) - zil_resume(dmu_objset_zil(os)); - dmu_objset_rele(os, &saa); - kmem_free(soa, sizeof (*soa)); - } - - dsl_sync_task_group_destroy(saa.saa_dstg); - spa_close(spa, FTAG); - return (rv); + return (dsl_sync_task(clone, + dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5)); } int @@ -1013,59 +878,12 @@ dmu_objset_snapshot_one(const char *fsname, const char *snapname) nvlist_t *snaps = fnvlist_alloc(); fnvlist_add_boolean(snaps, longsnap); - err = dmu_objset_snapshot(snaps, NULL, NULL); - fnvlist_free(snaps); strfree(longsnap); + err = dsl_dataset_snapshot(snaps, NULL, NULL); + fnvlist_free(snaps); return (err); } -int -dmu_objset_snapshot_tmp(const char *snapname, const char *tag, int cleanup_fd) -{ - dsl_sync_task_t *dst; - snapallarg_t saa = { 0 }; - spa_t *spa; - minor_t minor; - int err; - - err = spa_open(snapname, &spa, FTAG); - if (err) - return (err); - saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - saa.saa_htag = tag; - saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); - saa.saa_temporary = B_TRUE; - - if (cleanup_fd < 0) { - spa_close(spa, FTAG); - return (EINVAL); - } - if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) { - spa_close(spa, FTAG); - return (err); - } - - err = snapshot_one_impl(snapname, &saa); - - if (err == 0) - err = dsl_sync_task_group_wait(saa.saa_dstg); - - for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst; - dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) { - objset_t *os = dst->dst_arg1; - dsl_register_onexit_hold_cleanup(saa.saa_newds, tag, minor); - if (saa.saa_needsuspend) - zil_resume(dmu_objset_zil(os)); - dmu_objset_rele(os, &saa); - } - - zfs_onexit_fd_rele(cleanup_fd); - dsl_sync_task_group_destroy(saa.saa_dstg); - spa_close(spa, FTAG); - return (err); -} - - static void dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx) { @@ -1101,9 +919,9 @@ dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) objset_t *os = arg; dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; - ASSERT(bp == os->os_rootbp); - ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET); - ASSERT(BP_GET_LEVEL(bp) == 0); + ASSERT3P(bp, ==, os->os_rootbp); + ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); + ASSERT0(BP_GET_LEVEL(bp)); /* * Update rootbp fill count: it should be the number of objects @@ -1210,7 +1028,7 @@ dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; while (dr = list_head(list)) { - ASSERT(dr->dr_dbuf->db_level == 0); + ASSERT0(dr->dr_dbuf->db_level); list_remove(list, dr); if (dr->dr_zio) zio_nowait(dr->dr_zio); @@ -1505,12 +1323,12 @@ dmu_objset_userspace_upgrade(objset_t *os) return (EINTR); objerr = dmu_bonus_hold(os, obj, FTAG, &db); - if (objerr) + if (objerr != 0) continue; tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, obj); objerr = dmu_tx_assign(tx, TXG_WAIT); - if (objerr) { + if (objerr != 0) { dmu_tx_abort(tx); continue; } @@ -1593,6 +1411,8 @@ dmu_snapshot_list_next(objset_t *os, int namelen, char *name, zap_cursor_t cursor; zap_attribute_t attr; + ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); + if (ds->ds_phys->ds_snapnames_zapobj == 0) return (ENOENT); @@ -1659,64 +1479,34 @@ dmu_dir_list_next(objset_t *os, int namelen, char *name, return (0); } -struct findarg { - int (*func)(const char *, void *); - void *arg; -}; - -/* ARGSUSED */ -static int -findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) -{ - struct findarg *fa = arg; - return (fa->func(dsname, fa->arg)); -} - /* - * Find all objsets under name, and for each, call 'func(child_name, arg)'. - * Perhaps change all callers to use dmu_objset_find_spa()? + * Find objsets under and including ddobj, call func(ds) on each. */ int -dmu_objset_find(char *name, int func(const char *, void *), void *arg, - int flags) -{ - struct findarg fa; - fa.func = func; - fa.arg = arg; - return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags)); -} - -/* - * Find all objsets under name, call func on each - */ -int -dmu_objset_find_spa(spa_t *spa, const char *name, - int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags) +dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, + int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) { dsl_dir_t *dd; - dsl_pool_t *dp; dsl_dataset_t *ds; zap_cursor_t zc; zap_attribute_t *attr; - char *child; uint64_t thisobj; int err; - if (name == NULL) - name = spa_name(spa); - err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL); - if (err) + ASSERT(dsl_pool_config_held(dp)); + + err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd); + if (err != 0) return (err); /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ if (dd->dd_myname[0] == '$') { - dsl_dir_close(dd, FTAG); + dsl_dir_rele(dd, FTAG); return (0); } thisobj = dd->dd_phys->dd_head_dataset_obj; attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); - dp = dd->dd_pool; /* * Iterate over all children. @@ -1726,19 +1516,19 @@ dmu_objset_find_spa(spa_t *spa, const char *name, dd->dd_phys->dd_child_dir_zapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { - ASSERT(attr->za_integer_length == sizeof (uint64_t)); - ASSERT(attr->za_num_integers == 1); + ASSERT3U(attr->za_integer_length, ==, + sizeof (uint64_t)); + ASSERT3U(attr->za_num_integers, ==, 1); - child = kmem_asprintf("%s/%s", name, attr->za_name); - err = dmu_objset_find_spa(spa, child, func, arg, flags); - strfree(child); - if (err) + err = dmu_objset_find_dp(dp, attr->za_first_integer, + func, arg, flags); + if (err != 0) break; } zap_cursor_fini(&zc); - if (err) { - dsl_dir_close(dd, FTAG); + if (err != 0) { + dsl_dir_rele(dd, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); return (err); } @@ -1748,11 +1538,8 @@ dmu_objset_find_spa(spa_t *spa, const char *name, * Iterate over all snapshots. */ if (flags & DS_FIND_SNAPSHOTS) { - if (!dsl_pool_sync_context(dp)) - rw_enter(&dp->dp_config_rwlock, RW_READER); + dsl_dataset_t *ds; err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); - if (!dsl_pool_sync_context(dp)) - rw_exit(&dp->dp_config_rwlock); if (err == 0) { uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; @@ -1761,64 +1548,166 @@ dmu_objset_find_spa(spa_t *spa, const char *name, for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { - ASSERT(attr->za_integer_length == + ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); - ASSERT(attr->za_num_integers == 1); + ASSERT3U(attr->za_num_integers, ==, 1); - child = kmem_asprintf("%s@%s", - name, attr->za_name); - err = func(spa, attr->za_first_integer, - child, arg); - strfree(child); - if (err) + err = dsl_dataset_hold_obj(dp, + attr->za_first_integer, FTAG, &ds); + if (err != 0) + break; + err = func(dp, ds, arg); + dsl_dataset_rele(ds, FTAG); + if (err != 0) break; } zap_cursor_fini(&zc); } } - dsl_dir_close(dd, FTAG); + dsl_dir_rele(dd, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); - if (err) + if (err != 0) return (err); /* - * Apply to self if appropriate. + * Apply to self. */ - err = func(spa, thisobj, name, arg); + err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); + if (err != 0) + return (err); + err = func(dp, ds, arg); + dsl_dataset_rele(ds, FTAG); return (err); } -/* ARGSUSED */ -int -dmu_objset_prefetch(const char *name, void *arg) +/* + * Find all objsets under name, and for each, call 'func(child_name, arg)'. + * The dp_config_rwlock must not be held when this is called, and it + * will not be held when the callback is called. + * Therefore this function should only be used when the pool is not changing + * (e.g. in syncing context), or the callback can deal with the possible races. + */ +static int +dmu_objset_find_impl(spa_t *spa, const char *name, + int func(const char *, void *), void *arg, int flags) { + dsl_dir_t *dd; + dsl_pool_t *dp = spa_get_dsl(spa); dsl_dataset_t *ds; + zap_cursor_t zc; + zap_attribute_t *attr; + char *child; + uint64_t thisobj; + int err; - if (dsl_dataset_hold(name, FTAG, &ds)) - return (0); + dsl_pool_config_enter(dp, FTAG); - if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) { - mutex_enter(&ds->ds_opening_lock); - if (ds->ds_objset == NULL) { - uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH; - zbookmark_t zb; - - SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, - ZB_ROOT_LEVEL, ZB_ROOT_BLKID); - - (void) arc_read(NULL, dsl_dataset_get_spa(ds), - &ds->ds_phys->ds_bp, NULL, NULL, - ZIO_PRIORITY_ASYNC_READ, - ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, - &aflags, &zb); - } - mutex_exit(&ds->ds_opening_lock); + err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); + if (err != 0) { + dsl_pool_config_exit(dp, FTAG); + return (err); } - dsl_dataset_rele(ds, FTAG); - return (0); + /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ + if (dd->dd_myname[0] == '$') { + dsl_dir_rele(dd, FTAG); + dsl_pool_config_exit(dp, FTAG); + return (0); + } + + thisobj = dd->dd_phys->dd_head_dataset_obj; + attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); + + /* + * Iterate over all children. + */ + if (flags & DS_FIND_CHILDREN) { + for (zap_cursor_init(&zc, dp->dp_meta_objset, + dd->dd_phys->dd_child_dir_zapobj); + zap_cursor_retrieve(&zc, attr) == 0; + (void) zap_cursor_advance(&zc)) { + ASSERT3U(attr->za_integer_length, ==, + sizeof (uint64_t)); + ASSERT3U(attr->za_num_integers, ==, 1); + + child = kmem_asprintf("%s/%s", name, attr->za_name); + dsl_pool_config_exit(dp, FTAG); + err = dmu_objset_find_impl(spa, child, + func, arg, flags); + dsl_pool_config_enter(dp, FTAG); + strfree(child); + if (err != 0) + break; + } + zap_cursor_fini(&zc); + + if (err != 0) { + dsl_dir_rele(dd, FTAG); + dsl_pool_config_exit(dp, FTAG); + kmem_free(attr, sizeof (zap_attribute_t)); + return (err); + } + } + + /* + * Iterate over all snapshots. + */ + if (flags & DS_FIND_SNAPSHOTS) { + err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); + + if (err == 0) { + uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; + dsl_dataset_rele(ds, FTAG); + + for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); + zap_cursor_retrieve(&zc, attr) == 0; + (void) zap_cursor_advance(&zc)) { + ASSERT3U(attr->za_integer_length, ==, + sizeof (uint64_t)); + ASSERT3U(attr->za_num_integers, ==, 1); + + child = kmem_asprintf("%s@%s", + name, attr->za_name); + dsl_pool_config_exit(dp, FTAG); + err = func(child, arg); + dsl_pool_config_enter(dp, FTAG); + strfree(child); + if (err != 0) + break; + } + zap_cursor_fini(&zc); + } + } + + dsl_dir_rele(dd, FTAG); + kmem_free(attr, sizeof (zap_attribute_t)); + dsl_pool_config_exit(dp, FTAG); + + if (err != 0) + return (err); + + /* Apply to self. */ + return (func(name, arg)); +} + +/* + * See comment above dmu_objset_find_impl(). + */ +int +dmu_objset_find(char *name, int func(const char *, void *), void *arg, + int flags) +{ + spa_t *spa; + int error; + + error = spa_open(name, &spa, FTAG); + if (error != 0) + return (error); + error = dmu_objset_find_impl(spa, name, func, arg, flags); + spa_close(spa, FTAG); + return (error); } void @@ -1834,3 +1723,19 @@ dmu_objset_get_user(objset_t *os) ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); return (os->os_user_ptr); } + +/* + * Determine name of filesystem, given name of snapshot. + * buf must be at least MAXNAMELEN bytes + */ +int +dmu_fsname(const char *snapname, char *buf) +{ + char *atp = strchr(snapname, '@'); + if (atp == NULL) + return (EINVAL); + if (atp - snapname >= MAXNAMELEN) + return (ENAMETOOLONG); + (void) strlcpy(buf, snapname, atp - snapname + 1); + return (0); +} diff --git a/uts/common/fs/zfs/dmu_send.c b/uts/common/fs/zfs/dmu_send.c index c2493357d6d7..a2a3647b4b2e 100644 --- a/uts/common/fs/zfs/dmu_send.c +++ b/uts/common/fs/zfs/dmu_send.c @@ -46,11 +46,14 @@ #include #include #include +#include +#include /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ int zfs_send_corrupt_data = B_FALSE; static char *dmu_recv_tag = "dmu_recv_tag"; +static const char *recv_clone_name = "%recv"; static int dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) @@ -290,7 +293,7 @@ dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL)) return (EINTR); - if (dsp->dsa_err) + if (dsp->dsa_err != 0) return (EINTR); return (0); } @@ -340,7 +343,7 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, uint64_t dnobj = (zb->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i; err = dump_dnode(dsp, dnobj, blk+i); - if (err) + if (err != 0) break; } (void) arc_buf_remove_ref(abuf, &abuf); @@ -388,65 +391,33 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, } /* - * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. - * For example, they could both be snapshots of the same filesystem, and - * 'earlier' is before 'later'. Or 'earlier' could be the origin of - * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's - * filesystem. Or 'earlier' could be the origin's origin. + * Releases dp, ds, and fromds, using the specified tag. */ -static boolean_t -is_before(dsl_dataset_t *later, dsl_dataset_t *earlier) +static int +dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, + dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off) { - dsl_pool_t *dp = later->ds_dir->dd_pool; - int error; - boolean_t ret; - dsl_dataset_t *origin; - - if (earlier->ds_phys->ds_creation_txg >= - later->ds_phys->ds_creation_txg) - return (B_FALSE); - - if (later->ds_dir == earlier->ds_dir) - return (B_TRUE); - if (!dsl_dir_is_clone(later->ds_dir)) - return (B_FALSE); - - rw_enter(&dp->dp_config_rwlock, RW_READER); - if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) { - rw_exit(&dp->dp_config_rwlock); - return (B_TRUE); - } - error = dsl_dataset_hold_obj(dp, - later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); - rw_exit(&dp->dp_config_rwlock); - if (error != 0) - return (B_FALSE); - ret = is_before(origin, earlier); - dsl_dataset_rele(origin, FTAG); - return (ret); -} - -int -dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, - offset_t *off) -{ - dsl_dataset_t *ds = tosnap->os_dsl_dataset; - dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL; + objset_t *os; dmu_replay_record_t *drr; dmu_sendarg_t *dsp; int err; uint64_t fromtxg = 0; - /* tosnap must be a snapshot */ - if (ds->ds_phys->ds_next_snap_obj == 0) - return (EINVAL); - - /* - * fromsnap must be an earlier snapshot from the same fs as tosnap, - * or the origin's fs. - */ - if (fromds != NULL && !is_before(ds, fromds)) + if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) { + dsl_dataset_rele(fromds, tag); + dsl_dataset_rele(ds, tag); + dsl_pool_rele(dp, tag); return (EXDEV); + } + + err = dmu_objset_from_ds(ds, &os); + if (err != 0) { + if (fromds != NULL) + dsl_dataset_rele(fromds, tag); + dsl_dataset_rele(ds, tag); + dsl_pool_rele(dp, tag); + return (err); + } drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); drr->drr_type = DRR_BEGIN; @@ -455,13 +426,17 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, DMU_SUBSTREAM); #ifdef _KERNEL - if (dmu_objset_type(tosnap) == DMU_OST_ZFS) { + if (dmu_objset_type(os) == DMU_OST_ZFS) { uint64_t version; - if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) { + if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { kmem_free(drr, sizeof (dmu_replay_record_t)); + if (fromds != NULL) + dsl_dataset_rele(fromds, tag); + dsl_dataset_rele(ds, tag); + dsl_pool_rele(dp, tag); return (EINVAL); } - if (version == ZPL_VERSION_SA) { + if (version >= ZPL_VERSION_SA) { DMU_SET_FEATUREFLAGS( drr->drr_u.drr_begin.drr_versioninfo, DMU_BACKUP_FEATURE_SA_SPILL); @@ -471,19 +446,22 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, drr->drr_u.drr_begin.drr_creation_time = ds->ds_phys->ds_creation_time; - drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type; + drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); if (fromds != NULL && ds->ds_dir != fromds->ds_dir) drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid; if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; - if (fromds) + if (fromds != NULL) drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid; dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname); - if (fromds) + if (fromds != NULL) { fromtxg = fromds->ds_phys->ds_creation_txg; + dsl_dataset_rele(fromds, tag); + fromds = NULL; + } dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); @@ -491,7 +469,7 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, dsp->dsa_vp = vp; dsp->dsa_outfd = outfd; dsp->dsa_proc = curproc; - dsp->dsa_os = tosnap; + dsp->dsa_os = os; dsp->dsa_off = off; dsp->dsa_toguid = ds->ds_phys->ds_guid; ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0); @@ -506,6 +484,9 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, goto out; } + dsl_dataset_long_hold(ds, FTAG); + dsl_pool_rele(dp, tag); + err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH, backup_cb, dsp); @@ -513,8 +494,8 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) err = EINTR; - if (err) { - if (err == EINTR && dsp->dsa_err) + if (err != 0) { + if (err == EINTR && dsp->dsa_err != 0) err = dsp->dsa_err; goto out; } @@ -537,27 +518,96 @@ dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp, kmem_free(drr, sizeof (dmu_replay_record_t)); kmem_free(dsp, sizeof (dmu_sendarg_t)); + dsl_dataset_long_rele(ds, FTAG); + dsl_dataset_rele(ds, tag); + return (err); } int -dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep) +dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, + int outfd, vnode_t *vp, offset_t *off) +{ + dsl_pool_t *dp; + dsl_dataset_t *ds; + dsl_dataset_t *fromds = NULL; + int err; + + err = dsl_pool_hold(pool, FTAG, &dp); + if (err != 0) + return (err); + + err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); + if (err != 0) { + dsl_pool_rele(dp, FTAG); + return (err); + } + + if (fromsnap != 0) { + err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); + if (err != 0) { + dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); + return (err); + } + } + + return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off)); +} + +int +dmu_send(const char *tosnap, const char *fromsnap, + int outfd, vnode_t *vp, offset_t *off) +{ + dsl_pool_t *dp; + dsl_dataset_t *ds; + dsl_dataset_t *fromds = NULL; + int err; + + if (strchr(tosnap, '@') == NULL) + return (EINVAL); + if (fromsnap != NULL && strchr(fromsnap, '@') == NULL) + return (EINVAL); + + err = dsl_pool_hold(tosnap, FTAG, &dp); + if (err != 0) + return (err); + + err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); + if (err != 0) { + dsl_pool_rele(dp, FTAG); + return (err); + } + + if (fromsnap != NULL) { + err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); + if (err != 0) { + dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); + return (err); + } + } + return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off)); +} + +int +dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep) { - dsl_dataset_t *ds = tosnap->os_dsl_dataset; - dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL; dsl_pool_t *dp = ds->ds_dir->dd_pool; int err; uint64_t size; + ASSERT(dsl_pool_config_held(dp)); + /* tosnap must be a snapshot */ - if (ds->ds_phys->ds_next_snap_obj == 0) + if (!dsl_dataset_is_snapshot(ds)) return (EINVAL); /* * fromsnap must be an earlier snapshot from the same fs as tosnap, * or the origin's fs. */ - if (fromds != NULL && !is_before(ds, fromds)) + if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) return (EXDEV); /* Get uncompressed size estimate of changed data. */ @@ -567,7 +617,7 @@ dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep) uint64_t used, comp; err = dsl_dataset_space_written(fromds, ds, &used, &comp, &size); - if (err) + if (err != 0) return (err); } @@ -587,11 +637,8 @@ dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep) * block, which we observe in practice. */ uint64_t recordsize; - rw_enter(&dp->dp_config_rwlock, RW_READER); - err = dsl_prop_get_ds(ds, "recordsize", - sizeof (recordsize), 1, &recordsize, NULL); - rw_exit(&dp->dp_config_rwlock); - if (err) + err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize); + if (err != 0) return (err); size -= size / recordsize * sizeof (blkptr_t); @@ -603,93 +650,40 @@ dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep) return (0); } -struct recvbeginsyncarg { - const char *tofs; - const char *tosnap; - dsl_dataset_t *origin; - uint64_t fromguid; - dmu_objset_type_t type; - void *tag; - boolean_t force; - uint64_t dsflags; - char clonelastname[MAXNAMELEN]; - dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */ - cred_t *cr; -}; +typedef struct dmu_recv_begin_arg { + const char *drba_origin; + dmu_recv_cookie_t *drba_cookie; + cred_t *drba_cred; +} dmu_recv_begin_arg_t; -/* ARGSUSED */ static int -recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx) +recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, + uint64_t fromguid) { - dsl_dir_t *dd = arg1; - struct recvbeginsyncarg *rbsa = arg2; - objset_t *mos = dd->dd_pool->dp_meta_objset; - uint64_t val; - int err; - - err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj, - strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val); - - if (err != ENOENT) - return (err ? err : EEXIST); - - if (rbsa->origin) { - /* make sure it's a snap in the same pool */ - if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool) - return (EXDEV); - if (!dsl_dataset_is_snapshot(rbsa->origin)) - return (EINVAL); - if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid) - return (ENODEV); - } - - return (0); -} - -static void -recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dir_t *dd = arg1; - struct recvbeginsyncarg *rbsa = arg2; - uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; - uint64_t dsobj; - - /* Create and open new dataset. */ - dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1, - rbsa->origin, flags, rbsa->cr, tx); - VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj, - B_TRUE, dmu_recv_tag, &rbsa->ds)); - - if (rbsa->origin == NULL) { - (void) dmu_objset_create_impl(dd->dd_pool->dp_spa, - rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx); - } - - spa_history_log_internal_ds(rbsa->ds, "receive new", tx, ""); -} - -/* ARGSUSED */ -static int -recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dataset_t *ds = arg1; - struct recvbeginsyncarg *rbsa = arg2; - int err; uint64_t val; + int error; + dsl_pool_t *dp = ds->ds_dir->dd_pool; /* must not have any changes since most recent snapshot */ - if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds)) + if (!drba->drba_cookie->drc_force && + dsl_dataset_modified_since_lastsnap(ds)) return (ETXTBSY); - /* new snapshot name must not exist */ - err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, - ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val); - if (err == 0) - return (EEXIST); - if (err != ENOENT) - return (err); + /* temporary clone name must not exist */ + error = zap_lookup(dp->dp_meta_objset, + ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name, + 8, 1, &val); + if (error != ENOENT) + return (error == 0 ? EBUSY : error); - if (rbsa->fromguid) { + /* new snapshot name must not exist */ + error = zap_lookup(dp->dp_meta_objset, + ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap, + 8, 1, &val); + if (error != ENOENT) + return (error == 0 ? EEXIST : error); + + if (fromguid != 0) { /* if incremental, most recent snapshot must match fromguid */ if (ds->ds_prev == NULL) return (ENODEV); @@ -698,20 +692,20 @@ recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx) * most recent snapshot must match fromguid, or there are no * changes since the fromguid one */ - if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) { + if (ds->ds_prev->ds_phys->ds_guid != fromguid) { uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth; uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj; while (obj != 0) { dsl_dataset_t *snap; - err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool, - obj, FTAG, &snap); - if (err) + error = dsl_dataset_hold_obj(dp, obj, FTAG, + &snap); + if (error != 0) return (ENODEV); if (snap->ds_phys->ds_creation_txg < birth) { dsl_dataset_rele(snap, FTAG); return (ENODEV); } - if (snap->ds_phys->ds_guid == rbsa->fromguid) { + if (snap->ds_phys->ds_guid == fromguid) { dsl_dataset_rele(snap, FTAG); break; /* it's ok */ } @@ -727,58 +721,153 @@ recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx) return (ENODEV); } - /* temporary clone name must not exist */ - err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, - ds->ds_dir->dd_phys->dd_child_dir_zapobj, - rbsa->clonelastname, 8, 1, &val); - if (err == 0) - return (EEXIST); - if (err != ENOENT) - return (err); - return (0); + } -/* ARGSUSED */ -static void -recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx) +static int +dmu_recv_begin_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ohds = arg1; - struct recvbeginsyncarg *rbsa = arg2; - dsl_pool_t *dp = ohds->ds_dir->dd_pool; - dsl_dataset_t *cds; - uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags; - uint64_t dsobj; + dmu_recv_begin_arg_t *drba = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + struct drr_begin *drrb = drba->drba_cookie->drc_drrb; + uint64_t fromguid = drrb->drr_fromguid; + int flags = drrb->drr_flags; + int error; + dsl_dataset_t *ds; + const char *tofs = drba->drba_cookie->drc_tofs; - /* create and open the temporary clone */ - dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname, - ohds->ds_prev, flags, rbsa->cr, tx); - VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds)); + /* already checked */ + ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); + + if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == + DMU_COMPOUNDSTREAM || + drrb->drr_type >= DMU_OST_NUMTYPES || + ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) + return (EINVAL); + + /* Verify pool version supports SA if SA_SPILL feature set */ + if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & + DMU_BACKUP_FEATURE_SA_SPILL) && + spa_version(dp->dp_spa) < SPA_VERSION_SA) { + return (ENOTSUP); + } + + error = dsl_dataset_hold(dp, tofs, FTAG, &ds); + if (error == 0) { + /* target fs already exists; recv into temp clone */ + + /* Can't recv a clone into an existing fs */ + if (flags & DRR_FLAG_CLONE) { + dsl_dataset_rele(ds, FTAG); + return (EINVAL); + } + + error = recv_begin_check_existing_impl(drba, ds, fromguid); + dsl_dataset_rele(ds, FTAG); + } else if (error == ENOENT) { + /* target fs does not exist; must be a full backup or clone */ + char buf[MAXNAMELEN]; + + /* + * If it's a non-clone incremental, we are missing the + * target fs, so fail the recv. + */ + if (fromguid != 0 && !(flags & DRR_FLAG_CLONE)) + return (ENOENT); + + /* Open the parent of tofs */ + ASSERT3U(strlen(tofs), <, MAXNAMELEN); + (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); + error = dsl_dataset_hold(dp, buf, FTAG, &ds); + if (error != 0) + return (error); + + if (drba->drba_origin != NULL) { + dsl_dataset_t *origin; + error = dsl_dataset_hold(dp, drba->drba_origin, + FTAG, &origin); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } + if (!dsl_dataset_is_snapshot(origin)) { + dsl_dataset_rele(origin, FTAG); + dsl_dataset_rele(ds, FTAG); + return (EINVAL); + } + if (origin->ds_phys->ds_guid != fromguid) { + dsl_dataset_rele(origin, FTAG); + dsl_dataset_rele(ds, FTAG); + return (ENODEV); + } + dsl_dataset_rele(origin, FTAG); + } + dsl_dataset_rele(ds, FTAG); + error = 0; + } + return (error); +} + +static void +dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) +{ + dmu_recv_begin_arg_t *drba = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + struct drr_begin *drrb = drba->drba_cookie->drc_drrb; + const char *tofs = drba->drba_cookie->drc_tofs; + dsl_dataset_t *ds, *newds; + uint64_t dsobj; + int error; + uint64_t crflags; + + crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ? + DS_FLAG_CI_DATASET : 0; + + error = dsl_dataset_hold(dp, tofs, FTAG, &ds); + if (error == 0) { + /* create temporary clone */ + dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, + ds->ds_prev, crflags, drba->drba_cred, tx); + dsl_dataset_rele(ds, FTAG); + } else { + dsl_dir_t *dd; + const char *tail; + dsl_dataset_t *origin = NULL; + + VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); + + if (drba->drba_origin != NULL) { + VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, + FTAG, &origin)); + } + + /* Create new dataset. */ + dsobj = dsl_dataset_create_sync(dd, + strrchr(tofs, '/') + 1, + origin, crflags, drba->drba_cred, tx); + if (origin != NULL) + dsl_dataset_rele(origin, FTAG); + dsl_dir_rele(dd, FTAG); + drba->drba_cookie->drc_newfs = B_TRUE; + } + VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); + + dmu_buf_will_dirty(newds->ds_dbuf, tx); + newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; /* * If we actually created a non-clone, we need to create the * objset in our new dataset. */ - if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) { + if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { (void) dmu_objset_create_impl(dp->dp_spa, - cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx); + newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); } - rbsa->ds = cds; + drba->drba_cookie->drc_ds = newds; - spa_history_log_internal_ds(cds, "receive over existing", tx, ""); -} - -static boolean_t -dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb) -{ - int featureflags; - - featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); - - /* Verify pool version supports SA if SA_SPILL feature set */ - return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && - (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA)); + spa_history_log_internal_ds(newds, "receive", tx, ""); } /* @@ -786,132 +875,55 @@ dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb) * succeeds; otherwise we will leak the holds on the datasets. */ int -dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb, - boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc) +dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, + boolean_t force, char *origin, dmu_recv_cookie_t *drc) { - int err = 0; - boolean_t byteswap; - struct recvbeginsyncarg rbsa = { 0 }; - uint64_t versioninfo; - int flags; - dsl_dataset_t *ds; - - if (drrb->drr_magic == DMU_BACKUP_MAGIC) - byteswap = FALSE; - else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) - byteswap = TRUE; - else - return (EINVAL); - - rbsa.tofs = tofs; - rbsa.tosnap = tosnap; - rbsa.origin = origin ? origin->os_dsl_dataset : NULL; - rbsa.fromguid = drrb->drr_fromguid; - rbsa.type = drrb->drr_type; - rbsa.tag = FTAG; - rbsa.dsflags = 0; - rbsa.cr = CRED(); - versioninfo = drrb->drr_versioninfo; - flags = drrb->drr_flags; - - if (byteswap) { - rbsa.type = BSWAP_32(rbsa.type); - rbsa.fromguid = BSWAP_64(rbsa.fromguid); - versioninfo = BSWAP_64(versioninfo); - flags = BSWAP_32(flags); - } - - if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM || - rbsa.type >= DMU_OST_NUMTYPES || - ((flags & DRR_FLAG_CLONE) && origin == NULL)) - return (EINVAL); - - if (flags & DRR_FLAG_CI_DATA) - rbsa.dsflags = DS_FLAG_CI_DATASET; + dmu_recv_begin_arg_t drba = { 0 }; + dmu_replay_record_t *drr; bzero(drc, sizeof (dmu_recv_cookie_t)); drc->drc_drrb = drrb; drc->drc_tosnap = tosnap; - drc->drc_top_ds = top_ds; + drc->drc_tofs = tofs; drc->drc_force = force; - /* - * Process the begin in syncing context. - */ + if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) + drc->drc_byteswap = B_TRUE; + else if (drrb->drr_magic != DMU_BACKUP_MAGIC) + return (EINVAL); - /* open the dataset we are logically receiving into */ - err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds); - if (err == 0) { - if (dmu_recv_verify_features(ds, drrb)) { - dsl_dataset_rele(ds, dmu_recv_tag); - return (ENOTSUP); - } - /* target fs already exists; recv into temp clone */ + drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); + drr->drr_type = DRR_BEGIN; + drr->drr_u.drr_begin = *drc->drc_drrb; + if (drc->drc_byteswap) { + fletcher_4_incremental_byteswap(drr, + sizeof (dmu_replay_record_t), &drc->drc_cksum); + } else { + fletcher_4_incremental_native(drr, + sizeof (dmu_replay_record_t), &drc->drc_cksum); + } + kmem_free(drr, sizeof (dmu_replay_record_t)); - /* Can't recv a clone into an existing fs */ - if (flags & DRR_FLAG_CLONE) { - dsl_dataset_rele(ds, dmu_recv_tag); - return (EINVAL); - } - - /* must not have an incremental recv already in progress */ - if (!mutex_tryenter(&ds->ds_recvlock)) { - dsl_dataset_rele(ds, dmu_recv_tag); - return (EBUSY); - } - - /* tmp clone name is: tofs/%tosnap" */ - (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname), - "%%%s", tosnap); - rbsa.force = force; - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - recv_existing_check, recv_existing_sync, ds, &rbsa, 5); - if (err) { - mutex_exit(&ds->ds_recvlock); - dsl_dataset_rele(ds, dmu_recv_tag); - return (err); - } - drc->drc_logical_ds = ds; - drc->drc_real_ds = rbsa.ds; - } else if (err == ENOENT) { - /* target fs does not exist; must be a full backup or clone */ - char *cp; - - /* - * If it's a non-clone incremental, we are missing the - * target fs, so fail the recv. - */ - if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) - return (ENOENT); - - /* Open the parent of tofs */ - cp = strrchr(tofs, '/'); - *cp = '\0'; - err = dsl_dataset_hold(tofs, FTAG, &ds); - *cp = '/'; - if (err) - return (err); - - if (dmu_recv_verify_features(ds, drrb)) { - dsl_dataset_rele(ds, FTAG); - return (ENOTSUP); - } - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5); - dsl_dataset_rele(ds, FTAG); - if (err) - return (err); - drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds; - drc->drc_newfs = B_TRUE; + if (drc->drc_byteswap) { + drrb->drr_magic = BSWAP_64(drrb->drr_magic); + drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); + drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); + drrb->drr_type = BSWAP_32(drrb->drr_type); + drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); + drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); } - return (err); + drba.drba_origin = origin; + drba.drba_cookie = drc; + drba.drba_cred = CRED(); + + return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync, + &drba, 5)); } struct restorearg { int err; - int byteswap; + boolean_t byteswap; vnode_t *vp; char *buf; uint64_t voff; @@ -947,7 +959,7 @@ free_guid_map_onexit(void *arg) guid_map_entry_t *gmep; while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { - dsl_dataset_rele(gmep->gme_ds, ca); + dsl_dataset_long_rele(gmep->gme_ds, gmep); kmem_free(gmep, sizeof (guid_map_entry_t)); } avl_destroy(ca); @@ -975,7 +987,7 @@ restore_read(struct restorearg *ra, int len) ra->err = EINVAL; ra->voff += len - done - resid; done = len - resid; - if (ra->err) + if (ra->err != 0) return (NULL); } @@ -1094,7 +1106,7 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) if (drro->drr_bonuslen) { data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8)); - if (ra->err) + if (ra->err != 0) return (ra->err); } @@ -1103,7 +1115,7 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); err = dmu_tx_assign(tx, TXG_WAIT); - if (err) { + if (err != 0) { dmu_tx_abort(tx); return (err); } @@ -1117,14 +1129,14 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) drro->drr_type, drro->drr_blksz, drro->drr_bonustype, drro->drr_bonuslen); } - if (err) { + if (err != 0) { return (EINVAL); } tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, drro->drr_object); err = dmu_tx_assign(tx, TXG_WAIT); - if (err) { + if (err != 0) { dmu_tx_abort(tx); return (err); } @@ -1172,7 +1184,7 @@ restore_freeobjects(struct restorearg *ra, objset_t *os, continue; err = dmu_free_object(os, obj); - if (err) + if (err != 0) return (err); } return (0); @@ -1202,7 +1214,7 @@ restore_write(struct restorearg *ra, objset_t *os, dmu_tx_hold_write(tx, drrw->drr_object, drrw->drr_offset, drrw->drr_length); err = dmu_tx_assign(tx, TXG_WAIT); - if (err) { + if (err != 0) { dmu_tx_abort(tx); return (err); } @@ -1264,7 +1276,7 @@ restore_write_byref(struct restorearg *ra, objset_t *os, dmu_tx_hold_write(tx, drrwbr->drr_object, drrwbr->drr_offset, drrwbr->drr_length); err = dmu_tx_assign(tx, TXG_WAIT); - if (err) { + if (err != 0) { dmu_tx_abort(tx); return (err); } @@ -1305,7 +1317,7 @@ restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs) dmu_tx_hold_spill(tx, db->db_object); err = dmu_tx_assign(tx, TXG_WAIT); - if (err) { + if (err != 0) { dmu_buf_rele(db, FTAG); dmu_buf_rele(db_spill, FTAG); dmu_tx_abort(tx); @@ -1344,6 +1356,16 @@ restore_free(struct restorearg *ra, objset_t *os, return (err); } +/* used to destroy the drc_ds on error */ +static void +dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) +{ + char name[MAXNAMELEN]; + dsl_dataset_name(drc->drc_ds, name); + dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); + (void) dsl_destroy_head(name); +} + /* * NB: callers *must* call dmu_recv_end() if this succeeds. */ @@ -1357,52 +1379,24 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, zio_cksum_t pcksum; int featureflags; - if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) - ra.byteswap = TRUE; - - { - /* compute checksum of drr_begin record */ - dmu_replay_record_t *drr; - drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); - - drr->drr_type = DRR_BEGIN; - drr->drr_u.drr_begin = *drc->drc_drrb; - if (ra.byteswap) { - fletcher_4_incremental_byteswap(drr, - sizeof (dmu_replay_record_t), &ra.cksum); - } else { - fletcher_4_incremental_native(drr, - sizeof (dmu_replay_record_t), &ra.cksum); - } - kmem_free(drr, sizeof (dmu_replay_record_t)); - } - - if (ra.byteswap) { - struct drr_begin *drrb = drc->drc_drrb; - drrb->drr_magic = BSWAP_64(drrb->drr_magic); - drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); - drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); - drrb->drr_type = BSWAP_32(drrb->drr_type); - drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); - drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); - } - + ra.byteswap = drc->drc_byteswap; + ra.cksum = drc->drc_cksum; ra.vp = vp; ra.voff = *voffp; ra.bufsize = 1<<20; ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); /* these were verified in dmu_recv_begin */ - ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) == + ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, DMU_SUBSTREAM); - ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES); + ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); /* * Open the objset we are modifying. */ - VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0); + VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os)); - ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT); + ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT); featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); @@ -1415,7 +1409,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, goto out; } ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); - if (ra.err) { + if (ra.err != 0) { cleanup_fd = -1; goto out; } @@ -1429,12 +1423,12 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, ra.err = zfs_onexit_add_cb(minor, free_guid_map_onexit, ra.guid_to_ds_map, action_handlep); - if (ra.err) + if (ra.err != 0) goto out; } else { ra.err = zfs_onexit_cb_data(minor, *action_handlep, (void **)&ra.guid_to_ds_map); - if (ra.err) + if (ra.err != 0) goto out; } @@ -1528,14 +1522,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, * destroy what we created, so we don't leave it in the * inconsistent restoring state. */ - txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0); - - (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, - B_FALSE); - if (drc->drc_real_ds != drc->drc_logical_ds) { - mutex_exit(&drc->drc_logical_ds->ds_recvlock); - dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag); - } + dmu_recv_cleanup_ds(drc); } kmem_free(ra.buf, ra.bufsize); @@ -1543,142 +1530,176 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, return (ra.err); } -struct recvendsyncarg { - char *tosnap; - uint64_t creation_time; - uint64_t toguid; -}; - static int -recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx) +dmu_recv_end_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - struct recvendsyncarg *resa = arg2; + dmu_recv_cookie_t *drc = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + int error; - return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx)); + ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); + + if (!drc->drc_newfs) { + dsl_dataset_t *origin_head; + + error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); + if (error != 0) + return (error); + error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, + origin_head, drc->drc_force); + if (error != 0) { + dsl_dataset_rele(origin_head, FTAG); + return (error); + } + error = dsl_dataset_snapshot_check_impl(origin_head, + drc->drc_tosnap, tx); + dsl_dataset_rele(origin_head, FTAG); + if (error != 0) + return (error); + + error = dsl_destroy_head_check_impl(drc->drc_ds, 1); + } else { + error = dsl_dataset_snapshot_check_impl(drc->drc_ds, + drc->drc_tosnap, tx); + } + return (error); } static void -recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dmu_recv_end_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - struct recvendsyncarg *resa = arg2; + dmu_recv_cookie_t *drc = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); - dsl_dataset_snapshot_sync(ds, resa->tosnap, tx); + spa_history_log_internal_ds(drc->drc_ds, "finish receiving", + tx, "snap=%s", drc->drc_tosnap); - /* set snapshot's creation time and guid */ - dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); - ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time; - ds->ds_prev->ds_phys->ds_guid = resa->toguid; - ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; + if (!drc->drc_newfs) { + dsl_dataset_t *origin_head; - dmu_buf_will_dirty(ds->ds_dbuf, tx); - ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; - spa_history_log_internal_ds(ds, "finished receiving", tx, ""); + VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, + &origin_head)); + dsl_dataset_clone_swap_sync_impl(drc->drc_ds, + origin_head, tx); + dsl_dataset_snapshot_sync_impl(origin_head, + drc->drc_tosnap, tx); + + /* set snapshot's creation time and guid */ + dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); + origin_head->ds_prev->ds_phys->ds_creation_time = + drc->drc_drrb->drr_creation_time; + origin_head->ds_prev->ds_phys->ds_guid = + drc->drc_drrb->drr_toguid; + origin_head->ds_prev->ds_phys->ds_flags &= + ~DS_FLAG_INCONSISTENT; + + dmu_buf_will_dirty(origin_head->ds_dbuf, tx); + origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; + + dsl_dataset_rele(origin_head, FTAG); + dsl_destroy_head_sync_impl(drc->drc_ds, tx); + } else { + dsl_dataset_t *ds = drc->drc_ds; + + dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); + + /* set snapshot's creation time and guid */ + dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); + ds->ds_prev->ds_phys->ds_creation_time = + drc->drc_drrb->drr_creation_time; + ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid; + ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; + + dmu_buf_will_dirty(ds->ds_dbuf, tx); + ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT; + } + drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj; + /* + * Release the hold from dmu_recv_begin. This must be done before + * we return to open context, so that when we free the dataset's dnode, + * we can evict its bonus buffer. + */ + dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); + drc->drc_ds = NULL; } static int -add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds) +add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) { - dsl_pool_t *dp = ds->ds_dir->dd_pool; - uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj; + dsl_pool_t *dp; dsl_dataset_t *snapds; guid_map_entry_t *gmep; int err; ASSERT(guid_map != NULL); - rw_enter(&dp->dp_config_rwlock, RW_READER); - err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds); + err = dsl_pool_hold(name, FTAG, &dp); + if (err != 0) + return (err); + err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snapds); if (err == 0) { gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP); gmep->guid = snapds->ds_phys->ds_guid; gmep->gme_ds = snapds; avl_add(guid_map, gmep); + dsl_dataset_long_hold(snapds, gmep); + dsl_dataset_rele(snapds, FTAG); } - rw_exit(&dp->dp_config_rwlock); + dsl_pool_rele(dp, FTAG); return (err); } +static int dmu_recv_end_modified_blocks = 3; + static int dmu_recv_existing_end(dmu_recv_cookie_t *drc) { - struct recvendsyncarg resa; - dsl_dataset_t *ds = drc->drc_logical_ds; - int err, myerr; + int error; + char name[MAXNAMELEN]; - if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) { - err = dsl_dataset_clone_swap(drc->drc_real_ds, ds, - drc->drc_force); - if (err) - goto out; - } else { - mutex_exit(&ds->ds_recvlock); - dsl_dataset_rele(ds, dmu_recv_tag); - (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, - B_FALSE); - return (EBUSY); - } +#ifdef _KERNEL + /* + * We will be destroying the ds; make sure its origin is unmounted if + * necessary. + */ + dsl_dataset_name(drc->drc_ds, name); + zfs_destroy_unmount_origin(name); +#endif - resa.creation_time = drc->drc_drrb->drr_creation_time; - resa.toguid = drc->drc_drrb->drr_toguid; - resa.tosnap = drc->drc_tosnap; + error = dsl_sync_task(drc->drc_tofs, + dmu_recv_end_check, dmu_recv_end_sync, drc, + dmu_recv_end_modified_blocks); - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - recv_end_check, recv_end_sync, ds, &resa, 3); - if (err) { - /* swap back */ - (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE); - } - -out: - mutex_exit(&ds->ds_recvlock); - if (err == 0 && drc->drc_guid_to_ds_map != NULL) - (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds); - dsl_dataset_disown(ds, dmu_recv_tag); - myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE); - ASSERT0(myerr); - return (err); + if (error != 0) + dmu_recv_cleanup_ds(drc); + return (error); } static int dmu_recv_new_end(dmu_recv_cookie_t *drc) { - struct recvendsyncarg resa; - dsl_dataset_t *ds = drc->drc_logical_ds; - int err; + int error; - /* - * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean() - * expects it to have a ds_user_ptr (and zil), but clone_swap() - * can close it. - */ - txg_wait_synced(ds->ds_dir->dd_pool, 0); + error = dsl_sync_task(drc->drc_tofs, + dmu_recv_end_check, dmu_recv_end_sync, drc, + dmu_recv_end_modified_blocks); - resa.creation_time = drc->drc_drrb->drr_creation_time; - resa.toguid = drc->drc_drrb->drr_toguid; - resa.tosnap = drc->drc_tosnap; - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - recv_end_check, recv_end_sync, ds, &resa, 3); - if (err) { - /* clean up the fs we just recv'd into */ - (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE); - } else { - if (drc->drc_guid_to_ds_map != NULL) - (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds); - /* release the hold from dmu_recv_begin */ - dsl_dataset_disown(ds, dmu_recv_tag); + if (error != 0) { + dmu_recv_cleanup_ds(drc); + } else if (drc->drc_guid_to_ds_map != NULL) { + (void) add_ds_to_guidmap(drc->drc_tofs, + drc->drc_guid_to_ds_map, + drc->drc_newsnapobj); } - return (err); + return (error); } int dmu_recv_end(dmu_recv_cookie_t *drc) { - if (drc->drc_logical_ds != drc->drc_real_ds) - return (dmu_recv_existing_end(drc)); - else + if (drc->drc_newfs) return (dmu_recv_new_end(drc)); + else + return (dmu_recv_existing_end(drc)); } diff --git a/uts/common/fs/zfs/dmu_traverse.c b/uts/common/fs/zfs/dmu_traverse.c index f3d5069d476c..e9765179972a 100644 --- a/uts/common/fs/zfs/dmu_traverse.c +++ b/uts/common/fs/zfs/dmu_traverse.c @@ -265,7 +265,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); - if (err) + if (err != 0) return (err); cbp = buf->b_data; @@ -282,7 +282,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, zb->zb_level - 1, zb->zb_blkid * epb + i); err = traverse_visitbp(td, dnp, &cbp[i], &czb); - if (err) { + if (err != 0) { if (!hard) break; lasterr = err; @@ -295,7 +295,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); - if (err) + if (err != 0) return (err); dnp = buf->b_data; @@ -308,7 +308,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, for (i = 0; i < epb; i++) { err = traverse_dnode(td, &dnp[i], zb->zb_objset, zb->zb_blkid * epb + i); - if (err) { + if (err != 0) { if (!hard) break; lasterr = err; @@ -321,7 +321,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); - if (err) + if (err != 0) return (err); osp = buf->b_data; @@ -405,7 +405,7 @@ traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, for (j = 0; j < dnp->dn_nblkptr; j++) { SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb); - if (err) { + if (err != 0) { if (!hard) break; lasterr = err; @@ -415,7 +415,7 @@ traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb); - if (err) { + if (err != 0) { if (!hard) return (err); lasterr = err; @@ -514,14 +514,20 @@ traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); /* See comment on ZIL traversal in dsl_scan_visitds. */ - if (ds != NULL && !dsl_dataset_is_snapshot(ds)) { - objset_t *os; + if (ds != NULL && !dsl_dataset_is_snapshot(ds) && !BP_IS_HOLE(rootbp)) { + uint32_t flags = ARC_WAIT; + objset_phys_t *osp; + arc_buf_t *buf; - err = dmu_objset_from_ds(ds, &os); - if (err) + err = arc_read(NULL, td.td_spa, rootbp, + arc_getbuf_func, &buf, + ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL); + if (err != 0) return (err); - traverse_zil(&td, &os->os_zil_header); + osp = buf->b_data; + traverse_zil(&td, &osp->os_zil_header); + (void) arc_buf_remove_ref(buf, &buf); } if (!(flags & TRAVERSE_PREFETCH_DATA) || @@ -583,7 +589,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags, /* visit the MOS */ err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), txg_start, NULL, flags, func, arg); - if (err) + if (err != 0) return (err); /* visit each dataset */ @@ -592,7 +598,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags, dmu_object_info_t doi; err = dmu_object_info(mos, obj, &doi); - if (err) { + if (err != 0) { if (!hard) return (err); lasterr = err; @@ -603,10 +609,10 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags, dsl_dataset_t *ds; uint64_t txg = txg_start; - rw_enter(&dp->dp_config_rwlock, RW_READER); + dsl_pool_config_enter(dp, FTAG); err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - if (err) { + dsl_pool_config_exit(dp, FTAG); + if (err != 0) { if (!hard) return (err); lasterr = err; @@ -616,7 +622,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags, txg = ds->ds_phys->ds_prev_snap_txg; err = traverse_dataset(ds, txg, flags, func, arg); dsl_dataset_rele(ds, FTAG); - if (err) { + if (err != 0) { if (!hard) return (err); lasterr = err; diff --git a/uts/common/fs/zfs/dmu_tx.c b/uts/common/fs/zfs/dmu_tx.c index 556ae6a83a53..8e6beec4f1b9 100644 --- a/uts/common/fs/zfs/dmu_tx.c +++ b/uts/common/fs/zfs/dmu_tx.c @@ -898,7 +898,7 @@ dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db) #endif static int -dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how) +dmu_tx_try_assign(dmu_tx_t *tx, txg_how_t txg_how) { dmu_tx_hold_t *txh; spa_t *spa = tx->tx_pool->dp_spa; @@ -961,13 +961,6 @@ dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how) fudge += txh->txh_fudge; } - /* - * NB: This check must be after we've held the dnodes, so that - * the dmu_tx_unassign() logic will work properly - */ - if (txg_how >= TXG_INITIAL && txg_how != tx->tx_txg) - return (ERESTART); - /* * If a snapshot has been taken since we made our estimates, * assume that we won't be able to free or overwrite anything. @@ -1048,26 +1041,25 @@ dmu_tx_unassign(dmu_tx_t *tx) * * (1) TXG_WAIT. If the current open txg is full, waits until there's * a new one. This should be used when you're not holding locks. - * If will only fail if we're truly out of space (or over quota). + * It will only fail if we're truly out of space (or over quota). * * (2) TXG_NOWAIT. If we can't assign into the current open txg without * blocking, returns immediately with ERESTART. This should be used * whenever you're holding locks. On an ERESTART error, the caller * should drop locks, do a dmu_tx_wait(tx), and try again. - * - * (3) A specific txg. Use this if you need to ensure that multiple - * transactions all sync in the same txg. Like TXG_NOWAIT, it - * returns ERESTART if it can't assign you into the requested txg. */ int -dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how) +dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how) { int err; ASSERT(tx->tx_txg == 0); - ASSERT(txg_how != 0); + ASSERT(txg_how == TXG_WAIT || txg_how == TXG_NOWAIT); ASSERT(!dsl_pool_sync_context(tx->tx_pool)); + /* If we might wait, we must not hold the config lock. */ + ASSERT(txg_how != TXG_WAIT || !dsl_pool_config_held(tx->tx_pool)); + while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) { dmu_tx_unassign(tx); @@ -1088,6 +1080,7 @@ dmu_tx_wait(dmu_tx_t *tx) spa_t *spa = tx->tx_pool->dp_spa; ASSERT(tx->tx_txg == 0); + ASSERT(!dsl_pool_config_held(tx->tx_pool)); /* * It's possible that the pool has become active after this thread @@ -1214,6 +1207,14 @@ dmu_tx_get_txg(dmu_tx_t *tx) return (tx->tx_txg); } +dsl_pool_t * +dmu_tx_pool(dmu_tx_t *tx) +{ + ASSERT(tx->tx_pool != NULL); + return (tx->tx_pool); +} + + void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data) { diff --git a/uts/common/fs/zfs/dnode.c b/uts/common/fs/zfs/dnode.c index 6838576dcf09..5b5ece424a44 100644 --- a/uts/common/fs/zfs/dnode.c +++ b/uts/common/fs/zfs/dnode.c @@ -72,7 +72,11 @@ dnode_cons(void *arg, void *unused, int kmflag) mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); - refcount_create(&dn->dn_holds); + /* + * Every dbuf has a reference, and dropping a tracked reference is + * O(number of references), so don't track dn_holds. + */ + refcount_create_untracked(&dn->dn_holds); refcount_create(&dn->dn_tx_holds); list_link_init(&dn->dn_link); diff --git a/uts/common/fs/zfs/dnode_sync.c b/uts/common/fs/zfs/dnode_sync.c index 38dab665fc63..7d47ce02b41d 100644 --- a/uts/common/fs/zfs/dnode_sync.c +++ b/uts/common/fs/zfs/dnode_sync.c @@ -477,6 +477,7 @@ dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); dnode_evict_dbufs(dn); ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); + ASSERT3P(dn->dn_bonus, ==, NULL); /* * XXX - It would be nice to assert this, but we may still diff --git a/uts/common/fs/zfs/dsl_dataset.c b/uts/common/fs/zfs/dsl_dataset.c index 0a5ef837ccfd..5e0446d35137 100644 --- a/uts/common/fs/zfs/dsl_dataset.c +++ b/uts/common/fs/zfs/dsl_dataset.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012 by Delphix. All rights reserved. + * Copyright (c) 2013 by Delphix. All rights reserved. * Copyright (c) 2012, Joyent, Inc. All rights reserved. */ @@ -45,12 +45,8 @@ #include #include #include - -static char *dsl_reaper = "the grim reaper"; - -static dsl_checkfunc_t dsl_dataset_destroy_begin_check; -static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; -static dsl_syncfunc_t dsl_dataset_set_reservation_sync; +#include +#include #define SWITCH64(x, y) \ { \ @@ -63,9 +59,6 @@ static dsl_syncfunc_t dsl_dataset_set_reservation_sync; #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE -#define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) - - /* * Figure out how much of this delta should be propogated to the dsl_dir * layer. If there's a refreservation, that space has already been @@ -252,7 +245,7 @@ dsl_dataset_evict(dmu_buf_t *db, void *dsv) { dsl_dataset_t *ds = dsv; - ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); + ASSERT(ds->ds_owner == NULL); unique_remove(ds->ds_fsid_guid); @@ -260,32 +253,26 @@ dsl_dataset_evict(dmu_buf_t *db, void *dsv) dmu_objset_evict(ds->ds_objset); if (ds->ds_prev) { - dsl_dataset_drop_ref(ds->ds_prev, ds); + dsl_dataset_rele(ds->ds_prev, ds); ds->ds_prev = NULL; } bplist_destroy(&ds->ds_pending_deadlist); - if (db != NULL) { + if (ds->ds_phys->ds_deadlist_obj != 0) dsl_deadlist_close(&ds->ds_deadlist); - } else { - ASSERT(ds->ds_deadlist.dl_dbuf == NULL); - ASSERT(!ds->ds_deadlist.dl_oldfmt); - } if (ds->ds_dir) - dsl_dir_close(ds->ds_dir, ds); + dsl_dir_rele(ds->ds_dir, ds); ASSERT(!list_link_active(&ds->ds_synced_link)); mutex_destroy(&ds->ds_lock); - mutex_destroy(&ds->ds_recvlock); mutex_destroy(&ds->ds_opening_lock); - rw_destroy(&ds->ds_rwlock); - cv_destroy(&ds->ds_exclusive_cv); + refcount_destroy(&ds->ds_longholds); kmem_free(ds, sizeof (dsl_dataset_t)); } -static int +int dsl_dataset_get_snapname(dsl_dataset_t *ds) { dsl_dataset_phys_t *headphys; @@ -301,7 +288,7 @@ dsl_dataset_get_snapname(dsl_dataset_t *ds) err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &headdbuf); - if (err) + if (err != 0) return (err); headphys = headdbuf->db_data; err = zap_value_search(dp->dp_meta_objset, @@ -310,7 +297,7 @@ dsl_dataset_get_snapname(dsl_dataset_t *ds) return (err); } -static int +int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; @@ -330,8 +317,8 @@ dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) return (err); } -static int -dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) +int +dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; @@ -351,8 +338,8 @@ dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) return (err); } -static int -dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, +int +dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, dsl_dataset_t **dsp) { objset_t *mos = dp->dp_meta_objset; @@ -361,11 +348,10 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, int err; dmu_object_info_t doi; - ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || - dsl_pool_sync_context(dp)); + ASSERT(dsl_pool_config_held(dp)); err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); - if (err) + if (err != 0) return (err); /* Make sure dsobj has the correct object type. */ @@ -383,12 +369,9 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, ds->ds_phys = dbuf->db_data; mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); - mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); - - rw_init(&ds->ds_rwlock, 0, 0, 0); - cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); + refcount_create(&ds->ds_longholds); bplist_create(&ds->ds_pending_deadlist); dsl_deadlist_open(&ds->ds_deadlist, @@ -398,15 +381,13 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, offsetof(dmu_sendarg_t, dsa_link)); if (err == 0) { - err = dsl_dir_open_obj(dp, + err = dsl_dir_hold_obj(dp, ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); } - if (err) { + if (err != 0) { mutex_destroy(&ds->ds_lock); - mutex_destroy(&ds->ds_recvlock); mutex_destroy(&ds->ds_opening_lock); - rw_destroy(&ds->ds_rwlock); - cv_destroy(&ds->ds_exclusive_cv); + refcount_destroy(&ds->ds_longholds); bplist_destroy(&ds->ds_pending_deadlist); dsl_deadlist_close(&ds->ds_deadlist); kmem_free(ds, sizeof (dsl_dataset_t)); @@ -416,8 +397,8 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, if (!dsl_dataset_is_snapshot(ds)) { ds->ds_snapname[0] = '\0'; - if (ds->ds_phys->ds_prev_snap_obj) { - err = dsl_dataset_get_ref(dp, + if (ds->ds_phys->ds_prev_snap_obj != 0) { + err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev); } @@ -433,29 +414,14 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, } if (err == 0 && !dsl_dataset_is_snapshot(ds)) { - /* - * In sync context, we're called with either no lock - * or with the write lock. If we're not syncing, - * we're always called with the read lock held. - */ - boolean_t need_lock = - !RW_WRITE_HELD(&dp->dp_config_rwlock) && - dsl_pool_sync_context(dp); - - if (need_lock) - rw_enter(&dp->dp_config_rwlock, RW_READER); - - err = dsl_prop_get_ds(ds, - "refreservation", sizeof (uint64_t), 1, - &ds->ds_reserved, NULL); + err = dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_REFRESERVATION), + &ds->ds_reserved); if (err == 0) { - err = dsl_prop_get_ds(ds, - "refquota", sizeof (uint64_t), 1, - &ds->ds_quota, NULL); + err = dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_REFQUOTA), + &ds->ds_quota); } - - if (need_lock) - rw_exit(&dp->dp_config_rwlock); } else { ds->ds_reserved = ds->ds_quota = 0; } @@ -465,15 +431,13 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, bplist_destroy(&ds->ds_pending_deadlist); dsl_deadlist_close(&ds->ds_deadlist); if (ds->ds_prev) - dsl_dataset_drop_ref(ds->ds_prev, ds); - dsl_dir_close(ds->ds_dir, ds); + dsl_dataset_rele(ds->ds_prev, ds); + dsl_dir_rele(ds->ds_dir, ds); mutex_destroy(&ds->ds_lock); - mutex_destroy(&ds->ds_recvlock); mutex_destroy(&ds->ds_opening_lock); - rw_destroy(&ds->ds_rwlock); - cv_destroy(&ds->ds_exclusive_cv); + refcount_destroy(&ds->ds_longholds); kmem_free(ds, sizeof (dsl_dataset_t)); - if (err) { + if (err != 0) { dmu_buf_rele(dbuf, tag); return (err); } @@ -488,90 +452,68 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); - mutex_enter(&ds->ds_lock); - if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { - mutex_exit(&ds->ds_lock); - dmu_buf_rele(ds->ds_dbuf, tag); - return (ENOENT); - } - mutex_exit(&ds->ds_lock); *dsp = ds; return (0); } -static int -dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) +int +dsl_dataset_hold(dsl_pool_t *dp, const char *name, + void *tag, dsl_dataset_t **dsp) { - dsl_pool_t *dp = ds->ds_dir->dd_pool; + dsl_dir_t *dd; + const char *snapname; + uint64_t obj; + int err = 0; - /* - * In syncing context we don't want the rwlock lock: there - * may be an existing writer waiting for sync phase to - * finish. We don't need to worry about such writers, since - * sync phase is single-threaded, so the writer can't be - * doing anything while we are active. - */ - if (dsl_pool_sync_context(dp)) { - ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); - return (0); - } + err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); + if (err != 0) + return (err); - /* - * Normal users will hold the ds_rwlock as a READER until they - * are finished (i.e., call dsl_dataset_rele()). "Owners" will - * drop their READER lock after they set the ds_owner field. - * - * If the dataset is being destroyed, the destroy thread will - * obtain a WRITER lock for exclusive access after it's done its - * open-context work and then change the ds_owner to - * dsl_reaper once destruction is assured. So threads - * may block here temporarily, until the "destructability" of - * the dataset is determined. - */ - ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); - mutex_enter(&ds->ds_lock); - while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { - rw_exit(&dp->dp_config_rwlock); - cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); - if (DSL_DATASET_IS_DESTROYED(ds)) { - mutex_exit(&ds->ds_lock); - dsl_dataset_drop_ref(ds, tag); - rw_enter(&dp->dp_config_rwlock, RW_READER); + ASSERT(dsl_pool_config_held(dp)); + obj = dd->dd_phys->dd_head_dataset_obj; + if (obj != 0) + err = dsl_dataset_hold_obj(dp, obj, tag, dsp); + else + err = ENOENT; + + /* we may be looking for a snapshot */ + if (err == 0 && snapname != NULL) { + dsl_dataset_t *ds; + + if (*snapname++ != '@') { + dsl_dataset_rele(*dsp, tag); + dsl_dir_rele(dd, FTAG); return (ENOENT); } - /* - * The dp_config_rwlock lives above the ds_lock. And - * we need to check DSL_DATASET_IS_DESTROYED() while - * holding the ds_lock, so we have to drop and reacquire - * the ds_lock here. - */ - mutex_exit(&ds->ds_lock); - rw_enter(&dp->dp_config_rwlock, RW_READER); - mutex_enter(&ds->ds_lock); + + dprintf("looking for snapshot '%s'\n", snapname); + err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); + if (err == 0) + err = dsl_dataset_hold_obj(dp, obj, tag, &ds); + dsl_dataset_rele(*dsp, tag); + + if (err == 0) { + mutex_enter(&ds->ds_lock); + if (ds->ds_snapname[0] == 0) + (void) strlcpy(ds->ds_snapname, snapname, + sizeof (ds->ds_snapname)); + mutex_exit(&ds->ds_lock); + *dsp = ds; + } } - mutex_exit(&ds->ds_lock); - return (0); + + dsl_dir_rele(dd, FTAG); + return (err); } int -dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, - dsl_dataset_t **dsp) -{ - int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); - - if (err) - return (err); - return (dsl_dataset_hold_ref(*dsp, tag)); -} - -int -dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, +dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, dsl_dataset_t **dsp) { int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); - if (err) + if (err != 0) return (err); - if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { + if (!dsl_dataset_tryown(*dsp, tag)) { dsl_dataset_rele(*dsp, tag); *dsp = NULL; return (EBUSY); @@ -580,78 +522,48 @@ dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, } int -dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) -{ - dsl_dir_t *dd; - dsl_pool_t *dp; - const char *snapname; - uint64_t obj; - int err = 0; - - err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); - if (err) - return (err); - - dp = dd->dd_pool; - obj = dd->dd_phys->dd_head_dataset_obj; - rw_enter(&dp->dp_config_rwlock, RW_READER); - if (obj) - err = dsl_dataset_get_ref(dp, obj, tag, dsp); - else - err = ENOENT; - if (err) - goto out; - - err = dsl_dataset_hold_ref(*dsp, tag); - - /* we may be looking for a snapshot */ - if (err == 0 && snapname != NULL) { - dsl_dataset_t *ds = NULL; - - if (*snapname++ != '@') { - dsl_dataset_rele(*dsp, tag); - err = ENOENT; - goto out; - } - - dprintf("looking for snapshot '%s'\n", snapname); - err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); - if (err == 0) - err = dsl_dataset_get_ref(dp, obj, tag, &ds); - dsl_dataset_rele(*dsp, tag); - - ASSERT3U((err == 0), ==, (ds != NULL)); - - if (ds) { - mutex_enter(&ds->ds_lock); - if (ds->ds_snapname[0] == 0) - (void) strlcpy(ds->ds_snapname, snapname, - sizeof (ds->ds_snapname)); - mutex_exit(&ds->ds_lock); - err = dsl_dataset_hold_ref(ds, tag); - *dsp = err ? NULL : ds; - } - } -out: - rw_exit(&dp->dp_config_rwlock); - dsl_dir_close(dd, FTAG); - return (err); -} - -int -dsl_dataset_own(const char *name, boolean_t inconsistentok, +dsl_dataset_own(dsl_pool_t *dp, const char *name, void *tag, dsl_dataset_t **dsp) { - int err = dsl_dataset_hold(name, tag, dsp); - if (err) + int err = dsl_dataset_hold(dp, name, tag, dsp); + if (err != 0) return (err); - if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { + if (!dsl_dataset_tryown(*dsp, tag)) { dsl_dataset_rele(*dsp, tag); return (EBUSY); } return (0); } +/* + * See the comment above dsl_pool_hold() for details. In summary, a long + * hold is used to prevent destruction of a dataset while the pool hold + * is dropped, allowing other concurrent operations (e.g. spa_sync()). + * + * The dataset and pool must be held when this function is called. After it + * is called, the pool hold may be released while the dataset is still held + * and accessed. + */ +void +dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) +{ + ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); + (void) refcount_add(&ds->ds_longholds, tag); +} + +void +dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) +{ + (void) refcount_remove(&ds->ds_longholds, tag); +} + +/* Return B_TRUE if there are any long holds on this dataset. */ +boolean_t +dsl_dataset_long_held(dsl_dataset_t *ds) +{ + return (!refcount_is_zero(&ds->ds_longholds)); +} + void dsl_dataset_name(dsl_dataset_t *ds, char *name) { @@ -659,7 +571,7 @@ dsl_dataset_name(dsl_dataset_t *ds, char *name) (void) strcpy(name, "mos"); } else { dsl_dir_name(ds->ds_dir, name); - VERIFY(0 == dsl_dataset_get_snapname(ds)); + VERIFY0(dsl_dataset_get_snapname(ds)); if (ds->ds_snapname[0]) { (void) strcat(name, "@"); /* @@ -686,7 +598,7 @@ dsl_dataset_namelen(dsl_dataset_t *ds) result = 3; /* "mos" */ } else { result = dsl_dir_namelen(ds->ds_dir); - VERIFY(0 == dsl_dataset_get_snapname(ds)); + VERIFY0(dsl_dataset_get_snapname(ds)); if (ds->ds_snapname[0]) { ++result; /* adding one for the @-sign */ if (!MUTEX_HELD(&ds->ds_lock)) { @@ -703,64 +615,41 @@ dsl_dataset_namelen(dsl_dataset_t *ds) } void -dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) +dsl_dataset_rele(dsl_dataset_t *ds, void *tag) { dmu_buf_rele(ds->ds_dbuf, tag); } -void -dsl_dataset_rele(dsl_dataset_t *ds, void *tag) -{ - if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { - rw_exit(&ds->ds_rwlock); - } - dsl_dataset_drop_ref(ds, tag); -} - void dsl_dataset_disown(dsl_dataset_t *ds, void *tag) { - ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || - (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); + ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL); mutex_enter(&ds->ds_lock); ds->ds_owner = NULL; - if (RW_WRITE_HELD(&ds->ds_rwlock)) { - rw_exit(&ds->ds_rwlock); - cv_broadcast(&ds->ds_exclusive_cv); - } mutex_exit(&ds->ds_lock); - if (ds->ds_dbuf) - dsl_dataset_drop_ref(ds, tag); + dsl_dataset_long_rele(ds, tag); + if (ds->ds_dbuf != NULL) + dsl_dataset_rele(ds, tag); else dsl_dataset_evict(NULL, ds); } boolean_t -dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) +dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) { boolean_t gotit = FALSE; mutex_enter(&ds->ds_lock); - if (ds->ds_owner == NULL && - (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { + if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { ds->ds_owner = tag; - if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) - rw_exit(&ds->ds_rwlock); + dsl_dataset_long_hold(ds, tag); gotit = TRUE; } mutex_exit(&ds->ds_lock); return (gotit); } -void -dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) -{ - ASSERT3P(owner, ==, ds->ds_owner); - if (!RW_WRITE_HELD(&ds->ds_rwlock)) - rw_enter(&ds->ds_rwlock, RW_WRITER); -} - uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, uint64_t flags, dmu_tx_t *tx) @@ -781,7 +670,7 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); - VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); + VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); @@ -799,7 +688,7 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, if (origin == NULL) { dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); } else { - dsl_dataset_t *ohds; + dsl_dataset_t *ohds; /* head of the origin snapshot */ dsphys->ds_prev_snap_obj = origin->ds_object; dsphys->ds_prev_snap_txg = @@ -816,7 +705,7 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, dmu_buf_will_dirty(origin->ds_dbuf, tx); origin->ds_phys->ds_num_children++; - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, + VERIFY0(dsl_dataset_hold_obj(dp, origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); @@ -828,9 +717,8 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, zap_create(mos, DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); } - VERIFY(0 == zap_add_int(mos, - origin->ds_phys->ds_next_clones_obj, - dsobj, tx)); + VERIFY0(zap_add_int(mos, + origin->ds_phys->ds_next_clones_obj, dsobj, tx)); } dmu_buf_will_dirty(dd->dd_dbuf, tx); @@ -842,7 +730,7 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } - VERIFY3U(0, ==, zap_add_int(mos, + VERIFY0(zap_add_int(mos, origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); } } @@ -858,6 +746,16 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, return (dsobj); } +static void +dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) +{ + objset_t *os; + + VERIFY0(dmu_objset_from_ds(ds, &os)); + bzero(&os->os_zil_header, sizeof (os->os_zil_header)); + dsl_dataset_dirty(ds, tx); +} + uint64_t dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) @@ -866,29 +764,28 @@ dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, uint64_t dsobj, ddobj; dsl_dir_t *dd; + ASSERT(dmu_tx_is_syncing(tx)); ASSERT(lastname[0] != '@'); ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); - VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); + VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); - dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); + dsobj = dsl_dataset_create_sync_dd(dd, origin, + flags & ~DS_CREATE_FLAG_NODIRTY, tx); dsl_deleg_set_create_perms(dd, tx, cr); - dsl_dir_close(dd, FTAG); + dsl_dir_rele(dd, FTAG); /* * If we are creating a clone, make sure we zero out any stale * data from the origin snapshots zil header. */ - if (origin != NULL) { + if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { dsl_dataset_t *ds; - objset_t *os; - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); - VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); - bzero(&os->os_zil_header, sizeof (os->os_zil_header)); - dsl_dataset_dirty(ds, tx); + VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); + dsl_dataset_zero_zil(ds, tx); dsl_dataset_rele(ds, FTAG); } @@ -896,271 +793,65 @@ dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, } /* - * The snapshots must all be in the same pool. + * The unique space in the head dataset can be calculated by subtracting + * the space used in the most recent snapshot, that is still being used + * in this file system, from the space currently in use. To figure out + * the space in the most recent snapshot still in use, we need to take + * the total space used in the snapshot and subtract out the space that + * has been freed up since the snapshot was taken. */ -int -dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, - nvlist_t *errlist) +void +dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) { + uint64_t mrs_used; + uint64_t dlused, dlcomp, dluncomp; + + ASSERT(!dsl_dataset_is_snapshot(ds)); + + if (ds->ds_phys->ds_prev_snap_obj != 0) + mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; + else + mrs_used = 0; + + dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); + + ASSERT3U(dlused, <=, mrs_used); + ds->ds_phys->ds_unique_bytes = + ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); + + if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= + SPA_VERSION_UNIQUE_ACCURATE) + ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; +} + +void +dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, + dmu_tx_t *tx) +{ + objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; + uint64_t count; int err; - dsl_sync_task_t *dst; - spa_t *spa; - nvpair_t *pair; - dsl_sync_task_group_t *dstg; - - pair = nvlist_next_nvpair(snaps, NULL); - if (pair == NULL) - return (0); - - err = spa_open(nvpair_name(pair), &spa, FTAG); - if (err) - return (err); - dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - - for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; - pair = nvlist_next_nvpair(snaps, pair)) { - dsl_dataset_t *ds; - - err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds); - if (err == 0) { - struct dsl_ds_destroyarg *dsda; - - dsl_dataset_make_exclusive(ds, dstg); - dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), - KM_SLEEP); - dsda->ds = ds; - dsda->defer = defer; - dsl_sync_task_create(dstg, dsl_dataset_destroy_check, - dsl_dataset_destroy_sync, dsda, dstg, 0); - } else if (err == ENOENT) { - err = 0; - } else { - fnvlist_add_int32(errlist, nvpair_name(pair), err); - break; - } - } - - if (err == 0) - err = dsl_sync_task_group_wait(dstg); - - for (dst = list_head(&dstg->dstg_tasks); dst; - dst = list_next(&dstg->dstg_tasks, dst)) { - struct dsl_ds_destroyarg *dsda = dst->dst_arg1; - dsl_dataset_t *ds = dsda->ds; - - /* - * Return the snapshots that triggered the error. - */ - if (dst->dst_err != 0) { - char name[ZFS_MAXNAMELEN]; - dsl_dataset_name(ds, name); - fnvlist_add_int32(errlist, name, dst->dst_err); - } - ASSERT3P(dsda->rm_origin, ==, NULL); - dsl_dataset_disown(ds, dstg); - kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); - } - - dsl_sync_task_group_destroy(dstg); - spa_close(spa, FTAG); - return (err); - -} - -static boolean_t -dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) -{ - boolean_t might_destroy = B_FALSE; - - mutex_enter(&ds->ds_lock); - if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && - DS_IS_DEFER_DESTROY(ds)) - might_destroy = B_TRUE; - mutex_exit(&ds->ds_lock); - - return (might_destroy); -} - -/* - * If we're removing a clone, and these three conditions are true: - * 1) the clone's origin has no other children - * 2) the clone's origin has no user references - * 3) the clone's origin has been marked for deferred destruction - * Then, prepare to remove the origin as part of this sync task group. - */ -static int -dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) -{ - dsl_dataset_t *ds = dsda->ds; - dsl_dataset_t *origin = ds->ds_prev; - - if (dsl_dataset_might_destroy_origin(origin)) { - char *name; - int namelen; - int error; - - namelen = dsl_dataset_namelen(origin) + 1; - name = kmem_alloc(namelen, KM_SLEEP); - dsl_dataset_name(origin, name); -#ifdef _KERNEL - error = zfs_unmount_snap(name, NULL); - if (error) { - kmem_free(name, namelen); - return (error); - } -#endif - error = dsl_dataset_own(name, B_TRUE, tag, &origin); - kmem_free(name, namelen); - if (error) - return (error); - dsda->rm_origin = origin; - dsl_dataset_make_exclusive(origin, tag); - } - - return (0); -} - -/* - * ds must be opened as OWNER. On return (whether successful or not), - * ds will be closed and caller can no longer dereference it. - */ -int -dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) -{ - int err; - dsl_sync_task_group_t *dstg; - objset_t *os; - dsl_dir_t *dd; - uint64_t obj; - struct dsl_ds_destroyarg dsda = { 0 }; - - dsda.ds = ds; - - if (dsl_dataset_is_snapshot(ds)) { - /* Destroying a snapshot is simpler */ - dsl_dataset_make_exclusive(ds, tag); - - dsda.defer = defer; - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - dsl_dataset_destroy_check, dsl_dataset_destroy_sync, - &dsda, tag, 0); - ASSERT3P(dsda.rm_origin, ==, NULL); - goto out; - } else if (defer) { - err = EINVAL; - goto out; - } - - dd = ds->ds_dir; - - if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds), - &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { - /* - * Check for errors and mark this ds as inconsistent, in - * case we crash while freeing the objects. - */ - err = dsl_sync_task_do(dd->dd_pool, - dsl_dataset_destroy_begin_check, - dsl_dataset_destroy_begin_sync, ds, NULL, 0); - if (err) - goto out; - - err = dmu_objset_from_ds(ds, &os); - if (err) - goto out; - - /* - * Remove all objects while in the open context so that - * there is less work to do in the syncing context. - */ - for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, - ds->ds_phys->ds_prev_snap_txg)) { - /* - * Ignore errors, if there is not enough disk space - * we will deal with it in dsl_dataset_destroy_sync(). - */ - (void) dmu_free_object(os, obj); - } - if (err != ESRCH) - goto out; - - /* - * Sync out all in-flight IO. - */ - txg_wait_synced(dd->dd_pool, 0); - - /* - * If we managed to free all the objects in open - * context, the user space accounting should be zero. - */ - if (ds->ds_phys->ds_bp.blk_fill == 0 && - dmu_objset_userused_enabled(os)) { - uint64_t count; - - ASSERT(zap_count(os, DMU_USERUSED_OBJECT, - &count) != 0 || count == 0); - ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, - &count) != 0 || count == 0); - } - } - - rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); - err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); - rw_exit(&dd->dd_pool->dp_config_rwlock); - - if (err) - goto out; + ASSERT(ds->ds_phys->ds_num_children >= 2); + err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); /* - * Blow away the dsl_dir + head dataset. + * The err should not be ENOENT, but a bug in a previous version + * of the code could cause upgrade_clones_cb() to not set + * ds_next_snap_obj when it should, leading to a missing entry. + * If we knew that the pool was created after + * SPA_VERSION_NEXT_CLONES, we could assert that it isn't + * ENOENT. However, at least we can check that we don't have + * too many entries in the next_clones_obj even after failing to + * remove this one. */ - dsl_dataset_make_exclusive(ds, tag); - /* - * If we're removing a clone, we might also need to remove its - * origin. - */ - do { - dsda.need_prep = B_FALSE; - if (dsl_dir_is_clone(dd)) { - err = dsl_dataset_origin_rm_prep(&dsda, tag); - if (err) { - dsl_dir_close(dd, FTAG); - goto out; - } - } - - dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); - dsl_sync_task_create(dstg, dsl_dataset_destroy_check, - dsl_dataset_destroy_sync, &dsda, tag, 0); - dsl_sync_task_create(dstg, dsl_dir_destroy_check, - dsl_dir_destroy_sync, dd, FTAG, 0); - err = dsl_sync_task_group_wait(dstg); - dsl_sync_task_group_destroy(dstg); - - /* - * We could be racing against 'zfs release' or 'zfs destroy -d' - * on the origin snap, in which case we can get EBUSY if we - * needed to destroy the origin snap but were not ready to - * do so. - */ - if (dsda.need_prep) { - ASSERT(err == EBUSY); - ASSERT(dsl_dir_is_clone(dd)); - ASSERT(dsda.rm_origin == NULL); - } - } while (dsda.need_prep); - - if (dsda.rm_origin != NULL) - dsl_dataset_disown(dsda.rm_origin, tag); - - /* if it is successful, dsl_dir_destroy_sync will close the dd */ - if (err) - dsl_dir_close(dd, FTAG); -out: - dsl_dataset_disown(ds, tag); - return (err); + if (err != ENOENT) + VERIFY0(err); + ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, + &count)); + ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); } + blkptr_t * dsl_dataset_get_blkptr(dsl_dataset_t *ds) { @@ -1201,7 +892,7 @@ dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) dp = ds->ds_dir->dd_pool; - if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { + if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { /* up the hold count until we can be written out */ dmu_buf_add_ref(ds->ds_dbuf, ds); } @@ -1218,769 +909,6 @@ dsl_dataset_is_dirty(dsl_dataset_t *ds) return (B_FALSE); } -/* - * The unique space in the head dataset can be calculated by subtracting - * the space used in the most recent snapshot, that is still being used - * in this file system, from the space currently in use. To figure out - * the space in the most recent snapshot still in use, we need to take - * the total space used in the snapshot and subtract out the space that - * has been freed up since the snapshot was taken. - */ -static void -dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) -{ - uint64_t mrs_used; - uint64_t dlused, dlcomp, dluncomp; - - ASSERT(!dsl_dataset_is_snapshot(ds)); - - if (ds->ds_phys->ds_prev_snap_obj != 0) - mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; - else - mrs_used = 0; - - dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); - - ASSERT3U(dlused, <=, mrs_used); - ds->ds_phys->ds_unique_bytes = - ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); - - if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= - SPA_VERSION_UNIQUE_ACCURATE) - ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; -} - -struct killarg { - dsl_dataset_t *ds; - dmu_tx_t *tx; -}; - -/* ARGSUSED */ -static int -kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, - const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) -{ - struct killarg *ka = arg; - dmu_tx_t *tx = ka->tx; - - if (bp == NULL) - return (0); - - if (zb->zb_level == ZB_ZIL_LEVEL) { - ASSERT(zilog != NULL); - /* - * It's a block in the intent log. It has no - * accounting, so just free it. - */ - dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); - } else { - ASSERT(zilog == NULL); - ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); - (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); - } - - return (0); -} - -/* ARGSUSED */ -static int -dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dataset_t *ds = arg1; - objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; - uint64_t count; - int err; - - /* - * Can't delete a head dataset if there are snapshots of it. - * (Except if the only snapshots are from the branch we cloned - * from.) - */ - if (ds->ds_prev != NULL && - ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) - return (EBUSY); - - /* - * This is really a dsl_dir thing, but check it here so that - * we'll be less likely to leave this dataset inconsistent & - * nearly destroyed. - */ - err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); - if (err) - return (err); - if (count != 0) - return (EEXIST); - - return (0); -} - -/* ARGSUSED */ -static void -dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dataset_t *ds = arg1; - - /* Mark it as inconsistent on-disk, in case we crash */ - dmu_buf_will_dirty(ds->ds_dbuf, tx); - ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; - - spa_history_log_internal_ds(ds, "destroy begin", tx, ""); -} - -static int -dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, - dmu_tx_t *tx) -{ - dsl_dataset_t *ds = dsda->ds; - dsl_dataset_t *ds_prev = ds->ds_prev; - - if (dsl_dataset_might_destroy_origin(ds_prev)) { - struct dsl_ds_destroyarg ndsda = {0}; - - /* - * If we're not prepared to remove the origin, don't remove - * the clone either. - */ - if (dsda->rm_origin == NULL) { - dsda->need_prep = B_TRUE; - return (EBUSY); - } - - ndsda.ds = ds_prev; - ndsda.is_origin_rm = B_TRUE; - return (dsl_dataset_destroy_check(&ndsda, tag, tx)); - } - - /* - * If we're not going to remove the origin after all, - * undo the open context setup. - */ - if (dsda->rm_origin != NULL) { - dsl_dataset_disown(dsda->rm_origin, tag); - dsda->rm_origin = NULL; - } - - return (0); -} - -/* - * If you add new checks here, you may need to add - * additional checks to the "temporary" case in - * snapshot_check() in dmu_objset.c. - */ -/* ARGSUSED */ -int -dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) -{ - struct dsl_ds_destroyarg *dsda = arg1; - dsl_dataset_t *ds = dsda->ds; - - /* we have an owner hold, so noone else can destroy us */ - ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); - - /* - * Only allow deferred destroy on pools that support it. - * NOTE: deferred destroy is only supported on snapshots. - */ - if (dsda->defer) { - if (spa_version(ds->ds_dir->dd_pool->dp_spa) < - SPA_VERSION_USERREFS) - return (ENOTSUP); - ASSERT(dsl_dataset_is_snapshot(ds)); - return (0); - } - - /* - * Can't delete a head dataset if there are snapshots of it. - * (Except if the only snapshots are from the branch we cloned - * from.) - */ - if (ds->ds_prev != NULL && - ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) - return (EBUSY); - - /* - * If we made changes this txg, traverse_dsl_dataset won't find - * them. Try again. - */ - if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) - return (EAGAIN); - - if (dsl_dataset_is_snapshot(ds)) { - /* - * If this snapshot has an elevated user reference count, - * we can't destroy it yet. - */ - if (ds->ds_userrefs > 0 && !dsda->releasing) - return (EBUSY); - - mutex_enter(&ds->ds_lock); - /* - * Can't delete a branch point. However, if we're destroying - * a clone and removing its origin due to it having a user - * hold count of 0 and having been marked for deferred destroy, - * it's OK for the origin to have a single clone. - */ - if (ds->ds_phys->ds_num_children > - (dsda->is_origin_rm ? 2 : 1)) { - mutex_exit(&ds->ds_lock); - return (EEXIST); - } - mutex_exit(&ds->ds_lock); - } else if (dsl_dir_is_clone(ds->ds_dir)) { - return (dsl_dataset_origin_check(dsda, arg2, tx)); - } - - /* XXX we should do some i/o error checking... */ - return (0); -} - -struct refsarg { - kmutex_t lock; - boolean_t gone; - kcondvar_t cv; -}; - -/* ARGSUSED */ -static void -dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) -{ - struct refsarg *arg = argv; - - mutex_enter(&arg->lock); - arg->gone = TRUE; - cv_signal(&arg->cv); - mutex_exit(&arg->lock); -} - -static void -dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) -{ - struct refsarg arg; - - mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); - cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); - arg.gone = FALSE; - (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, - dsl_dataset_refs_gone); - dmu_buf_rele(ds->ds_dbuf, tag); - mutex_enter(&arg.lock); - while (!arg.gone) - cv_wait(&arg.cv, &arg.lock); - ASSERT(arg.gone); - mutex_exit(&arg.lock); - ds->ds_dbuf = NULL; - ds->ds_phys = NULL; - mutex_destroy(&arg.lock); - cv_destroy(&arg.cv); -} - -static void -remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) -{ - objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; - uint64_t count; - int err; - - ASSERT(ds->ds_phys->ds_num_children >= 2); - err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); - /* - * The err should not be ENOENT, but a bug in a previous version - * of the code could cause upgrade_clones_cb() to not set - * ds_next_snap_obj when it should, leading to a missing entry. - * If we knew that the pool was created after - * SPA_VERSION_NEXT_CLONES, we could assert that it isn't - * ENOENT. However, at least we can check that we don't have - * too many entries in the next_clones_obj even after failing to - * remove this one. - */ - if (err != ENOENT) { - VERIFY0(err); - } - ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, - &count)); - ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); -} - -static void -dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx) -{ - objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; - zap_cursor_t zc; - zap_attribute_t za; - - /* - * If it is the old version, dd_clones doesn't exist so we can't - * find the clones, but deadlist_remove_key() is a no-op so it - * doesn't matter. - */ - if (ds->ds_dir->dd_phys->dd_clones == 0) - return; - - for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones); - zap_cursor_retrieve(&zc, &za) == 0; - zap_cursor_advance(&zc)) { - dsl_dataset_t *clone; - - VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool, - za.za_first_integer, FTAG, &clone)); - if (clone->ds_dir->dd_origin_txg > mintxg) { - dsl_deadlist_remove_key(&clone->ds_deadlist, - mintxg, tx); - dsl_dataset_remove_clones_key(clone, mintxg, tx); - } - dsl_dataset_rele(clone, FTAG); - } - zap_cursor_fini(&zc); -} - -struct process_old_arg { - dsl_dataset_t *ds; - dsl_dataset_t *ds_prev; - boolean_t after_branch_point; - zio_t *pio; - uint64_t used, comp, uncomp; -}; - -static int -process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) -{ - struct process_old_arg *poa = arg; - dsl_pool_t *dp = poa->ds->ds_dir->dd_pool; - - if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) { - dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx); - if (poa->ds_prev && !poa->after_branch_point && - bp->blk_birth > - poa->ds_prev->ds_phys->ds_prev_snap_txg) { - poa->ds_prev->ds_phys->ds_unique_bytes += - bp_get_dsize_sync(dp->dp_spa, bp); - } - } else { - poa->used += bp_get_dsize_sync(dp->dp_spa, bp); - poa->comp += BP_GET_PSIZE(bp); - poa->uncomp += BP_GET_UCSIZE(bp); - dsl_free_sync(poa->pio, dp, tx->tx_txg, bp); - } - return (0); -} - -static void -process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, - dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx) -{ - struct process_old_arg poa = { 0 }; - dsl_pool_t *dp = ds->ds_dir->dd_pool; - objset_t *mos = dp->dp_meta_objset; - - ASSERT(ds->ds_deadlist.dl_oldfmt); - ASSERT(ds_next->ds_deadlist.dl_oldfmt); - - poa.ds = ds; - poa.ds_prev = ds_prev; - poa.after_branch_point = after_branch_point; - poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); - VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj, - process_old_cb, &poa, tx)); - VERIFY0(zio_wait(poa.pio)); - ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes); - - /* change snapused */ - dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, - -poa.used, -poa.comp, -poa.uncomp, tx); - - /* swap next's deadlist to our deadlist */ - dsl_deadlist_close(&ds->ds_deadlist); - dsl_deadlist_close(&ds_next->ds_deadlist); - SWITCH64(ds_next->ds_phys->ds_deadlist_obj, - ds->ds_phys->ds_deadlist_obj); - dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); - dsl_deadlist_open(&ds_next->ds_deadlist, mos, - ds_next->ds_phys->ds_deadlist_obj); -} - -static int -old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx) -{ - int err; - struct killarg ka; - - /* - * Free everything that we point to (that's born after - * the previous snapshot, if we are a clone) - * - * NB: this should be very quick, because we already - * freed all the objects in open context. - */ - ka.ds = ds; - ka.tx = tx; - err = traverse_dataset(ds, - ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST, - kill_blkptr, &ka); - ASSERT0(err); - ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0); - - return (err); -} - -void -dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) -{ - struct dsl_ds_destroyarg *dsda = arg1; - dsl_dataset_t *ds = dsda->ds; - int err; - int after_branch_point = FALSE; - dsl_pool_t *dp = ds->ds_dir->dd_pool; - objset_t *mos = dp->dp_meta_objset; - dsl_dataset_t *ds_prev = NULL; - boolean_t wont_destroy; - uint64_t obj; - - wont_destroy = (dsda->defer && - (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)); - - ASSERT(ds->ds_owner || wont_destroy); - ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); - ASSERT(ds->ds_prev == NULL || - ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); - ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); - - if (wont_destroy) { - ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); - dmu_buf_will_dirty(ds->ds_dbuf, tx); - ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; - spa_history_log_internal_ds(ds, "defer_destroy", tx, ""); - return; - } - - /* We need to log before removing it from the namespace. */ - spa_history_log_internal_ds(ds, "destroy", tx, ""); - - /* signal any waiters that this dataset is going away */ - mutex_enter(&ds->ds_lock); - ds->ds_owner = dsl_reaper; - cv_broadcast(&ds->ds_exclusive_cv); - mutex_exit(&ds->ds_lock); - - /* Remove our reservation */ - if (ds->ds_reserved != 0) { - dsl_prop_setarg_t psa; - uint64_t value = 0; - - dsl_prop_setarg_init_uint64(&psa, "refreservation", - (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), - &value); - psa.psa_effective_value = 0; /* predict default value */ - - dsl_dataset_set_reservation_sync(ds, &psa, tx); - ASSERT0(ds->ds_reserved); - } - - ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); - - dsl_scan_ds_destroyed(ds, tx); - - obj = ds->ds_object; - - if (ds->ds_phys->ds_prev_snap_obj != 0) { - if (ds->ds_prev) { - ds_prev = ds->ds_prev; - } else { - VERIFY(0 == dsl_dataset_hold_obj(dp, - ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); - } - after_branch_point = - (ds_prev->ds_phys->ds_next_snap_obj != obj); - - dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); - if (after_branch_point && - ds_prev->ds_phys->ds_next_clones_obj != 0) { - remove_from_next_clones(ds_prev, obj, tx); - if (ds->ds_phys->ds_next_snap_obj != 0) { - VERIFY(0 == zap_add_int(mos, - ds_prev->ds_phys->ds_next_clones_obj, - ds->ds_phys->ds_next_snap_obj, tx)); - } - } - if (after_branch_point && - ds->ds_phys->ds_next_snap_obj == 0) { - /* This clone is toast. */ - ASSERT(ds_prev->ds_phys->ds_num_children > 1); - ds_prev->ds_phys->ds_num_children--; - - /* - * If the clone's origin has no other clones, no - * user holds, and has been marked for deferred - * deletion, then we should have done the necessary - * destroy setup for it. - */ - if (ds_prev->ds_phys->ds_num_children == 1 && - ds_prev->ds_userrefs == 0 && - DS_IS_DEFER_DESTROY(ds_prev)) { - ASSERT3P(dsda->rm_origin, !=, NULL); - } else { - ASSERT3P(dsda->rm_origin, ==, NULL); - } - } else if (!after_branch_point) { - ds_prev->ds_phys->ds_next_snap_obj = - ds->ds_phys->ds_next_snap_obj; - } - } - - if (dsl_dataset_is_snapshot(ds)) { - dsl_dataset_t *ds_next; - uint64_t old_unique; - uint64_t used = 0, comp = 0, uncomp = 0; - - VERIFY(0 == dsl_dataset_hold_obj(dp, - ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); - ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); - - old_unique = ds_next->ds_phys->ds_unique_bytes; - - dmu_buf_will_dirty(ds_next->ds_dbuf, tx); - ds_next->ds_phys->ds_prev_snap_obj = - ds->ds_phys->ds_prev_snap_obj; - ds_next->ds_phys->ds_prev_snap_txg = - ds->ds_phys->ds_prev_snap_txg; - ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, - ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); - - - if (ds_next->ds_deadlist.dl_oldfmt) { - process_old_deadlist(ds, ds_prev, ds_next, - after_branch_point, tx); - } else { - /* Adjust prev's unique space. */ - if (ds_prev && !after_branch_point) { - dsl_deadlist_space_range(&ds_next->ds_deadlist, - ds_prev->ds_phys->ds_prev_snap_txg, - ds->ds_phys->ds_prev_snap_txg, - &used, &comp, &uncomp); - ds_prev->ds_phys->ds_unique_bytes += used; - } - - /* Adjust snapused. */ - dsl_deadlist_space_range(&ds_next->ds_deadlist, - ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, - &used, &comp, &uncomp); - dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, - -used, -comp, -uncomp, tx); - - /* Move blocks to be freed to pool's free list. */ - dsl_deadlist_move_bpobj(&ds_next->ds_deadlist, - &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg, - tx); - dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, - DD_USED_HEAD, used, comp, uncomp, tx); - - /* Merge our deadlist into next's and free it. */ - dsl_deadlist_merge(&ds_next->ds_deadlist, - ds->ds_phys->ds_deadlist_obj, tx); - } - dsl_deadlist_close(&ds->ds_deadlist); - dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); - - /* Collapse range in clone heads */ - dsl_dataset_remove_clones_key(ds, - ds->ds_phys->ds_creation_txg, tx); - - if (dsl_dataset_is_snapshot(ds_next)) { - dsl_dataset_t *ds_nextnext; - - /* - * Update next's unique to include blocks which - * were previously shared by only this snapshot - * and it. Those blocks will be born after the - * prev snap and before this snap, and will have - * died after the next snap and before the one - * after that (ie. be on the snap after next's - * deadlist). - */ - VERIFY(0 == dsl_dataset_hold_obj(dp, - ds_next->ds_phys->ds_next_snap_obj, - FTAG, &ds_nextnext)); - dsl_deadlist_space_range(&ds_nextnext->ds_deadlist, - ds->ds_phys->ds_prev_snap_txg, - ds->ds_phys->ds_creation_txg, - &used, &comp, &uncomp); - ds_next->ds_phys->ds_unique_bytes += used; - dsl_dataset_rele(ds_nextnext, FTAG); - ASSERT3P(ds_next->ds_prev, ==, NULL); - - /* Collapse range in this head. */ - dsl_dataset_t *hds; - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, - ds->ds_dir->dd_phys->dd_head_dataset_obj, - FTAG, &hds)); - dsl_deadlist_remove_key(&hds->ds_deadlist, - ds->ds_phys->ds_creation_txg, tx); - dsl_dataset_rele(hds, FTAG); - - } else { - ASSERT3P(ds_next->ds_prev, ==, ds); - dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); - ds_next->ds_prev = NULL; - if (ds_prev) { - VERIFY(0 == dsl_dataset_get_ref(dp, - ds->ds_phys->ds_prev_snap_obj, - ds_next, &ds_next->ds_prev)); - } - - dsl_dataset_recalc_head_uniq(ds_next); - - /* - * Reduce the amount of our unconsmed refreservation - * being charged to our parent by the amount of - * new unique data we have gained. - */ - if (old_unique < ds_next->ds_reserved) { - int64_t mrsdelta; - uint64_t new_unique = - ds_next->ds_phys->ds_unique_bytes; - - ASSERT(old_unique <= new_unique); - mrsdelta = MIN(new_unique - old_unique, - ds_next->ds_reserved - old_unique); - dsl_dir_diduse_space(ds->ds_dir, - DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); - } - } - dsl_dataset_rele(ds_next, FTAG); - } else { - zfeature_info_t *async_destroy = - &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]; - objset_t *os; - - /* - * There's no next snapshot, so this is a head dataset. - * Destroy the deadlist. Unless it's a clone, the - * deadlist should be empty. (If it's a clone, it's - * safe to ignore the deadlist contents.) - */ - dsl_deadlist_close(&ds->ds_deadlist); - dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); - ds->ds_phys->ds_deadlist_obj = 0; - - VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); - - if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) { - err = old_synchronous_dataset_destroy(ds, tx); - } else { - /* - * Move the bptree into the pool's list of trees to - * clean up and update space accounting information. - */ - uint64_t used, comp, uncomp; - - zil_destroy_sync(dmu_objset_zil(os), tx); - - if (!spa_feature_is_active(dp->dp_spa, async_destroy)) { - spa_feature_incr(dp->dp_spa, async_destroy, tx); - dp->dp_bptree_obj = bptree_alloc(mos, tx); - VERIFY(zap_add(mos, - DMU_POOL_DIRECTORY_OBJECT, - DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, - &dp->dp_bptree_obj, tx) == 0); - } - - used = ds->ds_dir->dd_phys->dd_used_bytes; - comp = ds->ds_dir->dd_phys->dd_compressed_bytes; - uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes; - - ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || - ds->ds_phys->ds_unique_bytes == used); - - bptree_add(mos, dp->dp_bptree_obj, - &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg, - used, comp, uncomp, tx); - dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, - -used, -comp, -uncomp, tx); - dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, - used, comp, uncomp, tx); - } - - if (ds->ds_prev != NULL) { - if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { - VERIFY3U(0, ==, zap_remove_int(mos, - ds->ds_prev->ds_dir->dd_phys->dd_clones, - ds->ds_object, tx)); - } - dsl_dataset_rele(ds->ds_prev, ds); - ds->ds_prev = ds_prev = NULL; - } - } - - /* - * This must be done after the dsl_traverse(), because it will - * re-open the objset. - */ - if (ds->ds_objset) { - dmu_objset_evict(ds->ds_objset); - ds->ds_objset = NULL; - } - - if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { - /* Erase the link in the dir */ - dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); - ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; - ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); - err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); - ASSERT(err == 0); - } else { - /* remove from snapshot namespace */ - dsl_dataset_t *ds_head; - ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); - VERIFY(0 == dsl_dataset_hold_obj(dp, - ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); - VERIFY(0 == dsl_dataset_get_snapname(ds)); -#ifdef ZFS_DEBUG - { - uint64_t val; - - err = dsl_dataset_snap_lookup(ds_head, - ds->ds_snapname, &val); - ASSERT0(err); - ASSERT3U(val, ==, obj); - } -#endif - err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); - ASSERT(err == 0); - dsl_dataset_rele(ds_head, FTAG); - } - - if (ds_prev && ds->ds_prev != ds_prev) - dsl_dataset_rele(ds_prev, FTAG); - - spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); - - if (ds->ds_phys->ds_next_clones_obj != 0) { - uint64_t count; - ASSERT(0 == zap_count(mos, - ds->ds_phys->ds_next_clones_obj, &count) && count == 0); - VERIFY(0 == dmu_object_free(mos, - ds->ds_phys->ds_next_clones_obj, tx)); - } - if (ds->ds_phys->ds_props_obj != 0) - VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); - if (ds->ds_phys->ds_userrefs_obj != 0) - VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); - dsl_dir_close(ds->ds_dir, ds); - ds->ds_dir = NULL; - dsl_dataset_drain_refs(ds, tag); - VERIFY(0 == dmu_object_free(mos, obj, tx)); - - if (dsda->rm_origin) { - /* - * Remove the origin of the clone we just destroyed. - */ - struct dsl_ds_destroyarg ndsda = {0}; - - ndsda.ds = dsda->rm_origin; - dsl_dataset_destroy_sync(&ndsda, tag, tx); - } -} - static int dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) { @@ -2009,13 +937,24 @@ dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) return (0); } +typedef struct dsl_dataset_snapshot_arg { + nvlist_t *ddsa_snaps; + nvlist_t *ddsa_props; + nvlist_t *ddsa_errors; +} dsl_dataset_snapshot_arg_t; + int -dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname, +dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx) { - int err; + int error; uint64_t value; + ds->ds_trysnap_txg = tx->tx_txg; + + if (!dmu_tx_is_syncing(tx)) + return (0); + /* * We don't allow multiple snapshots of the same txg. If there * is already one, try again. @@ -2026,39 +965,87 @@ dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname, /* * Check for conflicting snapshot name. */ - err = dsl_dataset_snap_lookup(ds, snapname, &value); - if (err == 0) + error = dsl_dataset_snap_lookup(ds, snapname, &value); + if (error == 0) return (EEXIST); - if (err != ENOENT) - return (err); + if (error != ENOENT) + return (error); - /* - * Check that the dataset's name is not too long. Name consists - * of the dataset's length + 1 for the @-sign + snapshot name's length - */ - if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) - return (ENAMETOOLONG); + error = dsl_dataset_snapshot_reserve_space(ds, tx); + if (error != 0) + return (error); - err = dsl_dataset_snapshot_reserve_space(ds, tx); - if (err) - return (err); - - ds->ds_trysnap_txg = tx->tx_txg; return (0); } +static int +dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_snapshot_arg_t *ddsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + int rv = 0; + + for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); + pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { + int error = 0; + dsl_dataset_t *ds; + char *name, *atp; + char dsname[MAXNAMELEN]; + + name = nvpair_name(pair); + if (strlen(name) >= MAXNAMELEN) + error = ENAMETOOLONG; + if (error == 0) { + atp = strchr(name, '@'); + if (atp == NULL) + error = EINVAL; + if (error == 0) + (void) strlcpy(dsname, name, atp - name + 1); + } + if (error == 0) + error = dsl_dataset_hold(dp, dsname, FTAG, &ds); + if (error == 0) { + error = dsl_dataset_snapshot_check_impl(ds, + atp + 1, tx); + dsl_dataset_rele(ds, FTAG); + } + + if (error != 0) { + if (ddsa->ddsa_errors != NULL) { + fnvlist_add_int32(ddsa->ddsa_errors, + name, error); + } + rv = error; + } + } + return (rv); +} + void -dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, +dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx) { + static zil_header_t zero_zil; + dsl_pool_t *dp = ds->ds_dir->dd_pool; dmu_buf_t *dbuf; dsl_dataset_phys_t *dsphys; uint64_t dsobj, crtxg; objset_t *mos = dp->dp_meta_objset; - int err; + objset_t *os; + + ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); + + /* + * If we are on an old pool, the zil must not be active, in which + * case it will be zeroed. Usually zil_suspend() accomplishes this. + */ + ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || + dmu_objset_from_ds(ds, &os) != 0 || + bcmp(&os->os_phys->os_zil_header, &zero_zil, + sizeof (zero_zil)) == 0); - ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); /* * The origin's ds_creation_txg has to be < TXG_INITIAL @@ -2070,7 +1057,7 @@ dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); - VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); + VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); @@ -2105,9 +1092,9 @@ dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, ds->ds_prev->ds_phys->ds_creation_txg); ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; } else if (next_clones_obj != 0) { - remove_from_next_clones(ds->ds_prev, + dsl_dataset_remove_from_next_clones(ds->ds_prev, dsphys->ds_next_snap_obj, tx); - VERIFY3U(0, ==, zap_add_int(mos, + VERIFY0(zap_add_int(mos, next_clones_obj, dsobj, tx)); } } @@ -2126,9 +1113,6 @@ dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, } dmu_buf_will_dirty(ds->ds_dbuf, tx); - zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu", - ds->ds_dir->dd_myname, snapname, dsobj, - ds->ds_phys->ds_prev_snap_txg); ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); dsl_deadlist_close(&ds->ds_deadlist); @@ -2143,13 +1127,12 @@ dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; - err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, - snapname, 8, 1, &dsobj, tx); - ASSERT(err == 0); + VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, + snapname, 8, 1, &dsobj, tx)); if (ds->ds_prev) - dsl_dataset_drop_ref(ds->ds_prev, ds); - VERIFY(0 == dsl_dataset_get_ref(dp, + dsl_dataset_rele(ds->ds_prev, ds); + VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); dsl_scan_ds_snapshotted(ds, tx); @@ -2159,6 +1142,198 @@ dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); } +static void +dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_snapshot_arg_t *ddsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + + for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); + pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { + dsl_dataset_t *ds; + char *name, *atp; + char dsname[MAXNAMELEN]; + + name = nvpair_name(pair); + atp = strchr(name, '@'); + (void) strlcpy(dsname, name, atp - name + 1); + VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); + + dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); + if (ddsa->ddsa_props != NULL) { + dsl_props_set_sync_impl(ds->ds_prev, + ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); + } + dsl_dataset_rele(ds, FTAG); + } +} + +/* + * The snapshots must all be in the same pool. + * All-or-nothing: if there are any failures, nothing will be modified. + */ +int +dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) +{ + dsl_dataset_snapshot_arg_t ddsa; + nvpair_t *pair; + boolean_t needsuspend; + int error; + spa_t *spa; + char *firstname; + nvlist_t *suspended = NULL; + + pair = nvlist_next_nvpair(snaps, NULL); + if (pair == NULL) + return (0); + firstname = nvpair_name(pair); + + error = spa_open(firstname, &spa, FTAG); + if (error != 0) + return (error); + needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); + spa_close(spa, FTAG); + + if (needsuspend) { + suspended = fnvlist_alloc(); + for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; + pair = nvlist_next_nvpair(snaps, pair)) { + char fsname[MAXNAMELEN]; + char *snapname = nvpair_name(pair); + char *atp; + void *cookie; + + atp = strchr(snapname, '@'); + if (atp == NULL) { + error = EINVAL; + break; + } + (void) strlcpy(fsname, snapname, atp - snapname + 1); + + error = zil_suspend(fsname, &cookie); + if (error != 0) + break; + fnvlist_add_uint64(suspended, fsname, + (uintptr_t)cookie); + } + } + + ddsa.ddsa_snaps = snaps; + ddsa.ddsa_props = props; + ddsa.ddsa_errors = errors; + + if (error == 0) { + error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, + dsl_dataset_snapshot_sync, &ddsa, + fnvlist_num_pairs(snaps) * 3); + } + + if (suspended != NULL) { + for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; + pair = nvlist_next_nvpair(suspended, pair)) { + zil_resume((void *)(uintptr_t) + fnvpair_value_uint64(pair)); + } + fnvlist_free(suspended); + } + + return (error); +} + +typedef struct dsl_dataset_snapshot_tmp_arg { + const char *ddsta_fsname; + const char *ddsta_snapname; + minor_t ddsta_cleanup_minor; + const char *ddsta_htag; +} dsl_dataset_snapshot_tmp_arg_t; + +static int +dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + + error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); + if (error != 0) + return (error); + + error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } + + if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { + dsl_dataset_rele(ds, FTAG); + return (ENOTSUP); + } + error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, + B_TRUE, tx); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } + + dsl_dataset_rele(ds, FTAG); + return (0); +} + +static void +dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); + + dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); + dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, + ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); + dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); + + dsl_dataset_rele(ds, FTAG); +} + +int +dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, + minor_t cleanup_minor, const char *htag) +{ + dsl_dataset_snapshot_tmp_arg_t ddsta; + int error; + spa_t *spa; + boolean_t needsuspend; + void *cookie; + + ddsta.ddsta_fsname = fsname; + ddsta.ddsta_snapname = snapname; + ddsta.ddsta_cleanup_minor = cleanup_minor; + ddsta.ddsta_htag = htag; + + error = spa_open(fsname, &spa, FTAG); + if (error != 0) + return (error); + needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); + spa_close(spa, FTAG); + + if (needsuspend) { + error = zil_suspend(fsname, &cookie); + if (error != 0) + return (error); + } + + error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, + dsl_dataset_snapshot_tmp_sync, &ddsta, 3); + + if (needsuspend) + zil_resume(cookie); + return (error); +} + + void dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) { @@ -2183,65 +1358,49 @@ get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; zap_cursor_t zc; zap_attribute_t za; - nvlist_t *propval; - nvlist_t *val; + nvlist_t *propval = fnvlist_alloc(); + nvlist_t *val = fnvlist_alloc(); - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); - VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); - VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0); + ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); /* - * There may me missing entries in ds_next_clones_obj + * There may be missing entries in ds_next_clones_obj * due to a bug in a previous version of the code. * Only trust it if it has the right number of entries. */ if (ds->ds_phys->ds_next_clones_obj != 0) { - ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, + ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, &count)); } - if (count != ds->ds_phys->ds_num_children - 1) { + if (count != ds->ds_phys->ds_num_children - 1) goto fail; - } for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { dsl_dataset_t *clone; char buf[ZFS_MAXNAMELEN]; - /* - * Even though we hold the dp_config_rwlock, the dataset - * may fail to open, returning ENOENT. If there is a - * thread concurrently attempting to destroy this - * dataset, it will have the ds_rwlock held for - * RW_WRITER. Our call to dsl_dataset_hold_obj() -> - * dsl_dataset_hold_ref() will fail its - * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the - * dp_config_rwlock, and wait for the destroy progress - * and signal ds_exclusive_cv. If the destroy was - * successful, we will see that - * DSL_DATASET_IS_DESTROYED(), and return ENOENT. - */ - if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool, - za.za_first_integer, FTAG, &clone) != 0) - continue; + VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, + za.za_first_integer, FTAG, &clone)); dsl_dir_name(clone->ds_dir, buf); - VERIFY(nvlist_add_boolean(val, buf) == 0); + fnvlist_add_boolean(val, buf); dsl_dataset_rele(clone, FTAG); } zap_cursor_fini(&zc); - VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0); - VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), - propval) == 0); + fnvlist_add_nvlist(propval, ZPROP_VALUE, val); + fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); fail: nvlist_free(val); nvlist_free(propval); - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); } void dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) { + dsl_pool_t *dp = ds->ds_dir->dd_pool; uint64_t refd, avail, uobjs, aobjs, ratio; + ASSERT(dsl_pool_config_held(dp)); + ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : (ds->ds_phys->ds_uncompressed_bytes * 100 / ds->ds_phys->ds_compressed_bytes); @@ -2287,10 +1446,8 @@ dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_dataset_t *prev; - rw_enter(&dp->dp_config_rwlock, RW_READER); int err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); - rw_exit(&dp->dp_config_rwlock); if (err == 0) { err = dsl_dataset_space_written(prev, ds, &written, &comp, &uncomp); @@ -2306,6 +1463,9 @@ dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) { + dsl_pool_t *dp = ds->ds_dir->dd_pool; + ASSERT(dsl_pool_config_held(dp)); + stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; stat->dds_guid = ds->ds_phys->ds_guid; @@ -2317,16 +1477,14 @@ dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) stat->dds_is_snapshot = B_FALSE; stat->dds_num_clones = 0; - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); if (dsl_dir_is_clone(ds->ds_dir)) { dsl_dataset_t *ods; - VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, + VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); dsl_dataset_name(ods, stat->dds_origin); - dsl_dataset_drop_ref(ods, FTAG); + dsl_dataset_rele(ods, FTAG); } - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); } } @@ -2364,8 +1522,7 @@ dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) { dsl_pool_t *dp = ds->ds_dir->dd_pool; - ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || - dsl_pool_sync_context(dp)); + ASSERT(dsl_pool_config_held(dp)); if (ds->ds_prev == NULL) return (B_FALSE); if (ds->ds_phys->ds_bp.blk_birth > @@ -2387,237 +1544,225 @@ dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) return (B_FALSE); } +typedef struct dsl_dataset_rename_snapshot_arg { + const char *ddrsa_fsname; + const char *ddrsa_oldsnapname; + const char *ddrsa_newsnapname; + boolean_t ddrsa_recursive; + dmu_tx_t *ddrsa_tx; +} dsl_dataset_rename_snapshot_arg_t; + /* ARGSUSED */ static int -dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, + dsl_dataset_t *hds, void *arg) { - dsl_dataset_t *ds = arg1; - char *newsnapname = arg2; - dsl_dir_t *dd = ds->ds_dir; - dsl_dataset_t *hds; + dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; + int error; uint64_t val; - int err; - err = dsl_dataset_hold_obj(dd->dd_pool, - dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); - if (err) - return (err); + error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); + if (error != 0) { + /* ignore nonexistent snapshots */ + return (error == ENOENT ? 0 : error); + } - /* new name better not be in use */ - err = dsl_dataset_snap_lookup(hds, newsnapname, &val); - dsl_dataset_rele(hds, FTAG); - - if (err == 0) - err = EEXIST; - else if (err == ENOENT) - err = 0; + /* new name should not exist */ + error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); + if (error == 0) + error = EEXIST; + else if (error == ENOENT) + error = 0; /* dataset name + 1 for the "@" + the new snapshot name must fit */ - if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) - err = ENAMETOOLONG; + if (dsl_dir_namelen(hds->ds_dir) + 1 + + strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) + error = ENAMETOOLONG; - return (err); + return (error); +} + +static int +dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *hds; + int error; + + error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); + if (error != 0) + return (error); + + if (ddrsa->ddrsa_recursive) { + error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, + dsl_dataset_rename_snapshot_check_impl, ddrsa, + DS_FIND_CHILDREN); + } else { + error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); + } + dsl_dataset_rele(hds, FTAG); + return (error); +} + +static int +dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, + dsl_dataset_t *hds, void *arg) +{ + dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; + dsl_dataset_t *ds; + uint64_t val; + dmu_tx_t *tx = ddrsa->ddrsa_tx; + int error; + + error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); + ASSERT(error == 0 || error == ENOENT); + if (error == ENOENT) { + /* ignore nonexistent snapshots */ + return (0); + } + + VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); + + /* log before we change the name */ + spa_history_log_internal_ds(ds, "rename", tx, + "-> @%s", ddrsa->ddrsa_newsnapname); + + VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx)); + mutex_enter(&ds->ds_lock); + (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); + mutex_exit(&ds->ds_lock); + VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, + ds->ds_snapname, 8, 1, &ds->ds_object, tx)); + + dsl_dataset_rele(ds, FTAG); + return (0); } static void -dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - const char *newsnapname = arg2; - dsl_dir_t *dd = ds->ds_dir; - objset_t *mos = dd->dd_pool->dp_meta_objset; + dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; - int err; - ASSERT(ds->ds_phys->ds_next_snap_obj != 0); - - VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, - dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); - - VERIFY(0 == dsl_dataset_get_snapname(ds)); - err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); - ASSERT0(err); - mutex_enter(&ds->ds_lock); - (void) strcpy(ds->ds_snapname, newsnapname); - mutex_exit(&ds->ds_lock); - err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, - ds->ds_snapname, 8, 1, &ds->ds_object, tx); - ASSERT0(err); - - spa_history_log_internal_ds(ds, "rename", tx, - "-> @%s", newsnapname); + VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); + ddrsa->ddrsa_tx = tx; + if (ddrsa->ddrsa_recursive) { + VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, + dsl_dataset_rename_snapshot_sync_impl, ddrsa, + DS_FIND_CHILDREN)); + } else { + VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); + } dsl_dataset_rele(hds, FTAG); } -struct renamesnaparg { - dsl_sync_task_group_t *dstg; - char failed[MAXPATHLEN]; - char *oldsnap; - char *newsnap; -}; - -static int -dsl_snapshot_rename_one(const char *name, void *arg) -{ - struct renamesnaparg *ra = arg; - dsl_dataset_t *ds = NULL; - char *snapname; - int err; - - snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); - (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); - - /* - * For recursive snapshot renames the parent won't be changing - * so we just pass name for both the to/from argument. - */ - err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); - if (err != 0) { - strfree(snapname); - return (err == ENOENT ? 0 : err); - } - -#ifdef _KERNEL - /* - * For all filesystems undergoing rename, we'll need to unmount it. - */ - (void) zfs_unmount_snap(snapname, NULL); -#endif - err = dsl_dataset_hold(snapname, ra->dstg, &ds); - strfree(snapname); - if (err != 0) - return (err == ENOENT ? 0 : err); - - dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, - dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); - - return (0); -} - -static int -dsl_recursive_rename(char *oldname, const char *newname) -{ - int err; - struct renamesnaparg *ra; - dsl_sync_task_t *dst; - spa_t *spa; - char *cp, *fsname = spa_strdup(oldname); - int len = strlen(oldname) + 1; - - /* truncate the snapshot name to get the fsname */ - cp = strchr(fsname, '@'); - *cp = '\0'; - - err = spa_open(fsname, &spa, FTAG); - if (err) { - kmem_free(fsname, len); - return (err); - } - ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); - ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - - ra->oldsnap = strchr(oldname, '@') + 1; - ra->newsnap = strchr(newname, '@') + 1; - *ra->failed = '\0'; - - err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, - DS_FIND_CHILDREN); - kmem_free(fsname, len); - - if (err == 0) { - err = dsl_sync_task_group_wait(ra->dstg); - } - - for (dst = list_head(&ra->dstg->dstg_tasks); dst; - dst = list_next(&ra->dstg->dstg_tasks, dst)) { - dsl_dataset_t *ds = dst->dst_arg1; - if (dst->dst_err) { - dsl_dir_name(ds->ds_dir, ra->failed); - (void) strlcat(ra->failed, "@", sizeof (ra->failed)); - (void) strlcat(ra->failed, ra->newsnap, - sizeof (ra->failed)); - } - dsl_dataset_rele(ds, ra->dstg); - } - - if (err) - (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); - - dsl_sync_task_group_destroy(ra->dstg); - kmem_free(ra, sizeof (struct renamesnaparg)); - spa_close(spa, FTAG); - return (err); -} - -static int -dsl_valid_rename(const char *oldname, void *arg) -{ - int delta = *(int *)arg; - - if (strlen(oldname) + delta >= MAXNAMELEN) - return (ENAMETOOLONG); - - return (0); -} - -#pragma weak dmu_objset_rename = dsl_dataset_rename int -dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) +dsl_dataset_rename_snapshot(const char *fsname, + const char *oldsnapname, const char *newsnapname, boolean_t recursive) { - dsl_dir_t *dd; + dsl_dataset_rename_snapshot_arg_t ddrsa; + + ddrsa.ddrsa_fsname = fsname; + ddrsa.ddrsa_oldsnapname = oldsnapname; + ddrsa.ddrsa_newsnapname = newsnapname; + ddrsa.ddrsa_recursive = recursive; + + return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, + dsl_dataset_rename_snapshot_sync, &ddrsa, 1)); +} + +static int +dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) +{ + const char *fsname = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; - const char *tail; - int err; + int64_t unused_refres_delta; + int error; - err = dsl_dir_open(oldname, FTAG, &dd, &tail); - if (err) - return (err); - - if (tail == NULL) { - int delta = strlen(newname) - strlen(oldname); - - /* if we're growing, validate child name lengths */ - if (delta > 0) - err = dmu_objset_find(oldname, dsl_valid_rename, - &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); - - if (err == 0) - err = dsl_dir_rename(dd, newname); - dsl_dir_close(dd, FTAG); - return (err); - } - - if (tail[0] != '@') { - /* the name ended in a nonexistent component */ - dsl_dir_close(dd, FTAG); - return (ENOENT); - } - - dsl_dir_close(dd, FTAG); - - /* new name must be snapshot in same filesystem */ - tail = strchr(newname, '@'); - if (tail == NULL) - return (EINVAL); - tail++; - if (strncmp(oldname, newname, tail - newname) != 0) - return (EXDEV); - - if (recursive) { - err = dsl_recursive_rename(oldname, newname); - } else { - err = dsl_dataset_hold(oldname, FTAG, &ds); - if (err) - return (err); - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - dsl_dataset_snapshot_rename_check, - dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); + error = dsl_dataset_hold(dp, fsname, FTAG, &ds); + if (error != 0) + return (error); + /* must not be a snapshot */ + if (dsl_dataset_is_snapshot(ds)) { dsl_dataset_rele(ds, FTAG); + return (EINVAL); } - return (err); + /* must have a most recent snapshot */ + if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { + dsl_dataset_rele(ds, FTAG); + return (EINVAL); + } + + if (dsl_dataset_long_held(ds)) { + dsl_dataset_rele(ds, FTAG); + return (EBUSY); + } + + /* + * Check if the snap we are rolling back to uses more than + * the refquota. + */ + if (ds->ds_quota != 0 && + ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) { + dsl_dataset_rele(ds, FTAG); + return (EDQUOT); + } + + /* + * When we do the clone swap, we will temporarily use more space + * due to the refreservation (the head will no longer have any + * unique space, so the entire amount of the refreservation will need + * to be free). We will immediately destroy the clone, freeing + * this space, but the freeing happens over many txg's. + */ + unused_refres_delta = (int64_t)MIN(ds->ds_reserved, + ds->ds_phys->ds_unique_bytes); + + if (unused_refres_delta > 0 && + unused_refres_delta > + dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { + dsl_dataset_rele(ds, FTAG); + return (ENOSPC); + } + + dsl_dataset_rele(ds, FTAG); + return (0); +} + +static void +dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) +{ + const char *fsname = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds, *clone; + uint64_t cloneobj; + + VERIFY0(dsl_dataset_hold(dp, fsname, FTAG, &ds)); + + cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", + ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); + + VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); + + dsl_dataset_clone_swap_sync_impl(clone, ds, tx); + dsl_dataset_zero_zil(ds, tx); + + dsl_destroy_head_sync_impl(clone, tx); + + dsl_dataset_rele(clone, FTAG); + dsl_dataset_rele(ds, FTAG); +} + +int +dsl_dataset_rollback(const char *fsname) +{ + return (dsl_sync_task(fsname, dsl_dataset_rollback_check, + dsl_dataset_rollback_sync, (void *)fsname, 1)); } struct promotenode { @@ -2625,49 +1770,66 @@ struct promotenode { dsl_dataset_t *ds; }; -struct promotearg { +typedef struct dsl_dataset_promote_arg { + const char *ddpa_clonename; + dsl_dataset_t *ddpa_clone; list_t shared_snaps, origin_snaps, clone_snaps; - dsl_dataset_t *origin_origin; + dsl_dataset_t *origin_origin; /* origin of the origin */ uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; char *err_ds; -}; +} dsl_dataset_promote_arg_t; static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); -static boolean_t snaplist_unstable(list_t *l); +static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, + void *tag); +static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); static int -dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *hds = arg1; - struct promotearg *pa = arg2; - struct promotenode *snap = list_head(&pa->shared_snaps); - dsl_dataset_t *origin_ds = snap->ds; + dsl_dataset_promote_arg_t *ddpa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *hds; + struct promotenode *snap; + dsl_dataset_t *origin_ds; int err; uint64_t unused; - /* Check that it is a real clone */ - if (!dsl_dir_is_clone(hds->ds_dir)) - return (EINVAL); + err = promote_hold(ddpa, dp, FTAG); + if (err != 0) + return (err); - /* Since this is so expensive, don't do the preliminary check */ - if (!dmu_tx_is_syncing(tx)) - return (0); + hds = ddpa->ddpa_clone; - if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) + if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { + promote_rele(ddpa, FTAG); return (EXDEV); + } + + /* + * Compute and check the amount of space to transfer. Since this is + * so expensive, don't do the preliminary check. + */ + if (!dmu_tx_is_syncing(tx)) { + promote_rele(ddpa, FTAG); + return (0); + } + + snap = list_head(&ddpa->shared_snaps); + origin_ds = snap->ds; /* compute origin's new unique space */ - snap = list_tail(&pa->clone_snaps); + snap = list_tail(&ddpa->clone_snaps); ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); dsl_deadlist_space_range(&snap->ds->ds_deadlist, origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, - &pa->unique, &unused, &unused); + &ddpa->unique, &unused, &unused); /* * Walk the snapshots that we are moving * * Compute space to transfer. Consider the incremental changes - * to used for each snapshot: + * to used by each snapshot: * (my used) = (prev's used) + (blocks born) - (blocks killed) * So each snapshot gave birth to: * (blocks born) = (my used) - (prev's used) + (blocks killed) @@ -2678,18 +1840,28 @@ dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) * Note however, if we stop before we reach the ORIGIN we get: * uN + kN + kN-1 + ... + kM - uM-1 */ - pa->used = origin_ds->ds_phys->ds_referenced_bytes; - pa->comp = origin_ds->ds_phys->ds_compressed_bytes; - pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; - for (snap = list_head(&pa->shared_snaps); snap; - snap = list_next(&pa->shared_snaps, snap)) { + ddpa->used = origin_ds->ds_phys->ds_referenced_bytes; + ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes; + ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; + for (snap = list_head(&ddpa->shared_snaps); snap; + snap = list_next(&ddpa->shared_snaps, snap)) { uint64_t val, dlused, dlcomp, dluncomp; dsl_dataset_t *ds = snap->ds; + /* + * If there are long holds, we won't be able to evict + * the objset. + */ + if (dsl_dataset_long_held(ds)) { + err = EBUSY; + goto out; + } + /* Check that the snapshot name does not conflict */ - VERIFY(0 == dsl_dataset_get_snapname(ds)); + VERIFY0(dsl_dataset_get_snapname(ds)); err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); if (err == 0) { + (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); err = EEXIST; goto out; } @@ -2702,26 +1874,27 @@ dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); - pa->used += dlused; - pa->comp += dlcomp; - pa->uncomp += dluncomp; + ddpa->used += dlused; + ddpa->comp += dlcomp; + ddpa->uncomp += dluncomp; } /* * If we are a clone of a clone then we never reached ORIGIN, * so we need to subtract out the clone origin's used space. */ - if (pa->origin_origin) { - pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes; - pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; - pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; + if (ddpa->origin_origin) { + ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes; + ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes; + ddpa->uncomp -= + ddpa->origin_origin->ds_phys->ds_uncompressed_bytes; } /* Check that there is enough space here */ err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, - pa->used); - if (err) - return (err); + ddpa->used); + if (err != 0) + goto out; /* * Compute the amounts of space that will be used by snapshots @@ -2739,68 +1912,75 @@ dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) * calls will be fast because they do not have to * iterate over all bps. */ - snap = list_head(&pa->origin_snaps); - err = snaplist_space(&pa->shared_snaps, - snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap); - if (err) - return (err); + snap = list_head(&ddpa->origin_snaps); + err = snaplist_space(&ddpa->shared_snaps, + snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); + if (err != 0) + goto out; - err = snaplist_space(&pa->clone_snaps, + err = snaplist_space(&ddpa->clone_snaps, snap->ds->ds_dir->dd_origin_txg, &space); - if (err) - return (err); - pa->cloneusedsnap += space; + if (err != 0) + goto out; + ddpa->cloneusedsnap += space; } if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { - err = snaplist_space(&pa->origin_snaps, - origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); - if (err) - return (err); + err = snaplist_space(&ddpa->origin_snaps, + origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap); + if (err != 0) + goto out; } - return (0); out: - pa->err_ds = snap->ds->ds_snapname; + promote_rele(ddpa, FTAG); return (err); } static void -dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *hds = arg1; - struct promotearg *pa = arg2; - struct promotenode *snap = list_head(&pa->shared_snaps); - dsl_dataset_t *origin_ds = snap->ds; + dsl_dataset_promote_arg_t *ddpa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *hds; + struct promotenode *snap; + dsl_dataset_t *origin_ds; dsl_dataset_t *origin_head; - dsl_dir_t *dd = hds->ds_dir; - dsl_pool_t *dp = hds->ds_dir->dd_pool; + dsl_dir_t *dd; dsl_dir_t *odd = NULL; uint64_t oldnext_obj; int64_t delta; - ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); + VERIFY0(promote_hold(ddpa, dp, FTAG)); + hds = ddpa->ddpa_clone; - snap = list_head(&pa->origin_snaps); + ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE); + + snap = list_head(&ddpa->shared_snaps); + origin_ds = snap->ds; + dd = hds->ds_dir; + + snap = list_head(&ddpa->origin_snaps); origin_head = snap->ds; /* * We need to explicitly open odd, since origin_ds's dd will be * changing. */ - VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, + VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, NULL, FTAG, &odd)); /* change origin's next snap */ dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; - snap = list_tail(&pa->clone_snaps); + snap = list_tail(&ddpa->clone_snaps); ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; /* change the origin's next clone */ if (origin_ds->ds_phys->ds_next_clones_obj) { - remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); - VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, + dsl_dataset_remove_from_next_clones(origin_ds, + snap->ds->ds_object, tx); + VERIFY0(zap_add_int(dp->dp_meta_objset, origin_ds->ds_phys->ds_next_clones_obj, oldnext_obj, tx)); } @@ -2817,39 +1997,43 @@ dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) /* change dd_clone entries */ if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { - VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, + VERIFY0(zap_remove_int(dp->dp_meta_objset, odd->dd_phys->dd_clones, hds->ds_object, tx)); - VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, - pa->origin_origin->ds_dir->dd_phys->dd_clones, + VERIFY0(zap_add_int(dp->dp_meta_objset, + ddpa->origin_origin->ds_dir->dd_phys->dd_clones, hds->ds_object, tx)); - VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, - pa->origin_origin->ds_dir->dd_phys->dd_clones, + VERIFY0(zap_remove_int(dp->dp_meta_objset, + ddpa->origin_origin->ds_dir->dd_phys->dd_clones, origin_head->ds_object, tx)); if (dd->dd_phys->dd_clones == 0) { dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } - VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, + VERIFY0(zap_add_int(dp->dp_meta_objset, dd->dd_phys->dd_clones, origin_head->ds_object, tx)); - } /* move snapshots to this dir */ - for (snap = list_head(&pa->shared_snaps); snap; - snap = list_next(&pa->shared_snaps, snap)) { + for (snap = list_head(&ddpa->shared_snaps); snap; + snap = list_next(&ddpa->shared_snaps, snap)) { dsl_dataset_t *ds = snap->ds; - /* unregister props as dsl_dir is changing */ + /* + * Property callbacks are registered to a particular + * dsl_dir. Since ours is changing, evict the objset + * so that they will be unregistered from the old dsl_dir. + */ if (ds->ds_objset) { dmu_objset_evict(ds->ds_objset); ds->ds_objset = NULL; } + /* move snap name entry */ - VERIFY(0 == dsl_dataset_get_snapname(ds)); - VERIFY(0 == dsl_dataset_snap_remove(origin_head, + VERIFY0(dsl_dataset_get_snapname(ds)); + VERIFY0(dsl_dataset_snap_remove(origin_head, ds->ds_snapname, tx)); - VERIFY(0 == zap_add(dp->dp_meta_objset, + VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 8, 1, &ds->ds_object, tx)); @@ -2858,8 +2042,8 @@ dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); ds->ds_phys->ds_dir_obj = dd->dd_object; ASSERT3P(ds->ds_dir, ==, odd); - dsl_dir_close(ds->ds_dir, ds); - VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, + dsl_dir_rele(ds->ds_dir, ds); + VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, NULL, ds, &ds->ds_dir)); /* move any clone references */ @@ -2883,20 +2067,20 @@ dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) continue; } - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, + VERIFY0(dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &cnds)); o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; - VERIFY3U(zap_remove_int(dp->dp_meta_objset, - odd->dd_phys->dd_clones, o, tx), ==, 0); - VERIFY3U(zap_add_int(dp->dp_meta_objset, - dd->dd_phys->dd_clones, o, tx), ==, 0); + VERIFY0(zap_remove_int(dp->dp_meta_objset, + odd->dd_phys->dd_clones, o, tx)); + VERIFY0(zap_add_int(dp->dp_meta_objset, + dd->dd_phys->dd_clones, o, tx)); dsl_dataset_rele(cnds, FTAG); } zap_cursor_fini(&zc); } - ASSERT0(dsl_prop_numcb(ds)); + ASSERT(!dsl_prop_hascb(ds)); } /* @@ -2906,31 +2090,31 @@ dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) * is true for each of {clone,origin} independently. */ - delta = pa->cloneusedsnap - + delta = ddpa->cloneusedsnap - dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, >=, 0); - ASSERT3U(pa->used, >=, delta); + ASSERT3U(ddpa->used, >=, delta); dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(dd, DD_USED_HEAD, - pa->used - delta, pa->comp, pa->uncomp, tx); + ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); - delta = pa->originusedsnap - + delta = ddpa->originusedsnap - odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, <=, 0); - ASSERT3U(pa->used, >=, -delta); + ASSERT3U(ddpa->used, >=, -delta); dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(odd, DD_USED_HEAD, - -pa->used - delta, -pa->comp, -pa->uncomp, tx); + -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); - origin_ds->ds_phys->ds_unique_bytes = pa->unique; + origin_ds->ds_phys->ds_unique_bytes = ddpa->unique; /* log history record */ spa_history_log_internal_ds(hds, "promote", tx, ""); - dsl_dir_close(odd, FTAG); + dsl_dir_rele(odd, FTAG); + promote_rele(ddpa, FTAG); } -static char *snaplist_tag = "snaplist"; /* * Make a list of dsl_dataset_t's for the snapshots between first_obj * (exclusive) and last_obj (inclusive). The list will be in reverse @@ -2938,13 +2122,11 @@ static char *snaplist_tag = "snaplist"; * snapshots back to this dataset's origin. */ static int -snaplist_make(dsl_pool_t *dp, boolean_t own, - uint64_t first_obj, uint64_t last_obj, list_t *l) +snaplist_make(dsl_pool_t *dp, + uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) { uint64_t obj = last_obj; - ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); - list_create(l, sizeof (struct promotenode), offsetof(struct promotenode, link)); @@ -2953,28 +2135,15 @@ snaplist_make(dsl_pool_t *dp, boolean_t own, struct promotenode *snap; int err; - if (own) { - err = dsl_dataset_own_obj(dp, obj, - 0, snaplist_tag, &ds); - if (err == 0) - dsl_dataset_make_exclusive(ds, snaplist_tag); - } else { - err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); - } - if (err == ENOENT) { - /* lost race with snapshot destroy */ - struct promotenode *last = list_tail(l); - ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); - obj = last->ds->ds_phys->ds_prev_snap_obj; - continue; - } else if (err) { + err = dsl_dataset_hold_obj(dp, obj, tag, &ds); + ASSERT(err != ENOENT); + if (err != 0) return (err); - } if (first_obj == 0) first_obj = ds->ds_dir->dd_phys->dd_origin_obj; - snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); + snap = kmem_alloc(sizeof (*snap), KM_SLEEP); snap->ds = ds; list_insert_tail(l, snap); obj = ds->ds_phys->ds_prev_snap_obj; @@ -2999,208 +2168,209 @@ snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) } static void -snaplist_destroy(list_t *l, boolean_t own) +snaplist_destroy(list_t *l, void *tag) { struct promotenode *snap; - if (!l || !list_link_active(&l->list_head)) + if (l == NULL || !list_link_active(&l->list_head)) return; while ((snap = list_tail(l)) != NULL) { list_remove(l, snap); - if (own) - dsl_dataset_disown(snap->ds, snaplist_tag); - else - dsl_dataset_rele(snap->ds, snaplist_tag); - kmem_free(snap, sizeof (struct promotenode)); + dsl_dataset_rele(snap->ds, tag); + kmem_free(snap, sizeof (*snap)); } list_destroy(l); } +static int +promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) +{ + int error; + dsl_dir_t *dd; + struct promotenode *snap; + + error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, + &ddpa->ddpa_clone); + if (error != 0) + return (error); + dd = ddpa->ddpa_clone->ds_dir; + + if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || + !dsl_dir_is_clone(dd)) { + dsl_dataset_rele(ddpa->ddpa_clone, tag); + return (EINVAL); + } + + error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj, + &ddpa->shared_snaps, tag); + if (error != 0) + goto out; + + error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, + &ddpa->clone_snaps, tag); + if (error != 0) + goto out; + + snap = list_head(&ddpa->shared_snaps); + ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); + error = snaplist_make(dp, dd->dd_phys->dd_origin_obj, + snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, + &ddpa->origin_snaps, tag); + if (error != 0) + goto out; + + if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { + error = dsl_dataset_hold_obj(dp, + snap->ds->ds_dir->dd_phys->dd_origin_obj, + tag, &ddpa->origin_origin); + if (error != 0) + goto out; + } +out: + if (error != 0) + promote_rele(ddpa, tag); + return (error); +} + +static void +promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) +{ + snaplist_destroy(&ddpa->shared_snaps, tag); + snaplist_destroy(&ddpa->clone_snaps, tag); + snaplist_destroy(&ddpa->origin_snaps, tag); + if (ddpa->origin_origin != NULL) + dsl_dataset_rele(ddpa->origin_origin, tag); + dsl_dataset_rele(ddpa->ddpa_clone, tag); +} + /* - * Promote a clone. Nomenclature note: - * "clone" or "cds": the original clone which is being promoted - * "origin" or "ods": the snapshot which is originally clone's origin - * "origin head" or "ohds": the dataset which is the head - * (filesystem/volume) for the origin - * "origin origin": the origin of the origin's filesystem (typically - * NULL, indicating that the clone is not a clone of a clone). + * Promote a clone. + * + * If it fails due to a conflicting snapshot name, "conflsnap" will be filled + * in with the name. (It must be at least MAXNAMELEN bytes long.) */ int dsl_dataset_promote(const char *name, char *conflsnap) { - dsl_dataset_t *ds; - dsl_dir_t *dd; - dsl_pool_t *dp; - dmu_object_info_t doi; - struct promotearg pa = { 0 }; - struct promotenode *snap; - int err; - - err = dsl_dataset_hold(name, FTAG, &ds); - if (err) - return (err); - dd = ds->ds_dir; - dp = dd->dd_pool; - - err = dmu_object_info(dp->dp_meta_objset, - ds->ds_phys->ds_snapnames_zapobj, &doi); - if (err) { - dsl_dataset_rele(ds, FTAG); - return (err); - } - - if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { - dsl_dataset_rele(ds, FTAG); - return (EINVAL); - } + dsl_dataset_promote_arg_t ddpa = { 0 }; + uint64_t numsnaps; + int error; + objset_t *os; /* - * We are going to inherit all the snapshots taken before our - * origin (i.e., our new origin will be our parent's origin). - * Take ownership of them so that we can rename them into our - * namespace. + * We will modify space proportional to the number of + * snapshots. Compute numsnaps. */ - rw_enter(&dp->dp_config_rwlock, RW_READER); + error = dmu_objset_hold(name, FTAG, &os); + if (error != 0) + return (error); + error = zap_count(dmu_objset_pool(os)->dp_meta_objset, + dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps); + dmu_objset_rele(os, FTAG); + if (error != 0) + return (error); - err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, - &pa.shared_snaps); - if (err != 0) - goto out; + ddpa.ddpa_clonename = name; + ddpa.err_ds = conflsnap; - err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); - if (err != 0) - goto out; - - snap = list_head(&pa.shared_snaps); - ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); - err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, - snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); - if (err != 0) - goto out; - - if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { - err = dsl_dataset_hold_obj(dp, - snap->ds->ds_dir->dd_phys->dd_origin_obj, - FTAG, &pa.origin_origin); - if (err != 0) - goto out; - } - -out: - rw_exit(&dp->dp_config_rwlock); - - /* - * Add in 128x the snapnames zapobj size, since we will be moving - * a bunch of snapnames to the promoted ds, and dirtying their - * bonus buffers. - */ - if (err == 0) { - err = dsl_sync_task_do(dp, dsl_dataset_promote_check, - dsl_dataset_promote_sync, ds, &pa, - 2 + 2 * doi.doi_physical_blocks_512); - if (err && pa.err_ds && conflsnap) - (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); - } - - snaplist_destroy(&pa.shared_snaps, B_TRUE); - snaplist_destroy(&pa.clone_snaps, B_FALSE); - snaplist_destroy(&pa.origin_snaps, B_FALSE); - if (pa.origin_origin) - dsl_dataset_rele(pa.origin_origin, FTAG); - dsl_dataset_rele(ds, FTAG); - return (err); + return (dsl_sync_task(name, dsl_dataset_promote_check, + dsl_dataset_promote_sync, &ddpa, 2 + numsnaps)); } -struct cloneswaparg { - dsl_dataset_t *cds; /* clone dataset */ - dsl_dataset_t *ohds; /* origin's head dataset */ - boolean_t force; - int64_t unused_refres_delta; /* change in unconsumed refreservation */ -}; - -/* ARGSUSED */ -static int -dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) +int +dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, + dsl_dataset_t *origin_head, boolean_t force) { - struct cloneswaparg *csa = arg1; + int64_t unused_refres_delta; /* they should both be heads */ - if (dsl_dataset_is_snapshot(csa->cds) || - dsl_dataset_is_snapshot(csa->ohds)) + if (dsl_dataset_is_snapshot(clone) || + dsl_dataset_is_snapshot(origin_head)) return (EINVAL); /* the branch point should be just before them */ - if (csa->cds->ds_prev != csa->ohds->ds_prev) + if (clone->ds_prev != origin_head->ds_prev) return (EINVAL); - /* cds should be the clone (unless they are unrelated) */ - if (csa->cds->ds_prev != NULL && - csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && - csa->ohds->ds_object != - csa->cds->ds_prev->ds_phys->ds_next_snap_obj) + /* clone should be the clone (unless they are unrelated) */ + if (clone->ds_prev != NULL && + clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && + origin_head->ds_object != + clone->ds_prev->ds_phys->ds_next_snap_obj) return (EINVAL); /* the clone should be a child of the origin */ - if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) + if (clone->ds_dir->dd_parent != origin_head->ds_dir) return (EINVAL); - /* ohds shouldn't be modified unless 'force' */ - if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) + /* origin_head shouldn't be modified unless 'force' */ + if (!force && dsl_dataset_modified_since_lastsnap(origin_head)) return (ETXTBSY); - /* adjust amount of any unconsumed refreservation */ - csa->unused_refres_delta = - (int64_t)MIN(csa->ohds->ds_reserved, - csa->ohds->ds_phys->ds_unique_bytes) - - (int64_t)MIN(csa->ohds->ds_reserved, - csa->cds->ds_phys->ds_unique_bytes); + /* origin_head should have no long holds (e.g. is not mounted) */ + if (dsl_dataset_long_held(origin_head)) + return (EBUSY); - if (csa->unused_refres_delta > 0 && - csa->unused_refres_delta > - dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) + /* check amount of any unconsumed refreservation */ + unused_refres_delta = + (int64_t)MIN(origin_head->ds_reserved, + origin_head->ds_phys->ds_unique_bytes) - + (int64_t)MIN(origin_head->ds_reserved, + clone->ds_phys->ds_unique_bytes); + + if (unused_refres_delta > 0 && + unused_refres_delta > + dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) return (ENOSPC); - if (csa->ohds->ds_quota != 0 && - csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) + /* clone can't be over the head's refquota */ + if (origin_head->ds_quota != 0 && + clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota) return (EDQUOT); return (0); } -/* ARGSUSED */ -static void -dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) +void +dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, + dsl_dataset_t *origin_head, dmu_tx_t *tx) { - struct cloneswaparg *csa = arg1; - dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; + dsl_pool_t *dp = dmu_tx_pool(tx); + int64_t unused_refres_delta; - ASSERT(csa->cds->ds_reserved == 0); - ASSERT(csa->ohds->ds_quota == 0 || - csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); + ASSERT(clone->ds_reserved == 0); + ASSERT(origin_head->ds_quota == 0 || + clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota); - dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); - dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); + dmu_buf_will_dirty(clone->ds_dbuf, tx); + dmu_buf_will_dirty(origin_head->ds_dbuf, tx); - if (csa->cds->ds_objset != NULL) { - dmu_objset_evict(csa->cds->ds_objset); - csa->cds->ds_objset = NULL; + if (clone->ds_objset != NULL) { + dmu_objset_evict(clone->ds_objset); + clone->ds_objset = NULL; } - if (csa->ohds->ds_objset != NULL) { - dmu_objset_evict(csa->ohds->ds_objset); - csa->ohds->ds_objset = NULL; + if (origin_head->ds_objset != NULL) { + dmu_objset_evict(origin_head->ds_objset); + origin_head->ds_objset = NULL; } + unused_refres_delta = + (int64_t)MIN(origin_head->ds_reserved, + origin_head->ds_phys->ds_unique_bytes) - + (int64_t)MIN(origin_head->ds_reserved, + clone->ds_phys->ds_unique_bytes); + /* * Reset origin's unique bytes, if it exists. */ - if (csa->cds->ds_prev) { - dsl_dataset_t *origin = csa->cds->ds_prev; + if (clone->ds_prev) { + dsl_dataset_t *origin = clone->ds_prev; uint64_t comp, uncomp; dmu_buf_will_dirty(origin->ds_dbuf, tx); - dsl_deadlist_space_range(&csa->cds->ds_deadlist, + dsl_deadlist_space_range(&clone->ds_deadlist, origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); } @@ -3208,9 +2378,9 @@ dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) /* swap blkptrs */ { blkptr_t tmp; - tmp = csa->ohds->ds_phys->ds_bp; - csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; - csa->cds->ds_phys->ds_bp = tmp; + tmp = origin_head->ds_phys->ds_bp; + origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp; + clone->ds_phys->ds_bp = tmp; } /* set dd_*_bytes */ @@ -3219,25 +2389,25 @@ dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) uint64_t cdl_used, cdl_comp, cdl_uncomp; uint64_t odl_used, odl_comp, odl_uncomp; - ASSERT3U(csa->cds->ds_dir->dd_phys-> + ASSERT3U(clone->ds_dir->dd_phys-> dd_used_breakdown[DD_USED_SNAP], ==, 0); - dsl_deadlist_space(&csa->cds->ds_deadlist, + dsl_deadlist_space(&clone->ds_deadlist, &cdl_used, &cdl_comp, &cdl_uncomp); - dsl_deadlist_space(&csa->ohds->ds_deadlist, + dsl_deadlist_space(&origin_head->ds_deadlist, &odl_used, &odl_comp, &odl_uncomp); - dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used - - (csa->ohds->ds_phys->ds_referenced_bytes + odl_used); - dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - - (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); - duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + + dused = clone->ds_phys->ds_referenced_bytes + cdl_used - + (origin_head->ds_phys->ds_referenced_bytes + odl_used); + dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp - + (origin_head->ds_phys->ds_compressed_bytes + odl_comp); + duncomp = clone->ds_phys->ds_uncompressed_bytes + cdl_uncomp - - (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); + (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp); - dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, + dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, dused, dcomp, duncomp, tx); - dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, + dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, -dused, -dcomp, -duncomp, tx); /* @@ -3246,86 +2416,46 @@ dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) * deadlist (since that's the only thing that's * changing that affects the snapused). */ - dsl_deadlist_space_range(&csa->cds->ds_deadlist, - csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, + dsl_deadlist_space_range(&clone->ds_deadlist, + origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &cdl_used, &cdl_comp, &cdl_uncomp); - dsl_deadlist_space_range(&csa->ohds->ds_deadlist, - csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, + dsl_deadlist_space_range(&origin_head->ds_deadlist, + origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &odl_used, &odl_comp, &odl_uncomp); - dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, + dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, DD_USED_HEAD, DD_USED_SNAP, tx); } /* swap ds_*_bytes */ - SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes, - csa->cds->ds_phys->ds_referenced_bytes); - SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, - csa->cds->ds_phys->ds_compressed_bytes); - SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, - csa->cds->ds_phys->ds_uncompressed_bytes); - SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, - csa->cds->ds_phys->ds_unique_bytes); + SWITCH64(origin_head->ds_phys->ds_referenced_bytes, + clone->ds_phys->ds_referenced_bytes); + SWITCH64(origin_head->ds_phys->ds_compressed_bytes, + clone->ds_phys->ds_compressed_bytes); + SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes, + clone->ds_phys->ds_uncompressed_bytes); + SWITCH64(origin_head->ds_phys->ds_unique_bytes, + clone->ds_phys->ds_unique_bytes); /* apply any parent delta for change in unconsumed refreservation */ - dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, - csa->unused_refres_delta, 0, 0, tx); + dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, + unused_refres_delta, 0, 0, tx); /* * Swap deadlists. */ - dsl_deadlist_close(&csa->cds->ds_deadlist); - dsl_deadlist_close(&csa->ohds->ds_deadlist); - SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, - csa->cds->ds_phys->ds_deadlist_obj); - dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, - csa->cds->ds_phys->ds_deadlist_obj); - dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, - csa->ohds->ds_phys->ds_deadlist_obj); + dsl_deadlist_close(&clone->ds_deadlist); + dsl_deadlist_close(&origin_head->ds_deadlist); + SWITCH64(origin_head->ds_phys->ds_deadlist_obj, + clone->ds_phys->ds_deadlist_obj); + dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, + clone->ds_phys->ds_deadlist_obj); + dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, + origin_head->ds_phys->ds_deadlist_obj); - dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx); + dsl_scan_ds_clone_swapped(origin_head, clone, tx); - spa_history_log_internal_ds(csa->cds, "clone swap", tx, - "parent=%s", csa->ohds->ds_dir->dd_myname); -} - -/* - * Swap 'clone' with its origin head datasets. Used at the end of "zfs - * recv" into an existing fs to swizzle the file system to the new - * version, and by "zfs rollback". Can also be used to swap two - * independent head datasets if neither has any snapshots. - */ -int -dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, - boolean_t force) -{ - struct cloneswaparg csa; - int error; - - ASSERT(clone->ds_owner); - ASSERT(origin_head->ds_owner); -retry: - /* - * Need exclusive access for the swap. If we're swapping these - * datasets back after an error, we already hold the locks. - */ - if (!RW_WRITE_HELD(&clone->ds_rwlock)) - rw_enter(&clone->ds_rwlock, RW_WRITER); - if (!RW_WRITE_HELD(&origin_head->ds_rwlock) && - !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { - rw_exit(&clone->ds_rwlock); - rw_enter(&origin_head->ds_rwlock, RW_WRITER); - if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { - rw_exit(&origin_head->ds_rwlock); - goto retry; - } - } - csa.cds = clone; - csa.ohds = origin_head; - csa.force = force; - error = dsl_sync_task_do(clone->ds_dir->dd_pool, - dsl_dataset_clone_swap_check, - dsl_dataset_clone_swap_sync, &csa, NULL, 9); - return (error); + spa_history_log_internal_ds(clone, "clone swap", tx, + "parent=%s", origin_head->ds_dir->dd_myname); } /* @@ -3335,21 +2465,20 @@ dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) { - spa_t *spa; dsl_pool_t *dp; dsl_dataset_t *ds; int error; - if ((error = spa_open(pname, &spa, FTAG)) != 0) + error = dsl_pool_hold(pname, FTAG, &dp); + if (error != 0) return (error); - dp = spa_get_dsl(spa); - rw_enter(&dp->dp_config_rwlock, RW_READER); - if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { + + error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); + if (error == 0) { dsl_dataset_name(ds, buf); dsl_dataset_rele(ds, FTAG); } - rw_exit(&dp->dp_config_rwlock); - spa_close(spa, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } @@ -3402,102 +2531,134 @@ dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, return (error); } +typedef struct dsl_dataset_set_qr_arg { + const char *ddsqra_name; + zprop_source_t ddsqra_source; + uint64_t ddsqra_value; +} dsl_dataset_set_qr_arg_t; + + /* ARGSUSED */ static int -dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_prop_setarg_t *psa = arg2; - int err; + dsl_dataset_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + uint64_t newval; - if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) + if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) return (ENOTSUP); - if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) - return (err); + error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); + if (error != 0) + return (error); - if (psa->psa_effective_value == 0) + if (dsl_dataset_is_snapshot(ds)) { + dsl_dataset_rele(ds, FTAG); + return (EINVAL); + } + + error = dsl_prop_predict(ds->ds_dir, + zfs_prop_to_name(ZFS_PROP_REFQUOTA), + ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } + + if (newval == 0) { + dsl_dataset_rele(ds, FTAG); return (0); + } - if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes || - psa->psa_effective_value < ds->ds_reserved) + if (newval < ds->ds_phys->ds_referenced_bytes || + newval < ds->ds_reserved) { + dsl_dataset_rele(ds, FTAG); return (ENOSPC); + } + dsl_dataset_rele(ds, FTAG); return (0); } -extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *); - -void -dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) +static void +dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_prop_setarg_t *psa = arg2; - uint64_t effective_value = psa->psa_effective_value; + dsl_dataset_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + uint64_t newval; - dsl_prop_set_sync(ds, psa, tx); - DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); + VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); - if (ds->ds_quota != effective_value) { + dsl_prop_set_sync_impl(ds, + zfs_prop_to_name(ZFS_PROP_REFQUOTA), + ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, + &ddsqra->ddsqra_value, tx); + + VERIFY0(dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); + + if (ds->ds_quota != newval) { dmu_buf_will_dirty(ds->ds_dbuf, tx); - ds->ds_quota = effective_value; + ds->ds_quota = newval; } + dsl_dataset_rele(ds, FTAG); } int -dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) +dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, + uint64_t refquota) { - dsl_dataset_t *ds; - dsl_prop_setarg_t psa; - int err; + dsl_dataset_set_qr_arg_t ddsqra; - dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); + ddsqra.ddsqra_name = dsname; + ddsqra.ddsqra_source = source; + ddsqra.ddsqra_value = refquota; - err = dsl_dataset_hold(dsname, FTAG, &ds); - if (err) - return (err); - - /* - * If someone removes a file, then tries to set the quota, we - * want to make sure the file freeing takes effect. - */ - txg_wait_open(ds->ds_dir->dd_pool, 0); - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, - ds, &psa, 0); - - dsl_dataset_rele(ds, FTAG); - return (err); + return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, + dsl_dataset_set_refquota_sync, &ddsqra, 0)); } static int -dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_prop_setarg_t *psa = arg2; - uint64_t effective_value; - uint64_t unique; - int err; + dsl_dataset_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + uint64_t newval, unique; - if (spa_version(ds->ds_dir->dd_pool->dp_spa) < - SPA_VERSION_REFRESERVATION) + if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) return (ENOTSUP); - if (dsl_dataset_is_snapshot(ds)) + error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); + if (error != 0) + return (error); + + if (dsl_dataset_is_snapshot(ds)) { + dsl_dataset_rele(ds, FTAG); return (EINVAL); + } - if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) - return (err); - - effective_value = psa->psa_effective_value; + error = dsl_prop_predict(ds->ds_dir, + zfs_prop_to_name(ZFS_PROP_REFRESERVATION), + ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } /* * If we are doing the preliminary check in open context, the * space estimates may be inaccurate. */ - if (!dmu_tx_is_syncing(tx)) + if (!dmu_tx_is_syncing(tx)) { + dsl_dataset_rele(ds, FTAG); return (0); + } mutex_enter(&ds->ds_lock); if (!DS_UNIQUE_IS_ACCURATE(ds)) @@ -3505,637 +2666,75 @@ dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) unique = ds->ds_phys->ds_unique_bytes; mutex_exit(&ds->ds_lock); - if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { - uint64_t delta = MAX(unique, effective_value) - + if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { + uint64_t delta = MAX(unique, newval) - MAX(unique, ds->ds_reserved); - if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) - return (ENOSPC); - if (ds->ds_quota > 0 && - effective_value > ds->ds_quota) + if (delta > + dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || + (ds->ds_quota > 0 && newval > ds->ds_quota)) { + dsl_dataset_rele(ds, FTAG); return (ENOSPC); + } } + dsl_dataset_rele(ds, FTAG); return (0); } -static void -dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) +void +dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, + zprop_source_t source, uint64_t value, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_prop_setarg_t *psa = arg2; - uint64_t effective_value = psa->psa_effective_value; + uint64_t newval; uint64_t unique; int64_t delta; - dsl_prop_set_sync(ds, psa, tx); - DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); + dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), + source, sizeof (value), 1, &value, tx); + + VERIFY0(dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); dmu_buf_will_dirty(ds->ds_dbuf, tx); - mutex_enter(&ds->ds_dir->dd_lock); mutex_enter(&ds->ds_lock); ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); unique = ds->ds_phys->ds_unique_bytes; - delta = MAX(0, (int64_t)(effective_value - unique)) - + delta = MAX(0, (int64_t)(newval - unique)) - MAX(0, (int64_t)(ds->ds_reserved - unique)); - ds->ds_reserved = effective_value; + ds->ds_reserved = newval; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); mutex_exit(&ds->ds_dir->dd_lock); } -int -dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, - uint64_t reservation) -{ - dsl_dataset_t *ds; - dsl_prop_setarg_t psa; - int err; - - dsl_prop_setarg_init_uint64(&psa, "refreservation", source, - &reservation); - - err = dsl_dataset_hold(dsname, FTAG, &ds); - if (err) - return (err); - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - dsl_dataset_set_reservation_check, - dsl_dataset_set_reservation_sync, ds, &psa, 0); - - dsl_dataset_rele(ds, FTAG); - return (err); -} - -typedef struct zfs_hold_cleanup_arg { - dsl_pool_t *dp; - uint64_t dsobj; - char htag[MAXNAMELEN]; -} zfs_hold_cleanup_arg_t; - static void -dsl_dataset_user_release_onexit(void *arg) +dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) { - zfs_hold_cleanup_arg_t *ca = arg; - - (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag, - B_TRUE); - kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t)); -} - -void -dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, - minor_t minor) -{ - zfs_hold_cleanup_arg_t *ca; - - ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP); - ca->dp = ds->ds_dir->dd_pool; - ca->dsobj = ds->ds_object; - (void) strlcpy(ca->htag, htag, sizeof (ca->htag)); - VERIFY3U(0, ==, zfs_onexit_add_cb(minor, - dsl_dataset_user_release_onexit, ca, NULL)); -} - -/* - * If you add new checks here, you may need to add - * additional checks to the "temporary" case in - * snapshot_check() in dmu_objset.c. - */ -static int -dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dataset_t *ds = arg1; - struct dsl_ds_holdarg *ha = arg2; - const char *htag = ha->htag; - objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; - int error = 0; - - if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) - return (ENOTSUP); - - if (!dsl_dataset_is_snapshot(ds)) - return (EINVAL); - - /* tags must be unique */ - mutex_enter(&ds->ds_lock); - if (ds->ds_phys->ds_userrefs_obj) { - error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, - 8, 1, tx); - if (error == 0) - error = EEXIST; - else if (error == ENOENT) - error = 0; - } - mutex_exit(&ds->ds_lock); - - if (error == 0 && ha->temphold && - strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) - error = E2BIG; - - return (error); -} - -void -dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dataset_t *ds = arg1; - struct dsl_ds_holdarg *ha = arg2; - const char *htag = ha->htag; - dsl_pool_t *dp = ds->ds_dir->dd_pool; - objset_t *mos = dp->dp_meta_objset; - uint64_t now = gethrestime_sec(); - uint64_t zapobj; - - mutex_enter(&ds->ds_lock); - if (ds->ds_phys->ds_userrefs_obj == 0) { - /* - * This is the first user hold for this dataset. Create - * the userrefs zap object. - */ - dmu_buf_will_dirty(ds->ds_dbuf, tx); - zapobj = ds->ds_phys->ds_userrefs_obj = - zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); - } else { - zapobj = ds->ds_phys->ds_userrefs_obj; - } - ds->ds_userrefs++; - mutex_exit(&ds->ds_lock); - - VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); - - if (ha->temphold) { - VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, - htag, &now, tx)); - } - - spa_history_log_internal_ds(ds, "hold", tx, - "tag = %s temp = %d holds now = %llu", - htag, (int)ha->temphold, ds->ds_userrefs); -} - -static int -dsl_dataset_user_hold_one(const char *dsname, void *arg) -{ - struct dsl_ds_holdarg *ha = arg; + dsl_dataset_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; - int error; - char *name; - /* alloc a buffer to hold dsname@snapname plus terminating NULL */ - name = kmem_asprintf("%s@%s", dsname, ha->snapname); - error = dsl_dataset_hold(name, ha->dstg, &ds); - strfree(name); - if (error == 0) { - ha->gotone = B_TRUE; - dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, - dsl_dataset_user_hold_sync, ds, ha, 0); - } else if (error == ENOENT && ha->recursive) { - error = 0; - } else { - (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); - } - return (error); -} - -int -dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag, - boolean_t temphold) -{ - struct dsl_ds_holdarg *ha; - int error; - - ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); - ha->htag = htag; - ha->temphold = temphold; - error = dsl_sync_task_do(ds->ds_dir->dd_pool, - dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync, - ds, ha, 0); - kmem_free(ha, sizeof (struct dsl_ds_holdarg)); - - return (error); -} - -int -dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, - boolean_t recursive, boolean_t temphold, int cleanup_fd) -{ - struct dsl_ds_holdarg *ha; - dsl_sync_task_t *dst; - spa_t *spa; - int error; - minor_t minor = 0; - - if (cleanup_fd != -1) { - /* Currently we only support cleanup-on-exit of tempholds. */ - if (!temphold) - return (EINVAL); - error = zfs_onexit_fd_hold(cleanup_fd, &minor); - if (error) - return (error); - } - - ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); - - (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); - - error = spa_open(dsname, &spa, FTAG); - if (error) { - kmem_free(ha, sizeof (struct dsl_ds_holdarg)); - if (cleanup_fd != -1) - zfs_onexit_fd_rele(cleanup_fd); - return (error); - } - - ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - ha->htag = htag; - ha->snapname = snapname; - ha->recursive = recursive; - ha->temphold = temphold; - - if (recursive) { - error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, - ha, DS_FIND_CHILDREN); - } else { - error = dsl_dataset_user_hold_one(dsname, ha); - } - if (error == 0) - error = dsl_sync_task_group_wait(ha->dstg); - - for (dst = list_head(&ha->dstg->dstg_tasks); dst; - dst = list_next(&ha->dstg->dstg_tasks, dst)) { - dsl_dataset_t *ds = dst->dst_arg1; - - if (dst->dst_err) { - dsl_dataset_name(ds, ha->failed); - *strchr(ha->failed, '@') = '\0'; - } else if (error == 0 && minor != 0 && temphold) { - /* - * If this hold is to be released upon process exit, - * register that action now. - */ - dsl_register_onexit_hold_cleanup(ds, htag, minor); - } - dsl_dataset_rele(ds, ha->dstg); - } - - if (error == 0 && recursive && !ha->gotone) - error = ENOENT; - - if (error) - (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); - - dsl_sync_task_group_destroy(ha->dstg); - - kmem_free(ha, sizeof (struct dsl_ds_holdarg)); - spa_close(spa, FTAG); - if (cleanup_fd != -1) - zfs_onexit_fd_rele(cleanup_fd); - return (error); -} - -struct dsl_ds_releasearg { - dsl_dataset_t *ds; - const char *htag; - boolean_t own; /* do we own or just hold ds? */ -}; - -static int -dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, - boolean_t *might_destroy) -{ - objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; - uint64_t zapobj; - uint64_t tmp; - int error; - - *might_destroy = B_FALSE; - - mutex_enter(&ds->ds_lock); - zapobj = ds->ds_phys->ds_userrefs_obj; - if (zapobj == 0) { - /* The tag can't possibly exist */ - mutex_exit(&ds->ds_lock); - return (ESRCH); - } - - /* Make sure the tag exists */ - error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); - if (error) { - mutex_exit(&ds->ds_lock); - if (error == ENOENT) - error = ESRCH; - return (error); - } - - if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && - DS_IS_DEFER_DESTROY(ds)) - *might_destroy = B_TRUE; - - mutex_exit(&ds->ds_lock); - return (0); -} - -static int -dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) -{ - struct dsl_ds_releasearg *ra = arg1; - dsl_dataset_t *ds = ra->ds; - boolean_t might_destroy; - int error; - - if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) - return (ENOTSUP); - - error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); - if (error) - return (error); - - if (might_destroy) { - struct dsl_ds_destroyarg dsda = {0}; - - if (dmu_tx_is_syncing(tx)) { - /* - * If we're not prepared to remove the snapshot, - * we can't allow the release to happen right now. - */ - if (!ra->own) - return (EBUSY); - } - dsda.ds = ds; - dsda.releasing = B_TRUE; - return (dsl_dataset_destroy_check(&dsda, tag, tx)); - } - - return (0); -} - -static void -dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx) -{ - struct dsl_ds_releasearg *ra = arg1; - dsl_dataset_t *ds = ra->ds; - dsl_pool_t *dp = ds->ds_dir->dd_pool; - objset_t *mos = dp->dp_meta_objset; - uint64_t zapobj; - uint64_t refs; - int error; - - mutex_enter(&ds->ds_lock); - ds->ds_userrefs--; - refs = ds->ds_userrefs; - mutex_exit(&ds->ds_lock); - error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); - VERIFY(error == 0 || error == ENOENT); - zapobj = ds->ds_phys->ds_userrefs_obj; - VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); - - spa_history_log_internal_ds(ds, "release", tx, - "tag = %s refs now = %lld", ra->htag, (longlong_t)refs); - - if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && - DS_IS_DEFER_DESTROY(ds)) { - struct dsl_ds_destroyarg dsda = {0}; - - ASSERT(ra->own); - dsda.ds = ds; - dsda.releasing = B_TRUE; - /* We already did the destroy_check */ - dsl_dataset_destroy_sync(&dsda, tag, tx); - } -} - -static int -dsl_dataset_user_release_one(const char *dsname, void *arg) -{ - struct dsl_ds_holdarg *ha = arg; - struct dsl_ds_releasearg *ra; - dsl_dataset_t *ds; - int error; - void *dtag = ha->dstg; - char *name; - boolean_t own = B_FALSE; - boolean_t might_destroy; - - /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ - name = kmem_asprintf("%s@%s", dsname, ha->snapname); - error = dsl_dataset_hold(name, dtag, &ds); - strfree(name); - if (error == ENOENT && ha->recursive) - return (0); - (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); - if (error) - return (error); - - ha->gotone = B_TRUE; - - ASSERT(dsl_dataset_is_snapshot(ds)); - - error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); - if (error) { - dsl_dataset_rele(ds, dtag); - return (error); - } - - if (might_destroy) { -#ifdef _KERNEL - name = kmem_asprintf("%s@%s", dsname, ha->snapname); - error = zfs_unmount_snap(name, NULL); - strfree(name); - if (error) { - dsl_dataset_rele(ds, dtag); - return (error); - } -#endif - if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { - dsl_dataset_rele(ds, dtag); - return (EBUSY); - } else { - own = B_TRUE; - dsl_dataset_make_exclusive(ds, dtag); - } - } - - ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); - ra->ds = ds; - ra->htag = ha->htag; - ra->own = own; - dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, - dsl_dataset_user_release_sync, ra, dtag, 0); - - return (0); -} - -int -dsl_dataset_user_release(char *dsname, char *snapname, char *htag, - boolean_t recursive) -{ - struct dsl_ds_holdarg *ha; - dsl_sync_task_t *dst; - spa_t *spa; - int error; - -top: - ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); - - (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); - - error = spa_open(dsname, &spa, FTAG); - if (error) { - kmem_free(ha, sizeof (struct dsl_ds_holdarg)); - return (error); - } - - ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); - ha->htag = htag; - ha->snapname = snapname; - ha->recursive = recursive; - if (recursive) { - error = dmu_objset_find(dsname, dsl_dataset_user_release_one, - ha, DS_FIND_CHILDREN); - } else { - error = dsl_dataset_user_release_one(dsname, ha); - } - if (error == 0) - error = dsl_sync_task_group_wait(ha->dstg); - - for (dst = list_head(&ha->dstg->dstg_tasks); dst; - dst = list_next(&ha->dstg->dstg_tasks, dst)) { - struct dsl_ds_releasearg *ra = dst->dst_arg1; - dsl_dataset_t *ds = ra->ds; - - if (dst->dst_err) - dsl_dataset_name(ds, ha->failed); - - if (ra->own) - dsl_dataset_disown(ds, ha->dstg); - else - dsl_dataset_rele(ds, ha->dstg); - - kmem_free(ra, sizeof (struct dsl_ds_releasearg)); - } - - if (error == 0 && recursive && !ha->gotone) - error = ENOENT; - - if (error && error != EBUSY) - (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); - - dsl_sync_task_group_destroy(ha->dstg); - kmem_free(ha, sizeof (struct dsl_ds_holdarg)); - spa_close(spa, FTAG); - - /* - * We can get EBUSY if we were racing with deferred destroy and - * dsl_dataset_user_release_check() hadn't done the necessary - * open context setup. We can also get EBUSY if we're racing - * with destroy and that thread is the ds_owner. Either way - * the busy condition should be transient, and we should retry - * the release operation. - */ - if (error == EBUSY) - goto top; - - return (error); -} - -/* - * Called at spa_load time (with retry == B_FALSE) to release a stale - * temporary user hold. Also called by the onexit code (with retry == B_TRUE). - */ -int -dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag, - boolean_t retry) -{ - dsl_dataset_t *ds; - char *snap; - char *name; - int namelen; - int error; - - do { - rw_enter(&dp->dp_config_rwlock, RW_READER); - error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - if (error) - return (error); - namelen = dsl_dataset_namelen(ds)+1; - name = kmem_alloc(namelen, KM_SLEEP); - dsl_dataset_name(ds, name); - dsl_dataset_rele(ds, FTAG); - - snap = strchr(name, '@'); - *snap = '\0'; - ++snap; - error = dsl_dataset_user_release(name, snap, htag, B_FALSE); - kmem_free(name, namelen); - - /* - * The object can't have been destroyed because we have a hold, - * but it might have been renamed, resulting in ENOENT. Retry - * if we've been requested to do so. - * - * It would be nice if we could use the dsobj all the way - * through and avoid ENOENT entirely. But we might need to - * unmount the snapshot, and there's currently no way to lookup - * a vfsp using a ZFS object id. - */ - } while ((error == ENOENT) && retry); - - return (error); -} - -int -dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) -{ - dsl_dataset_t *ds; - int err; - - err = dsl_dataset_hold(dsname, FTAG, &ds); - if (err) - return (err); - - VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); - if (ds->ds_phys->ds_userrefs_obj != 0) { - zap_attribute_t *za; - zap_cursor_t zc; - - za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); - for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, - ds->ds_phys->ds_userrefs_obj); - zap_cursor_retrieve(&zc, za) == 0; - zap_cursor_advance(&zc)) { - VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, - za->za_first_integer)); - } - zap_cursor_fini(&zc); - kmem_free(za, sizeof (zap_attribute_t)); - } + VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); + dsl_dataset_set_refreservation_sync_impl(ds, + ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); dsl_dataset_rele(ds, FTAG); - return (0); } -/* - * Note, this function is used as the callback for dmu_objset_find(). We - * always return 0 so that we will continue to find and process - * inconsistent datasets, even if we encounter an error trying to - * process one of them. - */ -/* ARGSUSED */ int -dsl_destroy_inconsistent(const char *dsname, void *arg) +dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, + uint64_t refreservation) { - dsl_dataset_t *ds; + dsl_dataset_set_qr_arg_t ddsqra; - if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { - if (DS_IS_INCONSISTENT(ds)) - (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); - else - dsl_dataset_disown(ds, FTAG); - } - return (0); + ddsqra.ddsqra_name = dsname; + ddsqra.ddsqra_source = source; + ddsqra.ddsqra_value = refreservation; + + return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, + dsl_dataset_set_refreservation_sync, &ddsqra, 0)); } /* @@ -4163,6 +2762,8 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, uint64_t snapobj; dsl_pool_t *dp = new->ds_dir->dd_pool; + ASSERT(dsl_pool_config_held(dp)); + *usedp = 0; *usedp += new->ds_phys->ds_referenced_bytes; *usedp -= oldsnap->ds_phys->ds_referenced_bytes; @@ -4175,7 +2776,6 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, *uncompp += new->ds_phys->ds_uncompressed_bytes; *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; - rw_enter(&dp->dp_config_rwlock, RW_READER); snapobj = new->ds_object; while (snapobj != oldsnap->ds_object) { dsl_dataset_t *snap; @@ -4224,7 +2824,6 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, } } - rw_exit(&dp->dp_config_rwlock); return (err); } @@ -4266,7 +2865,6 @@ dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, *usedp = *compp = *uncompp = 0; - rw_enter(&dp->dp_config_rwlock, RW_READER); snapobj = lastsnap->ds_phys->ds_next_snap_obj; while (snapobj != firstsnap->ds_object) { dsl_dataset_t *ds; @@ -4287,6 +2885,42 @@ dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, ASSERT3U(snapobj, !=, 0); dsl_dataset_rele(ds, FTAG); } - rw_exit(&dp->dp_config_rwlock); return (err); } + +/* + * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. + * For example, they could both be snapshots of the same filesystem, and + * 'earlier' is before 'later'. Or 'earlier' could be the origin of + * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's + * filesystem. Or 'earlier' could be the origin's origin. + */ +boolean_t +dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier) +{ + dsl_pool_t *dp = later->ds_dir->dd_pool; + int error; + boolean_t ret; + + ASSERT(dsl_pool_config_held(dp)); + + if (earlier->ds_phys->ds_creation_txg >= + later->ds_phys->ds_creation_txg) + return (B_FALSE); + + if (later->ds_dir == earlier->ds_dir) + return (B_TRUE); + if (!dsl_dir_is_clone(later->ds_dir)) + return (B_FALSE); + + if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) + return (B_TRUE); + dsl_dataset_t *origin; + error = dsl_dataset_hold_obj(dp, + later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); + if (error != 0) + return (B_FALSE); + ret = dsl_dataset_is_before(origin, earlier); + dsl_dataset_rele(origin, FTAG); + return (ret); +} diff --git a/uts/common/fs/zfs/dsl_deleg.c b/uts/common/fs/zfs/dsl_deleg.c index ba620bd6fbed..f09cb2f2ac29 100644 --- a/uts/common/fs/zfs/dsl_deleg.c +++ b/uts/common/fs/zfs/dsl_deleg.c @@ -147,28 +147,37 @@ dsl_deleg_can_unallow(char *ddname, nvlist_t *nvp, cred_t *cr) return (0); } -static void -dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dir_t *dd = arg1; - nvlist_t *nvp = arg2; - objset_t *mos = dd->dd_pool->dp_meta_objset; - nvpair_t *whopair = NULL; - uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj; +typedef struct dsl_deleg_arg { + const char *dda_name; + nvlist_t *dda_nvlist; +} dsl_deleg_arg_t; +static void +dsl_deleg_set_sync(void *arg, dmu_tx_t *tx) +{ + dsl_deleg_arg_t *dda = arg; + dsl_dir_t *dd; + dsl_pool_t *dp = dmu_tx_pool(tx); + objset_t *mos = dp->dp_meta_objset; + nvpair_t *whopair = NULL; + uint64_t zapobj; + + VERIFY0(dsl_dir_hold(dp, dda->dda_name, FTAG, &dd, NULL)); + + zapobj = dd->dd_phys->dd_deleg_zapobj; if (zapobj == 0) { dmu_buf_will_dirty(dd->dd_dbuf, tx); zapobj = dd->dd_phys->dd_deleg_zapobj = zap_create(mos, DMU_OT_DSL_PERMS, DMU_OT_NONE, 0, tx); } - while (whopair = nvlist_next_nvpair(nvp, whopair)) { + while (whopair = nvlist_next_nvpair(dda->dda_nvlist, whopair)) { const char *whokey = nvpair_name(whopair); nvlist_t *perms; nvpair_t *permpair = NULL; uint64_t jumpobj; - VERIFY(nvpair_value_nvlist(whopair, &perms) == 0); + perms = fnvpair_value_nvlist(whopair); if (zap_lookup(mos, zapobj, whokey, 8, 1, &jumpobj) != 0) { jumpobj = zap_create_link(mos, DMU_OT_DSL_PERMS, @@ -185,21 +194,27 @@ dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) "%s %s", whokey, perm); } } + dsl_dir_rele(dd, FTAG); } static void -dsl_deleg_unset_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_deleg_unset_sync(void *arg, dmu_tx_t *tx) { - dsl_dir_t *dd = arg1; - nvlist_t *nvp = arg2; - objset_t *mos = dd->dd_pool->dp_meta_objset; + dsl_deleg_arg_t *dda = arg; + dsl_dir_t *dd; + dsl_pool_t *dp = dmu_tx_pool(tx); + objset_t *mos = dp->dp_meta_objset; nvpair_t *whopair = NULL; - uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj; + uint64_t zapobj; - if (zapobj == 0) + VERIFY0(dsl_dir_hold(dp, dda->dda_name, FTAG, &dd, NULL)); + zapobj = dd->dd_phys->dd_deleg_zapobj; + if (zapobj == 0) { + dsl_dir_rele(dd, FTAG); return; + } - while (whopair = nvlist_next_nvpair(nvp, whopair)) { + while (whopair = nvlist_next_nvpair(dda->dda_nvlist, whopair)) { const char *whokey = nvpair_name(whopair); nvlist_t *perms; nvpair_t *permpair = NULL; @@ -234,35 +249,40 @@ dsl_deleg_unset_sync(void *arg1, void *arg2, dmu_tx_t *tx) "%s %s", whokey, perm); } } + dsl_dir_rele(dd, FTAG); +} + +static int +dsl_deleg_check(void *arg, dmu_tx_t *tx) +{ + dsl_deleg_arg_t *dda = arg; + dsl_dir_t *dd; + int error; + + if (spa_version(dmu_tx_pool(tx)->dp_spa) < + SPA_VERSION_DELEGATED_PERMS) { + return (ENOTSUP); + } + + error = dsl_dir_hold(dmu_tx_pool(tx), dda->dda_name, FTAG, &dd, NULL); + if (error == 0) + dsl_dir_rele(dd, FTAG); + return (error); } int dsl_deleg_set(const char *ddname, nvlist_t *nvp, boolean_t unset) { - dsl_dir_t *dd; - int error; - nvpair_t *whopair = NULL; - int blocks_modified = 0; + dsl_deleg_arg_t dda; - error = dsl_dir_open(ddname, FTAG, &dd, NULL); - if (error) - return (error); + /* nvp must already have been verified to be valid */ - if (spa_version(dmu_objset_spa(dd->dd_pool->dp_meta_objset)) < - SPA_VERSION_DELEGATED_PERMS) { - dsl_dir_close(dd, FTAG); - return (ENOTSUP); - } + dda.dda_name = ddname; + dda.dda_nvlist = nvp; - while (whopair = nvlist_next_nvpair(nvp, whopair)) - blocks_modified++; - - error = dsl_sync_task_do(dd->dd_pool, NULL, + return (dsl_sync_task(ddname, dsl_deleg_check, unset ? dsl_deleg_unset_sync : dsl_deleg_set_sync, - dd, nvp, blocks_modified); - dsl_dir_close(dd, FTAG); - - return (error); + &dda, fnvlist_num_pairs(nvp))); } /* @@ -290,16 +310,21 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp) int error; objset_t *mos; - error = dsl_dir_open(ddname, FTAG, &startdd, NULL); - if (error) + error = dsl_pool_hold(ddname, FTAG, &dp); + if (error != 0) return (error); + error = dsl_dir_hold(dp, ddname, FTAG, &startdd, NULL); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + dp = startdd->dd_pool; mos = dp->dp_meta_objset; VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); - rw_enter(&dp->dp_config_rwlock, RW_READER); for (dd = startdd; dd != NULL; dd = dd->dd_parent) { zap_cursor_t basezc; zap_attribute_t baseza; @@ -307,15 +332,12 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp) uint64_t n; char source[MAXNAMELEN]; - if (dd->dd_phys->dd_deleg_zapobj && - (zap_count(mos, dd->dd_phys->dd_deleg_zapobj, - &n) == 0) && n) { - VERIFY(nvlist_alloc(&sp_nvp, - NV_UNIQUE_NAME, KM_SLEEP) == 0); - } else { + if (dd->dd_phys->dd_deleg_zapobj == 0 || + zap_count(mos, dd->dd_phys->dd_deleg_zapobj, &n) != 0 || + n == 0) continue; - } + sp_nvp = fnvlist_alloc(); for (zap_cursor_init(&basezc, mos, dd->dd_phys->dd_deleg_zapobj); zap_cursor_retrieve(&basezc, &baseza) == 0; @@ -327,29 +349,26 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp) ASSERT(baseza.za_integer_length == 8); ASSERT(baseza.za_num_integers == 1); - VERIFY(nvlist_alloc(&perms_nvp, - NV_UNIQUE_NAME, KM_SLEEP) == 0); + perms_nvp = fnvlist_alloc(); for (zap_cursor_init(&zc, mos, baseza.za_first_integer); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { - VERIFY(nvlist_add_boolean(perms_nvp, - za.za_name) == 0); + fnvlist_add_boolean(perms_nvp, za.za_name); } zap_cursor_fini(&zc); - VERIFY(nvlist_add_nvlist(sp_nvp, baseza.za_name, - perms_nvp) == 0); - nvlist_free(perms_nvp); + fnvlist_add_nvlist(sp_nvp, baseza.za_name, perms_nvp); + fnvlist_free(perms_nvp); } zap_cursor_fini(&basezc); dsl_dir_name(dd, source); - VERIFY(nvlist_add_nvlist(*nvp, source, sp_nvp) == 0); + fnvlist_add_nvlist(*nvp, source, sp_nvp); nvlist_free(sp_nvp); } - rw_exit(&dp->dp_config_rwlock); - dsl_dir_close(startdd, FTAG); + dsl_dir_rele(startdd, FTAG); + dsl_pool_rele(dp, FTAG); return (0); } @@ -555,7 +574,7 @@ dsl_deleg_access_impl(dsl_dataset_t *ds, const char *perm, cred_t *cr) avl_create(&permsets, perm_set_compare, sizeof (perm_set_t), offsetof(perm_set_t, p_node)); - rw_enter(&dp->dp_config_rwlock, RW_READER); + ASSERT(dsl_pool_config_held(dp)); for (dd = ds->ds_dir; dd != NULL; dd = dd->dd_parent, checkflag = ZFS_DELEG_DESCENDENT) { uint64_t zapobj; @@ -616,7 +635,6 @@ dsl_deleg_access_impl(dsl_dataset_t *ds, const char *perm, cred_t *cr) } error = EPERM; success: - rw_exit(&dp->dp_config_rwlock); cookie = NULL; while ((setnode = avl_destroy_nodes(&permsets, &cookie)) != NULL) @@ -628,15 +646,19 @@ dsl_deleg_access_impl(dsl_dataset_t *ds, const char *perm, cred_t *cr) int dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr) { + dsl_pool_t *dp; dsl_dataset_t *ds; int error; - error = dsl_dataset_hold(dsname, FTAG, &ds); - if (error) + error = dsl_pool_hold(dsname, FTAG, &dp); + if (error != 0) return (error); - - error = dsl_deleg_access_impl(ds, perm, cr); - dsl_dataset_rele(ds, FTAG); + error = dsl_dataset_hold(dp, dsname, FTAG, &ds); + if (error == 0) { + error = dsl_deleg_access_impl(ds, perm, cr); + dsl_dataset_rele(ds, FTAG); + } + dsl_pool_rele(dp, FTAG); return (error); } diff --git a/uts/common/fs/zfs/dsl_destroy.c b/uts/common/fs/zfs/dsl_destroy.c new file mode 100644 index 000000000000..20d401f2590d --- /dev/null +++ b/uts/common/fs/zfs/dsl_destroy.c @@ -0,0 +1,926 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +typedef struct dmu_snapshots_destroy_arg { + nvlist_t *dsda_snaps; + nvlist_t *dsda_successful_snaps; + boolean_t dsda_defer; + nvlist_t *dsda_errlist; +} dmu_snapshots_destroy_arg_t; + +/* + * ds must be owned. + */ +static int +dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer) +{ + if (!dsl_dataset_is_snapshot(ds)) + return (EINVAL); + + if (dsl_dataset_long_held(ds)) + return (EBUSY); + + /* + * Only allow deferred destroy on pools that support it. + * NOTE: deferred destroy is only supported on snapshots. + */ + if (defer) { + if (spa_version(ds->ds_dir->dd_pool->dp_spa) < + SPA_VERSION_USERREFS) + return (ENOTSUP); + return (0); + } + + /* + * If this snapshot has an elevated user reference count, + * we can't destroy it yet. + */ + if (ds->ds_userrefs > 0) + return (EBUSY); + + /* + * Can't delete a branch point. + */ + if (ds->ds_phys->ds_num_children > 1) + return (EEXIST); + + return (0); +} + +static int +dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx) +{ + dmu_snapshots_destroy_arg_t *dsda = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + int error = 0; + + if (!dmu_tx_is_syncing(tx)) + return (0); + + for (pair = nvlist_next_nvpair(dsda->dsda_snaps, NULL); + pair != NULL; pair = nvlist_next_nvpair(dsda->dsda_snaps, pair)) { + dsl_dataset_t *ds; + + error = dsl_dataset_hold(dp, nvpair_name(pair), + FTAG, &ds); + + /* + * If the snapshot does not exist, silently ignore it + * (it's "already destroyed"). + */ + if (error == ENOENT) + continue; + + if (error == 0) { + error = dsl_destroy_snapshot_check_impl(ds, + dsda->dsda_defer); + dsl_dataset_rele(ds, FTAG); + } + + if (error == 0) { + fnvlist_add_boolean(dsda->dsda_successful_snaps, + nvpair_name(pair)); + } else { + fnvlist_add_int32(dsda->dsda_errlist, + nvpair_name(pair), error); + } + } + + pair = nvlist_next_nvpair(dsda->dsda_errlist, NULL); + if (pair != NULL) + return (fnvpair_value_int32(pair)); + return (0); +} + +struct process_old_arg { + dsl_dataset_t *ds; + dsl_dataset_t *ds_prev; + boolean_t after_branch_point; + zio_t *pio; + uint64_t used, comp, uncomp; +}; + +static int +process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) +{ + struct process_old_arg *poa = arg; + dsl_pool_t *dp = poa->ds->ds_dir->dd_pool; + + if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) { + dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx); + if (poa->ds_prev && !poa->after_branch_point && + bp->blk_birth > + poa->ds_prev->ds_phys->ds_prev_snap_txg) { + poa->ds_prev->ds_phys->ds_unique_bytes += + bp_get_dsize_sync(dp->dp_spa, bp); + } + } else { + poa->used += bp_get_dsize_sync(dp->dp_spa, bp); + poa->comp += BP_GET_PSIZE(bp); + poa->uncomp += BP_GET_UCSIZE(bp); + dsl_free_sync(poa->pio, dp, tx->tx_txg, bp); + } + return (0); +} + +static void +process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, + dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx) +{ + struct process_old_arg poa = { 0 }; + dsl_pool_t *dp = ds->ds_dir->dd_pool; + objset_t *mos = dp->dp_meta_objset; + uint64_t deadlist_obj; + + ASSERT(ds->ds_deadlist.dl_oldfmt); + ASSERT(ds_next->ds_deadlist.dl_oldfmt); + + poa.ds = ds; + poa.ds_prev = ds_prev; + poa.after_branch_point = after_branch_point; + poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); + VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj, + process_old_cb, &poa, tx)); + VERIFY0(zio_wait(poa.pio)); + ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes); + + /* change snapused */ + dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, + -poa.used, -poa.comp, -poa.uncomp, tx); + + /* swap next's deadlist to our deadlist */ + dsl_deadlist_close(&ds->ds_deadlist); + dsl_deadlist_close(&ds_next->ds_deadlist); + deadlist_obj = ds->ds_phys->ds_deadlist_obj; + ds->ds_phys->ds_deadlist_obj = ds_next->ds_phys->ds_deadlist_obj; + ds_next->ds_phys->ds_deadlist_obj = deadlist_obj; + dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); + dsl_deadlist_open(&ds_next->ds_deadlist, mos, + ds_next->ds_phys->ds_deadlist_obj); +} + +static void +dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx) +{ + objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; + zap_cursor_t zc; + zap_attribute_t za; + + /* + * If it is the old version, dd_clones doesn't exist so we can't + * find the clones, but dsl_deadlist_remove_key() is a no-op so it + * doesn't matter. + */ + if (ds->ds_dir->dd_phys->dd_clones == 0) + return; + + for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones); + zap_cursor_retrieve(&zc, &za) == 0; + zap_cursor_advance(&zc)) { + dsl_dataset_t *clone; + + VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, + za.za_first_integer, FTAG, &clone)); + if (clone->ds_dir->dd_origin_txg > mintxg) { + dsl_deadlist_remove_key(&clone->ds_deadlist, + mintxg, tx); + dsl_dataset_remove_clones_key(clone, mintxg, tx); + } + dsl_dataset_rele(clone, FTAG); + } + zap_cursor_fini(&zc); +} + +void +dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx) +{ + int err; + int after_branch_point = FALSE; + dsl_pool_t *dp = ds->ds_dir->dd_pool; + objset_t *mos = dp->dp_meta_objset; + dsl_dataset_t *ds_prev = NULL; + uint64_t obj; + + ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); + ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); + ASSERT(refcount_is_zero(&ds->ds_longholds)); + + if (defer && + (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)) { + ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); + dmu_buf_will_dirty(ds->ds_dbuf, tx); + ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; + spa_history_log_internal_ds(ds, "defer_destroy", tx, ""); + return; + } + + ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); + + /* We need to log before removing it from the namespace. */ + spa_history_log_internal_ds(ds, "destroy", tx, ""); + + dsl_scan_ds_destroyed(ds, tx); + + obj = ds->ds_object; + + if (ds->ds_phys->ds_prev_snap_obj != 0) { + ASSERT3P(ds->ds_prev, ==, NULL); + VERIFY0(dsl_dataset_hold_obj(dp, + ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); + after_branch_point = + (ds_prev->ds_phys->ds_next_snap_obj != obj); + + dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); + if (after_branch_point && + ds_prev->ds_phys->ds_next_clones_obj != 0) { + dsl_dataset_remove_from_next_clones(ds_prev, obj, tx); + if (ds->ds_phys->ds_next_snap_obj != 0) { + VERIFY0(zap_add_int(mos, + ds_prev->ds_phys->ds_next_clones_obj, + ds->ds_phys->ds_next_snap_obj, tx)); + } + } + if (!after_branch_point) { + ds_prev->ds_phys->ds_next_snap_obj = + ds->ds_phys->ds_next_snap_obj; + } + } + + dsl_dataset_t *ds_next; + uint64_t old_unique; + uint64_t used = 0, comp = 0, uncomp = 0; + + VERIFY0(dsl_dataset_hold_obj(dp, + ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); + ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); + + old_unique = ds_next->ds_phys->ds_unique_bytes; + + dmu_buf_will_dirty(ds_next->ds_dbuf, tx); + ds_next->ds_phys->ds_prev_snap_obj = + ds->ds_phys->ds_prev_snap_obj; + ds_next->ds_phys->ds_prev_snap_txg = + ds->ds_phys->ds_prev_snap_txg; + ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, + ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); + + if (ds_next->ds_deadlist.dl_oldfmt) { + process_old_deadlist(ds, ds_prev, ds_next, + after_branch_point, tx); + } else { + /* Adjust prev's unique space. */ + if (ds_prev && !after_branch_point) { + dsl_deadlist_space_range(&ds_next->ds_deadlist, + ds_prev->ds_phys->ds_prev_snap_txg, + ds->ds_phys->ds_prev_snap_txg, + &used, &comp, &uncomp); + ds_prev->ds_phys->ds_unique_bytes += used; + } + + /* Adjust snapused. */ + dsl_deadlist_space_range(&ds_next->ds_deadlist, + ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, + &used, &comp, &uncomp); + dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, + -used, -comp, -uncomp, tx); + + /* Move blocks to be freed to pool's free list. */ + dsl_deadlist_move_bpobj(&ds_next->ds_deadlist, + &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg, + tx); + dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, + DD_USED_HEAD, used, comp, uncomp, tx); + + /* Merge our deadlist into next's and free it. */ + dsl_deadlist_merge(&ds_next->ds_deadlist, + ds->ds_phys->ds_deadlist_obj, tx); + } + dsl_deadlist_close(&ds->ds_deadlist); + dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); + dmu_buf_will_dirty(ds->ds_dbuf, tx); + ds->ds_phys->ds_deadlist_obj = 0; + + /* Collapse range in clone heads */ + dsl_dataset_remove_clones_key(ds, + ds->ds_phys->ds_creation_txg, tx); + + if (dsl_dataset_is_snapshot(ds_next)) { + dsl_dataset_t *ds_nextnext; + + /* + * Update next's unique to include blocks which + * were previously shared by only this snapshot + * and it. Those blocks will be born after the + * prev snap and before this snap, and will have + * died after the next snap and before the one + * after that (ie. be on the snap after next's + * deadlist). + */ + VERIFY0(dsl_dataset_hold_obj(dp, + ds_next->ds_phys->ds_next_snap_obj, FTAG, &ds_nextnext)); + dsl_deadlist_space_range(&ds_nextnext->ds_deadlist, + ds->ds_phys->ds_prev_snap_txg, + ds->ds_phys->ds_creation_txg, + &used, &comp, &uncomp); + ds_next->ds_phys->ds_unique_bytes += used; + dsl_dataset_rele(ds_nextnext, FTAG); + ASSERT3P(ds_next->ds_prev, ==, NULL); + + /* Collapse range in this head. */ + dsl_dataset_t *hds; + VERIFY0(dsl_dataset_hold_obj(dp, + ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &hds)); + dsl_deadlist_remove_key(&hds->ds_deadlist, + ds->ds_phys->ds_creation_txg, tx); + dsl_dataset_rele(hds, FTAG); + + } else { + ASSERT3P(ds_next->ds_prev, ==, ds); + dsl_dataset_rele(ds_next->ds_prev, ds_next); + ds_next->ds_prev = NULL; + if (ds_prev) { + VERIFY0(dsl_dataset_hold_obj(dp, + ds->ds_phys->ds_prev_snap_obj, + ds_next, &ds_next->ds_prev)); + } + + dsl_dataset_recalc_head_uniq(ds_next); + + /* + * Reduce the amount of our unconsumed refreservation + * being charged to our parent by the amount of + * new unique data we have gained. + */ + if (old_unique < ds_next->ds_reserved) { + int64_t mrsdelta; + uint64_t new_unique = + ds_next->ds_phys->ds_unique_bytes; + + ASSERT(old_unique <= new_unique); + mrsdelta = MIN(new_unique - old_unique, + ds_next->ds_reserved - old_unique); + dsl_dir_diduse_space(ds->ds_dir, + DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); + } + } + dsl_dataset_rele(ds_next, FTAG); + + /* + * This must be done after the dsl_traverse(), because it will + * re-open the objset. + */ + if (ds->ds_objset) { + dmu_objset_evict(ds->ds_objset); + ds->ds_objset = NULL; + } + + /* remove from snapshot namespace */ + dsl_dataset_t *ds_head; + ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); + VERIFY0(dsl_dataset_hold_obj(dp, + ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); + VERIFY0(dsl_dataset_get_snapname(ds)); +#ifdef ZFS_DEBUG + { + uint64_t val; + + err = dsl_dataset_snap_lookup(ds_head, + ds->ds_snapname, &val); + ASSERT0(err); + ASSERT3U(val, ==, obj); + } +#endif + VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx)); + dsl_dataset_rele(ds_head, FTAG); + + if (ds_prev != NULL) + dsl_dataset_rele(ds_prev, FTAG); + + spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); + + if (ds->ds_phys->ds_next_clones_obj != 0) { + uint64_t count; + ASSERT0(zap_count(mos, + ds->ds_phys->ds_next_clones_obj, &count) && count == 0); + VERIFY0(dmu_object_free(mos, + ds->ds_phys->ds_next_clones_obj, tx)); + } + if (ds->ds_phys->ds_props_obj != 0) + VERIFY0(zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); + if (ds->ds_phys->ds_userrefs_obj != 0) + VERIFY0(zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); + dsl_dir_rele(ds->ds_dir, ds); + ds->ds_dir = NULL; + VERIFY0(dmu_object_free(mos, obj, tx)); +} + +static void +dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx) +{ + dmu_snapshots_destroy_arg_t *dsda = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + + for (pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, NULL); + pair != NULL; + pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, pair)) { + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds)); + + dsl_destroy_snapshot_sync_impl(ds, dsda->dsda_defer, tx); + dsl_dataset_rele(ds, FTAG); + } +} + +/* + * The semantics of this function are described in the comment above + * lzc_destroy_snaps(). To summarize: + * + * The snapshots must all be in the same pool. + * + * Snapshots that don't exist will be silently ignored (considered to be + * "already deleted"). + * + * On success, all snaps will be destroyed and this will return 0. + * On failure, no snaps will be destroyed, the errlist will be filled in, + * and this will return an errno. + */ +int +dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer, + nvlist_t *errlist) +{ + dmu_snapshots_destroy_arg_t dsda; + int error; + nvpair_t *pair; + + pair = nvlist_next_nvpair(snaps, NULL); + if (pair == NULL) + return (0); + + dsda.dsda_snaps = snaps; + dsda.dsda_successful_snaps = fnvlist_alloc(); + dsda.dsda_defer = defer; + dsda.dsda_errlist = errlist; + + error = dsl_sync_task(nvpair_name(pair), + dsl_destroy_snapshot_check, dsl_destroy_snapshot_sync, + &dsda, 0); + fnvlist_free(dsda.dsda_successful_snaps); + + return (error); +} + +int +dsl_destroy_snapshot(const char *name, boolean_t defer) +{ + int error; + nvlist_t *nvl = fnvlist_alloc(); + nvlist_t *errlist = fnvlist_alloc(); + + fnvlist_add_boolean(nvl, name); + error = dsl_destroy_snapshots_nvl(nvl, defer, errlist); + fnvlist_free(errlist); + fnvlist_free(nvl); + return (error); +} + +struct killarg { + dsl_dataset_t *ds; + dmu_tx_t *tx; +}; + +/* ARGSUSED */ +static int +kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, + const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) +{ + struct killarg *ka = arg; + dmu_tx_t *tx = ka->tx; + + if (bp == NULL) + return (0); + + if (zb->zb_level == ZB_ZIL_LEVEL) { + ASSERT(zilog != NULL); + /* + * It's a block in the intent log. It has no + * accounting, so just free it. + */ + dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); + } else { + ASSERT(zilog == NULL); + ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); + (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); + } + + return (0); +} + +static void +old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx) +{ + struct killarg ka; + + /* + * Free everything that we point to (that's born after + * the previous snapshot, if we are a clone) + * + * NB: this should be very quick, because we already + * freed all the objects in open context. + */ + ka.ds = ds; + ka.tx = tx; + VERIFY0(traverse_dataset(ds, + ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST, + kill_blkptr, &ka)); + ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0); +} + +typedef struct dsl_destroy_head_arg { + const char *ddha_name; +} dsl_destroy_head_arg_t; + +int +dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds) +{ + int error; + uint64_t count; + objset_t *mos; + + if (dsl_dataset_is_snapshot(ds)) + return (EINVAL); + + if (refcount_count(&ds->ds_longholds) != expected_holds) + return (EBUSY); + + mos = ds->ds_dir->dd_pool->dp_meta_objset; + + /* + * Can't delete a head dataset if there are snapshots of it. + * (Except if the only snapshots are from the branch we cloned + * from.) + */ + if (ds->ds_prev != NULL && + ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) + return (EBUSY); + + /* + * Can't delete if there are children of this fs. + */ + error = zap_count(mos, + ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); + if (error != 0) + return (error); + if (count != 0) + return (EEXIST); + + if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) && + ds->ds_prev->ds_phys->ds_num_children == 2 && + ds->ds_prev->ds_userrefs == 0) { + /* We need to remove the origin snapshot as well. */ + if (!refcount_is_zero(&ds->ds_prev->ds_longholds)) + return (EBUSY); + } + return (0); +} + +static int +dsl_destroy_head_check(void *arg, dmu_tx_t *tx) +{ + dsl_destroy_head_arg_t *ddha = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + + error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds); + if (error != 0) + return (error); + + error = dsl_destroy_head_check_impl(ds, 0); + dsl_dataset_rele(ds, FTAG); + return (error); +} + +static void +dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx) +{ + dsl_dir_t *dd; + dsl_pool_t *dp = dmu_tx_pool(tx); + objset_t *mos = dp->dp_meta_objset; + dd_used_t t; + + ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock)); + + VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd)); + + ASSERT0(dd->dd_phys->dd_head_dataset_obj); + + /* + * Remove our reservation. The impl() routine avoids setting the + * actual property, which would require the (already destroyed) ds. + */ + dsl_dir_set_reservation_sync_impl(dd, 0, tx); + + ASSERT0(dd->dd_phys->dd_used_bytes); + ASSERT0(dd->dd_phys->dd_reserved); + for (t = 0; t < DD_USED_NUM; t++) + ASSERT0(dd->dd_phys->dd_used_breakdown[t]); + + VERIFY0(zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx)); + VERIFY0(zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx)); + VERIFY0(dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx)); + VERIFY0(zap_remove(mos, + dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx)); + + dsl_dir_rele(dd, FTAG); + VERIFY0(dmu_object_free(mos, ddobj, tx)); +} + +void +dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx) +{ + dsl_pool_t *dp = dmu_tx_pool(tx); + objset_t *mos = dp->dp_meta_objset; + uint64_t obj, ddobj, prevobj = 0; + boolean_t rmorigin; + + ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); + ASSERT(ds->ds_prev == NULL || + ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); + ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); + ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); + + /* We need to log before removing it from the namespace. */ + spa_history_log_internal_ds(ds, "destroy", tx, ""); + + rmorigin = (dsl_dir_is_clone(ds->ds_dir) && + DS_IS_DEFER_DESTROY(ds->ds_prev) && + ds->ds_prev->ds_phys->ds_num_children == 2 && + ds->ds_prev->ds_userrefs == 0); + + /* Remove our reservation */ + if (ds->ds_reserved != 0) { + dsl_dataset_set_refreservation_sync_impl(ds, + (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), + 0, tx); + ASSERT0(ds->ds_reserved); + } + + dsl_scan_ds_destroyed(ds, tx); + + obj = ds->ds_object; + + if (ds->ds_phys->ds_prev_snap_obj != 0) { + /* This is a clone */ + ASSERT(ds->ds_prev != NULL); + ASSERT3U(ds->ds_prev->ds_phys->ds_next_snap_obj, !=, obj); + ASSERT0(ds->ds_phys->ds_next_snap_obj); + + dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); + if (ds->ds_prev->ds_phys->ds_next_clones_obj != 0) { + dsl_dataset_remove_from_next_clones(ds->ds_prev, + obj, tx); + } + + ASSERT3U(ds->ds_prev->ds_phys->ds_num_children, >, 1); + ds->ds_prev->ds_phys->ds_num_children--; + } + + zfeature_info_t *async_destroy = + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]; + objset_t *os; + + /* + * Destroy the deadlist. Unless it's a clone, the + * deadlist should be empty. (If it's a clone, it's + * safe to ignore the deadlist contents.) + */ + dsl_deadlist_close(&ds->ds_deadlist); + dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); + dmu_buf_will_dirty(ds->ds_dbuf, tx); + ds->ds_phys->ds_deadlist_obj = 0; + + VERIFY0(dmu_objset_from_ds(ds, &os)); + + if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) { + old_synchronous_dataset_destroy(ds, tx); + } else { + /* + * Move the bptree into the pool's list of trees to + * clean up and update space accounting information. + */ + uint64_t used, comp, uncomp; + + zil_destroy_sync(dmu_objset_zil(os), tx); + + if (!spa_feature_is_active(dp->dp_spa, async_destroy)) { + spa_feature_incr(dp->dp_spa, async_destroy, tx); + dp->dp_bptree_obj = bptree_alloc(mos, tx); + VERIFY0(zap_add(mos, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, + &dp->dp_bptree_obj, tx)); + } + + used = ds->ds_dir->dd_phys->dd_used_bytes; + comp = ds->ds_dir->dd_phys->dd_compressed_bytes; + uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes; + + ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || + ds->ds_phys->ds_unique_bytes == used); + + bptree_add(mos, dp->dp_bptree_obj, + &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg, + used, comp, uncomp, tx); + dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, + -used, -comp, -uncomp, tx); + dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, + used, comp, uncomp, tx); + } + + if (ds->ds_prev != NULL) { + if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { + VERIFY0(zap_remove_int(mos, + ds->ds_prev->ds_dir->dd_phys->dd_clones, + ds->ds_object, tx)); + } + prevobj = ds->ds_prev->ds_object; + dsl_dataset_rele(ds->ds_prev, ds); + ds->ds_prev = NULL; + } + + /* + * This must be done after the dsl_traverse(), because it will + * re-open the objset. + */ + if (ds->ds_objset) { + dmu_objset_evict(ds->ds_objset); + ds->ds_objset = NULL; + } + + /* Erase the link in the dir */ + dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); + ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; + ddobj = ds->ds_dir->dd_object; + ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); + VERIFY0(zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx)); + + spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); + + ASSERT0(ds->ds_phys->ds_next_clones_obj); + ASSERT0(ds->ds_phys->ds_props_obj); + ASSERT0(ds->ds_phys->ds_userrefs_obj); + dsl_dir_rele(ds->ds_dir, ds); + ds->ds_dir = NULL; + VERIFY0(dmu_object_free(mos, obj, tx)); + + dsl_dir_destroy_sync(ddobj, tx); + + if (rmorigin) { + dsl_dataset_t *prev; + VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev)); + dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx); + dsl_dataset_rele(prev, FTAG); + } +} + +static void +dsl_destroy_head_sync(void *arg, dmu_tx_t *tx) +{ + dsl_destroy_head_arg_t *ddha = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds)); + dsl_destroy_head_sync_impl(ds, tx); + dsl_dataset_rele(ds, FTAG); +} + +static void +dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx) +{ + dsl_destroy_head_arg_t *ddha = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds)); + + /* Mark it as inconsistent on-disk, in case we crash */ + dmu_buf_will_dirty(ds->ds_dbuf, tx); + ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; + + spa_history_log_internal_ds(ds, "destroy begin", tx, ""); + dsl_dataset_rele(ds, FTAG); +} + +int +dsl_destroy_head(const char *name) +{ + dsl_destroy_head_arg_t ddha; + int error; + spa_t *spa; + boolean_t isenabled; + +#ifdef _KERNEL + zfs_destroy_unmount_origin(name); +#endif + + error = spa_open(name, &spa, FTAG); + if (error != 0) + return (error); + isenabled = spa_feature_is_enabled(spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]); + spa_close(spa, FTAG); + + ddha.ddha_name = name; + + if (!isenabled) { + objset_t *os; + + error = dsl_sync_task(name, dsl_destroy_head_check, + dsl_destroy_head_begin_sync, &ddha, 0); + if (error != 0) + return (error); + + /* + * Head deletion is processed in one txg on old pools; + * remove the objects from open context so that the txg sync + * is not too long. + */ + error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, FTAG, &os); + if (error == 0) { + uint64_t prev_snap_txg = + dmu_objset_ds(os)->ds_phys->ds_prev_snap_txg; + for (uint64_t obj = 0; error == 0; + error = dmu_object_next(os, &obj, FALSE, + prev_snap_txg)) + (void) dmu_free_object(os, obj); + /* sync out all frees */ + txg_wait_synced(dmu_objset_pool(os), 0); + dmu_objset_disown(os, FTAG); + } + } + + return (dsl_sync_task(name, dsl_destroy_head_check, + dsl_destroy_head_sync, &ddha, 0)); +} + +/* + * Note, this function is used as the callback for dmu_objset_find(). We + * always return 0 so that we will continue to find and process + * inconsistent datasets, even if we encounter an error trying to + * process one of them. + */ +/* ARGSUSED */ +int +dsl_destroy_inconsistent(const char *dsname, void *arg) +{ + objset_t *os; + + if (dmu_objset_hold(dsname, FTAG, &os) == 0) { + boolean_t inconsistent = DS_IS_INCONSISTENT(dmu_objset_ds(os)); + dmu_objset_rele(os, FTAG); + if (inconsistent) + (void) dsl_destroy_head(dsname); + } + return (0); +} diff --git a/uts/common/fs/zfs/dsl_dir.c b/uts/common/fs/zfs/dsl_dir.c index 5ccb6862e9d0..1e7ba6d6cbef 100644 --- a/uts/common/fs/zfs/dsl_dir.c +++ b/uts/common/fs/zfs/dsl_dir.c @@ -40,8 +40,6 @@ #include "zfs_namecheck.h" static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); -static void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, - uint64_t value, dmu_tx_t *tx); /* ARGSUSED */ static void @@ -58,7 +56,7 @@ dsl_dir_evict(dmu_buf_t *db, void *arg) } if (dd->dd_parent) - dsl_dir_close(dd->dd_parent, dd); + dsl_dir_rele(dd->dd_parent, dd); spa_close(dd->dd_pool->dp_spa, dd); @@ -72,18 +70,17 @@ dsl_dir_evict(dmu_buf_t *db, void *arg) } int -dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, +dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, const char *tail, void *tag, dsl_dir_t **ddp) { dmu_buf_t *dbuf; dsl_dir_t *dd; int err; - ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || - dsl_pool_sync_context(dp)); + ASSERT(dsl_pool_config_held(dp)); err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); - if (err) + if (err != 0) return (err); dd = dmu_buf_get_user(dbuf); #ifdef ZFS_DEBUG @@ -110,9 +107,9 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, dsl_dir_snap_cmtime_update(dd); if (dd->dd_phys->dd_parent_obj) { - err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj, + err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj, NULL, dd, &dd->dd_parent); - if (err) + if (err != 0) goto errout; if (tail) { #ifdef ZFS_DEBUG @@ -129,7 +126,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, dd->dd_parent->dd_phys->dd_child_dir_zapobj, ddobj, 0, dd->dd_myname); } - if (err) + if (err != 0) goto errout; } else { (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); @@ -146,7 +143,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, */ err = dmu_bonus_hold(dp->dp_meta_objset, dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus); - if (err) + if (err != 0) goto errout; origin_phys = origin_bonus->db_data; dd->dd_origin_txg = @@ -158,7 +155,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, dsl_dir_evict); if (winner) { if (dd->dd_parent) - dsl_dir_close(dd->dd_parent, dd); + dsl_dir_rele(dd->dd_parent, dd); mutex_destroy(&dd->dd_lock); kmem_free(dd, sizeof (dsl_dir_t)); dd = winner; @@ -185,7 +182,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, errout: if (dd->dd_parent) - dsl_dir_close(dd->dd_parent, dd); + dsl_dir_rele(dd->dd_parent, dd); mutex_destroy(&dd->dd_lock); kmem_free(dd, sizeof (dsl_dir_t)); dmu_buf_rele(dbuf, tag); @@ -193,7 +190,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, } void -dsl_dir_close(dsl_dir_t *dd, void *tag) +dsl_dir_rele(dsl_dir_t *dd, void *tag) { dprintf_dd(dd, "%s\n", ""); spa_close(dd->dd_pool->dp_spa, tag); @@ -250,6 +247,7 @@ static int getcomponent(const char *path, char *component, const char **nextp) { char *p; + if ((path == NULL) || (path[0] == '\0')) return (ENOENT); /* This would be a good place to reserve some namespace... */ @@ -272,10 +270,10 @@ getcomponent(const char *path, char *component, const char **nextp) (void) strcpy(component, path); p = NULL; } else if (p[0] == '/') { - if (p-path >= MAXNAMELEN) + if (p - path >= MAXNAMELEN) return (ENAMETOOLONG); (void) strncpy(component, path, p - path); - component[p-path] = '\0'; + component[p - path] = '\0'; p++; } else if (p[0] == '@') { /* @@ -284,65 +282,54 @@ getcomponent(const char *path, char *component, const char **nextp) */ if (strchr(path, '/')) return (EINVAL); - if (p-path >= MAXNAMELEN) + if (p - path >= MAXNAMELEN) return (ENAMETOOLONG); (void) strncpy(component, path, p - path); - component[p-path] = '\0'; + component[p - path] = '\0'; } else { - ASSERT(!"invalid p"); + panic("invalid p=%p", (void *)p); } *nextp = p; return (0); } /* - * same as dsl_open_dir, ignore the first component of name and use the - * spa instead + * Return the dsl_dir_t, and possibly the last component which couldn't + * be found in *tail. The name must be in the specified dsl_pool_t. This + * thread must hold the dp_config_rwlock for the pool. Returns NULL if the + * path is bogus, or if tail==NULL and we couldn't parse the whole name. + * (*tail)[0] == '@' means that the last component is a snapshot. */ int -dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, +dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, dsl_dir_t **ddp, const char **tailp) { char buf[MAXNAMELEN]; - const char *next, *nextnext = NULL; + const char *spaname, *next, *nextnext = NULL; int err; dsl_dir_t *dd; - dsl_pool_t *dp; uint64_t ddobj; - int openedspa = FALSE; - - dprintf("%s\n", name); err = getcomponent(name, buf, &next); - if (err) + if (err != 0) return (err); - if (spa == NULL) { - err = spa_open(buf, &spa, FTAG); - if (err) { - dprintf("spa_open(%s) failed\n", buf); - return (err); - } - openedspa = TRUE; - /* XXX this assertion belongs in spa_open */ - ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa))); - } + /* Make sure the name is in the specified pool. */ + spaname = spa_name(dp->dp_spa); + if (strcmp(buf, spaname) != 0) + return (EINVAL); - dp = spa_get_dsl(spa); + ASSERT(dsl_pool_config_held(dp)); - rw_enter(&dp->dp_config_rwlock, RW_READER); - err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); - if (err) { - rw_exit(&dp->dp_config_rwlock); - if (openedspa) - spa_close(spa, FTAG); + err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); + if (err != 0) { return (err); } while (next != NULL) { dsl_dir_t *child_ds; err = getcomponent(next, buf, &nextnext); - if (err) + if (err != 0) break; ASSERT(next[0] != '\0'); if (next[0] == '@') @@ -353,25 +340,22 @@ dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, err = zap_lookup(dp->dp_meta_objset, dd->dd_phys->dd_child_dir_zapobj, buf, sizeof (ddobj), 1, &ddobj); - if (err) { + if (err != 0) { if (err == ENOENT) err = 0; break; } - err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds); - if (err) + err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds); + if (err != 0) break; - dsl_dir_close(dd, tag); + dsl_dir_rele(dd, tag); dd = child_ds; next = nextnext; } - rw_exit(&dp->dp_config_rwlock); - if (err) { - dsl_dir_close(dd, tag); - if (openedspa) - spa_close(spa, FTAG); + if (err != 0) { + dsl_dir_rele(dd, tag); return (err); } @@ -382,30 +366,16 @@ dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, if (next != NULL && (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { /* bad path name */ - dsl_dir_close(dd, tag); + dsl_dir_rele(dd, tag); dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); err = ENOENT; } - if (tailp) + if (tailp != NULL) *tailp = next; - if (openedspa) - spa_close(spa, FTAG); *ddp = dd; return (err); } -/* - * Return the dsl_dir_t, and possibly the last component which couldn't - * be found in *tail. Return NULL if the path is bogus, or if - * tail==NULL and we couldn't parse the whole name. (*tail)[0] == '@' - * means that the last component is a snapshot. - */ -int -dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp) -{ - return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp)); -} - uint64_t dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, dmu_tx_t *tx) @@ -443,71 +413,6 @@ dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, return (ddobj); } -/* ARGSUSED */ -int -dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) -{ - dsl_dir_t *dd = arg1; - dsl_pool_t *dp = dd->dd_pool; - objset_t *mos = dp->dp_meta_objset; - int err; - uint64_t count; - - /* - * There should be exactly two holds, both from - * dsl_dataset_destroy: one on the dd directory, and one on its - * head ds. If there are more holds, then a concurrent thread is - * performing a lookup inside this dir while we're trying to destroy - * it. To minimize this possibility, we perform this check only - * in syncing context and fail the operation if we encounter - * additional holds. The dp_config_rwlock ensures that nobody else - * opens it after we check. - */ - if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2) - return (EBUSY); - - err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count); - if (err) - return (err); - if (count != 0) - return (EEXIST); - - return (0); -} - -void -dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) -{ - dsl_dir_t *dd = arg1; - objset_t *mos = dd->dd_pool->dp_meta_objset; - uint64_t obj; - dd_used_t t; - - ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock)); - ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); - - /* - * Remove our reservation. The impl() routine avoids setting the - * actual property, which would require the (already destroyed) ds. - */ - dsl_dir_set_reservation_sync_impl(dd, 0, tx); - - ASSERT0(dd->dd_phys->dd_used_bytes); - ASSERT0(dd->dd_phys->dd_reserved); - for (t = 0; t < DD_USED_NUM; t++) - ASSERT0(dd->dd_phys->dd_used_breakdown[t]); - - VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx)); - VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx)); - VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx)); - VERIFY(0 == zap_remove(mos, - dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx)); - - obj = dd->dd_object; - dsl_dir_close(dd, tag); - VERIFY(0 == dmu_object_free(mos, obj, tx)); -} - boolean_t dsl_dir_is_clone(dsl_dir_t *dd) { @@ -545,18 +450,16 @@ dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) } mutex_exit(&dd->dd_lock); - rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); if (dsl_dir_is_clone(dd)) { dsl_dataset_t *ds; char buf[MAXNAMELEN]; - VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, + VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, dd->dd_phys->dd_origin_obj, FTAG, &ds)); dsl_dataset_name(ds, buf); dsl_dataset_rele(ds, FTAG); dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); } - rw_exit(&dd->dd_pool->dp_config_rwlock); } void @@ -566,7 +469,7 @@ dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) ASSERT(dd->dd_phys); - if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) { + if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { /* up the hold count until we can be written out */ dmu_buf_add_ref(dd->dd_dbuf, dd); } @@ -853,7 +756,7 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, FALSE, asize > usize, tr_list, tx, TRUE); } - if (err) + if (err != 0) dsl_dir_tempreserve_clear(tr_list, tx); else *tr_cookiep = tr_list; @@ -1004,115 +907,123 @@ dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, mutex_exit(&dd->dd_lock); } +typedef struct dsl_dir_set_qr_arg { + const char *ddsqra_name; + zprop_source_t ddsqra_source; + uint64_t ddsqra_value; +} dsl_dir_set_qr_arg_t; + static int -dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_dir_t *dd = ds->ds_dir; - dsl_prop_setarg_t *psa = arg2; - int err; - uint64_t towrite; + dsl_dir_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + uint64_t towrite, newval; - if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) - return (err); + error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); + if (error != 0) + return (error); - if (psa->psa_effective_value == 0) + error = dsl_prop_predict(ds->ds_dir, "quota", + ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } + + if (newval == 0) { + dsl_dataset_rele(ds, FTAG); return (0); + } - mutex_enter(&dd->dd_lock); + mutex_enter(&ds->ds_dir->dd_lock); /* * If we are doing the preliminary check in open context, and * there are pending changes, then don't fail it, since the * pending changes could under-estimate the amount of space to be * freed up. */ - towrite = dsl_dir_space_towrite(dd); + towrite = dsl_dir_space_towrite(ds->ds_dir); if ((dmu_tx_is_syncing(tx) || towrite == 0) && - (psa->psa_effective_value < dd->dd_phys->dd_reserved || - psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) { - err = ENOSPC; + (newval < ds->ds_dir->dd_phys->dd_reserved || + newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) { + error = ENOSPC; } - mutex_exit(&dd->dd_lock); - return (err); + mutex_exit(&ds->ds_dir->dd_lock); + dsl_dataset_rele(ds, FTAG); + return (error); } -extern dsl_syncfunc_t dsl_prop_set_sync; - static void -dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_dir_t *dd = ds->ds_dir; - dsl_prop_setarg_t *psa = arg2; - uint64_t effective_value = psa->psa_effective_value; + dsl_dir_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + uint64_t newval; - dsl_prop_set_sync(ds, psa, tx); - DSL_PROP_CHECK_PREDICTION(dd, psa); + VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); - dmu_buf_will_dirty(dd->dd_dbuf, tx); + dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), + ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, + &ddsqra->ddsqra_value, tx); - mutex_enter(&dd->dd_lock); - dd->dd_phys->dd_quota = effective_value; - mutex_exit(&dd->dd_lock); + VERIFY0(dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); + + dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); + mutex_enter(&ds->ds_dir->dd_lock); + ds->ds_dir->dd_phys->dd_quota = newval; + mutex_exit(&ds->ds_dir->dd_lock); + dsl_dataset_rele(ds, FTAG); } int dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) { - dsl_dir_t *dd; - dsl_dataset_t *ds; - dsl_prop_setarg_t psa; - int err; + dsl_dir_set_qr_arg_t ddsqra; - dsl_prop_setarg_init_uint64(&psa, "quota", source, "a); + ddsqra.ddsqra_name = ddname; + ddsqra.ddsqra_source = source; + ddsqra.ddsqra_value = quota; - err = dsl_dataset_hold(ddname, FTAG, &ds); - if (err) - return (err); - - err = dsl_dir_open(ddname, FTAG, &dd, NULL); - if (err) { - dsl_dataset_rele(ds, FTAG); - return (err); - } - - ASSERT(ds->ds_dir == dd); - - /* - * If someone removes a file, then tries to set the quota, we want to - * make sure the file freeing takes effect. - */ - txg_wait_open(dd->dd_pool, 0); - - err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check, - dsl_dir_set_quota_sync, ds, &psa, 0); - - dsl_dir_close(dd, FTAG); - dsl_dataset_rele(ds, FTAG); - return (err); + return (dsl_sync_task(ddname, dsl_dir_set_quota_check, + dsl_dir_set_quota_sync, &ddsqra, 0)); } int -dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_dir_t *dd = ds->ds_dir; - dsl_prop_setarg_t *psa = arg2; - uint64_t effective_value; - uint64_t used, avail; - int err; + dsl_dir_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + dsl_dir_t *dd; + uint64_t newval, used, avail; + int error; - if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) - return (err); - - effective_value = psa->psa_effective_value; + error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); + if (error != 0) + return (error); + dd = ds->ds_dir; /* * If we are doing the preliminary check in open context, the * space estimates may be inaccurate. */ - if (!dmu_tx_is_syncing(tx)) + if (!dmu_tx_is_syncing(tx)) { + dsl_dataset_rele(ds, FTAG); return (0); + } + + error = dsl_prop_predict(ds->ds_dir, + zfs_prop_to_name(ZFS_PROP_RESERVATION), + ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); + if (error != 0) { + dsl_dataset_rele(ds, FTAG); + return (error); + } mutex_enter(&dd->dd_lock); used = dd->dd_phys->dd_used_bytes; @@ -1125,21 +1036,21 @@ dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; } - if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) { - uint64_t delta = MAX(used, effective_value) - + if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) { + uint64_t delta = MAX(used, newval) - MAX(used, dd->dd_phys->dd_reserved); - if (delta > avail) - return (ENOSPC); - if (dd->dd_phys->dd_quota > 0 && - effective_value > dd->dd_phys->dd_quota) - return (ENOSPC); + if (delta > avail || + (dd->dd_phys->dd_quota > 0 && + newval > dd->dd_phys->dd_quota)) + error = ENOSPC; } - return (0); + dsl_dataset_rele(ds, FTAG); + return (error); } -static void +void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) { uint64_t used; @@ -1162,48 +1073,38 @@ dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) static void -dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_dir_t *dd = ds->ds_dir; - dsl_prop_setarg_t *psa = arg2; - uint64_t value = psa->psa_effective_value; + dsl_dir_set_qr_arg_t *ddsqra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + uint64_t newval; - dsl_prop_set_sync(ds, psa, tx); - DSL_PROP_CHECK_PREDICTION(dd, psa); + VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); - dsl_dir_set_reservation_sync_impl(dd, value, tx); + dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_RESERVATION), + ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, + &ddsqra->ddsqra_value, tx); + + VERIFY0(dsl_prop_get_int_ds(ds, + zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); + + dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); + dsl_dataset_rele(ds, FTAG); } int dsl_dir_set_reservation(const char *ddname, zprop_source_t source, uint64_t reservation) { - dsl_dir_t *dd; - dsl_dataset_t *ds; - dsl_prop_setarg_t psa; - int err; + dsl_dir_set_qr_arg_t ddsqra; - dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation); + ddsqra.ddsqra_name = ddname; + ddsqra.ddsqra_source = source; + ddsqra.ddsqra_value = reservation; - err = dsl_dataset_hold(ddname, FTAG, &ds); - if (err) - return (err); - - err = dsl_dir_open(ddname, FTAG, &dd, NULL); - if (err) { - dsl_dataset_rele(ds, FTAG); - return (err); - } - - ASSERT(ds->ds_dir == dd); - - err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check, - dsl_dir_set_reservation_sync, ds, &psa, 0); - - dsl_dir_close(dd, FTAG); - dsl_dataset_rele(ds, FTAG); - return (err); + return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, + dsl_dir_set_reservation_sync, &ddsqra, 0)); } static dsl_dir_t * @@ -1235,79 +1136,123 @@ would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) return (would_change(dd->dd_parent, delta, ancestor)); } -struct renamearg { - dsl_dir_t *newparent; - const char *mynewname; -}; +typedef struct dsl_dir_rename_arg { + const char *ddra_oldname; + const char *ddra_newname; +} dsl_dir_rename_arg_t; + +/* ARGSUSED */ +static int +dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) +{ + int *deltap = arg; + char namebuf[MAXNAMELEN]; + + dsl_dataset_name(ds, namebuf); + + if (strlen(namebuf) + *deltap >= MAXNAMELEN) + return (ENAMETOOLONG); + return (0); +} static int -dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_rename_check(void *arg, dmu_tx_t *tx) { - dsl_dir_t *dd = arg1; - struct renamearg *ra = arg2; - dsl_pool_t *dp = dd->dd_pool; - objset_t *mos = dp->dp_meta_objset; - int err; - uint64_t val; + dsl_dir_rename_arg_t *ddra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dir_t *dd, *newparent; + const char *mynewname; + int error; + int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); - /* - * There should only be one reference, from dmu_objset_rename(). - * Fleeting holds are also possible (eg, from "zfs list" getting - * stats), but any that are present in open context will likely - * be gone by syncing context, so only fail from syncing - * context. - */ - if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 1) - return (EBUSY); + /* target dir should exist */ + error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); + if (error != 0) + return (error); - /* check for existing name */ - err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, - ra->mynewname, 8, 1, &val); - if (err == 0) + /* new parent should exist */ + error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, + &newparent, &mynewname); + if (error != 0) { + dsl_dir_rele(dd, FTAG); + return (error); + } + + /* can't rename to different pool */ + if (dd->dd_pool != newparent->dd_pool) { + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); + return (ENXIO); + } + + /* new name should not already exist */ + if (mynewname == NULL) { + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); return (EEXIST); - if (err != ENOENT) - return (err); + } - if (ra->newparent != dd->dd_parent) { + /* if the name length is growing, validate child name lengths */ + if (delta > 0) { + error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, + &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); + if (error != 0) { + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); + return (error); + } + } + + if (newparent != dd->dd_parent) { /* is there enough space? */ uint64_t myspace = MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved); /* no rename into our descendant */ - if (closest_common_ancestor(dd, ra->newparent) == dd) + if (closest_common_ancestor(dd, newparent) == dd) { + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); return (EINVAL); + } - if (err = dsl_dir_transfer_possible(dd->dd_parent, - ra->newparent, myspace)) - return (err); + error = dsl_dir_transfer_possible(dd->dd_parent, + newparent, myspace); + if (error != 0) { + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); + return (error); + } } + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); return (0); } static void -dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) { - dsl_dir_t *dd = arg1; - struct renamearg *ra = arg2; - dsl_pool_t *dp = dd->dd_pool; + dsl_dir_rename_arg_t *ddra = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dir_t *dd, *newparent; + const char *mynewname; + int error; objset_t *mos = dp->dp_meta_objset; - int err; - char namebuf[MAXNAMELEN]; - ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2); + VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); + VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, + &mynewname)); /* Log this before we change the name. */ - dsl_dir_name(ra->newparent, namebuf); spa_history_log_internal_dd(dd, "rename", tx, - "-> %s/%s", namebuf, ra->mynewname); + "-> %s", ddra->ddra_newname); - if (ra->newparent != dd->dd_parent) { + if (newparent != dd->dd_parent) { dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, -dd->dd_phys->dd_used_bytes, -dd->dd_phys->dd_compressed_bytes, -dd->dd_phys->dd_uncompressed_bytes, tx); - dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD, + dsl_dir_diduse_space(newparent, DD_USED_CHILD, dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_compressed_bytes, dd->dd_phys->dd_uncompressed_bytes, tx); @@ -1318,7 +1263,7 @@ dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, -unused_rsrv, 0, 0, tx); - dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV, + dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, unused_rsrv, 0, 0, tx); } } @@ -1326,52 +1271,36 @@ dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) dmu_buf_will_dirty(dd->dd_dbuf, tx); /* remove from old parent zapobj */ - err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, + error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx); - ASSERT0(err); + ASSERT0(error); - (void) strcpy(dd->dd_myname, ra->mynewname); - dsl_dir_close(dd->dd_parent, dd); - dd->dd_phys->dd_parent_obj = ra->newparent->dd_object; - VERIFY(0 == dsl_dir_open_obj(dd->dd_pool, - ra->newparent->dd_object, NULL, dd, &dd->dd_parent)); + (void) strcpy(dd->dd_myname, mynewname); + dsl_dir_rele(dd->dd_parent, dd); + dd->dd_phys->dd_parent_obj = newparent->dd_object; + VERIFY0(dsl_dir_hold_obj(dp, + newparent->dd_object, NULL, dd, &dd->dd_parent)); /* add to new parent zapobj */ - err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, - dd->dd_myname, 8, 1, &dd->dd_object, tx); - ASSERT0(err); + VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj, + dd->dd_myname, 8, 1, &dd->dd_object, tx)); + dsl_prop_notify_all(dd); + + dsl_dir_rele(newparent, FTAG); + dsl_dir_rele(dd, FTAG); } int -dsl_dir_rename(dsl_dir_t *dd, const char *newname) +dsl_dir_rename(const char *oldname, const char *newname) { - struct renamearg ra; - int err; + dsl_dir_rename_arg_t ddra; - /* new parent should exist */ - err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname); - if (err) - return (err); + ddra.ddra_oldname = oldname; + ddra.ddra_newname = newname; - /* can't rename to different pool */ - if (dd->dd_pool != ra.newparent->dd_pool) { - err = ENXIO; - goto out; - } - - /* new name should not already exist */ - if (ra.mynewname == NULL) { - err = EEXIST; - goto out; - } - - err = dsl_sync_task_do(dd->dd_pool, - dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3); - -out: - dsl_dir_close(ra.newparent, FTAG); - return (err); + return (dsl_sync_task(oldname, + dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3)); } int diff --git a/uts/common/fs/zfs/dsl_pool.c b/uts/common/fs/zfs/dsl_pool.c index 38d656a43bea..6af631679e9e 100644 --- a/uts/common/fs/zfs/dsl_pool.c +++ b/uts/common/fs/zfs/dsl_pool.c @@ -43,6 +43,7 @@ #include #include #include +#include int zfs_no_write_throttle = 0; int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ @@ -69,7 +70,7 @@ dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) if (err) return (err); - return (dsl_dir_open_obj(dp, obj, name, dp, ddp)); + return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); } static dsl_pool_t * @@ -81,7 +82,7 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg) dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); dp->dp_spa = spa; dp->dp_meta_rootbp = *bp; - rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL); + rrw_init(&dp->dp_config_rwlock, B_TRUE); dp->dp_write_limit = zfs_write_limit_min; txg_init(dp, txg); @@ -92,7 +93,7 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg) txg_list_create(&dp->dp_dirty_dirs, offsetof(dsl_dir_t, dd_dirty_link)); txg_list_create(&dp->dp_sync_tasks, - offsetof(dsl_sync_task_group_t, dstg_node)); + offsetof(dsl_sync_task_t, dst_node)); mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); @@ -126,14 +127,14 @@ dsl_pool_open(dsl_pool_t *dp) dsl_dataset_t *ds; uint64_t obj; - rw_enter(&dp->dp_config_rwlock, RW_WRITER); + rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &dp->dp_root_dir_obj); if (err) goto out; - err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, + err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, dp, &dp->dp_root_dir); if (err) goto out; @@ -154,7 +155,7 @@ dsl_pool_open(dsl_pool_t *dp) &dp->dp_origin_snap); dsl_dataset_rele(ds, FTAG); } - dsl_dir_close(dd, dp); + dsl_dir_rele(dd, dp); if (err) goto out; } @@ -169,7 +170,7 @@ dsl_pool_open(dsl_pool_t *dp) DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); if (err) goto out; - VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, + VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); } @@ -202,7 +203,7 @@ dsl_pool_open(dsl_pool_t *dp) err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); out: - rw_exit(&dp->dp_config_rwlock); + rrw_exit(&dp->dp_config_rwlock, FTAG); return (err); } @@ -217,13 +218,13 @@ dsl_pool_close(dsl_pool_t *dp) * and not a hold, so just drop that here. */ if (dp->dp_origin_snap) - dsl_dataset_drop_ref(dp->dp_origin_snap, dp); + dsl_dataset_rele(dp->dp_origin_snap, dp); if (dp->dp_mos_dir) - dsl_dir_close(dp->dp_mos_dir, dp); + dsl_dir_rele(dp->dp_mos_dir, dp); if (dp->dp_free_dir) - dsl_dir_close(dp->dp_free_dir, dp); + dsl_dir_rele(dp->dp_free_dir, dp); if (dp->dp_root_dir) - dsl_dir_close(dp->dp_root_dir, dp); + dsl_dir_rele(dp->dp_root_dir, dp); bpobj_close(&dp->dp_free_bpobj); @@ -239,7 +240,7 @@ dsl_pool_close(dsl_pool_t *dp) arc_flush(dp->dp_spa); txg_fini(dp); dsl_scan_fini(dp); - rw_destroy(&dp->dp_config_rwlock); + rrw_destroy(&dp->dp_config_rwlock); mutex_destroy(&dp->dp_lock); taskq_destroy(dp->dp_vnrele_taskq); if (dp->dp_blkstats) @@ -257,6 +258,8 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) dsl_dataset_t *ds; uint64_t obj; + rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); + /* create and open the MOS (meta-objset) */ dp->dp_meta_objset = dmu_objset_create_impl(spa, NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); @@ -267,30 +270,30 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) ASSERT0(err); /* Initialize scan structures */ - VERIFY3U(0, ==, dsl_scan_init(dp, txg)); + VERIFY0(dsl_scan_init(dp, txg)); /* create and open the root dir */ dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); - VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj, + VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, dp, &dp->dp_root_dir)); /* create and open the meta-objset dir */ (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); - VERIFY(0 == dsl_pool_open_special_dir(dp, + VERIFY0(dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir)); if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { /* create and open the free dir */ (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); - VERIFY(0 == dsl_pool_open_special_dir(dp, + VERIFY0(dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir)); /* create and open the free_bplist */ obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); - VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, + VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); } @@ -301,7 +304,7 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); /* create the root objset */ - VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); + VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); os = dmu_objset_create_impl(dp->dp_spa, ds, dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); #ifdef _KERNEL @@ -311,6 +314,8 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) dmu_tx_commit(tx); + rrw_exit(&dp->dp_config_rwlock, FTAG); + return (dp); } @@ -333,10 +338,7 @@ static int deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_deadlist_t *dl = arg; - dsl_pool_t *dp = dmu_objset_pool(dl->dl_os); - rw_enter(&dp->dp_config_rwlock, RW_READER); dsl_deadlist_insert(dl, bp, tx); - rw_exit(&dp->dp_config_rwlock); return (0); } @@ -358,7 +360,7 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) /* * We need to copy dp_space_towrite() before doing - * dsl_sync_task_group_sync(), because + * dsl_sync_task_sync(), because * dsl_dataset_snapshot_reserve_space() will increase * dp_space_towrite but not actually write anything. */ @@ -472,14 +474,14 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) */ DTRACE_PROBE(pool_sync__3task); if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { - dsl_sync_task_group_t *dstg; + dsl_sync_task_t *dst; /* * No more sync tasks should have been added while we * were syncing. */ ASSERT(spa_sync_pass(dp->dp_spa) == 1); - while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) - dsl_sync_task_group_sync(dstg, tx); + while (dst = txg_list_remove(&dp->dp_sync_tasks, txg)) + dsl_sync_task_sync(dst, tx); } dmu_tx_commit(tx); @@ -654,14 +656,13 @@ dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) /* ARGSUSED */ static int -upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) +upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dmu_tx_t *tx = arg; dsl_dataset_t *ds, *prev = NULL; int err; - dsl_pool_t *dp = spa_get_dsl(spa); - err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); + err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); @@ -687,7 +688,7 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) * The $ORIGIN can't have any data, or the accounting * will be wrong. */ - ASSERT(prev->ds_phys->ds_bp.blk_birth == 0); + ASSERT0(prev->ds_phys->ds_bp.blk_birth); /* The origin doesn't get attached to itself */ if (ds->ds_object == prev->ds_object) { @@ -707,13 +708,13 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) if (ds->ds_phys->ds_next_snap_obj == 0) { ASSERT(ds->ds_prev == NULL); - VERIFY(0 == dsl_dataset_hold_obj(dp, + VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); } } - ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object); - ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object); + ASSERT3U(ds->ds_dir->dd_phys->dd_origin_obj, ==, prev->ds_object); + ASSERT3U(ds->ds_phys->ds_prev_snap_obj, ==, prev->ds_object); if (prev->ds_phys->ds_next_clones_obj == 0) { dmu_buf_will_dirty(prev->ds_dbuf, tx); @@ -721,7 +722,7 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) zap_create(dp->dp_meta_objset, DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); } - VERIFY(0 == zap_add_int(dp->dp_meta_objset, + VERIFY0(zap_add_int(dp->dp_meta_objset, prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); dsl_dataset_rele(ds, FTAG); @@ -736,25 +737,21 @@ dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dp->dp_origin_snap != NULL); - VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb, + VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, tx, DS_FIND_CHILDREN)); } /* ARGSUSED */ static int -upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) +upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) { dmu_tx_t *tx = arg; - dsl_dataset_t *ds; - dsl_pool_t *dp = spa_get_dsl(spa); objset_t *mos = dp->dp_meta_objset; - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); - - if (ds->ds_dir->dd_phys->dd_origin_obj) { + if (ds->ds_dir->dd_phys->dd_origin_obj != 0) { dsl_dataset_t *origin; - VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, + VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); if (origin->ds_dir->dd_phys->dd_clones == 0) { @@ -763,13 +760,11 @@ upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } - VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, - origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); + VERIFY0(zap_add_int(dp->dp_meta_objset, + origin->ds_dir->dd_phys->dd_clones, ds->ds_object, tx)); dsl_dataset_rele(origin, FTAG); } - - dsl_dataset_rele(ds, FTAG); return (0); } @@ -780,7 +775,7 @@ dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) uint64_t obj; (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); - VERIFY(0 == dsl_pool_open_special_dir(dp, + VERIFY0(dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir)); /* @@ -790,12 +785,11 @@ dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) */ obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); - VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, + VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); - VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, - dp->dp_meta_objset, obj)); + VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); - VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, + VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); } @@ -807,17 +801,16 @@ dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dp->dp_origin_snap == NULL); + ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); /* create the origin dir, ds, & snap-ds */ - rw_enter(&dp->dp_config_rwlock, RW_WRITER); dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, NULL, 0, kcred, tx); - VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); - dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx); - VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, + VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); + dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); + VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, dp, &dp->dp_origin_snap)); dsl_dataset_rele(ds, FTAG); - rw_exit(&dp->dp_config_rwlock); } taskq_t * @@ -852,7 +845,7 @@ dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) *htag = '\0'; ++htag; dsobj = strtonum(za.za_name, NULL); - (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE); + dsl_dataset_user_release_tmp(dp, dsobj, htag); } zap_cursor_fini(&zc); } @@ -874,7 +867,7 @@ dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) static int dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, - const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding) + const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) { objset_t *mos = dp->dp_meta_objset; uint64_t zapobj = dp->dp_tmp_userrefs_obj; @@ -899,7 +892,7 @@ dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); if (holding) - error = zap_add(mos, zapobj, name, 8, 1, now, tx); + error = zap_add(mos, zapobj, name, 8, 1, &now, tx); else error = zap_remove(mos, zapobj, name, tx); strfree(name); @@ -912,7 +905,7 @@ dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, */ int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, - uint64_t *now, dmu_tx_t *tx) + uint64_t now, dmu_tx_t *tx) { return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); } @@ -927,3 +920,106 @@ dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, tx, B_FALSE)); } + +/* + * DSL Pool Configuration Lock + * + * The dp_config_rwlock protects against changes to DSL state (e.g. dataset + * creation / destruction / rename / property setting). It must be held for + * read to hold a dataset or dsl_dir. I.e. you must call + * dsl_pool_config_enter() or dsl_pool_hold() before calling + * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock + * must be held continuously until all datasets and dsl_dirs are released. + * + * The only exception to this rule is that if a "long hold" is placed on + * a dataset, then the dp_config_rwlock may be dropped while the dataset + * is still held. The long hold will prevent the dataset from being + * destroyed -- the destroy will fail with EBUSY. A long hold can be + * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset + * (by calling dsl_{dataset,objset}_{try}own{_obj}). + * + * Legitimate long-holders (including owners) should be long-running, cancelable + * tasks that should cause "zfs destroy" to fail. This includes DMU + * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), + * "zfs send", and "zfs diff". There are several other long-holders whose + * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). + * + * The usual formula for long-holding would be: + * dsl_pool_hold() + * dsl_dataset_hold() + * ... perform checks ... + * dsl_dataset_long_hold() + * dsl_pool_rele() + * ... perform long-running task ... + * dsl_dataset_long_rele() + * dsl_dataset_rele() + * + * Note that when the long hold is released, the dataset is still held but + * the pool is not held. The dataset may change arbitrarily during this time + * (e.g. it could be destroyed). Therefore you shouldn't do anything to the + * dataset except release it. + * + * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only + * or modifying operations. + * + * Modifying operations should generally use dsl_sync_task(). The synctask + * infrastructure enforces proper locking strategy with respect to the + * dp_config_rwlock. See the comment above dsl_sync_task() for details. + * + * Read-only operations will manually hold the pool, then the dataset, obtain + * information from the dataset, then release the pool and dataset. + * dmu_objset_{hold,rele}() are convenience routines that also do the pool + * hold/rele. + */ + +int +dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) +{ + spa_t *spa; + int error; + + error = spa_open(name, &spa, tag); + if (error == 0) { + *dp = spa_get_dsl(spa); + dsl_pool_config_enter(*dp, tag); + } + return (error); +} + +void +dsl_pool_rele(dsl_pool_t *dp, void *tag) +{ + dsl_pool_config_exit(dp, tag); + spa_close(dp->dp_spa, tag); +} + +void +dsl_pool_config_enter(dsl_pool_t *dp, void *tag) +{ + /* + * We use a "reentrant" reader-writer lock, but not reentrantly. + * + * The rrwlock can (with the track_all flag) track all reading threads, + * which is very useful for debugging which code path failed to release + * the lock, and for verifying that the *current* thread does hold + * the lock. + * + * (Unlike a rwlock, which knows that N threads hold it for + * read, but not *which* threads, so rw_held(RW_READER) returns TRUE + * if any thread holds it for read, even if this thread doesn't). + */ + ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); + rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); +} + +void +dsl_pool_config_exit(dsl_pool_t *dp, void *tag) +{ + rrw_exit(&dp->dp_config_rwlock, tag); +} + +boolean_t +dsl_pool_config_held(dsl_pool_t *dp) +{ + return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); +} diff --git a/uts/common/fs/zfs/dsl_prop.c b/uts/common/fs/zfs/dsl_prop.c index 5bbe14ff691d..cd7c3ecb5f77 100644 --- a/uts/common/fs/zfs/dsl_prop.c +++ b/uts/common/fs/zfs/dsl_prop.c @@ -82,7 +82,7 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname, char *inheritstr; char *recvdstr; - ASSERT(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock)); + ASSERT(dsl_pool_config_held(dd->dd_pool)); if (setpoint) setpoint[0] = '\0'; @@ -97,8 +97,6 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname, * after this loop. */ for (; dd != NULL; dd = dd->dd_parent) { - ASSERT(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock)); - if (dd != target || snapshot) { if (!inheritable) break; @@ -167,7 +165,7 @@ dsl_prop_get_ds(dsl_dataset_t *ds, const char *propname, boolean_t snapshot; uint64_t zapobj; - ASSERT(RW_LOCK_HELD(&ds->ds_dir->dd_pool->dp_config_rwlock)); + ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); inheritable = (prop == ZPROP_INVAL || zfs_prop_inheritable(prop)); snapshot = (ds->ds_phys != NULL && dsl_dataset_is_snapshot(ds)); zapobj = (ds->ds_phys == NULL ? 0 : ds->ds_phys->ds_props_obj); @@ -235,18 +233,12 @@ dsl_prop_register(dsl_dataset_t *ds, const char *propname, uint64_t value; dsl_prop_cb_record_t *cbr; int err; - int need_rwlock; - need_rwlock = !RW_WRITE_HELD(&dp->dp_config_rwlock); - if (need_rwlock) - rw_enter(&dp->dp_config_rwlock, RW_READER); + ASSERT(dsl_pool_config_held(dp)); - err = dsl_prop_get_ds(ds, propname, 8, 1, &value, NULL); - if (err != 0) { - if (need_rwlock) - rw_exit(&dp->dp_config_rwlock); + err = dsl_prop_get_int_ds(ds, propname, &value); + if (err != 0) return (err); - } cbr = kmem_alloc(sizeof (dsl_prop_cb_record_t), KM_SLEEP); cbr->cbr_ds = ds; @@ -259,9 +251,6 @@ dsl_prop_register(dsl_dataset_t *ds, const char *propname, mutex_exit(&dd->dd_lock); cbr->cbr_func(cbr->cbr_arg, value); - - if (need_rwlock) - rw_exit(&dp->dp_config_rwlock); return (0); } @@ -269,19 +258,18 @@ int dsl_prop_get(const char *dsname, const char *propname, int intsz, int numints, void *buf, char *setpoint) { - dsl_dataset_t *ds; - int err; + objset_t *os; + int error; - err = dsl_dataset_hold(dsname, FTAG, &ds); - if (err) - return (err); + error = dmu_objset_hold(dsname, FTAG, &os); + if (error != 0) + return (error); - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); - err = dsl_prop_get_ds(ds, propname, intsz, numints, buf, setpoint); - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); + error = dsl_prop_get_ds(dmu_objset_ds(os), propname, + intsz, numints, buf, setpoint); - dsl_dataset_rele(ds, FTAG); - return (err); + dmu_objset_rele(os, FTAG); + return (error); } /* @@ -299,17 +287,11 @@ dsl_prop_get_integer(const char *ddname, const char *propname, return (dsl_prop_get(ddname, propname, 8, 1, valuep, setpoint)); } -void -dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname, - zprop_source_t source, uint64_t *value) +int +dsl_prop_get_int_ds(dsl_dataset_t *ds, const char *propname, + uint64_t *valuep) { - psa->psa_name = propname; - psa->psa_source = source; - psa->psa_intsz = 8; - psa->psa_numints = 1; - psa->psa_value = value; - - psa->psa_effective_value = -1ULL; + return (dsl_prop_get_ds(ds, propname, 8, 1, valuep, NULL)); } /* @@ -323,11 +305,10 @@ dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname, * a property not handled by this function. */ int -dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa) +dsl_prop_predict(dsl_dir_t *dd, const char *propname, + zprop_source_t source, uint64_t value, uint64_t *newvalp) { - const char *propname = psa->psa_name; zfs_prop_t prop = zfs_name_to_prop(propname); - zprop_source_t source = psa->psa_source; objset_t *mos; uint64_t zapobj; uint64_t version; @@ -359,36 +340,33 @@ dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa) switch (source) { case ZPROP_SRC_NONE: /* Revert to the received value, if any. */ - err = zap_lookup(mos, zapobj, recvdstr, 8, 1, - &psa->psa_effective_value); + err = zap_lookup(mos, zapobj, recvdstr, 8, 1, newvalp); if (err == ENOENT) - psa->psa_effective_value = 0; + *newvalp = 0; break; case ZPROP_SRC_LOCAL: - psa->psa_effective_value = *(uint64_t *)psa->psa_value; + *newvalp = value; break; case ZPROP_SRC_RECEIVED: /* * If there's no local setting, then the new received value will * be the effective value. */ - err = zap_lookup(mos, zapobj, propname, 8, 1, - &psa->psa_effective_value); + err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp); if (err == ENOENT) - psa->psa_effective_value = *(uint64_t *)psa->psa_value; + *newvalp = value; break; case (ZPROP_SRC_NONE | ZPROP_SRC_RECEIVED): /* * We're clearing the received value, so the local setting (if * it exists) remains the effective value. */ - err = zap_lookup(mos, zapobj, propname, 8, 1, - &psa->psa_effective_value); + err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp); if (err == ENOENT) - psa->psa_effective_value = 0; + *newvalp = 0; break; default: - cmn_err(CE_PANIC, "unexpected property source: %d", source); + panic("unexpected property source: %d", source); } strfree(recvdstr); @@ -399,37 +377,6 @@ dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa) return (err); } -#ifdef ZFS_DEBUG -void -dsl_prop_check_prediction(dsl_dir_t *dd, dsl_prop_setarg_t *psa) -{ - zfs_prop_t prop = zfs_name_to_prop(psa->psa_name); - uint64_t intval; - char setpoint[MAXNAMELEN]; - uint64_t version = spa_version(dd->dd_pool->dp_spa); - int err; - - if (version < SPA_VERSION_RECVD_PROPS) { - switch (prop) { - case ZFS_PROP_QUOTA: - case ZFS_PROP_RESERVATION: - return; - } - } - - err = dsl_prop_get_dd(dd, psa->psa_name, 8, 1, &intval, - setpoint, B_FALSE); - if (err == 0 && intval != psa->psa_effective_value) { - cmn_err(CE_PANIC, "%s property, source: %x, " - "predicted effective value: %llu, " - "actual effective value: %llu (setpoint: %s)", - psa->psa_name, psa->psa_source, - (unsigned long long)psa->psa_effective_value, - (unsigned long long)intval, setpoint); - } -} -#endif - /* * Unregister this callback. Return 0 on success, ENOENT if ddname is * invalid, ENOMSG if no matching callback registered. @@ -464,25 +411,57 @@ dsl_prop_unregister(dsl_dataset_t *ds, const char *propname, return (0); } -/* - * Return the number of callbacks that are registered for this dataset. - */ -int -dsl_prop_numcb(dsl_dataset_t *ds) +boolean_t +dsl_prop_hascb(dsl_dataset_t *ds) +{ + dsl_dir_t *dd = ds->ds_dir; + boolean_t rv = B_FALSE; + dsl_prop_cb_record_t *cbr; + + mutex_enter(&dd->dd_lock); + for (cbr = list_head(&dd->dd_prop_cbs); cbr; + cbr = list_next(&dd->dd_prop_cbs, cbr)) { + if (cbr->cbr_ds == ds) { + rv = B_TRUE; + break; + } + } + mutex_exit(&dd->dd_lock); + return (rv); +} + +/* ARGSUSED */ +static int +dsl_prop_notify_all_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) { dsl_dir_t *dd = ds->ds_dir; dsl_prop_cb_record_t *cbr; - int num = 0; mutex_enter(&dd->dd_lock); - for (cbr = list_head(&dd->dd_prop_cbs); - cbr; cbr = list_next(&dd->dd_prop_cbs, cbr)) { - if (cbr->cbr_ds == ds) - num++; + for (cbr = list_head(&dd->dd_prop_cbs); cbr; + cbr = list_next(&dd->dd_prop_cbs, cbr)) { + uint64_t value; + + if (dsl_prop_get_ds(cbr->cbr_ds, cbr->cbr_propname, + sizeof (value), 1, &value, NULL) == 0) + cbr->cbr_func(cbr->cbr_arg, value); } mutex_exit(&dd->dd_lock); - return (num); + return (0); +} + +/* + * Update all property values for ddobj & its descendants. This is used + * when renaming the dir. + */ +void +dsl_prop_notify_all(dsl_dir_t *dd) +{ + dsl_pool_t *dp = dd->dd_pool; + ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); + (void) dmu_objset_find_dp(dp, dd->dd_object, dsl_prop_notify_all_cb, + NULL, DS_FIND_CHILDREN); } static void @@ -496,8 +475,8 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj, zap_attribute_t *za; int err; - ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); - err = dsl_dir_open_obj(dp, ddobj, NULL, FTAG, &dd); + ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); + err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd); if (err) return; @@ -508,7 +487,7 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj, */ err = zap_contains(mos, dd->dd_phys->dd_props_zapobj, propname); if (err == 0) { - dsl_dir_close(dd, FTAG); + dsl_dir_rele(dd, FTAG); return; } ASSERT3U(err, ==, ENOENT); @@ -543,26 +522,24 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj, } kmem_free(za, sizeof (zap_attribute_t)); zap_cursor_fini(&zc); - dsl_dir_close(dd, FTAG); + dsl_dir_rele(dd, FTAG); } void -dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname, + zprop_source_t source, int intsz, int numints, const void *value, + dmu_tx_t *tx) { - dsl_dataset_t *ds = arg1; - dsl_prop_setarg_t *psa = arg2; objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t zapobj, intval, dummy; int isint; char valbuf[32]; - char *valstr = NULL; + const char *valstr = NULL; char *inheritstr; char *recvdstr; char *tbuf = NULL; int err; uint64_t version = spa_version(ds->ds_dir->dd_pool->dp_spa); - const char *propname = psa->psa_name; - zprop_source_t source = psa->psa_source; isint = (dodefault(propname, 8, 1, &intval) == 0); @@ -612,8 +589,8 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) */ err = zap_remove(mos, zapobj, inheritstr, tx); ASSERT(err == 0 || err == ENOENT); - VERIFY(0 == zap_update(mos, zapobj, propname, - psa->psa_intsz, psa->psa_numints, psa->psa_value, tx)); + VERIFY0(zap_update(mos, zapobj, propname, + intsz, numints, value, tx)); break; case ZPROP_SRC_INHERITED: /* @@ -624,12 +601,10 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) err = zap_remove(mos, zapobj, propname, tx); ASSERT(err == 0 || err == ENOENT); if (version >= SPA_VERSION_RECVD_PROPS && - dsl_prop_get_ds(ds, ZPROP_HAS_RECVD, 8, 1, &dummy, - NULL) == 0) { + dsl_prop_get_int_ds(ds, ZPROP_HAS_RECVD, &dummy) == 0) { dummy = 0; - err = zap_update(mos, zapobj, inheritstr, - 8, 1, &dummy, tx); - ASSERT(err == 0); + VERIFY0(zap_update(mos, zapobj, inheritstr, + 8, 1, &dummy, tx)); } break; case ZPROP_SRC_RECEIVED: @@ -637,7 +612,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) * set propname$recvd -> value */ err = zap_update(mos, zapobj, recvdstr, - psa->psa_intsz, psa->psa_numints, psa->psa_value, tx); + intsz, numints, value, tx); ASSERT(err == 0); break; case (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED): @@ -667,7 +642,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) strfree(recvdstr); if (isint) { - VERIFY(0 == dsl_prop_get_ds(ds, propname, 8, 1, &intval, NULL)); + VERIFY0(dsl_prop_get_int_ds(ds, propname, &intval)); if (ds->ds_phys != NULL && dsl_dataset_is_snapshot(ds)) { dsl_prop_cb_record_t *cbr; @@ -694,7 +669,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) valstr = valbuf; } else { if (source == ZPROP_SRC_LOCAL) { - valstr = (char *)psa->psa_value; + valstr = value; } else { tbuf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); if (dsl_prop_get_ds(ds, propname, 1, @@ -711,118 +686,73 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) kmem_free(tbuf, ZAP_MAXVALUELEN); } -void -dsl_props_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) +int +dsl_prop_set_int(const char *dsname, const char *propname, + zprop_source_t source, uint64_t value) { - dsl_dataset_t *ds = arg1; - dsl_props_arg_t *pa = arg2; - nvlist_t *props = pa->pa_props; - dsl_prop_setarg_t psa; - nvpair_t *elem = NULL; + nvlist_t *nvl = fnvlist_alloc(); + int error; - psa.psa_source = pa->pa_source; - - while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { - nvpair_t *pair = elem; - - psa.psa_name = nvpair_name(pair); - - if (nvpair_type(pair) == DATA_TYPE_NVLIST) { - /* - * dsl_prop_get_all_impl() returns properties in this - * format. - */ - nvlist_t *attrs; - VERIFY(nvpair_value_nvlist(pair, &attrs) == 0); - VERIFY(nvlist_lookup_nvpair(attrs, ZPROP_VALUE, - &pair) == 0); - } - - if (nvpair_type(pair) == DATA_TYPE_STRING) { - VERIFY(nvpair_value_string(pair, - (char **)&psa.psa_value) == 0); - psa.psa_intsz = 1; - psa.psa_numints = strlen(psa.psa_value) + 1; - } else { - uint64_t intval; - VERIFY(nvpair_value_uint64(pair, &intval) == 0); - psa.psa_intsz = sizeof (intval); - psa.psa_numints = 1; - psa.psa_value = &intval; - } - dsl_prop_set_sync(ds, &psa, tx); - } + fnvlist_add_uint64(nvl, propname, value); + error = dsl_props_set(dsname, source, nvl); + fnvlist_free(nvl); + return (error); } int -dsl_prop_set(const char *dsname, const char *propname, zprop_source_t source, - int intsz, int numints, const void *buf) +dsl_prop_set_string(const char *dsname, const char *propname, + zprop_source_t source, const char *value) { - dsl_dataset_t *ds; - uint64_t version; - int err; - dsl_prop_setarg_t psa; + nvlist_t *nvl = fnvlist_alloc(); + int error; - /* - * We must do these checks before we get to the syncfunc, since - * it can't fail. - */ - if (strlen(propname) >= ZAP_MAXNAMELEN) - return (ENAMETOOLONG); - - err = dsl_dataset_hold(dsname, FTAG, &ds); - if (err) - return (err); - - version = spa_version(ds->ds_dir->dd_pool->dp_spa); - if (intsz * numints >= (version < SPA_VERSION_STMF_PROP ? - ZAP_OLDMAXVALUELEN : ZAP_MAXVALUELEN)) { - dsl_dataset_rele(ds, FTAG); - return (E2BIG); - } - if (dsl_dataset_is_snapshot(ds) && - version < SPA_VERSION_SNAP_PROPS) { - dsl_dataset_rele(ds, FTAG); - return (ENOTSUP); - } - - psa.psa_name = propname; - psa.psa_source = source; - psa.psa_intsz = intsz; - psa.psa_numints = numints; - psa.psa_value = buf; - psa.psa_effective_value = -1ULL; - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - NULL, dsl_prop_set_sync, ds, &psa, 2); - - dsl_dataset_rele(ds, FTAG); - return (err); + fnvlist_add_string(nvl, propname, value); + error = dsl_props_set(dsname, source, nvl); + fnvlist_free(nvl); + return (error); } int -dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props) +dsl_prop_inherit(const char *dsname, const char *propname, + zprop_source_t source) { + nvlist_t *nvl = fnvlist_alloc(); + int error; + + fnvlist_add_boolean(nvl, propname); + error = dsl_props_set(dsname, source, nvl); + fnvlist_free(nvl); + return (error); +} + +typedef struct dsl_props_set_arg { + const char *dpsa_dsname; + zprop_source_t dpsa_source; + nvlist_t *dpsa_props; +} dsl_props_set_arg_t; + +static int +dsl_props_set_check(void *arg, dmu_tx_t *tx) +{ + dsl_props_set_arg_t *dpsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; uint64_t version; nvpair_t *elem = NULL; - dsl_props_arg_t pa; int err; - if (err = dsl_dataset_hold(dsname, FTAG, &ds)) + err = dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds); + if (err != 0) return (err); - /* - * Do these checks before the syncfunc, since it can't fail. - */ + version = spa_version(ds->ds_dir->dd_pool->dp_spa); - while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { + while ((elem = nvlist_next_nvpair(dpsa->dpsa_props, elem)) != NULL) { if (strlen(nvpair_name(elem)) >= ZAP_MAXNAMELEN) { dsl_dataset_rele(ds, FTAG); return (ENAMETOOLONG); } if (nvpair_type(elem) == DATA_TYPE_STRING) { - char *valstr; - VERIFY(nvpair_value_string(elem, &valstr) == 0); + char *valstr = fnvpair_value_string(elem); if (strlen(valstr) >= (version < SPA_VERSION_STMF_PROP ? ZAP_OLDMAXVALUELEN : ZAP_MAXVALUELEN)) { @@ -832,20 +762,83 @@ dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props) } } - if (dsl_dataset_is_snapshot(ds) && - version < SPA_VERSION_SNAP_PROPS) { + if (dsl_dataset_is_snapshot(ds) && version < SPA_VERSION_SNAP_PROPS) { dsl_dataset_rele(ds, FTAG); return (ENOTSUP); } - - pa.pa_props = props; - pa.pa_source = source; - - err = dsl_sync_task_do(ds->ds_dir->dd_pool, - NULL, dsl_props_set_sync, ds, &pa, 2); - dsl_dataset_rele(ds, FTAG); - return (err); + return (0); +} + +void +dsl_props_set_sync_impl(dsl_dataset_t *ds, zprop_source_t source, + nvlist_t *props, dmu_tx_t *tx) +{ + nvpair_t *elem = NULL; + + while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { + nvpair_t *pair = elem; + + if (nvpair_type(pair) == DATA_TYPE_NVLIST) { + /* + * dsl_prop_get_all_impl() returns properties in this + * format. + */ + nvlist_t *attrs = fnvpair_value_nvlist(pair); + pair = fnvlist_lookup_nvpair(attrs, ZPROP_VALUE); + } + + if (nvpair_type(pair) == DATA_TYPE_STRING) { + const char *value = fnvpair_value_string(pair); + dsl_prop_set_sync_impl(ds, nvpair_name(pair), + source, 1, strlen(value) + 1, value, tx); + } else if (nvpair_type(pair) == DATA_TYPE_UINT64) { + uint64_t intval = fnvpair_value_uint64(pair); + dsl_prop_set_sync_impl(ds, nvpair_name(pair), + source, sizeof (intval), 1, &intval, tx); + } else if (nvpair_type(pair) == DATA_TYPE_BOOLEAN) { + dsl_prop_set_sync_impl(ds, nvpair_name(pair), + source, 0, 0, NULL, tx); + } else { + panic("invalid nvpair type"); + } + } +} + +static void +dsl_props_set_sync(void *arg, dmu_tx_t *tx) +{ + dsl_props_set_arg_t *dpsa = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds)); + dsl_props_set_sync_impl(ds, dpsa->dpsa_source, dpsa->dpsa_props, tx); + dsl_dataset_rele(ds, FTAG); +} + +/* + * All-or-nothing; if any prop can't be set, nothing will be modified. + */ +int +dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props) +{ + dsl_props_set_arg_t dpsa; + int nblks = 0; + + dpsa.dpsa_dsname = dsname; + dpsa.dpsa_source = source; + dpsa.dpsa_props = props; + + /* + * If the source includes NONE, then we will only be removing entries + * from the ZAP object. In that case don't check for ENOSPC. + */ + if ((source & ZPROP_SRC_NONE) == 0) + nblks = 2 * fnvlist_num_pairs(props); + + return (dsl_sync_task(dsname, dsl_props_set_check, dsl_props_set_sync, + &dpsa, nblks)); } typedef enum dsl_prop_getflags { @@ -995,7 +988,7 @@ dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp, if (dsl_dataset_is_snapshot(ds)) flags |= DSL_PROP_GET_SNAPSHOT; - rw_enter(&dp->dp_config_rwlock, RW_READER); + ASSERT(dsl_pool_config_held(dp)); if (ds->ds_phys->ds_props_obj != 0) { ASSERT(flags & DSL_PROP_GET_SNAPSHOT); @@ -1020,58 +1013,51 @@ dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp, break; } out: - rw_exit(&dp->dp_config_rwlock); return (err); } boolean_t -dsl_prop_get_hasrecvd(objset_t *os) +dsl_prop_get_hasrecvd(const char *dsname) { - dsl_dataset_t *ds = os->os_dsl_dataset; - int rc; uint64_t dummy; - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); - rc = dsl_prop_get_ds(ds, ZPROP_HAS_RECVD, 8, 1, &dummy, NULL); - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); - ASSERT(rc != 0 || spa_version(os->os_spa) >= SPA_VERSION_RECVD_PROPS); - return (rc == 0); + return (0 == + dsl_prop_get_integer(dsname, ZPROP_HAS_RECVD, &dummy, NULL)); } -static void -dsl_prop_set_hasrecvd_impl(objset_t *os, zprop_source_t source) +static int +dsl_prop_set_hasrecvd_impl(const char *dsname, zprop_source_t source) { - dsl_dataset_t *ds = os->os_dsl_dataset; - uint64_t dummy = 0; - dsl_prop_setarg_t psa; + uint64_t version; + spa_t *spa; + int error = 0; - if (spa_version(os->os_spa) < SPA_VERSION_RECVD_PROPS) - return; + VERIFY0(spa_open(dsname, &spa, FTAG)); + version = spa_version(spa); + spa_close(spa, FTAG); - dsl_prop_setarg_init_uint64(&psa, ZPROP_HAS_RECVD, source, &dummy); - - (void) dsl_sync_task_do(ds->ds_dir->dd_pool, NULL, - dsl_prop_set_sync, ds, &psa, 2); + if (version >= SPA_VERSION_RECVD_PROPS) + error = dsl_prop_set_int(dsname, ZPROP_HAS_RECVD, source, 0); + return (error); } /* * Call after successfully receiving properties to ensure that only the first * receive on or after SPA_VERSION_RECVD_PROPS blows away local properties. */ -void -dsl_prop_set_hasrecvd(objset_t *os) +int +dsl_prop_set_hasrecvd(const char *dsname) { - if (dsl_prop_get_hasrecvd(os)) { - ASSERT(spa_version(os->os_spa) >= SPA_VERSION_RECVD_PROPS); - return; - } - dsl_prop_set_hasrecvd_impl(os, ZPROP_SRC_LOCAL); + int error = 0; + if (!dsl_prop_get_hasrecvd(dsname)) + error = dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_LOCAL); + return (error); } void -dsl_prop_unset_hasrecvd(objset_t *os) +dsl_prop_unset_hasrecvd(const char *dsname) { - dsl_prop_set_hasrecvd_impl(os, ZPROP_SRC_NONE); + VERIFY0(dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_NONE)); } int @@ -1081,16 +1067,25 @@ dsl_prop_get_all(objset_t *os, nvlist_t **nvp) } int -dsl_prop_get_received(objset_t *os, nvlist_t **nvp) +dsl_prop_get_received(const char *dsname, nvlist_t **nvp) { + objset_t *os; + int error; + /* * Received properties are not distinguishable from local properties * until the dataset has received properties on or after * SPA_VERSION_RECVD_PROPS. */ - dsl_prop_getflags_t flags = (dsl_prop_get_hasrecvd(os) ? + dsl_prop_getflags_t flags = (dsl_prop_get_hasrecvd(dsname) ? DSL_PROP_GET_RECEIVED : DSL_PROP_GET_LOCAL); - return (dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, flags)); + + error = dmu_objset_hold(dsname, FTAG, &os); + if (error != 0) + return (error); + error = dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, flags); + dmu_objset_rele(os, FTAG); + return (error); } void diff --git a/uts/common/fs/zfs/dsl_scan.c b/uts/common/fs/zfs/dsl_scan.c index e1717257990d..3de3c6e4d7f9 100644 --- a/uts/common/fs/zfs/dsl_scan.c +++ b/uts/common/fs/zfs/dsl_scan.c @@ -55,7 +55,7 @@ typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *); static scan_cb_t dsl_scan_defrag_cb; static scan_cb_t dsl_scan_scrub_cb; static scan_cb_t dsl_scan_remove_cb; -static dsl_syncfunc_t dsl_scan_cancel_sync; +static void dsl_scan_cancel_sync(void *, dmu_tx_t *); static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *tx); int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */ @@ -154,9 +154,9 @@ dsl_scan_fini(dsl_pool_t *dp) /* ARGSUSED */ static int -dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_scan_setup_check(void *arg, dmu_tx_t *tx) { - dsl_scan_t *scn = arg1; + dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (scn->scn_phys.scn_state == DSS_SCANNING) return (EBUSY); @@ -164,12 +164,11 @@ dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx) return (0); } -/* ARGSUSED */ static void -dsl_scan_setup_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_scan_setup_sync(void *arg, dmu_tx_t *tx) { - dsl_scan_t *scn = arg1; - pool_scan_func_t *funcp = arg2; + dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; + pool_scan_func_t *funcp = arg; dmu_object_type_t ot = 0; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; @@ -315,9 +314,9 @@ dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) /* ARGSUSED */ static int -dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_scan_cancel_check(void *arg, dmu_tx_t *tx) { - dsl_scan_t *scn = arg1; + dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (scn->scn_phys.scn_state != DSS_SCANNING) return (ENOENT); @@ -326,9 +325,9 @@ dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx) /* ARGSUSED */ static void -dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx) { - dsl_scan_t *scn = arg1; + dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dsl_scan_done(scn, B_FALSE, tx); dsl_scan_sync_state(scn, tx); @@ -337,12 +336,8 @@ dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx) int dsl_scan_cancel(dsl_pool_t *dp) { - boolean_t complete = B_FALSE; - int err; - - err = dsl_sync_task_do(dp, dsl_scan_cancel_check, - dsl_scan_cancel_sync, dp->dp_scan, &complete, 3); - return (err); + return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check, + dsl_scan_cancel_sync, NULL, 3)); } static void dsl_scan_visitbp(blkptr_t *bp, @@ -378,7 +373,7 @@ dsl_scan_ds_maxtxg(dsl_dataset_t *ds) static void dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx) { - VERIFY(0 == zap_update(scn->scn_dp->dp_meta_objset, + VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys, tx)); @@ -950,33 +945,33 @@ struct enqueue_clones_arg { /* ARGSUSED */ static int -enqueue_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) +enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { struct enqueue_clones_arg *eca = arg; dsl_dataset_t *ds; int err; - dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; - err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); + if (hds->ds_dir->dd_phys->dd_origin_obj != eca->originobj) + return (0); + + err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); - if (ds->ds_dir->dd_phys->dd_origin_obj == eca->originobj) { - while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) { - dsl_dataset_t *prev; - err = dsl_dataset_hold_obj(dp, - ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); + while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) { + dsl_dataset_t *prev; + err = dsl_dataset_hold_obj(dp, + ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); - dsl_dataset_rele(ds, FTAG); - if (err) - return (err); - ds = prev; - } - VERIFY(zap_add_int_key(dp->dp_meta_objset, - scn->scn_phys.scn_queue_obj, ds->ds_object, - ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0); + dsl_dataset_rele(ds, FTAG); + if (err) + return (err); + ds = prev; } + VERIFY(zap_add_int_key(dp->dp_meta_objset, + scn->scn_phys.scn_queue_obj, ds->ds_object, + ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0); dsl_dataset_rele(ds, FTAG); return (0); } @@ -1065,17 +1060,17 @@ dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) } if (usenext) { - VERIFY(zap_join_key(dp->dp_meta_objset, + VERIFY0(zap_join_key(dp->dp_meta_objset, ds->ds_phys->ds_next_clones_obj, scn->scn_phys.scn_queue_obj, - ds->ds_phys->ds_creation_txg, tx) == 0); + ds->ds_phys->ds_creation_txg, tx)); } else { struct enqueue_clones_arg eca; eca.tx = tx; eca.originobj = ds->ds_object; - (void) dmu_objset_find_spa(ds->ds_dir->dd_pool->dp_spa, - NULL, enqueue_clones_cb, &eca, DS_FIND_CHILDREN); + VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, + enqueue_clones_cb, &eca, DS_FIND_CHILDREN)); } } @@ -1085,15 +1080,14 @@ dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) /* ARGSUSED */ static int -enqueue_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) +enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dmu_tx_t *tx = arg; dsl_dataset_t *ds; int err; - dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; - err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); + err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); @@ -1248,8 +1242,8 @@ dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) return; if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { - VERIFY(0 == dmu_objset_find_spa(dp->dp_spa, - NULL, enqueue_cb, tx, DS_FIND_CHILDREN)); + VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, + enqueue_cb, tx, DS_FIND_CHILDREN)); } else { dsl_scan_visitds(scn, dp->dp_origin_snap->ds_object, tx); @@ -1384,7 +1378,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) func = POOL_SCAN_RESILVER; zfs_dbgmsg("restarting scan func=%u txg=%llu", func, tx->tx_txg); - dsl_scan_setup_sync(scn, &func, tx); + dsl_scan_setup_sync(&func, tx); } if (!dsl_scan_active(scn) || @@ -1418,21 +1412,21 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) err = bptree_iterate(dp->dp_meta_objset, dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx); - VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); - if (err != 0) - return; + VERIFY0(zio_wait(scn->scn_zio_root)); - /* disable async destroy feature */ - spa_feature_decr(spa, - &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY], tx); - ASSERT(!spa_feature_is_active(spa, - &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])); - VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset, - DMU_POOL_DIRECTORY_OBJECT, - DMU_POOL_BPTREE_OBJ, tx)); - VERIFY3U(0, ==, bptree_free(dp->dp_meta_objset, - dp->dp_bptree_obj, tx)); - dp->dp_bptree_obj = 0; + if (err == 0) { + zfeature_info_t *feat = &spa_feature_table + [SPA_FEATURE_ASYNC_DESTROY]; + /* finished; deactivate async destroy feature */ + spa_feature_decr(spa, feat, tx); + ASSERT(!spa_feature_is_active(spa, feat)); + VERIFY0(zap_remove(dp->dp_meta_objset, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_BPTREE_OBJ, tx)); + VERIFY0(bptree_free(dp->dp_meta_objset, + dp->dp_bptree_obj, tx)); + dp->dp_bptree_obj = 0; + } } if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " @@ -1479,7 +1473,9 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); + dsl_pool_config_enter(dp, FTAG); dsl_scan_visit(scn, tx); + dsl_pool_config_exit(dp, FTAG); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; @@ -1714,6 +1710,6 @@ dsl_scan(dsl_pool_t *dp, pool_scan_func_t func) spa->spa_scrub_reopen = B_FALSE; (void) spa_vdev_state_exit(spa, NULL, 0); - return (dsl_sync_task_do(dp, dsl_scan_setup_check, - dsl_scan_setup_sync, dp->dp_scan, &func, 0)); + return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check, + dsl_scan_setup_sync, &func, 0)); } diff --git a/uts/common/fs/zfs/dsl_synctask.c b/uts/common/fs/zfs/dsl_synctask.c index e2481281e06a..ecb45fbc38e6 100644 --- a/uts/common/fs/zfs/dsl_synctask.c +++ b/uts/common/fs/zfs/dsl_synctask.c @@ -34,136 +34,115 @@ /* ARGSUSED */ static int -dsl_null_checkfunc(void *arg1, void *arg2, dmu_tx_t *tx) +dsl_null_checkfunc(void *arg, dmu_tx_t *tx) { return (0); } -dsl_sync_task_group_t * -dsl_sync_task_group_create(dsl_pool_t *dp) -{ - dsl_sync_task_group_t *dstg; - - dstg = kmem_zalloc(sizeof (dsl_sync_task_group_t), KM_SLEEP); - list_create(&dstg->dstg_tasks, sizeof (dsl_sync_task_t), - offsetof(dsl_sync_task_t, dst_node)); - dstg->dstg_pool = dp; - - return (dstg); -} - -void -dsl_sync_task_create(dsl_sync_task_group_t *dstg, - dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc, - void *arg1, void *arg2, int blocks_modified) -{ - dsl_sync_task_t *dst; - - if (checkfunc == NULL) - checkfunc = dsl_null_checkfunc; - dst = kmem_zalloc(sizeof (dsl_sync_task_t), KM_SLEEP); - dst->dst_checkfunc = checkfunc; - dst->dst_syncfunc = syncfunc; - dst->dst_arg1 = arg1; - dst->dst_arg2 = arg2; - list_insert_tail(&dstg->dstg_tasks, dst); - - dstg->dstg_space += blocks_modified << DST_AVG_BLKSHIFT; -} - +/* + * Called from open context to perform a callback in syncing context. Waits + * for the operation to complete. + * + * The checkfunc will be called from open context as a preliminary check + * which can quickly fail. If it succeeds, it will be called again from + * syncing context. The checkfunc should generally be designed to work + * properly in either context, but if necessary it can check + * dmu_tx_is_syncing(tx). + * + * The synctask infrastructure enforces proper locking strategy with respect + * to the dp_config_rwlock -- the lock will always be held when the callbacks + * are called. It will be held for read during the open-context (preliminary) + * call to the checkfunc, and then held for write from syncing context during + * the calls to the check and sync funcs. + * + * A dataset or pool name can be passed as the first argument. Typically, + * the check func will hold, check the return value of the hold, and then + * release the dataset. The sync func will VERIFYO(hold()) the dataset. + * This is safe because no changes can be made between the check and sync funcs, + * and the sync func will only be called if the check func successfully opened + * the dataset. + */ int -dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg) +dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc, + dsl_syncfunc_t *syncfunc, void *arg, int blocks_modified) { + spa_t *spa; dmu_tx_t *tx; - uint64_t txg; - dsl_sync_task_t *dst; + int err; + dsl_sync_task_t dst = { 0 }; + dsl_pool_t *dp; + + err = spa_open(pool, &spa, FTAG); + if (err != 0) + return (err); + dp = spa_get_dsl(spa); top: - tx = dmu_tx_create_dd(dstg->dstg_pool->dp_mos_dir); - VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); + tx = dmu_tx_create_dd(dp->dp_mos_dir); + VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); - txg = dmu_tx_get_txg(tx); + dst.dst_pool = dp; + dst.dst_txg = dmu_tx_get_txg(tx); + dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT; + dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc; + dst.dst_syncfunc = syncfunc; + dst.dst_arg = arg; + dst.dst_error = 0; + dst.dst_nowaiter = B_FALSE; - /* Do a preliminary error check. */ - dstg->dstg_err = 0; -#ifdef ZFS_DEBUG - /* - * Only check half the time, otherwise, the sync-context - * check will almost never fail. - */ - if (spa_get_random(2) == 0) - goto skip; -#endif - rw_enter(&dstg->dstg_pool->dp_config_rwlock, RW_READER); - for (dst = list_head(&dstg->dstg_tasks); dst; - dst = list_next(&dstg->dstg_tasks, dst)) { - dst->dst_err = - dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx); - if (dst->dst_err) - dstg->dstg_err = dst->dst_err; - } - rw_exit(&dstg->dstg_pool->dp_config_rwlock); + dsl_pool_config_enter(dp, FTAG); + err = dst.dst_checkfunc(arg, tx); + dsl_pool_config_exit(dp, FTAG); - if (dstg->dstg_err) { + if (err != 0) { dmu_tx_commit(tx); - return (dstg->dstg_err); + spa_close(spa, FTAG); + return (err); } -skip: - /* - * We don't generally have many sync tasks, so pay the price of - * add_tail to get the tasks executed in the right order. - */ - VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks, - dstg, txg)); + VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg)); dmu_tx_commit(tx); - txg_wait_synced(dstg->dstg_pool, txg); + txg_wait_synced(dp, dst.dst_txg); - if (dstg->dstg_err == EAGAIN) { - txg_wait_synced(dstg->dstg_pool, txg + TXG_DEFER_SIZE); + if (dst.dst_error == EAGAIN) { + txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE); goto top; } - return (dstg->dstg_err); + spa_close(spa, FTAG); + return (dst.dst_error); } void -dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx) +dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg, + int blocks_modified, dmu_tx_t *tx) { - uint64_t txg; + dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP); - dstg->dstg_nowaiter = B_TRUE; - txg = dmu_tx_get_txg(tx); - /* - * We don't generally have many sync tasks, so pay the price of - * add_tail to get the tasks executed in the right order. - */ - VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks, - dstg, txg)); + dst->dst_pool = dp; + dst->dst_txg = dmu_tx_get_txg(tx); + dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT; + dst->dst_checkfunc = dsl_null_checkfunc; + dst->dst_syncfunc = syncfunc; + dst->dst_arg = arg; + dst->dst_error = 0; + dst->dst_nowaiter = B_TRUE; + + VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, dst, dst->dst_txg)); } +/* + * Called in syncing context to execute the synctask. + */ void -dsl_sync_task_group_destroy(dsl_sync_task_group_t *dstg) +dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx) { - dsl_sync_task_t *dst; - - while (dst = list_head(&dstg->dstg_tasks)) { - list_remove(&dstg->dstg_tasks, dst); - kmem_free(dst, sizeof (dsl_sync_task_t)); - } - kmem_free(dstg, sizeof (dsl_sync_task_group_t)); -} - -void -dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx) -{ - dsl_sync_task_t *dst; - dsl_pool_t *dp = dstg->dstg_pool; + dsl_pool_t *dp = dst->dst_pool; uint64_t quota, used; - ASSERT0(dstg->dstg_err); + ASSERT0(dst->dst_error); /* * Check for sufficient space. We just check against what's @@ -175,63 +154,21 @@ dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx) metaslab_class_get_deferred(spa_normal_class(dp->dp_spa)); used = dp->dp_root_dir->dd_phys->dd_used_bytes; /* MOS space is triple-dittoed, so we multiply by 3. */ - if (dstg->dstg_space > 0 && used + dstg->dstg_space * 3 > quota) { - dstg->dstg_err = ENOSPC; + if (dst->dst_space > 0 && used + dst->dst_space * 3 > quota) { + dst->dst_error = ENOSPC; + if (dst->dst_nowaiter) + kmem_free(dst, sizeof (*dst)); return; } /* - * Check for errors by calling checkfuncs. + * Check for errors by calling checkfunc. */ - rw_enter(&dp->dp_config_rwlock, RW_WRITER); - for (dst = list_head(&dstg->dstg_tasks); dst; - dst = list_next(&dstg->dstg_tasks, dst)) { - dst->dst_err = - dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx); - if (dst->dst_err) - dstg->dstg_err = dst->dst_err; - } - - if (dstg->dstg_err == 0) { - /* - * Execute sync tasks. - */ - for (dst = list_head(&dstg->dstg_tasks); dst; - dst = list_next(&dstg->dstg_tasks, dst)) { - dst->dst_syncfunc(dst->dst_arg1, dst->dst_arg2, tx); - } - } - rw_exit(&dp->dp_config_rwlock); - - if (dstg->dstg_nowaiter) - dsl_sync_task_group_destroy(dstg); -} - -int -dsl_sync_task_do(dsl_pool_t *dp, - dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc, - void *arg1, void *arg2, int blocks_modified) -{ - dsl_sync_task_group_t *dstg; - int err; - - ASSERT(spa_writeable(dp->dp_spa)); - - dstg = dsl_sync_task_group_create(dp); - dsl_sync_task_create(dstg, checkfunc, syncfunc, - arg1, arg2, blocks_modified); - err = dsl_sync_task_group_wait(dstg); - dsl_sync_task_group_destroy(dstg); - return (err); -} - -void -dsl_sync_task_do_nowait(dsl_pool_t *dp, - dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc, - void *arg1, void *arg2, int blocks_modified, dmu_tx_t *tx) -{ - dsl_sync_task_group_t *dstg = dsl_sync_task_group_create(dp); - dsl_sync_task_create(dstg, checkfunc, syncfunc, - arg1, arg2, blocks_modified); - dsl_sync_task_group_nowait(dstg, tx); + rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); + dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx); + if (dst->dst_error == 0) + dst->dst_syncfunc(dst->dst_arg, tx); + rrw_exit(&dp->dp_config_rwlock, FTAG); + if (dst->dst_nowaiter) + kmem_free(dst, sizeof (*dst)); } diff --git a/uts/common/fs/zfs/dsl_userhold.c b/uts/common/fs/zfs/dsl_userhold.c new file mode 100644 index 000000000000..e30169eade4d --- /dev/null +++ b/uts/common/fs/zfs/dsl_userhold.c @@ -0,0 +1,536 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +typedef struct dsl_dataset_user_hold_arg { + nvlist_t *dduha_holds; + nvlist_t *dduha_errlist; + minor_t dduha_minor; +} dsl_dataset_user_hold_arg_t; + +/* + * If you add new checks here, you may need to add additional checks to the + * "temporary" case in snapshot_check() in dmu_objset.c. + */ +int +dsl_dataset_user_hold_check_one(dsl_dataset_t *ds, const char *htag, + boolean_t temphold, dmu_tx_t *tx) +{ + dsl_pool_t *dp = dmu_tx_pool(tx); + objset_t *mos = dp->dp_meta_objset; + int error = 0; + + if (strlen(htag) > MAXNAMELEN) + return (E2BIG); + /* Tempholds have a more restricted length */ + if (temphold && strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) + return (E2BIG); + + /* tags must be unique (if ds already exists) */ + if (ds != NULL) { + mutex_enter(&ds->ds_lock); + if (ds->ds_phys->ds_userrefs_obj != 0) { + uint64_t value; + error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, + htag, 8, 1, &value); + if (error == 0) + error = EEXIST; + else if (error == ENOENT) + error = 0; + } + mutex_exit(&ds->ds_lock); + } + + return (error); +} + +static int +dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_hold_arg_t *dduha = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + int rv = 0; + + if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) + return (ENOTSUP); + + for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) { + int error = 0; + dsl_dataset_t *ds; + char *htag; + + /* must be a snapshot */ + if (strchr(nvpair_name(pair), '@') == NULL) + error = EINVAL; + + if (error == 0) + error = nvpair_value_string(pair, &htag); + if (error == 0) { + error = dsl_dataset_hold(dp, + nvpair_name(pair), FTAG, &ds); + } + if (error == 0) { + error = dsl_dataset_user_hold_check_one(ds, htag, + dduha->dduha_minor != 0, tx); + dsl_dataset_rele(ds, FTAG); + } + + if (error != 0) { + rv = error; + fnvlist_add_int32(dduha->dduha_errlist, + nvpair_name(pair), error); + } + } + return (rv); +} + +void +dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag, + minor_t minor, uint64_t now, dmu_tx_t *tx) +{ + dsl_pool_t *dp = ds->ds_dir->dd_pool; + objset_t *mos = dp->dp_meta_objset; + uint64_t zapobj; + + mutex_enter(&ds->ds_lock); + if (ds->ds_phys->ds_userrefs_obj == 0) { + /* + * This is the first user hold for this dataset. Create + * the userrefs zap object. + */ + dmu_buf_will_dirty(ds->ds_dbuf, tx); + zapobj = ds->ds_phys->ds_userrefs_obj = + zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); + } else { + zapobj = ds->ds_phys->ds_userrefs_obj; + } + ds->ds_userrefs++; + mutex_exit(&ds->ds_lock); + + VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx)); + + if (minor != 0) { + VERIFY0(dsl_pool_user_hold(dp, ds->ds_object, + htag, now, tx)); + dsl_register_onexit_hold_cleanup(ds, htag, minor); + } + + spa_history_log_internal_ds(ds, "hold", tx, + "tag=%s temp=%d refs=%llu", + htag, minor != 0, ds->ds_userrefs); +} + +static void +dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_hold_arg_t *dduha = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + uint64_t now = gethrestime_sec(); + + for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) { + dsl_dataset_t *ds; + VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds)); + dsl_dataset_user_hold_sync_one(ds, fnvpair_value_string(pair), + dduha->dduha_minor, now, tx); + dsl_dataset_rele(ds, FTAG); + } +} + +/* + * holds is nvl of snapname -> holdname + * errlist will be filled in with snapname -> error + * if cleanup_minor is not 0, the holds will be temporary, cleaned up + * when the process exits. + * + * if any fails, all will fail. + */ +int +dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist) +{ + dsl_dataset_user_hold_arg_t dduha; + nvpair_t *pair; + + pair = nvlist_next_nvpair(holds, NULL); + if (pair == NULL) + return (0); + + dduha.dduha_holds = holds; + dduha.dduha_errlist = errlist; + dduha.dduha_minor = cleanup_minor; + + return (dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check, + dsl_dataset_user_hold_sync, &dduha, fnvlist_num_pairs(holds))); +} + +typedef struct dsl_dataset_user_release_arg { + nvlist_t *ddura_holds; + nvlist_t *ddura_todelete; + nvlist_t *ddura_errlist; +} dsl_dataset_user_release_arg_t; + +static int +dsl_dataset_user_release_check_one(dsl_dataset_t *ds, + nvlist_t *holds, boolean_t *todelete) +{ + uint64_t zapobj; + nvpair_t *pair; + objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; + int error; + int numholds = 0; + + *todelete = B_FALSE; + + if (!dsl_dataset_is_snapshot(ds)) + return (EINVAL); + + zapobj = ds->ds_phys->ds_userrefs_obj; + if (zapobj == 0) + return (ESRCH); + + for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(holds, pair)) { + /* Make sure the hold exists */ + uint64_t tmp; + error = zap_lookup(mos, zapobj, nvpair_name(pair), 8, 1, &tmp); + if (error == ENOENT) + error = ESRCH; + if (error != 0) + return (error); + numholds++; + } + + if (DS_IS_DEFER_DESTROY(ds) && ds->ds_phys->ds_num_children == 1 && + ds->ds_userrefs == numholds) { + /* we need to destroy the snapshot as well */ + + if (dsl_dataset_long_held(ds)) + return (EBUSY); + *todelete = B_TRUE; + } + return (0); +} + +static int +dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_release_arg_t *ddura = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + int rv = 0; + + if (!dmu_tx_is_syncing(tx)) + return (0); + + for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) { + const char *name = nvpair_name(pair); + int error; + dsl_dataset_t *ds; + nvlist_t *holds; + + error = nvpair_value_nvlist(pair, &holds); + if (error != 0) + return (EINVAL); + + error = dsl_dataset_hold(dp, name, FTAG, &ds); + if (error == 0) { + boolean_t deleteme; + error = dsl_dataset_user_release_check_one(ds, + holds, &deleteme); + if (error == 0 && deleteme) { + fnvlist_add_boolean(ddura->ddura_todelete, + name); + } + dsl_dataset_rele(ds, FTAG); + } + if (error != 0) { + if (ddura->ddura_errlist != NULL) { + fnvlist_add_int32(ddura->ddura_errlist, + name, error); + } + rv = error; + } + } + return (rv); +} + +static void +dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds, + dmu_tx_t *tx) +{ + dsl_pool_t *dp = ds->ds_dir->dd_pool; + objset_t *mos = dp->dp_meta_objset; + uint64_t zapobj; + int error; + nvpair_t *pair; + + for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(holds, pair)) { + ds->ds_userrefs--; + error = dsl_pool_user_release(dp, ds->ds_object, + nvpair_name(pair), tx); + VERIFY(error == 0 || error == ENOENT); + zapobj = ds->ds_phys->ds_userrefs_obj; + VERIFY0(zap_remove(mos, zapobj, nvpair_name(pair), tx)); + + spa_history_log_internal_ds(ds, "release", tx, + "tag=%s refs=%lld", nvpair_name(pair), + (longlong_t)ds->ds_userrefs); + } +} + +static void +dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_release_arg_t *ddura = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + nvpair_t *pair; + + for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) { + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds)); + dsl_dataset_user_release_sync_one(ds, + fnvpair_value_nvlist(pair), tx); + if (nvlist_exists(ddura->ddura_todelete, + nvpair_name(pair))) { + ASSERT(ds->ds_userrefs == 0 && + ds->ds_phys->ds_num_children == 1 && + DS_IS_DEFER_DESTROY(ds)); + dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx); + } + dsl_dataset_rele(ds, FTAG); + } +} + +/* + * holds is nvl of snapname -> { holdname, ... } + * errlist will be filled in with snapname -> error + * + * if any fails, all will fail. + */ +int +dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist) +{ + dsl_dataset_user_release_arg_t ddura; + nvpair_t *pair; + int error; + + pair = nvlist_next_nvpair(holds, NULL); + if (pair == NULL) + return (0); + + ddura.ddura_holds = holds; + ddura.ddura_errlist = errlist; + ddura.ddura_todelete = fnvlist_alloc(); + + error = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_release_check, + dsl_dataset_user_release_sync, &ddura, fnvlist_num_pairs(holds)); + fnvlist_free(ddura.ddura_todelete); + return (error); +} + +typedef struct dsl_dataset_user_release_tmp_arg { + uint64_t ddurta_dsobj; + nvlist_t *ddurta_holds; + boolean_t ddurta_deleteme; +} dsl_dataset_user_release_tmp_arg_t; + +static int +dsl_dataset_user_release_tmp_check(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_release_tmp_arg_t *ddurta = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + int error; + + if (!dmu_tx_is_syncing(tx)) + return (0); + + error = dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds); + if (error) + return (error); + + error = dsl_dataset_user_release_check_one(ds, + ddurta->ddurta_holds, &ddurta->ddurta_deleteme); + dsl_dataset_rele(ds, FTAG); + return (error); +} + +static void +dsl_dataset_user_release_tmp_sync(void *arg, dmu_tx_t *tx) +{ + dsl_dataset_user_release_tmp_arg_t *ddurta = arg; + dsl_pool_t *dp = dmu_tx_pool(tx); + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds)); + dsl_dataset_user_release_sync_one(ds, ddurta->ddurta_holds, tx); + if (ddurta->ddurta_deleteme) { + ASSERT(ds->ds_userrefs == 0 && + ds->ds_phys->ds_num_children == 1 && + DS_IS_DEFER_DESTROY(ds)); + dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx); + } + dsl_dataset_rele(ds, FTAG); +} + +/* + * Called at spa_load time to release a stale temporary user hold. + * Also called by the onexit code. + */ +void +dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, const char *htag) +{ + dsl_dataset_user_release_tmp_arg_t ddurta; + dsl_dataset_t *ds; + int error; + +#ifdef _KERNEL + /* Make sure it is not mounted. */ + dsl_pool_config_enter(dp, FTAG); + error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); + if (error == 0) { + char name[MAXNAMELEN]; + dsl_dataset_name(ds, name); + dsl_dataset_rele(ds, FTAG); + dsl_pool_config_exit(dp, FTAG); + zfs_unmount_snap(name); + } else { + dsl_pool_config_exit(dp, FTAG); + } +#endif + + ddurta.ddurta_dsobj = dsobj; + ddurta.ddurta_holds = fnvlist_alloc(); + fnvlist_add_boolean(ddurta.ddurta_holds, htag); + + (void) dsl_sync_task(spa_name(dp->dp_spa), + dsl_dataset_user_release_tmp_check, + dsl_dataset_user_release_tmp_sync, &ddurta, 1); + fnvlist_free(ddurta.ddurta_holds); +} + +typedef struct zfs_hold_cleanup_arg { + char zhca_spaname[MAXNAMELEN]; + uint64_t zhca_spa_load_guid; + uint64_t zhca_dsobj; + char zhca_htag[MAXNAMELEN]; +} zfs_hold_cleanup_arg_t; + +static void +dsl_dataset_user_release_onexit(void *arg) +{ + zfs_hold_cleanup_arg_t *ca = arg; + spa_t *spa; + int error; + + error = spa_open(ca->zhca_spaname, &spa, FTAG); + if (error != 0) { + zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s " + "because pool is no longer loaded", + ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag); + return; + } + if (spa_load_guid(spa) != ca->zhca_spa_load_guid) { + zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s " + "because pool is no longer loaded (guid doesn't match)", + ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag); + spa_close(spa, FTAG); + return; + } + + dsl_dataset_user_release_tmp(spa_get_dsl(spa), + ca->zhca_dsobj, ca->zhca_htag); + kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t)); + spa_close(spa, FTAG); +} + +void +dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, + minor_t minor) +{ + zfs_hold_cleanup_arg_t *ca = kmem_alloc(sizeof (*ca), KM_SLEEP); + spa_t *spa = dsl_dataset_get_spa(ds); + (void) strlcpy(ca->zhca_spaname, spa_name(spa), + sizeof (ca->zhca_spaname)); + ca->zhca_spa_load_guid = spa_load_guid(spa); + ca->zhca_dsobj = ds->ds_object; + (void) strlcpy(ca->zhca_htag, htag, sizeof (ca->zhca_htag)); + VERIFY0(zfs_onexit_add_cb(minor, + dsl_dataset_user_release_onexit, ca, NULL)); +} + +int +dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl) +{ + dsl_pool_t *dp; + dsl_dataset_t *ds; + int err; + + err = dsl_pool_hold(dsname, FTAG, &dp); + if (err != 0) + return (err); + err = dsl_dataset_hold(dp, dsname, FTAG, &ds); + if (err != 0) { + dsl_pool_rele(dp, FTAG); + return (err); + } + + if (ds->ds_phys->ds_userrefs_obj != 0) { + zap_attribute_t *za; + zap_cursor_t zc; + + za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); + for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, + ds->ds_phys->ds_userrefs_obj); + zap_cursor_retrieve(&zc, za) == 0; + zap_cursor_advance(&zc)) { + fnvlist_add_uint64(nvl, za->za_name, + za->za_first_integer); + } + zap_cursor_fini(&zc); + kmem_free(za, sizeof (zap_attribute_t)); + } + dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); + return (0); +} diff --git a/uts/common/fs/zfs/metaslab.c b/uts/common/fs/zfs/metaslab.c index bf9889e183d5..aae2ccd5d9e9 100644 --- a/uts/common/fs/zfs/metaslab.c +++ b/uts/common/fs/zfs/metaslab.c @@ -1866,3 +1866,41 @@ metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg) return (error); } + +static void +checkmap(space_map_t *sm, uint64_t off, uint64_t size) +{ + space_seg_t *ss; + avl_index_t where; + + mutex_enter(sm->sm_lock); + ss = space_map_find(sm, off, size, &where); + if (ss != NULL) + panic("freeing free block; ss=%p", (void *)ss); + mutex_exit(sm->sm_lock); +} + +void +metaslab_check_free(spa_t *spa, const blkptr_t *bp) +{ + if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) + return; + + spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); + for (int i = 0; i < BP_GET_NDVAS(bp); i++) { + uint64_t vdid = DVA_GET_VDEV(&bp->blk_dva[i]); + vdev_t *vd = vdev_lookup_top(spa, vdid); + uint64_t off = DVA_GET_OFFSET(&bp->blk_dva[i]); + uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]); + metaslab_t *ms = vd->vdev_ms[off >> vd->vdev_ms_shift]; + + if (ms->ms_map->sm_loaded) + checkmap(ms->ms_map, off, size); + + for (int j = 0; j < TXG_SIZE; j++) + checkmap(ms->ms_freemap[j], off, size); + for (int j = 0; j < TXG_DEFER_SIZE; j++) + checkmap(ms->ms_defermap[j], off, size); + } + spa_config_exit(spa, SCL_VDEV, FTAG); +} diff --git a/uts/common/fs/zfs/refcount.c b/uts/common/fs/zfs/refcount.c index 3a8e144e9092..df0f25684958 100644 --- a/uts/common/fs/zfs/refcount.c +++ b/uts/common/fs/zfs/refcount.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include @@ -32,7 +33,7 @@ int reference_tracking_enable = FALSE; /* runs out of memory too easily */ #else int reference_tracking_enable = TRUE; #endif -int reference_history = 4; /* tunable */ +int reference_history = 3; /* tunable */ static kmem_cache_t *reference_cache; static kmem_cache_t *reference_history_cache; @@ -64,6 +65,14 @@ refcount_create(refcount_t *rc) offsetof(reference_t, ref_link)); rc->rc_count = 0; rc->rc_removed_count = 0; + rc->rc_tracked = reference_tracking_enable; +} + +void +refcount_create_untracked(refcount_t *rc) +{ + refcount_create(rc); + rc->rc_tracked = B_FALSE; } void @@ -96,14 +105,12 @@ refcount_destroy(refcount_t *rc) int refcount_is_zero(refcount_t *rc) { - ASSERT(rc->rc_count >= 0); return (rc->rc_count == 0); } int64_t refcount_count(refcount_t *rc) { - ASSERT(rc->rc_count >= 0); return (rc->rc_count); } @@ -113,14 +120,14 @@ refcount_add_many(refcount_t *rc, uint64_t number, void *holder) reference_t *ref = NULL; int64_t count; - if (reference_tracking_enable) { + if (rc->rc_tracked) { ref = kmem_cache_alloc(reference_cache, KM_SLEEP); ref->ref_holder = holder; ref->ref_number = number; } mutex_enter(&rc->rc_mtx); ASSERT(rc->rc_count >= 0); - if (reference_tracking_enable) + if (rc->rc_tracked) list_insert_head(&rc->rc_list, ref); rc->rc_count += number; count = rc->rc_count; @@ -144,7 +151,7 @@ refcount_remove_many(refcount_t *rc, uint64_t number, void *holder) mutex_enter(&rc->rc_mtx); ASSERT(rc->rc_count >= number); - if (!reference_tracking_enable) { + if (!rc->rc_tracked) { rc->rc_count -= number; count = rc->rc_count; mutex_exit(&rc->rc_mtx); @@ -161,7 +168,7 @@ refcount_remove_many(refcount_t *rc, uint64_t number, void *holder) KM_SLEEP); list_insert_head(&rc->rc_removed, ref); rc->rc_removed_count++; - if (rc->rc_removed_count >= reference_history) { + if (rc->rc_removed_count > reference_history) { ref = list_tail(&rc->rc_removed); list_remove(&rc->rc_removed, ref); kmem_cache_free(reference_history_cache, diff --git a/uts/common/fs/zfs/rrwlock.c b/uts/common/fs/zfs/rrwlock.c index 7f9290bd44c1..8e80166c7d14 100644 --- a/uts/common/fs/zfs/rrwlock.c +++ b/uts/common/fs/zfs/rrwlock.c @@ -75,8 +75,9 @@ uint_t rrw_tsd_key; typedef struct rrw_node { - struct rrw_node *rn_next; - rrwlock_t *rn_rrl; + struct rrw_node *rn_next; + rrwlock_t *rn_rrl; + void *rn_tag; } rrw_node_t; static rrw_node_t * @@ -98,13 +99,14 @@ rrn_find(rrwlock_t *rrl) * Add a node to the head of the singly linked list. */ static void -rrn_add(rrwlock_t *rrl) +rrn_add(rrwlock_t *rrl, void *tag) { rrw_node_t *rn; rn = kmem_alloc(sizeof (*rn), KM_SLEEP); rn->rn_rrl = rrl; rn->rn_next = tsd_get(rrw_tsd_key); + rn->rn_tag = tag; VERIFY(tsd_set(rrw_tsd_key, rn) == 0); } @@ -113,7 +115,7 @@ rrn_add(rrwlock_t *rrl) * thread's list and return TRUE; otherwise return FALSE. */ static boolean_t -rrn_find_and_remove(rrwlock_t *rrl) +rrn_find_and_remove(rrwlock_t *rrl, void *tag) { rrw_node_t *rn; rrw_node_t *prev = NULL; @@ -122,7 +124,7 @@ rrn_find_and_remove(rrwlock_t *rrl) return (B_FALSE); for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) { - if (rn->rn_rrl == rrl) { + if (rn->rn_rrl == rrl && rn->rn_tag == tag) { if (prev) prev->rn_next = rn->rn_next; else @@ -136,7 +138,7 @@ rrn_find_and_remove(rrwlock_t *rrl) } void -rrw_init(rrwlock_t *rrl) +rrw_init(rrwlock_t *rrl, boolean_t track_all) { mutex_init(&rrl->rr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&rrl->rr_cv, NULL, CV_DEFAULT, NULL); @@ -144,6 +146,7 @@ rrw_init(rrwlock_t *rrl) refcount_create(&rrl->rr_anon_rcount); refcount_create(&rrl->rr_linked_rcount); rrl->rr_writer_wanted = B_FALSE; + rrl->rr_track_all = track_all; } void @@ -156,12 +159,13 @@ rrw_destroy(rrwlock_t *rrl) refcount_destroy(&rrl->rr_linked_rcount); } -static void +void rrw_enter_read(rrwlock_t *rrl, void *tag) { mutex_enter(&rrl->rr_lock); #if !defined(DEBUG) && defined(_KERNEL) - if (!rrl->rr_writer && !rrl->rr_writer_wanted) { + if (rrl->rr_writer == NULL && !rrl->rr_writer_wanted && + !rrl->rr_track_all) { rrl->rr_anon_rcount.rc_count++; mutex_exit(&rrl->rr_lock); return; @@ -171,14 +175,14 @@ rrw_enter_read(rrwlock_t *rrl, void *tag) ASSERT(rrl->rr_writer != curthread); ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0); - while (rrl->rr_writer || (rrl->rr_writer_wanted && + while (rrl->rr_writer != NULL || (rrl->rr_writer_wanted && refcount_is_zero(&rrl->rr_anon_rcount) && rrn_find(rrl) == NULL)) cv_wait(&rrl->rr_cv, &rrl->rr_lock); - if (rrl->rr_writer_wanted) { + if (rrl->rr_writer_wanted || rrl->rr_track_all) { /* may or may not be a re-entrant enter */ - rrn_add(rrl); + rrn_add(rrl, tag); (void) refcount_add(&rrl->rr_linked_rcount, tag); } else { (void) refcount_add(&rrl->rr_anon_rcount, tag); @@ -187,7 +191,7 @@ rrw_enter_read(rrwlock_t *rrl, void *tag) mutex_exit(&rrl->rr_lock); } -static void +void rrw_enter_write(rrwlock_t *rrl) { mutex_enter(&rrl->rr_lock); @@ -233,10 +237,12 @@ rrw_exit(rrwlock_t *rrl, void *tag) if (rrl->rr_writer == NULL) { int64_t count; - if (rrn_find_and_remove(rrl)) + if (rrn_find_and_remove(rrl, tag)) { count = refcount_remove(&rrl->rr_linked_rcount, tag); - else + } else { + ASSERT(!rrl->rr_track_all); count = refcount_remove(&rrl->rr_anon_rcount, tag); + } if (count == 0) cv_broadcast(&rrl->rr_cv); } else { @@ -249,6 +255,11 @@ rrw_exit(rrwlock_t *rrl, void *tag) mutex_exit(&rrl->rr_lock); } +/* + * If the lock was created with track_all, rrw_held(RW_READER) will return + * B_TRUE iff the current thread has the lock for reader. Otherwise it may + * return B_TRUE if any thread has the lock for reader. + */ boolean_t rrw_held(rrwlock_t *rrl, krw_t rw) { @@ -259,7 +270,7 @@ rrw_held(rrwlock_t *rrl, krw_t rw) held = (rrl->rr_writer == curthread); } else { held = (!refcount_is_zero(&rrl->rr_anon_rcount) || - !refcount_is_zero(&rrl->rr_linked_rcount)); + rrn_find(rrl) != NULL); } mutex_exit(&rrl->rr_lock); diff --git a/uts/common/fs/zfs/sa.c b/uts/common/fs/zfs/sa.c index cd3a58b5fa93..05f329c64747 100644 --- a/uts/common/fs/zfs/sa.c +++ b/uts/common/fs/zfs/sa.c @@ -1001,10 +1001,10 @@ sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count, sa_attr_type_t *tb; int error; - mutex_enter(&os->os_lock); + mutex_enter(&os->os_user_ptr_lock); if (os->os_sa) { mutex_enter(&os->os_sa->sa_lock); - mutex_exit(&os->os_lock); + mutex_exit(&os->os_user_ptr_lock); tb = os->os_sa->sa_user_table; mutex_exit(&os->os_sa->sa_lock); *user_table = tb; @@ -1017,7 +1017,7 @@ sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count, os->os_sa = sa; mutex_enter(&sa->sa_lock); - mutex_exit(&os->os_lock); + mutex_exit(&os->os_user_ptr_lock); avl_create(&sa->sa_layout_num_tree, layout_num_compare, sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node)); avl_create(&sa->sa_layout_hash_tree, layout_hash_compare, diff --git a/uts/common/fs/zfs/spa.c b/uts/common/fs/zfs/spa.c index fdc28700c820..544a0407c1bc 100644 --- a/uts/common/fs/zfs/spa.c +++ b/uts/common/fs/zfs/spa.c @@ -63,6 +63,7 @@ #include #include #include +#include #ifdef _KERNEL #include @@ -129,10 +130,8 @@ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ }; -static dsl_syncfunc_t spa_sync_version; -static dsl_syncfunc_t spa_sync_props; -static dsl_checkfunc_t spa_change_guid_check; -static dsl_syncfunc_t spa_change_guid_sync; +static void spa_sync_version(void *arg, dmu_tx_t *tx); +static void spa_sync_props(void *arg, dmu_tx_t *tx); static boolean_t spa_has_active_shared_spare(spa_t *spa); static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, @@ -325,10 +324,10 @@ spa_prop_get(spa_t *spa, nvlist_t **nvp) dsl_dataset_t *ds = NULL; dp = spa_get_dsl(spa); - rw_enter(&dp->dp_config_rwlock, RW_READER); + dsl_pool_config_enter(dp, FTAG); if (err = dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &ds)) { - rw_exit(&dp->dp_config_rwlock); + dsl_pool_config_exit(dp, FTAG); break; } @@ -337,7 +336,7 @@ spa_prop_get(spa_t *spa, nvlist_t **nvp) KM_SLEEP); dsl_dataset_name(ds, strval); dsl_dataset_rele(ds, FTAG); - rw_exit(&dp->dp_config_rwlock); + dsl_pool_config_exit(dp, FTAG); } else { strval = NULL; intval = za.za_first_integer; @@ -491,9 +490,10 @@ spa_prop_validate(spa_t *spa, nvlist_t *props) if (dmu_objset_type(os) != DMU_OST_ZFS) { error = ENOTSUP; - } else if ((error = dsl_prop_get_integer(strval, + } else if ((error = + dsl_prop_get_int_ds(dmu_objset_ds(os), zfs_prop_to_name(ZFS_PROP_COMPRESSION), - &compress, NULL)) == 0 && + &compress)) == 0 && !BOOTFS_COMPRESS_VALID(compress)) { error = ENOTSUP; } else { @@ -660,8 +660,8 @@ spa_prop_set(spa_t *spa, nvlist_t *nvp) * read object, the features for write object, or the * feature descriptions object. */ - error = dsl_sync_task_do(spa_get_dsl(spa), NULL, - spa_sync_version, spa, &ver, 6); + error = dsl_sync_task(spa->spa_name, NULL, + spa_sync_version, &ver, 6); if (error) return (error); continue; @@ -672,8 +672,8 @@ spa_prop_set(spa_t *spa, nvlist_t *nvp) } if (need_sync) { - return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, - spa, nvp, 6)); + return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, + nvp, 6)); } return (0); @@ -695,10 +695,10 @@ spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) /*ARGSUSED*/ static int -spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx) +spa_change_guid_check(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - uint64_t *newguid = arg2; + uint64_t *newguid = arg; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; vdev_t *rvd = spa->spa_root_vdev; uint64_t vdev_state; @@ -715,10 +715,10 @@ spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx) } static void -spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx) +spa_change_guid_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - uint64_t *newguid = arg2; + uint64_t *newguid = arg; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; uint64_t oldguid; vdev_t *rvd = spa->spa_root_vdev; @@ -752,8 +752,8 @@ spa_change_guid(spa_t *spa) mutex_enter(&spa_namespace_lock); guid = spa_generate_guid(NULL); - error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check, - spa_change_guid_sync, spa, &guid, 5); + error = dsl_sync_task(spa->spa_name, spa_change_guid_check, + spa_change_guid_sync, &guid, 5); if (error == 0) { spa_config_sync(spa, B_FALSE, B_TRUE); @@ -1687,21 +1687,22 @@ spa_config_valid(spa_t *spa, nvlist_t *config) /* * Check for missing log devices */ -static int +static boolean_t spa_check_logs(spa_t *spa) { + boolean_t rv = B_FALSE; + switch (spa->spa_log_state) { case SPA_LOG_MISSING: /* need to recheck in case slog has been restored */ case SPA_LOG_UNKNOWN: - if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, - DS_FIND_CHILDREN)) { + rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, + NULL, DS_FIND_CHILDREN) != 0); + if (rv) spa_set_log_state(spa, SPA_LOG_MISSING); - return (1); - } break; } - return (0); + return (rv); } static boolean_t @@ -1747,11 +1748,11 @@ spa_activate_log(spa_t *spa) int spa_offline_log(spa_t *spa) { - int error = 0; - - if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline, - NULL, DS_FIND_CHILDREN)) == 0) { + int error; + error = dmu_objset_find(spa_name(spa), zil_vdev_offline, + NULL, DS_FIND_CHILDREN); + if (error == 0) { /* * We successfully offlined the log device, sync out the * current txg so that the "stubby" block can be removed @@ -3549,7 +3550,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, if (props != NULL) { spa_configfile_set(spa, props, B_FALSE); - spa_sync_props(spa, props, tx); + spa_sync_props(props, tx); } dmu_tx_commit(tx); @@ -5813,10 +5814,11 @@ spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) } static void -spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx) +spa_sync_version(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - uint64_t version = *(uint64_t *)arg2; + uint64_t *versionp = arg; + uint64_t version = *versionp; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; /* * Setting the version is special cased when first creating the pool. @@ -5835,11 +5837,11 @@ spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx) * Set zpool properties. */ static void -spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) +spa_sync_props(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; + nvlist_t *nvp = arg; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = spa->spa_meta_objset; - nvlist_t *nvp = arg2; nvpair_t *elem = NULL; mutex_enter(&spa->spa_props_lock); @@ -5990,6 +5992,8 @@ spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) ASSERT(spa->spa_sync_pass == 1); + rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); + if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { dsl_pool_create_origin(dp, tx); @@ -6015,6 +6019,7 @@ spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { spa_feature_create_zap_objects(spa, tx); } + rrw_exit(&dp->dp_config_rwlock, FTAG); } /* diff --git a/uts/common/fs/zfs/spa_history.c b/uts/common/fs/zfs/spa_history.c index 9ae28739f377..eef642aa1323 100644 --- a/uts/common/fs/zfs/spa_history.c +++ b/uts/common/fs/zfs/spa_history.c @@ -195,10 +195,10 @@ spa_history_zone(void) */ /*ARGSUSED*/ static void -spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx) +spa_history_log_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - nvlist_t *nvl = arg2; + nvlist_t *nvl = arg; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = spa->spa_meta_objset; dmu_buf_t *dbp; spa_history_phys_t *shpp; @@ -220,7 +220,7 @@ spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx) * Get the offset of where we need to write via the bonus buffer. * Update the offset when the write completes. */ - VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); + VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)); shpp = dbp->db_data; dmu_buf_will_dirty(dbp, tx); @@ -321,8 +321,8 @@ spa_history_log_nvl(spa_t *spa, nvlist_t *nvl) fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED())); /* Kick this off asynchronously; errors are ignored. */ - dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL, - spa_history_log_sync, spa, nvarg, 0, tx); + dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync, + nvarg, 0, tx); dmu_tx_commit(tx); /* spa_history_log_sync will free nvl */ @@ -455,10 +455,10 @@ log_internal(nvlist_t *nvl, const char *operation, spa_t *spa, fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg); if (dmu_tx_is_syncing(tx)) { - spa_history_log_sync(spa, nvl, tx); + spa_history_log_sync(nvl, tx); } else { - dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL, - spa_history_log_sync, spa, nvl, 0, tx); + dsl_sync_task_nowait(spa_get_dsl(spa), + spa_history_log_sync, nvl, 0, tx); } /* spa_history_log_sync() will free nvl */ } @@ -530,15 +530,9 @@ spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation, void spa_history_log_version(spa_t *spa, const char *operation) { -#ifdef _KERNEL - uint64_t current_vers = spa_version(spa); - spa_history_log_internal(spa, operation, NULL, "pool version %llu; software version %llu/%d; uts %s %s %s %s", - (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION, + (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION, utsname.nodename, utsname.release, utsname.version, utsname.machine); - cmn_err(CE_CONT, "!%s version %llu pool %s using %llu", operation, - (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION); -#endif } diff --git a/uts/common/fs/zfs/spa_misc.c b/uts/common/fs/zfs/spa_misc.c index 405d93c6ced4..733d2609e55e 100644 --- a/uts/common/fs/zfs/spa_misc.c +++ b/uts/common/fs/zfs/spa_misc.c @@ -237,8 +237,8 @@ kmem_cache_t *spa_buffer_pool; int spa_mode_global; #ifdef ZFS_DEBUG -/* Everything except dprintf is on by default in debug builds */ -int zfs_flags = ~ZFS_DEBUG_DPRINTF; +/* Everything except dprintf and spa is on by default in debug builds */ +int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); #else int zfs_flags = 0; #endif @@ -282,7 +282,7 @@ spa_config_lock_init(spa_t *spa) spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); - refcount_create(&scl->scl_count); + refcount_create_untracked(&scl->scl_count); scl->scl_writer = NULL; scl->scl_write_wanted = 0; } @@ -335,6 +335,8 @@ spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) { int wlocks_held = 0; + ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); + for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (scl->scl_writer == curthread) @@ -413,27 +415,22 @@ spa_lookup(const char *name) static spa_t search; /* spa_t is large; don't allocate on stack */ spa_t *spa; avl_index_t where; - char c; char *cp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); + (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); + /* * If it's a full dataset name, figure out the pool name and * just use that. */ - cp = strpbrk(name, "/@"); - if (cp) { - c = *cp; + cp = strpbrk(search.spa_name, "/@"); + if (cp != NULL) *cp = '\0'; - } - (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); spa = avl_find(&spa_namespace_avl, &search, &where); - if (cp) - *cp = c; - return (spa); } @@ -567,6 +564,8 @@ spa_add(const char *name, nvlist_t *config, const char *altroot) kstat_install(spa->spa_iokstat); } + spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); + return (spa); } diff --git a/uts/common/fs/zfs/space_map.c b/uts/common/fs/zfs/space_map.c index 30a35c85dab2..fb30b344704f 100644 --- a/uts/common/fs/zfs/space_map.c +++ b/uts/common/fs/zfs/space_map.c @@ -102,7 +102,7 @@ void space_map_add(space_map_t *sm, uint64_t start, uint64_t size) { avl_index_t where; - space_seg_t ssearch, *ss_before, *ss_after, *ss; + space_seg_t *ss_before, *ss_after, *ss; uint64_t end = start + size; int merge_before, merge_after; @@ -115,11 +115,8 @@ space_map_add(space_map_t *sm, uint64_t start, uint64_t size) VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0); VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); - ssearch.ss_start = start; - ssearch.ss_end = end; - ss = avl_find(&sm->sm_root, &ssearch, &where); - - if (ss != NULL && ss->ss_start <= start && ss->ss_end >= end) { + ss = space_map_find(sm, start, size, &where); + if (ss != NULL) { zfs_panic_recover("zfs: allocating allocated segment" "(offset=%llu size=%llu)\n", (longlong_t)start, (longlong_t)size); @@ -171,19 +168,12 @@ void space_map_remove(space_map_t *sm, uint64_t start, uint64_t size) { avl_index_t where; - space_seg_t ssearch, *ss, *newseg; + space_seg_t *ss, *newseg; uint64_t end = start + size; int left_over, right_over; - ASSERT(MUTEX_HELD(sm->sm_lock)); VERIFY(!sm->sm_condensing); - VERIFY(size != 0); - VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0); - VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); - - ssearch.ss_start = start; - ssearch.ss_end = end; - ss = avl_find(&sm->sm_root, &ssearch, &where); + ss = space_map_find(sm, start, size, &where); /* Make sure we completely overlap with someone */ if (ss == NULL) { @@ -226,12 +216,11 @@ space_map_remove(space_map_t *sm, uint64_t start, uint64_t size) sm->sm_space -= size; } -boolean_t -space_map_contains(space_map_t *sm, uint64_t start, uint64_t size) +space_seg_t * +space_map_find(space_map_t *sm, uint64_t start, uint64_t size, + avl_index_t *wherep) { - avl_index_t where; space_seg_t ssearch, *ss; - uint64_t end = start + size; ASSERT(MUTEX_HELD(sm->sm_lock)); VERIFY(size != 0); @@ -239,10 +228,20 @@ space_map_contains(space_map_t *sm, uint64_t start, uint64_t size) VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0); ssearch.ss_start = start; - ssearch.ss_end = end; - ss = avl_find(&sm->sm_root, &ssearch, &where); + ssearch.ss_end = start + size; + ss = avl_find(&sm->sm_root, &ssearch, wherep); - return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end); + if (ss != NULL && ss->ss_start <= start && ss->ss_end >= start + size) + return (ss); + return (NULL); +} + +boolean_t +space_map_contains(space_map_t *sm, uint64_t start, uint64_t size) +{ + avl_index_t where; + + return (space_map_find(sm, start, size, &where) != 0); } void diff --git a/uts/common/fs/zfs/sys/arc.h b/uts/common/fs/zfs/sys/arc.h index 916d2abf1477..0e86290f2dfb 100644 --- a/uts/common/fs/zfs/sys/arc.h +++ b/uts/common/fs/zfs/sys/arc.h @@ -89,7 +89,7 @@ arc_buf_t *arc_loan_buf(spa_t *spa, int size); void arc_return_buf(arc_buf_t *buf, void *tag); void arc_loan_inuse_buf(arc_buf_t *buf, void *tag); void arc_buf_add_ref(arc_buf_t *buf, void *tag); -int arc_buf_remove_ref(arc_buf_t *buf, void *tag); +boolean_t arc_buf_remove_ref(arc_buf_t *buf, void *tag); int arc_buf_size(arc_buf_t *buf); void arc_release(arc_buf_t *buf, void *tag); int arc_released(arc_buf_t *buf); diff --git a/uts/common/fs/zfs/sys/dbuf.h b/uts/common/fs/zfs/sys/dbuf.h index 8591f1585128..a29f7b3ccf19 100644 --- a/uts/common/fs/zfs/sys/dbuf.h +++ b/uts/common/fs/zfs/sys/dbuf.h @@ -311,20 +311,17 @@ void dbuf_fini(void); boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); -#define DBUF_IS_METADATA(_db) \ - (dbuf_is_metadata(_db)) - #define DBUF_GET_BUFC_TYPE(_db) \ - (DBUF_IS_METADATA(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) + (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) #define DBUF_IS_CACHEABLE(_db) \ ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ - (DBUF_IS_METADATA(_db) && \ + (dbuf_is_metadata(_db) && \ ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) #define DBUF_IS_L2CACHEABLE(_db) \ ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \ - (DBUF_IS_METADATA(_db) && \ + (dbuf_is_metadata(_db) && \ ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA))) #ifdef ZFS_DEBUG diff --git a/uts/common/fs/zfs/sys/dmu.h b/uts/common/fs/zfs/sys/dmu.h index ef0a6a7c3731..1366a998fd5a 100644 --- a/uts/common/fs/zfs/sys/dmu.h +++ b/uts/common/fs/zfs/sys/dmu.h @@ -217,6 +217,11 @@ typedef enum dmu_object_type { DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE), } dmu_object_type_t; +typedef enum txg_how { + TXG_WAIT = 1, + TXG_NOWAIT, +} txg_how_t; + void byteswap_uint64_array(void *buf, size_t size); void byteswap_uint32_array(void *buf, size_t size); void byteswap_uint16_array(void *buf, size_t size); @@ -255,22 +260,19 @@ void dmu_objset_rele(objset_t *os, void *tag); void dmu_objset_disown(objset_t *os, void *tag); int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp); -int dmu_objset_evict_dbufs(objset_t *os); +void dmu_objset_evict_dbufs(objset_t *os); int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg); -int dmu_objset_clone(const char *name, struct dsl_dataset *clone_origin, - uint64_t flags); -int dmu_objset_destroy(const char *name, boolean_t defer); -int dmu_snapshots_destroy_nvl(struct nvlist *snaps, boolean_t defer, +int dmu_objset_clone(const char *name, const char *origin); +int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, struct nvlist *errlist); -int dmu_objset_snapshot(struct nvlist *snaps, struct nvlist *, struct nvlist *); int dmu_objset_snapshot_one(const char *fsname, const char *snapname); int dmu_objset_snapshot_tmp(const char *, const char *, int); -int dmu_objset_rename(const char *name, const char *newname, - boolean_t recursive); int dmu_objset_find(char *name, int func(const char *, void *), void *arg, int flags); void dmu_objset_byteswap(void *buf, size_t size); +int dsl_dataset_rename_snapshot(const char *fsname, + const char *oldsnapname, const char *newsnapname, boolean_t recursive); typedef struct dmu_buf { uint64_t db_object; /* object that this buffer is part of */ @@ -545,7 +547,7 @@ void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object); void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow); void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size); void dmu_tx_abort(dmu_tx_t *tx); -int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how); +int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how); void dmu_tx_wait(dmu_tx_t *tx); void dmu_tx_commit(dmu_tx_t *tx); @@ -788,36 +790,8 @@ typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp, void dmu_traverse_objset(objset_t *os, uint64_t txg_start, dmu_traverse_cb_t cb, void *arg); -int dmu_send(objset_t *tosnap, objset_t *fromsnap, - int outfd, struct vnode *vp, offset_t *off); -int dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep); - -typedef struct dmu_recv_cookie { - /* - * This structure is opaque! - * - * If logical and real are different, we are recving the stream - * into the "real" temporary clone, and then switching it with - * the "logical" target. - */ - struct dsl_dataset *drc_logical_ds; - struct dsl_dataset *drc_real_ds; - struct drr_begin *drc_drrb; - char *drc_tosnap; - char *drc_top_ds; - boolean_t drc_newfs; - boolean_t drc_force; - struct avl_tree *drc_guid_to_ds_map; -} dmu_recv_cookie_t; - -int dmu_recv_begin(char *tofs, char *tosnap, char *topds, struct drr_begin *, - boolean_t force, objset_t *origin, dmu_recv_cookie_t *); -int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp, - int cleanup_fd, uint64_t *action_handlep); -int dmu_recv_end(dmu_recv_cookie_t *drc); - -int dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp, - offset_t *off); +int dmu_diff(const char *tosnap_name, const char *fromsnap_name, + struct vnode *vp, offset_t *offp); /* CRC64 table */ #define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */ diff --git a/uts/common/fs/zfs/sys/dmu_objset.h b/uts/common/fs/zfs/sys/dmu_objset.h index 578b94732ed0..143e594d1d1d 100644 --- a/uts/common/fs/zfs/sys/dmu_objset.h +++ b/uts/common/fs/zfs/sys/dmu_objset.h @@ -43,6 +43,7 @@ extern "C" { extern krwlock_t os_lock; +struct dsl_pool; struct dsl_dataset; struct dmu_tx; @@ -114,8 +115,6 @@ struct objset { /* stuff we store for the user */ kmutex_t os_user_ptr_lock; void *os_user_ptr; - - /* SA layout/attribute registration */ sa_os_t *os_sa; }; @@ -143,10 +142,11 @@ void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp); uint64_t dmu_objset_fsid_guid(objset_t *os); -int dmu_objset_find_spa(spa_t *spa, const char *name, - int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags); +int dmu_objset_find_dp(struct dsl_pool *dp, uint64_t ddobj, + int func(struct dsl_pool *, struct dsl_dataset *, void *), + void *arg, int flags); int dmu_objset_prefetch(const char *name, void *arg); -int dmu_objset_evict_dbufs(objset_t *os); +void dmu_objset_evict_dbufs(objset_t *os); timestruc_t dmu_objset_snap_cmtime(objset_t *os); /* called from dsl */ @@ -162,6 +162,7 @@ void dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx); boolean_t dmu_objset_userused_enabled(objset_t *os); int dmu_objset_userspace_upgrade(objset_t *os); boolean_t dmu_objset_userspace_present(objset_t *os); +int dmu_fsname(const char *snapname, char *buf); void dmu_objset_init(void); void dmu_objset_fini(void); diff --git a/uts/common/fs/zfs/sys/dmu_send.h b/uts/common/fs/zfs/sys/dmu_send.h new file mode 100644 index 000000000000..ee0885a60ff6 --- /dev/null +++ b/uts/common/fs/zfs/sys/dmu_send.h @@ -0,0 +1,66 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. + * Copyright 2011 Nexenta Systems, Inc. All rights reserved. + * Copyright (c) 2012, Joyent, Inc. All rights reserved. + */ + +#ifndef _DMU_SEND_H +#define _DMU_SEND_H + +#include +#include + +struct vnode; +struct dsl_dataset; +struct drr_begin; +struct avl_tree; + +int dmu_send(const char *tosnap, const char *fromsnap, int outfd, + struct vnode *vp, offset_t *off); +int dmu_send_estimate(struct dsl_dataset *ds, struct dsl_dataset *fromds, + uint64_t *sizep); +int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, + int outfd, struct vnode *vp, offset_t *off); + +typedef struct dmu_recv_cookie { + struct dsl_dataset *drc_ds; + struct drr_begin *drc_drrb; + const char *drc_tofs; + const char *drc_tosnap; + boolean_t drc_newfs; + boolean_t drc_byteswap; + boolean_t drc_force; + struct avl_tree *drc_guid_to_ds_map; + zio_cksum_t drc_cksum; + uint64_t drc_newsnapobj; +} dmu_recv_cookie_t; + +int dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb, + boolean_t force, char *origin, dmu_recv_cookie_t *drc); +int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp, + int cleanup_fd, uint64_t *action_handlep); +int dmu_recv_end(dmu_recv_cookie_t *drc); + +#endif /* _DMU_SEND_H */ diff --git a/uts/common/fs/zfs/sys/dmu_tx.h b/uts/common/fs/zfs/sys/dmu_tx.h index c5ea50fa8d82..dbd2242540ef 100644 --- a/uts/common/fs/zfs/sys/dmu_tx.h +++ b/uts/common/fs/zfs/sys/dmu_tx.h @@ -22,6 +22,9 @@ * Copyright 2010 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ +/* + * Copyright (c) 2012 by Delphix. All rights reserved. + */ #ifndef _SYS_DMU_TX_H #define _SYS_DMU_TX_H @@ -108,10 +111,11 @@ typedef struct dmu_tx_callback { * These routines are defined in dmu.h, and are called by the user. */ dmu_tx_t *dmu_tx_create(objset_t *dd); -int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how); +int dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how); void dmu_tx_commit(dmu_tx_t *tx); void dmu_tx_abort(dmu_tx_t *tx); uint64_t dmu_tx_get_txg(dmu_tx_t *tx); +struct dsl_pool *dmu_tx_pool(dmu_tx_t *tx); void dmu_tx_wait(dmu_tx_t *tx); void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func, diff --git a/uts/common/fs/zfs/sys/dsl_dataset.h b/uts/common/fs/zfs/sys/dsl_dataset.h index 272c3ecde249..6729f9f05a96 100644 --- a/uts/common/fs/zfs/sys/dsl_dataset.h +++ b/uts/common/fs/zfs/sys/dsl_dataset.h @@ -35,6 +35,7 @@ #include #include #include +#include #ifdef __cplusplus extern "C" { @@ -48,10 +49,8 @@ struct dsl_pool; #define DS_IS_INCONSISTENT(ds) \ ((ds)->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) /* - * NB: nopromote can not yet be set, but we want support for it in this - * on-disk version, so that we don't need to upgrade for it later. It - * will be needed when we implement 'zfs split' (where the split off - * clone should not be promoted). + * Note: nopromote can not yet be set, but we want support for it in this + * on-disk version, so that we don't need to upgrade for it later. */ #define DS_FLAG_NOPROMOTE (1ULL<<1) @@ -76,6 +75,8 @@ struct dsl_pool; */ #define DS_FLAG_CI_DATASET (1ULL<<16) +#define DS_CREATE_FLAG_NODIRTY (1ULL<<24) + typedef struct dsl_dataset_phys { uint64_t ds_dir_obj; /* DMU_OT_DSL_DIR */ uint64_t ds_prev_snap_obj; /* DMU_OT_DSL_DATASET */ @@ -125,9 +126,6 @@ typedef struct dsl_dataset { dsl_deadlist_t ds_deadlist; bplist_t ds_pending_deadlist; - /* to protect against multiple concurrent incremental recv */ - kmutex_t ds_recvlock; - /* protected by lock on pool's dp_dirty_datasets list */ txg_node_t ds_dirty_link; list_node_t ds_synced_link; @@ -139,13 +137,15 @@ typedef struct dsl_dataset { kmutex_t ds_lock; objset_t *ds_objset; uint64_t ds_userrefs; + void *ds_owner; /* - * ds_owner is protected by the ds_rwlock and the ds_lock + * Long holds prevent the ds from being destroyed; they allow the + * ds to remain held even after dropping the dp_config_rwlock. + * Owning counts as a long hold. See the comments above + * dsl_pool_hold() for details. */ - krwlock_t ds_rwlock; - kcondvar_t ds_exclusive_cv; - void *ds_owner; + refcount_t ds_longholds; /* no locking; only for making guesses */ uint64_t ds_trysnap_txg; @@ -163,76 +163,44 @@ typedef struct dsl_dataset { char ds_snapname[MAXNAMELEN]; } dsl_dataset_t; -struct dsl_ds_destroyarg { - dsl_dataset_t *ds; /* ds to destroy */ - dsl_dataset_t *rm_origin; /* also remove our origin? */ - boolean_t is_origin_rm; /* set if removing origin snap */ - boolean_t defer; /* destroy -d requested? */ - boolean_t releasing; /* destroying due to release? */ - boolean_t need_prep; /* do we need to retry due to EBUSY? */ -}; - /* * The max length of a temporary tag prefix is the number of hex digits * required to express UINT64_MAX plus one for the hyphen. */ #define MAX_TAG_PREFIX_LEN 17 -struct dsl_ds_holdarg { - dsl_sync_task_group_t *dstg; - const char *htag; - char *snapname; - boolean_t recursive; - boolean_t gotone; - boolean_t temphold; - char failed[MAXPATHLEN]; -}; - #define dsl_dataset_is_snapshot(ds) \ ((ds)->ds_phys->ds_num_children != 0) #define DS_UNIQUE_IS_ACCURATE(ds) \ (((ds)->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0) -int dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp); -int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, - void *tag, dsl_dataset_t **); -int dsl_dataset_own(const char *name, boolean_t inconsistentok, +int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag, + dsl_dataset_t **dsp); +int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag, + dsl_dataset_t **); +void dsl_dataset_rele(dsl_dataset_t *ds, void *tag); +int dsl_dataset_own(struct dsl_pool *dp, const char *name, void *tag, dsl_dataset_t **dsp); int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj, - boolean_t inconsistentok, void *tag, dsl_dataset_t **dsp); -void dsl_dataset_name(dsl_dataset_t *ds, char *name); -void dsl_dataset_rele(dsl_dataset_t *ds, void *tag); + void *tag, dsl_dataset_t **dsp); void dsl_dataset_disown(dsl_dataset_t *ds, void *tag); -void dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag); -boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, - void *tag); -void dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *tag); +void dsl_dataset_name(dsl_dataset_t *ds, char *name); +boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag); void dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, minor_t minor); uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname, dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *); uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, uint64_t flags, dmu_tx_t *tx); -int dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer); -dsl_checkfunc_t dsl_dataset_destroy_check; -dsl_syncfunc_t dsl_dataset_destroy_sync; -dsl_syncfunc_t dsl_dataset_user_hold_sync; -int dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *, dmu_tx_t *tx); -void dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *, dmu_tx_t *tx); -int dsl_dataset_rename(char *name, const char *newname, boolean_t recursive); +int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors); int dsl_dataset_promote(const char *name, char *conflsnap); int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, boolean_t force); -int dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, - boolean_t recursive, boolean_t temphold, int cleanup_fd); -int dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag, - boolean_t temphold); -int dsl_dataset_user_release(char *dsname, char *snapname, char *htag, - boolean_t recursive); -int dsl_dataset_user_release_tmp(struct dsl_pool *dp, uint64_t dsobj, - char *htag, boolean_t retry); -int dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp); +int dsl_dataset_rename_snapshot(const char *fsname, + const char *oldsnapname, const char *newsnapname, boolean_t recursive); +int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, + minor_t cleanup_minor, const char *htag); blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds); void dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx); @@ -269,13 +237,35 @@ int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf); int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv); -int dsl_dataset_set_quota(const char *dsname, zprop_source_t source, +int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, uint64_t quota); -dsl_syncfunc_t dsl_dataset_set_quota_sync; -int dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, +int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, uint64_t reservation); -int dsl_destroy_inconsistent(const char *dsname, void *arg); +boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier); +void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag); +void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag); +boolean_t dsl_dataset_long_held(dsl_dataset_t *ds); + +int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, + dsl_dataset_t *origin_head, boolean_t force); +void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, + dsl_dataset_t *origin_head, dmu_tx_t *tx); +int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, + dmu_tx_t *tx); +void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, + dmu_tx_t *tx); + +void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, + dmu_tx_t *tx); +void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds); +int dsl_dataset_get_snapname(dsl_dataset_t *ds); +int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, + uint64_t *value); +int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx); +void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, + zprop_source_t source, uint64_t value, dmu_tx_t *tx); +int dsl_dataset_rollback(const char *fsname); #ifdef ZFS_DEBUG #define dprintf_ds(ds, fmt, ...) do { \ diff --git a/uts/common/fs/zfs/sys/dsl_destroy.h b/uts/common/fs/zfs/sys/dsl_destroy.h new file mode 100644 index 000000000000..c5a70bb90e4c --- /dev/null +++ b/uts/common/fs/zfs/sys/dsl_destroy.h @@ -0,0 +1,52 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. + * Copyright (c) 2012, Joyent, Inc. All rights reserved. + */ + +#ifndef _SYS_DSL_DESTROY_H +#define _SYS_DSL_DESTROY_H + +#ifdef __cplusplus +extern "C" { +#endif + +struct nvlist; +struct dsl_dataset; +struct dmu_tx; + +int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer, + struct nvlist *errlist); +int dsl_destroy_snapshot(const char *name, boolean_t defer); +int dsl_destroy_head(const char *name); +int dsl_destroy_head_check_impl(struct dsl_dataset *ds, int expected_holds); +void dsl_destroy_head_sync_impl(struct dsl_dataset *ds, struct dmu_tx *tx); +int dsl_destroy_inconsistent(const char *dsname, void *arg); +void dsl_destroy_snapshot_sync_impl(struct dsl_dataset *ds, + boolean_t defer, struct dmu_tx *tx); + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DSL_DESTROY_H */ diff --git a/uts/common/fs/zfs/sys/dsl_dir.h b/uts/common/fs/zfs/sys/dsl_dir.h index 2191635dd813..641bcfcdd300 100644 --- a/uts/common/fs/zfs/sys/dsl_dir.h +++ b/uts/common/fs/zfs/sys/dsl_dir.h @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_DSL_DIR_H @@ -101,18 +102,15 @@ struct dsl_dir { char dd_myname[MAXNAMELEN]; }; -void dsl_dir_close(dsl_dir_t *dd, void *tag); -int dsl_dir_open(const char *name, void *tag, dsl_dir_t **, const char **tail); -int dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, dsl_dir_t **, - const char **tailp); -int dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj, +void dsl_dir_rele(dsl_dir_t *dd, void *tag); +int dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, + dsl_dir_t **, const char **tail); +int dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, const char *tail, void *tag, dsl_dir_t **); void dsl_dir_name(dsl_dir_t *dd, char *buf); int dsl_dir_namelen(dsl_dir_t *dd); uint64_t dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, dmu_tx_t *tx); -dsl_checkfunc_t dsl_dir_destroy_check; -dsl_syncfunc_t dsl_dir_destroy_sync; void dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv); uint64_t dsl_dir_space_available(dsl_dir_t *dd, dsl_dir_t *ancestor, int64_t delta, int ondiskonly); @@ -131,14 +129,15 @@ int dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota); int dsl_dir_set_reservation(const char *ddname, zprop_source_t source, uint64_t reservation); -int dsl_dir_rename(dsl_dir_t *dd, const char *newname); +int dsl_dir_rename(const char *oldname, const char *newname); int dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space); -int dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx); boolean_t dsl_dir_is_clone(dsl_dir_t *dd); void dsl_dir_new_refreservation(dsl_dir_t *dd, struct dsl_dataset *ds, uint64_t reservation, cred_t *cr, dmu_tx_t *tx); void dsl_dir_snap_cmtime_update(dsl_dir_t *dd); timestruc_t dsl_dir_snap_cmtime(dsl_dir_t *dd); +void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, + dmu_tx_t *tx); /* internal reserved dir name */ #define MOS_DIR_NAME "$MOS" diff --git a/uts/common/fs/zfs/sys/dsl_pool.h b/uts/common/fs/zfs/sys/dsl_pool.h index ab1229a2e615..b0160edfb1ff 100644 --- a/uts/common/fs/zfs/sys/dsl_pool.h +++ b/uts/common/fs/zfs/sys/dsl_pool.h @@ -36,6 +36,7 @@ #include #include #include +#include #ifdef __cplusplus extern "C" { @@ -113,7 +114,7 @@ typedef struct dsl_pool { * syncing context does not need to ever have it for read, since * nobody else could possibly have it for write. */ - krwlock_t dp_config_rwlock; + rrwlock_t dp_config_rwlock; zfs_all_blkstats_t *dp_blkstats; } dsl_pool_t; @@ -139,15 +140,20 @@ void dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx); void dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx); void dsl_pool_mos_diduse_space(dsl_pool_t *dp, int64_t used, int64_t comp, int64_t uncomp); +void dsl_pool_config_enter(dsl_pool_t *dp, void *tag); +void dsl_pool_config_exit(dsl_pool_t *dp, void *tag); +boolean_t dsl_pool_config_held(dsl_pool_t *dp); taskq_t *dsl_pool_vnrele_taskq(dsl_pool_t *dp); -extern int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, - const char *tag, uint64_t *now, dmu_tx_t *tx); -extern int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, +int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, + const char *tag, uint64_t now, dmu_tx_t *tx); +int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, dmu_tx_t *tx); -extern void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp); +void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp); int dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **); +int dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp); +void dsl_pool_rele(dsl_pool_t *dp, void *tag); #ifdef __cplusplus } diff --git a/uts/common/fs/zfs/sys/dsl_prop.h b/uts/common/fs/zfs/sys/dsl_prop.h index b0d9a52cdfd7..5fe18d6a7c55 100644 --- a/uts/common/fs/zfs/sys/dsl_prop.h +++ b/uts/common/fs/zfs/sys/dsl_prop.h @@ -54,58 +54,47 @@ typedef struct dsl_props_arg { zprop_source_t pa_source; } dsl_props_arg_t; -typedef struct dsl_prop_set_arg { - const char *psa_name; - zprop_source_t psa_source; - int psa_intsz; - int psa_numints; - const void *psa_value; - - /* - * Used to handle the special requirements of the quota and reservation - * properties. - */ - uint64_t psa_effective_value; -} dsl_prop_setarg_t; - int dsl_prop_register(struct dsl_dataset *ds, const char *propname, dsl_prop_changed_cb_t *callback, void *cbarg); int dsl_prop_unregister(struct dsl_dataset *ds, const char *propname, dsl_prop_changed_cb_t *callback, void *cbarg); -int dsl_prop_numcb(struct dsl_dataset *ds); +void dsl_prop_notify_all(struct dsl_dir *dd); +boolean_t dsl_prop_hascb(struct dsl_dataset *ds); int dsl_prop_get(const char *ddname, const char *propname, int intsz, int numints, void *buf, char *setpoint); int dsl_prop_get_integer(const char *ddname, const char *propname, uint64_t *valuep, char *setpoint); int dsl_prop_get_all(objset_t *os, nvlist_t **nvp); -int dsl_prop_get_received(objset_t *os, nvlist_t **nvp); +int dsl_prop_get_received(const char *dsname, nvlist_t **nvp); int dsl_prop_get_ds(struct dsl_dataset *ds, const char *propname, int intsz, int numints, void *buf, char *setpoint); +int dsl_prop_get_int_ds(struct dsl_dataset *ds, const char *propname, + uint64_t *valuep); int dsl_prop_get_dd(struct dsl_dir *dd, const char *propname, int intsz, int numints, void *buf, char *setpoint, boolean_t snapshot); -dsl_syncfunc_t dsl_props_set_sync; -int dsl_prop_set(const char *ddname, const char *propname, - zprop_source_t source, int intsz, int numints, const void *buf); +void dsl_props_set_sync_impl(struct dsl_dataset *ds, zprop_source_t source, + nvlist_t *props, dmu_tx_t *tx); +void dsl_prop_set_sync_impl(struct dsl_dataset *ds, const char *propname, + zprop_source_t source, int intsz, int numints, const void *value, + dmu_tx_t *tx); int dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *nvl); +int dsl_prop_set_int(const char *dsname, const char *propname, + zprop_source_t source, uint64_t value); +int dsl_prop_set_string(const char *dsname, const char *propname, + zprop_source_t source, const char *value); +int dsl_prop_inherit(const char *dsname, const char *propname, + zprop_source_t source); -void dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname, - zprop_source_t source, uint64_t *value); -int dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa); -#ifdef ZFS_DEBUG -void dsl_prop_check_prediction(dsl_dir_t *dd, dsl_prop_setarg_t *psa); -#define DSL_PROP_CHECK_PREDICTION(dd, psa) \ - dsl_prop_check_prediction((dd), (psa)) -#else -#define DSL_PROP_CHECK_PREDICTION(dd, psa) /* nothing */ -#endif +int dsl_prop_predict(dsl_dir_t *dd, const char *propname, + zprop_source_t source, uint64_t value, uint64_t *newvalp); /* flag first receive on or after SPA_VERSION_RECVD_PROPS */ -boolean_t dsl_prop_get_hasrecvd(objset_t *os); -void dsl_prop_set_hasrecvd(objset_t *os); -void dsl_prop_unset_hasrecvd(objset_t *os); +boolean_t dsl_prop_get_hasrecvd(const char *dsname); +int dsl_prop_set_hasrecvd(const char *dsname); +void dsl_prop_unset_hasrecvd(const char *dsname); void dsl_prop_nvlist_add_uint64(nvlist_t *nv, zfs_prop_t prop, uint64_t value); void dsl_prop_nvlist_add_string(nvlist_t *nv, diff --git a/uts/common/fs/zfs/sys/dsl_synctask.h b/uts/common/fs/zfs/sys/dsl_synctask.h index 9126290cdb5b..ef86fb64cf0c 100644 --- a/uts/common/fs/zfs/sys/dsl_synctask.h +++ b/uts/common/fs/zfs/sys/dsl_synctask.h @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_DSL_SYNCTASK_H @@ -34,43 +35,26 @@ extern "C" { struct dsl_pool; -typedef int (dsl_checkfunc_t)(void *, void *, dmu_tx_t *); -typedef void (dsl_syncfunc_t)(void *, void *, dmu_tx_t *); +typedef int (dsl_checkfunc_t)(void *, dmu_tx_t *); +typedef void (dsl_syncfunc_t)(void *, dmu_tx_t *); typedef struct dsl_sync_task { - list_node_t dst_node; + txg_node_t dst_node; + struct dsl_pool *dst_pool; + uint64_t dst_txg; + int dst_space; dsl_checkfunc_t *dst_checkfunc; dsl_syncfunc_t *dst_syncfunc; - void *dst_arg1; - void *dst_arg2; - int dst_err; + void *dst_arg; + int dst_error; + boolean_t dst_nowaiter; } dsl_sync_task_t; -typedef struct dsl_sync_task_group { - txg_node_t dstg_node; - list_t dstg_tasks; - struct dsl_pool *dstg_pool; - uint64_t dstg_txg; - int dstg_err; - int dstg_space; - boolean_t dstg_nowaiter; -} dsl_sync_task_group_t; - -dsl_sync_task_group_t *dsl_sync_task_group_create(struct dsl_pool *dp); -void dsl_sync_task_create(dsl_sync_task_group_t *dstg, - dsl_checkfunc_t *, dsl_syncfunc_t *, - void *arg1, void *arg2, int blocks_modified); -int dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg); -void dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx); -void dsl_sync_task_group_destroy(dsl_sync_task_group_t *dstg); -void dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx); - -int dsl_sync_task_do(struct dsl_pool *dp, - dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc, - void *arg1, void *arg2, int blocks_modified); -void dsl_sync_task_do_nowait(struct dsl_pool *dp, - dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc, - void *arg1, void *arg2, int blocks_modified, dmu_tx_t *tx); +void dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx); +int dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc, + dsl_syncfunc_t *syncfunc, void *arg, int blocks_modified); +void dsl_sync_task_nowait(struct dsl_pool *dp, dsl_syncfunc_t *syncfunc, + void *arg, int blocks_modified, dmu_tx_t *tx); #ifdef __cplusplus } diff --git a/uts/common/fs/zfs/sys/dsl_userhold.h b/uts/common/fs/zfs/sys/dsl_userhold.h new file mode 100644 index 000000000000..56c6c8f47a87 --- /dev/null +++ b/uts/common/fs/zfs/sys/dsl_userhold.h @@ -0,0 +1,57 @@ + +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. + * Copyright (c) 2012, Joyent, Inc. All rights reserved. + */ + +#ifndef _SYS_DSL_USERHOLD_H +#define _SYS_DSL_USERHOLD_H + +#include +#include + +#ifdef __cplusplus +extern "C" { +#endif + +struct dsl_pool; +struct dsl_dataset; +struct dmu_tx; + +int dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, + nvlist_t *errlist); +int dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist); +int dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl); +void dsl_dataset_user_release_tmp(struct dsl_pool *dp, uint64_t dsobj, + const char *htag); +int dsl_dataset_user_hold_check_one(struct dsl_dataset *ds, const char *htag, + boolean_t temphold, struct dmu_tx *tx); +void dsl_dataset_user_hold_sync_one(struct dsl_dataset *ds, const char *htag, + minor_t minor, uint64_t now, struct dmu_tx *tx); + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DSL_USERHOLD_H */ diff --git a/uts/common/fs/zfs/sys/metaslab.h b/uts/common/fs/zfs/sys/metaslab.h index 2cf4d2b489bd..d6c0bf4c94cd 100644 --- a/uts/common/fs/zfs/sys/metaslab.h +++ b/uts/common/fs/zfs/sys/metaslab.h @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_METASLAB_H @@ -56,6 +56,7 @@ extern int metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, extern void metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now); extern int metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg); +extern void metaslab_check_free(spa_t *spa, const blkptr_t *bp); extern metaslab_class_t *metaslab_class_create(spa_t *spa, space_map_ops_t *ops); diff --git a/uts/common/fs/zfs/sys/refcount.h b/uts/common/fs/zfs/sys/refcount.h index 1dcd467f91ae..9efc5f1e445b 100644 --- a/uts/common/fs/zfs/sys/refcount.h +++ b/uts/common/fs/zfs/sys/refcount.h @@ -51,15 +51,17 @@ typedef struct reference { typedef struct refcount { kmutex_t rc_mtx; + boolean_t rc_tracked; list_t rc_list; list_t rc_removed; uint64_t rc_count; uint64_t rc_removed_count; } refcount_t; -/* Note: refcount_t must be initialized with refcount_create() */ +/* Note: refcount_t must be initialized with refcount_create[_untracked]() */ void refcount_create(refcount_t *rc); +void refcount_create_untracked(refcount_t *rc); void refcount_destroy(refcount_t *rc); void refcount_destroy_many(refcount_t *rc, uint64_t number); int refcount_is_zero(refcount_t *rc); @@ -80,6 +82,7 @@ typedef struct refcount { } refcount_t; #define refcount_create(rc) ((rc)->rc_count = 0) +#define refcount_create_untracked(rc) ((rc)->rc_count = 0) #define refcount_destroy(rc) ((rc)->rc_count = 0) #define refcount_destroy_many(rc, number) ((rc)->rc_count = 0) #define refcount_is_zero(rc) ((rc)->rc_count == 0) diff --git a/uts/common/fs/zfs/sys/rrwlock.h b/uts/common/fs/zfs/sys/rrwlock.h index 239268bd58e7..e1e6d3122373 100644 --- a/uts/common/fs/zfs/sys/rrwlock.h +++ b/uts/common/fs/zfs/sys/rrwlock.h @@ -58,6 +58,7 @@ typedef struct rrwlock { refcount_t rr_anon_rcount; refcount_t rr_linked_rcount; boolean_t rr_writer_wanted; + boolean_t rr_track_all; } rrwlock_t; /* @@ -65,15 +66,19 @@ typedef struct rrwlock { * 'tag' must be the same in a rrw_enter() as in its * corresponding rrw_exit(). */ -void rrw_init(rrwlock_t *rrl); +void rrw_init(rrwlock_t *rrl, boolean_t track_all); void rrw_destroy(rrwlock_t *rrl); void rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag); +void rrw_enter_read(rrwlock_t *rrl, void *tag); +void rrw_enter_write(rrwlock_t *rrl); void rrw_exit(rrwlock_t *rrl, void *tag); boolean_t rrw_held(rrwlock_t *rrl, krw_t rw); void rrw_tsd_destroy(void *arg); #define RRW_READ_HELD(x) rrw_held(x, RW_READER) #define RRW_WRITE_HELD(x) rrw_held(x, RW_WRITER) +#define RRW_LOCK_HELD(x) \ + (rrw_held(x, RW_WRITER) || rrw_held(x, RW_READER)) #ifdef __cplusplus } diff --git a/uts/common/fs/zfs/sys/space_map.h b/uts/common/fs/zfs/sys/space_map.h index 2da50fb7b3b5..64223daf6206 100644 --- a/uts/common/fs/zfs/sys/space_map.h +++ b/uts/common/fs/zfs/sys/space_map.h @@ -149,6 +149,8 @@ extern void space_map_add(space_map_t *sm, uint64_t start, uint64_t size); extern void space_map_remove(space_map_t *sm, uint64_t start, uint64_t size); extern boolean_t space_map_contains(space_map_t *sm, uint64_t start, uint64_t size); +extern space_seg_t *space_map_find(space_map_t *sm, uint64_t start, + uint64_t size, avl_index_t *wherep); extern void space_map_swap(space_map_t **msrc, space_map_t **mdest); extern void space_map_vacate(space_map_t *sm, space_map_func_t *func, space_map_t *mdest); diff --git a/uts/common/fs/zfs/sys/txg.h b/uts/common/fs/zfs/sys/txg.h index 1287f09c7ee8..2df33f0fb0a7 100644 --- a/uts/common/fs/zfs/sys/txg.h +++ b/uts/common/fs/zfs/sys/txg.h @@ -45,9 +45,6 @@ extern "C" { /* Number of txgs worth of frees we defer adding to in-core spacemaps */ #define TXG_DEFER_SIZE 2 -#define TXG_WAIT 1ULL -#define TXG_NOWAIT 2ULL - typedef struct tx_cpu tx_cpu_t; typedef struct txg_handle { @@ -119,11 +116,11 @@ extern boolean_t txg_sync_waiting(struct dsl_pool *dp); extern void txg_list_create(txg_list_t *tl, size_t offset); extern void txg_list_destroy(txg_list_t *tl); extern boolean_t txg_list_empty(txg_list_t *tl, uint64_t txg); -extern int txg_list_add(txg_list_t *tl, void *p, uint64_t txg); -extern int txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg); +extern boolean_t txg_list_add(txg_list_t *tl, void *p, uint64_t txg); +extern boolean_t txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg); extern void *txg_list_remove(txg_list_t *tl, uint64_t txg); extern void *txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg); -extern int txg_list_member(txg_list_t *tl, void *p, uint64_t txg); +extern boolean_t txg_list_member(txg_list_t *tl, void *p, uint64_t txg); extern void *txg_list_head(txg_list_t *tl, uint64_t txg); extern void *txg_list_next(txg_list_t *tl, void *p, uint64_t txg); diff --git a/uts/common/fs/zfs/sys/zfeature.h b/uts/common/fs/zfs/sys/zfeature.h index 481e85b1bad0..1a081e422d4a 100644 --- a/uts/common/fs/zfs/sys/zfeature.h +++ b/uts/common/fs/zfs/sys/zfeature.h @@ -26,7 +26,6 @@ #ifndef _SYS_ZFEATURE_H #define _SYS_ZFEATURE_H -#include #include #include "zfeature_common.h" @@ -34,14 +33,18 @@ extern "C" { #endif -extern boolean_t feature_is_supported(objset_t *os, uint64_t obj, +struct spa; +struct dmu_tx; +struct objset; + +extern boolean_t feature_is_supported(struct objset *os, uint64_t obj, uint64_t desc_obj, nvlist_t *unsup_feat, nvlist_t *enabled_feat); -struct spa; -extern void spa_feature_create_zap_objects(struct spa *, dmu_tx_t *); -extern void spa_feature_enable(struct spa *, zfeature_info_t *, dmu_tx_t *); -extern void spa_feature_incr(struct spa *, zfeature_info_t *, dmu_tx_t *); -extern void spa_feature_decr(struct spa *, zfeature_info_t *, dmu_tx_t *); +extern void spa_feature_create_zap_objects(struct spa *, struct dmu_tx *); +extern void spa_feature_enable(struct spa *, zfeature_info_t *, + struct dmu_tx *); +extern void spa_feature_incr(struct spa *, zfeature_info_t *, struct dmu_tx *); +extern void spa_feature_decr(struct spa *, zfeature_info_t *, struct dmu_tx *); extern boolean_t spa_feature_is_enabled(struct spa *, zfeature_info_t *); extern boolean_t spa_feature_is_active(struct spa *, zfeature_info_t *); diff --git a/uts/common/fs/zfs/sys/zfs_debug.h b/uts/common/fs/zfs/sys/zfs_debug.h index 94626229add7..14eb2abdc1cf 100644 --- a/uts/common/fs/zfs/sys/zfs_debug.h +++ b/uts/common/fs/zfs/sys/zfs_debug.h @@ -48,11 +48,13 @@ extern "C" { extern int zfs_flags; -#define ZFS_DEBUG_DPRINTF 0x0001 -#define ZFS_DEBUG_DBUF_VERIFY 0x0002 -#define ZFS_DEBUG_DNODE_VERIFY 0x0004 -#define ZFS_DEBUG_SNAPNAMES 0x0008 -#define ZFS_DEBUG_MODIFY 0x0010 +#define ZFS_DEBUG_DPRINTF (1<<0) +#define ZFS_DEBUG_DBUF_VERIFY (1<<1) +#define ZFS_DEBUG_DNODE_VERIFY (1<<2) +#define ZFS_DEBUG_SNAPNAMES (1<<3) +#define ZFS_DEBUG_MODIFY (1<<4) +#define ZFS_DEBUG_SPA (1<<5) +#define ZFS_DEBUG_ZIO_FREE (1<<6) #ifdef ZFS_DEBUG extern void __dprintf(const char *file, const char *func, diff --git a/uts/common/fs/zfs/sys/zfs_ioctl.h b/uts/common/fs/zfs/sys/zfs_ioctl.h index 86e901be0d7a..874d422568d9 100644 --- a/uts/common/fs/zfs/sys/zfs_ioctl.h +++ b/uts/common/fs/zfs/sys/zfs_ioctl.h @@ -293,7 +293,6 @@ typedef struct zfs_cmd { uint64_t zc_history; /* really (char *) */ char zc_value[MAXPATHLEN * 2]; char zc_string[MAXNAMELEN]; - char zc_top_ds[MAXPATHLEN]; uint64_t zc_guid; uint64_t zc_nvlist_conf; /* really (char *) */ uint64_t zc_nvlist_conf_size; @@ -345,7 +344,8 @@ extern int zfs_secpolicy_rename_perms(const char *from, const char *to, cred_t *cr); extern int zfs_secpolicy_destroy_perms(const char *name, cred_t *cr); extern int zfs_busy(void); -extern int zfs_unmount_snap(const char *, void *); +extern void zfs_unmount_snap(const char *); +extern void zfs_destroy_unmount_origin(const char *); /* * ZFS minor numbers can refer to either a control device instance or diff --git a/uts/common/fs/zfs/sys/zfs_znode.h b/uts/common/fs/zfs/sys/zfs_znode.h index 3e9621a0ee24..cf0bbee2cab3 100644 --- a/uts/common/fs/zfs/sys/zfs_znode.h +++ b/uts/common/fs/zfs/sys/zfs_znode.h @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_FS_ZFS_ZNODE_H @@ -240,7 +241,7 @@ typedef struct znode { */ #define ZFS_ENTER(zfsvfs) \ { \ - rrw_enter(&(zfsvfs)->z_teardown_lock, RW_READER, FTAG); \ + rrw_enter_read(&(zfsvfs)->z_teardown_lock, FTAG); \ if ((zfsvfs)->z_unmounted) { \ ZFS_EXIT(zfsvfs); \ return (EIO); \ diff --git a/uts/common/fs/zfs/sys/zil.h b/uts/common/fs/zfs/sys/zil.h index e52c65bb762f..a212e4f0e111 100644 --- a/uts/common/fs/zfs/sys/zil.h +++ b/uts/common/fs/zfs/sys/zil.h @@ -411,8 +411,8 @@ extern int zil_check_log_chain(const char *osname, void *txarg); extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); extern void zil_clean(zilog_t *zilog, uint64_t synced_txg); -extern int zil_suspend(zilog_t *zilog); -extern void zil_resume(zilog_t *zilog); +extern int zil_suspend(const char *osname, void **cookiep); +extern void zil_resume(void *cookie); extern void zil_add_block(zilog_t *zilog, const blkptr_t *bp); extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp); diff --git a/uts/common/fs/zfs/txg.c b/uts/common/fs/zfs/txg.c index 47603872afb1..58690e325fee 100644 --- a/uts/common/fs/zfs/txg.c +++ b/uts/common/fs/zfs/txg.c @@ -576,6 +576,8 @@ txg_wait_synced(dsl_pool_t *dp, uint64_t txg) { tx_state_t *tx = &dp->dp_tx; + ASSERT(!dsl_pool_config_held(dp)); + mutex_enter(&tx->tx_sync_lock); ASSERT(tx->tx_threads == 2); if (txg == 0) @@ -599,6 +601,8 @@ txg_wait_open(dsl_pool_t *dp, uint64_t txg) { tx_state_t *tx = &dp->dp_tx; + ASSERT(!dsl_pool_config_held(dp)); + mutex_enter(&tx->tx_sync_lock); ASSERT(tx->tx_threads == 2); if (txg == 0) @@ -664,42 +668,43 @@ txg_list_empty(txg_list_t *tl, uint64_t txg) } /* - * Add an entry to the list. - * Returns 0 if it's a new entry, 1 if it's already there. + * Add an entry to the list (unless it's already on the list). + * Returns B_TRUE if it was actually added. */ -int +boolean_t txg_list_add(txg_list_t *tl, void *p, uint64_t txg) { int t = txg & TXG_MASK; txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); - int already_on_list; + boolean_t add; mutex_enter(&tl->tl_lock); - already_on_list = tn->tn_member[t]; - if (!already_on_list) { + add = (tn->tn_member[t] == 0); + if (add) { tn->tn_member[t] = 1; tn->tn_next[t] = tl->tl_head[t]; tl->tl_head[t] = tn; } mutex_exit(&tl->tl_lock); - return (already_on_list); + return (add); } /* - * Add an entry to the end of the list (walks list to find end). - * Returns 0 if it's a new entry, 1 if it's already there. + * Add an entry to the end of the list, unless it's already on the list. + * (walks list to find end) + * Returns B_TRUE if it was actually added. */ -int +boolean_t txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg) { int t = txg & TXG_MASK; txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); - int already_on_list; + boolean_t add; mutex_enter(&tl->tl_lock); - already_on_list = tn->tn_member[t]; - if (!already_on_list) { + add = (tn->tn_member[t] == 0); + if (add) { txg_node_t **tp; for (tp = &tl->tl_head[t]; *tp != NULL; tp = &(*tp)->tn_next[t]) @@ -711,7 +716,7 @@ txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg) } mutex_exit(&tl->tl_lock); - return (already_on_list); + return (add); } /* @@ -762,13 +767,13 @@ txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg) return (NULL); } -int +boolean_t txg_list_member(txg_list_t *tl, void *p, uint64_t txg) { int t = txg & TXG_MASK; txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); - return (tn->tn_member[t]); + return (tn->tn_member[t] != 0); } /* diff --git a/uts/common/fs/zfs/zfs_ctldir.c b/uts/common/fs/zfs/zfs_ctldir.c index d902ff637c38..ef9a5611a255 100644 --- a/uts/common/fs/zfs/zfs_ctldir.c +++ b/uts/common/fs/zfs/zfs_ctldir.c @@ -72,6 +72,7 @@ #include #include #include +#include #include #include #include @@ -615,7 +616,7 @@ zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, zfsvfs_t *zfsvfs; avl_index_t where; char from[MAXNAMELEN], to[MAXNAMELEN]; - char real[MAXNAMELEN]; + char real[MAXNAMELEN], fsname[MAXNAMELEN]; int err; zfsvfs = sdvp->v_vfsp->vfs_data; @@ -634,12 +635,14 @@ zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, ZFS_EXIT(zfsvfs); + dmu_objset_name(zfsvfs->z_os, fsname); + err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); - if (!err) + if (err == 0) err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); - if (!err) + if (err == 0) err = zfs_secpolicy_rename_perms(from, to, cr); - if (err) + if (err != 0) return (err); /* @@ -659,7 +662,7 @@ zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, return (ENOENT); } - err = dmu_objset_rename(from, to, B_FALSE); + err = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE); if (err == 0) zfsctl_rename_snap(sdp, sep, tnm); @@ -699,9 +702,9 @@ zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, ZFS_EXIT(zfsvfs); err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); - if (!err) + if (err == 0) err = zfs_secpolicy_destroy_perms(snapname, cr); - if (err) + if (err != 0) return (err); mutex_enter(&sdp->sd_lock); @@ -711,10 +714,10 @@ zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, if (sep) { avl_remove(&sdp->sd_snaps, sep); err = zfsctl_unmount_snap(sep, MS_FORCE, cr); - if (err) + if (err != 0) avl_add(&sdp->sd_snaps, sep); else - err = dmu_objset_destroy(snapname, B_FALSE); + err = dsl_destroy_snapshot(snapname, B_FALSE); } else { err = ENOENT; } @@ -746,12 +749,12 @@ zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, *vpp = NULL; err = zfs_secpolicy_snapshot_perms(name, cr); - if (err) + if (err != 0) return (err); if (err == 0) { err = dmu_objset_snapshot_one(name, dirname); - if (err) + if (err != 0) return (err); err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); } @@ -831,7 +834,7 @@ zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, *vpp = sep->se_root; VN_HOLD(*vpp); err = traverse(vpp); - if (err) { + if (err != 0) { VN_RELE(*vpp); *vpp = NULL; } else if (*vpp == sep->se_root) { @@ -857,7 +860,7 @@ zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, * The requested snapshot is not currently mounted, look it up. */ err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); - if (err) { + if (err != 0) { mutex_exit(&sdp->sd_lock); ZFS_EXIT(zfsvfs); /* @@ -930,7 +933,7 @@ zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, * If we had an error, drop our hold on the vnode and * zfsctl_snapshot_inactive() will clean up. */ - if (err) { + if (err != 0) { VN_RELE(*vpp); *vpp = NULL; } @@ -982,8 +985,10 @@ zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, ZFS_ENTER(zfsvfs); cookie = *offp; + dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, &cookie, &case_conflict); + dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); if (error) { ZFS_EXIT(zfsvfs); if (error == ENOENT) { diff --git a/uts/common/fs/zfs/zfs_ioctl.c b/uts/common/fs/zfs/zfs_ioctl.c index 38adc1940ff8..caad34c5a70f 100644 --- a/uts/common/fs/zfs/zfs_ioctl.c +++ b/uts/common/fs/zfs/zfs_ioctl.c @@ -156,6 +156,7 @@ #include #include #include +#include #include #include #include @@ -173,6 +174,9 @@ #include #include #include +#include +#include +#include #include #include "zfs_namecheck.h" @@ -237,11 +241,7 @@ static int zfs_fill_zplprops_root(uint64_t, nvlist_t *, nvlist_t *, int zfs_set_prop_nvlist(const char *, zprop_source_t, nvlist_t *, nvlist_t *); static int get_nvlist(uint64_t nvl, uint64_t size, int iflag, nvlist_t **nvp); -static int zfs_prop_activate_feature(dsl_pool_t *dp, zfeature_info_t *feature); -static int zfs_prop_activate_feature_check(void *arg1, void *arg2, - dmu_tx_t *tx); -static void zfs_prop_activate_feature_sync(void *arg1, void *arg2, - dmu_tx_t *tx); +static int zfs_prop_activate_feature(spa_t *spa, zfeature_info_t *feature); /* _NOTE(PRINTFLIKE(4)) - this is printf-like, but lint is too whiney */ void @@ -461,37 +461,12 @@ zfs_dozonecheck_ds(const char *dataset, dsl_dataset_t *ds, cred_t *cr) { uint64_t zoned; - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); - if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL)) { - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); + if (dsl_prop_get_int_ds(ds, "zoned", &zoned)) return (ENOENT); - } - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); return (zfs_dozonecheck_impl(dataset, zoned, cr)); } -static int -zfs_secpolicy_write_perms(const char *name, const char *perm, cred_t *cr) -{ - int error; - dsl_dataset_t *ds; - - error = dsl_dataset_hold(name, FTAG, &ds); - if (error != 0) - return (error); - - error = zfs_dozonecheck_ds(name, ds, cr); - if (error == 0) { - error = secpolicy_zfs(cr); - if (error) - error = dsl_deleg_access_impl(ds, perm, cr); - } - - dsl_dataset_rele(ds, FTAG); - return (error); -} - static int zfs_secpolicy_write_perms_ds(const char *name, dsl_dataset_t *ds, const char *perm, cred_t *cr) @@ -501,12 +476,36 @@ zfs_secpolicy_write_perms_ds(const char *name, dsl_dataset_t *ds, error = zfs_dozonecheck_ds(name, ds, cr); if (error == 0) { error = secpolicy_zfs(cr); - if (error) + if (error != 0) error = dsl_deleg_access_impl(ds, perm, cr); } return (error); } +static int +zfs_secpolicy_write_perms(const char *name, const char *perm, cred_t *cr) +{ + int error; + dsl_dataset_t *ds; + dsl_pool_t *dp; + + error = dsl_pool_hold(name, FTAG, &dp); + if (error != 0) + return (error); + + error = dsl_dataset_hold(dp, name, FTAG, &ds); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + + error = zfs_secpolicy_write_perms_ds(name, ds, perm, cr); + + dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); + return (error); +} + /* * Policy for setting the security label property. * @@ -525,7 +524,7 @@ zfs_set_slabel_policy(const char *name, char *strval, cred_t *cr) /* First get the existing dataset label. */ error = dsl_prop_get(name, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1, sizeof (ds_hexsl), &ds_hexsl, NULL); - if (error) + if (error != 0) return (EPERM); if (strcasecmp(strval, ZFS_MLSLABEL_DEFAULT) == 0) @@ -575,7 +574,7 @@ zfs_set_slabel_policy(const char *name, char *strval, cred_t *cr) */ error = dmu_objset_own(name, DMU_OST_ZFS, B_TRUE, setsl_tag, &os); - if (error) + if (error != 0) return (EPERM); dmu_objset_disown(os, setsl_tag); @@ -663,7 +662,7 @@ zfs_secpolicy_set_fsacl(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) int error; error = zfs_dozonecheck(zc->zc_name, cr); - if (error) + if (error != 0) return (error); /* @@ -685,7 +684,6 @@ zfs_secpolicy_rollback(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) static int zfs_secpolicy_send(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) { - spa_t *spa; dsl_pool_t *dp; dsl_dataset_t *ds; char *cp; @@ -698,23 +696,22 @@ zfs_secpolicy_send(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) cp = strchr(zc->zc_name, '@'); if (cp == NULL) return (EINVAL); - error = spa_open(zc->zc_name, &spa, FTAG); - if (error) + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); + if (error != 0) return (error); - dp = spa_get_dsl(spa); - rw_enter(&dp->dp_config_rwlock, RW_READER); error = dsl_dataset_hold_obj(dp, zc->zc_sendobj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - spa_close(spa, FTAG); - if (error) + if (error != 0) { + dsl_pool_rele(dp, FTAG); return (error); + } dsl_dataset_name(ds, zc->zc_name); error = zfs_secpolicy_write_perms_ds(zc->zc_name, ds, ZFS_DELEG_PERM_SEND, cr); dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } @@ -835,12 +832,21 @@ zfs_secpolicy_destroy_snaps(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) return (EINVAL); for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; pair = nextpair) { + dsl_pool_t *dp; dsl_dataset_t *ds; + error = dsl_pool_hold(nvpair_name(pair), FTAG, &dp); + if (error != 0) + break; nextpair = nvlist_next_nvpair(snaps, pair); - error = dsl_dataset_hold(nvpair_name(pair), FTAG, &ds); - if (error == 0) { + error = dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds); + if (error == 0) dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); + + if (error == 0) { + error = zfs_secpolicy_destroy_perms(nvpair_name(pair), + cr); } else if (error == ENOENT) { /* * Ignore any snapshots that don't exist (we consider @@ -852,11 +858,7 @@ zfs_secpolicy_destroy_snaps(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) */ fnvlist_remove_nvpair(snaps, pair); error = 0; - continue; - } else { - break; } - error = zfs_secpolicy_destroy_perms(nvpair_name(pair), cr); if (error != 0) break; } @@ -904,41 +906,47 @@ zfs_secpolicy_rename(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) static int zfs_secpolicy_promote(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) { - char parentname[MAXNAMELEN]; - objset_t *clone; + dsl_pool_t *dp; + dsl_dataset_t *clone; int error; error = zfs_secpolicy_write_perms(zc->zc_name, ZFS_DELEG_PERM_PROMOTE, cr); - if (error) + if (error != 0) return (error); - error = dmu_objset_hold(zc->zc_name, FTAG, &clone); + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); + if (error != 0) + return (error); + + error = dsl_dataset_hold(dp, zc->zc_name, FTAG, &clone); if (error == 0) { - dsl_dataset_t *pclone = NULL; + char parentname[MAXNAMELEN]; + dsl_dataset_t *origin = NULL; dsl_dir_t *dd; - dd = clone->os_dsl_dataset->ds_dir; + dd = clone->ds_dir; - rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); error = dsl_dataset_hold_obj(dd->dd_pool, - dd->dd_phys->dd_origin_obj, FTAG, &pclone); - rw_exit(&dd->dd_pool->dp_config_rwlock); - if (error) { - dmu_objset_rele(clone, FTAG); + dd->dd_phys->dd_origin_obj, FTAG, &origin); + if (error != 0) { + dsl_dataset_rele(clone, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } - error = zfs_secpolicy_write_perms(zc->zc_name, + error = zfs_secpolicy_write_perms_ds(zc->zc_name, clone, ZFS_DELEG_PERM_MOUNT, cr); - dsl_dataset_name(pclone, parentname); - dmu_objset_rele(clone, FTAG); - dsl_dataset_rele(pclone, FTAG); - if (error == 0) - error = zfs_secpolicy_write_perms(parentname, + dsl_dataset_name(origin, parentname); + if (error == 0) { + error = zfs_secpolicy_write_perms_ds(parentname, origin, ZFS_DELEG_PERM_PROMOTE, cr); + } + dsl_dataset_rele(clone, FTAG); + dsl_dataset_rele(origin, FTAG); } + dsl_pool_rele(dp, FTAG); return (error); } @@ -1147,16 +1155,47 @@ zfs_secpolicy_userspace_upgrade(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) static int zfs_secpolicy_hold(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) { - return (zfs_secpolicy_write_perms(zc->zc_name, - ZFS_DELEG_PERM_HOLD, cr)); + nvpair_t *pair; + nvlist_t *holds; + int error; + + error = nvlist_lookup_nvlist(innvl, "holds", &holds); + if (error != 0) + return (EINVAL); + + for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(holds, pair)) { + char fsname[MAXNAMELEN]; + error = dmu_fsname(nvpair_name(pair), fsname); + if (error != 0) + return (error); + error = zfs_secpolicy_write_perms(fsname, + ZFS_DELEG_PERM_HOLD, cr); + if (error != 0) + return (error); + } + return (0); } /* ARGSUSED */ static int zfs_secpolicy_release(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) { - return (zfs_secpolicy_write_perms(zc->zc_name, - ZFS_DELEG_PERM_RELEASE, cr)); + nvpair_t *pair; + int error; + + for (pair = nvlist_next_nvpair(innvl, NULL); pair != NULL; + pair = nvlist_next_nvpair(innvl, pair)) { + char fsname[MAXNAMELEN]; + error = dmu_fsname(nvpair_name(pair), fsname); + if (error != 0) + return (error); + error = zfs_secpolicy_write_perms(fsname, + ZFS_DELEG_PERM_RELEASE, cr); + if (error != 0) + return (error); + } + return (0); } /* @@ -1177,11 +1216,11 @@ zfs_secpolicy_tmp_snapshot(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) return (0); error = zfs_secpolicy_snapshot_perms(zc->zc_name, cr); - if (!error) + if (error == 0) error = zfs_secpolicy_hold(zc, innvl, cr); - if (!error) + if (error == 0) error = zfs_secpolicy_release(zc, innvl, cr); - if (!error) + if (error == 0) error = zfs_secpolicy_destroy(zc, innvl, cr); return (error); } @@ -1291,7 +1330,7 @@ getzfsvfs(const char *dsname, zfsvfs_t **zfvp) int error; error = dmu_objset_hold(dsname, FTAG, &os); - if (error) + if (error != 0) return (error); if (dmu_objset_type(os) != DMU_OST_ZFS) { dmu_objset_rele(os, FTAG); @@ -1394,7 +1433,7 @@ zfs_ioc_pool_create(zfs_cmd_t *zc) VERIFY(nvlist_alloc(&zplprops, NV_UNIQUE_NAME, KM_SLEEP) == 0); error = zfs_fill_zplprops_root(version, rootprops, zplprops, NULL); - if (error) + if (error != 0) goto pool_props_bad; } @@ -1667,12 +1706,7 @@ zfs_ioc_pool_reguid(zfs_cmd_t *zc) static int zfs_ioc_dsobj_to_dsname(zfs_cmd_t *zc) { - int error; - - if (error = dsl_dsobj_to_dsname(zc->zc_name, zc->zc_obj, zc->zc_value)) - return (error); - - return (0); + return (dsl_dsobj_to_dsname(zc->zc_name, zc->zc_obj, zc->zc_value)); } /* @@ -1988,15 +2022,14 @@ zfs_ioc_objset_stats_impl(zfs_cmd_t *zc, objset_t *os) static int zfs_ioc_objset_stats(zfs_cmd_t *zc) { - objset_t *os = NULL; + objset_t *os; int error; - if (error = dmu_objset_hold(zc->zc_name, FTAG, &os)) - return (error); - - error = zfs_ioc_objset_stats_impl(zc, os); - - dmu_objset_rele(os, FTAG); + error = dmu_objset_hold(zc->zc_name, FTAG, &os); + if (error == 0) { + error = zfs_ioc_objset_stats_impl(zc, os); + dmu_objset_rele(os, FTAG); + } return (error); } @@ -2017,30 +2050,23 @@ zfs_ioc_objset_stats(zfs_cmd_t *zc) static int zfs_ioc_objset_recvd_props(zfs_cmd_t *zc) { - objset_t *os = NULL; - int error; + int error = 0; nvlist_t *nv; - if (error = dmu_objset_hold(zc->zc_name, FTAG, &os)) - return (error); - /* * Without this check, we would return local property values if the * caller has not already received properties on or after * SPA_VERSION_RECVD_PROPS. */ - if (!dsl_prop_get_hasrecvd(os)) { - dmu_objset_rele(os, FTAG); + if (!dsl_prop_get_hasrecvd(zc->zc_name)) return (ENOTSUP); - } if (zc->zc_nvlist_dst != 0 && - (error = dsl_prop_get_received(os, &nv)) == 0) { + (error = dsl_prop_get_received(zc->zc_name, &nv)) == 0) { error = put_nvlist(zc, nv); nvlist_free(nv); } - dmu_objset_rele(os, FTAG); return (error); } @@ -2155,20 +2181,6 @@ zfs_ioc_dataset_list_next(zfs_cmd_t *zc) (void) strlcat(zc->zc_name, "/", sizeof (zc->zc_name)); p = zc->zc_name + strlen(zc->zc_name); - /* - * Pre-fetch the datasets. dmu_objset_prefetch() always returns 0 - * but is not declared void because its called by dmu_objset_find(). - */ - if (zc->zc_cookie == 0) { - uint64_t cookie = 0; - int len = sizeof (zc->zc_name) - (p - zc->zc_name); - - while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { - if (!dataset_name_hidden(zc->zc_name)) - (void) dmu_objset_prefetch(zc->zc_name, NULL); - } - } - do { error = dmu_dir_list_next(os, sizeof (zc->zc_name) - (p - zc->zc_name), p, @@ -2211,14 +2223,10 @@ zfs_ioc_snapshot_list_next(zfs_cmd_t *zc) objset_t *os; int error; -top: - if (zc->zc_cookie == 0) - (void) dmu_objset_find(zc->zc_name, dmu_objset_prefetch, - NULL, DS_FIND_SNAPSHOTS); - error = dmu_objset_hold(zc->zc_name, FTAG, &os); - if (error) + if (error != 0) { return (error == ENOENT ? ESRCH : error); + } /* * A dataset name of maximum length cannot have any snapshots, @@ -2238,24 +2246,8 @@ zfs_ioc_snapshot_list_next(zfs_cmd_t *zc) dsl_dataset_t *ds; dsl_pool_t *dp = os->os_dsl_dataset->ds_dir->dd_pool; - /* - * Since we probably don't have a hold on this snapshot, - * it's possible that the objsetid could have been destroyed - * and reused for a new objset. It's OK if this happens during - * a zfs send operation, since the new createtxg will be - * beyond the range we're interested in. - */ - rw_enter(&dp->dp_config_rwlock, RW_READER); error = dsl_dataset_hold_obj(dp, zc->zc_obj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - if (error) { - if (error == ENOENT) { - /* Racing with destroy, get the next one. */ - *strchr(zc->zc_name, '@') = '\0'; - dmu_objset_rele(os, FTAG); - goto top; - } - } else { + if (error == 0) { objset_t *ossnap; error = dmu_objset_from_ds(ds, &ossnap); @@ -2269,7 +2261,7 @@ zfs_ioc_snapshot_list_next(zfs_cmd_t *zc) dmu_objset_rele(os, FTAG); /* if we failed, undo the @ that we tacked on to zc_name */ - if (error) + if (error != 0) *strchr(zc->zc_name, '@') = '\0'; return (error); } @@ -2359,13 +2351,13 @@ zfs_prop_set_special(const char *dsname, zprop_source_t source, err = dsl_dir_set_quota(dsname, source, intval); break; case ZFS_PROP_REFQUOTA: - err = dsl_dataset_set_quota(dsname, source, intval); + err = dsl_dataset_set_refquota(dsname, source, intval); break; case ZFS_PROP_RESERVATION: err = dsl_dir_set_reservation(dsname, source, intval); break; case ZFS_PROP_REFRESERVATION: - err = dsl_dataset_set_reservation(dsname, source, intval); + err = dsl_dataset_set_refreservation(dsname, source, intval); break; case ZFS_PROP_VOLSIZE: err = zvol_set_volsize(dsname, intval); @@ -2396,19 +2388,16 @@ zfs_prop_set_special(const char *dsname, zprop_source_t source, zfeature_info_t *feature = &spa_feature_table[SPA_FEATURE_LZ4_COMPRESS]; spa_t *spa; - dsl_pool_t *dp; if ((err = spa_open(dsname, &spa, FTAG)) != 0) return (err); - dp = spa->spa_dsl_pool; - /* * Setting the LZ4 compression algorithm activates * the feature. */ if (!spa_feature_is_active(spa, feature)) { - if ((err = zfs_prop_activate_feature(dp, + if ((err = zfs_prop_activate_feature(spa, feature)) != 0) { spa_close(spa, FTAG); return (err); @@ -2567,12 +2556,12 @@ zfs_set_prop_nvlist(const char *dsname, zprop_source_t source, nvlist_t *nvl, if (nvpair_type(propval) == DATA_TYPE_STRING) { strval = fnvpair_value_string(propval); - err = dsl_prop_set(dsname, propname, source, 1, - strlen(strval) + 1, strval); + err = dsl_prop_set_string(dsname, propname, + source, strval); } else { intval = fnvpair_value_uint64(propval); - err = dsl_prop_set(dsname, propname, source, 8, - 1, &intval); + err = dsl_prop_set_int(dsname, propname, source, + intval); } if (err != 0) { @@ -2638,7 +2627,7 @@ props_skip(nvlist_t *props, nvlist_t *skipped, nvlist_t **newprops) } static int -clear_received_props(objset_t *os, const char *fs, nvlist_t *props, +clear_received_props(const char *dsname, nvlist_t *props, nvlist_t *skipped) { int err = 0; @@ -2650,8 +2639,8 @@ clear_received_props(objset_t *os, const char *fs, nvlist_t *props, * properties at least once on or after SPA_VERSION_RECVD_PROPS. */ zprop_source_t flags = (ZPROP_SRC_NONE | - (dsl_prop_get_hasrecvd(os) ? ZPROP_SRC_RECEIVED : 0)); - err = zfs_set_prop_nvlist(fs, flags, cleared_props, NULL); + (dsl_prop_get_hasrecvd(dsname) ? ZPROP_SRC_RECEIVED : 0)); + err = zfs_set_prop_nvlist(dsname, flags, cleared_props, NULL); } nvlist_free(cleared_props); return (err); @@ -2683,22 +2672,19 @@ zfs_ioc_set_prop(zfs_cmd_t *zc) if (received) { nvlist_t *origprops; - objset_t *os; - if (dmu_objset_hold(zc->zc_name, FTAG, &os) == 0) { - if (dsl_prop_get_received(os, &origprops) == 0) { - (void) clear_received_props(os, - zc->zc_name, origprops, nvl); - nvlist_free(origprops); - } - - dsl_prop_set_hasrecvd(os); - dmu_objset_rele(os, FTAG); + if (dsl_prop_get_received(zc->zc_name, &origprops) == 0) { + (void) clear_received_props(zc->zc_name, + origprops, nvl); + nvlist_free(origprops); } + + error = dsl_prop_set_hasrecvd(zc->zc_name); } errors = fnvlist_alloc(); - error = zfs_set_prop_nvlist(zc->zc_name, source, nvl, errors); + if (error == 0) + error = zfs_set_prop_nvlist(zc->zc_name, source, nvl, errors); if (zc->zc_nvlist_dst != NULL && errors != NULL) { (void) put_nvlist(zc, errors); @@ -2781,7 +2767,7 @@ zfs_ioc_inherit_prop(zfs_cmd_t *zc) } /* property name has been validated by zfs_secpolicy_inherit_prop() */ - return (dsl_prop_set(zc->zc_name, zc->zc_value, source, 0, 0, NULL)); + return (dsl_prop_inherit(zc->zc_name, zc->zc_value, source)); } static int @@ -2893,7 +2879,7 @@ zfs_ioc_set_fsacl(zfs_cmd_t *zc) */ error = secpolicy_zfs(CRED()); - if (error) { + if (error != 0) { if (zc->zc_perm_action == B_FALSE) { error = dsl_deleg_can_allow(zc->zc_name, fsaclnv, CRED()); @@ -3221,7 +3207,7 @@ zfs_ioc_create(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) error = zfs_set_prop_nvlist(fsname, ZPROP_SRC_LOCAL, nvprops, outnvl); if (error != 0) - (void) dmu_objset_destroy(fsname, B_FALSE); + (void) dsl_destroy_head(fsname); } return (error); } @@ -3240,7 +3226,6 @@ zfs_ioc_clone(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) int error = 0; nvlist_t *nvprops = NULL; char *origin_name; - dsl_dataset_t *origin; if (nvlist_lookup_string(innvl, "origin", &origin_name) != 0) return (EINVAL); @@ -3252,14 +3237,8 @@ zfs_ioc_clone(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) if (dataset_namecheck(origin_name, NULL, NULL) != 0) return (EINVAL); - - error = dsl_dataset_hold(origin_name, FTAG, &origin); - if (error) - return (error); - - error = dmu_objset_clone(fsname, origin, 0); - dsl_dataset_rele(origin, FTAG); - if (error) + error = dmu_objset_clone(fsname, origin_name); + if (error != 0) return (error); /* @@ -3269,7 +3248,7 @@ zfs_ioc_clone(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) error = zfs_set_prop_nvlist(fsname, ZPROP_SRC_LOCAL, nvprops, outnvl); if (error != 0) - (void) dmu_objset_destroy(fsname, B_FALSE); + (void) dsl_destroy_head(fsname); } return (error); } @@ -3281,7 +3260,6 @@ zfs_ioc_clone(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) * } * * outnvl: snapshot -> error code (int32) - * */ static int zfs_ioc_snapshot(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl) @@ -3331,7 +3309,7 @@ zfs_ioc_snapshot(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl) } } - error = dmu_objset_snapshot(snaps, props, outnvl); + error = dsl_dataset_snapshot(snaps, props, outnvl); return (error); } @@ -3376,30 +3354,73 @@ zfs_ioc_log_history(const char *unused, nvlist_t *innvl, nvlist_t *outnvl) return (error); } -/* ARGSUSED */ -int -zfs_unmount_snap(const char *name, void *arg) +/* + * The dp_config_rwlock must not be held when calling this, because the + * unmount may need to write out data. + * + * This function is best-effort. Callers must deal gracefully if it + * remains mounted (or is remounted after this call). + */ +void +zfs_unmount_snap(const char *snapname) { vfs_t *vfsp; - int err; + zfsvfs_t *zfsvfs; - if (strchr(name, '@') == NULL) - return (0); + if (strchr(snapname, '@') == NULL) + return; - vfsp = zfs_get_vfs(name); + vfsp = zfs_get_vfs(snapname); if (vfsp == NULL) - return (0); + return; - if ((err = vn_vfswlock(vfsp->vfs_vnodecovered)) != 0) { + zfsvfs = vfsp->vfs_data; + ASSERT(!dsl_pool_config_held(dmu_objset_pool(zfsvfs->z_os))); + + if (vn_vfswlock(vfsp->vfs_vnodecovered) != 0) { VFS_RELE(vfsp); - return (err); + return; } VFS_RELE(vfsp); /* * Always force the unmount for snapshots. */ - return (dounmount(vfsp, MS_FORCE, kcred)); + (void) dounmount(vfsp, MS_FORCE, kcred); +} + +/* ARGSUSED */ +static int +zfs_unmount_snap_cb(const char *snapname, void *arg) +{ + zfs_unmount_snap(snapname); + return (0); +} + +/* + * When a clone is destroyed, its origin may also need to be destroyed, + * in which case it must be unmounted. This routine will do that unmount + * if necessary. + */ +void +zfs_destroy_unmount_origin(const char *fsname) +{ + int error; + objset_t *os; + dsl_dataset_t *ds; + + error = dmu_objset_hold(fsname, FTAG, &os); + if (error != 0) + return; + ds = dmu_objset_ds(os); + if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev)) { + char originname[MAXNAMELEN]; + dsl_dataset_name(ds->ds_prev, originname); + dmu_objset_rele(os, FTAG); + zfs_unmount_snap(originname); + } else { + dmu_objset_rele(os, FTAG); + } } /* @@ -3435,14 +3456,10 @@ zfs_ioc_destroy_snaps(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl) (name[poollen] != '/' && name[poollen] != '@')) return (EXDEV); - /* - * Ignore failures to unmount; dmu_snapshots_destroy_nvl() - * will deal with this gracefully (by filling in outnvl). - */ - (void) zfs_unmount_snap(name, NULL); + zfs_unmount_snap(name); } - return (dmu_snapshots_destroy_nvl(snaps, defer, outnvl)); + return (dsl_destroy_snapshots_nvl(snaps, defer, outnvl)); } /* @@ -3457,13 +3474,13 @@ static int zfs_ioc_destroy(zfs_cmd_t *zc) { int err; - if (strchr(zc->zc_name, '@') && zc->zc_objset_type == DMU_OST_ZFS) { - err = zfs_unmount_snap(zc->zc_name, NULL); - if (err) - return (err); - } + if (strchr(zc->zc_name, '@') && zc->zc_objset_type == DMU_OST_ZFS) + zfs_unmount_snap(zc->zc_name); - err = dmu_objset_destroy(zc->zc_name, zc->zc_defer_destroy); + if (strchr(zc->zc_name, '@')) + err = dsl_destroy_snapshot(zc->zc_name, zc->zc_defer_destroy); + else + err = dsl_destroy_head(zc->zc_name); if (zc->zc_objset_type == DMU_OST_ZVOL && err == 0) (void) zvol_remove_minor(zc->zc_name); return (err); @@ -3478,81 +3495,36 @@ zfs_ioc_destroy(zfs_cmd_t *zc) static int zfs_ioc_rollback(zfs_cmd_t *zc) { - dsl_dataset_t *ds, *clone; - int error; zfsvfs_t *zfsvfs; - char *clone_name; + int error; - error = dsl_dataset_hold(zc->zc_name, FTAG, &ds); - if (error) - return (error); - - /* must not be a snapshot */ - if (dsl_dataset_is_snapshot(ds)) { - dsl_dataset_rele(ds, FTAG); - return (EINVAL); - } - - /* must have a most recent snapshot */ - if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { - dsl_dataset_rele(ds, FTAG); - return (EINVAL); - } - - /* - * Create clone of most recent snapshot. - */ - clone_name = kmem_asprintf("%s/%%rollback", zc->zc_name); - error = dmu_objset_clone(clone_name, ds->ds_prev, DS_FLAG_INCONSISTENT); - if (error) - goto out; - - error = dsl_dataset_own(clone_name, B_TRUE, FTAG, &clone); - if (error) - goto out; - - /* - * Do clone swap. - */ if (getzfsvfs(zc->zc_name, &zfsvfs) == 0) { error = zfs_suspend_fs(zfsvfs); if (error == 0) { int resume_err; - if (dsl_dataset_tryown(ds, B_FALSE, FTAG)) { - error = dsl_dataset_clone_swap(clone, ds, - B_TRUE); - dsl_dataset_disown(ds, FTAG); - ds = NULL; - } else { - error = EBUSY; - } + error = dsl_dataset_rollback(zc->zc_name); resume_err = zfs_resume_fs(zfsvfs, zc->zc_name); error = error ? error : resume_err; } VFS_RELE(zfsvfs->z_vfs); } else { - if (dsl_dataset_tryown(ds, B_FALSE, FTAG)) { - error = dsl_dataset_clone_swap(clone, ds, B_TRUE); - dsl_dataset_disown(ds, FTAG); - ds = NULL; - } else { - error = EBUSY; - } + error = dsl_dataset_rollback(zc->zc_name); } - - /* - * Destroy clone (which also closes it). - */ - (void) dsl_dataset_destroy(clone, FTAG, B_FALSE); - -out: - strfree(clone_name); - if (ds) - dsl_dataset_rele(ds, FTAG); return (error); } +static int +recursive_unmount(const char *fsname, void *arg) +{ + const char *snapname = arg; + char fullname[MAXNAMELEN]; + + (void) snprintf(fullname, sizeof (fullname), "%s@%s", fsname, snapname); + zfs_unmount_snap(fullname); + return (0); +} + /* * inputs: * zc_name old name of dataset @@ -3565,26 +3537,33 @@ static int zfs_ioc_rename(zfs_cmd_t *zc) { boolean_t recursive = zc->zc_cookie & 1; + char *at; zc->zc_value[sizeof (zc->zc_value) - 1] = '\0'; if (dataset_namecheck(zc->zc_value, NULL, NULL) != 0 || strchr(zc->zc_value, '%')) return (EINVAL); - /* - * Unmount snapshot unless we're doing a recursive rename, - * in which case the dataset code figures out which snapshots - * to unmount. - */ - if (!recursive && strchr(zc->zc_name, '@') != NULL && - zc->zc_objset_type == DMU_OST_ZFS) { - int err = zfs_unmount_snap(zc->zc_name, NULL); - if (err) - return (err); + at = strchr(zc->zc_name, '@'); + if (at != NULL) { + /* snaps must be in same fs */ + if (strncmp(zc->zc_name, zc->zc_value, at - zc->zc_name + 1)) + return (EXDEV); + *at = '\0'; + if (zc->zc_objset_type == DMU_OST_ZFS) { + int error = dmu_objset_find(zc->zc_name, + recursive_unmount, at + 1, + recursive ? DS_FIND_CHILDREN : 0); + if (error != 0) + return (error); + } + return (dsl_dataset_rename_snapshot(zc->zc_name, + at + 1, strchr(zc->zc_value, '@') + 1, recursive)); + } else { + if (zc->zc_objset_type == DMU_OST_ZVOL) + (void) zvol_remove_minor(zc->zc_name); + return (dsl_dir_rename(zc->zc_name, zc->zc_value)); } - if (zc->zc_objset_type == DMU_OST_ZVOL) - (void) zvol_remove_minor(zc->zc_name); - return (dmu_objset_rename(zc->zc_name, zc->zc_value, recursive)); } static int @@ -3727,36 +3706,15 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr) return (zfs_secpolicy_setprop(dsname, prop, pair, CRED())); } -/* - * Activates a feature on a pool in response to a property setting. This - * creates a new sync task which modifies the pool to reflect the feature - * as being active. - */ -static int -zfs_prop_activate_feature(dsl_pool_t *dp, zfeature_info_t *feature) -{ - int err; - - /* EBUSY here indicates that the feature is already active */ - err = dsl_sync_task_do(dp, zfs_prop_activate_feature_check, - zfs_prop_activate_feature_sync, dp->dp_spa, feature, 2); - - if (err != 0 && err != EBUSY) - return (err); - else - return (0); -} - /* * Checks for a race condition to make sure we don't increment a feature flag * multiple times. */ -/*ARGSUSED*/ static int -zfs_prop_activate_feature_check(void *arg1, void *arg2, dmu_tx_t *tx) +zfs_prop_activate_feature_check(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - zfeature_info_t *feature = arg2; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + zfeature_info_t *feature = arg; if (!spa_feature_is_active(spa, feature)) return (0); @@ -3769,14 +3727,35 @@ zfs_prop_activate_feature_check(void *arg1, void *arg2, dmu_tx_t *tx) * zfs_prop_activate_feature. */ static void -zfs_prop_activate_feature_sync(void *arg1, void *arg2, dmu_tx_t *tx) +zfs_prop_activate_feature_sync(void *arg, dmu_tx_t *tx) { - spa_t *spa = arg1; - zfeature_info_t *feature = arg2; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + zfeature_info_t *feature = arg; spa_feature_incr(spa, feature, tx); } +/* + * Activates a feature on a pool in response to a property setting. This + * creates a new sync task which modifies the pool to reflect the feature + * as being active. + */ +static int +zfs_prop_activate_feature(spa_t *spa, zfeature_info_t *feature) +{ + int err; + + /* EBUSY here indicates that the feature is already active */ + err = dsl_sync_task(spa_name(spa), + zfs_prop_activate_feature_check, zfs_prop_activate_feature_sync, + feature, 2); + + if (err != 0 && err != EBUSY) + return (err); + else + return (0); +} + /* * Removes properties from the given props list that fail permission checks * needed to clear them and to restore them in case of a receive error. For each @@ -3931,7 +3910,6 @@ static int zfs_ioc_recv(zfs_cmd_t *zc) { file_t *fp; - objset_t *os; dmu_recv_cookie_t drc; boolean_t force = (boolean_t)zc->zc_guid; int fd; @@ -3941,7 +3919,7 @@ zfs_ioc_recv(zfs_cmd_t *zc) offset_t off; nvlist_t *props = NULL; /* sent properties */ nvlist_t *origprops = NULL; /* existing properties */ - objset_t *origin = NULL; + char *origin = NULL; char *tosnap; char tofs[ZFS_MAXNAMELEN]; boolean_t first_recvd_props = B_FALSE; @@ -3969,18 +3947,31 @@ zfs_ioc_recv(zfs_cmd_t *zc) VERIFY(nvlist_alloc(&errors, NV_UNIQUE_NAME, KM_SLEEP) == 0); - if (props && dmu_objset_hold(tofs, FTAG, &os) == 0) { - if ((spa_version(os->os_spa) >= SPA_VERSION_RECVD_PROPS) && - !dsl_prop_get_hasrecvd(os)) { + if (zc->zc_string[0]) + origin = zc->zc_string; + + error = dmu_recv_begin(tofs, tosnap, + &zc->zc_begin_record, force, origin, &drc); + if (error != 0) + goto out; + + /* + * Set properties before we receive the stream so that they are applied + * to the new data. Note that we must call dmu_recv_stream() if + * dmu_recv_begin() succeeds. + */ + if (props != NULL && !drc.drc_newfs) { + if (spa_version(dsl_dataset_get_spa(drc.drc_ds)) >= + SPA_VERSION_RECVD_PROPS && + !dsl_prop_get_hasrecvd(tofs)) first_recvd_props = B_TRUE; - } /* * If new received properties are supplied, they are to * completely replace the existing received properties, so stash * away the existing ones. */ - if (dsl_prop_get_received(os, &origprops) == 0) { + if (dsl_prop_get_received(tofs, &origprops) == 0) { nvlist_t *errlist = NULL; /* * Don't bother writing a property if its value won't @@ -3992,53 +3983,25 @@ zfs_ioc_recv(zfs_cmd_t *zc) */ if (!first_recvd_props) props_reduce(props, origprops); - if (zfs_check_clearable(tofs, origprops, - &errlist) != 0) + if (zfs_check_clearable(tofs, origprops, &errlist) != 0) (void) nvlist_merge(errors, errlist, 0); nvlist_free(errlist); - } - dmu_objset_rele(os, FTAG); - } - - if (zc->zc_string[0]) { - error = dmu_objset_hold(zc->zc_string, FTAG, &origin); - if (error) - goto out; - } - - error = dmu_recv_begin(tofs, tosnap, zc->zc_top_ds, - &zc->zc_begin_record, force, origin, &drc); - if (origin) - dmu_objset_rele(origin, FTAG); - if (error) - goto out; - - /* - * Set properties before we receive the stream so that they are applied - * to the new data. Note that we must call dmu_recv_stream() if - * dmu_recv_begin() succeeds. - */ - if (props) { - if (dmu_objset_from_ds(drc.drc_logical_ds, &os) == 0) { - if (drc.drc_newfs) { - if (spa_version(os->os_spa) >= - SPA_VERSION_RECVD_PROPS) - first_recvd_props = B_TRUE; - } else if (origprops != NULL) { - if (clear_received_props(os, tofs, origprops, - first_recvd_props ? NULL : props) != 0) - zc->zc_obj |= ZPROP_ERR_NOCLEAR; - } else { + if (clear_received_props(tofs, origprops, + first_recvd_props ? NULL : props) != 0) zc->zc_obj |= ZPROP_ERR_NOCLEAR; - } - dsl_prop_set_hasrecvd(os); - } else if (!drc.drc_newfs) { + } else { zc->zc_obj |= ZPROP_ERR_NOCLEAR; } + } - (void) zfs_set_prop_nvlist(tofs, ZPROP_SRC_RECEIVED, - props, errors); + if (props != NULL) { + props_error = dsl_prop_set_hasrecvd(tofs); + + if (props_error == 0) { + (void) zfs_set_prop_nvlist(tofs, ZPROP_SRC_RECEIVED, + props, errors); + } } if (zc->zc_nvlist_dst_size != 0 && @@ -4090,22 +4053,16 @@ zfs_ioc_recv(zfs_cmd_t *zc) /* * On error, restore the original props. */ - if (error && props) { - if (dmu_objset_hold(tofs, FTAG, &os) == 0) { - if (clear_received_props(os, tofs, props, NULL) != 0) { - /* - * We failed to clear the received properties. - * Since we may have left a $recvd value on the - * system, we can't clear the $hasrecvd flag. - */ - zc->zc_obj |= ZPROP_ERR_NORESTORE; - } else if (first_recvd_props) { - dsl_prop_unset_hasrecvd(os); - } - dmu_objset_rele(os, FTAG); - } else if (!drc.drc_newfs) { - /* We failed to clear the received properties. */ + if (error != 0 && props != NULL && !drc.drc_newfs) { + if (clear_received_props(tofs, props, NULL) != 0) { + /* + * We failed to clear the received properties. + * Since we may have left a $recvd value on the + * system, we can't clear the $hasrecvd flag. + */ zc->zc_obj |= ZPROP_ERR_NORESTORE; + } else if (first_recvd_props) { + dsl_prop_unset_hasrecvd(tofs); } if (origprops == NULL && !drc.drc_newfs) { @@ -4157,100 +4114,75 @@ zfs_ioc_recv(zfs_cmd_t *zc) static int zfs_ioc_send(zfs_cmd_t *zc) { - objset_t *fromsnap = NULL; - objset_t *tosnap; int error; offset_t off; - dsl_dataset_t *ds; - dsl_dataset_t *dsfrom = NULL; - spa_t *spa; - dsl_pool_t *dp; boolean_t estimate = (zc->zc_guid != 0); - error = spa_open(zc->zc_name, &spa, FTAG); - if (error) - return (error); + if (zc->zc_obj != 0) { + dsl_pool_t *dp; + dsl_dataset_t *tosnap; - dp = spa_get_dsl(spa); - rw_enter(&dp->dp_config_rwlock, RW_READER); - error = dsl_dataset_hold_obj(dp, zc->zc_sendobj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - spa_close(spa, FTAG); - if (error) - return (error); + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); + if (error != 0) + return (error); - error = dmu_objset_from_ds(ds, &tosnap); - if (error) { - dsl_dataset_rele(ds, FTAG); - return (error); - } - - if (zc->zc_fromobj != 0) { - rw_enter(&dp->dp_config_rwlock, RW_READER); - error = dsl_dataset_hold_obj(dp, zc->zc_fromobj, FTAG, &dsfrom); - rw_exit(&dp->dp_config_rwlock); - if (error) { - dsl_dataset_rele(ds, FTAG); + error = dsl_dataset_hold_obj(dp, zc->zc_sendobj, FTAG, &tosnap); + if (error != 0) { + dsl_pool_rele(dp, FTAG); return (error); } - error = dmu_objset_from_ds(dsfrom, &fromsnap); - if (error) { - dsl_dataset_rele(dsfrom, FTAG); - dsl_dataset_rele(ds, FTAG); - return (error); - } - } - if (zc->zc_obj) { - dsl_pool_t *dp = ds->ds_dir->dd_pool; - - if (fromsnap != NULL) { - dsl_dataset_rele(dsfrom, FTAG); - dsl_dataset_rele(ds, FTAG); - return (EINVAL); - } - - if (dsl_dir_is_clone(ds->ds_dir)) { - rw_enter(&dp->dp_config_rwlock, RW_READER); - error = dsl_dataset_hold_obj(dp, - ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &dsfrom); - rw_exit(&dp->dp_config_rwlock); - if (error) { - dsl_dataset_rele(ds, FTAG); - return (error); - } - error = dmu_objset_from_ds(dsfrom, &fromsnap); - if (error) { - dsl_dataset_rele(dsfrom, FTAG); - dsl_dataset_rele(ds, FTAG); - return (error); - } - } + if (dsl_dir_is_clone(tosnap->ds_dir)) + zc->zc_fromobj = tosnap->ds_dir->dd_phys->dd_origin_obj; + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); } if (estimate) { - error = dmu_send_estimate(tosnap, fromsnap, - &zc->zc_objset_type); - } else { - file_t *fp = getf(zc->zc_cookie); - if (fp == NULL) { - dsl_dataset_rele(ds, FTAG); - if (dsfrom) - dsl_dataset_rele(dsfrom, FTAG); - return (EBADF); + dsl_pool_t *dp; + dsl_dataset_t *tosnap; + dsl_dataset_t *fromsnap = NULL; + + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); + if (error != 0) + return (error); + + error = dsl_dataset_hold_obj(dp, zc->zc_sendobj, FTAG, &tosnap); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); } + if (zc->zc_fromobj != 0) { + error = dsl_dataset_hold_obj(dp, zc->zc_fromobj, + FTAG, &fromsnap); + if (error != 0) { + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); + return (error); + } + } + + error = dmu_send_estimate(tosnap, fromsnap, + &zc->zc_objset_type); + + if (fromsnap != NULL) + dsl_dataset_rele(fromsnap, FTAG); + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); + } else { + file_t *fp = getf(zc->zc_cookie); + if (fp == NULL) + return (EBADF); + off = fp->f_offset; - error = dmu_send(tosnap, fromsnap, - zc->zc_cookie, fp->f_vnode, &off); + error = dmu_send_obj(zc->zc_name, zc->zc_sendobj, + zc->zc_fromobj, zc->zc_cookie, fp->f_vnode, &off); if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0) fp->f_offset = off; releasef(zc->zc_cookie); } - if (dsfrom) - dsl_dataset_rele(dsfrom, FTAG); - dsl_dataset_rele(ds, FTAG); return (error); } @@ -4265,13 +4197,21 @@ zfs_ioc_send(zfs_cmd_t *zc) static int zfs_ioc_send_progress(zfs_cmd_t *zc) { + dsl_pool_t *dp; dsl_dataset_t *ds; dmu_sendarg_t *dsp = NULL; int error; - if ((error = dsl_dataset_hold(zc->zc_name, FTAG, &ds)) != 0) + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); + if (error != 0) return (error); + error = dsl_dataset_hold(dp, zc->zc_name, FTAG, &ds); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + mutex_enter(&ds->ds_sendstream_lock); /* @@ -4294,6 +4234,7 @@ zfs_ioc_send_progress(zfs_cmd_t *zc) mutex_exit(&ds->ds_sendstream_lock); dsl_dataset_rele(ds, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } @@ -4400,7 +4341,7 @@ zfs_ioc_clear(zfs_cmd_t *zc) } } - if (error) + if (error != 0) return (error); spa_vdev_state_enter(spa, SCL_NONE); @@ -4438,7 +4379,7 @@ zfs_ioc_pool_reopen(zfs_cmd_t *zc) int error; error = spa_open(zc->zc_name, &spa, FTAG); - if (error) + if (error != 0) return (error); spa_vdev_state_enter(spa, SCL_NONE); @@ -4478,7 +4419,7 @@ zfs_ioc_promote(zfs_cmd_t *zc) if (cp) *cp = '\0'; (void) dmu_objset_find(zc->zc_value, - zfs_unmount_snap, NULL, DS_FIND_SNAPSHOTS); + zfs_unmount_snap_cb, NULL, DS_FIND_SNAPSHOTS); return (dsl_dataset_promote(zc->zc_name, zc->zc_string)); } @@ -4504,7 +4445,7 @@ zfs_ioc_userspace_one(zfs_cmd_t *zc) return (EINVAL); error = zfsvfs_hold(zc->zc_name, FTAG, &zfsvfs, B_FALSE); - if (error) + if (error != 0) return (error); error = zfs_userspace_one(zfsvfs, @@ -4535,7 +4476,7 @@ zfs_ioc_userspace_many(zfs_cmd_t *zc) return (ENOMEM); int error = zfsvfs_hold(zc->zc_name, FTAG, &zfsvfs, B_FALSE); - if (error) + if (error != 0) return (error); void *buf = kmem_alloc(bufsize, KM_SLEEP); @@ -4585,7 +4526,7 @@ zfs_ioc_userspace_upgrade(zfs_cmd_t *zc) } else { /* XXX kind of reading contents without owning */ error = dmu_objset_hold(zc->zc_name, FTAG, &os); - if (error) + if (error != 0) return (error); error = dmu_objset_userspace_upgrade(os); @@ -4658,7 +4599,7 @@ zfs_ioc_share(zfs_cmd_t *zc) return (ENOSYS); } error = zfs_init_sharefs(); - if (error) { + if (error != 0) { mutex_exit(&zfs_share_lock); return (ENOSYS); } @@ -4683,7 +4624,7 @@ zfs_ioc_share(zfs_cmd_t *zc) return (ENOSYS); } error = zfs_init_sharefs(); - if (error) { + if (error != 0) { mutex_exit(&zfs_share_lock); return (ENOSYS); } @@ -4748,7 +4689,7 @@ zfs_ioc_next_obj(zfs_cmd_t *zc) int error; error = dmu_objset_hold(zc->zc_name, FTAG, &os); - if (error) + if (error != 0) return (error); error = dmu_object_next(os, &zc->zc_obj, B_FALSE, @@ -4771,25 +4712,26 @@ static int zfs_ioc_tmp_snapshot(zfs_cmd_t *zc) { char *snap_name; + char *hold_name; int error; + minor_t minor; - snap_name = kmem_asprintf("%s@%s-%016llx", zc->zc_name, zc->zc_value, - (u_longlong_t)ddi_get_lbolt64()); - - if (strlen(snap_name) >= MAXPATHLEN) { - strfree(snap_name); - return (E2BIG); - } - - error = dmu_objset_snapshot_tmp(snap_name, "%temp", zc->zc_cleanup_fd); - if (error != 0) { - strfree(snap_name); + error = zfs_onexit_fd_hold(zc->zc_cleanup_fd, &minor); + if (error != 0) return (error); - } - (void) strcpy(zc->zc_value, strchr(snap_name, '@') + 1); + snap_name = kmem_asprintf("%s-%016llx", zc->zc_value, + (u_longlong_t)ddi_get_lbolt64()); + hold_name = kmem_asprintf("%%%s", zc->zc_value); + + error = dsl_dataset_snapshot_tmp(zc->zc_name, snap_name, minor, + hold_name); + if (error == 0) + (void) strcpy(zc->zc_value, snap_name); strfree(snap_name); - return (0); + strfree(hold_name); + zfs_onexit_fd_rele(zc->zc_cleanup_fd); + return (error); } /* @@ -4804,39 +4746,22 @@ zfs_ioc_tmp_snapshot(zfs_cmd_t *zc) static int zfs_ioc_diff(zfs_cmd_t *zc) { - objset_t *fromsnap; - objset_t *tosnap; file_t *fp; offset_t off; int error; - error = dmu_objset_hold(zc->zc_name, FTAG, &tosnap); - if (error) - return (error); - - error = dmu_objset_hold(zc->zc_value, FTAG, &fromsnap); - if (error) { - dmu_objset_rele(tosnap, FTAG); - return (error); - } - fp = getf(zc->zc_cookie); - if (fp == NULL) { - dmu_objset_rele(fromsnap, FTAG); - dmu_objset_rele(tosnap, FTAG); + if (fp == NULL) return (EBADF); - } off = fp->f_offset; - error = dmu_diff(tosnap, fromsnap, fp->f_vnode, &off); + error = dmu_diff(zc->zc_name, zc->zc_value, fp->f_vnode, &off); if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0) fp->f_offset = off; releasef(zc->zc_cookie); - dmu_objset_rele(fromsnap, FTAG); - dmu_objset_rele(tosnap, FTAG); return (error); } @@ -4905,13 +4830,13 @@ zfs_ioc_smb_acl(zfs_cmd_t *zc) ZFS_SHARES_DIR); dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); error = dmu_tx_assign(tx, TXG_WAIT); - if (error) { + if (error != 0) { dmu_tx_abort(tx); } else { error = zfs_create_share_dir(zfsvfs, tx); dmu_tx_commit(tx); } - if (error) { + if (error != 0) { mutex_exit(&zfsvfs->z_lock); VN_RELE(vp); ZFS_EXIT(zfsvfs); @@ -4990,124 +4915,82 @@ zfs_ioc_smb_acl(zfs_cmd_t *zc) } /* - * inputs: - * zc_name name of filesystem - * zc_value short name of snap - * zc_string user-supplied tag for this hold - * zc_cookie recursive flag - * zc_temphold set if hold is temporary - * zc_cleanup_fd cleanup-on-exit file descriptor for calling process - * zc_sendobj if non-zero, the objid for zc_name@zc_value - * zc_createtxg if zc_sendobj is non-zero, snap must have zc_createtxg + * innvl: { + * "holds" -> { snapname -> holdname (string), ... } + * (optional) "cleanup_fd" -> fd (int32) + * } * - * outputs: none + * outnvl: { + * snapname -> error value (int32) + * ... + * } */ +/* ARGSUSED */ static int -zfs_ioc_hold(zfs_cmd_t *zc) +zfs_ioc_hold(const char *pool, nvlist_t *args, nvlist_t *errlist) { - boolean_t recursive = zc->zc_cookie; - spa_t *spa; - dsl_pool_t *dp; - dsl_dataset_t *ds; + nvlist_t *holds; + int cleanup_fd = -1; int error; minor_t minor = 0; - if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0) + error = nvlist_lookup_nvlist(args, "holds", &holds); + if (error != 0) return (EINVAL); - if (zc->zc_sendobj == 0) { - return (dsl_dataset_user_hold(zc->zc_name, zc->zc_value, - zc->zc_string, recursive, zc->zc_temphold, - zc->zc_cleanup_fd)); + if (nvlist_lookup_int32(args, "cleanup_fd", &cleanup_fd) == 0) { + error = zfs_onexit_fd_hold(cleanup_fd, &minor); + if (error != 0) + return (error); } - if (recursive) - return (EINVAL); + error = dsl_dataset_user_hold(holds, minor, errlist); + if (minor != 0) + zfs_onexit_fd_rele(cleanup_fd); + return (error); +} - error = spa_open(zc->zc_name, &spa, FTAG); - if (error) - return (error); +/* + * innvl is not used. + * + * outnvl: { + * holdname -> time added (uint64 seconds since epoch) + * ... + * } + */ +/* ARGSUSED */ +static int +zfs_ioc_get_holds(const char *snapname, nvlist_t *args, nvlist_t *outnvl) +{ + return (dsl_dataset_get_holds(snapname, outnvl)); +} - dp = spa_get_dsl(spa); - rw_enter(&dp->dp_config_rwlock, RW_READER); - error = dsl_dataset_hold_obj(dp, zc->zc_sendobj, FTAG, &ds); - rw_exit(&dp->dp_config_rwlock); - spa_close(spa, FTAG); - if (error) - return (error); +/* + * innvl: { + * snapname -> { holdname, ... } + * ... + * } + * + * outnvl: { + * snapname -> error value (int32) + * ... + * } + */ +/* ARGSUSED */ +static int +zfs_ioc_release(const char *pool, nvlist_t *holds, nvlist_t *errlist) +{ + nvpair_t *pair; /* - * Until we have a hold on this snapshot, it's possible that - * zc_sendobj could've been destroyed and reused as part - * of a later txg. Make sure we're looking at the right object. + * The release may cause the snapshot to be destroyed; make sure it + * is not mounted. */ - if (zc->zc_createtxg != ds->ds_phys->ds_creation_txg) { - dsl_dataset_rele(ds, FTAG); - return (ENOENT); - } + for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL; + pair = nvlist_next_nvpair(holds, pair)) + zfs_unmount_snap(nvpair_name(pair)); - if (zc->zc_cleanup_fd != -1 && zc->zc_temphold) { - error = zfs_onexit_fd_hold(zc->zc_cleanup_fd, &minor); - if (error) { - dsl_dataset_rele(ds, FTAG); - return (error); - } - } - - error = dsl_dataset_user_hold_for_send(ds, zc->zc_string, - zc->zc_temphold); - if (minor != 0) { - if (error == 0) { - dsl_register_onexit_hold_cleanup(ds, zc->zc_string, - minor); - } - zfs_onexit_fd_rele(zc->zc_cleanup_fd); - } - dsl_dataset_rele(ds, FTAG); - - return (error); -} - -/* - * inputs: - * zc_name name of dataset from which we're releasing a user hold - * zc_value short name of snap - * zc_string user-supplied tag for this hold - * zc_cookie recursive flag - * - * outputs: none - */ -static int -zfs_ioc_release(zfs_cmd_t *zc) -{ - boolean_t recursive = zc->zc_cookie; - - if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0) - return (EINVAL); - - return (dsl_dataset_user_release(zc->zc_name, zc->zc_value, - zc->zc_string, recursive)); -} - -/* - * inputs: - * zc_name name of filesystem - * - * outputs: - * zc_nvlist_src{_size} nvlist of snapshot holds - */ -static int -zfs_ioc_get_holds(zfs_cmd_t *zc) -{ - nvlist_t *nvp; - int error; - - if ((error = dsl_dataset_get_holds(zc->zc_name, &nvp)) == 0) { - error = put_nvlist(zc, nvp); - nvlist_free(nvp); - } - - return (error); + return (dsl_dataset_user_release(holds, errlist)); } /* @@ -5124,14 +5007,21 @@ static int zfs_ioc_space_written(zfs_cmd_t *zc) { int error; + dsl_pool_t *dp; dsl_dataset_t *new, *old; - error = dsl_dataset_hold(zc->zc_name, FTAG, &new); + error = dsl_pool_hold(zc->zc_name, FTAG, &dp); if (error != 0) return (error); - error = dsl_dataset_hold(zc->zc_value, FTAG, &old); + error = dsl_dataset_hold(dp, zc->zc_name, FTAG, &new); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + error = dsl_dataset_hold(dp, zc->zc_value, FTAG, &old); if (error != 0) { dsl_dataset_rele(new, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } @@ -5139,8 +5029,10 @@ zfs_ioc_space_written(zfs_cmd_t *zc) &zc->zc_objset_type, &zc->zc_perm_action); dsl_dataset_rele(old, FTAG); dsl_dataset_rele(new, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } + /* * innvl: { * "firstsnap" -> snapshot name @@ -5156,6 +5048,7 @@ static int zfs_ioc_space_snaps(const char *lastsnap, nvlist_t *innvl, nvlist_t *outnvl) { int error; + dsl_pool_t *dp; dsl_dataset_t *new, *old; char *firstsnap; uint64_t used, comp, uncomp; @@ -5163,18 +5056,26 @@ zfs_ioc_space_snaps(const char *lastsnap, nvlist_t *innvl, nvlist_t *outnvl) if (nvlist_lookup_string(innvl, "firstsnap", &firstsnap) != 0) return (EINVAL); - error = dsl_dataset_hold(lastsnap, FTAG, &new); + error = dsl_pool_hold(lastsnap, FTAG, &dp); if (error != 0) return (error); - error = dsl_dataset_hold(firstsnap, FTAG, &old); + + error = dsl_dataset_hold(dp, lastsnap, FTAG, &new); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + error = dsl_dataset_hold(dp, firstsnap, FTAG, &old); if (error != 0) { dsl_dataset_rele(new, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } error = dsl_dataset_space_wouldfree(old, new, &used, &comp, &uncomp); dsl_dataset_rele(old, FTAG); dsl_dataset_rele(new, FTAG); + dsl_pool_rele(dp, FTAG); fnvlist_add_uint64(outnvl, "used", used); fnvlist_add_uint64(outnvl, "compressed", comp); fnvlist_add_uint64(outnvl, "uncompressed", uncomp); @@ -5193,47 +5094,27 @@ zfs_ioc_space_snaps(const char *lastsnap, nvlist_t *innvl, nvlist_t *outnvl) static int zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) { - objset_t *fromsnap = NULL; - objset_t *tosnap; int error; offset_t off; - char *fromname; + char *fromname = NULL; int fd; error = nvlist_lookup_int32(innvl, "fd", &fd); if (error != 0) return (EINVAL); - error = dmu_objset_hold(snapname, FTAG, &tosnap); - if (error) - return (error); - - error = nvlist_lookup_string(innvl, "fromsnap", &fromname); - if (error == 0) { - error = dmu_objset_hold(fromname, FTAG, &fromsnap); - if (error) { - dmu_objset_rele(tosnap, FTAG); - return (error); - } - } + (void) nvlist_lookup_string(innvl, "fromsnap", &fromname); file_t *fp = getf(fd); - if (fp == NULL) { - dmu_objset_rele(tosnap, FTAG); - if (fromsnap != NULL) - dmu_objset_rele(fromsnap, FTAG); + if (fp == NULL) return (EBADF); - } off = fp->f_offset; - error = dmu_send(tosnap, fromsnap, fd, fp->f_vnode, &off); + error = dmu_send(snapname, fromname, fd, fp->f_vnode, &off); if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0) fp->f_offset = off; releasef(fd); - if (fromsnap != NULL) - dmu_objset_rele(fromsnap, FTAG); - dmu_objset_rele(tosnap, FTAG); return (error); } @@ -5252,21 +5133,29 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) static int zfs_ioc_send_space(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) { - objset_t *fromsnap = NULL; - objset_t *tosnap; + dsl_pool_t *dp; + dsl_dataset_t *fromsnap = NULL; + dsl_dataset_t *tosnap; int error; char *fromname; uint64_t space; - error = dmu_objset_hold(snapname, FTAG, &tosnap); - if (error) + error = dsl_pool_hold(snapname, FTAG, &dp); + if (error != 0) return (error); + error = dsl_dataset_hold(dp, snapname, FTAG, &tosnap); + if (error != 0) { + dsl_pool_rele(dp, FTAG); + return (error); + } + error = nvlist_lookup_string(innvl, "fromsnap", &fromname); if (error == 0) { - error = dmu_objset_hold(fromname, FTAG, &fromsnap); - if (error) { - dmu_objset_rele(tosnap, FTAG); + error = dsl_dataset_hold(dp, fromname, FTAG, &fromsnap); + if (error != 0) { + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } } @@ -5275,8 +5164,9 @@ zfs_ioc_send_space(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) fnvlist_add_uint64(outnvl, "space", space); if (fromsnap != NULL) - dmu_objset_rele(fromsnap, FTAG); - dmu_objset_rele(tosnap, FTAG); + dsl_dataset_rele(fromsnap, FTAG); + dsl_dataset_rele(tosnap, FTAG); + dsl_pool_rele(dp, FTAG); return (error); } @@ -5421,6 +5311,17 @@ zfs_ioctl_init(void) zfs_ioc_destroy_snaps, zfs_secpolicy_destroy_snaps, POOL_NAME, POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE); + zfs_ioctl_register("hold", ZFS_IOC_HOLD, + zfs_ioc_hold, zfs_secpolicy_hold, POOL_NAME, + POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE); + zfs_ioctl_register("release", ZFS_IOC_RELEASE, + zfs_ioc_release, zfs_secpolicy_release, POOL_NAME, + POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE); + + zfs_ioctl_register("get_holds", ZFS_IOC_GET_HOLDS, + zfs_ioc_get_holds, zfs_secpolicy_read, DATASET_NAME, + POOL_CHECK_SUSPENDED, B_FALSE, B_FALSE); + /* IOCTLS that use the legacy function signature */ zfs_ioctl_register_legacy(ZFS_IOC_POOL_FREEZE, zfs_ioc_pool_freeze, @@ -5498,8 +5399,6 @@ zfs_ioctl_init(void) zfs_ioctl_register_dataset_read(ZFS_IOC_SPACE_WRITTEN, zfs_ioc_space_written); - zfs_ioctl_register_dataset_read(ZFS_IOC_GET_HOLDS, - zfs_ioc_get_holds); zfs_ioctl_register_dataset_read(ZFS_IOC_OBJSET_RECVD_PROPS, zfs_ioc_objset_recvd_props); zfs_ioctl_register_dataset_read(ZFS_IOC_NEXT_OBJ, @@ -5542,10 +5441,6 @@ zfs_ioctl_init(void) zfs_secpolicy_recv); zfs_ioctl_register_dataset_modify(ZFS_IOC_PROMOTE, zfs_ioc_promote, zfs_secpolicy_promote); - zfs_ioctl_register_dataset_modify(ZFS_IOC_HOLD, zfs_ioc_hold, - zfs_secpolicy_hold); - zfs_ioctl_register_dataset_modify(ZFS_IOC_RELEASE, zfs_ioc_release, - zfs_secpolicy_release); zfs_ioctl_register_dataset_modify(ZFS_IOC_INHERIT_PROP, zfs_ioc_inherit_prop, zfs_secpolicy_inherit_prop); zfs_ioctl_register_dataset_modify(ZFS_IOC_SET_FSACL, zfs_ioc_set_fsacl, diff --git a/uts/common/fs/zfs/zfs_vfsops.c b/uts/common/fs/zfs/zfs_vfsops.c index 2a2501707b48..42486ea23c59 100644 --- a/uts/common/fs/zfs/zfs_vfsops.c +++ b/uts/common/fs/zfs/zfs_vfsops.c @@ -513,27 +513,31 @@ zfs_register_callbacks(vfs_t *vfsp) * overboard... */ ds = dmu_objset_ds(os); - error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs); + dsl_pool_config_enter(dmu_objset_pool(os), FTAG); + error = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "xattr", xattr_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "recordsize", blksz_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "readonly", readonly_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "devices", devices_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "setuid", setuid_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "exec", exec_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "snapdir", snapdir_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "aclmode", acl_mode_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, - "aclinherit", acl_inherit_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, + zfsvfs); error = error ? error : dsl_prop_register(ds, - "vscan", vscan_changed_cb, zfsvfs); + zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); + dsl_pool_config_exit(dmu_objset_pool(os), FTAG); if (error) goto unregister; @@ -563,28 +567,35 @@ zfs_register_callbacks(vfs_t *vfsp) * registered, but this is OK; it will simply return ENOMSG, * which we will ignore. */ - (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs); - (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, - zfsvfs); - (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME), + atime_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR), + xattr_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), + blksz_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY), + readonly_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), + devices_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID), + setuid_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC), + exec_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), + snapdir_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE), + acl_mode_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), + acl_inherit_changed_cb, zfsvfs); + (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), + vscan_changed_cb, zfsvfs); return (error); - } static int zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, uint64_t *userp, uint64_t *groupp) { - int error = 0; - /* * Is it a valid type of object to track? */ @@ -641,7 +652,7 @@ zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, *groupp = BSWAP_64(*groupp); } } - return (error); + return (0); } static void @@ -993,7 +1004,7 @@ zfsvfs_create(const char *osname, zfsvfs_t **zfvp) mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), offsetof(znode_t, z_link_node)); - rrw_init(&zfsvfs->z_teardown_lock); + rrw_init(&zfsvfs->z_teardown_lock, B_FALSE); rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) @@ -1444,8 +1455,9 @@ zfs_mount_label_policy(vfs_t *vfsp, char *osname) char *str = NULL; if (l_to_str_internal(mnt_sl, &str) == 0 && - dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), - ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0) + dsl_prop_set_string(osname, + zfs_prop_to_name(ZFS_PROP_MLSLABEL), + ZPROP_SRC_LOCAL, str) == 0) retv = 0; if (str != NULL) kmem_free(str, strlen(str) + 1); @@ -1856,7 +1868,7 @@ zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); - (void) dmu_objset_evict_dbufs(zfsvfs->z_os); + dmu_objset_evict_dbufs(zfsvfs->z_os); return (0); } diff --git a/uts/common/fs/zfs/zil.c b/uts/common/fs/zfs/zil.c index 81d2bb5a9755..e9616f8f65e0 100644 --- a/uts/common/fs/zfs/zil.c +++ b/uts/common/fs/zfs/zil.c @@ -235,7 +235,7 @@ zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, } } - VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); + VERIFY(arc_buf_remove_ref(abuf, &abuf)); } return (error); @@ -332,7 +332,7 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, break; error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end); - if (error) + if (error != 0) break; for (lrp = lrbuf; lrp < end; lrp += reclen) { @@ -467,7 +467,7 @@ zilog_dirty(zilog_t *zilog, uint64_t txg) if (dsl_dataset_is_snapshot(ds)) panic("dirtying snapshot!"); - if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg) == 0) { + if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) { /* up the hold count until we can be written out */ dmu_buf_add_ref(ds->ds_dbuf, zilog); } @@ -626,8 +626,8 @@ zil_claim(const char *osname, void *txarg) objset_t *os; int error; - error = dmu_objset_hold(osname, FTAG, &os); - if (error) { + error = dmu_objset_own(osname, DMU_OST_ANY, B_FALSE, FTAG, &os); + if (error != 0) { cmn_err(CE_WARN, "can't open objset for %s", osname); return (0); } @@ -640,7 +640,7 @@ zil_claim(const char *osname, void *txarg) zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log); BP_ZERO(&zh->zh_log); dsl_dataset_dirty(dmu_objset_ds(os), tx); - dmu_objset_rele(os, FTAG); + dmu_objset_disown(os, FTAG); return (0); } @@ -665,7 +665,7 @@ zil_claim(const char *osname, void *txarg) } ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); - dmu_objset_rele(os, FTAG); + dmu_objset_disown(os, FTAG); return (0); } @@ -685,7 +685,7 @@ zil_check_log_chain(const char *osname, void *tx) ASSERT(tx == NULL); error = dmu_objset_hold(osname, FTAG, &os); - if (error) { + if (error != 0) { cmn_err(CE_WARN, "can't open objset for %s", osname); return (0); } @@ -973,7 +973,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) /* pass the old blkptr in order to spread log blocks across devs */ error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz, USE_SLOG(zilog)); - if (!error) { + if (error == 0) { ASSERT3U(bp->blk_birth, ==, txg); bp->blk_cksum = lwb->lwb_blk.blk_cksum; bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; @@ -1084,7 +1084,7 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) txg_wait_synced(zilog->zl_dmu_pool, txg); return (lwb); } - if (error) { + if (error != 0) { ASSERT(error == ENOENT || error == EEXIST || error == EALREADY); return (lwb); @@ -1708,6 +1708,9 @@ zil_free(zilog_t *zilog) { zilog->zl_stop_sync = 1; + ASSERT0(zilog->zl_suspend); + ASSERT0(zilog->zl_suspending); + ASSERT(list_is_empty(&zilog->zl_lwb_list)); list_destroy(&zilog->zl_lwb_list); @@ -1803,32 +1806,100 @@ zil_close(zilog_t *zilog) mutex_exit(&zilog->zl_lock); } +static char *suspend_tag = "zil suspending"; + /* * Suspend an intent log. While in suspended mode, we still honor * synchronous semantics, but we rely on txg_wait_synced() to do it. - * We suspend the log briefly when taking a snapshot so that the snapshot - * contains all the data it's supposed to, and has an empty intent log. + * On old version pools, we suspend the log briefly when taking a + * snapshot so that it will have an empty intent log. + * + * Long holds are not really intended to be used the way we do here -- + * held for such a short time. A concurrent caller of dsl_dataset_long_held() + * could fail. Therefore we take pains to only put a long hold if it is + * actually necessary. Fortunately, it will only be necessary if the + * objset is currently mounted (or the ZVOL equivalent). In that case it + * will already have a long hold, so we are not really making things any worse. + * + * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or + * zvol_state_t), and use their mechanism to prevent their hold from being + * dropped (e.g. VFS_HOLD()). However, that would be even more pain for + * very little gain. + * + * if cookiep == NULL, this does both the suspend & resume. + * Otherwise, it returns with the dataset "long held", and the cookie + * should be passed into zil_resume(). */ int -zil_suspend(zilog_t *zilog) +zil_suspend(const char *osname, void **cookiep) { - const zil_header_t *zh = zilog->zl_header; + objset_t *os; + zilog_t *zilog; + const zil_header_t *zh; + int error; + + error = dmu_objset_hold(osname, suspend_tag, &os); + if (error != 0) + return (error); + zilog = dmu_objset_zil(os); mutex_enter(&zilog->zl_lock); + zh = zilog->zl_header; + if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ mutex_exit(&zilog->zl_lock); + dmu_objset_rele(os, suspend_tag); return (EBUSY); } - if (zilog->zl_suspend++ != 0) { + + /* + * Don't put a long hold in the cases where we can avoid it. This + * is when there is no cookie so we are doing a suspend & resume + * (i.e. called from zil_vdev_offline()), and there's nothing to do + * for the suspend because it's already suspended, or there's no ZIL. + */ + if (cookiep == NULL && !zilog->zl_suspending && + (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) { + mutex_exit(&zilog->zl_lock); + dmu_objset_rele(os, suspend_tag); + return (0); + } + + dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag); + dsl_pool_rele(dmu_objset_pool(os), suspend_tag); + + zilog->zl_suspend++; + + if (zilog->zl_suspend > 1) { /* - * Someone else already began a suspend. + * Someone else is already suspending it. * Just wait for them to finish. */ + while (zilog->zl_suspending) cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); mutex_exit(&zilog->zl_lock); + + if (cookiep == NULL) + zil_resume(os); + else + *cookiep = os; return (0); } + + /* + * If there is no pointer to an on-disk block, this ZIL must not + * be active (e.g. filesystem not mounted), so there's nothing + * to clean up. + */ + if (BP_IS_HOLE(&zh->zh_log)) { + ASSERT(cookiep != NULL); /* fast path already handled */ + + *cookiep = os; + mutex_exit(&zilog->zl_lock); + return (0); + } + zilog->zl_suspending = B_TRUE; mutex_exit(&zilog->zl_lock); @@ -1841,16 +1912,25 @@ zil_suspend(zilog_t *zilog) cv_broadcast(&zilog->zl_cv_suspend); mutex_exit(&zilog->zl_lock); + if (cookiep == NULL) + zil_resume(os); + else + *cookiep = os; return (0); } void -zil_resume(zilog_t *zilog) +zil_resume(void *cookie) { + objset_t *os = cookie; + zilog_t *zilog = dmu_objset_zil(os); + mutex_enter(&zilog->zl_lock); ASSERT(zilog->zl_suspend != 0); zilog->zl_suspend--; mutex_exit(&zilog->zl_lock); + dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag); + dsl_dataset_rele(dmu_objset_ds(os), suspend_tag); } typedef struct zil_replay_arg { @@ -1923,7 +2003,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { error = zil_read_log_data(zilog, (lr_write_t *)lr, zr->zr_lr + reclen); - if (error) + if (error != 0) return (zil_replay_error(zilog, lr, error)); } @@ -1944,7 +2024,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) * is updated if we are in replay mode. */ error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap); - if (error) { + if (error != 0) { /* * The DMU's dnode layer doesn't see removes until the txg * commits, so a subsequent claim can spuriously fail with @@ -1954,7 +2034,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) */ txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE); - if (error) + if (error != 0) return (zil_replay_error(zilog, lr, error)); } return (0); @@ -2026,19 +2106,10 @@ zil_replaying(zilog_t *zilog, dmu_tx_t *tx) int zil_vdev_offline(const char *osname, void *arg) { - objset_t *os; - zilog_t *zilog; int error; - error = dmu_objset_hold(osname, FTAG, &os); - if (error) - return (error); - - zilog = dmu_objset_zil(os); - if (zil_suspend(zilog) != 0) - error = EEXIST; - else - zil_resume(zilog); - dmu_objset_rele(os, FTAG); - return (error); + error = zil_suspend(osname, NULL); + if (error != 0) + return (EEXIST); + return (0); } diff --git a/uts/common/fs/zfs/zio.c b/uts/common/fs/zfs/zio.c index 432a992b26f4..7940e201ef80 100644 --- a/uts/common/fs/zfs/zio.c +++ b/uts/common/fs/zfs/zio.c @@ -697,6 +697,7 @@ zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) { + metaslab_check_free(spa, bp); bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); } @@ -713,6 +714,8 @@ zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, ASSERT(spa_syncing_txg(spa) == txg); ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); + metaslab_check_free(spa, bp); + zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE); @@ -2010,7 +2013,7 @@ zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) bcmp(abuf->b_data, zio->io_orig_data, zio->io_orig_size) != 0) error = EEXIST; - VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); + VERIFY(arc_buf_remove_ref(abuf, &abuf)); } ddt_enter(ddt); @@ -2600,8 +2603,9 @@ zio_vdev_io_assess(zio_t *zio) * set vdev_cant_write so that we stop trying to allocate from it. */ if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && - vd != NULL && !vd->vdev_ops->vdev_op_leaf) + vd != NULL && !vd->vdev_ops->vdev_op_leaf) { vd->vdev_cant_write = B_TRUE; + } if (zio->io_error) zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; diff --git a/uts/common/fs/zfs/zvol.c b/uts/common/fs/zfs/zvol.c index b413f5ed8b00..5911fd3d2d07 100644 --- a/uts/common/fs/zfs/zvol.c +++ b/uts/common/fs/zfs/zvol.c @@ -653,7 +653,7 @@ zvol_last_close(zvol_state_t *zv) if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && !(zv->zv_flags & ZVOL_RDONLY)) txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); - (void) dmu_objset_evict_dbufs(zv->zv_objset); + dmu_objset_evict_dbufs(zv->zv_objset); dmu_objset_disown(zv->zv_objset, zvol_tag); zv->zv_objset = NULL; @@ -698,7 +698,7 @@ zvol_prealloc(zvol_state_t *zv) return (0); } -int +static int zvol_update_volsize(objset_t *os, uint64_t volsize) { dmu_tx_t *tx; @@ -749,13 +749,12 @@ zvol_remove_minors(const char *name) } static int -zvol_set_volsize_impl(objset_t *os, zvol_state_t *zv, uint64_t volsize) +zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) { uint64_t old_volsize = 0ULL; - int error; + int error = 0; ASSERT(MUTEX_HELD(&zfsdev_state_lock)); - error = zvol_update_volsize(os, volsize); /* * Reinitialize the dump area to the new size. If we @@ -764,27 +763,25 @@ zvol_set_volsize_impl(objset_t *os, zvol_state_t *zv, uint64_t volsize) * to calling dumpvp_resize() to ensure that the devices' * size(9P) is not visible by the dump subsystem. */ - if (zv && error == 0) { - old_volsize = zv->zv_volsize; - zvol_size_changed(zv, volsize); + old_volsize = zv->zv_volsize; + zvol_size_changed(zv, volsize); - if (zv->zv_flags & ZVOL_DUMPIFIED) { - if ((error = zvol_dumpify(zv)) != 0 || - (error = dumpvp_resize()) != 0) { - int dumpify_error; + if (zv->zv_flags & ZVOL_DUMPIFIED) { + if ((error = zvol_dumpify(zv)) != 0 || + (error = dumpvp_resize()) != 0) { + int dumpify_error; - (void) zvol_update_volsize(os, old_volsize); - zvol_size_changed(zv, old_volsize); - dumpify_error = zvol_dumpify(zv); - error = dumpify_error ? dumpify_error : error; - } + (void) zvol_update_volsize(zv->zv_objset, old_volsize); + zvol_size_changed(zv, old_volsize); + dumpify_error = zvol_dumpify(zv); + error = dumpify_error ? dumpify_error : error; } } /* * Generate a LUN expansion event. */ - if (zv && error == 0) { + if (error == 0) { sysevent_id_t eid; nvlist_t *attr; char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); @@ -812,29 +809,45 @@ zvol_set_volsize(const char *name, uint64_t volsize) int error; dmu_object_info_t doi; uint64_t readonly; + boolean_t owned = B_FALSE; + + error = dsl_prop_get_integer(name, + zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); + if (error != 0) + return (error); + if (readonly) + return (EROFS); mutex_enter(&zfsdev_state_lock); zv = zvol_minor_lookup(name); - if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { - mutex_exit(&zfsdev_state_lock); - return (error); + + if (zv == NULL || zv->zv_objset == NULL) { + if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, + FTAG, &os)) != 0) { + mutex_exit(&zfsdev_state_lock); + return (error); + } + owned = B_TRUE; + if (zv != NULL) + zv->zv_objset = os; + } else { + os = zv->zv_objset; } if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || - (error = zvol_check_volsize(volsize, - doi.doi_data_block_size)) != 0) + (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) goto out; - VERIFY3U(dsl_prop_get_integer(name, - zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL), ==, 0); - if (readonly) { - error = EROFS; - goto out; - } + error = zvol_update_volsize(os, volsize); - error = zvol_set_volsize_impl(os, zv, volsize); + if (error == 0 && zv != NULL) + error = zvol_update_live_volsize(zv, volsize); out: - dmu_objset_rele(os, FTAG); + if (owned) { + dmu_objset_disown(os, FTAG); + if (zv != NULL) + zv->zv_objset = NULL; + } mutex_exit(&zfsdev_state_lock); return (error); } @@ -1155,6 +1168,9 @@ zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, ze = list_next(&zv->zv_extents, ze); } + if (ze == NULL) + return (EINVAL); + if (!ddi_in_panic()) spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); @@ -1308,6 +1324,9 @@ zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) if (zv == NULL) return (ENXIO); + if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) + return (EINVAL); + boff = ldbtob(blkno); resid = ldbtob(nblocks); diff --git a/uts/common/sys/nvpair.h b/uts/common/sys/nvpair.h index ad25effc2994..e4d637b0073c 100644 --- a/uts/common/sys/nvpair.h +++ b/uts/common/sys/nvpair.h @@ -284,6 +284,7 @@ void fnvlist_pack_free(char *, size_t); nvlist_t *fnvlist_unpack(char *, size_t); nvlist_t *fnvlist_dup(nvlist_t *); void fnvlist_merge(nvlist_t *, nvlist_t *); +size_t fnvlist_num_pairs(nvlist_t *); void fnvlist_add_boolean(nvlist_t *, const char *); void fnvlist_add_boolean_value(nvlist_t *, const char *, boolean_t); From 0c75791b4cd5947e74ef52fb08958459b762d1d6 Mon Sep 17 00:00:00 2001 From: Martin Matuska Date: Tue, 5 Mar 2013 15:33:28 +0000 Subject: [PATCH 3/4] Update vendor/illumos/dist and vendor-sys/illumos/dist to illumos-gate 13975:ef6409bc370f Illumos ZFS issues: 3582 zfs_delay() should support a variable resolution 3584 DTrace sdt probes for ZFS txg states --- lib/libzpool/common/kernel.c | 35 +++++++++++++++++++++++++++ lib/libzpool/common/sys/zfs_context.h | 2 ++ uts/common/fs/zfs/dsl_dir.c | 3 ++- uts/common/fs/zfs/dsl_pool.c | 18 +++++++++----- uts/common/fs/zfs/dsl_scan.c | 8 +++--- uts/common/fs/zfs/spa_misc.c | 6 ++--- uts/common/fs/zfs/sys/txg.h | 9 ++----- uts/common/fs/zfs/sys/txg_impl.h | 12 ++++++--- uts/common/fs/zfs/txg.c | 27 +++++++++++++-------- 9 files changed, 85 insertions(+), 35 deletions(-) diff --git a/lib/libzpool/common/kernel.c b/lib/libzpool/common/kernel.c index 96280941a646..4dd614f7c1bc 100644 --- a/lib/libzpool/common/kernel.c +++ b/lib/libzpool/common/kernel.c @@ -329,6 +329,41 @@ cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime) return (1); } +/*ARGSUSED*/ +clock_t +cv_timedwait_hires(kcondvar_t *cv, kmutex_t *mp, hrtime_t tim, hrtime_t res, + int flag) +{ + int error; + timestruc_t ts; + hrtime_t delta; + + ASSERT(flag == 0); + +top: + delta = tim - gethrtime(); + if (delta <= 0) + return (-1); + + ts.tv_sec = delta / NANOSEC; + ts.tv_nsec = delta % NANOSEC; + + ASSERT(mutex_owner(mp) == curthread); + mp->m_owner = NULL; + error = cond_reltimedwait(cv, &mp->m_lock, &ts); + mp->m_owner = curthread; + + if (error == ETIME) + return (-1); + + if (error == EINTR) + goto top; + + ASSERT(error == 0); + + return (1); +} + void cv_signal(kcondvar_t *cv) { diff --git a/lib/libzpool/common/sys/zfs_context.h b/lib/libzpool/common/sys/zfs_context.h index 7802da4e2b00..a46c657a1d70 100644 --- a/lib/libzpool/common/sys/zfs_context.h +++ b/lib/libzpool/common/sys/zfs_context.h @@ -254,6 +254,8 @@ extern void cv_init(kcondvar_t *cv, char *name, int type, void *arg); extern void cv_destroy(kcondvar_t *cv); extern void cv_wait(kcondvar_t *cv, kmutex_t *mp); extern clock_t cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime); +extern clock_t cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim, + hrtime_t res, int flag); extern void cv_signal(kcondvar_t *cv); extern void cv_broadcast(kcondvar_t *cv); diff --git a/uts/common/fs/zfs/dsl_dir.c b/uts/common/fs/zfs/dsl_dir.c index 1e7ba6d6cbef..c53c3c8e877c 100644 --- a/uts/common/fs/zfs/dsl_dir.c +++ b/uts/common/fs/zfs/dsl_dir.c @@ -738,7 +738,8 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx); } else { if (err == EAGAIN) { - txg_delay(dd->dd_pool, tx->tx_txg, 1); + txg_delay(dd->dd_pool, tx->tx_txg, + MSEC2NSEC(10), MSEC2NSEC(10)); err = ERESTART; } dsl_pool_memory_pressure(dd->dd_pool); diff --git a/uts/common/fs/zfs/dsl_pool.c b/uts/common/fs/zfs/dsl_pool.c index 6af631679e9e..c83068a00101 100644 --- a/uts/common/fs/zfs/dsl_pool.c +++ b/uts/common/fs/zfs/dsl_pool.c @@ -58,6 +58,9 @@ kmutex_t zfs_write_limit_lock; static pgcnt_t old_physmem = 0; +hrtime_t zfs_throttle_delay = MSEC2NSEC(10); +hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); + int dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) { @@ -511,12 +514,13 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) * Weight the throughput calculation towards the current value: * thru = 3/4 old_thru + 1/4 new_thru * - * Note: write_time is in nanosecs, so write_time/MICROSEC - * yields millisecs + * Note: write_time is in nanosecs while dp_throughput is expressed in + * bytes per millisecond. */ ASSERT(zfs_write_limit_min > 0); - if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) { - uint64_t throughput = data_written / (write_time / MICROSEC); + if (data_written > zfs_write_limit_min / 8 && + write_time > MSEC2NSEC(1)) { + uint64_t throughput = data_written / NSEC2MSEC(write_time); if (dp->dp_throughput) dp->dp_throughput = throughput / 4 + @@ -614,8 +618,10 @@ dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) * the caller 1 clock tick. This will slow down the "fill" * rate until the sync process can catch up with us. */ - if (reserved && reserved > (write_limit - (write_limit >> 3))) - txg_delay(dp, tx->tx_txg, 1); + if (reserved && reserved > (write_limit - (write_limit >> 3))) { + txg_delay(dp, tx->tx_txg, zfs_throttle_delay, + zfs_throttle_resolution); + } return (0); } diff --git a/uts/common/fs/zfs/dsl_scan.c b/uts/common/fs/zfs/dsl_scan.c index 3de3c6e4d7f9..167f3825e5c6 100644 --- a/uts/common/fs/zfs/dsl_scan.c +++ b/uts/common/fs/zfs/dsl_scan.c @@ -403,7 +403,7 @@ dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_t *zb) zfs_resilver_min_time_ms : zfs_scan_min_time_ms; elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || - (elapsed_nanosecs / MICROSEC > mintime && + (NSEC2MSEC(elapsed_nanosecs) > mintime && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)) { if (zb) { @@ -1308,7 +1308,7 @@ dsl_scan_free_should_pause(dsl_scan_t *scn) elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || - (elapsed_nanosecs / MICROSEC > zfs_free_min_time_ms && + (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)); } @@ -1433,7 +1433,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) "free_bpobj/bptree txg %llu", (longlong_t)scn->scn_visited_this_txg, (longlong_t) - (gethrtime() - scn->scn_sync_start_time) / MICROSEC, + NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)tx->tx_txg); scn->scn_visited_this_txg = 0; /* @@ -1481,7 +1481,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) zfs_dbgmsg("visited %llu blocks in %llums", (longlong_t)scn->scn_visited_this_txg, - (longlong_t)(gethrtime() - scn->scn_sync_start_time) / MICROSEC); + (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time)); if (!scn->scn_pausing) { /* finished with scan. */ diff --git a/uts/common/fs/zfs/spa_misc.c b/uts/common/fs/zfs/spa_misc.c index 733d2609e55e..1663abbb5eac 100644 --- a/uts/common/fs/zfs/spa_misc.c +++ b/uts/common/fs/zfs/spa_misc.c @@ -499,8 +499,8 @@ spa_add(const char *name, nvlist_t *config, const char *altroot) hdlr.cyh_arg = spa; hdlr.cyh_level = CY_LOW_LEVEL; - spa->spa_deadman_synctime = zfs_deadman_synctime * - zfs_txg_synctime_ms * MICROSEC; + spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime * + zfs_txg_synctime_ms); /* * This determines how often we need to check for hung I/Os after @@ -508,7 +508,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot) * an expensive operation we don't want to check too frequently. * Instead wait for 5 synctimes before checking again. */ - when.cyt_interval = 5ULL * zfs_txg_synctime_ms * MICROSEC; + when.cyt_interval = MSEC2NSEC(5 * zfs_txg_synctime_ms); when.cyt_when = CY_INFINITY; mutex_enter(&cpu_lock); spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); diff --git a/uts/common/fs/zfs/sys/txg.h b/uts/common/fs/zfs/sys/txg.h index 2df33f0fb0a7..1529e5ac6da4 100644 --- a/uts/common/fs/zfs/sys/txg.h +++ b/uts/common/fs/zfs/sys/txg.h @@ -74,13 +74,8 @@ extern void txg_rele_to_quiesce(txg_handle_t *txghp); extern void txg_rele_to_sync(txg_handle_t *txghp); extern void txg_register_callbacks(txg_handle_t *txghp, list_t *tx_callbacks); -/* - * Delay the caller by the specified number of ticks or until - * the txg closes (whichever comes first). This is intended - * to be used to throttle writers when the system nears its - * capacity. - */ -extern void txg_delay(struct dsl_pool *dp, uint64_t txg, int ticks); +extern void txg_delay(struct dsl_pool *dp, uint64_t txg, hrtime_t delta, + hrtime_t resolution); /* * Wait until the given transaction group has finished syncing. diff --git a/uts/common/fs/zfs/sys/txg_impl.h b/uts/common/fs/zfs/sys/txg_impl.h index 7b356eac1293..8f1b21b3d460 100644 --- a/uts/common/fs/zfs/sys/txg_impl.h +++ b/uts/common/fs/zfs/sys/txg_impl.h @@ -23,6 +23,10 @@ * Use is subject to license terms. */ +/* + * Copyright (c) 2012 by Delphix. All rights reserved. + */ + #ifndef _SYS_TXG_IMPL_H #define _SYS_TXG_IMPL_H @@ -36,14 +40,14 @@ extern "C" { struct tx_cpu { kmutex_t tc_lock; kcondvar_t tc_cv[TXG_SIZE]; - uint64_t tc_count[TXG_SIZE]; + uint64_t tc_count[TXG_SIZE]; /* tx hold count on each txg */ list_t tc_callbacks[TXG_SIZE]; /* commit cb list */ - char tc_pad[16]; + char tc_pad[16]; /* pad to fill 3 cache lines */ }; typedef struct tx_state { - tx_cpu_t *tx_cpu; /* protects right to enter txg */ - kmutex_t tx_sync_lock; /* protects tx_state_t */ + tx_cpu_t *tx_cpu; /* protects access to tx_open_txg */ + kmutex_t tx_sync_lock; /* protects the rest of this struct */ uint64_t tx_open_txg; /* currently open txg id */ uint64_t tx_quiesced_txg; /* quiesced txg waiting for sync */ uint64_t tx_syncing_txg; /* currently syncing txg id */ diff --git a/uts/common/fs/zfs/txg.c b/uts/common/fs/zfs/txg.c index 58690e325fee..232cdd961edb 100644 --- a/uts/common/fs/zfs/txg.c +++ b/uts/common/fs/zfs/txg.c @@ -232,7 +232,7 @@ txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp) } static void -txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time) +txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, clock_t time) { CALLB_CPR_SAFE_BEGIN(cpr); @@ -353,6 +353,9 @@ txg_quiesce(dsl_pool_t *dp, uint64_t txg) ASSERT(txg == tx->tx_open_txg); tx->tx_open_txg++; + DTRACE_PROBE2(txg__quiescing, dsl_pool_t *, dp, uint64_t, txg); + DTRACE_PROBE2(txg__opened, dsl_pool_t *, dp, uint64_t, tx->tx_open_txg); + /* * Now that we've incremented tx_open_txg, we can let threads * enter the next transaction group. @@ -475,6 +478,7 @@ txg_sync_thread(dsl_pool_t *dp) txg = tx->tx_quiesced_txg; tx->tx_quiesced_txg = 0; tx->tx_syncing_txg = txg; + DTRACE_PROBE2(txg__syncing, dsl_pool_t *, dp, uint64_t, txg); cv_broadcast(&tx->tx_quiesce_more_cv); dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", @@ -488,6 +492,7 @@ txg_sync_thread(dsl_pool_t *dp) mutex_enter(&tx->tx_sync_lock); tx->tx_synced_txg = txg; tx->tx_syncing_txg = 0; + DTRACE_PROBE2(txg__synced, dsl_pool_t *, dp, uint64_t, txg); cv_broadcast(&tx->tx_sync_done_cv); /* @@ -536,21 +541,22 @@ txg_quiesce_thread(dsl_pool_t *dp) */ dprintf("quiesce done, handing off txg %llu\n", txg); tx->tx_quiesced_txg = txg; + DTRACE_PROBE2(txg__quiesced, dsl_pool_t *, dp, uint64_t, txg); cv_broadcast(&tx->tx_sync_more_cv); cv_broadcast(&tx->tx_quiesce_done_cv); } } /* - * Delay this thread by 'ticks' if we are still in the open transaction - * group and there is already a waiting txg quiesing or quiesced. Abort - * the delay if this txg stalls or enters the quiesing state. + * Delay this thread by delay nanoseconds if we are still in the open + * transaction group and there is already a waiting txg quiesing or quiesced. + * Abort the delay if this txg stalls or enters the quiesing state. */ void -txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks) +txg_delay(dsl_pool_t *dp, uint64_t txg, hrtime_t delay, hrtime_t resolution) { tx_state_t *tx = &dp->dp_tx; - clock_t timeout = ddi_get_lbolt() + ticks; + hrtime_t start = gethrtime(); /* don't delay if this txg could transition to quiesing immediately */ if (tx->tx_open_txg > txg || @@ -563,10 +569,11 @@ txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks) return; } - while (ddi_get_lbolt() < timeout && - tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) - (void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock, - timeout); + while (gethrtime() - start < delay && + tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) { + (void) cv_timedwait_hires(&tx->tx_quiesce_more_cv, + &tx->tx_sync_lock, delay, resolution, 0); + } mutex_exit(&tx->tx_sync_lock); } From c433297d86d2122d31cc566f4424c53aa22f9807 Mon Sep 17 00:00:00 2001 From: Martin Matuska Date: Tue, 5 Mar 2013 15:35:57 +0000 Subject: [PATCH 4/4] Update vendor/illumos/dist and vendor-sys/illumos/dist to illumos-gate 13977:af7f2603d4d7 Illumos ZFS issues: 3603 panic from bpobj_enqueue_subobj() 3604 zdb should print bpobjs more verbosely --- cmd/zdb/zdb.c | 70 +++++++++++++++++++++++++++++---------- uts/common/fs/zfs/bpobj.c | 8 ++++- uts/common/fs/zfs/dmu.c | 2 +- 3 files changed, 61 insertions(+), 19 deletions(-) diff --git a/cmd/zdb/zdb.c b/cmd/zdb/zdb.c index 46a630fb4230..a851977cc536 100644 --- a/cmd/zdb/zdb.c +++ b/cmd/zdb/zdb.c @@ -1222,7 +1222,7 @@ dump_bpobj_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) } static void -dump_bpobj(bpobj_t *bpo, char *name) +dump_bpobj(bpobj_t *bpo, char *name, int indent) { char bytes[32]; char comp[32]; @@ -1232,31 +1232,56 @@ dump_bpobj(bpobj_t *bpo, char *name) return; zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes); - if (bpo->bpo_havesubobj) { + if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) { zdb_nicenum(bpo->bpo_phys->bpo_comp, comp); zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp); - (void) printf("\n %s: %llu local blkptrs, %llu subobjs, " - "%s (%s/%s comp)\n", - name, (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, + (void) printf(" %*s: object %llu, %llu local blkptrs, " + "%llu subobjs, %s (%s/%s comp)\n", + indent * 8, name, + (u_longlong_t)bpo->bpo_object, + (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs, bytes, comp, uncomp); + + for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) { + uint64_t subobj; + bpobj_t subbpo; + int error; + VERIFY0(dmu_read(bpo->bpo_os, + bpo->bpo_phys->bpo_subobjs, + i * sizeof (subobj), sizeof (subobj), &subobj, 0)); + error = bpobj_open(&subbpo, bpo->bpo_os, subobj); + if (error != 0) { + (void) printf("ERROR %u while trying to open " + "subobj id %llu\n", + error, (u_longlong_t)subobj); + continue; + } + dump_bpobj(&subbpo, "subobj", indent + 1); + } } else { - (void) printf("\n %s: %llu blkptrs, %s\n", - name, (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, bytes); + (void) printf(" %*s: object %llu, %llu blkptrs, %s\n", + indent * 8, name, + (u_longlong_t)bpo->bpo_object, + (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, + bytes); } if (dump_opt['d'] < 5) return; - (void) printf("\n"); - (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL); + if (indent == 0) { + (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL); + (void) printf("\n"); + } } static void dump_deadlist(dsl_deadlist_t *dl) { dsl_deadlist_entry_t *dle; + uint64_t unused; char bytes[32]; char comp[32]; char uncomp[32]; @@ -1275,14 +1300,24 @@ dump_deadlist(dsl_deadlist_t *dl) (void) printf("\n"); + /* force the tree to be loaded */ + dsl_deadlist_space_range(dl, 0, UINT64_MAX, &unused, &unused, &unused); + for (dle = avl_first(&dl->dl_tree); dle; dle = AVL_NEXT(&dl->dl_tree, dle)) { - (void) printf(" mintxg %llu -> obj %llu\n", - (longlong_t)dle->dle_mintxg, - (longlong_t)dle->dle_bpobj.bpo_object); + if (dump_opt['d'] >= 5) { + char buf[128]; + (void) snprintf(buf, sizeof (buf), "mintxg %llu -> ", + (longlong_t)dle->dle_mintxg, + (longlong_t)dle->dle_bpobj.bpo_object); - if (dump_opt['d'] >= 5) - dump_bpobj(&dle->dle_bpobj, ""); + dump_bpobj(&dle->dle_bpobj, buf, 0); + } else { + (void) printf("mintxg %llu -> obj %llu\n", + (longlong_t)dle->dle_mintxg, + (longlong_t)dle->dle_bpobj.bpo_object); + + } } } @@ -1305,7 +1340,7 @@ fuid_table_destroy() * print uid or gid information. * For normal POSIX id just the id is printed in decimal format. * For CIFS files with FUID the fuid is printed in hex followed by - * the doman-rid string. + * the domain-rid string. */ static void print_idstr(uint64_t id, const char *id_type) @@ -2598,10 +2633,11 @@ dump_zpool(spa_t *spa) if (dump_opt['d'] || dump_opt['i']) { dump_dir(dp->dp_meta_objset); if (dump_opt['d'] >= 3) { - dump_bpobj(&spa->spa_deferred_bpobj, "Deferred frees"); + dump_bpobj(&spa->spa_deferred_bpobj, + "Deferred frees", 0); if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { dump_bpobj(&spa->spa_dsl_pool->dp_free_bpobj, - "Pool snapshot frees"); + "Pool snapshot frees", 0); } if (spa_feature_is_active(spa, diff --git a/uts/common/fs/zfs/bpobj.c b/uts/common/fs/zfs/bpobj.c index bcb5f331f0b3..1b9baa779d55 100644 --- a/uts/common/fs/zfs/bpobj.c +++ b/uts/common/fs/zfs/bpobj.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012 by Delphix. All rights reserved. + * Copyright (c) 2013 by Delphix. All rights reserved. */ #include @@ -418,6 +418,12 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx) VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs, 0, FTAG, &subdb, 0)); + /* + * Make sure that we are not asking dmu_write() + * to write more data than we have in our buffer. + */ + VERIFY3U(subdb->db_size, >=, + numsubsub * sizeof (subobj)); dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj), numsubsub * sizeof (subobj), subdb->db_data, tx); diff --git a/uts/common/fs/zfs/dmu.c b/uts/common/fs/zfs/dmu.c index 6ee37ac8b774..3e09873f2d90 100644 --- a/uts/common/fs/zfs/dmu.c +++ b/uts/common/fs/zfs/dmu.c @@ -1698,7 +1698,7 @@ dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) doi->doi_checksum = dn->dn_checksum; doi->doi_compress = dn->dn_compress; doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; - doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; + doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; doi->doi_fill_count = 0; for (int i = 0; i < dnp->dn_nblkptr; i++) doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;