From 0945ce02d2314be0ecc3766528cb9ef289bcd418 Mon Sep 17 00:00:00 2001 From: Xin LI Date: Sat, 8 Nov 2014 06:43:33 +0000 Subject: [PATCH] 5027 zfs large block support Reviewed by: Alek Pinchuk Reviewed by: George Wilson Reviewed by: Josef 'Jeff' Sipek Reviewed by: Richard Elling Reviewed by: Saso Kiselkov Reviewed by: Brian Behlendorf Approved by: Dan McDonald Author: Matthew Ahrens illumos/illumos-gate@b515258426fed6c7311fd3f1dea697cfbd4085c6 --- cmd/zdb/zdb.c | 32 +++++++- cmd/zfs/zfs_main.c | 11 ++- cmd/zstreamdump/zstreamdump.c | 19 ++++- cmd/ztest/ztest.c | 14 +++- common/zfs/zfeature_common.c | 12 ++- common/zfs/zfeature_common.h | 1 + common/zfs/zfs_prop.c | 4 +- common/zfs/zpool_prop.c | 2 + lib/libzfs/common/libzfs.h | 3 + lib/libzfs/common/libzfs_dataset.c | 38 +++++++--- lib/libzfs/common/libzfs_sendrecv.c | 15 ++-- lib/libzfs_core/common/libzfs_core.c | 6 ++ lib/libzfs_core/common/libzfs_core.h | 3 +- man/man1m/zfs.1m | 42 +++++++++-- man/man5/zpool-features.5 | 28 +++++-- uts/common/fs/zfs/bpobj.c | 5 +- uts/common/fs/zfs/bptree.c | 2 +- uts/common/fs/zfs/dbuf.c | 6 +- uts/common/fs/zfs/dmu_objset.c | 16 ++++ uts/common/fs/zfs/dmu_send.c | 83 ++++++++++++++++----- uts/common/fs/zfs/dmu_tx.c | 24 ++++-- uts/common/fs/zfs/dnode.c | 10 +-- uts/common/fs/zfs/dsl_dataset.c | 105 ++++++++++++++++++++++++++- uts/common/fs/zfs/dsl_deadlist.c | 8 +- uts/common/fs/zfs/dsl_destroy.c | 7 ++ uts/common/fs/zfs/dsl_pool.c | 4 +- uts/common/fs/zfs/metaslab.c | 2 +- uts/common/fs/zfs/sa.c | 6 +- uts/common/fs/zfs/spa.c | 26 ++++++- uts/common/fs/zfs/spa_history.c | 2 +- uts/common/fs/zfs/spa_misc.c | 9 +++ uts/common/fs/zfs/sys/dmu.h | 3 +- uts/common/fs/zfs/sys/dmu_objset.h | 1 + uts/common/fs/zfs/sys/dmu_send.h | 6 +- uts/common/fs/zfs/sys/dsl_dataset.h | 11 +++ uts/common/fs/zfs/sys/spa.h | 22 ++++-- uts/common/fs/zfs/sys/zap_impl.h | 3 +- uts/common/fs/zfs/sys/zfs_ioctl.h | 5 +- uts/common/fs/zfs/sys/zfs_znode.h | 2 - uts/common/fs/zfs/sys/zil.h | 1 - uts/common/fs/zfs/sys/zil_impl.h | 2 +- uts/common/fs/zfs/vdev.c | 6 +- uts/common/fs/zfs/vdev_queue.c | 2 +- uts/common/fs/zfs/vdev_raidz.c | 6 +- uts/common/fs/zfs/zap_micro.c | 16 ++-- uts/common/fs/zfs/zfs_ioctl.c | 54 ++++++++++++-- uts/common/fs/zfs/zfs_log.c | 2 +- uts/common/fs/zfs/zfs_vfsops.c | 9 +-- uts/common/fs/zfs/zfs_vnops.c | 8 +- uts/common/fs/zfs/zfs_znode.c | 8 +- uts/common/fs/zfs/zil.c | 11 ++- uts/common/fs/zfs/zio.c | 9 +-- uts/common/fs/zfs/zvol.c | 11 +-- uts/common/sys/fs/zfs.h | 1 + 54 files changed, 579 insertions(+), 165 deletions(-) diff --git a/cmd/zdb/zdb.c b/cmd/zdb/zdb.c index 3747e6106798..36c4a8515b36 100644 --- a/cmd/zdb/zdb.c +++ b/cmd/zdb/zdb.c @@ -2120,6 +2120,8 @@ dump_label(const char *dev) (void) close(fd); } +static uint64_t num_large_blocks; + /*ARGSUSED*/ static int dump_one_dir(const char *dsname, void *arg) @@ -2132,6 +2134,8 @@ dump_one_dir(const char *dsname, void *arg) (void) printf("Could not open %s, error %d\n", dsname, error); return (0); } + if (dmu_objset_ds(os)->ds_large_blocks) + num_large_blocks++; dump_dir(os); dmu_objset_disown(os, FTAG); fuid_table_destroy(); @@ -2142,7 +2146,7 @@ dump_one_dir(const char *dsname, void *arg) /* * Block statistics. */ -#define PSIZE_HISTO_SIZE (SPA_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1) +#define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2) typedef struct zdb_blkstats { uint64_t zb_asize; uint64_t zb_lsize; @@ -2207,7 +2211,15 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp, zb->zb_lsize += BP_GET_LSIZE(bp); zb->zb_psize += BP_GET_PSIZE(bp); zb->zb_count++; - zb->zb_psize_histogram[BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT]++; + + /* + * The histogram is only big enough to record blocks up to + * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last, + * "other", bucket. + */ + int idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT; + idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1); + zb->zb_psize_histogram[idx]++; zb->zb_gangs += BP_COUNT_GANG(bp); @@ -2917,6 +2929,7 @@ dump_zpool(spa_t *spa) dump_metaslab_groups(spa); if (dump_opt['d'] || dump_opt['i']) { + uint64_t refcount; dump_dir(dp->dp_meta_objset); if (dump_opt['d'] >= 3) { dump_bpobj(&spa->spa_deferred_bpobj, @@ -2936,8 +2949,21 @@ dump_zpool(spa_t *spa) } (void) dmu_objset_find(spa_name(spa), dump_one_dir, NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); + + (void) feature_get_refcount(spa, + &spa_feature_table[SPA_FEATURE_LARGE_BLOCKS], &refcount); + if (num_large_blocks != refcount) { + (void) printf("large_blocks feature refcount mismatch: " + "expected %lld != actual %lld\n", + (longlong_t)num_large_blocks, + (longlong_t)refcount); + rc = 2; + } else { + (void) printf("Verified large_blocks feature refcount " + "is correct (%llu)\n", (longlong_t)refcount); + } } - if (dump_opt['b'] || dump_opt['c']) + if (rc == 0 && (dump_opt['b'] || dump_opt['c'])) rc = dump_block_stats(spa); if (rc == 0) diff --git a/cmd/zfs/zfs_main.c b/cmd/zfs/zfs_main.c index 1c642b5e3dae..34ddad2c38b5 100644 --- a/cmd/zfs/zfs_main.c +++ b/cmd/zfs/zfs_main.c @@ -256,9 +256,9 @@ get_usage(zfs_help_t idx) case HELP_ROLLBACK: return (gettext("\trollback [-rRf] \n")); case HELP_SEND: - return (gettext("\tsend [-DnPpRve] [-[iI] snapshot] " + return (gettext("\tsend [-DnPpRvLe] [-[iI] snapshot] " "\n" - "\tsend [-e] [-i snapshot|bookmark] " + "\tsend [-Le] [-i snapshot|bookmark] " "\n")); case HELP_SET: return (gettext("\tset " @@ -3640,7 +3640,7 @@ zfs_do_send(int argc, char **argv) boolean_t extraverbose = B_FALSE; /* check options */ - while ((c = getopt(argc, argv, ":i:I:RDpvnPe")) != -1) { + while ((c = getopt(argc, argv, ":i:I:RDpvnPLe")) != -1) { switch (c) { case 'i': if (fromname) @@ -3675,6 +3675,9 @@ zfs_do_send(int argc, char **argv) case 'n': flags.dryrun = B_TRUE; break; + case 'L': + flags.largeblock = B_TRUE; + break; case 'e': flags.embed_data = B_TRUE; break; @@ -3731,6 +3734,8 @@ zfs_do_send(int argc, char **argv) if (zhp == NULL) return (1); + if (flags.largeblock) + lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK; if (flags.embed_data) lzc_flags |= LZC_SEND_FLAG_EMBED_DATA; diff --git a/cmd/zstreamdump/zstreamdump.c b/cmd/zstreamdump/zstreamdump.c index dce1cb3d765b..d99d8014f049 100644 --- a/cmd/zstreamdump/zstreamdump.c +++ b/cmd/zstreamdump/zstreamdump.c @@ -54,7 +54,6 @@ uint64_t total_stream_len = 0; FILE *send_stream = 0; boolean_t do_byteswap = B_FALSE; boolean_t do_cksum = B_TRUE; -#define INITIAL_BUFLEN (1<<20) static void usage(void) @@ -67,6 +66,18 @@ usage(void) exit(1); } +static void * +safe_malloc(size_t size) +{ + void *rv = malloc(size); + if (rv == NULL) { + (void) fprintf(stderr, "ERROR; failed to allocate %zu bytes\n", + size); + abort(); + } + return (rv); +} + /* * ssread - send stream read. * @@ -158,7 +169,7 @@ print_block(char *buf, int length) int main(int argc, char *argv[]) { - char *buf = malloc(INITIAL_BUFLEN); + char *buf = safe_malloc(SPA_MAXBLOCKSIZE); uint64_t drr_record_count[DRR_NUMTYPES] = { 0 }; uint64_t total_records = 0; dmu_replay_record_t thedrr; @@ -307,9 +318,9 @@ main(int argc, char *argv[]) nvlist_t *nv; int sz = drr->drr_payloadlen; - if (sz > INITIAL_BUFLEN) { + if (sz > SPA_MAXBLOCKSIZE) { free(buf); - buf = malloc(sz); + buf = safe_malloc(sz); } (void) ssread(buf, sz, &zc); if (ferror(send_stream)) diff --git a/cmd/ztest/ztest.c b/cmd/ztest/ztest.c index 6a29b2f32cab..a5f960080465 100644 --- a/cmd/ztest/ztest.c +++ b/cmd/ztest/ztest.c @@ -985,9 +985,15 @@ ztest_spa_get_ashift() { static int ztest_random_blocksize(void) { - // Choose a block size >= the ashift. - uint64_t block_shift = - ztest_random(SPA_MAXBLOCKSHIFT - ztest_spa_get_ashift() + 1); + uint64_t block_shift; + /* + * Choose a block size >= the ashift. + * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks. + */ + int maxbs = SPA_OLD_MAXBLOCKSHIFT; + if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE) + maxbs = 20; + block_shift = ztest_random(maxbs - ztest_spa_get_ashift() + 1); return (1 << (SPA_MINBLOCKSHIFT + block_shift)); } @@ -4787,7 +4793,7 @@ ztest_fault_inject(ztest_ds_t *zd, uint64_t id) char path0[MAXPATHLEN]; char pathrand[MAXPATHLEN]; size_t fsize; - int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ + int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ int iters = 1000; int maxfaults; int mirror_save; diff --git a/common/zfs/zfeature_common.c b/common/zfs/zfeature_common.c index 9b046ab07db1..3358e5eb87ba 100644 --- a/common/zfs/zfeature_common.c +++ b/common/zfs/zfeature_common.c @@ -57,7 +57,8 @@ valid_char(char c, boolean_t after_colon) { return ((c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') || - c == (after_colon ? '_' : '.')); + (after_colon && c == '_') || + (!after_colon && (c == '.' || c == '-'))); } /* @@ -221,4 +222,13 @@ zpool_feature_init(void) "com.delphix:embedded_data", "embedded_data", "Blocks which compress very well use even less space.", B_FALSE, B_TRUE, B_TRUE, NULL); + + static const spa_feature_t large_blocks_deps[] = { + SPA_FEATURE_EXTENSIBLE_DATASET, + SPA_FEATURE_NONE + }; + zfeature_register(SPA_FEATURE_LARGE_BLOCKS, + "org.open-zfs:large_blocks", "large_blocks", + "Support for blocks larger than 128KB.", B_FALSE, B_FALSE, B_FALSE, + large_blocks_deps); } diff --git a/common/zfs/zfeature_common.h b/common/zfs/zfeature_common.h index be2111be9184..3f54b392be42 100644 --- a/common/zfs/zfeature_common.h +++ b/common/zfs/zfeature_common.h @@ -51,6 +51,7 @@ typedef enum spa_feature { SPA_FEATURE_EMBEDDED_DATA, SPA_FEATURE_BOOKMARKS, SPA_FEATURE_FS_SS_LIMIT, + SPA_FEATURE_LARGE_BLOCKS, SPA_FEATURES } spa_feature_t; diff --git a/common/zfs/zfs_prop.c b/common/zfs/zfs_prop.c index 0b4927243fb6..e145b1c866e3 100644 --- a/common/zfs/zfs_prop.c +++ b/common/zfs/zfs_prop.c @@ -397,8 +397,8 @@ zfs_prop_init(void) /* inherit number properties */ zprop_register_number(ZFS_PROP_RECORDSIZE, "recordsize", - SPA_MAXBLOCKSIZE, PROP_INHERIT, - ZFS_TYPE_FILESYSTEM, "512 to 128k, power of 2", "RECSIZE"); + SPA_OLD_MAXBLOCKSIZE, PROP_INHERIT, + ZFS_TYPE_FILESYSTEM, "512 to 1M, power of 2", "RECSIZE"); /* hidden properties */ zprop_register_hidden(ZFS_PROP_CREATETXG, "createtxg", PROP_TYPE_NUMBER, diff --git a/common/zfs/zpool_prop.c b/common/zfs/zpool_prop.c index a400f821e2e2..4d906b02bc02 100644 --- a/common/zfs/zpool_prop.c +++ b/common/zfs/zpool_prop.c @@ -127,6 +127,8 @@ zpool_prop_init(void) /* hidden properties */ zprop_register_hidden(ZPOOL_PROP_NAME, "name", PROP_TYPE_STRING, PROP_READONLY, ZFS_TYPE_POOL, "NAME"); + zprop_register_hidden(ZPOOL_PROP_MAXBLOCKSIZE, "maxblocksize", + PROP_TYPE_NUMBER, PROP_READONLY, ZFS_TYPE_POOL, "MAXBLOCKSIZE"); } /* diff --git a/lib/libzfs/common/libzfs.h b/lib/libzfs/common/libzfs.h index 44b97062b75d..9aa5ba590564 100644 --- a/lib/libzfs/common/libzfs.h +++ b/lib/libzfs/common/libzfs.h @@ -591,6 +591,9 @@ typedef struct sendflags { /* show progress (ie. -v) */ boolean_t progress; + /* large blocks (>128K) are permitted */ + boolean_t largeblock; + /* WRITE_EMBEDDED records of type DATA are permitted */ boolean_t embed_data; } sendflags_t; diff --git a/lib/libzfs/common/libzfs_dataset.c b/lib/libzfs/common/libzfs_dataset.c index 1a61d2694abc..27c3cca8e82e 100644 --- a/lib/libzfs/common/libzfs_dataset.c +++ b/lib/libzfs/common/libzfs_dataset.c @@ -1048,21 +1048,36 @@ zfs_valid_proplist(libzfs_handle_t *hdl, zfs_type_t type, nvlist_t *nvl, break; } - case ZFS_PROP_RECORDSIZE: case ZFS_PROP_VOLBLOCKSIZE: - /* must be power of two within SPA_{MIN,MAX}BLOCKSIZE */ + case ZFS_PROP_RECORDSIZE: + { + int maxbs = SPA_MAXBLOCKSIZE; + if (zhp != NULL) { + maxbs = zpool_get_prop_int(zhp->zpool_hdl, + ZPOOL_PROP_MAXBLOCKSIZE, NULL); + } + /* + * Volumes are limited to a volblocksize of 128KB, + * because they typically service workloads with + * small random writes, which incur a large performance + * penalty with large blocks. + */ + if (prop == ZFS_PROP_VOLBLOCKSIZE) + maxbs = SPA_OLD_MAXBLOCKSIZE; + /* + * The value must be a power of two between + * SPA_MINBLOCKSIZE and maxbs. + */ if (intval < SPA_MINBLOCKSIZE || - intval > SPA_MAXBLOCKSIZE || !ISP2(intval)) { + intval > maxbs || !ISP2(intval)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, - "'%s' must be power of 2 from %u " - "to %uk"), propname, - (uint_t)SPA_MINBLOCKSIZE, - (uint_t)SPA_MAXBLOCKSIZE >> 10); + "'%s' must be power of 2 from 512B " + "to %uKB"), propname, maxbs >> 10); (void) zfs_error(hdl, EZFS_BADPROP, errbuf); goto error; } break; - + } case ZFS_PROP_MLSLABEL: { /* @@ -1437,7 +1452,8 @@ zfs_setprop_error(libzfs_handle_t *hdl, zfs_prop_t prop, int err, break; case ERANGE: - if (prop == ZFS_PROP_COMPRESSION) { + if (prop == ZFS_PROP_COMPRESSION || + prop == ZFS_PROP_RECORDSIZE) { (void) zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "property setting is not allowed on " "bootable datasets")); @@ -3136,9 +3152,7 @@ zfs_create(libzfs_handle_t *hdl, const char *path, zfs_type_t type, case EDOM: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "volume block size must be power of 2 from " - "%u to %uk"), - (uint_t)SPA_MINBLOCKSIZE, - (uint_t)SPA_MAXBLOCKSIZE >> 10); + "512B to 128KB")); return (zfs_error(hdl, EZFS_BADPROP, errbuf)); diff --git a/lib/libzfs/common/libzfs_sendrecv.c b/lib/libzfs/common/libzfs_sendrecv.c index 6697b5283150..c4944438aa2c 100644 --- a/lib/libzfs/common/libzfs_sendrecv.c +++ b/lib/libzfs/common/libzfs_sendrecv.c @@ -206,7 +206,7 @@ static void * cksummer(void *arg) { dedup_arg_t *dda = arg; - char *buf = malloc(1<<20); + char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); dmu_replay_record_t thedrr; dmu_replay_record_t *drr = &thedrr; struct drr_begin *drrb = &thedrr.drr_u.drr_begin; @@ -271,9 +271,9 @@ cksummer(void *arg) DMU_COMPOUNDSTREAM && drr->drr_payloadlen != 0) { int sz = drr->drr_payloadlen; - if (sz > 1<<20) { - free(buf); - buf = malloc(sz); + if (sz > SPA_MAXBLOCKSIZE) { + buf = zfs_realloc(dda->dedup_hdl, buf, + SPA_MAXBLOCKSIZE, sz); } (void) ssread(buf, sz, ofp); if (ferror(stdin)) @@ -806,7 +806,7 @@ typedef struct send_dump_data { char prevsnap[ZFS_MAXNAMELEN]; uint64_t prevsnap_obj; boolean_t seenfrom, seento, replicate, doall, fromorigin; - boolean_t verbose, dryrun, parsable, progress, embed_data; + boolean_t verbose, dryrun, parsable, progress, embed_data, large_block; int outfd; boolean_t err; nvlist_t *fss; @@ -1153,6 +1153,8 @@ dump_snapshot(zfs_handle_t *zhp, void *arg) } enum lzc_send_flags flags = 0; + if (sdd->large_block) + flags |= LZC_SEND_FLAG_LARGE_BLOCK; if (sdd->embed_data) flags |= LZC_SEND_FLAG_EMBED_DATA; @@ -1501,6 +1503,7 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, sdd.parsable = flags->parsable; sdd.progress = flags->progress; sdd.dryrun = flags->dryrun; + sdd.large_block = flags->largeblock; sdd.embed_data = flags->embed_data; sdd.filter_cb = filter_func; sdd.filter_cb_arg = cb_arg; @@ -2506,7 +2509,7 @@ static int recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) { dmu_replay_record_t *drr; - void *buf = malloc(1<<20); + void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, diff --git a/lib/libzfs_core/common/libzfs_core.c b/lib/libzfs_core/common/libzfs_core.c index 6f36568667b7..06221fab4ae6 100644 --- a/lib/libzfs_core/common/libzfs_core.c +++ b/lib/libzfs_core/common/libzfs_core.c @@ -455,6 +455,10 @@ lzc_get_holds(const char *snapname, nvlist_t **holdsp) * * "fd" is the file descriptor to write the send stream to. * + * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted + * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT + * records with drr_blksz > 128K. + * * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, * which the receiving system must support (as indicated by support @@ -471,6 +475,8 @@ lzc_send(const char *snapname, const char *from, int fd, fnvlist_add_int32(args, "fd", fd); if (from != NULL) fnvlist_add_string(args, "fromsnap", from); + if (flags & LZC_SEND_FLAG_LARGE_BLOCK) + fnvlist_add_boolean(args, "largeblockok"); if (flags & LZC_SEND_FLAG_EMBED_DATA) fnvlist_add_boolean(args, "embedok"); err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); diff --git a/lib/libzfs_core/common/libzfs_core.h b/lib/libzfs_core/common/libzfs_core.h index d7d767055d33..bdd6c951ee49 100644 --- a/lib/libzfs_core/common/libzfs_core.h +++ b/lib/libzfs_core/common/libzfs_core.h @@ -53,7 +53,8 @@ int lzc_release(nvlist_t *, nvlist_t **); int lzc_get_holds(const char *, nvlist_t **); enum lzc_send_flags { - LZC_SEND_FLAG_EMBED_DATA = 1 << 0 + LZC_SEND_FLAG_EMBED_DATA = 1 << 0, + LZC_SEND_FLAG_LARGE_BLOCK = 1 << 1 }; int lzc_send(const char *, const char *, int, enum lzc_send_flags); diff --git a/man/man1m/zfs.1m b/man/man1m/zfs.1m index 60cef0ae9509..4ab55f1a3ccd 100644 --- a/man/man1m/zfs.1m +++ b/man/man1m/zfs.1m @@ -176,12 +176,12 @@ zfs \- configures ZFS file systems .LP .nf -\fBzfs\fR \fBsend\fR [\fB-DnPpRve\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR +\fBzfs\fR \fBsend\fR [\fB-DnPpRveL\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR .fi .LP .nf -\fBzfs\fR \fBsend\fR [\fB-e\fR] [\fB-i \fIsnapshot\fR|\fIbookmark\fR]\fR \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR +\fBzfs\fR \fBsend\fR [\fB-eL\fR] [\fB-i \fIsnapshot\fR|\fIbookmark\fR]\fR \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR .fi .LP @@ -1244,7 +1244,9 @@ significant performance gains. Use of this property for general purpose file systems is strongly discouraged, and may adversely affect performance. .sp The size specified must be a power of two greater than or equal to 512 and less -than or equal to 128 Kbytes. +than or equal to 128 Kbytes. If the \fBlarge_blocks\fR feature is enabled +on the pool, the size may be up to 1 Mbyte. See \fBzpool-features\fR(5) +for details on ZFS feature flags. .sp Changing the file system's \fBrecordsize\fR affects only files created afterward; existing files are unaffected. @@ -2923,7 +2925,7 @@ See \fBzpool-features\fR(5) for details on ZFS feature flags and the .sp .ne 2 .na -\fBzfs send\fR [\fB-DnPpRve\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR +\fBzfs send\fR [\fB-DnPpRveL\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR .ad .sp .6 .RS 4n @@ -2996,6 +2998,21 @@ will be much better if the filesystem uses a dedup-capable checksum (eg. \fBsha256\fR). .RE +.sp +.ne 2 +.na +\fB\fB-L\fR\fR +.ad +.sp .6 +.RS 4n +Generate a stream which may contain blocks larger than 128KB. This flag +has no effect if the \fBlarge_blocks\fR pool feature is disabled, or if +the \fRrecordsize\fR property of this filesystem has never been set above +128KB. The receiving system must have the \fBlarge_blocks\fR pool feature +enabled as well. See \fBzpool-features\fR(5) for details on ZFS feature +flags and the \fBlarge_blocks\fR feature. +.RE + .sp .ne 2 .na @@ -3064,7 +3081,7 @@ on future versions of \fBZFS\fR. .sp .ne 2 .na -\fBzfs send\fR [\fB-e\fR] [\fB-i\fR \fIsnapshot\fR|\fIbookmark\fR] \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR +\fBzfs send\fR [\fB-eL\fR] [\fB-i\fR \fIsnapshot\fR|\fIbookmark\fR] \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR .ad .sp .6 .RS 4n @@ -3092,6 +3109,21 @@ be the origin snapshot, or an earlier snapshot in the origin's filesystem, or the origin's origin, etc. .RE +.sp +.ne 2 +.na +\fB\fB-L\fR\fR +.ad +.sp .6 +.RS 4n +Generate a stream which may contain blocks larger than 128KB. This flag +has no effect if the \fBlarge_blocks\fR pool feature is disabled, or if +the \fRrecordsize\fR property of this filesystem has never been set above +128KB. The receiving system must have the \fBlarge_blocks\fR pool feature +enabled as well. See \fBzpool-features\fR(5) for details on ZFS feature +flags and the \fBlarge_blocks\fR feature. +.RE + .sp .ne 2 .na diff --git a/man/man5/zpool-features.5 b/man/man5/zpool-features.5 index 24398124f0f7..fbefb37b8091 100644 --- a/man/man5/zpool-features.5 +++ b/man/man5/zpool-features.5 @@ -18,7 +18,6 @@ .SH NAME zpool\-features \- ZFS pool feature descriptions .SH DESCRIPTION -.sp .LP ZFS pool on\-disk format versions are specified via "features" which replace the old on\-disk format numbers (the last supported on\-disk format number is @@ -36,7 +35,6 @@ format of the pool is specified by the set of all features marked as \fBactive\fR on the pool. If the pool was created by another software version this set may include unsupported features. .SS "Identifying features" -.sp .LP Every feature has a guid of the form \fIcom.example:feature_name\fR. The reverse DNS name ensures that the feature's guid is unique across all ZFS @@ -51,7 +49,6 @@ name is the portion of its guid which follows the ':' (e.g. however a feature's short name may differ across ZFS implementations if following the convention would result in name conflicts. .SS "Feature states" -.sp .LP Features can be in one of three states: .sp @@ -97,7 +94,6 @@ cannot be disabled once they have been enabled. The state of supported features is exposed through pool properties of the form \fIfeature@short_name\fR. .SS "Read\-only compatibility" -.sp .LP Some features may make on\-disk format changes that do not interfere with other software's ability to read from the pool. These features are referred to as @@ -106,7 +102,6 @@ compatible, the pool can be imported in read\-only mode by setting the \fBreadonly\fR property during import (see \fBzpool\fR(1M) for details on importing pools). .SS "Unsupported features" -.sp .LP For each unsupported feature enabled on an imported pool a pool property named \fIunsupported@feature_guid\fR will indicate why the import was allowed @@ -133,13 +128,11 @@ read\-only mode. .RE .SS "Feature dependencies" -.sp .LP Some features depend on other features being enabled in order to function properly. Enabling a feature will automatically enable any features it depends on. .SH FEATURES -.sp .LP The following features are supported on this system: .sp @@ -430,5 +423,26 @@ never return to being \fBenabled\fR. .RE +.sp +.ne 2 +.na +\fB\fBlarge_blocks\fR\fR +.ad +.RS 4n +.TS +l l . +GUID org.open-zfs:large_block +READ\-ONLY COMPATIBLE no +DEPENDENCIES extensible_dataset +.TE + +The \fBlarge_block\fR feature allows the record size on a dataset to be +set larger than 128KB. + +This feature becomes \fBactive\fR once a \fBrecordsize\fR property has been +set larger than 128KB, and will return to being \fBenabled\fR once all +filesystems that have ever had their recordsize larger than 128KB are destroyed. +.RE + .SH "SEE ALSO" \fBzpool\fR(1M) diff --git a/uts/common/fs/zfs/bpobj.c b/uts/common/fs/zfs/bpobj.c index e75ae72f9e9c..da4d38a3a9cf 100644 --- a/uts/common/fs/zfs/bpobj.c +++ b/uts/common/fs/zfs/bpobj.c @@ -43,7 +43,7 @@ bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx) if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) { ASSERT0(dp->dp_empty_bpobj); dp->dp_empty_bpobj = - bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx); + bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx); VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, @@ -396,7 +396,8 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx) dmu_buf_will_dirty(bpo->bpo_dbuf, tx); if (bpo->bpo_phys->bpo_subobjs == 0) { bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os, - DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx); + DMU_OT_BPOBJ_SUBOBJ, SPA_OLD_MAXBLOCKSIZE, + DMU_OT_NONE, 0, tx); } dmu_object_info_t doi; diff --git a/uts/common/fs/zfs/bptree.c b/uts/common/fs/zfs/bptree.c index c724ed074103..5f7d76f0e2ad 100644 --- a/uts/common/fs/zfs/bptree.c +++ b/uts/common/fs/zfs/bptree.c @@ -65,7 +65,7 @@ bptree_alloc(objset_t *os, dmu_tx_t *tx) bptree_phys_t *bt; obj = dmu_object_alloc(os, DMU_OTN_UINT64_METADATA, - SPA_MAXBLOCKSIZE, DMU_OTN_UINT64_METADATA, + SPA_OLD_MAXBLOCKSIZE, DMU_OTN_UINT64_METADATA, sizeof (bptree_phys_t), tx); /* diff --git a/uts/common/fs/zfs/dbuf.c b/uts/common/fs/zfs/dbuf.c index 8d064daf25ce..a5816e12ccf0 100644 --- a/uts/common/fs/zfs/dbuf.c +++ b/uts/common/fs/zfs/dbuf.c @@ -2022,10 +2022,8 @@ dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx) return (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; - if (blksz > SPA_MAXBLOCKSIZE) - blksz = SPA_MAXBLOCKSIZE; - else - blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); + ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset))); + blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); DB_DNODE_ENTER(db); dn = DB_DNODE(db); diff --git a/uts/common/fs/zfs/dmu_objset.c b/uts/common/fs/zfs/dmu_objset.c index 73b8e056cc8c..e7aeed17fb9d 100644 --- a/uts/common/fs/zfs/dmu_objset.c +++ b/uts/common/fs/zfs/dmu_objset.c @@ -255,6 +255,14 @@ logbias_changed_cb(void *arg, uint64_t newval) zil_set_logbias(os->os_zil, newval); } +static void +recordsize_changed_cb(void *arg, uint64_t newval) +{ + objset_t *os = arg; + + os->os_recordsize = newval; +} + void dmu_objset_byteswap(void *buf, size_t size) { @@ -384,6 +392,11 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, ZFS_PROP_REDUNDANT_METADATA), redundant_metadata_changed_cb, os); } + if (err == 0) { + err = dsl_prop_register(ds, + zfs_prop_to_name(ZFS_PROP_RECORDSIZE), + recordsize_changed_cb, os); + } } if (err != 0) { VERIFY(arc_buf_remove_ref(os->os_phys_buf, @@ -642,6 +655,9 @@ dmu_objset_evict(objset_t *os) VERIFY0(dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA), redundant_metadata_changed_cb, os)); + VERIFY0(dsl_prop_unregister(ds, + zfs_prop_to_name(ZFS_PROP_RECORDSIZE), + recordsize_changed_cb, os)); } VERIFY0(dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), diff --git a/uts/common/fs/zfs/dmu_send.c b/uts/common/fs/zfs/dmu_send.c index fed1b86a1b1e..2c08e7075f66 100644 --- a/uts/common/fs/zfs/dmu_send.c +++ b/uts/common/fs/zfs/dmu_send.c @@ -206,11 +206,12 @@ dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, drrw->drr_offset = offset; drrw->drr_length = blksz; drrw->drr_toguid = dsp->dsa_toguid; - if (BP_IS_EMBEDDED(bp)) { + if (bp == NULL || BP_IS_EMBEDDED(bp)) { /* - * There's no pre-computed checksum of embedded BP's, so - * (like fletcher4-checkummed blocks) userland will have - * to compute a dedup-capable checksum itself. + * There's no pre-computed checksum for partial-block + * writes or embedded BP's, so (like + * fletcher4-checkummed blocks) userland will have to + * compute a dedup-capable checksum itself. */ drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; } else { @@ -372,6 +373,10 @@ dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) drro->drr_compress = dnp->dn_compress; drro->drr_toguid = dsp->dsa_toguid; + if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && + drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) + drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; + if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) return (SET_ERROR(EINTR)); @@ -491,6 +496,7 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, uint32_t aflags = ARC_WAIT; arc_buf_t *abuf; int blksz = BP_GET_LSIZE(bp); + uint64_t offset; ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); ASSERT0(zb->zb_level); @@ -511,8 +517,24 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, } } - err = dump_write(dsp, type, zb->zb_object, zb->zb_blkid * blksz, - blksz, bp, abuf->b_data); + offset = zb->zb_blkid * blksz; + + if (!(dsp->dsa_featureflags & + DMU_BACKUP_FEATURE_LARGE_BLOCKS) && + blksz > SPA_OLD_MAXBLOCKSIZE) { + char *buf = abuf->b_data; + while (blksz > 0 && err == 0) { + int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); + err = dump_write(dsp, type, zb->zb_object, + offset, n, NULL, buf); + offset += n; + buf += n; + blksz -= n; + } + } else { + err = dump_write(dsp, type, zb->zb_object, + offset, blksz, bp, abuf->b_data); + } (void) arc_buf_remove_ref(abuf, &abuf); } @@ -526,7 +548,7 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, static int dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok, - int outfd, vnode_t *vp, offset_t *off) + boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off) { objset_t *os; dmu_replay_record_t *drr; @@ -561,6 +583,8 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, } #endif + if (large_block_ok && ds->ds_large_blocks) + featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; if (embedok && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; @@ -656,7 +680,8 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds, int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, - boolean_t embedok, int outfd, vnode_t *vp, offset_t *off) + boolean_t embedok, boolean_t large_block_ok, + int outfd, vnode_t *vp, offset_t *off) { dsl_pool_t *dp; dsl_dataset_t *ds; @@ -690,18 +715,19 @@ dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, zb.zbm_guid = fromds->ds_phys->ds_guid; is_clone = (fromds->ds_dir != ds->ds_dir); dsl_dataset_rele(fromds, FTAG); - err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok, - outfd, vp, off); + err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, + embedok, large_block_ok, outfd, vp, off); } else { - err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok, - outfd, vp, off); + err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, + embedok, large_block_ok, outfd, vp, off); } dsl_dataset_rele(ds, FTAG); return (err); } int -dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, +dmu_send(const char *tosnap, const char *fromsnap, + boolean_t embedok, boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off) { dsl_pool_t *dp; @@ -768,11 +794,11 @@ dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, dsl_pool_rele(dp, FTAG); return (err); } - err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok, - outfd, vp, off); + err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, + embedok, large_block_ok, outfd, vp, off); } else { - err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok, - outfd, vp, off); + err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, + embedok, large_block_ok, outfd, vp, off); } if (owned) dsl_dataset_disown(ds, FTAG); @@ -972,6 +998,15 @@ dmu_recv_begin_check(void *arg, dmu_tx_t *tx) !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) return (SET_ERROR(ENOTSUP)); + /* + * The receiving code doesn't know how to translate large blocks + * to smaller ones, so the pool must have the LARGE_BLOCKS + * feature enabled if the stream has LARGE_BLOCKS. + */ + if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && + !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) + return (SET_ERROR(ENOTSUP)); + error = dsl_dataset_hold(dp, tofs, FTAG, &ds); if (error == 0) { /* target fs already exists; recv into temp clone */ @@ -1097,6 +1132,13 @@ dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) } VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); + if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & + DMU_BACKUP_FEATURE_LARGE_BLOCKS) && + !newds->ds_large_blocks) { + dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); + newds->ds_large_blocks = B_TRUE; + } + dmu_buf_will_dirty(newds->ds_dbuf, tx); newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; @@ -1222,6 +1264,7 @@ restore_read(struct restorearg *ra, int len, char *buf) /* some things will require 8-byte alignment, so everything must */ ASSERT0(len % 8); + ASSERT3U(len, <=, ra->bufsize); while (done < len) { ssize_t resid; @@ -1361,7 +1404,7 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || drro->drr_blksz < SPA_MINBLOCKSIZE || - drro->drr_blksz > SPA_MAXBLOCKSIZE || + drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) || drro->drr_bonuslen > DN_MAX_BONUSLEN) { return (SET_ERROR(EINVAL)); } @@ -1634,7 +1677,7 @@ restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs) int err; if (drrs->drr_length < SPA_MINBLOCKSIZE || - drrs->drr_length > SPA_MAXBLOCKSIZE) + drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os))) return (SET_ERROR(EINVAL)); data = restore_read(ra, drrs->drr_length, NULL); @@ -1721,7 +1764,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, ra.cksum = drc->drc_cksum; ra.vp = vp; ra.voff = *voffp; - ra.bufsize = 1<<20; + ra.bufsize = SPA_MAXBLOCKSIZE; ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP); /* these were verified in dmu_recv_begin */ diff --git a/uts/common/fs/zfs/dmu_tx.c b/uts/common/fs/zfs/dmu_tx.c index 55ce31eda85c..d249762e5a2c 100644 --- a/uts/common/fs/zfs/dmu_tx.c +++ b/uts/common/fs/zfs/dmu_tx.c @@ -224,7 +224,7 @@ dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) return; min_bs = SPA_MINBLOCKSHIFT; - max_bs = SPA_MAXBLOCKSHIFT; + max_bs = highbit64(txh->txh_tx->tx_objset->os_recordsize) - 1; min_ibs = DN_MIN_INDBLKSHIFT; max_ibs = DN_MAX_INDBLKSHIFT; @@ -293,6 +293,14 @@ dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) */ ASSERT(dn->dn_datablkshift != 0); min_bs = max_bs = dn->dn_datablkshift; + } else { + /* + * The blocksize can increase up to the recordsize, + * or if it is already more than the recordsize, + * up to the next power of 2. + */ + min_bs = highbit64(dn->dn_datablksz - 1); + max_bs = MAX(max_bs, highbit64(dn->dn_datablksz - 1)); } /* @@ -750,11 +758,11 @@ dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name) bp = &dn->dn_phys->dn_blkptr[0]; if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, bp, bp->blk_birth)) - txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; + txh->txh_space_tooverwrite += MZAP_MAX_BLKSZ; else - txh->txh_space_towrite += SPA_MAXBLOCKSIZE; + txh->txh_space_towrite += MZAP_MAX_BLKSZ; if (!BP_IS_HOLE(bp)) - txh->txh_space_tounref += SPA_MAXBLOCKSIZE; + txh->txh_space_tounref += MZAP_MAX_BLKSZ; return; } @@ -1543,18 +1551,18 @@ dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object) /* If blkptr doesn't exist then add space to towrite */ if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { - txh->txh_space_towrite += SPA_MAXBLOCKSIZE; + txh->txh_space_towrite += SPA_OLD_MAXBLOCKSIZE; } else { blkptr_t *bp; bp = &dn->dn_phys->dn_spill; if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, bp, bp->blk_birth)) - txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; + txh->txh_space_tooverwrite += SPA_OLD_MAXBLOCKSIZE; else - txh->txh_space_towrite += SPA_MAXBLOCKSIZE; + txh->txh_space_towrite += SPA_OLD_MAXBLOCKSIZE; if (!BP_IS_HOLE(bp)) - txh->txh_space_tounref += SPA_MAXBLOCKSIZE; + txh->txh_space_tounref += SPA_OLD_MAXBLOCKSIZE; } } diff --git a/uts/common/fs/zfs/dnode.c b/uts/common/fs/zfs/dnode.c index 175157714d06..9c70c2bfb1b0 100644 --- a/uts/common/fs/zfs/dnode.c +++ b/uts/common/fs/zfs/dnode.c @@ -510,10 +510,10 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, { int i; + ASSERT3U(blocksize, <=, + spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); if (blocksize == 0) blocksize = 1 << zfs_default_bs; - else if (blocksize > SPA_MAXBLOCKSIZE) - blocksize = SPA_MAXBLOCKSIZE; else blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); @@ -594,7 +594,8 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int nblkptr; ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); - ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); + ASSERT3U(blocksize, <=, + spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); ASSERT0(blocksize % SPA_MINBLOCKSIZE); ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); ASSERT(tx->tx_txg != 0); @@ -1347,10 +1348,9 @@ dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) dmu_buf_impl_t *db; int err; + ASSERT3U(size, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); if (size == 0) size = SPA_MINBLOCKSIZE; - if (size > SPA_MAXBLOCKSIZE) - size = SPA_MAXBLOCKSIZE; else size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); diff --git a/uts/common/fs/zfs/dsl_dataset.c b/uts/common/fs/zfs/dsl_dataset.c index f1b92f3eaa33..e7ed75090208 100644 --- a/uts/common/fs/zfs/dsl_dataset.c +++ b/uts/common/fs/zfs/dsl_dataset.c @@ -50,6 +50,17 @@ #include #include +/* + * The SPA supports block sizes up to 16MB. However, very large blocks + * can have an impact on i/o latency (e.g. tying up a spinning disk for + * ~300ms), and also potentially on the memory allocator. Therefore, + * we do not allow the recordsize to be set larger than zfs_max_recordsize + * (default 1MB). Larger blocks can be created by changing this tunable, + * and pools with larger blocks can always be imported and used, regardless + * of this setting. + */ +int zfs_max_recordsize = 1 * 1024 * 1024; + #define SWITCH64(x, y) \ { \ uint64_t __tmp = (x); \ @@ -59,8 +70,6 @@ #define DS_REF_MAX (1ULL << 62) -#define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE - /* * 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 @@ -110,6 +119,8 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) ds->ds_phys->ds_compressed_bytes += compressed; ds->ds_phys->ds_uncompressed_bytes += uncompressed; ds->ds_phys->ds_unique_bytes += used; + if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) + ds->ds_need_large_blocks = B_TRUE; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, compressed, uncompressed, tx); @@ -387,6 +398,14 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), offsetof(dmu_sendarg_t, dsa_link)); + if (doi.doi_type == DMU_OTN_ZAP_METADATA) { + err = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS); + if (err == 0) + ds->ds_large_blocks = B_TRUE; + else + ASSERT3U(err, ==, ENOENT); + } + if (err == 0) { err = dsl_dir_hold_obj(dp, ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); @@ -700,6 +719,9 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, dsphys->ds_flags |= origin->ds_phys->ds_flags & (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); + if (origin->ds_large_blocks) + dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); + dmu_buf_will_dirty(origin->ds_dbuf, tx); origin->ds_phys->ds_num_children++; @@ -1213,6 +1235,9 @@ dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, dsphys->ds_bp = ds->ds_phys->ds_bp; dmu_buf_rele(dbuf, FTAG); + if (ds->ds_large_blocks) + dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); + ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); if (ds->ds_prev) { uint64_t next_clones_obj = @@ -1486,6 +1511,11 @@ dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; dmu_objset_sync(ds->ds_objset, zio, tx); + + if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { + dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); + ds->ds_large_blocks = B_TRUE; + } } static void @@ -3128,6 +3158,77 @@ dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, return (err); } +static int +dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) +{ + const char *dsname = arg; + dsl_dataset_t *ds; + dsl_pool_t *dp = dmu_tx_pool(tx); + int error = 0; + + if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) + return (SET_ERROR(ENOTSUP)); + + ASSERT(spa_feature_is_enabled(dp->dp_spa, + SPA_FEATURE_EXTENSIBLE_DATASET)); + + error = dsl_dataset_hold(dp, dsname, FTAG, &ds); + if (error != 0) + return (error); + + if (ds->ds_large_blocks) + error = EALREADY; + dsl_dataset_rele(ds, FTAG); + + return (error); +} + +void +dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) +{ + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; + uint64_t zero = 0; + + spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); + dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); + + VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, + sizeof (zero), 1, &zero, tx)); +} + +static void +dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) +{ + const char *dsname = arg; + dsl_dataset_t *ds; + + VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); + + dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); + ASSERT(!ds->ds_large_blocks); + ds->ds_large_blocks = B_TRUE; + dsl_dataset_rele(ds, FTAG); +} + +int +dsl_dataset_activate_large_blocks(const char *dsname) +{ + int error; + + error = dsl_sync_task(dsname, + dsl_dataset_activate_large_blocks_check, + dsl_dataset_activate_large_blocks_sync, (void *)dsname, + 1, ZFS_SPACE_CHECK_RESERVED); + + /* + * EALREADY indicates that this dataset already supports large blocks. + */ + if (error == EALREADY) + error = 0; + return (error); +} + /* * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. * For example, they could both be snapshots of the same filesystem, and diff --git a/uts/common/fs/zfs/dsl_deadlist.c b/uts/common/fs/zfs/dsl_deadlist.c index 4f39c397a064..8c8e3746eecb 100644 --- a/uts/common/fs/zfs/dsl_deadlist.c +++ b/uts/common/fs/zfs/dsl_deadlist.c @@ -143,7 +143,7 @@ uint64_t dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx) { if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS) - return (bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx)); + return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx)); return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR, sizeof (dsl_deadlist_phys_t), tx)); } @@ -180,7 +180,7 @@ dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle, { if (dle->dle_bpobj.bpo_object == dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) { - uint64_t obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + uint64_t obj = bpobj_alloc(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx); bpobj_close(&dle->dle_bpobj); bpobj_decr_empty(dl->dl_os, tx); VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj)); @@ -254,7 +254,7 @@ dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx) dle = kmem_alloc(sizeof (*dle), KM_SLEEP); dle->dle_mintxg = mintxg; - obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx); VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj)); avl_add(&dl->dl_tree, dle); @@ -338,7 +338,7 @@ dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg, if (dle->dle_mintxg >= maxtxg) break; - obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx); VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj, dle->dle_mintxg, obj, tx)); } diff --git a/uts/common/fs/zfs/dsl_destroy.c b/uts/common/fs/zfs/dsl_destroy.c index f8a4546535e4..1237641583a1 100644 --- a/uts/common/fs/zfs/dsl_destroy.c +++ b/uts/common/fs/zfs/dsl_destroy.c @@ -264,6 +264,10 @@ dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx) obj = ds->ds_object; + if (ds->ds_large_blocks) { + ASSERT0(zap_contains(mos, obj, DS_FIELD_LARGE_BLOCKS)); + spa_feature_decr(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS, tx); + } if (ds->ds_phys->ds_prev_snap_obj != 0) { ASSERT3P(ds->ds_prev, ==, NULL); VERIFY0(dsl_dataset_hold_obj(dp, @@ -720,6 +724,9 @@ dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx) ASSERT0(ds->ds_reserved); } + if (ds->ds_large_blocks) + spa_feature_decr(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS, tx); + dsl_scan_ds_destroyed(ds, tx); obj = ds->ds_object; diff --git a/uts/common/fs/zfs/dsl_pool.c b/uts/common/fs/zfs/dsl_pool.c index ad26a9070b68..1d246b8e688d 100644 --- a/uts/common/fs/zfs/dsl_pool.c +++ b/uts/common/fs/zfs/dsl_pool.c @@ -367,7 +367,7 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) FREE_DIR_NAME, &dp->dp_free_dir)); /* create and open the free_bplist */ - obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); + obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); VERIFY0(bpobj_open(&dp->dp_free_bpobj, @@ -792,7 +792,7 @@ dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) * subobj support. So call dmu_object_alloc() directly. */ obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, - SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); + SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); diff --git a/uts/common/fs/zfs/metaslab.c b/uts/common/fs/zfs/metaslab.c index a33ec7f62810..0f05e8b00a87 100644 --- a/uts/common/fs/zfs/metaslab.c +++ b/uts/common/fs/zfs/metaslab.c @@ -125,7 +125,7 @@ int metaslab_debug_unload = 0; * an allocation of this size then it switches to using more * aggressive strategy (i.e search by size rather than offset). */ -uint64_t metaslab_df_alloc_threshold = SPA_MAXBLOCKSIZE; +uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE; /* * The minimum free space, in percent, which must be available diff --git a/uts/common/fs/zfs/sa.c b/uts/common/fs/zfs/sa.c index bd2ebce5df34..8b3963aed9f5 100644 --- a/uts/common/fs/zfs/sa.c +++ b/uts/common/fs/zfs/sa.c @@ -500,7 +500,7 @@ sa_resize_spill(sa_handle_t *hdl, uint32_t size, dmu_tx_t *tx) if (size == 0) { blocksize = SPA_MINBLOCKSIZE; - } else if (size > SPA_MAXBLOCKSIZE) { + } else if (size > SPA_OLD_MAXBLOCKSIZE) { ASSERT(0); return (SET_ERROR(EFBIG)); } else { @@ -675,7 +675,7 @@ sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count, hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus, SA_BONUS, &i, &used, &spilling); - if (used > SPA_MAXBLOCKSIZE) + if (used > SPA_OLD_MAXBLOCKSIZE) return (SET_ERROR(EFBIG)); VERIFY(0 == dmu_set_bonus(hdl->sa_bonus, spilling ? @@ -699,7 +699,7 @@ sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count, attr_count - i, hdl->sa_spill, SA_SPILL, &i, &spill_used, &dummy); - if (spill_used > SPA_MAXBLOCKSIZE) + if (spill_used > SPA_OLD_MAXBLOCKSIZE) return (SET_ERROR(EFBIG)); buf_space = hdl->sa_spill->db_size - spillhdrsize; diff --git a/uts/common/fs/zfs/spa.c b/uts/common/fs/zfs/spa.c index be308e2f8730..634967c46f1b 100644 --- a/uts/common/fs/zfs/spa.c +++ b/uts/common/fs/zfs/spa.c @@ -267,6 +267,14 @@ spa_prop_get_config(spa_t *spa, nvlist_t **nvp) spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 0, ZPROP_SRC_LOCAL); + if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { + spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, + MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); + } else { + spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, + SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); + } + if ((dp = list_head(&spa->spa_config_list)) != NULL) { if (dp->scd_path == NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, @@ -481,7 +489,7 @@ spa_prop_validate(spa_t *spa, nvlist_t *props) if (!error) { objset_t *os; - uint64_t compress; + uint64_t propval; if (strval == NULL || strval[0] == '\0') { objnum = zpool_prop_default_numeric( @@ -492,15 +500,25 @@ spa_prop_validate(spa_t *spa, nvlist_t *props) if (error = dmu_objset_hold(strval, FTAG, &os)) break; - /* Must be ZPL and not gzip compressed. */ + /* + * Must be ZPL, and its property settings + * must be supported by GRUB (compression + * is not gzip, and large blocks are not used). + */ if (dmu_objset_type(os) != DMU_OST_ZFS) { error = SET_ERROR(ENOTSUP); } else if ((error = dsl_prop_get_int_ds(dmu_objset_ds(os), zfs_prop_to_name(ZFS_PROP_COMPRESSION), - &compress)) == 0 && - !BOOTFS_COMPRESS_VALID(compress)) { + &propval)) == 0 && + !BOOTFS_COMPRESS_VALID(propval)) { + error = SET_ERROR(ENOTSUP); + } else if ((error = + dsl_prop_get_int_ds(dmu_objset_ds(os), + zfs_prop_to_name(ZFS_PROP_RECORDSIZE), + &propval)) == 0 && + propval > SPA_OLD_MAXBLOCKSIZE) { error = SET_ERROR(ENOTSUP); } else { objnum = dmu_objset_id(os); diff --git a/uts/common/fs/zfs/spa_history.c b/uts/common/fs/zfs/spa_history.c index cf72c8ad8827..ce64f70b28c3 100644 --- a/uts/common/fs/zfs/spa_history.c +++ b/uts/common/fs/zfs/spa_history.c @@ -90,7 +90,7 @@ spa_history_create_obj(spa_t *spa, dmu_tx_t *tx) ASSERT(spa->spa_history == 0); spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY, - SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS, + SPA_OLD_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS, sizeof (spa_history_phys_t), tx); VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, diff --git a/uts/common/fs/zfs/spa_misc.c b/uts/common/fs/zfs/spa_misc.c index 050436272696..1729ba0503be 100644 --- a/uts/common/fs/zfs/spa_misc.c +++ b/uts/common/fs/zfs/spa_misc.c @@ -1963,3 +1963,12 @@ spa_debug_enabled(spa_t *spa) { return (spa->spa_debug); } + +int +spa_maxblocksize(spa_t *spa) +{ + if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) + return (SPA_MAXBLOCKSIZE); + else + return (SPA_OLD_MAXBLOCKSIZE); +} diff --git a/uts/common/fs/zfs/sys/dmu.h b/uts/common/fs/zfs/sys/dmu.h index e9fa39d5d6f1..93165a95951d 100644 --- a/uts/common/fs/zfs/sys/dmu.h +++ b/uts/common/fs/zfs/sys/dmu.h @@ -249,7 +249,7 @@ void zfs_znode_byteswap(void *buf, size_t size); * The maximum number of bytes that can be accessed as part of one * operation, including metadata. */ -#define DMU_MAX_ACCESS (10<<20) /* 10MB */ +#define DMU_MAX_ACCESS (32 * 1024 * 1024) /* 32MB */ #define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */ #define DMU_USERUSED_OBJECT (-1ULL) @@ -637,6 +637,7 @@ void xuio_stat_wbuf_copied(); void xuio_stat_wbuf_nocopy(); extern int zfs_prefetch_disable; +extern int zfs_max_recordsize; /* * Asynchronously try to read in the data. diff --git a/uts/common/fs/zfs/sys/dmu_objset.h b/uts/common/fs/zfs/sys/dmu_objset.h index 23d88fd048bd..804f0c182b6f 100644 --- a/uts/common/fs/zfs/sys/dmu_objset.h +++ b/uts/common/fs/zfs/sys/dmu_objset.h @@ -95,6 +95,7 @@ struct objset { zfs_cache_type_t os_secondary_cache; zfs_sync_type_t os_sync; zfs_redundant_metadata_type_t os_redundant_metadata; + int os_recordsize; /* no lock needed: */ struct dmu_tx *os_synctx; /* XXX sketchy */ diff --git a/uts/common/fs/zfs/sys/dmu_send.h b/uts/common/fs/zfs/sys/dmu_send.h index dc183c02c350..3a8dc89abd4a 100644 --- a/uts/common/fs/zfs/sys/dmu_send.h +++ b/uts/common/fs/zfs/sys/dmu_send.h @@ -37,12 +37,14 @@ struct dsl_dataset; struct drr_begin; struct avl_tree; -int dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, +int dmu_send(const char *tosnap, const char *fromsnap, + boolean_t embedok, boolean_t large_block_ok, 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, - boolean_t embedok, int outfd, vnode_t *vp, offset_t *off); + boolean_t embedok, boolean_t large_block_ok, + int outfd, struct vnode *vp, offset_t *off); typedef struct dmu_recv_cookie { struct dsl_dataset *drc_ds; diff --git a/uts/common/fs/zfs/sys/dsl_dataset.h b/uts/common/fs/zfs/sys/dsl_dataset.h index d9552b2260af..ff90f8b439cc 100644 --- a/uts/common/fs/zfs/sys/dsl_dataset.h +++ b/uts/common/fs/zfs/sys/dsl_dataset.h @@ -82,6 +82,13 @@ struct dsl_pool; */ #define DS_FIELD_BOOKMARK_NAMES "com.delphix:bookmarks" +/* + * This field is present (with value=0) if this dataset may contain large + * blocks (>128KB). If it is present, then this dataset + * is counted in the refcount of the SPA_FEATURE_LARGE_BLOCKS feature. + */ +#define DS_FIELD_LARGE_BLOCKS "org.open-zfs:large_blocks" + /* * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose * name lookups should be performed case-insensitively. @@ -135,6 +142,8 @@ typedef struct dsl_dataset { /* only used in syncing context, only valid for non-snapshots: */ struct dsl_dataset *ds_prev; uint64_t ds_bookmarks; /* DMU_OTN_ZAP_METADATA */ + boolean_t ds_large_blocks; + boolean_t ds_need_large_blocks; /* has internal locking: */ dsl_deadlist_t ds_deadlist; @@ -244,6 +253,8 @@ int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *last, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp); boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds); +int dsl_dataset_activate_large_blocks(const char *dsname); +void dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx); int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf); diff --git a/uts/common/fs/zfs/sys/spa.h b/uts/common/fs/zfs/sys/spa.h index 6c7d34cb8235..e4731ae5a739 100644 --- a/uts/common/fs/zfs/sys/spa.h +++ b/uts/common/fs/zfs/sys/spa.h @@ -94,17 +94,26 @@ _NOTE(CONSTCOND) } while (0) _NOTE(CONSTCOND) } while (0) /* - * We currently support nine block sizes, from 512 bytes to 128K. - * We could go higher, but the benefits are near-zero and the cost - * of COWing a giant block to modify one byte would become excessive. + * We currently support block sizes from 512 bytes to 16MB. + * The benefits of larger blocks, and thus larger IO, need to be weighed + * against the cost of COWing a giant block to modify one byte, and the + * large latency of reading or writing a large block. + * + * Note that although blocks up to 16MB are supported, the recordsize + * property can not be set larger than zfs_max_recordsize (default 1MB). + * See the comment near zfs_max_recordsize in dsl_dataset.c for details. + * + * Note that although the LSIZE field of the blkptr_t can store sizes up + * to 32MB, the dnode's dn_datablkszsec can only store sizes up to + * 32MB - 512 bytes. Therefore, we limit SPA_MAXBLOCKSIZE to 16MB. */ #define SPA_MINBLOCKSHIFT 9 -#define SPA_MAXBLOCKSHIFT 17 +#define SPA_OLD_MAXBLOCKSHIFT 17 +#define SPA_MAXBLOCKSHIFT 24 #define SPA_MINBLOCKSIZE (1ULL << SPA_MINBLOCKSHIFT) +#define SPA_OLD_MAXBLOCKSIZE (1ULL << SPA_OLD_MAXBLOCKSHIFT) #define SPA_MAXBLOCKSIZE (1ULL << SPA_MAXBLOCKSHIFT) -#define SPA_BLOCKSIZES (SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1) - /* * Size of block to hold the configuration data (a packed nvlist) */ @@ -781,6 +790,7 @@ extern boolean_t spa_has_slogs(spa_t *spa); extern boolean_t spa_is_root(spa_t *spa); extern boolean_t spa_writeable(spa_t *spa); extern boolean_t spa_has_pending_synctask(spa_t *spa); +extern int spa_maxblocksize(spa_t *spa); extern int spa_mode(spa_t *spa); extern uint64_t strtonum(const char *str, char **nptr); diff --git a/uts/common/fs/zfs/sys/zap_impl.h b/uts/common/fs/zfs/sys/zap_impl.h index 466aab02bacb..8b4a8b2b56ed 100644 --- a/uts/common/fs/zfs/sys/zap_impl.h +++ b/uts/common/fs/zfs/sys/zap_impl.h @@ -42,8 +42,7 @@ extern int fzap_default_block_shift; #define MZAP_ENT_LEN 64 #define MZAP_NAME_LEN (MZAP_ENT_LEN - 8 - 4 - 2) -#define MZAP_MAX_BLKSHIFT SPA_MAXBLOCKSHIFT -#define MZAP_MAX_BLKSZ (1 << MZAP_MAX_BLKSHIFT) +#define MZAP_MAX_BLKSZ SPA_OLD_MAXBLOCKSIZE #define ZAP_NEED_CD (-1U) diff --git a/uts/common/fs/zfs/sys/zfs_ioctl.h b/uts/common/fs/zfs/sys/zfs_ioctl.h index bf9f83c37672..62f6ff997d16 100644 --- a/uts/common/fs/zfs/sys/zfs_ioctl.h +++ b/uts/common/fs/zfs/sys/zfs_ioctl.h @@ -85,13 +85,16 @@ typedef enum drr_headertype { /* flags #3 - #15 are reserved for incompatible closed-source implementations */ #define DMU_BACKUP_FEATURE_EMBED_DATA (1<<16) #define DMU_BACKUP_FEATURE_EMBED_DATA_LZ4 (1<<17) +/* flag #18 is reserved for a Delphix feature */ +#define DMU_BACKUP_FEATURE_LARGE_BLOCKS (1<<19) /* * Mask of all supported backup features */ #define DMU_BACKUP_FEATURE_MASK (DMU_BACKUP_FEATURE_DEDUP | \ DMU_BACKUP_FEATURE_DEDUPPROPS | DMU_BACKUP_FEATURE_SA_SPILL | \ - DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) + DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_EMBED_DATA_LZ4 | \ + DMU_BACKUP_FEATURE_LARGE_BLOCKS) /* Are all features in the given flag word currently supported? */ #define DMU_STREAM_SUPPORTED(x) (!((x) & ~DMU_BACKUP_FEATURE_MASK)) diff --git a/uts/common/fs/zfs/sys/zfs_znode.h b/uts/common/fs/zfs/sys/zfs_znode.h index bc389b4a4366..df08aad7b41e 100644 --- a/uts/common/fs/zfs/sys/zfs_znode.h +++ b/uts/common/fs/zfs/sys/zfs_znode.h @@ -136,8 +136,6 @@ extern "C" { #define ZFS_SHARES_DIR "SHARES" #define ZFS_SA_ATTRS "SA_ATTRS" -#define ZFS_MAX_BLOCKSIZE (SPA_MAXBLOCKSIZE) - /* * Path component length * diff --git a/uts/common/fs/zfs/sys/zil.h b/uts/common/fs/zfs/sys/zil.h index 15ef2aa8bf98..895d632a2620 100644 --- a/uts/common/fs/zfs/sys/zil.h +++ b/uts/common/fs/zfs/sys/zil.h @@ -90,7 +90,6 @@ typedef struct zil_chain { } zil_chain_t; #define ZIL_MIN_BLKSZ 4096ULL -#define ZIL_MAX_BLKSZ SPA_MAXBLOCKSIZE /* * The words of a log block checksum. diff --git a/uts/common/fs/zfs/sys/zil_impl.h b/uts/common/fs/zfs/sys/zil_impl.h index 58566203b697..b5c666c02b73 100644 --- a/uts/common/fs/zfs/sys/zil_impl.h +++ b/uts/common/fs/zfs/sys/zil_impl.h @@ -139,7 +139,7 @@ typedef struct zil_bp_node { avl_node_t zn_node; } zil_bp_node_t; -#define ZIL_MAX_LOG_DATA (SPA_MAXBLOCKSIZE - sizeof (zil_chain_t) - \ +#define ZIL_MAX_LOG_DATA (SPA_OLD_MAXBLOCKSIZE - sizeof (zil_chain_t) - \ sizeof (lr_write_t)) #ifdef __cplusplus diff --git a/uts/common/fs/zfs/vdev.c b/uts/common/fs/zfs/vdev.c index 7571b21a5f8c..67b58edeed0b 100644 --- a/uts/common/fs/zfs/vdev.c +++ b/uts/common/fs/zfs/vdev.c @@ -828,9 +828,9 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg) /* * Compute the raidz-deflation ratio. Note, we hard-code - * in 128k (1 << 17) because it is the current "typical" blocksize. - * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, - * or we will inconsistently account for existing bp's. + * in 128k (1 << 17) because it is the "typical" blocksize. + * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, + * otherwise it would inconsistently account for existing bp's. */ vd->vdev_deflate_ratio = (1 << 17) / (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); diff --git a/uts/common/fs/zfs/vdev_queue.c b/uts/common/fs/zfs/vdev_queue.c index 561e6a4482eb..f47c4cd1e26d 100644 --- a/uts/common/fs/zfs/vdev_queue.c +++ b/uts/common/fs/zfs/vdev_queue.c @@ -162,7 +162,7 @@ int zfs_vdev_async_write_active_max_dirty_percent = 60; * we include spans of optional I/Os to aid aggregation at the disk even when * they aren't able to help us aggregate at this level. */ -int zfs_vdev_aggregation_limit = SPA_MAXBLOCKSIZE; +int zfs_vdev_aggregation_limit = SPA_OLD_MAXBLOCKSIZE; int zfs_vdev_read_gap_limit = 32 << 10; int zfs_vdev_write_gap_limit = 4 << 10; diff --git a/uts/common/fs/zfs/vdev_raidz.c b/uts/common/fs/zfs/vdev_raidz.c index f686d56e9e5f..085d1250a1ad 100644 --- a/uts/common/fs/zfs/vdev_raidz.c +++ b/uts/common/fs/zfs/vdev_raidz.c @@ -1604,7 +1604,7 @@ vdev_raidz_physio(vdev_t *vd, caddr_t data, size_t size, /* * Don't write past the end of the block */ - VERIFY3U(offset + size, <=, origoffset + SPA_MAXBLOCKSIZE); + VERIFY3U(offset + size, <=, origoffset + SPA_OLD_MAXBLOCKSIZE); start = offset; end = start + size; @@ -1619,8 +1619,8 @@ vdev_raidz_physio(vdev_t *vd, caddr_t data, size_t size, * KB size. */ rm = vdev_raidz_map_alloc(data - (offset - origoffset), - SPA_MAXBLOCKSIZE, origoffset, tvd->vdev_ashift, vd->vdev_children, - vd->vdev_nparity); + SPA_OLD_MAXBLOCKSIZE, origoffset, tvd->vdev_ashift, + vd->vdev_children, vd->vdev_nparity); coloffset = origoffset; diff --git a/uts/common/fs/zfs/zap_micro.c b/uts/common/fs/zfs/zap_micro.c index 1152f9072f99..59a9f970448a 100644 --- a/uts/common/fs/zfs/zap_micro.c +++ b/uts/common/fs/zfs/zap_micro.c @@ -33,6 +33,7 @@ #include #include #include +#include #ifdef _KERNEL #include @@ -653,9 +654,9 @@ zap_create_flags(objset_t *os, int normflags, zap_flags_t flags, uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx); ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT && - leaf_blockshift <= SPA_MAXBLOCKSHIFT && + leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT && indirect_blockshift >= SPA_MINBLOCKSHIFT && - indirect_blockshift <= SPA_MAXBLOCKSHIFT); + indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT); VERIFY(dmu_object_set_blocksize(os, obj, 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0); @@ -1345,7 +1346,6 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add, zap_t *zap; int err = 0; - /* * Since, we don't have a name, we cannot figure out which blocks will * be affected in this operation. So, account for the worst case : @@ -1358,7 +1358,7 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add, * large microzap results in a promotion to fatzap. */ if (name == NULL) { - *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE; + *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE; return (err); } @@ -1382,7 +1382,7 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add, /* * We treat this case as similar to (name == NULL) */ - *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE; + *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE; } } else { /* @@ -1401,12 +1401,12 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add, * ptrtbl blocks */ if (dmu_buf_freeable(zap->zap_dbuf)) - *tooverwrite += SPA_MAXBLOCKSIZE; + *tooverwrite += MZAP_MAX_BLKSZ; else - *towrite += SPA_MAXBLOCKSIZE; + *towrite += MZAP_MAX_BLKSZ; if (add) { - *towrite += 4 * SPA_MAXBLOCKSIZE; + *towrite += 4 * MZAP_MAX_BLKSZ; } } diff --git a/uts/common/fs/zfs/zfs_ioctl.c b/uts/common/fs/zfs/zfs_ioctl.c index 0cec1c2adad3..a05e7b2a29b3 100644 --- a/uts/common/fs/zfs/zfs_ioctl.c +++ b/uts/common/fs/zfs/zfs_ioctl.c @@ -2383,7 +2383,7 @@ zfs_prop_set_special(const char *dsname, zprop_source_t source, const char *propname = nvpair_name(pair); zfs_prop_t prop = zfs_name_to_prop(propname); uint64_t intval; - int err; + int err = -1; if (prop == ZPROP_INVAL) { if (zfs_prop_userquota(propname)) @@ -3772,8 +3772,7 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr) * the SPA supports it. We ignore any errors here since * we'll catch them later. */ - if (nvpair_type(pair) == DATA_TYPE_UINT64 && - nvpair_value_uint64(pair, &intval) == 0) { + if (nvpair_value_uint64(pair, &intval) == 0) { if (intval >= ZIO_COMPRESS_GZIP_1 && intval <= ZIO_COMPRESS_GZIP_9 && zfs_earlier_version(dsname, @@ -3824,6 +3823,42 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr) return (SET_ERROR(ENOTSUP)); break; + case ZFS_PROP_RECORDSIZE: + /* Record sizes above 128k need the feature to be enabled */ + if (nvpair_value_uint64(pair, &intval) == 0 && + intval > SPA_OLD_MAXBLOCKSIZE) { + spa_t *spa; + + /* + * If this is a bootable dataset then + * the we don't allow large (>128K) blocks, + * because GRUB doesn't support them. + */ + if (zfs_is_bootfs(dsname) && + intval > SPA_OLD_MAXBLOCKSIZE) { + return (SET_ERROR(EDOM)); + } + + /* + * We don't allow setting the property above 1MB, + * unless the tunable has been changed. + */ + if (intval > zfs_max_recordsize || + intval > SPA_MAXBLOCKSIZE) + return (SET_ERROR(EDOM)); + + if ((err = spa_open(dsname, &spa, FTAG)) != 0) + return (err); + + if (!spa_feature_is_enabled(spa, + SPA_FEATURE_LARGE_BLOCKS)) { + spa_close(spa, FTAG); + return (SET_ERROR(ENOTSUP)); + } + spa_close(spa, FTAG); + } + break; + case ZFS_PROP_SHARESMB: if (zpl_earlier_version(dsname, ZPL_VERSION_FUID)) return (SET_ERROR(ENOTSUP)); @@ -4245,7 +4280,7 @@ zfs_ioc_recv(zfs_cmd_t *zc) * zc_fromobj objsetid of incremental fromsnap (may be zero) * zc_guid if set, estimate size of stream only. zc_cookie is ignored. * output size in zc_objset_type. - * zc_flags if =1, WRITE_EMBEDDED records are permitted + * zc_flags lzc_send_flags * * outputs: * zc_objset_type estimated size, if zc_guid is set @@ -4257,6 +4292,7 @@ zfs_ioc_send(zfs_cmd_t *zc) offset_t off; boolean_t estimate = (zc->zc_guid != 0); boolean_t embedok = (zc->zc_flags & 0x1); + boolean_t large_block_ok = (zc->zc_flags & 0x2); if (zc->zc_obj != 0) { dsl_pool_t *dp; @@ -4317,7 +4353,8 @@ zfs_ioc_send(zfs_cmd_t *zc) off = fp->f_offset; error = dmu_send_obj(zc->zc_name, zc->zc_sendobj, - zc->zc_fromobj, embedok, zc->zc_cookie, fp->f_vnode, &off); + zc->zc_fromobj, embedok, large_block_ok, + zc->zc_cookie, fp->f_vnode, &off); if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0) fp->f_offset = off; @@ -5219,6 +5256,8 @@ zfs_ioc_space_snaps(const char *lastsnap, nvlist_t *innvl, nvlist_t *outnvl) * innvl: { * "fd" -> file descriptor to write stream to (int32) * (optional) "fromsnap" -> full snap name to send an incremental from + * (optional) "largeblockok" -> (value ignored) + * indicates that blocks > 128KB are permitted * (optional) "embedok" -> (value ignored) * presence indicates DRR_WRITE_EMBEDDED records are permitted * } @@ -5233,6 +5272,7 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) offset_t off; char *fromname = NULL; int fd; + boolean_t largeblockok; boolean_t embedok; error = nvlist_lookup_int32(innvl, "fd", &fd); @@ -5241,6 +5281,7 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) (void) nvlist_lookup_string(innvl, "fromsnap", &fromname); + largeblockok = nvlist_exists(innvl, "largeblockok"); embedok = nvlist_exists(innvl, "embedok"); file_t *fp = getf(fd); @@ -5248,7 +5289,8 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl) return (SET_ERROR(EBADF)); off = fp->f_offset; - error = dmu_send(snapname, fromname, embedok, fd, fp->f_vnode, &off); + error = dmu_send(snapname, fromname, embedok, largeblockok, + fd, fp->f_vnode, &off); if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0) fp->f_offset = off; diff --git a/uts/common/fs/zfs/zfs_log.c b/uts/common/fs/zfs/zfs_log.c index aeaba2233a19..47d32a45c39d 100644 --- a/uts/common/fs/zfs/zfs_log.c +++ b/uts/common/fs/zfs/zfs_log.c @@ -485,7 +485,7 @@ zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype, * If the write would overflow the largest block then split it. */ if (write_state != WR_INDIRECT && resid > ZIL_MAX_LOG_DATA) - len = SPA_MAXBLOCKSIZE >> 1; + len = SPA_OLD_MAXBLOCKSIZE >> 1; else len = resid; diff --git a/uts/common/fs/zfs/zfs_vfsops.c b/uts/common/fs/zfs/zfs_vfsops.c index 0b0b0a99cc6e..8cf83b399a15 100644 --- a/uts/common/fs/zfs/zfs_vfsops.c +++ b/uts/common/fs/zfs/zfs_vfsops.c @@ -272,10 +272,9 @@ static void blksz_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; - - if (newval < SPA_MINBLOCKSIZE || - newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) - newval = SPA_MAXBLOCKSIZE; + ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); + ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); + ASSERT(ISP2(newval)); zfsvfs->z_max_blksz = newval; zfsvfs->z_vfs->vfs_bsize = newval; @@ -906,7 +905,7 @@ zfsvfs_create(const char *osname, zfsvfs_t **zfvp) */ zfsvfs->z_vfs = NULL; zfsvfs->z_parent = zfsvfs; - zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; + zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; zfsvfs->z_os = os; diff --git a/uts/common/fs/zfs/zfs_vnops.c b/uts/common/fs/zfs/zfs_vnops.c index c608a79fd2a0..5060412afd9b 100644 --- a/uts/common/fs/zfs/zfs_vnops.c +++ b/uts/common/fs/zfs/zfs_vnops.c @@ -817,8 +817,14 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) uint64_t new_blksz; if (zp->z_blksz > max_blksz) { + /* + * File's blocksize is already larger than the + * "recordsize" property. Only let it grow to + * the next power of 2. + */ ASSERT(!ISP2(zp->z_blksz)); - new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); + new_blksz = MIN(end_size, + 1 << highbit64(zp->z_blksz)); } else { new_blksz = MIN(end_size, max_blksz); } diff --git a/uts/common/fs/zfs/zfs_znode.c b/uts/common/fs/zfs/zfs_znode.c index 7577250408df..4664899d13f8 100644 --- a/uts/common/fs/zfs/zfs_znode.c +++ b/uts/common/fs/zfs/zfs_znode.c @@ -58,6 +58,7 @@ #endif /* _KERNEL */ #include +#include #include #include #include @@ -1474,8 +1475,13 @@ zfs_extend(znode_t *zp, uint64_t end) * We are growing the file past the current block size. */ if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { + /* + * File's blocksize is already larger than the + * "recordsize" property. Only let it grow to + * the next power of 2. + */ ASSERT(!ISP2(zp->z_blksz)); - newblksz = MIN(end, SPA_MAXBLOCKSIZE); + newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); } else { newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); } diff --git a/uts/common/fs/zfs/zil.c b/uts/common/fs/zfs/zil.c index 6377285fe2a9..6beacadf710d 100644 --- a/uts/common/fs/zfs/zil.c +++ b/uts/common/fs/zfs/zil.c @@ -220,6 +220,7 @@ zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) { error = SET_ERROR(ECKSUM); } else { + ASSERT3U(len, <=, SPA_OLD_MAXBLOCKSIZE); bcopy(lr, dst, len); *end = (char *)dst + len; *nbp = zilc->zc_next_blk; @@ -234,6 +235,8 @@ zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, (zilc->zc_nused > (size - sizeof (*zilc)))) { error = SET_ERROR(ECKSUM); } else { + ASSERT3U(zilc->zc_nused, <=, + SPA_OLD_MAXBLOCKSIZE); bcopy(lr, dst, zilc->zc_nused); *end = (char *)dst + zilc->zc_nused; *nbp = zilc->zc_next_blk; @@ -317,7 +320,7 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, * If the log has been claimed, stop if we encounter a sequence * number greater than the highest claimed sequence number. */ - lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE); + lrbuf = zio_buf_alloc(SPA_OLD_MAXBLOCKSIZE); zil_bp_tree_init(zilog); for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { @@ -364,7 +367,7 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq)); zil_bp_tree_fini(zilog); - zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE); + zio_buf_free(lrbuf, SPA_OLD_MAXBLOCKSIZE); return (error); } @@ -896,7 +899,7 @@ zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) * * These must be a multiple of 4KB. Note only the amount used (again * aligned to 4KB) actually gets written. However, we can't always just - * allocate SPA_MAXBLOCKSIZE as the slog space could be exhausted. + * allocate SPA_OLD_MAXBLOCKSIZE as the slog space could be exhausted. */ uint64_t zil_block_buckets[] = { 4096, /* non TX_WRITE */ @@ -978,7 +981,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) continue; zil_blksz = zil_block_buckets[i]; if (zil_blksz == UINT64_MAX) - zil_blksz = SPA_MAXBLOCKSIZE; + zil_blksz = SPA_OLD_MAXBLOCKSIZE; zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz; for (i = 0; i < ZIL_PREV_BLKS; i++) zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]); diff --git a/uts/common/fs/zfs/zio.c b/uts/common/fs/zfs/zio.c index 302d692c64a6..d67e1361b4f8 100644 --- a/uts/common/fs/zfs/zio.c +++ b/uts/common/fs/zfs/zio.c @@ -112,9 +112,8 @@ zio_init(void) /* * For small buffers, we want a cache for each multiple of - * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache - * for each quarter-power of 2. For large buffers, we want - * a cache for each multiple of PAGESIZE. + * SPA_MINBLOCKSIZE. For larger buffers, we want a cache + * for each quarter-power of 2. */ for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { size_t size = (c + 1) << SPA_MINBLOCKSHIFT; @@ -137,10 +136,8 @@ zio_init(void) #endif if (size <= 4 * SPA_MINBLOCKSIZE) { align = SPA_MINBLOCKSIZE; - } else if (IS_P2ALIGNED(size, PAGESIZE)) { - align = PAGESIZE; } else if (IS_P2ALIGNED(size, p2 >> 2)) { - align = p2 >> 2; + align = MIN(p2 >> 2, PAGESIZE); } if (align != 0) { diff --git a/uts/common/fs/zfs/zvol.c b/uts/common/fs/zfs/zvol.c index ca05d5cf214c..80888103fe48 100644 --- a/uts/common/fs/zfs/zvol.c +++ b/uts/common/fs/zfs/zvol.c @@ -191,7 +191,7 @@ int zvol_check_volblocksize(uint64_t volblocksize) { if (volblocksize < SPA_MINBLOCKSIZE || - volblocksize > SPA_MAXBLOCKSIZE || + volblocksize > SPA_OLD_MAXBLOCKSIZE || !ISP2(volblocksize)) return (SET_ERROR(EDOM)); @@ -692,7 +692,7 @@ zvol_prealloc(zvol_state_t *zv) while (resid != 0) { int error; - uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); + uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); tx = dmu_tx_create(os); dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); @@ -1661,7 +1661,8 @@ zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) (void) strcpy(dki.dki_dname, "zvol"); dki.dki_ctype = DKC_UNKNOWN; dki.dki_unit = getminor(dev); - dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); + dki.dki_maxtransfer = + 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); mutex_exit(&zfsdev_state_lock); if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) error = SET_ERROR(EFAULT); @@ -1978,14 +1979,14 @@ zvol_dump_init(zvol_state_t *zv, boolean_t resize) zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs, tx); error = error ? error : dmu_object_set_blocksize( - os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); + os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); if (version >= SPA_VERSION_DEDUP) { error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup, tx); } if (error == 0) - zv->zv_volblocksize = SPA_MAXBLOCKSIZE; + zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; } dmu_tx_commit(tx); diff --git a/uts/common/sys/fs/zfs.h b/uts/common/sys/fs/zfs.h index 7155d9702bb2..569fae20915d 100644 --- a/uts/common/sys/fs/zfs.h +++ b/uts/common/sys/fs/zfs.h @@ -192,6 +192,7 @@ typedef enum { ZPOOL_PROP_FREEING, ZPOOL_PROP_FRAGMENTATION, ZPOOL_PROP_LEAKED, + ZPOOL_PROP_MAXBLOCKSIZE, ZPOOL_NUM_PROPS } zpool_prop_t;