From 37f9dac592bf5889c3efb305c48ac39b4c7dd140 Mon Sep 17 00:00:00 2001 From: Richard Yao Date: Fri, 4 Jul 2014 18:43:47 -0400 Subject: [PATCH] zvol processing should use struct bio Internally, zvols are files exposed through the block device API. This is intended to reduce overhead when things require block devices. However, the ZoL zvol code emulates a traditional block device in that it has a top half and a bottom half. This is an unnecessary source of overhead that does not exist on any other OpenZFS platform does this. This patch removes it. Early users of this patch reported double digit performance gains in IOPS on zvols in the range of 50% to 80%. Comments in the code suggest that the current implementation was done to obtain IO merging from Linux's IO elevator. However, the DMU already does write merging while arc_read() should implicitly merge read IOs because only 1 thread is permitted to fetch the buffer into ARC. In addition, commercial ZFSOnLinux distributions report that regular files are more performant than zvols under the current implementation, and the main consumers of zvols are VMs and iSCSI targets, which have their own elevators to merge IOs. Some minor refactoring allows us to register zfs_request() as our ->make_request() handler in place of the generic_make_request() function. This eliminates the layer of code that broke IO requests on zvols into a top half and a bottom half. This has several benefits: 1. No per zvol spinlocks. 2. No redundant IO elevator processing. 3. Interrupts are disabled only when actually necessary. 4. No redispatching of IOs when all taskq threads are busy. 5. Linux's page out routines will properly block. 6. Many autotools checks become obsolete. An unfortunate consequence of eliminating the layer that generic_make_request() is that we no longer calls the instrumentation hooks for block IO accounting. Those hooks are GPL-exported, so we cannot call them ourselves and consequently, we lose the ability to do IO monitoring via iostat. Since zvols are internally files mapped as block devices, this should be okay. Anyone who is willing to accept the performance penalty for the block IO layer's accounting could use the loop device in between the zvol and its consumer. Alternatively, perf and ftrace likely could be used. Also, tools like latencytop will still work. Tools such as latencytop sometimes provide a better view of performance bottlenecks than the traditional block IO accounting tools do. Lastly, if direct reclaim occurs during spacemap loading and swap is on a zvol, this code will deadlock. That deadlock could already occur with sync=always on zvols. Given that swap on zvols is not yet production ready, this is not a blocker. Signed-off-by: Richard Yao --- config/kernel-bio-rw-barrier.m4 | 25 +++ config/kernel-bio-rw-discard.m4 | 25 +++ config/kernel-current_bio_tail.m4 | 33 ++++ config/kernel-mk-request-fn.m4 | 43 +++++ config/kernel.m4 | 4 + include/linux/blkdev_compat.h | 21 ++- include/sys/dmu.h | 4 +- man/man5/zfs-module-parameters.5 | 11 -- module/zfs/dmu.c | 57 +++---- module/zfs/vdev_disk.c | 22 ++- module/zfs/zvol.c | 257 ++++++++++-------------------- 11 files changed, 280 insertions(+), 222 deletions(-) create mode 100644 config/kernel-bio-rw-barrier.m4 create mode 100644 config/kernel-bio-rw-discard.m4 create mode 100644 config/kernel-current_bio_tail.m4 create mode 100644 config/kernel-mk-request-fn.m4 diff --git a/config/kernel-bio-rw-barrier.m4 b/config/kernel-bio-rw-barrier.m4 new file mode 100644 index 000000000000..bcf0f7ea00b0 --- /dev/null +++ b/config/kernel-bio-rw-barrier.m4 @@ -0,0 +1,25 @@ +dnl # +dnl # Interface for issuing a discard bio: +dnl # 2.6.28-2.6.35: BIO_RW_BARRIER +dnl # 2.6.36-3.x: REQ_BARRIER +dnl # + +dnl # Since REQ_BARRIER is a preprocessor definition, there is no need for an +dnl # autotools check for it. Also, REQ_BARRIER existed in the request layer +dnl # until torvalds/linux@7b6d91daee5cac6402186ff224c3af39d79f4a0e unified the +dnl # request layer and bio layer flags, so it would be wrong to assume that +dnl # the APIs are mutually exclusive contrary to the typical case. +AC_DEFUN([ZFS_AC_KERNEL_BIO_RW_BARRIER], [ + AC_MSG_CHECKING([whether BIO_RW_BARRIER is defined]) + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + int flags __attribute__ ((unused)); + flags = BIO_RW_BARRIER; + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_BIO_RW_BARRIER, 1, [BIO_RW_BARRIER is defined]) + ],[ + AC_MSG_RESULT(no) + ]) +]) diff --git a/config/kernel-bio-rw-discard.m4 b/config/kernel-bio-rw-discard.m4 new file mode 100644 index 000000000000..0554b9a9dae0 --- /dev/null +++ b/config/kernel-bio-rw-discard.m4 @@ -0,0 +1,25 @@ +dnl # +dnl # Interface for issuing a discard bio: +dnl # 2.6.28-2.6.35: BIO_RW_DISCARD +dnl # 2.6.36-3.x: REQ_DISCARD +dnl # + +dnl # Since REQ_DISCARD is a preprocessor definition, there is no need for an +dnl # autotools check for it. Also, REQ_DISCARD existed in the request layer +dnl # until torvalds/linux@7b6d91daee5cac6402186ff224c3af39d79f4a0e unified the +dnl # request layer and bio layer flags, so it would be wrong to assume that +dnl # the APIs are mutually exclusive contrary to the typical case. +AC_DEFUN([ZFS_AC_KERNEL_BIO_RW_DISCARD], [ + AC_MSG_CHECKING([whether BIO_RW_DISCARD is defined]) + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + int flags __attribute__ ((unused)); + flags = BIO_RW_DISCARD; + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_BIO_RW_DISCARD, 1, [BIO_RW_DISCARD is defined]) + ],[ + AC_MSG_RESULT(no) + ]) +]) diff --git a/config/kernel-current_bio_tail.m4 b/config/kernel-current_bio_tail.m4 new file mode 100644 index 000000000000..b72f21e8a356 --- /dev/null +++ b/config/kernel-current_bio_tail.m4 @@ -0,0 +1,33 @@ +dnl # +dnl # 2.6.34 API change +dnl # current->bio_tail and current->bio_list were struct bio pointers prior to +dnl # Linux 2.6.34. They were refactored into a struct bio_list pointer called +dnl # current->bio_list in Linux 2.6.34. +dnl # +AC_DEFUN([ZFS_AC_KERNEL_CURRENT_BIO_TAIL], [ + AC_MSG_CHECKING([whether current->bio_tail exists]) + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + current->bio_tail = (struct bio **) NULL; + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_CURRENT_BIO_TAIL, 1, + [current->bio_tail exists]) + ],[ + AC_MSG_RESULT(no) + AC_MSG_CHECKING([whether current->bio_list exists]) + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + current->bio_list = (struct bio_list *) NULL; + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_CURRENT_BIO_LIST, 1, + [current->bio_list exists]) + ],[ + AC_MSG_ERROR(no - Please file a bug report at + https://github.com/zfsonlinux/zfs/issues/new) + ]) + ]) +]) diff --git a/config/kernel-mk-request-fn.m4 b/config/kernel-mk-request-fn.m4 new file mode 100644 index 000000000000..88ee2ebb3d00 --- /dev/null +++ b/config/kernel-mk-request-fn.m4 @@ -0,0 +1,43 @@ +dnl # +dnl # Linux 3.2 API Change +dnl # make_request_fn returns void instead of int. +dnl # +AC_DEFUN([ZFS_AC_KERNEL_MAKE_REQUEST_FN], [ + AC_MSG_CHECKING([whether make_request_fn() returns int]) + ZFS_LINUX_TRY_COMPILE([ + #include + + int make_request(struct request_queue *q, struct bio *bio) + { + return (0); + } + ],[ + blk_queue_make_request(NULL, &make_request); + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(MAKE_REQUEST_FN_RET, int, + [make_request_fn() returns int]) + AC_DEFINE(HAVE_MAKE_REQUEST_FN_RET_INT, 1, + [Noting that make_request_fn() returns int]) + ],[ + AC_MSG_RESULT(no) + AC_MSG_CHECKING([whether make_request_fn() returns void]) + ZFS_LINUX_TRY_COMPILE([ + #include + + void make_request(struct request_queue *q, struct bio *bio) + { + return; + } + ],[ + blk_queue_make_request(NULL, &make_request); + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(MAKE_REQUEST_FN_RET, void, + [make_request_fn() returns void]) + ],[ + AC_MSG_ERROR(no - Please file a bug report at + https://github.com/zfsonlinux/zfs/issues/new) + ]) + ]) +]) diff --git a/config/kernel.m4 b/config/kernel.m4 index 09d8003f12a0..b3dd7232c226 100644 --- a/config/kernel.m4 +++ b/config/kernel.m4 @@ -7,6 +7,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_TEST_MODULE ZFS_AC_KERNEL_CONFIG ZFS_AC_KERNEL_DECLARE_EVENT_CLASS + ZFS_AC_KERNEL_CURRENT_BIO_TAIL ZFS_AC_KERNEL_BDEV_BLOCK_DEVICE_OPERATIONS ZFS_AC_KERNEL_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID ZFS_AC_KERNEL_TYPE_FMODE_T @@ -22,6 +23,8 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_KERNEL_BIO_FAILFAST_DTD ZFS_AC_KERNEL_REQ_FAILFAST_MASK ZFS_AC_KERNEL_BIO_END_IO_T_ARGS + ZFS_AC_KERNEL_BIO_RW_BARRIER + ZFS_AC_KERNEL_BIO_RW_DISCARD ZFS_AC_KERNEL_BIO_RW_SYNC ZFS_AC_KERNEL_BIO_RW_SYNCIO ZFS_AC_KERNEL_REQ_SYNC @@ -100,6 +103,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_KERNEL_VFS_RW_ITERATE ZFS_AC_KERNEL_KMAP_ATOMIC_ARGS ZFS_AC_KERNEL_FOLLOW_DOWN_ONE + ZFS_AC_KERNEL_MAKE_REQUEST_FN AS_IF([test "$LINUX_OBJ" != "$LINUX"], [ KERNELMAKE_PARAMS="$KERNELMAKE_PARAMS O=$LINUX_OBJ" diff --git a/include/linux/blkdev_compat.h b/include/linux/blkdev_compat.h index cdcf12c42090..0b5c8af429e1 100644 --- a/include/linux/blkdev_compat.h +++ b/include/linux/blkdev_compat.h @@ -295,10 +295,16 @@ struct req_iterator { #define BIO_BI_SECTOR(bio) (bio)->bi_iter.bi_sector #define BIO_BI_SIZE(bio) (bio)->bi_iter.bi_size #define BIO_BI_IDX(bio) (bio)->bi_iter.bi_idx +#define bio_for_each_segment4(bv, bvp, b, i) \ + bio_for_each_segment((bv), (b), (i)) +typedef struct bvec_iter bvec_iterator_t; #else #define BIO_BI_SECTOR(bio) (bio)->bi_sector #define BIO_BI_SIZE(bio) (bio)->bi_size #define BIO_BI_IDX(bio) (bio)->bi_idx +#define bio_for_each_segment4(bv, bvp, b, i) \ + bio_for_each_segment((bvp), (b), (i)) +typedef int bvec_iterator_t; #endif /* @@ -457,17 +463,30 @@ bio_set_flags_failfast(struct block_device *bdev, int *flags) #define VDEV_REQ_FUA REQ_FUA #else #define VDEV_WRITE_FLUSH_FUA WRITE_BARRIER +#ifdef HAVE_BIO_RW_BARRIER +#define VDEV_REQ_FLUSH (1 << BIO_RW_BARRIER) +#define VDEV_REQ_FUA (1 << BIO_RW_BARRIER) +#else #define VDEV_REQ_FLUSH REQ_HARDBARRIER #define VDEV_REQ_FUA REQ_FUA #endif +#endif /* * 2.6.32 API change * Use the normal I/O patch for discards. */ -#ifdef REQ_DISCARD +#ifdef QUEUE_FLAG_DISCARD +#ifdef HAVE_BIO_RW_DISCARD +#define VDEV_REQ_DISCARD (1 << BIO_RW_DISCARD) +#else #define VDEV_REQ_DISCARD REQ_DISCARD #endif +#else +#error "Allowing the build will cause discard requests to become writes " + "potentially triggering the DMU_MAX_ACCESS assertion. Please file a " + "an issue report at: https://github.com/zfsonlinux/zfs/issues/new" +#endif /* * 2.6.33 API change diff --git a/include/sys/dmu.h b/include/sys/dmu.h index 4ad496ae0dc3..d9434db46383 100644 --- a/include/sys/dmu.h +++ b/include/sys/dmu.h @@ -710,8 +710,8 @@ void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, dmu_tx_t *tx); #ifdef _KERNEL #include -int dmu_read_req(objset_t *os, uint64_t object, struct request *req); -int dmu_write_req(objset_t *os, uint64_t object, struct request *req, +int dmu_read_bio(objset_t *os, uint64_t object, struct bio *bio); +int dmu_write_bio(objset_t *os, uint64_t object, struct bio *bio, dmu_tx_t *tx); int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); diff --git a/man/man5/zfs-module-parameters.5 b/man/man5/zfs-module-parameters.5 index 35ea9d9f6fbb..2ceb6551961c 100644 --- a/man/man5/zfs-module-parameters.5 +++ b/man/man5/zfs-module-parameters.5 @@ -1591,17 +1591,6 @@ Max number of blocks to discard at once Default value: \fB16,384\fR. .RE -.sp -.ne 2 -.na -\fBzvol_threads\fR (uint) -.ad -.RS 12n -Max number of threads to handle zvol I/O requests -.sp -Default value: \fB32\fR. -.RE - .SH ZFS I/O SCHEDULER ZFS issues I/O operations to leaf vdevs to satisfy and complete I/Os. The I/O scheduler determines when and in what order those operations are diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c index ac7499d0176e..5e2a1db601b4 100644 --- a/module/zfs/dmu.c +++ b/module/zfs/dmu.c @@ -1049,15 +1049,16 @@ xuio_stat_wbuf_nocopy() * return value is the number of bytes successfully copied to arg_buf. */ static int -dmu_req_copy(void *arg_buf, int size, struct request *req, size_t req_offset) +dmu_bio_copy(void *arg_buf, int size, struct bio *bio, size_t bio_offset) { - struct bio_vec bv, *bvp; - struct req_iterator iter; + struct bio_vec bv, *bvp = &bv; + bvec_iterator_t iter; char *bv_buf; int tocpy, bv_len, bv_offset; int offset = 0; - rq_for_each_segment4(bv, bvp, req, iter) { + bio_for_each_segment4(bv, bvp, bio, iter) { + /* * Fully consumed the passed arg_buf. We use goto here because * rq_for_each_segment is a double loop @@ -1066,23 +1067,23 @@ dmu_req_copy(void *arg_buf, int size, struct request *req, size_t req_offset) if (size == offset) goto out; - /* Skip already copied bv */ - if (req_offset >= bv.bv_len) { - req_offset -= bv.bv_len; + /* Skip already copied bvp */ + if (bio_offset >= bvp->bv_len) { + bio_offset -= bvp->bv_len; continue; } - bv_len = bv.bv_len - req_offset; - bv_offset = bv.bv_offset + req_offset; - req_offset = 0; + bv_len = bvp->bv_len - bio_offset; + bv_offset = bvp->bv_offset + bio_offset; + bio_offset = 0; tocpy = MIN(bv_len, size - offset); ASSERT3S(tocpy, >=, 0); - bv_buf = page_address(bv.bv_page) + bv_offset; + bv_buf = page_address(bvp->bv_page) + bv_offset; ASSERT3P(bv_buf, !=, NULL); - if (rq_data_dir(req) == WRITE) + if (bio_data_dir(bio) == WRITE) memcpy(arg_buf + offset, bv_buf, tocpy); else memcpy(bv_buf, arg_buf + offset, tocpy); @@ -1094,13 +1095,13 @@ dmu_req_copy(void *arg_buf, int size, struct request *req, size_t req_offset) } int -dmu_read_req(objset_t *os, uint64_t object, struct request *req) +dmu_read_bio(objset_t *os, uint64_t object, struct bio *bio) { - uint64_t size = blk_rq_bytes(req); - uint64_t offset = blk_rq_pos(req) << 9; + uint64_t offset = BIO_BI_SECTOR(bio) << 9; + uint64_t size = BIO_BI_SIZE(bio); dmu_buf_t **dbp; int numbufs, i, err; - size_t req_offset; + size_t bio_offset; /* * NB: we could do this block-at-a-time, but it's nice @@ -1111,7 +1112,7 @@ dmu_read_req(objset_t *os, uint64_t object, struct request *req) if (err) return (err); - req_offset = 0; + bio_offset = 0; for (i = 0; i < numbufs; i++) { uint64_t tocpy; int64_t bufoff; @@ -1125,8 +1126,8 @@ dmu_read_req(objset_t *os, uint64_t object, struct request *req) if (tocpy == 0) break; - didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req, - req_offset); + didcpy = dmu_bio_copy(db->db_data + bufoff, tocpy, bio, + bio_offset); if (didcpy < tocpy) err = EIO; @@ -1136,7 +1137,7 @@ dmu_read_req(objset_t *os, uint64_t object, struct request *req) size -= tocpy; offset += didcpy; - req_offset += didcpy; + bio_offset += didcpy; err = 0; } dmu_buf_rele_array(dbp, numbufs, FTAG); @@ -1145,13 +1146,13 @@ dmu_read_req(objset_t *os, uint64_t object, struct request *req) } int -dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) +dmu_write_bio(objset_t *os, uint64_t object, struct bio *bio, dmu_tx_t *tx) { - uint64_t size = blk_rq_bytes(req); - uint64_t offset = blk_rq_pos(req) << 9; + uint64_t offset = BIO_BI_SECTOR(bio) << 9; + uint64_t size = BIO_BI_SIZE(bio); dmu_buf_t **dbp; int numbufs, i, err; - size_t req_offset; + size_t bio_offset; if (size == 0) return (0); @@ -1161,7 +1162,7 @@ dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) if (err) return (err); - req_offset = 0; + bio_offset = 0; for (i = 0; i < numbufs; i++) { uint64_t tocpy; int64_t bufoff; @@ -1182,8 +1183,8 @@ dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) else dmu_buf_will_dirty(db, tx); - didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req, - req_offset); + didcpy = dmu_bio_copy(db->db_data + bufoff, tocpy, bio, + bio_offset); if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); @@ -1196,7 +1197,7 @@ dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) size -= tocpy; offset += didcpy; - req_offset += didcpy; + bio_offset += didcpy; err = 0; } diff --git a/module/zfs/vdev_disk.c b/module/zfs/vdev_disk.c index 380ede35b517..e7e2b3b93f40 100644 --- a/module/zfs/vdev_disk.c +++ b/module/zfs/vdev_disk.c @@ -496,6 +496,22 @@ bio_map(struct bio *bio, void *bio_ptr, unsigned int bio_size) return (bio_size); } +static inline void +vdev_submit_bio(int rw, struct bio *bio) +{ +#ifdef HAVE_CURRENT_BIO_TAIL + struct bio **bio_tail = current->bio_tail; + current->bio_tail = NULL; + submit_bio(rw, bio); + current->bio_tail = bio_tail; +#else + struct bio_list *bio_list = current->bio_list; + current->bio_list = NULL; + submit_bio(rw, bio); + current->bio_list = bio_list; +#endif +} + static int __vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr, size_t kbuf_size, uint64_t kbuf_offset, int flags) @@ -571,7 +587,7 @@ __vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr, bio_offset += BIO_BI_SIZE(dr->dr_bio[i]); } - /* Extra reference to protect dio_request during submit_bio */ + /* Extra reference to protect dio_request during vdev_submit_bio */ vdev_disk_dio_get(dr); if (zio) zio->io_delay = jiffies_64; @@ -579,7 +595,7 @@ __vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr, /* Submit all bio's associated with this dio */ for (i = 0; i < dr->dr_bio_count; i++) if (dr->dr_bio[i]) - submit_bio(dr->dr_rw, dr->dr_bio[i]); + vdev_submit_bio(dr->dr_rw, dr->dr_bio[i]); /* * On synchronous blocking requests we wait for all bio the completion @@ -645,7 +661,7 @@ vdev_disk_io_flush(struct block_device *bdev, zio_t *zio) bio->bi_private = zio; bio->bi_bdev = bdev; zio->io_delay = jiffies_64; - submit_bio(VDEV_WRITE_FLUSH_FUA, bio); + vdev_submit_bio(VDEV_WRITE_FLUSH_FUA, bio); invalidate_bdev(bdev); return (0); diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c index 0c6cddef4205..074ec51e6f9e 100644 --- a/module/zfs/zvol.c +++ b/module/zfs/zvol.c @@ -50,10 +50,8 @@ unsigned int zvol_inhibit_dev = 0; unsigned int zvol_major = ZVOL_MAJOR; -unsigned int zvol_threads = 32; unsigned long zvol_max_discard_blocks = 16384; -static taskq_t *zvol_taskq; static kmutex_t zvol_state_lock; static list_t zvol_state_list; static char *zvol_tag = "zvol_tag"; @@ -590,34 +588,24 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset, } } -/* - * Common write path running under the zvol taskq context. This function - * is responsible for copying the request structure data in to the DMU and - * signaling the request queue with the result of the copy. - */ -static void -zvol_write(void *arg) +static int +zvol_write(struct bio *bio) { - struct request *req = (struct request *)arg; - struct request_queue *q = req->q; - zvol_state_t *zv = q->queuedata; - fstrans_cookie_t cookie = spl_fstrans_mark(); - uint64_t offset = blk_rq_pos(req) << 9; - uint64_t size = blk_rq_bytes(req); + zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data; + uint64_t offset = BIO_BI_SECTOR(bio) << 9; + uint64_t size = BIO_BI_SIZE(bio); int error = 0; dmu_tx_t *tx; rl_t *rl; - if (req->cmd_flags & VDEV_REQ_FLUSH) + if (bio->bi_rw & VDEV_REQ_FLUSH) zil_commit(zv->zv_zilog, ZVOL_OBJ); /* * Some requests are just for flush and nothing else. */ - if (size == 0) { - error = 0; + if (size == 0) goto out; - } rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER); @@ -632,96 +620,77 @@ zvol_write(void *arg) goto out; } - error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx); + error = dmu_write_bio(zv->zv_objset, ZVOL_OBJ, bio, tx); if (error == 0) zvol_log_write(zv, tx, offset, size, - req->cmd_flags & VDEV_REQ_FUA); + !!(bio->bi_rw & VDEV_REQ_FUA)); dmu_tx_commit(tx); zfs_range_unlock(rl); - if ((req->cmd_flags & VDEV_REQ_FUA) || + if ((bio->bi_rw & VDEV_REQ_FUA) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zv->zv_zilog, ZVOL_OBJ); out: - blk_end_request(req, -error, size); - spl_fstrans_unmark(cookie); + return (error); } -#ifdef HAVE_BLK_QUEUE_DISCARD -static void -zvol_discard(void *arg) +static int +zvol_discard(struct bio *bio) { - struct request *req = (struct request *)arg; - struct request_queue *q = req->q; - zvol_state_t *zv = q->queuedata; - fstrans_cookie_t cookie = spl_fstrans_mark(); - uint64_t start = blk_rq_pos(req) << 9; - uint64_t end = start + blk_rq_bytes(req); + zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data; + uint64_t start = BIO_BI_SECTOR(bio) << 9; + uint64_t size = BIO_BI_SIZE(bio); + uint64_t end = start + size; int error; rl_t *rl; - if (end > zv->zv_volsize) { - error = EIO; - goto out; - } + if (end > zv->zv_volsize) + return (SET_ERROR(EIO)); /* * Align the request to volume block boundaries. If we don't, * then this will force dnode_free_range() to zero out the * unaligned parts, which is slow (read-modify-write) and * useless since we are not freeing any space by doing so. + * XXX: We should handle secure discard by zeroing out unaligned parts. */ start = P2ROUNDUP(start, zv->zv_volblocksize); end = P2ALIGN(end, zv->zv_volblocksize); - if (start >= end) { - error = 0; - goto out; - } + if (start >= end) + return (0); - rl = zfs_range_lock(&zv->zv_znode, start, end - start, RL_WRITER); + rl = zfs_range_lock(&zv->zv_znode, start, size, RL_WRITER); - error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, start, end-start); + error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, start, size); /* * TODO: maybe we should add the operation to the log. */ zfs_range_unlock(rl); -out: - blk_end_request(req, -error, blk_rq_bytes(req)); - spl_fstrans_unmark(cookie); -} -#endif /* HAVE_BLK_QUEUE_DISCARD */ -/* - * Common read path running under the zvol taskq context. This function - * is responsible for copying the requested data out of the DMU and in to - * a linux request structure. It then must signal the request queue with - * an error code describing the result of the copy. - */ -static void -zvol_read(void *arg) + return (error); +} + +static int +zvol_read(struct bio *bio) { - struct request *req = (struct request *)arg; - struct request_queue *q = req->q; - zvol_state_t *zv = q->queuedata; - fstrans_cookie_t cookie = spl_fstrans_mark(); - uint64_t offset = blk_rq_pos(req) << 9; - uint64_t size = blk_rq_bytes(req); + zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data; + uint64_t offset = BIO_BI_SECTOR(bio) << 9; + uint64_t len = BIO_BI_SIZE(bio); int error; rl_t *rl; - if (size == 0) { - error = 0; - goto out; - } + if (len == 0) + return (0); - rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); - error = dmu_read_req(zv->zv_objset, ZVOL_OBJ, req); + rl = zfs_range_lock(&zv->zv_znode, offset, len, RL_READER); + + error = dmu_read_bio(zv->zv_objset, ZVOL_OBJ, bio); zfs_range_unlock(rl); @@ -729,91 +698,50 @@ zvol_read(void *arg) if (error == ECKSUM) error = SET_ERROR(EIO); -out: - blk_end_request(req, -error, size); - spl_fstrans_unmark(cookie); + return (error); } -/* - * Request will be added back to the request queue and retried if - * it cannot be immediately dispatched to the taskq for handling - */ -static inline void -zvol_dispatch(task_func_t func, struct request *req) -{ - if (!taskq_dispatch(zvol_taskq, func, (void *)req, TQ_NOSLEEP)) - blk_requeue_request(req->q, req); -} - -/* - * Common request path. Rather than registering a custom make_request() - * function we use the generic Linux version. This is done because it allows - * us to easily merge read requests which would otherwise we performed - * synchronously by the DMU. This is less critical in write case where the - * DMU will perform the correct merging within a transaction group. Using - * the generic make_request() also let's use leverage the fact that the - * elevator with ensure correct ordering in regards to barrior IOs. On - * the downside it means that in the write case we end up doing request - * merging twice once in the elevator and once in the DMU. - * - * The request handler is called under a spin lock so all the real work - * is handed off to be done in the context of the zvol taskq. This function - * simply performs basic request sanity checking and hands off the request. - */ -static void -zvol_request(struct request_queue *q) +static MAKE_REQUEST_FN_RET +zvol_request(struct request_queue *q, struct bio *bio) { zvol_state_t *zv = q->queuedata; - struct request *req; - unsigned int size; + fstrans_cookie_t cookie = spl_fstrans_mark(); + uint64_t offset = BIO_BI_SECTOR(bio); + unsigned int sectors = bio_sectors(bio); + int error = 0; - while ((req = blk_fetch_request(q)) != NULL) { - size = blk_rq_bytes(req); - - if (size != 0 && blk_rq_pos(req) + blk_rq_sectors(req) > - get_capacity(zv->zv_disk)) { - printk(KERN_INFO - "%s: bad access: block=%llu, count=%lu\n", - req->rq_disk->disk_name, - (long long unsigned)blk_rq_pos(req), - (long unsigned)blk_rq_sectors(req)); - __blk_end_request(req, -EIO, size); - continue; - } - - if (!blk_fs_request(req)) { - printk(KERN_INFO "%s: non-fs cmd\n", - req->rq_disk->disk_name); - __blk_end_request(req, -EIO, size); - continue; - } - - switch ((int)rq_data_dir(req)) { - case READ: - zvol_dispatch(zvol_read, req); - break; - case WRITE: - if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { - __blk_end_request(req, -EROFS, size); - break; - } - -#ifdef HAVE_BLK_QUEUE_DISCARD - if (req->cmd_flags & VDEV_REQ_DISCARD) { - zvol_dispatch(zvol_discard, req); - break; - } -#endif /* HAVE_BLK_QUEUE_DISCARD */ - - zvol_dispatch(zvol_write, req); - break; - default: - printk(KERN_INFO "%s: unknown cmd: %d\n", - req->rq_disk->disk_name, (int)rq_data_dir(req)); - __blk_end_request(req, -EIO, size); - break; - } + if (bio_has_data(bio) && offset + sectors > + get_capacity(zv->zv_disk)) { + printk(KERN_INFO + "%s: bad access: block=%llu, count=%lu\n", + zv->zv_disk->disk_name, + (long long unsigned)offset, + (long unsigned)sectors); + error = SET_ERROR(EIO); + goto out; } + + if (bio_data_dir(bio) == WRITE) { + if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { + error = SET_ERROR(EROFS); + goto out; + } + + if (bio->bi_rw & VDEV_REQ_DISCARD) { + error = zvol_discard(bio); + goto out; + } + + error = zvol_write(bio); + } else + error = zvol_read(bio); + +out: + bio_endio(bio, -error); + spl_fstrans_unmark(cookie); +#ifdef HAVE_MAKE_REQUEST_FN_RET_INT + return (0); +#endif } static void @@ -1259,25 +1187,17 @@ static zvol_state_t * zvol_alloc(dev_t dev, const char *name) { zvol_state_t *zv; - int error = 0; zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); spin_lock_init(&zv->zv_lock); list_link_init(&zv->zv_next); - zv->zv_queue = blk_init_queue(zvol_request, &zv->zv_lock); + zv->zv_queue = blk_alloc_queue(GFP_ATOMIC); if (zv->zv_queue == NULL) goto out_kmem; -#ifdef HAVE_ELEVATOR_CHANGE - error = elevator_change(zv->zv_queue, "noop"); -#endif /* HAVE_ELEVATOR_CHANGE */ - if (error) { - printk("ZFS: Unable to set \"%s\" scheduler for zvol %s: %d\n", - "noop", name, error); - goto out_queue; - } + blk_queue_make_request(zv->zv_queue, zvol_request); #ifdef HAVE_BLK_QUEUE_FLUSH blk_queue_flush(zv->zv_queue, VDEV_REQ_FLUSH | VDEV_REQ_FUA); @@ -1418,13 +1338,11 @@ __zvol_create_minor(const char *name, boolean_t ignore_snapdev) blk_queue_max_segment_size(zv->zv_queue, UINT_MAX); blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize); blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize); -#ifdef HAVE_BLK_QUEUE_DISCARD blk_queue_max_discard_sectors(zv->zv_queue, (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize); queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue); -#endif -#ifdef HAVE_BLK_QUEUE_NONROT +#ifdef QUEUE_FLAG_NONROT queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue); #endif #ifdef QUEUE_FLAG_ADD_RANDOM @@ -1651,7 +1569,6 @@ zvol_set_snapdev(const char *dsname, uint64_t snapdev) { int zvol_init(void) { - int threads = MIN(MAX(zvol_threads, 1), 1024); int error; list_create(&zvol_state_list, sizeof (zvol_state_t), @@ -1659,18 +1576,10 @@ zvol_init(void) mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); - zvol_taskq = taskq_create(ZVOL_DRIVER, threads, maxclsyspri, - threads * 2, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); - if (zvol_taskq == NULL) { - printk(KERN_INFO "ZFS: taskq_create() failed\n"); - error = -ENOMEM; - goto out1; - } - error = register_blkdev(zvol_major, ZVOL_DRIVER); if (error) { printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); - goto out2; + goto out; } blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS, @@ -1678,9 +1587,7 @@ zvol_init(void) return (0); -out2: - taskq_destroy(zvol_taskq); -out1: +out: mutex_destroy(&zvol_state_lock); list_destroy(&zvol_state_list); @@ -1693,7 +1600,6 @@ zvol_fini(void) zvol_remove_minors(NULL); blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS); unregister_blkdev(zvol_major, ZVOL_DRIVER); - taskq_destroy(zvol_taskq); mutex_destroy(&zvol_state_lock); list_destroy(&zvol_state_list); } @@ -1704,8 +1610,5 @@ MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes"); module_param(zvol_major, uint, 0444); MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); -module_param(zvol_threads, uint, 0444); -MODULE_PARM_DESC(zvol_threads, "Max number of threads to handle I/O requests"); - module_param(zvol_max_discard_blocks, ulong, 0444); MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard");