4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
27 * ZFS volume emulation driver.
29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30 * Volumes are accessed through the symbolic links named:
32 * /dev/<pool_name>/<dataset_name>
34 * Volumes are persistent through reboot and module load. No user command
35 * needs to be run before opening and using a device.
38 #include <sys/dmu_traverse.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/dsl_prop.h>
42 #include <sys/zil_impl.h>
44 #include <sys/zfs_rlock.h>
45 #include <sys/zfs_znode.h>
47 #include <linux/blkdev_compat.h>
49 unsigned int zvol_inhibit_dev = 0;
50 unsigned int zvol_major = ZVOL_MAJOR;
51 unsigned int zvol_threads = 32;
52 unsigned long zvol_max_discard_blocks = 16384;
54 static taskq_t *zvol_taskq;
55 static kmutex_t zvol_state_lock;
56 static list_t zvol_state_list;
57 static char *zvol_tag = "zvol_tag";
60 * The in-core state of each volume.
62 typedef struct zvol_state {
63 char zv_name[MAXNAMELEN]; /* name */
64 uint64_t zv_volsize; /* advertised space */
65 uint64_t zv_volblocksize;/* volume block size */
66 objset_t *zv_objset; /* objset handle */
67 uint32_t zv_flags; /* ZVOL_* flags */
68 uint32_t zv_open_count; /* open counts */
69 uint32_t zv_changed; /* disk changed */
70 zilog_t *zv_zilog; /* ZIL handle */
71 znode_t zv_znode; /* for range locking */
72 dmu_buf_t *zv_dbuf; /* bonus handle */
73 dev_t zv_dev; /* device id */
74 struct gendisk *zv_disk; /* generic disk */
75 struct request_queue *zv_queue; /* request queue */
76 spinlock_t zv_lock; /* request queue lock */
77 list_node_t zv_next; /* next zvol_state_t linkage */
80 #define ZVOL_RDONLY 0x1
83 * Find the next available range of ZVOL_MINORS minor numbers. The
84 * zvol_state_list is kept in ascending minor order so we simply need
85 * to scan the list for the first gap in the sequence. This allows us
86 * to recycle minor number as devices are created and removed.
89 zvol_find_minor(unsigned *minor)
94 ASSERT(MUTEX_HELD(&zvol_state_lock));
95 for (zv = list_head(&zvol_state_list); zv != NULL;
96 zv = list_next(&zvol_state_list, zv), *minor += ZVOL_MINORS) {
97 if (MINOR(zv->zv_dev) != MINOR(*minor))
101 /* All minors are in use */
102 if (*minor >= (1 << MINORBITS))
109 * Find a zvol_state_t given the full major+minor dev_t.
111 static zvol_state_t *
112 zvol_find_by_dev(dev_t dev)
116 ASSERT(MUTEX_HELD(&zvol_state_lock));
117 for (zv = list_head(&zvol_state_list); zv != NULL;
118 zv = list_next(&zvol_state_list, zv)) {
119 if (zv->zv_dev == dev)
127 * Find a zvol_state_t given the name provided at zvol_alloc() time.
129 static zvol_state_t *
130 zvol_find_by_name(const char *name)
134 ASSERT(MUTEX_HELD(&zvol_state_lock));
135 for (zv = list_head(&zvol_state_list); zv != NULL;
136 zv = list_next(&zvol_state_list, zv)) {
137 if (!strncmp(zv->zv_name, name, MAXNAMELEN))
145 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
148 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
150 zfs_creat_t *zct = arg;
151 nvlist_t *nvprops = zct->zct_props;
153 uint64_t volblocksize, volsize;
155 VERIFY(nvlist_lookup_uint64(nvprops,
156 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
157 if (nvlist_lookup_uint64(nvprops,
158 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
159 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
162 * These properties must be removed from the list so the generic
163 * property setting step won't apply to them.
165 VERIFY(nvlist_remove_all(nvprops,
166 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
167 (void) nvlist_remove_all(nvprops,
168 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
170 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
174 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
178 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
183 * ZFS_IOC_OBJSET_STATS entry point.
186 zvol_get_stats(objset_t *os, nvlist_t *nv)
189 dmu_object_info_t *doi;
192 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
196 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
197 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
198 error = dmu_object_info(os, ZVOL_OBJ, doi);
201 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
202 doi->doi_data_block_size);
205 kmem_free(doi, sizeof(dmu_object_info_t));
211 * Sanity check volume size.
214 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
219 if (volsize % blocksize != 0)
223 if (volsize - 1 > MAXOFFSET_T)
230 * Ensure the zap is flushed then inform the VFS of the capacity change.
233 zvol_update_volsize(zvol_state_t *zv, uint64_t volsize, objset_t *os)
235 struct block_device *bdev;
239 ASSERT(MUTEX_HELD(&zvol_state_lock));
241 tx = dmu_tx_create(os);
242 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
243 error = dmu_tx_assign(tx, TXG_WAIT);
249 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
256 error = dmu_free_long_range(os,
257 ZVOL_OBJ, volsize, DMU_OBJECT_END);
261 bdev = bdget_disk(zv->zv_disk, 0);
266 * Added check_disk_size_change() helper function.
268 #ifdef HAVE_CHECK_DISK_SIZE_CHANGE
269 set_capacity(zv->zv_disk, volsize >> 9);
270 zv->zv_volsize = volsize;
271 check_disk_size_change(zv->zv_disk, bdev);
273 zv->zv_volsize = volsize;
275 (void) check_disk_change(bdev);
276 #endif /* HAVE_CHECK_DISK_SIZE_CHANGE */
284 * Set ZFS_PROP_VOLSIZE set entry point.
287 zvol_set_volsize(const char *name, uint64_t volsize)
290 dmu_object_info_t *doi;
295 mutex_enter(&zvol_state_lock);
297 zv = zvol_find_by_name(name);
303 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
305 error = dmu_objset_hold(name, FTAG, &os);
309 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) != 0 ||
310 (error = zvol_check_volsize(volsize,doi->doi_data_block_size)) != 0)
313 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, NULL) == 0);
319 if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) {
324 error = zvol_update_volsize(zv, volsize, os);
326 kmem_free(doi, sizeof(dmu_object_info_t));
329 dmu_objset_rele(os, FTAG);
331 mutex_exit(&zvol_state_lock);
337 * Sanity check volume block size.
340 zvol_check_volblocksize(uint64_t volblocksize)
342 if (volblocksize < SPA_MINBLOCKSIZE ||
343 volblocksize > SPA_MAXBLOCKSIZE ||
351 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
354 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
360 mutex_enter(&zvol_state_lock);
362 zv = zvol_find_by_name(name);
368 if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) {
373 tx = dmu_tx_create(zv->zv_objset);
374 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
375 error = dmu_tx_assign(tx, TXG_WAIT);
379 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
380 volblocksize, 0, tx);
381 if (error == ENOTSUP)
385 zv->zv_volblocksize = volblocksize;
388 mutex_exit(&zvol_state_lock);
394 * Replay a TX_WRITE ZIL transaction that didn't get committed
395 * after a system failure
398 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
400 objset_t *os = zv->zv_objset;
401 char *data = (char *)(lr + 1); /* data follows lr_write_t */
402 uint64_t off = lr->lr_offset;
403 uint64_t len = lr->lr_length;
408 byteswap_uint64_array(lr, sizeof (*lr));
410 tx = dmu_tx_create(os);
411 dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
412 error = dmu_tx_assign(tx, TXG_WAIT);
416 dmu_write(os, ZVOL_OBJ, off, len, data, tx);
424 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
430 * Callback vectors for replaying records.
431 * Only TX_WRITE is needed for zvol.
433 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
434 (zil_replay_func_t *)zvol_replay_err, /* no such transaction type */
435 (zil_replay_func_t *)zvol_replay_err, /* TX_CREATE */
436 (zil_replay_func_t *)zvol_replay_err, /* TX_MKDIR */
437 (zil_replay_func_t *)zvol_replay_err, /* TX_MKXATTR */
438 (zil_replay_func_t *)zvol_replay_err, /* TX_SYMLINK */
439 (zil_replay_func_t *)zvol_replay_err, /* TX_REMOVE */
440 (zil_replay_func_t *)zvol_replay_err, /* TX_RMDIR */
441 (zil_replay_func_t *)zvol_replay_err, /* TX_LINK */
442 (zil_replay_func_t *)zvol_replay_err, /* TX_RENAME */
443 (zil_replay_func_t *)zvol_replay_write, /* TX_WRITE */
444 (zil_replay_func_t *)zvol_replay_err, /* TX_TRUNCATE */
445 (zil_replay_func_t *)zvol_replay_err, /* TX_SETATTR */
446 (zil_replay_func_t *)zvol_replay_err, /* TX_ACL */
450 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
452 * We store data in the log buffers if it's small enough.
453 * Otherwise we will later flush the data out via dmu_sync().
455 ssize_t zvol_immediate_write_sz = 32768;
458 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx,
459 uint64_t offset, uint64_t size, int sync)
461 uint32_t blocksize = zv->zv_volblocksize;
462 zilog_t *zilog = zv->zv_zilog;
464 ssize_t immediate_write_sz;
466 if (zil_replaying(zilog, tx))
469 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
470 ? 0 : zvol_immediate_write_sz;
471 slogging = spa_has_slogs(zilog->zl_spa) &&
472 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
478 itx_wr_state_t write_state;
481 * Unlike zfs_log_write() we can be called with
482 * up to DMU_MAX_ACCESS/2 (5MB) writes.
484 if (blocksize > immediate_write_sz && !slogging &&
485 size >= blocksize && offset % blocksize == 0) {
486 write_state = WR_INDIRECT; /* uses dmu_sync */
489 write_state = WR_COPIED;
490 len = MIN(ZIL_MAX_LOG_DATA, size);
492 write_state = WR_NEED_COPY;
493 len = MIN(ZIL_MAX_LOG_DATA, size);
496 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
497 (write_state == WR_COPIED ? len : 0));
498 lr = (lr_write_t *)&itx->itx_lr;
499 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
500 ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
501 zil_itx_destroy(itx);
502 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
503 lr = (lr_write_t *)&itx->itx_lr;
504 write_state = WR_NEED_COPY;
507 itx->itx_wr_state = write_state;
508 if (write_state == WR_NEED_COPY)
510 lr->lr_foid = ZVOL_OBJ;
511 lr->lr_offset = offset;
514 BP_ZERO(&lr->lr_blkptr);
516 itx->itx_private = zv;
517 itx->itx_sync = sync;
519 (void) zil_itx_assign(zilog, itx, tx);
527 * Common write path running under the zvol taskq context. This function
528 * is responsible for copying the request structure data in to the DMU and
529 * signaling the request queue with the result of the copy.
532 zvol_write(void *arg)
534 struct request *req = (struct request *)arg;
535 struct request_queue *q = req->q;
536 zvol_state_t *zv = q->queuedata;
537 uint64_t offset = blk_rq_pos(req) << 9;
538 uint64_t size = blk_rq_bytes(req);
544 * Annotate this call path with a flag that indicates that it is
545 * unsafe to use KM_SLEEP during memory allocations due to the
546 * potential for a deadlock. KM_PUSHPAGE should be used instead.
548 ASSERT(!(current->flags & PF_NOFS));
549 current->flags |= PF_NOFS;
551 if (req->cmd_flags & VDEV_REQ_FLUSH)
552 zil_commit(zv->zv_zilog, ZVOL_OBJ);
555 * Some requests are just for flush and nothing else.
558 blk_end_request(req, 0, size);
562 rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
564 tx = dmu_tx_create(zv->zv_objset);
565 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
567 /* This will only fail for ENOSPC */
568 error = dmu_tx_assign(tx, TXG_WAIT);
571 zfs_range_unlock(rl);
572 blk_end_request(req, -error, size);
576 error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx);
578 zvol_log_write(zv, tx, offset, size,
579 req->cmd_flags & VDEV_REQ_FUA);
582 zfs_range_unlock(rl);
584 if ((req->cmd_flags & VDEV_REQ_FUA) ||
585 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
586 zil_commit(zv->zv_zilog, ZVOL_OBJ);
588 blk_end_request(req, -error, size);
590 current->flags &= ~PF_NOFS;
593 #ifdef HAVE_BLK_QUEUE_DISCARD
595 zvol_discard(void *arg)
597 struct request *req = (struct request *)arg;
598 struct request_queue *q = req->q;
599 zvol_state_t *zv = q->queuedata;
600 uint64_t start = blk_rq_pos(req) << 9;
601 uint64_t end = start + blk_rq_bytes(req);
606 * Annotate this call path with a flag that indicates that it is
607 * unsafe to use KM_SLEEP during memory allocations due to the
608 * potential for a deadlock. KM_PUSHPAGE should be used instead.
610 ASSERT(!(current->flags & PF_NOFS));
611 current->flags |= PF_NOFS;
613 if (end > zv->zv_volsize) {
614 blk_end_request(req, -EIO, blk_rq_bytes(req));
619 * Align the request to volume block boundaries. If we don't,
620 * then this will force dnode_free_range() to zero out the
621 * unaligned parts, which is slow (read-modify-write) and
622 * useless since we are not freeing any space by doing so.
624 start = P2ROUNDUP(start, zv->zv_volblocksize);
625 end = P2ALIGN(end, zv->zv_volblocksize);
628 blk_end_request(req, 0, blk_rq_bytes(req));
632 rl = zfs_range_lock(&zv->zv_znode, start, end - start, RL_WRITER);
634 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, start, end - start);
637 * TODO: maybe we should add the operation to the log.
640 zfs_range_unlock(rl);
642 blk_end_request(req, -error, blk_rq_bytes(req));
644 current->flags &= ~PF_NOFS;
646 #endif /* HAVE_BLK_QUEUE_DISCARD */
649 * Common read path running under the zvol taskq context. This function
650 * is responsible for copying the requested data out of the DMU and in to
651 * a linux request structure. It then must signal the request queue with
652 * an error code describing the result of the copy.
657 struct request *req = (struct request *)arg;
658 struct request_queue *q = req->q;
659 zvol_state_t *zv = q->queuedata;
660 uint64_t offset = blk_rq_pos(req) << 9;
661 uint64_t size = blk_rq_bytes(req);
666 blk_end_request(req, 0, size);
670 rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
672 error = dmu_read_req(zv->zv_objset, ZVOL_OBJ, req);
674 zfs_range_unlock(rl);
676 /* convert checksum errors into IO errors */
680 blk_end_request(req, -error, size);
684 * Request will be added back to the request queue and retried if
685 * it cannot be immediately dispatched to the taskq for handling
688 zvol_dispatch(task_func_t func, struct request *req)
690 if (!taskq_dispatch(zvol_taskq, func, (void *)req, TQ_NOSLEEP))
691 blk_requeue_request(req->q, req);
695 * Common request path. Rather than registering a custom make_request()
696 * function we use the generic Linux version. This is done because it allows
697 * us to easily merge read requests which would otherwise we performed
698 * synchronously by the DMU. This is less critical in write case where the
699 * DMU will perform the correct merging within a transaction group. Using
700 * the generic make_request() also let's use leverage the fact that the
701 * elevator with ensure correct ordering in regards to barrior IOs. On
702 * the downside it means that in the write case we end up doing request
703 * merging twice once in the elevator and once in the DMU.
705 * The request handler is called under a spin lock so all the real work
706 * is handed off to be done in the context of the zvol taskq. This function
707 * simply performs basic request sanity checking and hands off the request.
710 zvol_request(struct request_queue *q)
712 zvol_state_t *zv = q->queuedata;
716 while ((req = blk_fetch_request(q)) != NULL) {
717 size = blk_rq_bytes(req);
719 if (size != 0 && blk_rq_pos(req) + blk_rq_sectors(req) >
720 get_capacity(zv->zv_disk)) {
722 "%s: bad access: block=%llu, count=%lu\n",
723 req->rq_disk->disk_name,
724 (long long unsigned)blk_rq_pos(req),
725 (long unsigned)blk_rq_sectors(req));
726 __blk_end_request(req, -EIO, size);
730 if (!blk_fs_request(req)) {
731 printk(KERN_INFO "%s: non-fs cmd\n",
732 req->rq_disk->disk_name);
733 __blk_end_request(req, -EIO, size);
737 switch (rq_data_dir(req)) {
739 zvol_dispatch(zvol_read, req);
742 if (unlikely(get_disk_ro(zv->zv_disk)) ||
743 unlikely(zv->zv_flags & ZVOL_RDONLY)) {
744 __blk_end_request(req, -EROFS, size);
748 #ifdef HAVE_BLK_QUEUE_DISCARD
749 if (req->cmd_flags & VDEV_REQ_DISCARD) {
750 zvol_dispatch(zvol_discard, req);
753 #endif /* HAVE_BLK_QUEUE_DISCARD */
755 zvol_dispatch(zvol_write, req);
758 printk(KERN_INFO "%s: unknown cmd: %d\n",
759 req->rq_disk->disk_name, (int)rq_data_dir(req));
760 __blk_end_request(req, -EIO, size);
767 zvol_get_done(zgd_t *zgd, int error)
770 dmu_buf_rele(zgd->zgd_db, zgd);
772 zfs_range_unlock(zgd->zgd_rl);
774 if (error == 0 && zgd->zgd_bp)
775 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
777 kmem_free(zgd, sizeof (zgd_t));
781 * Get data to generate a TX_WRITE intent log record.
784 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
786 zvol_state_t *zv = arg;
787 objset_t *os = zv->zv_objset;
788 uint64_t offset = lr->lr_offset;
789 uint64_t size = lr->lr_length;
797 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE);
798 zgd->zgd_zilog = zv->zv_zilog;
799 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
802 * Write records come in two flavors: immediate and indirect.
803 * For small writes it's cheaper to store the data with the
804 * log record (immediate); for large writes it's cheaper to
805 * sync the data and get a pointer to it (indirect) so that
806 * we don't have to write the data twice.
808 if (buf != NULL) { /* immediate write */
809 error = dmu_read(os, ZVOL_OBJ, offset, size, buf,
810 DMU_READ_NO_PREFETCH);
812 size = zv->zv_volblocksize;
813 offset = P2ALIGN_TYPED(offset, size, uint64_t);
814 error = dmu_buf_hold(os, ZVOL_OBJ, offset, zgd, &db,
815 DMU_READ_NO_PREFETCH);
818 zgd->zgd_bp = &lr->lr_blkptr;
821 ASSERT(db->db_offset == offset);
822 ASSERT(db->db_size == size);
824 error = dmu_sync(zio, lr->lr_common.lrc_txg,
832 zvol_get_done(zgd, error);
838 * The zvol_state_t's are inserted in increasing MINOR(dev_t) order.
841 zvol_insert(zvol_state_t *zv_insert)
843 zvol_state_t *zv = NULL;
845 ASSERT(MUTEX_HELD(&zvol_state_lock));
846 ASSERT3U(MINOR(zv_insert->zv_dev) & ZVOL_MINOR_MASK, ==, 0);
847 for (zv = list_head(&zvol_state_list); zv != NULL;
848 zv = list_next(&zvol_state_list, zv)) {
849 if (MINOR(zv->zv_dev) > MINOR(zv_insert->zv_dev))
853 list_insert_before(&zvol_state_list, zv, zv_insert);
857 * Simply remove the zvol from to list of zvols.
860 zvol_remove(zvol_state_t *zv_remove)
862 ASSERT(MUTEX_HELD(&zvol_state_lock));
863 list_remove(&zvol_state_list, zv_remove);
867 zvol_first_open(zvol_state_t *zv)
874 /* lie and say we're read-only */
875 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
879 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
881 dmu_objset_disown(os, zvol_tag);
886 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
888 dmu_objset_disown(os, zvol_tag);
892 set_capacity(zv->zv_disk, volsize >> 9);
893 zv->zv_volsize = volsize;
894 zv->zv_zilog = zil_open(os, zvol_get_data);
896 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0);
897 if (ro || dmu_objset_is_snapshot(os)) {
898 set_disk_ro(zv->zv_disk, 1);
899 zv->zv_flags |= ZVOL_RDONLY;
901 set_disk_ro(zv->zv_disk, 0);
902 zv->zv_flags &= ~ZVOL_RDONLY;
909 zvol_last_close(zvol_state_t *zv)
911 zil_close(zv->zv_zilog);
914 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
920 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
921 !(zv->zv_flags & ZVOL_RDONLY))
922 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
923 (void) dmu_objset_evict_dbufs(zv->zv_objset);
925 dmu_objset_disown(zv->zv_objset, zvol_tag);
926 zv->zv_objset = NULL;
930 zvol_open(struct block_device *bdev, fmode_t flag)
932 zvol_state_t *zv = bdev->bd_disk->private_data;
933 int error = 0, drop_mutex = 0;
936 * If the caller is already holding the mutex do not take it
937 * again, this will happen as part of zvol_create_minor().
938 * Once add_disk() is called the device is live and the kernel
939 * will attempt to open it to read the partition information.
941 if (!mutex_owned(&zvol_state_lock)) {
942 mutex_enter(&zvol_state_lock);
946 ASSERT3P(zv, !=, NULL);
948 if (zv->zv_open_count == 0) {
949 error = zvol_first_open(zv);
954 if ((flag & FMODE_WRITE) &&
955 (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY))) {
963 if (zv->zv_open_count == 0)
968 mutex_exit(&zvol_state_lock);
970 check_disk_change(bdev);
976 zvol_release(struct gendisk *disk, fmode_t mode)
978 zvol_state_t *zv = disk->private_data;
981 if (!mutex_owned(&zvol_state_lock)) {
982 mutex_enter(&zvol_state_lock);
986 ASSERT3P(zv, !=, NULL);
987 ASSERT3U(zv->zv_open_count, >, 0);
989 if (zv->zv_open_count == 0)
993 mutex_exit(&zvol_state_lock);
999 zvol_ioctl(struct block_device *bdev, fmode_t mode,
1000 unsigned int cmd, unsigned long arg)
1002 zvol_state_t *zv = bdev->bd_disk->private_data;
1010 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1013 error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN);
1025 #ifdef CONFIG_COMPAT
1027 zvol_compat_ioctl(struct block_device *bdev, fmode_t mode,
1028 unsigned cmd, unsigned long arg)
1030 return zvol_ioctl(bdev, mode, cmd, arg);
1033 #define zvol_compat_ioctl NULL
1036 static int zvol_media_changed(struct gendisk *disk)
1038 zvol_state_t *zv = disk->private_data;
1040 return zv->zv_changed;
1043 static int zvol_revalidate_disk(struct gendisk *disk)
1045 zvol_state_t *zv = disk->private_data;
1048 set_capacity(zv->zv_disk, zv->zv_volsize >> 9);
1054 * Provide a simple virtual geometry for legacy compatibility. For devices
1055 * smaller than 1 MiB a small head and sector count is used to allow very
1056 * tiny devices. For devices over 1 Mib a standard head and sector count
1057 * is used to keep the cylinders count reasonable.
1060 zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1062 zvol_state_t *zv = bdev->bd_disk->private_data;
1063 sector_t sectors = get_capacity(zv->zv_disk);
1065 if (sectors > 2048) {
1074 geo->cylinders = sectors / (geo->heads * geo->sectors);
1079 static struct kobject *
1080 zvol_probe(dev_t dev, int *part, void *arg)
1083 struct kobject *kobj;
1085 mutex_enter(&zvol_state_lock);
1086 zv = zvol_find_by_dev(dev);
1087 kobj = zv ? get_disk(zv->zv_disk) : NULL;
1088 mutex_exit(&zvol_state_lock);
1093 #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS
1094 static struct block_device_operations zvol_ops = {
1096 .release = zvol_release,
1097 .ioctl = zvol_ioctl,
1098 .compat_ioctl = zvol_compat_ioctl,
1099 .media_changed = zvol_media_changed,
1100 .revalidate_disk = zvol_revalidate_disk,
1101 .getgeo = zvol_getgeo,
1102 .owner = THIS_MODULE,
1105 #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1108 zvol_open_by_inode(struct inode *inode, struct file *file)
1110 return zvol_open(inode->i_bdev, file->f_mode);
1114 zvol_release_by_inode(struct inode *inode, struct file *file)
1116 return zvol_release(inode->i_bdev->bd_disk, file->f_mode);
1120 zvol_ioctl_by_inode(struct inode *inode, struct file *file,
1121 unsigned int cmd, unsigned long arg)
1123 if (file == NULL || inode == NULL)
1125 return zvol_ioctl(inode->i_bdev, file->f_mode, cmd, arg);
1128 # ifdef CONFIG_COMPAT
1130 zvol_compat_ioctl_by_inode(struct file *file,
1131 unsigned int cmd, unsigned long arg)
1135 return zvol_compat_ioctl(file->f_dentry->d_inode->i_bdev,
1136 file->f_mode, cmd, arg);
1139 # define zvol_compat_ioctl_by_inode NULL
1142 static struct block_device_operations zvol_ops = {
1143 .open = zvol_open_by_inode,
1144 .release = zvol_release_by_inode,
1145 .ioctl = zvol_ioctl_by_inode,
1146 .compat_ioctl = zvol_compat_ioctl_by_inode,
1147 .media_changed = zvol_media_changed,
1148 .revalidate_disk = zvol_revalidate_disk,
1149 .getgeo = zvol_getgeo,
1150 .owner = THIS_MODULE,
1152 #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1155 * Allocate memory for a new zvol_state_t and setup the required
1156 * request queue and generic disk structures for the block device.
1158 static zvol_state_t *
1159 zvol_alloc(dev_t dev, const char *name)
1164 zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
1168 zv->zv_queue = blk_init_queue(zvol_request, &zv->zv_lock);
1169 if (zv->zv_queue == NULL)
1172 #ifdef HAVE_ELEVATOR_CHANGE
1173 error = elevator_change(zv->zv_queue, "noop");
1174 #endif /* HAVE_ELEVATOR_CHANGE */
1176 printk("ZFS: Unable to set \"%s\" scheduler for zvol %s: %d\n",
1177 "noop", name, error);
1181 #ifdef HAVE_BLK_QUEUE_FLUSH
1182 blk_queue_flush(zv->zv_queue, VDEV_REQ_FLUSH | VDEV_REQ_FUA);
1184 blk_queue_ordered(zv->zv_queue, QUEUE_ORDERED_DRAIN, NULL);
1185 #endif /* HAVE_BLK_QUEUE_FLUSH */
1187 zv->zv_disk = alloc_disk(ZVOL_MINORS);
1188 if (zv->zv_disk == NULL)
1191 zv->zv_queue->queuedata = zv;
1193 zv->zv_open_count = 0;
1194 strlcpy(zv->zv_name, name, MAXNAMELEN);
1196 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
1197 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
1198 sizeof (rl_t), offsetof(rl_t, r_node));
1199 zv->zv_znode.z_is_zvol = TRUE;
1201 spin_lock_init(&zv->zv_lock);
1202 list_link_init(&zv->zv_next);
1204 zv->zv_disk->major = zvol_major;
1205 zv->zv_disk->first_minor = (dev & MINORMASK);
1206 zv->zv_disk->fops = &zvol_ops;
1207 zv->zv_disk->private_data = zv;
1208 zv->zv_disk->queue = zv->zv_queue;
1209 snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s%d",
1210 ZVOL_DEV_NAME, (dev & MINORMASK));
1215 blk_cleanup_queue(zv->zv_queue);
1217 kmem_free(zv, sizeof (zvol_state_t));
1223 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1226 zvol_free(zvol_state_t *zv)
1228 avl_destroy(&zv->zv_znode.z_range_avl);
1229 mutex_destroy(&zv->zv_znode.z_range_lock);
1231 del_gendisk(zv->zv_disk);
1232 blk_cleanup_queue(zv->zv_queue);
1233 put_disk(zv->zv_disk);
1235 kmem_free(zv, sizeof (zvol_state_t));
1239 __zvol_create_minor(const char *name)
1243 dmu_object_info_t *doi;
1248 ASSERT(MUTEX_HELD(&zvol_state_lock));
1250 zv = zvol_find_by_name(name);
1256 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
1258 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os);
1262 error = dmu_object_info(os, ZVOL_OBJ, doi);
1264 goto out_dmu_objset_disown;
1266 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1268 goto out_dmu_objset_disown;
1270 error = zvol_find_minor(&minor);
1272 goto out_dmu_objset_disown;
1274 zv = zvol_alloc(MKDEV(zvol_major, minor), name);
1277 goto out_dmu_objset_disown;
1280 if (dmu_objset_is_snapshot(os))
1281 zv->zv_flags |= ZVOL_RDONLY;
1283 zv->zv_volblocksize = doi->doi_data_block_size;
1284 zv->zv_volsize = volsize;
1287 set_capacity(zv->zv_disk, zv->zv_volsize >> 9);
1289 blk_queue_max_hw_sectors(zv->zv_queue, UINT_MAX);
1290 blk_queue_max_segments(zv->zv_queue, UINT16_MAX);
1291 blk_queue_max_segment_size(zv->zv_queue, UINT_MAX);
1292 blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize);
1293 blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize);
1294 #ifdef HAVE_BLK_QUEUE_DISCARD
1295 blk_queue_max_discard_sectors(zv->zv_queue,
1296 (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9);
1297 blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize);
1298 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue);
1300 #ifdef HAVE_BLK_QUEUE_NONROT
1301 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue);
1304 if (zil_replay_disable)
1305 zil_destroy(dmu_objset_zil(os), B_FALSE);
1307 zil_replay(os, zv, zvol_replay_vector);
1309 out_dmu_objset_disown:
1310 dmu_objset_disown(os, zvol_tag);
1311 zv->zv_objset = NULL;
1313 kmem_free(doi, sizeof(dmu_object_info_t));
1318 add_disk(zv->zv_disk);
1325 * Create a block device minor node and setup the linkage between it
1326 * and the specified volume. Once this function returns the block
1327 * device is live and ready for use.
1330 zvol_create_minor(const char *name)
1334 mutex_enter(&zvol_state_lock);
1335 error = __zvol_create_minor(name);
1336 mutex_exit(&zvol_state_lock);
1342 __zvol_remove_minor(const char *name)
1346 ASSERT(MUTEX_HELD(&zvol_state_lock));
1348 zv = zvol_find_by_name(name);
1352 if (zv->zv_open_count > 0)
1362 * Remove a block device minor node for the specified volume.
1365 zvol_remove_minor(const char *name)
1369 mutex_enter(&zvol_state_lock);
1370 error = __zvol_remove_minor(name);
1371 mutex_exit(&zvol_state_lock);
1377 zvol_create_minors_cb(spa_t *spa, uint64_t dsobj,
1378 const char *dsname, void *arg)
1380 if (strchr(dsname, '/') == NULL)
1383 (void) __zvol_create_minor(dsname);
1388 * Create minors for specified pool, if pool is NULL create minors
1389 * for all available pools.
1392 zvol_create_minors(const char *pool)
1397 if (zvol_inhibit_dev)
1400 mutex_enter(&zvol_state_lock);
1402 error = dmu_objset_find_spa(NULL, pool, zvol_create_minors_cb,
1403 NULL, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1405 mutex_enter(&spa_namespace_lock);
1406 while ((spa = spa_next(spa)) != NULL) {
1407 error = dmu_objset_find_spa(NULL,
1408 spa_name(spa), zvol_create_minors_cb, NULL,
1409 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1413 mutex_exit(&spa_namespace_lock);
1415 mutex_exit(&zvol_state_lock);
1421 * Remove minors for specified pool, if pool is NULL remove all minors.
1424 zvol_remove_minors(const char *pool)
1426 zvol_state_t *zv, *zv_next;
1429 if (zvol_inhibit_dev)
1432 str = kmem_zalloc(MAXNAMELEN, KM_SLEEP);
1434 (void) strncpy(str, pool, strlen(pool));
1435 (void) strcat(str, "/");
1438 mutex_enter(&zvol_state_lock);
1439 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1440 zv_next = list_next(&zvol_state_list, zv);
1442 if (pool == NULL || !strncmp(str, zv->zv_name, strlen(str))) {
1447 mutex_exit(&zvol_state_lock);
1448 kmem_free(str, MAXNAMELEN);
1456 zvol_taskq = taskq_create(ZVOL_DRIVER, zvol_threads, maxclsyspri,
1457 zvol_threads, INT_MAX, TASKQ_PREPOPULATE);
1458 if (zvol_taskq == NULL) {
1459 printk(KERN_INFO "ZFS: taskq_create() failed\n");
1463 error = register_blkdev(zvol_major, ZVOL_DRIVER);
1465 printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error);
1466 taskq_destroy(zvol_taskq);
1470 blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS,
1471 THIS_MODULE, zvol_probe, NULL, NULL);
1473 mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1474 list_create(&zvol_state_list, sizeof (zvol_state_t),
1475 offsetof(zvol_state_t, zv_next));
1477 (void) zvol_create_minors(NULL);
1485 zvol_remove_minors(NULL);
1486 blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS);
1487 unregister_blkdev(zvol_major, ZVOL_DRIVER);
1488 taskq_destroy(zvol_taskq);
1489 mutex_destroy(&zvol_state_lock);
1490 list_destroy(&zvol_state_list);
1493 module_param(zvol_inhibit_dev, uint, 0644);
1494 MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes");
1496 module_param(zvol_major, uint, 0444);
1497 MODULE_PARM_DESC(zvol_major, "Major number for zvol device");
1499 module_param(zvol_threads, uint, 0444);
1500 MODULE_PARM_DESC(zvol_threads, "Number of threads for zvol device");
1502 module_param(zvol_max_discard_blocks, ulong, 0444);
1503 MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard at once");