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;
53 static taskq_t *zvol_taskq;
54 static kmutex_t zvol_state_lock;
55 static list_t zvol_state_list;
56 static char *zvol_tag = "zvol_tag";
59 * The in-core state of each volume.
61 typedef struct zvol_state {
62 char zv_name[MAXNAMELEN]; /* name */
63 uint64_t zv_volsize; /* advertised space */
64 uint64_t zv_volblocksize;/* volume block size */
65 objset_t *zv_objset; /* objset handle */
66 uint32_t zv_flags; /* ZVOL_* flags */
67 uint32_t zv_open_count; /* open counts */
68 uint32_t zv_changed; /* disk changed */
69 zilog_t *zv_zilog; /* ZIL handle */
70 znode_t zv_znode; /* for range locking */
71 dmu_buf_t *zv_dbuf; /* bonus handle */
72 dev_t zv_dev; /* device id */
73 struct gendisk *zv_disk; /* generic disk */
74 struct request_queue *zv_queue; /* request queue */
75 spinlock_t zv_lock; /* request queue lock */
76 list_node_t zv_next; /* next zvol_state_t linkage */
79 #define ZVOL_RDONLY 0x1
82 * Find the next available range of ZVOL_MINORS minor numbers. The
83 * zvol_state_list is kept in ascending minor order so we simply need
84 * to scan the list for the first gap in the sequence. This allows us
85 * to recycle minor number as devices are created and removed.
88 zvol_find_minor(unsigned *minor)
93 ASSERT(MUTEX_HELD(&zvol_state_lock));
94 for (zv = list_head(&zvol_state_list); zv != NULL;
95 zv = list_next(&zvol_state_list, zv), *minor += ZVOL_MINORS) {
96 if (MINOR(zv->zv_dev) != MINOR(*minor))
100 /* All minors are in use */
101 if (*minor >= (1 << MINORBITS))
108 * Find a zvol_state_t given the full major+minor dev_t.
110 static zvol_state_t *
111 zvol_find_by_dev(dev_t dev)
115 ASSERT(MUTEX_HELD(&zvol_state_lock));
116 for (zv = list_head(&zvol_state_list); zv != NULL;
117 zv = list_next(&zvol_state_list, zv)) {
118 if (zv->zv_dev == dev)
126 * Find a zvol_state_t given the name provided at zvol_alloc() time.
128 static zvol_state_t *
129 zvol_find_by_name(const char *name)
133 ASSERT(MUTEX_HELD(&zvol_state_lock));
134 for (zv = list_head(&zvol_state_list); zv != NULL;
135 zv = list_next(&zvol_state_list, zv)) {
136 if (!strncmp(zv->zv_name, name, MAXNAMELEN))
144 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
147 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
149 zfs_creat_t *zct = arg;
150 nvlist_t *nvprops = zct->zct_props;
152 uint64_t volblocksize, volsize;
154 VERIFY(nvlist_lookup_uint64(nvprops,
155 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
156 if (nvlist_lookup_uint64(nvprops,
157 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
158 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
161 * These properties must be removed from the list so the generic
162 * property setting step won't apply to them.
164 VERIFY(nvlist_remove_all(nvprops,
165 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
166 (void) nvlist_remove_all(nvprops,
167 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
169 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
173 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
177 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
182 * ZFS_IOC_OBJSET_STATS entry point.
185 zvol_get_stats(objset_t *os, nvlist_t *nv)
188 dmu_object_info_t *doi;
191 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
195 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
196 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
197 error = dmu_object_info(os, ZVOL_OBJ, doi);
200 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
201 doi->doi_data_block_size);
204 kmem_free(doi, sizeof(dmu_object_info_t));
210 * Sanity check volume size.
213 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
218 if (volsize % blocksize != 0)
222 if (volsize - 1 > MAXOFFSET_T)
229 * Ensure the zap is flushed then inform the VFS of the capacity change.
232 zvol_update_volsize(zvol_state_t *zv, uint64_t volsize, objset_t *os)
234 struct block_device *bdev;
238 ASSERT(MUTEX_HELD(&zvol_state_lock));
240 tx = dmu_tx_create(os);
241 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
242 error = dmu_tx_assign(tx, TXG_WAIT);
248 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
255 error = dmu_free_long_range(os,
256 ZVOL_OBJ, volsize, DMU_OBJECT_END);
260 bdev = bdget_disk(zv->zv_disk, 0);
265 * Added check_disk_size_change() helper function.
267 #ifdef HAVE_CHECK_DISK_SIZE_CHANGE
268 set_capacity(zv->zv_disk, volsize >> 9);
269 zv->zv_volsize = volsize;
270 check_disk_size_change(zv->zv_disk, bdev);
272 zv->zv_volsize = volsize;
274 (void) check_disk_change(bdev);
275 #endif /* HAVE_CHECK_DISK_SIZE_CHANGE */
283 * Set ZFS_PROP_VOLSIZE set entry point.
286 zvol_set_volsize(const char *name, uint64_t volsize)
289 dmu_object_info_t *doi;
294 mutex_enter(&zvol_state_lock);
296 zv = zvol_find_by_name(name);
302 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
304 error = dmu_objset_hold(name, FTAG, &os);
308 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) != 0 ||
309 (error = zvol_check_volsize(volsize,doi->doi_data_block_size)) != 0)
312 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, NULL) == 0);
318 if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) {
323 error = zvol_update_volsize(zv, volsize, os);
325 kmem_free(doi, sizeof(dmu_object_info_t));
328 dmu_objset_rele(os, FTAG);
330 mutex_exit(&zvol_state_lock);
336 * Sanity check volume block size.
339 zvol_check_volblocksize(uint64_t volblocksize)
341 if (volblocksize < SPA_MINBLOCKSIZE ||
342 volblocksize > SPA_MAXBLOCKSIZE ||
350 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
353 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
359 mutex_enter(&zvol_state_lock);
361 zv = zvol_find_by_name(name);
367 if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) {
372 tx = dmu_tx_create(zv->zv_objset);
373 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
374 error = dmu_tx_assign(tx, TXG_WAIT);
378 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
379 volblocksize, 0, tx);
380 if (error == ENOTSUP)
384 zv->zv_volblocksize = volblocksize;
387 mutex_exit(&zvol_state_lock);
393 * Replay a TX_WRITE ZIL transaction that didn't get committed
394 * after a system failure
397 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
399 objset_t *os = zv->zv_objset;
400 char *data = (char *)(lr + 1); /* data follows lr_write_t */
401 uint64_t off = lr->lr_offset;
402 uint64_t len = lr->lr_length;
407 byteswap_uint64_array(lr, sizeof (*lr));
409 tx = dmu_tx_create(os);
410 dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
411 error = dmu_tx_assign(tx, TXG_WAIT);
415 dmu_write(os, ZVOL_OBJ, off, len, data, tx);
423 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
429 * Callback vectors for replaying records.
430 * Only TX_WRITE is needed for zvol.
432 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
433 (zil_replay_func_t *)zvol_replay_err, /* no such transaction type */
434 (zil_replay_func_t *)zvol_replay_err, /* TX_CREATE */
435 (zil_replay_func_t *)zvol_replay_err, /* TX_MKDIR */
436 (zil_replay_func_t *)zvol_replay_err, /* TX_MKXATTR */
437 (zil_replay_func_t *)zvol_replay_err, /* TX_SYMLINK */
438 (zil_replay_func_t *)zvol_replay_err, /* TX_REMOVE */
439 (zil_replay_func_t *)zvol_replay_err, /* TX_RMDIR */
440 (zil_replay_func_t *)zvol_replay_err, /* TX_LINK */
441 (zil_replay_func_t *)zvol_replay_err, /* TX_RENAME */
442 (zil_replay_func_t *)zvol_replay_write, /* TX_WRITE */
443 (zil_replay_func_t *)zvol_replay_err, /* TX_TRUNCATE */
444 (zil_replay_func_t *)zvol_replay_err, /* TX_SETATTR */
445 (zil_replay_func_t *)zvol_replay_err, /* TX_ACL */
449 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
451 * We store data in the log buffers if it's small enough.
452 * Otherwise we will later flush the data out via dmu_sync().
454 ssize_t zvol_immediate_write_sz = 32768;
457 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx,
458 uint64_t offset, uint64_t size, int sync)
460 uint32_t blocksize = zv->zv_volblocksize;
461 zilog_t *zilog = zv->zv_zilog;
463 ssize_t immediate_write_sz;
465 if (zil_replaying(zilog, tx))
468 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
469 ? 0 : zvol_immediate_write_sz;
470 slogging = spa_has_slogs(zilog->zl_spa) &&
471 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
477 itx_wr_state_t write_state;
480 * Unlike zfs_log_write() we can be called with
481 * up to DMU_MAX_ACCESS/2 (5MB) writes.
483 if (blocksize > immediate_write_sz && !slogging &&
484 size >= blocksize && offset % blocksize == 0) {
485 write_state = WR_INDIRECT; /* uses dmu_sync */
488 write_state = WR_COPIED;
489 len = MIN(ZIL_MAX_LOG_DATA, size);
491 write_state = WR_NEED_COPY;
492 len = MIN(ZIL_MAX_LOG_DATA, size);
495 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
496 (write_state == WR_COPIED ? len : 0));
497 lr = (lr_write_t *)&itx->itx_lr;
498 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
499 ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
500 zil_itx_destroy(itx);
501 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
502 lr = (lr_write_t *)&itx->itx_lr;
503 write_state = WR_NEED_COPY;
506 itx->itx_wr_state = write_state;
507 if (write_state == WR_NEED_COPY)
509 lr->lr_foid = ZVOL_OBJ;
510 lr->lr_offset = offset;
513 BP_ZERO(&lr->lr_blkptr);
515 itx->itx_private = zv;
516 itx->itx_sync = sync;
518 (void) zil_itx_assign(zilog, itx, tx);
526 * Common write path running under the zvol taskq context. This function
527 * is responsible for copying the request structure data in to the DMU and
528 * signaling the request queue with the result of the copy.
531 zvol_write(void *arg)
533 struct request *req = (struct request *)arg;
534 struct request_queue *q = req->q;
535 zvol_state_t *zv = q->queuedata;
536 uint64_t offset = blk_rq_pos(req) << 9;
537 uint64_t size = blk_rq_bytes(req);
542 if (req->cmd_flags & VDEV_REQ_FLUSH)
543 zil_commit(zv->zv_zilog, ZVOL_OBJ);
546 * Some requests are just for flush and nothing else.
549 blk_end_request(req, 0, size);
553 rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
555 tx = dmu_tx_create(zv->zv_objset);
556 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
558 /* This will only fail for ENOSPC */
559 error = dmu_tx_assign(tx, TXG_WAIT);
562 zfs_range_unlock(rl);
563 blk_end_request(req, -error, size);
567 error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx);
569 zvol_log_write(zv, tx, offset, size,
570 req->cmd_flags & VDEV_REQ_FUA);
573 zfs_range_unlock(rl);
575 if ((req->cmd_flags & VDEV_REQ_FUA) ||
576 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
577 zil_commit(zv->zv_zilog, ZVOL_OBJ);
579 blk_end_request(req, -error, size);
582 #ifdef HAVE_BLK_QUEUE_DISCARD
584 zvol_discard(void *arg)
586 struct request *req = (struct request *)arg;
587 struct request_queue *q = req->q;
588 zvol_state_t *zv = q->queuedata;
589 uint64_t offset = blk_rq_pos(req) << 9;
590 uint64_t size = blk_rq_bytes(req);
594 if (offset + size > zv->zv_volsize) {
595 blk_end_request(req, -EIO, size);
600 blk_end_request(req, 0, size);
604 rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
606 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, size);
609 * TODO: maybe we should add the operation to the log.
612 zfs_range_unlock(rl);
614 blk_end_request(req, -error, size);
616 #endif /* HAVE_BLK_QUEUE_DISCARD */
619 * Common read path running under the zvol taskq context. This function
620 * is responsible for copying the requested data out of the DMU and in to
621 * a linux request structure. It then must signal the request queue with
622 * an error code describing the result of the copy.
627 struct request *req = (struct request *)arg;
628 struct request_queue *q = req->q;
629 zvol_state_t *zv = q->queuedata;
630 uint64_t offset = blk_rq_pos(req) << 9;
631 uint64_t size = blk_rq_bytes(req);
636 blk_end_request(req, 0, size);
640 rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
642 error = dmu_read_req(zv->zv_objset, ZVOL_OBJ, req);
644 zfs_range_unlock(rl);
646 /* convert checksum errors into IO errors */
650 blk_end_request(req, -error, size);
654 * Request will be added back to the request queue and retried if
655 * it cannot be immediately dispatched to the taskq for handling
658 zvol_dispatch(task_func_t func, struct request *req)
660 if (!taskq_dispatch(zvol_taskq, func, (void *)req, TQ_NOSLEEP))
661 blk_requeue_request(req->q, req);
665 * Common request path. Rather than registering a custom make_request()
666 * function we use the generic Linux version. This is done because it allows
667 * us to easily merge read requests which would otherwise we performed
668 * synchronously by the DMU. This is less critical in write case where the
669 * DMU will perform the correct merging within a transaction group. Using
670 * the generic make_request() also let's use leverage the fact that the
671 * elevator with ensure correct ordering in regards to barrior IOs. On
672 * the downside it means that in the write case we end up doing request
673 * merging twice once in the elevator and once in the DMU.
675 * The request handler is called under a spin lock so all the real work
676 * is handed off to be done in the context of the zvol taskq. This function
677 * simply performs basic request sanity checking and hands off the request.
680 zvol_request(struct request_queue *q)
682 zvol_state_t *zv = q->queuedata;
686 while ((req = blk_fetch_request(q)) != NULL) {
687 size = blk_rq_bytes(req);
689 if (size != 0 && blk_rq_pos(req) + blk_rq_sectors(req) >
690 get_capacity(zv->zv_disk)) {
692 "%s: bad access: block=%llu, count=%lu\n",
693 req->rq_disk->disk_name,
694 (long long unsigned)blk_rq_pos(req),
695 (long unsigned)blk_rq_sectors(req));
696 __blk_end_request(req, -EIO, size);
700 if (!blk_fs_request(req)) {
701 printk(KERN_INFO "%s: non-fs cmd\n",
702 req->rq_disk->disk_name);
703 __blk_end_request(req, -EIO, size);
707 switch (rq_data_dir(req)) {
709 zvol_dispatch(zvol_read, req);
712 if (unlikely(get_disk_ro(zv->zv_disk)) ||
713 unlikely(zv->zv_flags & ZVOL_RDONLY)) {
714 __blk_end_request(req, -EROFS, size);
718 #ifdef HAVE_BLK_QUEUE_DISCARD
719 if (req->cmd_flags & VDEV_REQ_DISCARD) {
720 zvol_dispatch(zvol_discard, req);
723 #endif /* HAVE_BLK_QUEUE_DISCARD */
725 zvol_dispatch(zvol_write, req);
728 printk(KERN_INFO "%s: unknown cmd: %d\n",
729 req->rq_disk->disk_name, (int)rq_data_dir(req));
730 __blk_end_request(req, -EIO, size);
737 zvol_get_done(zgd_t *zgd, int error)
740 dmu_buf_rele(zgd->zgd_db, zgd);
742 zfs_range_unlock(zgd->zgd_rl);
744 if (error == 0 && zgd->zgd_bp)
745 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
747 kmem_free(zgd, sizeof (zgd_t));
751 * Get data to generate a TX_WRITE intent log record.
754 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
756 zvol_state_t *zv = arg;
757 objset_t *os = zv->zv_objset;
758 uint64_t offset = lr->lr_offset;
759 uint64_t size = lr->lr_length;
767 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
768 zgd->zgd_zilog = zv->zv_zilog;
769 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
772 * Write records come in two flavors: immediate and indirect.
773 * For small writes it's cheaper to store the data with the
774 * log record (immediate); for large writes it's cheaper to
775 * sync the data and get a pointer to it (indirect) so that
776 * we don't have to write the data twice.
778 if (buf != NULL) { /* immediate write */
779 error = dmu_read(os, ZVOL_OBJ, offset, size, buf,
780 DMU_READ_NO_PREFETCH);
782 size = zv->zv_volblocksize;
783 offset = P2ALIGN_TYPED(offset, size, uint64_t);
784 error = dmu_buf_hold(os, ZVOL_OBJ, offset, zgd, &db,
785 DMU_READ_NO_PREFETCH);
788 zgd->zgd_bp = &lr->lr_blkptr;
791 ASSERT(db->db_offset == offset);
792 ASSERT(db->db_size == size);
794 error = dmu_sync(zio, lr->lr_common.lrc_txg,
802 zvol_get_done(zgd, error);
808 * The zvol_state_t's are inserted in increasing MINOR(dev_t) order.
811 zvol_insert(zvol_state_t *zv_insert)
813 zvol_state_t *zv = NULL;
815 ASSERT(MUTEX_HELD(&zvol_state_lock));
816 ASSERT3U(MINOR(zv_insert->zv_dev) & ZVOL_MINOR_MASK, ==, 0);
817 for (zv = list_head(&zvol_state_list); zv != NULL;
818 zv = list_next(&zvol_state_list, zv)) {
819 if (MINOR(zv->zv_dev) > MINOR(zv_insert->zv_dev))
823 list_insert_before(&zvol_state_list, zv, zv_insert);
827 * Simply remove the zvol from to list of zvols.
830 zvol_remove(zvol_state_t *zv_remove)
832 ASSERT(MUTEX_HELD(&zvol_state_lock));
833 list_remove(&zvol_state_list, zv_remove);
837 zvol_first_open(zvol_state_t *zv)
844 /* lie and say we're read-only */
845 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
849 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
851 dmu_objset_disown(os, zvol_tag);
856 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
858 dmu_objset_disown(os, zvol_tag);
862 set_capacity(zv->zv_disk, volsize >> 9);
863 zv->zv_volsize = volsize;
864 zv->zv_zilog = zil_open(os, zvol_get_data);
866 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0);
867 if (ro || dmu_objset_is_snapshot(os)) {
868 set_disk_ro(zv->zv_disk, 1);
869 zv->zv_flags |= ZVOL_RDONLY;
871 set_disk_ro(zv->zv_disk, 0);
872 zv->zv_flags &= ~ZVOL_RDONLY;
879 zvol_last_close(zvol_state_t *zv)
881 zil_close(zv->zv_zilog);
883 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
885 dmu_objset_disown(zv->zv_objset, zvol_tag);
886 zv->zv_objset = NULL;
890 zvol_open(struct block_device *bdev, fmode_t flag)
892 zvol_state_t *zv = bdev->bd_disk->private_data;
893 int error = 0, drop_mutex = 0;
896 * If the caller is already holding the mutex do not take it
897 * again, this will happen as part of zvol_create_minor().
898 * Once add_disk() is called the device is live and the kernel
899 * will attempt to open it to read the partition information.
901 if (!mutex_owned(&zvol_state_lock)) {
902 mutex_enter(&zvol_state_lock);
906 ASSERT3P(zv, !=, NULL);
908 if (zv->zv_open_count == 0) {
909 error = zvol_first_open(zv);
914 if ((flag & FMODE_WRITE) &&
915 (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY))) {
923 if (zv->zv_open_count == 0)
928 mutex_exit(&zvol_state_lock);
930 check_disk_change(bdev);
936 zvol_release(struct gendisk *disk, fmode_t mode)
938 zvol_state_t *zv = disk->private_data;
941 if (!mutex_owned(&zvol_state_lock)) {
942 mutex_enter(&zvol_state_lock);
946 ASSERT3P(zv, !=, NULL);
947 ASSERT3U(zv->zv_open_count, >, 0);
949 if (zv->zv_open_count == 0)
953 mutex_exit(&zvol_state_lock);
959 zvol_ioctl(struct block_device *bdev, fmode_t mode,
960 unsigned int cmd, unsigned long arg)
962 zvol_state_t *zv = bdev->bd_disk->private_data;
970 zil_commit(zv->zv_zilog, ZVOL_OBJ);
973 error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN);
987 zvol_compat_ioctl(struct block_device *bdev, fmode_t mode,
988 unsigned cmd, unsigned long arg)
990 return zvol_ioctl(bdev, mode, cmd, arg);
993 #define zvol_compat_ioctl NULL
996 static int zvol_media_changed(struct gendisk *disk)
998 zvol_state_t *zv = disk->private_data;
1000 return zv->zv_changed;
1003 static int zvol_revalidate_disk(struct gendisk *disk)
1005 zvol_state_t *zv = disk->private_data;
1008 set_capacity(zv->zv_disk, zv->zv_volsize >> 9);
1014 * Provide a simple virtual geometry for legacy compatibility. For devices
1015 * smaller than 1 MiB a small head and sector count is used to allow very
1016 * tiny devices. For devices over 1 Mib a standard head and sector count
1017 * is used to keep the cylinders count reasonable.
1020 zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1022 zvol_state_t *zv = bdev->bd_disk->private_data;
1023 sector_t sectors = get_capacity(zv->zv_disk);
1025 if (sectors > 2048) {
1034 geo->cylinders = sectors / (geo->heads * geo->sectors);
1039 static struct kobject *
1040 zvol_probe(dev_t dev, int *part, void *arg)
1043 struct kobject *kobj;
1045 mutex_enter(&zvol_state_lock);
1046 zv = zvol_find_by_dev(dev);
1047 kobj = zv ? get_disk(zv->zv_disk) : ERR_PTR(-ENOENT);
1048 mutex_exit(&zvol_state_lock);
1053 #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS
1054 static struct block_device_operations zvol_ops = {
1056 .release = zvol_release,
1057 .ioctl = zvol_ioctl,
1058 .compat_ioctl = zvol_compat_ioctl,
1059 .media_changed = zvol_media_changed,
1060 .revalidate_disk = zvol_revalidate_disk,
1061 .getgeo = zvol_getgeo,
1062 .owner = THIS_MODULE,
1065 #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1068 zvol_open_by_inode(struct inode *inode, struct file *file)
1070 return zvol_open(inode->i_bdev, file->f_mode);
1074 zvol_release_by_inode(struct inode *inode, struct file *file)
1076 return zvol_release(inode->i_bdev->bd_disk, file->f_mode);
1080 zvol_ioctl_by_inode(struct inode *inode, struct file *file,
1081 unsigned int cmd, unsigned long arg)
1083 if (file == NULL || inode == NULL)
1085 return zvol_ioctl(inode->i_bdev, file->f_mode, cmd, arg);
1088 # ifdef CONFIG_COMPAT
1090 zvol_compat_ioctl_by_inode(struct file *file,
1091 unsigned int cmd, unsigned long arg)
1095 return zvol_compat_ioctl(file->f_dentry->d_inode->i_bdev,
1096 file->f_mode, cmd, arg);
1099 # define zvol_compat_ioctl_by_inode NULL
1102 static struct block_device_operations zvol_ops = {
1103 .open = zvol_open_by_inode,
1104 .release = zvol_release_by_inode,
1105 .ioctl = zvol_ioctl_by_inode,
1106 .compat_ioctl = zvol_compat_ioctl_by_inode,
1107 .media_changed = zvol_media_changed,
1108 .revalidate_disk = zvol_revalidate_disk,
1109 .getgeo = zvol_getgeo,
1110 .owner = THIS_MODULE,
1112 #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1115 * Allocate memory for a new zvol_state_t and setup the required
1116 * request queue and generic disk structures for the block device.
1118 static zvol_state_t *
1119 zvol_alloc(dev_t dev, const char *name)
1123 zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
1127 zv->zv_queue = blk_init_queue(zvol_request, &zv->zv_lock);
1128 if (zv->zv_queue == NULL)
1131 #ifdef HAVE_BLK_QUEUE_FLUSH
1132 blk_queue_flush(zv->zv_queue, VDEV_REQ_FLUSH | VDEV_REQ_FUA);
1134 blk_queue_ordered(zv->zv_queue, QUEUE_ORDERED_DRAIN, NULL);
1135 #endif /* HAVE_BLK_QUEUE_FLUSH */
1137 zv->zv_disk = alloc_disk(ZVOL_MINORS);
1138 if (zv->zv_disk == NULL)
1141 zv->zv_queue->queuedata = zv;
1143 zv->zv_open_count = 0;
1144 strlcpy(zv->zv_name, name, MAXNAMELEN);
1146 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
1147 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
1148 sizeof (rl_t), offsetof(rl_t, r_node));
1149 zv->zv_znode.z_is_zvol = TRUE;
1151 spin_lock_init(&zv->zv_lock);
1152 list_link_init(&zv->zv_next);
1154 zv->zv_disk->major = zvol_major;
1155 zv->zv_disk->first_minor = (dev & MINORMASK);
1156 zv->zv_disk->fops = &zvol_ops;
1157 zv->zv_disk->private_data = zv;
1158 zv->zv_disk->queue = zv->zv_queue;
1159 snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s%d",
1160 ZVOL_DEV_NAME, (dev & MINORMASK));
1165 blk_cleanup_queue(zv->zv_queue);
1167 kmem_free(zv, sizeof (zvol_state_t));
1173 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1176 zvol_free(zvol_state_t *zv)
1178 avl_destroy(&zv->zv_znode.z_range_avl);
1179 mutex_destroy(&zv->zv_znode.z_range_lock);
1181 del_gendisk(zv->zv_disk);
1182 blk_cleanup_queue(zv->zv_queue);
1183 put_disk(zv->zv_disk);
1185 kmem_free(zv, sizeof (zvol_state_t));
1189 __zvol_create_minor(const char *name)
1193 dmu_object_info_t *doi;
1198 ASSERT(MUTEX_HELD(&zvol_state_lock));
1200 zv = zvol_find_by_name(name);
1206 doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);
1208 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os);
1212 error = dmu_object_info(os, ZVOL_OBJ, doi);
1214 goto out_dmu_objset_disown;
1216 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1218 goto out_dmu_objset_disown;
1220 error = zvol_find_minor(&minor);
1222 goto out_dmu_objset_disown;
1224 zv = zvol_alloc(MKDEV(zvol_major, minor), name);
1227 goto out_dmu_objset_disown;
1230 if (dmu_objset_is_snapshot(os))
1231 zv->zv_flags |= ZVOL_RDONLY;
1233 zv->zv_volblocksize = doi->doi_data_block_size;
1234 zv->zv_volsize = volsize;
1237 set_capacity(zv->zv_disk, zv->zv_volsize >> 9);
1239 blk_queue_max_hw_sectors(zv->zv_queue, UINT_MAX);
1240 blk_queue_max_segments(zv->zv_queue, UINT16_MAX);
1241 blk_queue_max_segment_size(zv->zv_queue, UINT_MAX);
1242 blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize);
1243 blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize);
1244 #ifdef HAVE_BLK_QUEUE_DISCARD
1245 blk_queue_max_discard_sectors(zv->zv_queue, UINT_MAX);
1246 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue);
1248 #ifdef HAVE_BLK_QUEUE_NONROT
1249 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue);
1252 if (zil_replay_disable)
1253 zil_destroy(dmu_objset_zil(os), B_FALSE);
1255 zil_replay(os, zv, zvol_replay_vector);
1257 out_dmu_objset_disown:
1258 dmu_objset_disown(os, zvol_tag);
1259 zv->zv_objset = NULL;
1261 kmem_free(doi, sizeof(dmu_object_info_t));
1266 add_disk(zv->zv_disk);
1273 * Create a block device minor node and setup the linkage between it
1274 * and the specified volume. Once this function returns the block
1275 * device is live and ready for use.
1278 zvol_create_minor(const char *name)
1282 mutex_enter(&zvol_state_lock);
1283 error = __zvol_create_minor(name);
1284 mutex_exit(&zvol_state_lock);
1290 __zvol_remove_minor(const char *name)
1294 ASSERT(MUTEX_HELD(&zvol_state_lock));
1296 zv = zvol_find_by_name(name);
1300 if (zv->zv_open_count > 0)
1310 * Remove a block device minor node for the specified volume.
1313 zvol_remove_minor(const char *name)
1317 mutex_enter(&zvol_state_lock);
1318 error = __zvol_remove_minor(name);
1319 mutex_exit(&zvol_state_lock);
1325 zvol_create_minors_cb(spa_t *spa, uint64_t dsobj,
1326 const char *dsname, void *arg)
1328 if (strchr(dsname, '/') == NULL)
1331 (void) __zvol_create_minor(dsname);
1336 * Create minors for specified pool, if pool is NULL create minors
1337 * for all available pools.
1340 zvol_create_minors(const char *pool)
1345 if (zvol_inhibit_dev)
1348 mutex_enter(&zvol_state_lock);
1350 error = dmu_objset_find_spa(NULL, pool, zvol_create_minors_cb,
1351 NULL, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1353 mutex_enter(&spa_namespace_lock);
1354 while ((spa = spa_next(spa)) != NULL) {
1355 error = dmu_objset_find_spa(NULL,
1356 spa_name(spa), zvol_create_minors_cb, NULL,
1357 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1361 mutex_exit(&spa_namespace_lock);
1363 mutex_exit(&zvol_state_lock);
1369 * Remove minors for specified pool, if pool is NULL remove all minors.
1372 zvol_remove_minors(const char *pool)
1374 zvol_state_t *zv, *zv_next;
1377 if (zvol_inhibit_dev)
1380 str = kmem_zalloc(MAXNAMELEN, KM_SLEEP);
1382 (void) strncpy(str, pool, strlen(pool));
1383 (void) strcat(str, "/");
1386 mutex_enter(&zvol_state_lock);
1387 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1388 zv_next = list_next(&zvol_state_list, zv);
1390 if (pool == NULL || !strncmp(str, zv->zv_name, strlen(str))) {
1395 mutex_exit(&zvol_state_lock);
1396 kmem_free(str, MAXNAMELEN);
1404 zvol_taskq = taskq_create(ZVOL_DRIVER, zvol_threads, maxclsyspri,
1405 zvol_threads, INT_MAX, TASKQ_PREPOPULATE);
1406 if (zvol_taskq == NULL) {
1407 printk(KERN_INFO "ZFS: taskq_create() failed\n");
1411 error = register_blkdev(zvol_major, ZVOL_DRIVER);
1413 printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error);
1414 taskq_destroy(zvol_taskq);
1418 blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS,
1419 THIS_MODULE, zvol_probe, NULL, NULL);
1421 mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1422 list_create(&zvol_state_list, sizeof (zvol_state_t),
1423 offsetof(zvol_state_t, zv_next));
1425 (void) zvol_create_minors(NULL);
1433 zvol_remove_minors(NULL);
1434 blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS);
1435 unregister_blkdev(zvol_major, ZVOL_DRIVER);
1436 taskq_destroy(zvol_taskq);
1437 mutex_destroy(&zvol_state_lock);
1438 list_destroy(&zvol_state_list);
1441 module_param(zvol_inhibit_dev, uint, 0644);
1442 MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes");
1444 module_param(zvol_major, uint, 0444);
1445 MODULE_PARM_DESC(zvol_major, "Major number for zvol device");
1447 module_param(zvol_threads, uint, 0444);
1448 MODULE_PARM_DESC(zvol_threads, "Number of threads for zvol device");