2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <linux/list_sort.h>
31 #include <asm/div64.h>
33 #include "extent_map.h"
35 #include "transaction.h"
36 #include "print-tree.h"
39 #include "async-thread.h"
40 #include "check-integrity.h"
41 #include "rcu-string.h"
43 #include "dev-replace.h"
46 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
47 [BTRFS_RAID_RAID10] = {
50 .devs_max = 0, /* 0 == as many as possible */
52 .tolerated_failures = 1,
56 [BTRFS_RAID_RAID1] = {
61 .tolerated_failures = 1,
70 .tolerated_failures = 0,
74 [BTRFS_RAID_RAID0] = {
79 .tolerated_failures = 0,
83 [BTRFS_RAID_SINGLE] = {
88 .tolerated_failures = 0,
92 [BTRFS_RAID_RAID5] = {
97 .tolerated_failures = 1,
101 [BTRFS_RAID_RAID6] = {
106 .tolerated_failures = 2,
112 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
113 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
114 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
115 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
116 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
117 [BTRFS_RAID_SINGLE] = 0,
118 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
119 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
123 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
124 * condition is not met. Zero means there's no corresponding
125 * BTRFS_ERROR_DEV_*_NOT_MET value.
127 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
128 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
129 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
130 [BTRFS_RAID_DUP] = 0,
131 [BTRFS_RAID_RAID0] = 0,
132 [BTRFS_RAID_SINGLE] = 0,
133 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
134 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
137 static int init_first_rw_device(struct btrfs_trans_handle *trans,
138 struct btrfs_fs_info *fs_info);
139 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
144 enum btrfs_map_op op,
145 u64 logical, u64 *length,
146 struct btrfs_bio **bbio_ret,
147 int mirror_num, int need_raid_map);
153 * There are several mutexes that protect manipulation of devices and low-level
154 * structures like chunks but not block groups, extents or files
156 * uuid_mutex (global lock)
157 * ------------------------
158 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
159 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
160 * device) or requested by the device= mount option
162 * the mutex can be very coarse and can cover long-running operations
164 * protects: updates to fs_devices counters like missing devices, rw devices,
165 * seeding, structure cloning, openning/closing devices at mount/umount time
167 * global::fs_devs - add, remove, updates to the global list
169 * does not protect: manipulation of the fs_devices::devices list!
171 * btrfs_device::name - renames (write side), read is RCU
173 * fs_devices::device_list_mutex (per-fs, with RCU)
174 * ------------------------------------------------
175 * protects updates to fs_devices::devices, ie. adding and deleting
177 * simple list traversal with read-only actions can be done with RCU protection
179 * may be used to exclude some operations from running concurrently without any
180 * modifications to the list (see write_all_supers)
184 * coarse lock owned by a mounted filesystem; used to exclude some operations
185 * that cannot run in parallel and affect the higher-level properties of the
186 * filesystem like: device add/deleting/resize/replace, or balance
190 * protects balance structures (status, state) and context accessed from
191 * several places (internally, ioctl)
195 * protects chunks, adding or removing during allocation, trim or when a new
196 * device is added/removed
200 * a big lock that is held by the cleaner thread and prevents running subvolume
201 * cleaning together with relocation or delayed iputs
214 DEFINE_MUTEX(uuid_mutex);
215 static LIST_HEAD(fs_uuids);
216 struct list_head *btrfs_get_fs_uuids(void)
222 * alloc_fs_devices - allocate struct btrfs_fs_devices
223 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
225 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
226 * The returned struct is not linked onto any lists and can be destroyed with
227 * kfree() right away.
229 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
231 struct btrfs_fs_devices *fs_devs;
233 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
235 return ERR_PTR(-ENOMEM);
237 mutex_init(&fs_devs->device_list_mutex);
239 INIT_LIST_HEAD(&fs_devs->devices);
240 INIT_LIST_HEAD(&fs_devs->resized_devices);
241 INIT_LIST_HEAD(&fs_devs->alloc_list);
242 INIT_LIST_HEAD(&fs_devs->list);
244 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
249 static void free_device(struct btrfs_device *device)
251 rcu_string_free(device->name);
252 bio_put(device->flush_bio);
256 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
258 struct btrfs_device *device;
259 WARN_ON(fs_devices->opened);
260 while (!list_empty(&fs_devices->devices)) {
261 device = list_entry(fs_devices->devices.next,
262 struct btrfs_device, dev_list);
263 list_del(&device->dev_list);
269 static void btrfs_kobject_uevent(struct block_device *bdev,
270 enum kobject_action action)
274 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
276 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
278 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
279 &disk_to_dev(bdev->bd_disk)->kobj);
282 void __exit btrfs_cleanup_fs_uuids(void)
284 struct btrfs_fs_devices *fs_devices;
286 while (!list_empty(&fs_uuids)) {
287 fs_devices = list_entry(fs_uuids.next,
288 struct btrfs_fs_devices, list);
289 list_del(&fs_devices->list);
290 free_fs_devices(fs_devices);
295 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
296 * Returned struct is not linked onto any lists and must be destroyed using
299 static struct btrfs_device *__alloc_device(void)
301 struct btrfs_device *dev;
303 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
305 return ERR_PTR(-ENOMEM);
308 * Preallocate a bio that's always going to be used for flushing device
309 * barriers and matches the device lifespan
311 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
312 if (!dev->flush_bio) {
314 return ERR_PTR(-ENOMEM);
317 INIT_LIST_HEAD(&dev->dev_list);
318 INIT_LIST_HEAD(&dev->dev_alloc_list);
319 INIT_LIST_HEAD(&dev->resized_list);
321 spin_lock_init(&dev->io_lock);
323 atomic_set(&dev->reada_in_flight, 0);
324 atomic_set(&dev->dev_stats_ccnt, 0);
325 btrfs_device_data_ordered_init(dev);
326 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
327 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
333 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
336 * If devid and uuid are both specified, the match must be exact, otherwise
337 * only devid is used.
339 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
340 u64 devid, const u8 *uuid)
342 struct list_head *head = &fs_devices->devices;
343 struct btrfs_device *dev;
345 list_for_each_entry(dev, head, dev_list) {
346 if (dev->devid == devid &&
347 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
354 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
356 struct btrfs_fs_devices *fs_devices;
358 list_for_each_entry(fs_devices, &fs_uuids, list) {
359 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
366 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
367 int flush, struct block_device **bdev,
368 struct buffer_head **bh)
372 *bdev = blkdev_get_by_path(device_path, flags, holder);
375 ret = PTR_ERR(*bdev);
380 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
381 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
383 blkdev_put(*bdev, flags);
386 invalidate_bdev(*bdev);
387 *bh = btrfs_read_dev_super(*bdev);
390 blkdev_put(*bdev, flags);
402 static void requeue_list(struct btrfs_pending_bios *pending_bios,
403 struct bio *head, struct bio *tail)
406 struct bio *old_head;
408 old_head = pending_bios->head;
409 pending_bios->head = head;
410 if (pending_bios->tail)
411 tail->bi_next = old_head;
413 pending_bios->tail = tail;
417 * we try to collect pending bios for a device so we don't get a large
418 * number of procs sending bios down to the same device. This greatly
419 * improves the schedulers ability to collect and merge the bios.
421 * But, it also turns into a long list of bios to process and that is sure
422 * to eventually make the worker thread block. The solution here is to
423 * make some progress and then put this work struct back at the end of
424 * the list if the block device is congested. This way, multiple devices
425 * can make progress from a single worker thread.
427 static noinline void run_scheduled_bios(struct btrfs_device *device)
429 struct btrfs_fs_info *fs_info = device->fs_info;
431 struct backing_dev_info *bdi;
432 struct btrfs_pending_bios *pending_bios;
436 unsigned long num_run;
437 unsigned long batch_run = 0;
438 unsigned long last_waited = 0;
440 int sync_pending = 0;
441 struct blk_plug plug;
444 * this function runs all the bios we've collected for
445 * a particular device. We don't want to wander off to
446 * another device without first sending all of these down.
447 * So, setup a plug here and finish it off before we return
449 blk_start_plug(&plug);
451 bdi = device->bdev->bd_bdi;
454 spin_lock(&device->io_lock);
459 /* take all the bios off the list at once and process them
460 * later on (without the lock held). But, remember the
461 * tail and other pointers so the bios can be properly reinserted
462 * into the list if we hit congestion
464 if (!force_reg && device->pending_sync_bios.head) {
465 pending_bios = &device->pending_sync_bios;
468 pending_bios = &device->pending_bios;
472 pending = pending_bios->head;
473 tail = pending_bios->tail;
474 WARN_ON(pending && !tail);
477 * if pending was null this time around, no bios need processing
478 * at all and we can stop. Otherwise it'll loop back up again
479 * and do an additional check so no bios are missed.
481 * device->running_pending is used to synchronize with the
484 if (device->pending_sync_bios.head == NULL &&
485 device->pending_bios.head == NULL) {
487 device->running_pending = 0;
490 device->running_pending = 1;
493 pending_bios->head = NULL;
494 pending_bios->tail = NULL;
496 spin_unlock(&device->io_lock);
501 /* we want to work on both lists, but do more bios on the
502 * sync list than the regular list
505 pending_bios != &device->pending_sync_bios &&
506 device->pending_sync_bios.head) ||
507 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
508 device->pending_bios.head)) {
509 spin_lock(&device->io_lock);
510 requeue_list(pending_bios, pending, tail);
515 pending = pending->bi_next;
518 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
521 * if we're doing the sync list, record that our
522 * plug has some sync requests on it
524 * If we're doing the regular list and there are
525 * sync requests sitting around, unplug before
528 if (pending_bios == &device->pending_sync_bios) {
530 } else if (sync_pending) {
531 blk_finish_plug(&plug);
532 blk_start_plug(&plug);
536 btrfsic_submit_bio(cur);
543 * we made progress, there is more work to do and the bdi
544 * is now congested. Back off and let other work structs
547 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
548 fs_info->fs_devices->open_devices > 1) {
549 struct io_context *ioc;
551 ioc = current->io_context;
554 * the main goal here is that we don't want to
555 * block if we're going to be able to submit
556 * more requests without blocking.
558 * This code does two great things, it pokes into
559 * the elevator code from a filesystem _and_
560 * it makes assumptions about how batching works.
562 if (ioc && ioc->nr_batch_requests > 0 &&
563 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
565 ioc->last_waited == last_waited)) {
567 * we want to go through our batch of
568 * requests and stop. So, we copy out
569 * the ioc->last_waited time and test
570 * against it before looping
572 last_waited = ioc->last_waited;
576 spin_lock(&device->io_lock);
577 requeue_list(pending_bios, pending, tail);
578 device->running_pending = 1;
580 spin_unlock(&device->io_lock);
581 btrfs_queue_work(fs_info->submit_workers,
591 spin_lock(&device->io_lock);
592 if (device->pending_bios.head || device->pending_sync_bios.head)
594 spin_unlock(&device->io_lock);
597 blk_finish_plug(&plug);
600 static void pending_bios_fn(struct btrfs_work *work)
602 struct btrfs_device *device;
604 device = container_of(work, struct btrfs_device, work);
605 run_scheduled_bios(device);
609 * Search and remove all stale (devices which are not mounted) devices.
610 * When both inputs are NULL, it will search and release all stale devices.
611 * path: Optional. When provided will it release all unmounted devices
612 * matching this path only.
613 * skip_dev: Optional. Will skip this device when searching for the stale
616 static void btrfs_free_stale_devices(const char *path,
617 struct btrfs_device *skip_dev)
619 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
620 struct btrfs_device *dev, *tmp_dev;
622 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) {
627 list_for_each_entry_safe(dev, tmp_dev,
628 &fs_devs->devices, dev_list) {
631 if (skip_dev && skip_dev == dev)
633 if (path && !dev->name)
638 not_found = strcmp(rcu_str_deref(dev->name),
644 /* delete the stale device */
645 if (fs_devs->num_devices == 1) {
646 btrfs_sysfs_remove_fsid(fs_devs);
647 list_del(&fs_devs->list);
648 free_fs_devices(fs_devs);
651 fs_devs->num_devices--;
652 list_del(&dev->dev_list);
659 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
660 struct btrfs_device *device, fmode_t flags,
663 struct request_queue *q;
664 struct block_device *bdev;
665 struct buffer_head *bh;
666 struct btrfs_super_block *disk_super;
675 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
680 disk_super = (struct btrfs_super_block *)bh->b_data;
681 devid = btrfs_stack_device_id(&disk_super->dev_item);
682 if (devid != device->devid)
685 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
688 device->generation = btrfs_super_generation(disk_super);
690 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
691 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
692 fs_devices->seeding = 1;
694 if (bdev_read_only(bdev))
695 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
697 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
700 q = bdev_get_queue(bdev);
701 if (!blk_queue_nonrot(q))
702 fs_devices->rotating = 1;
705 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
706 device->mode = flags;
708 fs_devices->open_devices++;
709 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
710 device->devid != BTRFS_DEV_REPLACE_DEVID) {
711 fs_devices->rw_devices++;
712 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
720 blkdev_put(bdev, flags);
726 * Add new device to list of registered devices
729 * device pointer which was just added or updated when successful
730 * error pointer when failed
732 static noinline struct btrfs_device *device_list_add(const char *path,
733 struct btrfs_super_block *disk_super)
735 struct btrfs_device *device;
736 struct btrfs_fs_devices *fs_devices;
737 struct rcu_string *name;
738 u64 found_transid = btrfs_super_generation(disk_super);
739 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
741 fs_devices = find_fsid(disk_super->fsid);
743 fs_devices = alloc_fs_devices(disk_super->fsid);
744 if (IS_ERR(fs_devices))
745 return ERR_CAST(fs_devices);
747 list_add(&fs_devices->list, &fs_uuids);
751 device = find_device(fs_devices, devid,
752 disk_super->dev_item.uuid);
756 if (fs_devices->opened)
757 return ERR_PTR(-EBUSY);
759 device = btrfs_alloc_device(NULL, &devid,
760 disk_super->dev_item.uuid);
761 if (IS_ERR(device)) {
762 /* we can safely leave the fs_devices entry around */
766 name = rcu_string_strdup(path, GFP_NOFS);
769 return ERR_PTR(-ENOMEM);
771 rcu_assign_pointer(device->name, name);
773 mutex_lock(&fs_devices->device_list_mutex);
774 list_add_rcu(&device->dev_list, &fs_devices->devices);
775 fs_devices->num_devices++;
776 mutex_unlock(&fs_devices->device_list_mutex);
778 device->fs_devices = fs_devices;
779 btrfs_free_stale_devices(path, device);
781 if (disk_super->label[0])
782 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
783 disk_super->label, devid, found_transid, path);
785 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
786 disk_super->fsid, devid, found_transid, path);
788 } else if (!device->name || strcmp(device->name->str, path)) {
790 * When FS is already mounted.
791 * 1. If you are here and if the device->name is NULL that
792 * means this device was missing at time of FS mount.
793 * 2. If you are here and if the device->name is different
794 * from 'path' that means either
795 * a. The same device disappeared and reappeared with
797 * b. The missing-disk-which-was-replaced, has
800 * We must allow 1 and 2a above. But 2b would be a spurious
803 * Further in case of 1 and 2a above, the disk at 'path'
804 * would have missed some transaction when it was away and
805 * in case of 2a the stale bdev has to be updated as well.
806 * 2b must not be allowed at all time.
810 * For now, we do allow update to btrfs_fs_device through the
811 * btrfs dev scan cli after FS has been mounted. We're still
812 * tracking a problem where systems fail mount by subvolume id
813 * when we reject replacement on a mounted FS.
815 if (!fs_devices->opened && found_transid < device->generation) {
817 * That is if the FS is _not_ mounted and if you
818 * are here, that means there is more than one
819 * disk with same uuid and devid.We keep the one
820 * with larger generation number or the last-in if
821 * generation are equal.
823 return ERR_PTR(-EEXIST);
826 name = rcu_string_strdup(path, GFP_NOFS);
828 return ERR_PTR(-ENOMEM);
829 rcu_string_free(device->name);
830 rcu_assign_pointer(device->name, name);
831 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
832 fs_devices->missing_devices--;
833 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
838 * Unmount does not free the btrfs_device struct but would zero
839 * generation along with most of the other members. So just update
840 * it back. We need it to pick the disk with largest generation
843 if (!fs_devices->opened)
844 device->generation = found_transid;
846 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
851 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
853 struct btrfs_fs_devices *fs_devices;
854 struct btrfs_device *device;
855 struct btrfs_device *orig_dev;
857 fs_devices = alloc_fs_devices(orig->fsid);
858 if (IS_ERR(fs_devices))
861 mutex_lock(&orig->device_list_mutex);
862 fs_devices->total_devices = orig->total_devices;
864 /* We have held the volume lock, it is safe to get the devices. */
865 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
866 struct rcu_string *name;
868 device = btrfs_alloc_device(NULL, &orig_dev->devid,
874 * This is ok to do without rcu read locked because we hold the
875 * uuid mutex so nothing we touch in here is going to disappear.
877 if (orig_dev->name) {
878 name = rcu_string_strdup(orig_dev->name->str,
884 rcu_assign_pointer(device->name, name);
887 list_add(&device->dev_list, &fs_devices->devices);
888 device->fs_devices = fs_devices;
889 fs_devices->num_devices++;
891 mutex_unlock(&orig->device_list_mutex);
894 mutex_unlock(&orig->device_list_mutex);
895 free_fs_devices(fs_devices);
896 return ERR_PTR(-ENOMEM);
899 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
901 struct btrfs_device *device, *next;
902 struct btrfs_device *latest_dev = NULL;
904 mutex_lock(&uuid_mutex);
906 /* This is the initialized path, it is safe to release the devices. */
907 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
908 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
909 &device->dev_state)) {
910 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
911 &device->dev_state) &&
913 device->generation > latest_dev->generation)) {
919 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
921 * In the first step, keep the device which has
922 * the correct fsid and the devid that is used
923 * for the dev_replace procedure.
924 * In the second step, the dev_replace state is
925 * read from the device tree and it is known
926 * whether the procedure is really active or
927 * not, which means whether this device is
928 * used or whether it should be removed.
930 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
931 &device->dev_state)) {
936 blkdev_put(device->bdev, device->mode);
938 fs_devices->open_devices--;
940 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
941 list_del_init(&device->dev_alloc_list);
942 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
943 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
945 fs_devices->rw_devices--;
947 list_del_init(&device->dev_list);
948 fs_devices->num_devices--;
952 if (fs_devices->seed) {
953 fs_devices = fs_devices->seed;
957 fs_devices->latest_bdev = latest_dev->bdev;
959 mutex_unlock(&uuid_mutex);
962 static void free_device_rcu(struct rcu_head *head)
964 struct btrfs_device *device;
966 device = container_of(head, struct btrfs_device, rcu);
970 static void btrfs_close_bdev(struct btrfs_device *device)
975 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
976 sync_blockdev(device->bdev);
977 invalidate_bdev(device->bdev);
980 blkdev_put(device->bdev, device->mode);
983 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
985 struct btrfs_fs_devices *fs_devices = device->fs_devices;
986 struct btrfs_device *new_device;
987 struct rcu_string *name;
990 fs_devices->open_devices--;
992 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
993 device->devid != BTRFS_DEV_REPLACE_DEVID) {
994 list_del_init(&device->dev_alloc_list);
995 fs_devices->rw_devices--;
998 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
999 fs_devices->missing_devices--;
1001 new_device = btrfs_alloc_device(NULL, &device->devid,
1003 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1005 /* Safe because we are under uuid_mutex */
1007 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1008 BUG_ON(!name); /* -ENOMEM */
1009 rcu_assign_pointer(new_device->name, name);
1012 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1013 new_device->fs_devices = device->fs_devices;
1016 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1018 struct btrfs_device *device, *tmp;
1019 struct list_head pending_put;
1021 INIT_LIST_HEAD(&pending_put);
1023 if (--fs_devices->opened > 0)
1026 mutex_lock(&fs_devices->device_list_mutex);
1027 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1028 btrfs_prepare_close_one_device(device);
1029 list_add(&device->dev_list, &pending_put);
1031 mutex_unlock(&fs_devices->device_list_mutex);
1034 * btrfs_show_devname() is using the device_list_mutex,
1035 * sometimes call to blkdev_put() leads vfs calling
1036 * into this func. So do put outside of device_list_mutex,
1039 while (!list_empty(&pending_put)) {
1040 device = list_first_entry(&pending_put,
1041 struct btrfs_device, dev_list);
1042 list_del(&device->dev_list);
1043 btrfs_close_bdev(device);
1044 call_rcu(&device->rcu, free_device_rcu);
1047 WARN_ON(fs_devices->open_devices);
1048 WARN_ON(fs_devices->rw_devices);
1049 fs_devices->opened = 0;
1050 fs_devices->seeding = 0;
1055 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1057 struct btrfs_fs_devices *seed_devices = NULL;
1060 mutex_lock(&uuid_mutex);
1061 ret = __btrfs_close_devices(fs_devices);
1062 if (!fs_devices->opened) {
1063 seed_devices = fs_devices->seed;
1064 fs_devices->seed = NULL;
1066 mutex_unlock(&uuid_mutex);
1068 while (seed_devices) {
1069 fs_devices = seed_devices;
1070 seed_devices = fs_devices->seed;
1071 __btrfs_close_devices(fs_devices);
1072 free_fs_devices(fs_devices);
1077 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1078 fmode_t flags, void *holder)
1080 struct list_head *head = &fs_devices->devices;
1081 struct btrfs_device *device;
1082 struct btrfs_device *latest_dev = NULL;
1085 flags |= FMODE_EXCL;
1087 list_for_each_entry(device, head, dev_list) {
1088 /* Just open everything we can; ignore failures here */
1089 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1093 device->generation > latest_dev->generation)
1094 latest_dev = device;
1096 if (fs_devices->open_devices == 0) {
1100 fs_devices->opened = 1;
1101 fs_devices->latest_bdev = latest_dev->bdev;
1102 fs_devices->total_rw_bytes = 0;
1107 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1109 struct btrfs_device *dev1, *dev2;
1111 dev1 = list_entry(a, struct btrfs_device, dev_list);
1112 dev2 = list_entry(b, struct btrfs_device, dev_list);
1114 if (dev1->devid < dev2->devid)
1116 else if (dev1->devid > dev2->devid)
1121 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1122 fmode_t flags, void *holder)
1126 mutex_lock(&uuid_mutex);
1127 if (fs_devices->opened) {
1128 fs_devices->opened++;
1131 list_sort(NULL, &fs_devices->devices, devid_cmp);
1132 ret = __btrfs_open_devices(fs_devices, flags, holder);
1134 mutex_unlock(&uuid_mutex);
1138 static void btrfs_release_disk_super(struct page *page)
1144 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1146 struct btrfs_super_block **disk_super)
1151 /* make sure our super fits in the device */
1152 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1155 /* make sure our super fits in the page */
1156 if (sizeof(**disk_super) > PAGE_SIZE)
1159 /* make sure our super doesn't straddle pages on disk */
1160 index = bytenr >> PAGE_SHIFT;
1161 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1164 /* pull in the page with our super */
1165 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1168 if (IS_ERR_OR_NULL(*page))
1173 /* align our pointer to the offset of the super block */
1174 *disk_super = p + (bytenr & ~PAGE_MASK);
1176 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1177 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1178 btrfs_release_disk_super(*page);
1182 if ((*disk_super)->label[0] &&
1183 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1184 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1190 * Look for a btrfs signature on a device. This may be called out of the mount path
1191 * and we are not allowed to call set_blocksize during the scan. The superblock
1192 * is read via pagecache
1194 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1195 struct btrfs_fs_devices **fs_devices_ret)
1197 struct btrfs_super_block *disk_super;
1198 struct btrfs_device *device;
1199 struct block_device *bdev;
1205 * we would like to check all the supers, but that would make
1206 * a btrfs mount succeed after a mkfs from a different FS.
1207 * So, we need to add a special mount option to scan for
1208 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1210 bytenr = btrfs_sb_offset(0);
1211 flags |= FMODE_EXCL;
1212 mutex_lock(&uuid_mutex);
1214 bdev = blkdev_get_by_path(path, flags, holder);
1216 ret = PTR_ERR(bdev);
1220 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1222 goto error_bdev_put;
1225 device = device_list_add(path, disk_super);
1227 ret = PTR_ERR(device);
1229 *fs_devices_ret = device->fs_devices;
1231 btrfs_release_disk_super(page);
1234 blkdev_put(bdev, flags);
1236 mutex_unlock(&uuid_mutex);
1240 /* helper to account the used device space in the range */
1241 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1242 u64 end, u64 *length)
1244 struct btrfs_key key;
1245 struct btrfs_root *root = device->fs_info->dev_root;
1246 struct btrfs_dev_extent *dev_extent;
1247 struct btrfs_path *path;
1251 struct extent_buffer *l;
1255 if (start >= device->total_bytes ||
1256 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1259 path = btrfs_alloc_path();
1262 path->reada = READA_FORWARD;
1264 key.objectid = device->devid;
1266 key.type = BTRFS_DEV_EXTENT_KEY;
1268 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1272 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1279 slot = path->slots[0];
1280 if (slot >= btrfs_header_nritems(l)) {
1281 ret = btrfs_next_leaf(root, path);
1289 btrfs_item_key_to_cpu(l, &key, slot);
1291 if (key.objectid < device->devid)
1294 if (key.objectid > device->devid)
1297 if (key.type != BTRFS_DEV_EXTENT_KEY)
1300 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1301 extent_end = key.offset + btrfs_dev_extent_length(l,
1303 if (key.offset <= start && extent_end > end) {
1304 *length = end - start + 1;
1306 } else if (key.offset <= start && extent_end > start)
1307 *length += extent_end - start;
1308 else if (key.offset > start && extent_end <= end)
1309 *length += extent_end - key.offset;
1310 else if (key.offset > start && key.offset <= end) {
1311 *length += end - key.offset + 1;
1313 } else if (key.offset > end)
1321 btrfs_free_path(path);
1325 static int contains_pending_extent(struct btrfs_transaction *transaction,
1326 struct btrfs_device *device,
1327 u64 *start, u64 len)
1329 struct btrfs_fs_info *fs_info = device->fs_info;
1330 struct extent_map *em;
1331 struct list_head *search_list = &fs_info->pinned_chunks;
1333 u64 physical_start = *start;
1336 search_list = &transaction->pending_chunks;
1338 list_for_each_entry(em, search_list, list) {
1339 struct map_lookup *map;
1342 map = em->map_lookup;
1343 for (i = 0; i < map->num_stripes; i++) {
1346 if (map->stripes[i].dev != device)
1348 if (map->stripes[i].physical >= physical_start + len ||
1349 map->stripes[i].physical + em->orig_block_len <=
1353 * Make sure that while processing the pinned list we do
1354 * not override our *start with a lower value, because
1355 * we can have pinned chunks that fall within this
1356 * device hole and that have lower physical addresses
1357 * than the pending chunks we processed before. If we
1358 * do not take this special care we can end up getting
1359 * 2 pending chunks that start at the same physical
1360 * device offsets because the end offset of a pinned
1361 * chunk can be equal to the start offset of some
1364 end = map->stripes[i].physical + em->orig_block_len;
1371 if (search_list != &fs_info->pinned_chunks) {
1372 search_list = &fs_info->pinned_chunks;
1381 * find_free_dev_extent_start - find free space in the specified device
1382 * @device: the device which we search the free space in
1383 * @num_bytes: the size of the free space that we need
1384 * @search_start: the position from which to begin the search
1385 * @start: store the start of the free space.
1386 * @len: the size of the free space. that we find, or the size
1387 * of the max free space if we don't find suitable free space
1389 * this uses a pretty simple search, the expectation is that it is
1390 * called very infrequently and that a given device has a small number
1393 * @start is used to store the start of the free space if we find. But if we
1394 * don't find suitable free space, it will be used to store the start position
1395 * of the max free space.
1397 * @len is used to store the size of the free space that we find.
1398 * But if we don't find suitable free space, it is used to store the size of
1399 * the max free space.
1401 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1402 struct btrfs_device *device, u64 num_bytes,
1403 u64 search_start, u64 *start, u64 *len)
1405 struct btrfs_fs_info *fs_info = device->fs_info;
1406 struct btrfs_root *root = fs_info->dev_root;
1407 struct btrfs_key key;
1408 struct btrfs_dev_extent *dev_extent;
1409 struct btrfs_path *path;
1414 u64 search_end = device->total_bytes;
1417 struct extent_buffer *l;
1420 * We don't want to overwrite the superblock on the drive nor any area
1421 * used by the boot loader (grub for example), so we make sure to start
1422 * at an offset of at least 1MB.
1424 search_start = max_t(u64, search_start, SZ_1M);
1426 path = btrfs_alloc_path();
1430 max_hole_start = search_start;
1434 if (search_start >= search_end ||
1435 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1440 path->reada = READA_FORWARD;
1441 path->search_commit_root = 1;
1442 path->skip_locking = 1;
1444 key.objectid = device->devid;
1445 key.offset = search_start;
1446 key.type = BTRFS_DEV_EXTENT_KEY;
1448 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1452 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1459 slot = path->slots[0];
1460 if (slot >= btrfs_header_nritems(l)) {
1461 ret = btrfs_next_leaf(root, path);
1469 btrfs_item_key_to_cpu(l, &key, slot);
1471 if (key.objectid < device->devid)
1474 if (key.objectid > device->devid)
1477 if (key.type != BTRFS_DEV_EXTENT_KEY)
1480 if (key.offset > search_start) {
1481 hole_size = key.offset - search_start;
1484 * Have to check before we set max_hole_start, otherwise
1485 * we could end up sending back this offset anyway.
1487 if (contains_pending_extent(transaction, device,
1490 if (key.offset >= search_start) {
1491 hole_size = key.offset - search_start;
1498 if (hole_size > max_hole_size) {
1499 max_hole_start = search_start;
1500 max_hole_size = hole_size;
1504 * If this free space is greater than which we need,
1505 * it must be the max free space that we have found
1506 * until now, so max_hole_start must point to the start
1507 * of this free space and the length of this free space
1508 * is stored in max_hole_size. Thus, we return
1509 * max_hole_start and max_hole_size and go back to the
1512 if (hole_size >= num_bytes) {
1518 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1519 extent_end = key.offset + btrfs_dev_extent_length(l,
1521 if (extent_end > search_start)
1522 search_start = extent_end;
1529 * At this point, search_start should be the end of
1530 * allocated dev extents, and when shrinking the device,
1531 * search_end may be smaller than search_start.
1533 if (search_end > search_start) {
1534 hole_size = search_end - search_start;
1536 if (contains_pending_extent(transaction, device, &search_start,
1538 btrfs_release_path(path);
1542 if (hole_size > max_hole_size) {
1543 max_hole_start = search_start;
1544 max_hole_size = hole_size;
1549 if (max_hole_size < num_bytes)
1555 btrfs_free_path(path);
1556 *start = max_hole_start;
1558 *len = max_hole_size;
1562 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1563 struct btrfs_device *device, u64 num_bytes,
1564 u64 *start, u64 *len)
1566 /* FIXME use last free of some kind */
1567 return find_free_dev_extent_start(trans->transaction, device,
1568 num_bytes, 0, start, len);
1571 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1572 struct btrfs_device *device,
1573 u64 start, u64 *dev_extent_len)
1575 struct btrfs_fs_info *fs_info = device->fs_info;
1576 struct btrfs_root *root = fs_info->dev_root;
1578 struct btrfs_path *path;
1579 struct btrfs_key key;
1580 struct btrfs_key found_key;
1581 struct extent_buffer *leaf = NULL;
1582 struct btrfs_dev_extent *extent = NULL;
1584 path = btrfs_alloc_path();
1588 key.objectid = device->devid;
1590 key.type = BTRFS_DEV_EXTENT_KEY;
1592 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1594 ret = btrfs_previous_item(root, path, key.objectid,
1595 BTRFS_DEV_EXTENT_KEY);
1598 leaf = path->nodes[0];
1599 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1600 extent = btrfs_item_ptr(leaf, path->slots[0],
1601 struct btrfs_dev_extent);
1602 BUG_ON(found_key.offset > start || found_key.offset +
1603 btrfs_dev_extent_length(leaf, extent) < start);
1605 btrfs_release_path(path);
1607 } else if (ret == 0) {
1608 leaf = path->nodes[0];
1609 extent = btrfs_item_ptr(leaf, path->slots[0],
1610 struct btrfs_dev_extent);
1612 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1616 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1618 ret = btrfs_del_item(trans, root, path);
1620 btrfs_handle_fs_error(fs_info, ret,
1621 "Failed to remove dev extent item");
1623 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1626 btrfs_free_path(path);
1630 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1631 struct btrfs_device *device,
1632 u64 chunk_offset, u64 start, u64 num_bytes)
1635 struct btrfs_path *path;
1636 struct btrfs_fs_info *fs_info = device->fs_info;
1637 struct btrfs_root *root = fs_info->dev_root;
1638 struct btrfs_dev_extent *extent;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key key;
1642 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1643 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1644 path = btrfs_alloc_path();
1648 key.objectid = device->devid;
1650 key.type = BTRFS_DEV_EXTENT_KEY;
1651 ret = btrfs_insert_empty_item(trans, root, path, &key,
1656 leaf = path->nodes[0];
1657 extent = btrfs_item_ptr(leaf, path->slots[0],
1658 struct btrfs_dev_extent);
1659 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1660 BTRFS_CHUNK_TREE_OBJECTID);
1661 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1662 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1663 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1665 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1666 btrfs_mark_buffer_dirty(leaf);
1668 btrfs_free_path(path);
1672 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1674 struct extent_map_tree *em_tree;
1675 struct extent_map *em;
1679 em_tree = &fs_info->mapping_tree.map_tree;
1680 read_lock(&em_tree->lock);
1681 n = rb_last(&em_tree->map);
1683 em = rb_entry(n, struct extent_map, rb_node);
1684 ret = em->start + em->len;
1686 read_unlock(&em_tree->lock);
1691 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1695 struct btrfs_key key;
1696 struct btrfs_key found_key;
1697 struct btrfs_path *path;
1699 path = btrfs_alloc_path();
1703 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1704 key.type = BTRFS_DEV_ITEM_KEY;
1705 key.offset = (u64)-1;
1707 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1711 BUG_ON(ret == 0); /* Corruption */
1713 ret = btrfs_previous_item(fs_info->chunk_root, path,
1714 BTRFS_DEV_ITEMS_OBJECTID,
1715 BTRFS_DEV_ITEM_KEY);
1719 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1721 *devid_ret = found_key.offset + 1;
1725 btrfs_free_path(path);
1730 * the device information is stored in the chunk root
1731 * the btrfs_device struct should be fully filled in
1733 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1734 struct btrfs_fs_info *fs_info,
1735 struct btrfs_device *device)
1737 struct btrfs_root *root = fs_info->chunk_root;
1739 struct btrfs_path *path;
1740 struct btrfs_dev_item *dev_item;
1741 struct extent_buffer *leaf;
1742 struct btrfs_key key;
1745 path = btrfs_alloc_path();
1749 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1750 key.type = BTRFS_DEV_ITEM_KEY;
1751 key.offset = device->devid;
1753 ret = btrfs_insert_empty_item(trans, root, path, &key,
1758 leaf = path->nodes[0];
1759 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1761 btrfs_set_device_id(leaf, dev_item, device->devid);
1762 btrfs_set_device_generation(leaf, dev_item, 0);
1763 btrfs_set_device_type(leaf, dev_item, device->type);
1764 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1765 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1766 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1767 btrfs_set_device_total_bytes(leaf, dev_item,
1768 btrfs_device_get_disk_total_bytes(device));
1769 btrfs_set_device_bytes_used(leaf, dev_item,
1770 btrfs_device_get_bytes_used(device));
1771 btrfs_set_device_group(leaf, dev_item, 0);
1772 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1773 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1774 btrfs_set_device_start_offset(leaf, dev_item, 0);
1776 ptr = btrfs_device_uuid(dev_item);
1777 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1778 ptr = btrfs_device_fsid(dev_item);
1779 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1780 btrfs_mark_buffer_dirty(leaf);
1784 btrfs_free_path(path);
1789 * Function to update ctime/mtime for a given device path.
1790 * Mainly used for ctime/mtime based probe like libblkid.
1792 static void update_dev_time(const char *path_name)
1796 filp = filp_open(path_name, O_RDWR, 0);
1799 file_update_time(filp);
1800 filp_close(filp, NULL);
1803 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1804 struct btrfs_device *device)
1806 struct btrfs_root *root = fs_info->chunk_root;
1808 struct btrfs_path *path;
1809 struct btrfs_key key;
1810 struct btrfs_trans_handle *trans;
1812 path = btrfs_alloc_path();
1816 trans = btrfs_start_transaction(root, 0);
1817 if (IS_ERR(trans)) {
1818 btrfs_free_path(path);
1819 return PTR_ERR(trans);
1821 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1822 key.type = BTRFS_DEV_ITEM_KEY;
1823 key.offset = device->devid;
1825 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1829 btrfs_abort_transaction(trans, ret);
1830 btrfs_end_transaction(trans);
1834 ret = btrfs_del_item(trans, root, path);
1836 btrfs_abort_transaction(trans, ret);
1837 btrfs_end_transaction(trans);
1841 btrfs_free_path(path);
1843 ret = btrfs_commit_transaction(trans);
1848 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1849 * filesystem. It's up to the caller to adjust that number regarding eg. device
1852 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1860 seq = read_seqbegin(&fs_info->profiles_lock);
1862 all_avail = fs_info->avail_data_alloc_bits |
1863 fs_info->avail_system_alloc_bits |
1864 fs_info->avail_metadata_alloc_bits;
1865 } while (read_seqretry(&fs_info->profiles_lock, seq));
1867 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1868 if (!(all_avail & btrfs_raid_group[i]))
1871 if (num_devices < btrfs_raid_array[i].devs_min) {
1872 int ret = btrfs_raid_mindev_error[i];
1882 static struct btrfs_device * btrfs_find_next_active_device(
1883 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1885 struct btrfs_device *next_device;
1887 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1888 if (next_device != device &&
1889 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1890 && next_device->bdev)
1898 * Helper function to check if the given device is part of s_bdev / latest_bdev
1899 * and replace it with the provided or the next active device, in the context
1900 * where this function called, there should be always be another device (or
1901 * this_dev) which is active.
1903 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1904 struct btrfs_device *device, struct btrfs_device *this_dev)
1906 struct btrfs_device *next_device;
1909 next_device = this_dev;
1911 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1913 ASSERT(next_device);
1915 if (fs_info->sb->s_bdev &&
1916 (fs_info->sb->s_bdev == device->bdev))
1917 fs_info->sb->s_bdev = next_device->bdev;
1919 if (fs_info->fs_devices->latest_bdev == device->bdev)
1920 fs_info->fs_devices->latest_bdev = next_device->bdev;
1923 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1926 struct btrfs_device *device;
1927 struct btrfs_fs_devices *cur_devices;
1931 mutex_lock(&fs_info->volume_mutex);
1932 mutex_lock(&uuid_mutex);
1934 num_devices = fs_info->fs_devices->num_devices;
1935 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1936 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1937 WARN_ON(num_devices < 1);
1940 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1942 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1946 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1951 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1952 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1956 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1957 fs_info->fs_devices->rw_devices == 1) {
1958 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1962 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1963 mutex_lock(&fs_info->chunk_mutex);
1964 list_del_init(&device->dev_alloc_list);
1965 device->fs_devices->rw_devices--;
1966 mutex_unlock(&fs_info->chunk_mutex);
1969 mutex_unlock(&uuid_mutex);
1970 ret = btrfs_shrink_device(device, 0);
1971 mutex_lock(&uuid_mutex);
1976 * TODO: the superblock still includes this device in its num_devices
1977 * counter although write_all_supers() is not locked out. This
1978 * could give a filesystem state which requires a degraded mount.
1980 ret = btrfs_rm_dev_item(fs_info, device);
1984 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1985 btrfs_scrub_cancel_dev(fs_info, device);
1988 * the device list mutex makes sure that we don't change
1989 * the device list while someone else is writing out all
1990 * the device supers. Whoever is writing all supers, should
1991 * lock the device list mutex before getting the number of
1992 * devices in the super block (super_copy). Conversely,
1993 * whoever updates the number of devices in the super block
1994 * (super_copy) should hold the device list mutex.
1997 cur_devices = device->fs_devices;
1998 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1999 list_del_rcu(&device->dev_list);
2001 device->fs_devices->num_devices--;
2002 device->fs_devices->total_devices--;
2004 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2005 device->fs_devices->missing_devices--;
2007 btrfs_assign_next_active_device(fs_info, device, NULL);
2010 device->fs_devices->open_devices--;
2011 /* remove sysfs entry */
2012 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2015 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2016 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2017 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2020 * at this point, the device is zero sized and detached from
2021 * the devices list. All that's left is to zero out the old
2022 * supers and free the device.
2024 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2025 btrfs_scratch_superblocks(device->bdev, device->name->str);
2027 btrfs_close_bdev(device);
2028 call_rcu(&device->rcu, free_device_rcu);
2030 if (cur_devices->open_devices == 0) {
2031 struct btrfs_fs_devices *fs_devices;
2032 fs_devices = fs_info->fs_devices;
2033 while (fs_devices) {
2034 if (fs_devices->seed == cur_devices) {
2035 fs_devices->seed = cur_devices->seed;
2038 fs_devices = fs_devices->seed;
2040 cur_devices->seed = NULL;
2041 __btrfs_close_devices(cur_devices);
2042 free_fs_devices(cur_devices);
2046 mutex_unlock(&uuid_mutex);
2047 mutex_unlock(&fs_info->volume_mutex);
2051 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2052 mutex_lock(&fs_info->chunk_mutex);
2053 list_add(&device->dev_alloc_list,
2054 &fs_info->fs_devices->alloc_list);
2055 device->fs_devices->rw_devices++;
2056 mutex_unlock(&fs_info->chunk_mutex);
2061 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2062 struct btrfs_device *srcdev)
2064 struct btrfs_fs_devices *fs_devices;
2066 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2069 * in case of fs with no seed, srcdev->fs_devices will point
2070 * to fs_devices of fs_info. However when the dev being replaced is
2071 * a seed dev it will point to the seed's local fs_devices. In short
2072 * srcdev will have its correct fs_devices in both the cases.
2074 fs_devices = srcdev->fs_devices;
2076 list_del_rcu(&srcdev->dev_list);
2077 list_del(&srcdev->dev_alloc_list);
2078 fs_devices->num_devices--;
2079 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2080 fs_devices->missing_devices--;
2082 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2083 fs_devices->rw_devices--;
2086 fs_devices->open_devices--;
2089 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2090 struct btrfs_device *srcdev)
2092 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2094 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2095 /* zero out the old super if it is writable */
2096 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2099 btrfs_close_bdev(srcdev);
2100 call_rcu(&srcdev->rcu, free_device_rcu);
2102 /* if this is no devs we rather delete the fs_devices */
2103 if (!fs_devices->num_devices) {
2104 struct btrfs_fs_devices *tmp_fs_devices;
2107 * On a mounted FS, num_devices can't be zero unless it's a
2108 * seed. In case of a seed device being replaced, the replace
2109 * target added to the sprout FS, so there will be no more
2110 * device left under the seed FS.
2112 ASSERT(fs_devices->seeding);
2114 tmp_fs_devices = fs_info->fs_devices;
2115 while (tmp_fs_devices) {
2116 if (tmp_fs_devices->seed == fs_devices) {
2117 tmp_fs_devices->seed = fs_devices->seed;
2120 tmp_fs_devices = tmp_fs_devices->seed;
2122 fs_devices->seed = NULL;
2123 __btrfs_close_devices(fs_devices);
2124 free_fs_devices(fs_devices);
2128 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2129 struct btrfs_device *tgtdev)
2131 mutex_lock(&uuid_mutex);
2133 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2135 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2138 fs_info->fs_devices->open_devices--;
2140 fs_info->fs_devices->num_devices--;
2142 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2144 list_del_rcu(&tgtdev->dev_list);
2146 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2147 mutex_unlock(&uuid_mutex);
2150 * The update_dev_time() with in btrfs_scratch_superblocks()
2151 * may lead to a call to btrfs_show_devname() which will try
2152 * to hold device_list_mutex. And here this device
2153 * is already out of device list, so we don't have to hold
2154 * the device_list_mutex lock.
2156 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2158 btrfs_close_bdev(tgtdev);
2159 call_rcu(&tgtdev->rcu, free_device_rcu);
2162 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2163 const char *device_path,
2164 struct btrfs_device **device)
2167 struct btrfs_super_block *disk_super;
2170 struct block_device *bdev;
2171 struct buffer_head *bh;
2174 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2175 fs_info->bdev_holder, 0, &bdev, &bh);
2178 disk_super = (struct btrfs_super_block *)bh->b_data;
2179 devid = btrfs_stack_device_id(&disk_super->dev_item);
2180 dev_uuid = disk_super->dev_item.uuid;
2181 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2185 blkdev_put(bdev, FMODE_READ);
2189 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2190 const char *device_path,
2191 struct btrfs_device **device)
2194 if (strcmp(device_path, "missing") == 0) {
2195 struct list_head *devices;
2196 struct btrfs_device *tmp;
2198 devices = &fs_info->fs_devices->devices;
2200 * It is safe to read the devices since the volume_mutex
2201 * is held by the caller.
2203 list_for_each_entry(tmp, devices, dev_list) {
2204 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2205 &tmp->dev_state) && !tmp->bdev) {
2212 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2216 return btrfs_find_device_by_path(fs_info, device_path, device);
2221 * Lookup a device given by device id, or the path if the id is 0.
2223 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2224 const char *devpath,
2225 struct btrfs_device **device)
2231 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2235 if (!devpath || !devpath[0])
2238 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2245 * does all the dirty work required for changing file system's UUID.
2247 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2249 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2250 struct btrfs_fs_devices *old_devices;
2251 struct btrfs_fs_devices *seed_devices;
2252 struct btrfs_super_block *disk_super = fs_info->super_copy;
2253 struct btrfs_device *device;
2256 BUG_ON(!mutex_is_locked(&uuid_mutex));
2257 if (!fs_devices->seeding)
2260 seed_devices = alloc_fs_devices(NULL);
2261 if (IS_ERR(seed_devices))
2262 return PTR_ERR(seed_devices);
2264 old_devices = clone_fs_devices(fs_devices);
2265 if (IS_ERR(old_devices)) {
2266 kfree(seed_devices);
2267 return PTR_ERR(old_devices);
2270 list_add(&old_devices->list, &fs_uuids);
2272 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2273 seed_devices->opened = 1;
2274 INIT_LIST_HEAD(&seed_devices->devices);
2275 INIT_LIST_HEAD(&seed_devices->alloc_list);
2276 mutex_init(&seed_devices->device_list_mutex);
2278 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2279 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2281 list_for_each_entry(device, &seed_devices->devices, dev_list)
2282 device->fs_devices = seed_devices;
2284 mutex_lock(&fs_info->chunk_mutex);
2285 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2286 mutex_unlock(&fs_info->chunk_mutex);
2288 fs_devices->seeding = 0;
2289 fs_devices->num_devices = 0;
2290 fs_devices->open_devices = 0;
2291 fs_devices->missing_devices = 0;
2292 fs_devices->rotating = 0;
2293 fs_devices->seed = seed_devices;
2295 generate_random_uuid(fs_devices->fsid);
2296 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2297 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2298 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2300 super_flags = btrfs_super_flags(disk_super) &
2301 ~BTRFS_SUPER_FLAG_SEEDING;
2302 btrfs_set_super_flags(disk_super, super_flags);
2308 * Store the expected generation for seed devices in device items.
2310 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2311 struct btrfs_fs_info *fs_info)
2313 struct btrfs_root *root = fs_info->chunk_root;
2314 struct btrfs_path *path;
2315 struct extent_buffer *leaf;
2316 struct btrfs_dev_item *dev_item;
2317 struct btrfs_device *device;
2318 struct btrfs_key key;
2319 u8 fs_uuid[BTRFS_FSID_SIZE];
2320 u8 dev_uuid[BTRFS_UUID_SIZE];
2324 path = btrfs_alloc_path();
2328 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2330 key.type = BTRFS_DEV_ITEM_KEY;
2333 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2337 leaf = path->nodes[0];
2339 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2340 ret = btrfs_next_leaf(root, path);
2345 leaf = path->nodes[0];
2346 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2347 btrfs_release_path(path);
2351 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2352 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2353 key.type != BTRFS_DEV_ITEM_KEY)
2356 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2357 struct btrfs_dev_item);
2358 devid = btrfs_device_id(leaf, dev_item);
2359 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2361 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2363 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2364 BUG_ON(!device); /* Logic error */
2366 if (device->fs_devices->seeding) {
2367 btrfs_set_device_generation(leaf, dev_item,
2368 device->generation);
2369 btrfs_mark_buffer_dirty(leaf);
2377 btrfs_free_path(path);
2381 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2383 struct btrfs_root *root = fs_info->dev_root;
2384 struct request_queue *q;
2385 struct btrfs_trans_handle *trans;
2386 struct btrfs_device *device;
2387 struct block_device *bdev;
2388 struct list_head *devices;
2389 struct super_block *sb = fs_info->sb;
2390 struct rcu_string *name;
2392 int seeding_dev = 0;
2394 bool unlocked = false;
2396 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2399 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2400 fs_info->bdev_holder);
2402 return PTR_ERR(bdev);
2404 if (fs_info->fs_devices->seeding) {
2406 down_write(&sb->s_umount);
2407 mutex_lock(&uuid_mutex);
2410 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2412 devices = &fs_info->fs_devices->devices;
2414 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2415 list_for_each_entry(device, devices, dev_list) {
2416 if (device->bdev == bdev) {
2419 &fs_info->fs_devices->device_list_mutex);
2423 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2425 device = btrfs_alloc_device(fs_info, NULL, NULL);
2426 if (IS_ERR(device)) {
2427 /* we can safely leave the fs_devices entry around */
2428 ret = PTR_ERR(device);
2432 name = rcu_string_strdup(device_path, GFP_KERNEL);
2435 goto error_free_device;
2437 rcu_assign_pointer(device->name, name);
2439 trans = btrfs_start_transaction(root, 0);
2440 if (IS_ERR(trans)) {
2441 ret = PTR_ERR(trans);
2442 goto error_free_device;
2445 q = bdev_get_queue(bdev);
2446 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2447 device->generation = trans->transid;
2448 device->io_width = fs_info->sectorsize;
2449 device->io_align = fs_info->sectorsize;
2450 device->sector_size = fs_info->sectorsize;
2451 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2452 fs_info->sectorsize);
2453 device->disk_total_bytes = device->total_bytes;
2454 device->commit_total_bytes = device->total_bytes;
2455 device->fs_info = fs_info;
2456 device->bdev = bdev;
2457 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2458 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2459 device->mode = FMODE_EXCL;
2460 device->dev_stats_valid = 1;
2461 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2464 sb->s_flags &= ~SB_RDONLY;
2465 ret = btrfs_prepare_sprout(fs_info);
2467 btrfs_abort_transaction(trans, ret);
2472 device->fs_devices = fs_info->fs_devices;
2474 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2475 mutex_lock(&fs_info->chunk_mutex);
2476 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2477 list_add(&device->dev_alloc_list,
2478 &fs_info->fs_devices->alloc_list);
2479 fs_info->fs_devices->num_devices++;
2480 fs_info->fs_devices->open_devices++;
2481 fs_info->fs_devices->rw_devices++;
2482 fs_info->fs_devices->total_devices++;
2483 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2485 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2487 if (!blk_queue_nonrot(q))
2488 fs_info->fs_devices->rotating = 1;
2490 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2491 btrfs_set_super_total_bytes(fs_info->super_copy,
2492 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2494 tmp = btrfs_super_num_devices(fs_info->super_copy);
2495 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2497 /* add sysfs device entry */
2498 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2501 * we've got more storage, clear any full flags on the space
2504 btrfs_clear_space_info_full(fs_info);
2506 mutex_unlock(&fs_info->chunk_mutex);
2507 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2510 mutex_lock(&fs_info->chunk_mutex);
2511 ret = init_first_rw_device(trans, fs_info);
2512 mutex_unlock(&fs_info->chunk_mutex);
2514 btrfs_abort_transaction(trans, ret);
2519 ret = btrfs_add_dev_item(trans, fs_info, device);
2521 btrfs_abort_transaction(trans, ret);
2526 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2528 ret = btrfs_finish_sprout(trans, fs_info);
2530 btrfs_abort_transaction(trans, ret);
2534 /* Sprouting would change fsid of the mounted root,
2535 * so rename the fsid on the sysfs
2537 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2539 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2541 "sysfs: failed to create fsid for sprout");
2544 ret = btrfs_commit_transaction(trans);
2547 mutex_unlock(&uuid_mutex);
2548 up_write(&sb->s_umount);
2551 if (ret) /* transaction commit */
2554 ret = btrfs_relocate_sys_chunks(fs_info);
2556 btrfs_handle_fs_error(fs_info, ret,
2557 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2558 trans = btrfs_attach_transaction(root);
2559 if (IS_ERR(trans)) {
2560 if (PTR_ERR(trans) == -ENOENT)
2562 ret = PTR_ERR(trans);
2566 ret = btrfs_commit_transaction(trans);
2569 /* Update ctime/mtime for libblkid */
2570 update_dev_time(device_path);
2574 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2577 sb->s_flags |= SB_RDONLY;
2579 btrfs_end_transaction(trans);
2581 free_device(device);
2583 blkdev_put(bdev, FMODE_EXCL);
2584 if (seeding_dev && !unlocked) {
2585 mutex_unlock(&uuid_mutex);
2586 up_write(&sb->s_umount);
2591 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2592 const char *device_path,
2593 struct btrfs_device *srcdev,
2594 struct btrfs_device **device_out)
2596 struct btrfs_device *device;
2597 struct block_device *bdev;
2598 struct list_head *devices;
2599 struct rcu_string *name;
2600 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2604 if (fs_info->fs_devices->seeding) {
2605 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2609 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2610 fs_info->bdev_holder);
2612 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2613 return PTR_ERR(bdev);
2616 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2618 devices = &fs_info->fs_devices->devices;
2619 list_for_each_entry(device, devices, dev_list) {
2620 if (device->bdev == bdev) {
2622 "target device is in the filesystem!");
2629 if (i_size_read(bdev->bd_inode) <
2630 btrfs_device_get_total_bytes(srcdev)) {
2632 "target device is smaller than source device!");
2638 device = btrfs_alloc_device(NULL, &devid, NULL);
2639 if (IS_ERR(device)) {
2640 ret = PTR_ERR(device);
2644 name = rcu_string_strdup(device_path, GFP_KERNEL);
2646 free_device(device);
2650 rcu_assign_pointer(device->name, name);
2652 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2653 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2654 device->generation = 0;
2655 device->io_width = fs_info->sectorsize;
2656 device->io_align = fs_info->sectorsize;
2657 device->sector_size = fs_info->sectorsize;
2658 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2659 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2660 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2661 device->commit_total_bytes = srcdev->commit_total_bytes;
2662 device->commit_bytes_used = device->bytes_used;
2663 device->fs_info = fs_info;
2664 device->bdev = bdev;
2665 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2666 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2667 device->mode = FMODE_EXCL;
2668 device->dev_stats_valid = 1;
2669 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2670 device->fs_devices = fs_info->fs_devices;
2671 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2672 fs_info->fs_devices->num_devices++;
2673 fs_info->fs_devices->open_devices++;
2674 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2676 *device_out = device;
2680 blkdev_put(bdev, FMODE_EXCL);
2684 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2685 struct btrfs_device *device)
2688 struct btrfs_path *path;
2689 struct btrfs_root *root = device->fs_info->chunk_root;
2690 struct btrfs_dev_item *dev_item;
2691 struct extent_buffer *leaf;
2692 struct btrfs_key key;
2694 path = btrfs_alloc_path();
2698 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2699 key.type = BTRFS_DEV_ITEM_KEY;
2700 key.offset = device->devid;
2702 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2711 leaf = path->nodes[0];
2712 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2714 btrfs_set_device_id(leaf, dev_item, device->devid);
2715 btrfs_set_device_type(leaf, dev_item, device->type);
2716 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2717 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2718 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2719 btrfs_set_device_total_bytes(leaf, dev_item,
2720 btrfs_device_get_disk_total_bytes(device));
2721 btrfs_set_device_bytes_used(leaf, dev_item,
2722 btrfs_device_get_bytes_used(device));
2723 btrfs_mark_buffer_dirty(leaf);
2726 btrfs_free_path(path);
2730 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2731 struct btrfs_device *device, u64 new_size)
2733 struct btrfs_fs_info *fs_info = device->fs_info;
2734 struct btrfs_super_block *super_copy = fs_info->super_copy;
2735 struct btrfs_fs_devices *fs_devices;
2739 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2742 new_size = round_down(new_size, fs_info->sectorsize);
2744 mutex_lock(&fs_info->chunk_mutex);
2745 old_total = btrfs_super_total_bytes(super_copy);
2746 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2748 if (new_size <= device->total_bytes ||
2749 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2750 mutex_unlock(&fs_info->chunk_mutex);
2754 fs_devices = fs_info->fs_devices;
2756 btrfs_set_super_total_bytes(super_copy,
2757 round_down(old_total + diff, fs_info->sectorsize));
2758 device->fs_devices->total_rw_bytes += diff;
2760 btrfs_device_set_total_bytes(device, new_size);
2761 btrfs_device_set_disk_total_bytes(device, new_size);
2762 btrfs_clear_space_info_full(device->fs_info);
2763 if (list_empty(&device->resized_list))
2764 list_add_tail(&device->resized_list,
2765 &fs_devices->resized_devices);
2766 mutex_unlock(&fs_info->chunk_mutex);
2768 return btrfs_update_device(trans, device);
2771 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2772 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2774 struct btrfs_root *root = fs_info->chunk_root;
2776 struct btrfs_path *path;
2777 struct btrfs_key key;
2779 path = btrfs_alloc_path();
2783 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2784 key.offset = chunk_offset;
2785 key.type = BTRFS_CHUNK_ITEM_KEY;
2787 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2790 else if (ret > 0) { /* Logic error or corruption */
2791 btrfs_handle_fs_error(fs_info, -ENOENT,
2792 "Failed lookup while freeing chunk.");
2797 ret = btrfs_del_item(trans, root, path);
2799 btrfs_handle_fs_error(fs_info, ret,
2800 "Failed to delete chunk item.");
2802 btrfs_free_path(path);
2806 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2808 struct btrfs_super_block *super_copy = fs_info->super_copy;
2809 struct btrfs_disk_key *disk_key;
2810 struct btrfs_chunk *chunk;
2817 struct btrfs_key key;
2819 mutex_lock(&fs_info->chunk_mutex);
2820 array_size = btrfs_super_sys_array_size(super_copy);
2822 ptr = super_copy->sys_chunk_array;
2825 while (cur < array_size) {
2826 disk_key = (struct btrfs_disk_key *)ptr;
2827 btrfs_disk_key_to_cpu(&key, disk_key);
2829 len = sizeof(*disk_key);
2831 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2832 chunk = (struct btrfs_chunk *)(ptr + len);
2833 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2834 len += btrfs_chunk_item_size(num_stripes);
2839 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2840 key.offset == chunk_offset) {
2841 memmove(ptr, ptr + len, array_size - (cur + len));
2843 btrfs_set_super_sys_array_size(super_copy, array_size);
2849 mutex_unlock(&fs_info->chunk_mutex);
2853 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2854 u64 logical, u64 length)
2856 struct extent_map_tree *em_tree;
2857 struct extent_map *em;
2859 em_tree = &fs_info->mapping_tree.map_tree;
2860 read_lock(&em_tree->lock);
2861 em = lookup_extent_mapping(em_tree, logical, length);
2862 read_unlock(&em_tree->lock);
2865 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2867 return ERR_PTR(-EINVAL);
2870 if (em->start > logical || em->start + em->len < logical) {
2872 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2873 logical, length, em->start, em->start + em->len);
2874 free_extent_map(em);
2875 return ERR_PTR(-EINVAL);
2878 /* callers are responsible for dropping em's ref. */
2882 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2883 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2885 struct extent_map *em;
2886 struct map_lookup *map;
2887 u64 dev_extent_len = 0;
2889 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2891 em = get_chunk_map(fs_info, chunk_offset, 1);
2894 * This is a logic error, but we don't want to just rely on the
2895 * user having built with ASSERT enabled, so if ASSERT doesn't
2896 * do anything we still error out.
2901 map = em->map_lookup;
2902 mutex_lock(&fs_info->chunk_mutex);
2903 check_system_chunk(trans, fs_info, map->type);
2904 mutex_unlock(&fs_info->chunk_mutex);
2907 * Take the device list mutex to prevent races with the final phase of
2908 * a device replace operation that replaces the device object associated
2909 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2911 mutex_lock(&fs_devices->device_list_mutex);
2912 for (i = 0; i < map->num_stripes; i++) {
2913 struct btrfs_device *device = map->stripes[i].dev;
2914 ret = btrfs_free_dev_extent(trans, device,
2915 map->stripes[i].physical,
2918 mutex_unlock(&fs_devices->device_list_mutex);
2919 btrfs_abort_transaction(trans, ret);
2923 if (device->bytes_used > 0) {
2924 mutex_lock(&fs_info->chunk_mutex);
2925 btrfs_device_set_bytes_used(device,
2926 device->bytes_used - dev_extent_len);
2927 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2928 btrfs_clear_space_info_full(fs_info);
2929 mutex_unlock(&fs_info->chunk_mutex);
2932 if (map->stripes[i].dev) {
2933 ret = btrfs_update_device(trans, map->stripes[i].dev);
2935 mutex_unlock(&fs_devices->device_list_mutex);
2936 btrfs_abort_transaction(trans, ret);
2941 mutex_unlock(&fs_devices->device_list_mutex);
2943 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2945 btrfs_abort_transaction(trans, ret);
2949 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2951 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2952 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2954 btrfs_abort_transaction(trans, ret);
2959 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2961 btrfs_abort_transaction(trans, ret);
2967 free_extent_map(em);
2971 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2973 struct btrfs_root *root = fs_info->chunk_root;
2974 struct btrfs_trans_handle *trans;
2978 * Prevent races with automatic removal of unused block groups.
2979 * After we relocate and before we remove the chunk with offset
2980 * chunk_offset, automatic removal of the block group can kick in,
2981 * resulting in a failure when calling btrfs_remove_chunk() below.
2983 * Make sure to acquire this mutex before doing a tree search (dev
2984 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2985 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2986 * we release the path used to search the chunk/dev tree and before
2987 * the current task acquires this mutex and calls us.
2989 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2991 ret = btrfs_can_relocate(fs_info, chunk_offset);
2995 /* step one, relocate all the extents inside this chunk */
2996 btrfs_scrub_pause(fs_info);
2997 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2998 btrfs_scrub_continue(fs_info);
3002 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3004 if (IS_ERR(trans)) {
3005 ret = PTR_ERR(trans);
3006 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3011 * step two, delete the device extents and the
3012 * chunk tree entries
3014 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3015 btrfs_end_transaction(trans);
3019 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3021 struct btrfs_root *chunk_root = fs_info->chunk_root;
3022 struct btrfs_path *path;
3023 struct extent_buffer *leaf;
3024 struct btrfs_chunk *chunk;
3025 struct btrfs_key key;
3026 struct btrfs_key found_key;
3028 bool retried = false;
3032 path = btrfs_alloc_path();
3037 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3038 key.offset = (u64)-1;
3039 key.type = BTRFS_CHUNK_ITEM_KEY;
3042 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3043 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3045 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3048 BUG_ON(ret == 0); /* Corruption */
3050 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3053 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3059 leaf = path->nodes[0];
3060 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3062 chunk = btrfs_item_ptr(leaf, path->slots[0],
3063 struct btrfs_chunk);
3064 chunk_type = btrfs_chunk_type(leaf, chunk);
3065 btrfs_release_path(path);
3067 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3068 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3074 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3076 if (found_key.offset == 0)
3078 key.offset = found_key.offset - 1;
3081 if (failed && !retried) {
3085 } else if (WARN_ON(failed && retried)) {
3089 btrfs_free_path(path);
3094 * return 1 : allocate a data chunk successfully,
3095 * return <0: errors during allocating a data chunk,
3096 * return 0 : no need to allocate a data chunk.
3098 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3101 struct btrfs_block_group_cache *cache;
3105 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3107 chunk_type = cache->flags;
3108 btrfs_put_block_group(cache);
3110 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3111 spin_lock(&fs_info->data_sinfo->lock);
3112 bytes_used = fs_info->data_sinfo->bytes_used;
3113 spin_unlock(&fs_info->data_sinfo->lock);
3116 struct btrfs_trans_handle *trans;
3119 trans = btrfs_join_transaction(fs_info->tree_root);
3121 return PTR_ERR(trans);
3123 ret = btrfs_force_chunk_alloc(trans, fs_info,
3124 BTRFS_BLOCK_GROUP_DATA);
3125 btrfs_end_transaction(trans);
3135 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3136 struct btrfs_balance_control *bctl)
3138 struct btrfs_root *root = fs_info->tree_root;
3139 struct btrfs_trans_handle *trans;
3140 struct btrfs_balance_item *item;
3141 struct btrfs_disk_balance_args disk_bargs;
3142 struct btrfs_path *path;
3143 struct extent_buffer *leaf;
3144 struct btrfs_key key;
3147 path = btrfs_alloc_path();
3151 trans = btrfs_start_transaction(root, 0);
3152 if (IS_ERR(trans)) {
3153 btrfs_free_path(path);
3154 return PTR_ERR(trans);
3157 key.objectid = BTRFS_BALANCE_OBJECTID;
3158 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3161 ret = btrfs_insert_empty_item(trans, root, path, &key,
3166 leaf = path->nodes[0];
3167 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3169 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3171 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3172 btrfs_set_balance_data(leaf, item, &disk_bargs);
3173 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3174 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3175 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3176 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3178 btrfs_set_balance_flags(leaf, item, bctl->flags);
3180 btrfs_mark_buffer_dirty(leaf);
3182 btrfs_free_path(path);
3183 err = btrfs_commit_transaction(trans);
3189 static int del_balance_item(struct btrfs_fs_info *fs_info)
3191 struct btrfs_root *root = fs_info->tree_root;
3192 struct btrfs_trans_handle *trans;
3193 struct btrfs_path *path;
3194 struct btrfs_key key;
3197 path = btrfs_alloc_path();
3201 trans = btrfs_start_transaction(root, 0);
3202 if (IS_ERR(trans)) {
3203 btrfs_free_path(path);
3204 return PTR_ERR(trans);
3207 key.objectid = BTRFS_BALANCE_OBJECTID;
3208 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3211 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3219 ret = btrfs_del_item(trans, root, path);
3221 btrfs_free_path(path);
3222 err = btrfs_commit_transaction(trans);
3229 * This is a heuristic used to reduce the number of chunks balanced on
3230 * resume after balance was interrupted.
3232 static void update_balance_args(struct btrfs_balance_control *bctl)
3235 * Turn on soft mode for chunk types that were being converted.
3237 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3238 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3239 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3240 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3241 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3242 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3245 * Turn on usage filter if is not already used. The idea is
3246 * that chunks that we have already balanced should be
3247 * reasonably full. Don't do it for chunks that are being
3248 * converted - that will keep us from relocating unconverted
3249 * (albeit full) chunks.
3251 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3252 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3253 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3254 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3255 bctl->data.usage = 90;
3257 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3258 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3259 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3260 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3261 bctl->sys.usage = 90;
3263 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3264 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3265 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3266 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3267 bctl->meta.usage = 90;
3272 * Should be called with both balance and volume mutexes held to
3273 * serialize other volume operations (add_dev/rm_dev/resize) with
3274 * restriper. Same goes for unset_balance_control.
3276 static void set_balance_control(struct btrfs_balance_control *bctl)
3278 struct btrfs_fs_info *fs_info = bctl->fs_info;
3280 BUG_ON(fs_info->balance_ctl);
3282 spin_lock(&fs_info->balance_lock);
3283 fs_info->balance_ctl = bctl;
3284 spin_unlock(&fs_info->balance_lock);
3287 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3289 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3291 BUG_ON(!fs_info->balance_ctl);
3293 spin_lock(&fs_info->balance_lock);
3294 fs_info->balance_ctl = NULL;
3295 spin_unlock(&fs_info->balance_lock);
3301 * Balance filters. Return 1 if chunk should be filtered out
3302 * (should not be balanced).
3304 static int chunk_profiles_filter(u64 chunk_type,
3305 struct btrfs_balance_args *bargs)
3307 chunk_type = chunk_to_extended(chunk_type) &
3308 BTRFS_EXTENDED_PROFILE_MASK;
3310 if (bargs->profiles & chunk_type)
3316 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3317 struct btrfs_balance_args *bargs)
3319 struct btrfs_block_group_cache *cache;
3321 u64 user_thresh_min;
3322 u64 user_thresh_max;
3325 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3326 chunk_used = btrfs_block_group_used(&cache->item);
3328 if (bargs->usage_min == 0)
3329 user_thresh_min = 0;
3331 user_thresh_min = div_factor_fine(cache->key.offset,
3334 if (bargs->usage_max == 0)
3335 user_thresh_max = 1;
3336 else if (bargs->usage_max > 100)
3337 user_thresh_max = cache->key.offset;
3339 user_thresh_max = div_factor_fine(cache->key.offset,
3342 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3345 btrfs_put_block_group(cache);
3349 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3350 u64 chunk_offset, struct btrfs_balance_args *bargs)
3352 struct btrfs_block_group_cache *cache;
3353 u64 chunk_used, user_thresh;
3356 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3357 chunk_used = btrfs_block_group_used(&cache->item);
3359 if (bargs->usage_min == 0)
3361 else if (bargs->usage > 100)
3362 user_thresh = cache->key.offset;
3364 user_thresh = div_factor_fine(cache->key.offset,
3367 if (chunk_used < user_thresh)
3370 btrfs_put_block_group(cache);
3374 static int chunk_devid_filter(struct extent_buffer *leaf,
3375 struct btrfs_chunk *chunk,
3376 struct btrfs_balance_args *bargs)
3378 struct btrfs_stripe *stripe;
3379 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3382 for (i = 0; i < num_stripes; i++) {
3383 stripe = btrfs_stripe_nr(chunk, i);
3384 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3391 /* [pstart, pend) */
3392 static int chunk_drange_filter(struct extent_buffer *leaf,
3393 struct btrfs_chunk *chunk,
3394 struct btrfs_balance_args *bargs)
3396 struct btrfs_stripe *stripe;
3397 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3403 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3406 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3407 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3408 factor = num_stripes / 2;
3409 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3410 factor = num_stripes - 1;
3411 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3412 factor = num_stripes - 2;
3414 factor = num_stripes;
3417 for (i = 0; i < num_stripes; i++) {
3418 stripe = btrfs_stripe_nr(chunk, i);
3419 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3422 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3423 stripe_length = btrfs_chunk_length(leaf, chunk);
3424 stripe_length = div_u64(stripe_length, factor);
3426 if (stripe_offset < bargs->pend &&
3427 stripe_offset + stripe_length > bargs->pstart)
3434 /* [vstart, vend) */
3435 static int chunk_vrange_filter(struct extent_buffer *leaf,
3436 struct btrfs_chunk *chunk,
3438 struct btrfs_balance_args *bargs)
3440 if (chunk_offset < bargs->vend &&
3441 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3442 /* at least part of the chunk is inside this vrange */
3448 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3449 struct btrfs_chunk *chunk,
3450 struct btrfs_balance_args *bargs)
3452 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3454 if (bargs->stripes_min <= num_stripes
3455 && num_stripes <= bargs->stripes_max)
3461 static int chunk_soft_convert_filter(u64 chunk_type,
3462 struct btrfs_balance_args *bargs)
3464 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3467 chunk_type = chunk_to_extended(chunk_type) &
3468 BTRFS_EXTENDED_PROFILE_MASK;
3470 if (bargs->target == chunk_type)
3476 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3477 struct extent_buffer *leaf,
3478 struct btrfs_chunk *chunk, u64 chunk_offset)
3480 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3481 struct btrfs_balance_args *bargs = NULL;
3482 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3485 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3486 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3490 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3491 bargs = &bctl->data;
3492 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3494 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3495 bargs = &bctl->meta;
3497 /* profiles filter */
3498 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3499 chunk_profiles_filter(chunk_type, bargs)) {
3504 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3505 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3507 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3508 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3513 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3514 chunk_devid_filter(leaf, chunk, bargs)) {
3518 /* drange filter, makes sense only with devid filter */
3519 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3520 chunk_drange_filter(leaf, chunk, bargs)) {
3525 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3526 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3530 /* stripes filter */
3531 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3532 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3536 /* soft profile changing mode */
3537 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3538 chunk_soft_convert_filter(chunk_type, bargs)) {
3543 * limited by count, must be the last filter
3545 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3546 if (bargs->limit == 0)
3550 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3552 * Same logic as the 'limit' filter; the minimum cannot be
3553 * determined here because we do not have the global information
3554 * about the count of all chunks that satisfy the filters.
3556 if (bargs->limit_max == 0)
3565 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3567 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3568 struct btrfs_root *chunk_root = fs_info->chunk_root;
3569 struct btrfs_root *dev_root = fs_info->dev_root;
3570 struct list_head *devices;
3571 struct btrfs_device *device;
3575 struct btrfs_chunk *chunk;
3576 struct btrfs_path *path = NULL;
3577 struct btrfs_key key;
3578 struct btrfs_key found_key;
3579 struct btrfs_trans_handle *trans;
3580 struct extent_buffer *leaf;
3583 int enospc_errors = 0;
3584 bool counting = true;
3585 /* The single value limit and min/max limits use the same bytes in the */
3586 u64 limit_data = bctl->data.limit;
3587 u64 limit_meta = bctl->meta.limit;
3588 u64 limit_sys = bctl->sys.limit;
3592 int chunk_reserved = 0;
3594 /* step one make some room on all the devices */
3595 devices = &fs_info->fs_devices->devices;
3596 list_for_each_entry(device, devices, dev_list) {
3597 old_size = btrfs_device_get_total_bytes(device);
3598 size_to_free = div_factor(old_size, 1);
3599 size_to_free = min_t(u64, size_to_free, SZ_1M);
3600 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3601 btrfs_device_get_total_bytes(device) -
3602 btrfs_device_get_bytes_used(device) > size_to_free ||
3603 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3606 ret = btrfs_shrink_device(device, old_size - size_to_free);
3610 /* btrfs_shrink_device never returns ret > 0 */
3615 trans = btrfs_start_transaction(dev_root, 0);
3616 if (IS_ERR(trans)) {
3617 ret = PTR_ERR(trans);
3618 btrfs_info_in_rcu(fs_info,
3619 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3620 rcu_str_deref(device->name), ret,
3621 old_size, old_size - size_to_free);
3625 ret = btrfs_grow_device(trans, device, old_size);
3627 btrfs_end_transaction(trans);
3628 /* btrfs_grow_device never returns ret > 0 */
3630 btrfs_info_in_rcu(fs_info,
3631 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3632 rcu_str_deref(device->name), ret,
3633 old_size, old_size - size_to_free);
3637 btrfs_end_transaction(trans);
3640 /* step two, relocate all the chunks */
3641 path = btrfs_alloc_path();
3647 /* zero out stat counters */
3648 spin_lock(&fs_info->balance_lock);
3649 memset(&bctl->stat, 0, sizeof(bctl->stat));
3650 spin_unlock(&fs_info->balance_lock);
3654 * The single value limit and min/max limits use the same bytes
3657 bctl->data.limit = limit_data;
3658 bctl->meta.limit = limit_meta;
3659 bctl->sys.limit = limit_sys;
3661 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3662 key.offset = (u64)-1;
3663 key.type = BTRFS_CHUNK_ITEM_KEY;
3666 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3667 atomic_read(&fs_info->balance_cancel_req)) {
3672 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3673 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3675 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3680 * this shouldn't happen, it means the last relocate
3684 BUG(); /* FIXME break ? */
3686 ret = btrfs_previous_item(chunk_root, path, 0,
3687 BTRFS_CHUNK_ITEM_KEY);
3689 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3694 leaf = path->nodes[0];
3695 slot = path->slots[0];
3696 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3698 if (found_key.objectid != key.objectid) {
3699 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3703 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3704 chunk_type = btrfs_chunk_type(leaf, chunk);
3707 spin_lock(&fs_info->balance_lock);
3708 bctl->stat.considered++;
3709 spin_unlock(&fs_info->balance_lock);
3712 ret = should_balance_chunk(fs_info, leaf, chunk,
3715 btrfs_release_path(path);
3717 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3722 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3723 spin_lock(&fs_info->balance_lock);
3724 bctl->stat.expected++;
3725 spin_unlock(&fs_info->balance_lock);
3727 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3729 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3731 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3738 * Apply limit_min filter, no need to check if the LIMITS
3739 * filter is used, limit_min is 0 by default
3741 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3742 count_data < bctl->data.limit_min)
3743 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3744 count_meta < bctl->meta.limit_min)
3745 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3746 count_sys < bctl->sys.limit_min)) {
3747 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3751 if (!chunk_reserved) {
3753 * We may be relocating the only data chunk we have,
3754 * which could potentially end up with losing data's
3755 * raid profile, so lets allocate an empty one in
3758 ret = btrfs_may_alloc_data_chunk(fs_info,
3761 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3763 } else if (ret == 1) {
3768 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3769 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3770 if (ret && ret != -ENOSPC)
3772 if (ret == -ENOSPC) {
3775 spin_lock(&fs_info->balance_lock);
3776 bctl->stat.completed++;
3777 spin_unlock(&fs_info->balance_lock);
3780 if (found_key.offset == 0)
3782 key.offset = found_key.offset - 1;
3786 btrfs_release_path(path);
3791 btrfs_free_path(path);
3792 if (enospc_errors) {
3793 btrfs_info(fs_info, "%d enospc errors during balance",
3803 * alloc_profile_is_valid - see if a given profile is valid and reduced
3804 * @flags: profile to validate
3805 * @extended: if true @flags is treated as an extended profile
3807 static int alloc_profile_is_valid(u64 flags, int extended)
3809 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3810 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3812 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3814 /* 1) check that all other bits are zeroed */
3818 /* 2) see if profile is reduced */
3820 return !extended; /* "0" is valid for usual profiles */
3822 /* true if exactly one bit set */
3823 return (flags & (flags - 1)) == 0;
3826 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3828 /* cancel requested || normal exit path */
3829 return atomic_read(&fs_info->balance_cancel_req) ||
3830 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3831 atomic_read(&fs_info->balance_cancel_req) == 0);
3834 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3838 unset_balance_control(fs_info);
3839 ret = del_balance_item(fs_info);
3841 btrfs_handle_fs_error(fs_info, ret, NULL);
3843 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3846 /* Non-zero return value signifies invalidity */
3847 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3850 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3851 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3852 (bctl_arg->target & ~allowed)));
3856 * Should be called with both balance and volume mutexes held
3858 int btrfs_balance(struct btrfs_balance_control *bctl,
3859 struct btrfs_ioctl_balance_args *bargs)
3861 struct btrfs_fs_info *fs_info = bctl->fs_info;
3862 u64 meta_target, data_target;
3869 if (btrfs_fs_closing(fs_info) ||
3870 atomic_read(&fs_info->balance_pause_req) ||
3871 atomic_read(&fs_info->balance_cancel_req)) {
3876 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3877 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3881 * In case of mixed groups both data and meta should be picked,
3882 * and identical options should be given for both of them.
3884 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3885 if (mixed && (bctl->flags & allowed)) {
3886 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3887 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3888 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3890 "with mixed groups data and metadata balance options must be the same");
3896 num_devices = fs_info->fs_devices->num_devices;
3897 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3898 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3899 BUG_ON(num_devices < 1);
3902 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3903 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3904 if (num_devices > 1)
3905 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3906 if (num_devices > 2)
3907 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3908 if (num_devices > 3)
3909 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3910 BTRFS_BLOCK_GROUP_RAID6);
3911 if (validate_convert_profile(&bctl->data, allowed)) {
3913 "unable to start balance with target data profile %llu",
3918 if (validate_convert_profile(&bctl->meta, allowed)) {
3920 "unable to start balance with target metadata profile %llu",
3925 if (validate_convert_profile(&bctl->sys, allowed)) {
3927 "unable to start balance with target system profile %llu",
3933 /* allow to reduce meta or sys integrity only if force set */
3934 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3935 BTRFS_BLOCK_GROUP_RAID10 |
3936 BTRFS_BLOCK_GROUP_RAID5 |
3937 BTRFS_BLOCK_GROUP_RAID6;
3939 seq = read_seqbegin(&fs_info->profiles_lock);
3941 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3942 (fs_info->avail_system_alloc_bits & allowed) &&
3943 !(bctl->sys.target & allowed)) ||
3944 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3945 (fs_info->avail_metadata_alloc_bits & allowed) &&
3946 !(bctl->meta.target & allowed))) {
3947 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3949 "force reducing metadata integrity");
3952 "balance will reduce metadata integrity, use force if you want this");
3957 } while (read_seqretry(&fs_info->profiles_lock, seq));
3959 /* if we're not converting, the target field is uninitialized */
3960 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3961 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3962 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3963 bctl->data.target : fs_info->avail_data_alloc_bits;
3964 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3965 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3967 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3968 meta_target, data_target);
3971 ret = insert_balance_item(fs_info, bctl);
3972 if (ret && ret != -EEXIST)
3975 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3976 BUG_ON(ret == -EEXIST);
3977 set_balance_control(bctl);
3979 BUG_ON(ret != -EEXIST);
3980 spin_lock(&fs_info->balance_lock);
3981 update_balance_args(bctl);
3982 spin_unlock(&fs_info->balance_lock);
3985 atomic_inc(&fs_info->balance_running);
3986 mutex_unlock(&fs_info->balance_mutex);
3988 ret = __btrfs_balance(fs_info);
3990 mutex_lock(&fs_info->balance_mutex);
3991 atomic_dec(&fs_info->balance_running);
3994 memset(bargs, 0, sizeof(*bargs));
3995 update_ioctl_balance_args(fs_info, 0, bargs);
3998 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3999 balance_need_close(fs_info)) {
4000 __cancel_balance(fs_info);
4003 wake_up(&fs_info->balance_wait_q);
4007 if (bctl->flags & BTRFS_BALANCE_RESUME)
4008 __cancel_balance(fs_info);
4011 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4016 static int balance_kthread(void *data)
4018 struct btrfs_fs_info *fs_info = data;
4021 mutex_lock(&fs_info->volume_mutex);
4022 mutex_lock(&fs_info->balance_mutex);
4024 if (fs_info->balance_ctl) {
4025 btrfs_info(fs_info, "continuing balance");
4026 ret = btrfs_balance(fs_info->balance_ctl, NULL);
4029 mutex_unlock(&fs_info->balance_mutex);
4030 mutex_unlock(&fs_info->volume_mutex);
4035 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4037 struct task_struct *tsk;
4039 spin_lock(&fs_info->balance_lock);
4040 if (!fs_info->balance_ctl) {
4041 spin_unlock(&fs_info->balance_lock);
4044 spin_unlock(&fs_info->balance_lock);
4046 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4047 btrfs_info(fs_info, "force skipping balance");
4051 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4052 return PTR_ERR_OR_ZERO(tsk);
4055 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4057 struct btrfs_balance_control *bctl;
4058 struct btrfs_balance_item *item;
4059 struct btrfs_disk_balance_args disk_bargs;
4060 struct btrfs_path *path;
4061 struct extent_buffer *leaf;
4062 struct btrfs_key key;
4065 path = btrfs_alloc_path();
4069 key.objectid = BTRFS_BALANCE_OBJECTID;
4070 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4073 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4076 if (ret > 0) { /* ret = -ENOENT; */
4081 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4087 leaf = path->nodes[0];
4088 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4090 bctl->fs_info = fs_info;
4091 bctl->flags = btrfs_balance_flags(leaf, item);
4092 bctl->flags |= BTRFS_BALANCE_RESUME;
4094 btrfs_balance_data(leaf, item, &disk_bargs);
4095 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4096 btrfs_balance_meta(leaf, item, &disk_bargs);
4097 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4098 btrfs_balance_sys(leaf, item, &disk_bargs);
4099 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4101 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4103 mutex_lock(&fs_info->volume_mutex);
4104 mutex_lock(&fs_info->balance_mutex);
4106 set_balance_control(bctl);
4108 mutex_unlock(&fs_info->balance_mutex);
4109 mutex_unlock(&fs_info->volume_mutex);
4111 btrfs_free_path(path);
4115 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4119 mutex_lock(&fs_info->balance_mutex);
4120 if (!fs_info->balance_ctl) {
4121 mutex_unlock(&fs_info->balance_mutex);
4125 if (atomic_read(&fs_info->balance_running)) {
4126 atomic_inc(&fs_info->balance_pause_req);
4127 mutex_unlock(&fs_info->balance_mutex);
4129 wait_event(fs_info->balance_wait_q,
4130 atomic_read(&fs_info->balance_running) == 0);
4132 mutex_lock(&fs_info->balance_mutex);
4133 /* we are good with balance_ctl ripped off from under us */
4134 BUG_ON(atomic_read(&fs_info->balance_running));
4135 atomic_dec(&fs_info->balance_pause_req);
4140 mutex_unlock(&fs_info->balance_mutex);
4144 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4146 if (sb_rdonly(fs_info->sb))
4149 mutex_lock(&fs_info->balance_mutex);
4150 if (!fs_info->balance_ctl) {
4151 mutex_unlock(&fs_info->balance_mutex);
4155 atomic_inc(&fs_info->balance_cancel_req);
4157 * if we are running just wait and return, balance item is
4158 * deleted in btrfs_balance in this case
4160 if (atomic_read(&fs_info->balance_running)) {
4161 mutex_unlock(&fs_info->balance_mutex);
4162 wait_event(fs_info->balance_wait_q,
4163 atomic_read(&fs_info->balance_running) == 0);
4164 mutex_lock(&fs_info->balance_mutex);
4166 /* __cancel_balance needs volume_mutex */
4167 mutex_unlock(&fs_info->balance_mutex);
4168 mutex_lock(&fs_info->volume_mutex);
4169 mutex_lock(&fs_info->balance_mutex);
4171 if (fs_info->balance_ctl)
4172 __cancel_balance(fs_info);
4174 mutex_unlock(&fs_info->volume_mutex);
4177 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4178 atomic_dec(&fs_info->balance_cancel_req);
4179 mutex_unlock(&fs_info->balance_mutex);
4183 static int btrfs_uuid_scan_kthread(void *data)
4185 struct btrfs_fs_info *fs_info = data;
4186 struct btrfs_root *root = fs_info->tree_root;
4187 struct btrfs_key key;
4188 struct btrfs_path *path = NULL;
4190 struct extent_buffer *eb;
4192 struct btrfs_root_item root_item;
4194 struct btrfs_trans_handle *trans = NULL;
4196 path = btrfs_alloc_path();
4203 key.type = BTRFS_ROOT_ITEM_KEY;
4207 ret = btrfs_search_forward(root, &key, path, 0);
4214 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4215 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4216 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4217 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4220 eb = path->nodes[0];
4221 slot = path->slots[0];
4222 item_size = btrfs_item_size_nr(eb, slot);
4223 if (item_size < sizeof(root_item))
4226 read_extent_buffer(eb, &root_item,
4227 btrfs_item_ptr_offset(eb, slot),
4228 (int)sizeof(root_item));
4229 if (btrfs_root_refs(&root_item) == 0)
4232 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4233 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4237 btrfs_release_path(path);
4239 * 1 - subvol uuid item
4240 * 1 - received_subvol uuid item
4242 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4243 if (IS_ERR(trans)) {
4244 ret = PTR_ERR(trans);
4252 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4253 ret = btrfs_uuid_tree_add(trans, fs_info,
4255 BTRFS_UUID_KEY_SUBVOL,
4258 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4264 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4265 ret = btrfs_uuid_tree_add(trans, fs_info,
4266 root_item.received_uuid,
4267 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4270 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4278 ret = btrfs_end_transaction(trans);
4284 btrfs_release_path(path);
4285 if (key.offset < (u64)-1) {
4287 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4289 key.type = BTRFS_ROOT_ITEM_KEY;
4290 } else if (key.objectid < (u64)-1) {
4292 key.type = BTRFS_ROOT_ITEM_KEY;
4301 btrfs_free_path(path);
4302 if (trans && !IS_ERR(trans))
4303 btrfs_end_transaction(trans);
4305 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4307 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4308 up(&fs_info->uuid_tree_rescan_sem);
4313 * Callback for btrfs_uuid_tree_iterate().
4315 * 0 check succeeded, the entry is not outdated.
4316 * < 0 if an error occurred.
4317 * > 0 if the check failed, which means the caller shall remove the entry.
4319 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4320 u8 *uuid, u8 type, u64 subid)
4322 struct btrfs_key key;
4324 struct btrfs_root *subvol_root;
4326 if (type != BTRFS_UUID_KEY_SUBVOL &&
4327 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4330 key.objectid = subid;
4331 key.type = BTRFS_ROOT_ITEM_KEY;
4332 key.offset = (u64)-1;
4333 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4334 if (IS_ERR(subvol_root)) {
4335 ret = PTR_ERR(subvol_root);
4342 case BTRFS_UUID_KEY_SUBVOL:
4343 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4346 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4347 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4357 static int btrfs_uuid_rescan_kthread(void *data)
4359 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4363 * 1st step is to iterate through the existing UUID tree and
4364 * to delete all entries that contain outdated data.
4365 * 2nd step is to add all missing entries to the UUID tree.
4367 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4369 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4370 up(&fs_info->uuid_tree_rescan_sem);
4373 return btrfs_uuid_scan_kthread(data);
4376 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4378 struct btrfs_trans_handle *trans;
4379 struct btrfs_root *tree_root = fs_info->tree_root;
4380 struct btrfs_root *uuid_root;
4381 struct task_struct *task;
4388 trans = btrfs_start_transaction(tree_root, 2);
4390 return PTR_ERR(trans);
4392 uuid_root = btrfs_create_tree(trans, fs_info,
4393 BTRFS_UUID_TREE_OBJECTID);
4394 if (IS_ERR(uuid_root)) {
4395 ret = PTR_ERR(uuid_root);
4396 btrfs_abort_transaction(trans, ret);
4397 btrfs_end_transaction(trans);
4401 fs_info->uuid_root = uuid_root;
4403 ret = btrfs_commit_transaction(trans);
4407 down(&fs_info->uuid_tree_rescan_sem);
4408 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4410 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4411 btrfs_warn(fs_info, "failed to start uuid_scan task");
4412 up(&fs_info->uuid_tree_rescan_sem);
4413 return PTR_ERR(task);
4419 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4421 struct task_struct *task;
4423 down(&fs_info->uuid_tree_rescan_sem);
4424 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4426 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4427 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4428 up(&fs_info->uuid_tree_rescan_sem);
4429 return PTR_ERR(task);
4436 * shrinking a device means finding all of the device extents past
4437 * the new size, and then following the back refs to the chunks.
4438 * The chunk relocation code actually frees the device extent
4440 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4442 struct btrfs_fs_info *fs_info = device->fs_info;
4443 struct btrfs_root *root = fs_info->dev_root;
4444 struct btrfs_trans_handle *trans;
4445 struct btrfs_dev_extent *dev_extent = NULL;
4446 struct btrfs_path *path;
4452 bool retried = false;
4453 bool checked_pending_chunks = false;
4454 struct extent_buffer *l;
4455 struct btrfs_key key;
4456 struct btrfs_super_block *super_copy = fs_info->super_copy;
4457 u64 old_total = btrfs_super_total_bytes(super_copy);
4458 u64 old_size = btrfs_device_get_total_bytes(device);
4461 new_size = round_down(new_size, fs_info->sectorsize);
4462 diff = round_down(old_size - new_size, fs_info->sectorsize);
4464 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4467 path = btrfs_alloc_path();
4471 path->reada = READA_FORWARD;
4473 mutex_lock(&fs_info->chunk_mutex);
4475 btrfs_device_set_total_bytes(device, new_size);
4476 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4477 device->fs_devices->total_rw_bytes -= diff;
4478 atomic64_sub(diff, &fs_info->free_chunk_space);
4480 mutex_unlock(&fs_info->chunk_mutex);
4483 key.objectid = device->devid;
4484 key.offset = (u64)-1;
4485 key.type = BTRFS_DEV_EXTENT_KEY;
4488 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4489 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4491 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4495 ret = btrfs_previous_item(root, path, 0, key.type);
4497 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4502 btrfs_release_path(path);
4507 slot = path->slots[0];
4508 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4510 if (key.objectid != device->devid) {
4511 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4512 btrfs_release_path(path);
4516 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4517 length = btrfs_dev_extent_length(l, dev_extent);
4519 if (key.offset + length <= new_size) {
4520 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4521 btrfs_release_path(path);
4525 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4526 btrfs_release_path(path);
4529 * We may be relocating the only data chunk we have,
4530 * which could potentially end up with losing data's
4531 * raid profile, so lets allocate an empty one in
4534 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4536 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4540 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4541 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4542 if (ret && ret != -ENOSPC)
4546 } while (key.offset-- > 0);
4548 if (failed && !retried) {
4552 } else if (failed && retried) {
4557 /* Shrinking succeeded, else we would be at "done". */
4558 trans = btrfs_start_transaction(root, 0);
4559 if (IS_ERR(trans)) {
4560 ret = PTR_ERR(trans);
4564 mutex_lock(&fs_info->chunk_mutex);
4567 * We checked in the above loop all device extents that were already in
4568 * the device tree. However before we have updated the device's
4569 * total_bytes to the new size, we might have had chunk allocations that
4570 * have not complete yet (new block groups attached to transaction
4571 * handles), and therefore their device extents were not yet in the
4572 * device tree and we missed them in the loop above. So if we have any
4573 * pending chunk using a device extent that overlaps the device range
4574 * that we can not use anymore, commit the current transaction and
4575 * repeat the search on the device tree - this way we guarantee we will
4576 * not have chunks using device extents that end beyond 'new_size'.
4578 if (!checked_pending_chunks) {
4579 u64 start = new_size;
4580 u64 len = old_size - new_size;
4582 if (contains_pending_extent(trans->transaction, device,
4584 mutex_unlock(&fs_info->chunk_mutex);
4585 checked_pending_chunks = true;
4588 ret = btrfs_commit_transaction(trans);
4595 btrfs_device_set_disk_total_bytes(device, new_size);
4596 if (list_empty(&device->resized_list))
4597 list_add_tail(&device->resized_list,
4598 &fs_info->fs_devices->resized_devices);
4600 WARN_ON(diff > old_total);
4601 btrfs_set_super_total_bytes(super_copy,
4602 round_down(old_total - diff, fs_info->sectorsize));
4603 mutex_unlock(&fs_info->chunk_mutex);
4605 /* Now btrfs_update_device() will change the on-disk size. */
4606 ret = btrfs_update_device(trans, device);
4607 btrfs_end_transaction(trans);
4609 btrfs_free_path(path);
4611 mutex_lock(&fs_info->chunk_mutex);
4612 btrfs_device_set_total_bytes(device, old_size);
4613 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4614 device->fs_devices->total_rw_bytes += diff;
4615 atomic64_add(diff, &fs_info->free_chunk_space);
4616 mutex_unlock(&fs_info->chunk_mutex);
4621 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4622 struct btrfs_key *key,
4623 struct btrfs_chunk *chunk, int item_size)
4625 struct btrfs_super_block *super_copy = fs_info->super_copy;
4626 struct btrfs_disk_key disk_key;
4630 mutex_lock(&fs_info->chunk_mutex);
4631 array_size = btrfs_super_sys_array_size(super_copy);
4632 if (array_size + item_size + sizeof(disk_key)
4633 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4634 mutex_unlock(&fs_info->chunk_mutex);
4638 ptr = super_copy->sys_chunk_array + array_size;
4639 btrfs_cpu_key_to_disk(&disk_key, key);
4640 memcpy(ptr, &disk_key, sizeof(disk_key));
4641 ptr += sizeof(disk_key);
4642 memcpy(ptr, chunk, item_size);
4643 item_size += sizeof(disk_key);
4644 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4645 mutex_unlock(&fs_info->chunk_mutex);
4651 * sort the devices in descending order by max_avail, total_avail
4653 static int btrfs_cmp_device_info(const void *a, const void *b)
4655 const struct btrfs_device_info *di_a = a;
4656 const struct btrfs_device_info *di_b = b;
4658 if (di_a->max_avail > di_b->max_avail)
4660 if (di_a->max_avail < di_b->max_avail)
4662 if (di_a->total_avail > di_b->total_avail)
4664 if (di_a->total_avail < di_b->total_avail)
4669 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4671 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4674 btrfs_set_fs_incompat(info, RAID56);
4677 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4678 - sizeof(struct btrfs_chunk)) \
4679 / sizeof(struct btrfs_stripe) + 1)
4681 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4682 - 2 * sizeof(struct btrfs_disk_key) \
4683 - 2 * sizeof(struct btrfs_chunk)) \
4684 / sizeof(struct btrfs_stripe) + 1)
4686 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4687 u64 start, u64 type)
4689 struct btrfs_fs_info *info = trans->fs_info;
4690 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4691 struct btrfs_device *device;
4692 struct map_lookup *map = NULL;
4693 struct extent_map_tree *em_tree;
4694 struct extent_map *em;
4695 struct btrfs_device_info *devices_info = NULL;
4697 int num_stripes; /* total number of stripes to allocate */
4698 int data_stripes; /* number of stripes that count for
4700 int sub_stripes; /* sub_stripes info for map */
4701 int dev_stripes; /* stripes per dev */
4702 int devs_max; /* max devs to use */
4703 int devs_min; /* min devs needed */
4704 int devs_increment; /* ndevs has to be a multiple of this */
4705 int ncopies; /* how many copies to data has */
4707 u64 max_stripe_size;
4716 BUG_ON(!alloc_profile_is_valid(type, 0));
4718 if (list_empty(&fs_devices->alloc_list)) {
4719 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4720 btrfs_debug(info, "%s: no writable device", __func__);
4724 index = btrfs_bg_flags_to_raid_index(type);
4726 sub_stripes = btrfs_raid_array[index].sub_stripes;
4727 dev_stripes = btrfs_raid_array[index].dev_stripes;
4728 devs_max = btrfs_raid_array[index].devs_max;
4729 devs_min = btrfs_raid_array[index].devs_min;
4730 devs_increment = btrfs_raid_array[index].devs_increment;
4731 ncopies = btrfs_raid_array[index].ncopies;
4733 if (type & BTRFS_BLOCK_GROUP_DATA) {
4734 max_stripe_size = SZ_1G;
4735 max_chunk_size = 10 * max_stripe_size;
4737 devs_max = BTRFS_MAX_DEVS(info);
4738 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4739 /* for larger filesystems, use larger metadata chunks */
4740 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4741 max_stripe_size = SZ_1G;
4743 max_stripe_size = SZ_256M;
4744 max_chunk_size = max_stripe_size;
4746 devs_max = BTRFS_MAX_DEVS(info);
4747 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4748 max_stripe_size = SZ_32M;
4749 max_chunk_size = 2 * max_stripe_size;
4751 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4753 btrfs_err(info, "invalid chunk type 0x%llx requested",
4758 /* we don't want a chunk larger than 10% of writeable space */
4759 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4762 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4768 * in the first pass through the devices list, we gather information
4769 * about the available holes on each device.
4772 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4776 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4778 "BTRFS: read-only device in alloc_list\n");
4782 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4783 &device->dev_state) ||
4784 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4787 if (device->total_bytes > device->bytes_used)
4788 total_avail = device->total_bytes - device->bytes_used;
4792 /* If there is no space on this device, skip it. */
4793 if (total_avail == 0)
4796 ret = find_free_dev_extent(trans, device,
4797 max_stripe_size * dev_stripes,
4798 &dev_offset, &max_avail);
4799 if (ret && ret != -ENOSPC)
4803 max_avail = max_stripe_size * dev_stripes;
4805 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4806 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4808 "%s: devid %llu has no free space, have=%llu want=%u",
4809 __func__, device->devid, max_avail,
4810 BTRFS_STRIPE_LEN * dev_stripes);
4814 if (ndevs == fs_devices->rw_devices) {
4815 WARN(1, "%s: found more than %llu devices\n",
4816 __func__, fs_devices->rw_devices);
4819 devices_info[ndevs].dev_offset = dev_offset;
4820 devices_info[ndevs].max_avail = max_avail;
4821 devices_info[ndevs].total_avail = total_avail;
4822 devices_info[ndevs].dev = device;
4827 * now sort the devices by hole size / available space
4829 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4830 btrfs_cmp_device_info, NULL);
4832 /* round down to number of usable stripes */
4833 ndevs = round_down(ndevs, devs_increment);
4835 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4837 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4839 "%s: not enough devices with free space: have=%d minimum required=%d",
4840 __func__, ndevs, min(devs_min,
4841 devs_increment * sub_stripes));
4846 ndevs = min(ndevs, devs_max);
4849 * The primary goal is to maximize the number of stripes, so use as
4850 * many devices as possible, even if the stripes are not maximum sized.
4852 * The DUP profile stores more than one stripe per device, the
4853 * max_avail is the total size so we have to adjust.
4855 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4856 num_stripes = ndevs * dev_stripes;
4859 * this will have to be fixed for RAID1 and RAID10 over
4862 data_stripes = num_stripes / ncopies;
4864 if (type & BTRFS_BLOCK_GROUP_RAID5)
4865 data_stripes = num_stripes - 1;
4867 if (type & BTRFS_BLOCK_GROUP_RAID6)
4868 data_stripes = num_stripes - 2;
4871 * Use the number of data stripes to figure out how big this chunk
4872 * is really going to be in terms of logical address space,
4873 * and compare that answer with the max chunk size
4875 if (stripe_size * data_stripes > max_chunk_size) {
4876 stripe_size = div_u64(max_chunk_size, data_stripes);
4878 /* bump the answer up to a 16MB boundary */
4879 stripe_size = round_up(stripe_size, SZ_16M);
4882 * But don't go higher than the limits we found while searching
4885 stripe_size = min(devices_info[ndevs - 1].max_avail,
4889 /* align to BTRFS_STRIPE_LEN */
4890 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4892 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4897 map->num_stripes = num_stripes;
4899 for (i = 0; i < ndevs; ++i) {
4900 for (j = 0; j < dev_stripes; ++j) {
4901 int s = i * dev_stripes + j;
4902 map->stripes[s].dev = devices_info[i].dev;
4903 map->stripes[s].physical = devices_info[i].dev_offset +
4907 map->stripe_len = BTRFS_STRIPE_LEN;
4908 map->io_align = BTRFS_STRIPE_LEN;
4909 map->io_width = BTRFS_STRIPE_LEN;
4911 map->sub_stripes = sub_stripes;
4913 num_bytes = stripe_size * data_stripes;
4915 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4917 em = alloc_extent_map();
4923 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4924 em->map_lookup = map;
4926 em->len = num_bytes;
4927 em->block_start = 0;
4928 em->block_len = em->len;
4929 em->orig_block_len = stripe_size;
4931 em_tree = &info->mapping_tree.map_tree;
4932 write_lock(&em_tree->lock);
4933 ret = add_extent_mapping(em_tree, em, 0);
4935 write_unlock(&em_tree->lock);
4936 free_extent_map(em);
4940 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4941 refcount_inc(&em->refs);
4942 write_unlock(&em_tree->lock);
4944 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4946 goto error_del_extent;
4948 for (i = 0; i < map->num_stripes; i++) {
4949 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4950 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4953 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4955 free_extent_map(em);
4956 check_raid56_incompat_flag(info, type);
4958 kfree(devices_info);
4962 write_lock(&em_tree->lock);
4963 remove_extent_mapping(em_tree, em);
4964 write_unlock(&em_tree->lock);
4966 /* One for our allocation */
4967 free_extent_map(em);
4968 /* One for the tree reference */
4969 free_extent_map(em);
4970 /* One for the pending_chunks list reference */
4971 free_extent_map(em);
4973 kfree(devices_info);
4977 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4978 struct btrfs_fs_info *fs_info,
4979 u64 chunk_offset, u64 chunk_size)
4981 struct btrfs_root *extent_root = fs_info->extent_root;
4982 struct btrfs_root *chunk_root = fs_info->chunk_root;
4983 struct btrfs_key key;
4984 struct btrfs_device *device;
4985 struct btrfs_chunk *chunk;
4986 struct btrfs_stripe *stripe;
4987 struct extent_map *em;
4988 struct map_lookup *map;
4995 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4999 map = em->map_lookup;
5000 item_size = btrfs_chunk_item_size(map->num_stripes);
5001 stripe_size = em->orig_block_len;
5003 chunk = kzalloc(item_size, GFP_NOFS);
5010 * Take the device list mutex to prevent races with the final phase of
5011 * a device replace operation that replaces the device object associated
5012 * with the map's stripes, because the device object's id can change
5013 * at any time during that final phase of the device replace operation
5014 * (dev-replace.c:btrfs_dev_replace_finishing()).
5016 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5017 for (i = 0; i < map->num_stripes; i++) {
5018 device = map->stripes[i].dev;
5019 dev_offset = map->stripes[i].physical;
5021 ret = btrfs_update_device(trans, device);
5024 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5025 dev_offset, stripe_size);
5030 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5034 stripe = &chunk->stripe;
5035 for (i = 0; i < map->num_stripes; i++) {
5036 device = map->stripes[i].dev;
5037 dev_offset = map->stripes[i].physical;
5039 btrfs_set_stack_stripe_devid(stripe, device->devid);
5040 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5041 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5044 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5046 btrfs_set_stack_chunk_length(chunk, chunk_size);
5047 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5048 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5049 btrfs_set_stack_chunk_type(chunk, map->type);
5050 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5051 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5052 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5053 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5054 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5056 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5057 key.type = BTRFS_CHUNK_ITEM_KEY;
5058 key.offset = chunk_offset;
5060 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5061 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5063 * TODO: Cleanup of inserted chunk root in case of
5066 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5071 free_extent_map(em);
5076 * Chunk allocation falls into two parts. The first part does works
5077 * that make the new allocated chunk useable, but not do any operation
5078 * that modifies the chunk tree. The second part does the works that
5079 * require modifying the chunk tree. This division is important for the
5080 * bootstrap process of adding storage to a seed btrfs.
5082 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5083 struct btrfs_fs_info *fs_info, u64 type)
5087 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5088 chunk_offset = find_next_chunk(fs_info);
5089 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5092 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5093 struct btrfs_fs_info *fs_info)
5096 u64 sys_chunk_offset;
5100 chunk_offset = find_next_chunk(fs_info);
5101 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5102 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5106 sys_chunk_offset = find_next_chunk(fs_info);
5107 alloc_profile = btrfs_system_alloc_profile(fs_info);
5108 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5112 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5116 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5117 BTRFS_BLOCK_GROUP_RAID10 |
5118 BTRFS_BLOCK_GROUP_RAID5 |
5119 BTRFS_BLOCK_GROUP_DUP)) {
5121 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5130 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5132 struct extent_map *em;
5133 struct map_lookup *map;
5138 em = get_chunk_map(fs_info, chunk_offset, 1);
5142 map = em->map_lookup;
5143 for (i = 0; i < map->num_stripes; i++) {
5144 if (test_bit(BTRFS_DEV_STATE_MISSING,
5145 &map->stripes[i].dev->dev_state)) {
5149 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5150 &map->stripes[i].dev->dev_state)) {
5157 * If the number of missing devices is larger than max errors,
5158 * we can not write the data into that chunk successfully, so
5161 if (miss_ndevs > btrfs_chunk_max_errors(map))
5164 free_extent_map(em);
5168 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5170 extent_map_tree_init(&tree->map_tree);
5173 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5175 struct extent_map *em;
5178 write_lock(&tree->map_tree.lock);
5179 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5181 remove_extent_mapping(&tree->map_tree, em);
5182 write_unlock(&tree->map_tree.lock);
5186 free_extent_map(em);
5187 /* once for the tree */
5188 free_extent_map(em);
5192 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5194 struct extent_map *em;
5195 struct map_lookup *map;
5198 em = get_chunk_map(fs_info, logical, len);
5201 * We could return errors for these cases, but that could get
5202 * ugly and we'd probably do the same thing which is just not do
5203 * anything else and exit, so return 1 so the callers don't try
5204 * to use other copies.
5208 map = em->map_lookup;
5209 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5210 ret = map->num_stripes;
5211 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5212 ret = map->sub_stripes;
5213 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5215 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5217 * There could be two corrupted data stripes, we need
5218 * to loop retry in order to rebuild the correct data.
5220 * Fail a stripe at a time on every retry except the
5221 * stripe under reconstruction.
5223 ret = map->num_stripes;
5226 free_extent_map(em);
5228 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5229 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5230 fs_info->dev_replace.tgtdev)
5232 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5237 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5240 struct extent_map *em;
5241 struct map_lookup *map;
5242 unsigned long len = fs_info->sectorsize;
5244 em = get_chunk_map(fs_info, logical, len);
5246 if (!WARN_ON(IS_ERR(em))) {
5247 map = em->map_lookup;
5248 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5249 len = map->stripe_len * nr_data_stripes(map);
5250 free_extent_map(em);
5255 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5257 struct extent_map *em;
5258 struct map_lookup *map;
5261 em = get_chunk_map(fs_info, logical, len);
5263 if(!WARN_ON(IS_ERR(em))) {
5264 map = em->map_lookup;
5265 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5267 free_extent_map(em);
5272 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5273 struct map_lookup *map, int first, int num,
5274 int optimal, int dev_replace_is_ongoing)
5278 struct btrfs_device *srcdev;
5280 if (dev_replace_is_ongoing &&
5281 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5282 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5283 srcdev = fs_info->dev_replace.srcdev;
5288 * try to avoid the drive that is the source drive for a
5289 * dev-replace procedure, only choose it if no other non-missing
5290 * mirror is available
5292 for (tolerance = 0; tolerance < 2; tolerance++) {
5293 if (map->stripes[optimal].dev->bdev &&
5294 (tolerance || map->stripes[optimal].dev != srcdev))
5296 for (i = first; i < first + num; i++) {
5297 if (map->stripes[i].dev->bdev &&
5298 (tolerance || map->stripes[i].dev != srcdev))
5303 /* we couldn't find one that doesn't fail. Just return something
5304 * and the io error handling code will clean up eventually
5309 static inline int parity_smaller(u64 a, u64 b)
5314 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5315 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5317 struct btrfs_bio_stripe s;
5324 for (i = 0; i < num_stripes - 1; i++) {
5325 if (parity_smaller(bbio->raid_map[i],
5326 bbio->raid_map[i+1])) {
5327 s = bbio->stripes[i];
5328 l = bbio->raid_map[i];
5329 bbio->stripes[i] = bbio->stripes[i+1];
5330 bbio->raid_map[i] = bbio->raid_map[i+1];
5331 bbio->stripes[i+1] = s;
5332 bbio->raid_map[i+1] = l;
5340 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5342 struct btrfs_bio *bbio = kzalloc(
5343 /* the size of the btrfs_bio */
5344 sizeof(struct btrfs_bio) +
5345 /* plus the variable array for the stripes */
5346 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5347 /* plus the variable array for the tgt dev */
5348 sizeof(int) * (real_stripes) +
5350 * plus the raid_map, which includes both the tgt dev
5353 sizeof(u64) * (total_stripes),
5354 GFP_NOFS|__GFP_NOFAIL);
5356 atomic_set(&bbio->error, 0);
5357 refcount_set(&bbio->refs, 1);
5362 void btrfs_get_bbio(struct btrfs_bio *bbio)
5364 WARN_ON(!refcount_read(&bbio->refs));
5365 refcount_inc(&bbio->refs);
5368 void btrfs_put_bbio(struct btrfs_bio *bbio)
5372 if (refcount_dec_and_test(&bbio->refs))
5376 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5378 * Please note that, discard won't be sent to target device of device
5381 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5382 u64 logical, u64 length,
5383 struct btrfs_bio **bbio_ret)
5385 struct extent_map *em;
5386 struct map_lookup *map;
5387 struct btrfs_bio *bbio;
5391 u64 stripe_end_offset;
5398 u32 sub_stripes = 0;
5399 u64 stripes_per_dev = 0;
5400 u32 remaining_stripes = 0;
5401 u32 last_stripe = 0;
5405 /* discard always return a bbio */
5408 em = get_chunk_map(fs_info, logical, length);
5412 map = em->map_lookup;
5413 /* we don't discard raid56 yet */
5414 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5419 offset = logical - em->start;
5420 length = min_t(u64, em->len - offset, length);
5422 stripe_len = map->stripe_len;
5424 * stripe_nr counts the total number of stripes we have to stride
5425 * to get to this block
5427 stripe_nr = div64_u64(offset, stripe_len);
5429 /* stripe_offset is the offset of this block in its stripe */
5430 stripe_offset = offset - stripe_nr * stripe_len;
5432 stripe_nr_end = round_up(offset + length, map->stripe_len);
5433 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5434 stripe_cnt = stripe_nr_end - stripe_nr;
5435 stripe_end_offset = stripe_nr_end * map->stripe_len -
5438 * after this, stripe_nr is the number of stripes on this
5439 * device we have to walk to find the data, and stripe_index is
5440 * the number of our device in the stripe array
5444 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5445 BTRFS_BLOCK_GROUP_RAID10)) {
5446 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5449 sub_stripes = map->sub_stripes;
5451 factor = map->num_stripes / sub_stripes;
5452 num_stripes = min_t(u64, map->num_stripes,
5453 sub_stripes * stripe_cnt);
5454 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5455 stripe_index *= sub_stripes;
5456 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5457 &remaining_stripes);
5458 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5459 last_stripe *= sub_stripes;
5460 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5461 BTRFS_BLOCK_GROUP_DUP)) {
5462 num_stripes = map->num_stripes;
5464 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5468 bbio = alloc_btrfs_bio(num_stripes, 0);
5474 for (i = 0; i < num_stripes; i++) {
5475 bbio->stripes[i].physical =
5476 map->stripes[stripe_index].physical +
5477 stripe_offset + stripe_nr * map->stripe_len;
5478 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5480 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5481 BTRFS_BLOCK_GROUP_RAID10)) {
5482 bbio->stripes[i].length = stripes_per_dev *
5485 if (i / sub_stripes < remaining_stripes)
5486 bbio->stripes[i].length +=
5490 * Special for the first stripe and
5493 * |-------|...|-------|
5497 if (i < sub_stripes)
5498 bbio->stripes[i].length -=
5501 if (stripe_index >= last_stripe &&
5502 stripe_index <= (last_stripe +
5504 bbio->stripes[i].length -=
5507 if (i == sub_stripes - 1)
5510 bbio->stripes[i].length = length;
5514 if (stripe_index == map->num_stripes) {
5521 bbio->map_type = map->type;
5522 bbio->num_stripes = num_stripes;
5524 free_extent_map(em);
5529 * In dev-replace case, for repair case (that's the only case where the mirror
5530 * is selected explicitly when calling btrfs_map_block), blocks left of the
5531 * left cursor can also be read from the target drive.
5533 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5535 * For READ, it also needs to be supported using the same mirror number.
5537 * If the requested block is not left of the left cursor, EIO is returned. This
5538 * can happen because btrfs_num_copies() returns one more in the dev-replace
5541 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5542 u64 logical, u64 length,
5543 u64 srcdev_devid, int *mirror_num,
5546 struct btrfs_bio *bbio = NULL;
5548 int index_srcdev = 0;
5550 u64 physical_of_found = 0;
5554 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5555 logical, &length, &bbio, 0, 0);
5557 ASSERT(bbio == NULL);
5561 num_stripes = bbio->num_stripes;
5562 if (*mirror_num > num_stripes) {
5564 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5565 * that means that the requested area is not left of the left
5568 btrfs_put_bbio(bbio);
5573 * process the rest of the function using the mirror_num of the source
5574 * drive. Therefore look it up first. At the end, patch the device
5575 * pointer to the one of the target drive.
5577 for (i = 0; i < num_stripes; i++) {
5578 if (bbio->stripes[i].dev->devid != srcdev_devid)
5582 * In case of DUP, in order to keep it simple, only add the
5583 * mirror with the lowest physical address
5586 physical_of_found <= bbio->stripes[i].physical)
5591 physical_of_found = bbio->stripes[i].physical;
5594 btrfs_put_bbio(bbio);
5600 *mirror_num = index_srcdev + 1;
5601 *physical = physical_of_found;
5605 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5606 struct btrfs_bio **bbio_ret,
5607 struct btrfs_dev_replace *dev_replace,
5608 int *num_stripes_ret, int *max_errors_ret)
5610 struct btrfs_bio *bbio = *bbio_ret;
5611 u64 srcdev_devid = dev_replace->srcdev->devid;
5612 int tgtdev_indexes = 0;
5613 int num_stripes = *num_stripes_ret;
5614 int max_errors = *max_errors_ret;
5617 if (op == BTRFS_MAP_WRITE) {
5618 int index_where_to_add;
5621 * duplicate the write operations while the dev replace
5622 * procedure is running. Since the copying of the old disk to
5623 * the new disk takes place at run time while the filesystem is
5624 * mounted writable, the regular write operations to the old
5625 * disk have to be duplicated to go to the new disk as well.
5627 * Note that device->missing is handled by the caller, and that
5628 * the write to the old disk is already set up in the stripes
5631 index_where_to_add = num_stripes;
5632 for (i = 0; i < num_stripes; i++) {
5633 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5634 /* write to new disk, too */
5635 struct btrfs_bio_stripe *new =
5636 bbio->stripes + index_where_to_add;
5637 struct btrfs_bio_stripe *old =
5640 new->physical = old->physical;
5641 new->length = old->length;
5642 new->dev = dev_replace->tgtdev;
5643 bbio->tgtdev_map[i] = index_where_to_add;
5644 index_where_to_add++;
5649 num_stripes = index_where_to_add;
5650 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5651 int index_srcdev = 0;
5653 u64 physical_of_found = 0;
5656 * During the dev-replace procedure, the target drive can also
5657 * be used to read data in case it is needed to repair a corrupt
5658 * block elsewhere. This is possible if the requested area is
5659 * left of the left cursor. In this area, the target drive is a
5660 * full copy of the source drive.
5662 for (i = 0; i < num_stripes; i++) {
5663 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5665 * In case of DUP, in order to keep it simple,
5666 * only add the mirror with the lowest physical
5670 physical_of_found <=
5671 bbio->stripes[i].physical)
5675 physical_of_found = bbio->stripes[i].physical;
5679 struct btrfs_bio_stripe *tgtdev_stripe =
5680 bbio->stripes + num_stripes;
5682 tgtdev_stripe->physical = physical_of_found;
5683 tgtdev_stripe->length =
5684 bbio->stripes[index_srcdev].length;
5685 tgtdev_stripe->dev = dev_replace->tgtdev;
5686 bbio->tgtdev_map[index_srcdev] = num_stripes;
5693 *num_stripes_ret = num_stripes;
5694 *max_errors_ret = max_errors;
5695 bbio->num_tgtdevs = tgtdev_indexes;
5699 static bool need_full_stripe(enum btrfs_map_op op)
5701 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5704 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5705 enum btrfs_map_op op,
5706 u64 logical, u64 *length,
5707 struct btrfs_bio **bbio_ret,
5708 int mirror_num, int need_raid_map)
5710 struct extent_map *em;
5711 struct map_lookup *map;
5721 int tgtdev_indexes = 0;
5722 struct btrfs_bio *bbio = NULL;
5723 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5724 int dev_replace_is_ongoing = 0;
5725 int num_alloc_stripes;
5726 int patch_the_first_stripe_for_dev_replace = 0;
5727 u64 physical_to_patch_in_first_stripe = 0;
5728 u64 raid56_full_stripe_start = (u64)-1;
5730 if (op == BTRFS_MAP_DISCARD)
5731 return __btrfs_map_block_for_discard(fs_info, logical,
5734 em = get_chunk_map(fs_info, logical, *length);
5738 map = em->map_lookup;
5739 offset = logical - em->start;
5741 stripe_len = map->stripe_len;
5744 * stripe_nr counts the total number of stripes we have to stride
5745 * to get to this block
5747 stripe_nr = div64_u64(stripe_nr, stripe_len);
5749 stripe_offset = stripe_nr * stripe_len;
5750 if (offset < stripe_offset) {
5752 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5753 stripe_offset, offset, em->start, logical,
5755 free_extent_map(em);
5759 /* stripe_offset is the offset of this block in its stripe*/
5760 stripe_offset = offset - stripe_offset;
5762 /* if we're here for raid56, we need to know the stripe aligned start */
5763 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5764 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5765 raid56_full_stripe_start = offset;
5767 /* allow a write of a full stripe, but make sure we don't
5768 * allow straddling of stripes
5770 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5772 raid56_full_stripe_start *= full_stripe_len;
5775 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5777 /* For writes to RAID[56], allow a full stripeset across all disks.
5778 For other RAID types and for RAID[56] reads, just allow a single
5779 stripe (on a single disk). */
5780 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5781 (op == BTRFS_MAP_WRITE)) {
5782 max_len = stripe_len * nr_data_stripes(map) -
5783 (offset - raid56_full_stripe_start);
5785 /* we limit the length of each bio to what fits in a stripe */
5786 max_len = stripe_len - stripe_offset;
5788 *length = min_t(u64, em->len - offset, max_len);
5790 *length = em->len - offset;
5793 /* This is for when we're called from btrfs_merge_bio_hook() and all
5794 it cares about is the length */
5798 btrfs_dev_replace_lock(dev_replace, 0);
5799 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5800 if (!dev_replace_is_ongoing)
5801 btrfs_dev_replace_unlock(dev_replace, 0);
5803 btrfs_dev_replace_set_lock_blocking(dev_replace);
5805 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5806 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5807 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5808 dev_replace->srcdev->devid,
5810 &physical_to_patch_in_first_stripe);
5814 patch_the_first_stripe_for_dev_replace = 1;
5815 } else if (mirror_num > map->num_stripes) {
5821 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5822 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5824 if (!need_full_stripe(op))
5826 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5827 if (need_full_stripe(op))
5828 num_stripes = map->num_stripes;
5829 else if (mirror_num)
5830 stripe_index = mirror_num - 1;
5832 stripe_index = find_live_mirror(fs_info, map, 0,
5834 current->pid % map->num_stripes,
5835 dev_replace_is_ongoing);
5836 mirror_num = stripe_index + 1;
5839 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5840 if (need_full_stripe(op)) {
5841 num_stripes = map->num_stripes;
5842 } else if (mirror_num) {
5843 stripe_index = mirror_num - 1;
5848 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5849 u32 factor = map->num_stripes / map->sub_stripes;
5851 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5852 stripe_index *= map->sub_stripes;
5854 if (need_full_stripe(op))
5855 num_stripes = map->sub_stripes;
5856 else if (mirror_num)
5857 stripe_index += mirror_num - 1;
5859 int old_stripe_index = stripe_index;
5860 stripe_index = find_live_mirror(fs_info, map,
5862 map->sub_stripes, stripe_index +
5863 current->pid % map->sub_stripes,
5864 dev_replace_is_ongoing);
5865 mirror_num = stripe_index - old_stripe_index + 1;
5868 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5869 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5870 /* push stripe_nr back to the start of the full stripe */
5871 stripe_nr = div64_u64(raid56_full_stripe_start,
5872 stripe_len * nr_data_stripes(map));
5874 /* RAID[56] write or recovery. Return all stripes */
5875 num_stripes = map->num_stripes;
5876 max_errors = nr_parity_stripes(map);
5878 *length = map->stripe_len;
5883 * Mirror #0 or #1 means the original data block.
5884 * Mirror #2 is RAID5 parity block.
5885 * Mirror #3 is RAID6 Q block.
5887 stripe_nr = div_u64_rem(stripe_nr,
5888 nr_data_stripes(map), &stripe_index);
5890 stripe_index = nr_data_stripes(map) +
5893 /* We distribute the parity blocks across stripes */
5894 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5896 if (!need_full_stripe(op) && mirror_num <= 1)
5901 * after this, stripe_nr is the number of stripes on this
5902 * device we have to walk to find the data, and stripe_index is
5903 * the number of our device in the stripe array
5905 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5907 mirror_num = stripe_index + 1;
5909 if (stripe_index >= map->num_stripes) {
5911 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5912 stripe_index, map->num_stripes);
5917 num_alloc_stripes = num_stripes;
5918 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5919 if (op == BTRFS_MAP_WRITE)
5920 num_alloc_stripes <<= 1;
5921 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5922 num_alloc_stripes++;
5923 tgtdev_indexes = num_stripes;
5926 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5931 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5932 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5934 /* build raid_map */
5935 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5936 (need_full_stripe(op) || mirror_num > 1)) {
5940 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5941 sizeof(struct btrfs_bio_stripe) *
5943 sizeof(int) * tgtdev_indexes);
5945 /* Work out the disk rotation on this stripe-set */
5946 div_u64_rem(stripe_nr, num_stripes, &rot);
5948 /* Fill in the logical address of each stripe */
5949 tmp = stripe_nr * nr_data_stripes(map);
5950 for (i = 0; i < nr_data_stripes(map); i++)
5951 bbio->raid_map[(i+rot) % num_stripes] =
5952 em->start + (tmp + i) * map->stripe_len;
5954 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5955 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5956 bbio->raid_map[(i+rot+1) % num_stripes] =
5961 for (i = 0; i < num_stripes; i++) {
5962 bbio->stripes[i].physical =
5963 map->stripes[stripe_index].physical +
5965 stripe_nr * map->stripe_len;
5966 bbio->stripes[i].dev =
5967 map->stripes[stripe_index].dev;
5971 if (need_full_stripe(op))
5972 max_errors = btrfs_chunk_max_errors(map);
5975 sort_parity_stripes(bbio, num_stripes);
5977 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5978 need_full_stripe(op)) {
5979 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5984 bbio->map_type = map->type;
5985 bbio->num_stripes = num_stripes;
5986 bbio->max_errors = max_errors;
5987 bbio->mirror_num = mirror_num;
5990 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5991 * mirror_num == num_stripes + 1 && dev_replace target drive is
5992 * available as a mirror
5994 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5995 WARN_ON(num_stripes > 1);
5996 bbio->stripes[0].dev = dev_replace->tgtdev;
5997 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5998 bbio->mirror_num = map->num_stripes + 1;
6001 if (dev_replace_is_ongoing) {
6002 btrfs_dev_replace_clear_lock_blocking(dev_replace);
6003 btrfs_dev_replace_unlock(dev_replace, 0);
6005 free_extent_map(em);
6009 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6010 u64 logical, u64 *length,
6011 struct btrfs_bio **bbio_ret, int mirror_num)
6013 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6017 /* For Scrub/replace */
6018 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6019 u64 logical, u64 *length,
6020 struct btrfs_bio **bbio_ret)
6022 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6025 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
6026 u64 chunk_start, u64 physical, u64 devid,
6027 u64 **logical, int *naddrs, int *stripe_len)
6029 struct extent_map *em;
6030 struct map_lookup *map;
6038 em = get_chunk_map(fs_info, chunk_start, 1);
6042 map = em->map_lookup;
6044 rmap_len = map->stripe_len;
6046 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6047 length = div_u64(length, map->num_stripes / map->sub_stripes);
6048 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6049 length = div_u64(length, map->num_stripes);
6050 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6051 length = div_u64(length, nr_data_stripes(map));
6052 rmap_len = map->stripe_len * nr_data_stripes(map);
6055 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6056 BUG_ON(!buf); /* -ENOMEM */
6058 for (i = 0; i < map->num_stripes; i++) {
6059 if (devid && map->stripes[i].dev->devid != devid)
6061 if (map->stripes[i].physical > physical ||
6062 map->stripes[i].physical + length <= physical)
6065 stripe_nr = physical - map->stripes[i].physical;
6066 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6068 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6069 stripe_nr = stripe_nr * map->num_stripes + i;
6070 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6071 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6072 stripe_nr = stripe_nr * map->num_stripes + i;
6073 } /* else if RAID[56], multiply by nr_data_stripes().
6074 * Alternatively, just use rmap_len below instead of
6075 * map->stripe_len */
6077 bytenr = chunk_start + stripe_nr * rmap_len;
6078 WARN_ON(nr >= map->num_stripes);
6079 for (j = 0; j < nr; j++) {
6080 if (buf[j] == bytenr)
6084 WARN_ON(nr >= map->num_stripes);
6091 *stripe_len = rmap_len;
6093 free_extent_map(em);
6097 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6099 bio->bi_private = bbio->private;
6100 bio->bi_end_io = bbio->end_io;
6103 btrfs_put_bbio(bbio);
6106 static void btrfs_end_bio(struct bio *bio)
6108 struct btrfs_bio *bbio = bio->bi_private;
6109 int is_orig_bio = 0;
6111 if (bio->bi_status) {
6112 atomic_inc(&bbio->error);
6113 if (bio->bi_status == BLK_STS_IOERR ||
6114 bio->bi_status == BLK_STS_TARGET) {
6115 unsigned int stripe_index =
6116 btrfs_io_bio(bio)->stripe_index;
6117 struct btrfs_device *dev;
6119 BUG_ON(stripe_index >= bbio->num_stripes);
6120 dev = bbio->stripes[stripe_index].dev;
6122 if (bio_op(bio) == REQ_OP_WRITE)
6123 btrfs_dev_stat_inc_and_print(dev,
6124 BTRFS_DEV_STAT_WRITE_ERRS);
6126 btrfs_dev_stat_inc_and_print(dev,
6127 BTRFS_DEV_STAT_READ_ERRS);
6128 if (bio->bi_opf & REQ_PREFLUSH)
6129 btrfs_dev_stat_inc_and_print(dev,
6130 BTRFS_DEV_STAT_FLUSH_ERRS);
6135 if (bio == bbio->orig_bio)
6138 btrfs_bio_counter_dec(bbio->fs_info);
6140 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6143 bio = bbio->orig_bio;
6146 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6147 /* only send an error to the higher layers if it is
6148 * beyond the tolerance of the btrfs bio
6150 if (atomic_read(&bbio->error) > bbio->max_errors) {
6151 bio->bi_status = BLK_STS_IOERR;
6154 * this bio is actually up to date, we didn't
6155 * go over the max number of errors
6157 bio->bi_status = BLK_STS_OK;
6160 btrfs_end_bbio(bbio, bio);
6161 } else if (!is_orig_bio) {
6167 * see run_scheduled_bios for a description of why bios are collected for
6170 * This will add one bio to the pending list for a device and make sure
6171 * the work struct is scheduled.
6173 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6176 struct btrfs_fs_info *fs_info = device->fs_info;
6177 int should_queue = 1;
6178 struct btrfs_pending_bios *pending_bios;
6180 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6186 /* don't bother with additional async steps for reads, right now */
6187 if (bio_op(bio) == REQ_OP_READ) {
6188 btrfsic_submit_bio(bio);
6192 WARN_ON(bio->bi_next);
6193 bio->bi_next = NULL;
6195 spin_lock(&device->io_lock);
6196 if (op_is_sync(bio->bi_opf))
6197 pending_bios = &device->pending_sync_bios;
6199 pending_bios = &device->pending_bios;
6201 if (pending_bios->tail)
6202 pending_bios->tail->bi_next = bio;
6204 pending_bios->tail = bio;
6205 if (!pending_bios->head)
6206 pending_bios->head = bio;
6207 if (device->running_pending)
6210 spin_unlock(&device->io_lock);
6213 btrfs_queue_work(fs_info->submit_workers, &device->work);
6216 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6217 u64 physical, int dev_nr, int async)
6219 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6220 struct btrfs_fs_info *fs_info = bbio->fs_info;
6222 bio->bi_private = bbio;
6223 btrfs_io_bio(bio)->stripe_index = dev_nr;
6224 bio->bi_end_io = btrfs_end_bio;
6225 bio->bi_iter.bi_sector = physical >> 9;
6228 struct rcu_string *name;
6231 name = rcu_dereference(dev->name);
6232 btrfs_debug(fs_info,
6233 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6234 bio_op(bio), bio->bi_opf,
6235 (u64)bio->bi_iter.bi_sector,
6236 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6237 bio->bi_iter.bi_size);
6241 bio_set_dev(bio, dev->bdev);
6243 btrfs_bio_counter_inc_noblocked(fs_info);
6246 btrfs_schedule_bio(dev, bio);
6248 btrfsic_submit_bio(bio);
6251 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6253 atomic_inc(&bbio->error);
6254 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6255 /* Should be the original bio. */
6256 WARN_ON(bio != bbio->orig_bio);
6258 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6259 bio->bi_iter.bi_sector = logical >> 9;
6260 if (atomic_read(&bbio->error) > bbio->max_errors)
6261 bio->bi_status = BLK_STS_IOERR;
6263 bio->bi_status = BLK_STS_OK;
6264 btrfs_end_bbio(bbio, bio);
6268 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6269 int mirror_num, int async_submit)
6271 struct btrfs_device *dev;
6272 struct bio *first_bio = bio;
6273 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6279 struct btrfs_bio *bbio = NULL;
6281 length = bio->bi_iter.bi_size;
6282 map_length = length;
6284 btrfs_bio_counter_inc_blocked(fs_info);
6285 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6286 &map_length, &bbio, mirror_num, 1);
6288 btrfs_bio_counter_dec(fs_info);
6289 return errno_to_blk_status(ret);
6292 total_devs = bbio->num_stripes;
6293 bbio->orig_bio = first_bio;
6294 bbio->private = first_bio->bi_private;
6295 bbio->end_io = first_bio->bi_end_io;
6296 bbio->fs_info = fs_info;
6297 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6299 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6300 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6301 /* In this case, map_length has been set to the length of
6302 a single stripe; not the whole write */
6303 if (bio_op(bio) == REQ_OP_WRITE) {
6304 ret = raid56_parity_write(fs_info, bio, bbio,
6307 ret = raid56_parity_recover(fs_info, bio, bbio,
6308 map_length, mirror_num, 1);
6311 btrfs_bio_counter_dec(fs_info);
6312 return errno_to_blk_status(ret);
6315 if (map_length < length) {
6317 "mapping failed logical %llu bio len %llu len %llu",
6318 logical, length, map_length);
6322 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6323 dev = bbio->stripes[dev_nr].dev;
6324 if (!dev || !dev->bdev ||
6325 (bio_op(first_bio) == REQ_OP_WRITE &&
6326 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6327 bbio_error(bbio, first_bio, logical);
6331 if (dev_nr < total_devs - 1)
6332 bio = btrfs_bio_clone(first_bio);
6336 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6337 dev_nr, async_submit);
6339 btrfs_bio_counter_dec(fs_info);
6343 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6346 struct btrfs_device *device;
6347 struct btrfs_fs_devices *cur_devices;
6349 cur_devices = fs_info->fs_devices;
6350 while (cur_devices) {
6352 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6353 device = find_device(cur_devices, devid, uuid);
6357 cur_devices = cur_devices->seed;
6362 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6363 u64 devid, u8 *dev_uuid)
6365 struct btrfs_device *device;
6367 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6371 list_add(&device->dev_list, &fs_devices->devices);
6372 device->fs_devices = fs_devices;
6373 fs_devices->num_devices++;
6375 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6376 fs_devices->missing_devices++;
6382 * btrfs_alloc_device - allocate struct btrfs_device
6383 * @fs_info: used only for generating a new devid, can be NULL if
6384 * devid is provided (i.e. @devid != NULL).
6385 * @devid: a pointer to devid for this device. If NULL a new devid
6387 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6390 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6391 * on error. Returned struct is not linked onto any lists and must be
6392 * destroyed with free_device.
6394 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6398 struct btrfs_device *dev;
6401 if (WARN_ON(!devid && !fs_info))
6402 return ERR_PTR(-EINVAL);
6404 dev = __alloc_device();
6413 ret = find_next_devid(fs_info, &tmp);
6416 return ERR_PTR(ret);
6422 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6424 generate_random_uuid(dev->uuid);
6426 btrfs_init_work(&dev->work, btrfs_submit_helper,
6427 pending_bios_fn, NULL, NULL);
6432 /* Return -EIO if any error, otherwise return 0. */
6433 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6434 struct extent_buffer *leaf,
6435 struct btrfs_chunk *chunk, u64 logical)
6443 length = btrfs_chunk_length(leaf, chunk);
6444 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6445 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6446 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6447 type = btrfs_chunk_type(leaf, chunk);
6450 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6454 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6455 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6458 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6459 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6460 btrfs_chunk_sector_size(leaf, chunk));
6463 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6464 btrfs_err(fs_info, "invalid chunk length %llu", length);
6467 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6468 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6472 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6474 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6475 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6476 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6477 btrfs_chunk_type(leaf, chunk));
6480 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6481 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6482 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6483 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6484 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6485 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6486 num_stripes != 1)) {
6488 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6489 num_stripes, sub_stripes,
6490 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6497 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6498 u64 devid, u8 *uuid, bool error)
6501 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6504 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6508 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6509 struct extent_buffer *leaf,
6510 struct btrfs_chunk *chunk)
6512 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6513 struct map_lookup *map;
6514 struct extent_map *em;
6518 u8 uuid[BTRFS_UUID_SIZE];
6523 logical = key->offset;
6524 length = btrfs_chunk_length(leaf, chunk);
6525 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6527 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6531 read_lock(&map_tree->map_tree.lock);
6532 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6533 read_unlock(&map_tree->map_tree.lock);
6535 /* already mapped? */
6536 if (em && em->start <= logical && em->start + em->len > logical) {
6537 free_extent_map(em);
6540 free_extent_map(em);
6543 em = alloc_extent_map();
6546 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6548 free_extent_map(em);
6552 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6553 em->map_lookup = map;
6554 em->start = logical;
6557 em->block_start = 0;
6558 em->block_len = em->len;
6560 map->num_stripes = num_stripes;
6561 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6562 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6563 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6564 map->type = btrfs_chunk_type(leaf, chunk);
6565 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6566 for (i = 0; i < num_stripes; i++) {
6567 map->stripes[i].physical =
6568 btrfs_stripe_offset_nr(leaf, chunk, i);
6569 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6570 read_extent_buffer(leaf, uuid, (unsigned long)
6571 btrfs_stripe_dev_uuid_nr(chunk, i),
6573 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6575 if (!map->stripes[i].dev &&
6576 !btrfs_test_opt(fs_info, DEGRADED)) {
6577 free_extent_map(em);
6578 btrfs_report_missing_device(fs_info, devid, uuid, true);
6581 if (!map->stripes[i].dev) {
6582 map->stripes[i].dev =
6583 add_missing_dev(fs_info->fs_devices, devid,
6585 if (IS_ERR(map->stripes[i].dev)) {
6586 free_extent_map(em);
6588 "failed to init missing dev %llu: %ld",
6589 devid, PTR_ERR(map->stripes[i].dev));
6590 return PTR_ERR(map->stripes[i].dev);
6592 btrfs_report_missing_device(fs_info, devid, uuid, false);
6594 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6595 &(map->stripes[i].dev->dev_state));
6599 write_lock(&map_tree->map_tree.lock);
6600 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6601 write_unlock(&map_tree->map_tree.lock);
6602 BUG_ON(ret); /* Tree corruption */
6603 free_extent_map(em);
6608 static void fill_device_from_item(struct extent_buffer *leaf,
6609 struct btrfs_dev_item *dev_item,
6610 struct btrfs_device *device)
6614 device->devid = btrfs_device_id(leaf, dev_item);
6615 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6616 device->total_bytes = device->disk_total_bytes;
6617 device->commit_total_bytes = device->disk_total_bytes;
6618 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6619 device->commit_bytes_used = device->bytes_used;
6620 device->type = btrfs_device_type(leaf, dev_item);
6621 device->io_align = btrfs_device_io_align(leaf, dev_item);
6622 device->io_width = btrfs_device_io_width(leaf, dev_item);
6623 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6624 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6625 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6627 ptr = btrfs_device_uuid(dev_item);
6628 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6631 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6634 struct btrfs_fs_devices *fs_devices;
6637 BUG_ON(!mutex_is_locked(&uuid_mutex));
6640 fs_devices = fs_info->fs_devices->seed;
6641 while (fs_devices) {
6642 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6645 fs_devices = fs_devices->seed;
6648 fs_devices = find_fsid(fsid);
6650 if (!btrfs_test_opt(fs_info, DEGRADED))
6651 return ERR_PTR(-ENOENT);
6653 fs_devices = alloc_fs_devices(fsid);
6654 if (IS_ERR(fs_devices))
6657 fs_devices->seeding = 1;
6658 fs_devices->opened = 1;
6662 fs_devices = clone_fs_devices(fs_devices);
6663 if (IS_ERR(fs_devices))
6666 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6667 fs_info->bdev_holder);
6669 free_fs_devices(fs_devices);
6670 fs_devices = ERR_PTR(ret);
6674 if (!fs_devices->seeding) {
6675 __btrfs_close_devices(fs_devices);
6676 free_fs_devices(fs_devices);
6677 fs_devices = ERR_PTR(-EINVAL);
6681 fs_devices->seed = fs_info->fs_devices->seed;
6682 fs_info->fs_devices->seed = fs_devices;
6687 static int read_one_dev(struct btrfs_fs_info *fs_info,
6688 struct extent_buffer *leaf,
6689 struct btrfs_dev_item *dev_item)
6691 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6692 struct btrfs_device *device;
6695 u8 fs_uuid[BTRFS_FSID_SIZE];
6696 u8 dev_uuid[BTRFS_UUID_SIZE];
6698 devid = btrfs_device_id(leaf, dev_item);
6699 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6701 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6704 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6705 fs_devices = open_seed_devices(fs_info, fs_uuid);
6706 if (IS_ERR(fs_devices))
6707 return PTR_ERR(fs_devices);
6710 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6712 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6713 btrfs_report_missing_device(fs_info, devid,
6718 device = add_missing_dev(fs_devices, devid, dev_uuid);
6719 if (IS_ERR(device)) {
6721 "failed to add missing dev %llu: %ld",
6722 devid, PTR_ERR(device));
6723 return PTR_ERR(device);
6725 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6727 if (!device->bdev) {
6728 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6729 btrfs_report_missing_device(fs_info,
6730 devid, dev_uuid, true);
6733 btrfs_report_missing_device(fs_info, devid,
6737 if (!device->bdev &&
6738 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6740 * this happens when a device that was properly setup
6741 * in the device info lists suddenly goes bad.
6742 * device->bdev is NULL, and so we have to set
6743 * device->missing to one here
6745 device->fs_devices->missing_devices++;
6746 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6749 /* Move the device to its own fs_devices */
6750 if (device->fs_devices != fs_devices) {
6751 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6752 &device->dev_state));
6754 list_move(&device->dev_list, &fs_devices->devices);
6755 device->fs_devices->num_devices--;
6756 fs_devices->num_devices++;
6758 device->fs_devices->missing_devices--;
6759 fs_devices->missing_devices++;
6761 device->fs_devices = fs_devices;
6765 if (device->fs_devices != fs_info->fs_devices) {
6766 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6767 if (device->generation !=
6768 btrfs_device_generation(leaf, dev_item))
6772 fill_device_from_item(leaf, dev_item, device);
6773 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6774 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6775 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6776 device->fs_devices->total_rw_bytes += device->total_bytes;
6777 atomic64_add(device->total_bytes - device->bytes_used,
6778 &fs_info->free_chunk_space);
6784 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6786 struct btrfs_root *root = fs_info->tree_root;
6787 struct btrfs_super_block *super_copy = fs_info->super_copy;
6788 struct extent_buffer *sb;
6789 struct btrfs_disk_key *disk_key;
6790 struct btrfs_chunk *chunk;
6792 unsigned long sb_array_offset;
6799 struct btrfs_key key;
6801 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6803 * This will create extent buffer of nodesize, superblock size is
6804 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6805 * overallocate but we can keep it as-is, only the first page is used.
6807 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6810 set_extent_buffer_uptodate(sb);
6811 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6813 * The sb extent buffer is artificial and just used to read the system array.
6814 * set_extent_buffer_uptodate() call does not properly mark all it's
6815 * pages up-to-date when the page is larger: extent does not cover the
6816 * whole page and consequently check_page_uptodate does not find all
6817 * the page's extents up-to-date (the hole beyond sb),
6818 * write_extent_buffer then triggers a WARN_ON.
6820 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6821 * but sb spans only this function. Add an explicit SetPageUptodate call
6822 * to silence the warning eg. on PowerPC 64.
6824 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6825 SetPageUptodate(sb->pages[0]);
6827 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6828 array_size = btrfs_super_sys_array_size(super_copy);
6830 array_ptr = super_copy->sys_chunk_array;
6831 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6834 while (cur_offset < array_size) {
6835 disk_key = (struct btrfs_disk_key *)array_ptr;
6836 len = sizeof(*disk_key);
6837 if (cur_offset + len > array_size)
6838 goto out_short_read;
6840 btrfs_disk_key_to_cpu(&key, disk_key);
6843 sb_array_offset += len;
6846 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6847 chunk = (struct btrfs_chunk *)sb_array_offset;
6849 * At least one btrfs_chunk with one stripe must be
6850 * present, exact stripe count check comes afterwards
6852 len = btrfs_chunk_item_size(1);
6853 if (cur_offset + len > array_size)
6854 goto out_short_read;
6856 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6859 "invalid number of stripes %u in sys_array at offset %u",
6860 num_stripes, cur_offset);
6865 type = btrfs_chunk_type(sb, chunk);
6866 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6868 "invalid chunk type %llu in sys_array at offset %u",
6874 len = btrfs_chunk_item_size(num_stripes);
6875 if (cur_offset + len > array_size)
6876 goto out_short_read;
6878 ret = read_one_chunk(fs_info, &key, sb, chunk);
6883 "unexpected item type %u in sys_array at offset %u",
6884 (u32)key.type, cur_offset);
6889 sb_array_offset += len;
6892 clear_extent_buffer_uptodate(sb);
6893 free_extent_buffer_stale(sb);
6897 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6899 clear_extent_buffer_uptodate(sb);
6900 free_extent_buffer_stale(sb);
6905 * Check if all chunks in the fs are OK for read-write degraded mount
6907 * If the @failing_dev is specified, it's accounted as missing.
6909 * Return true if all chunks meet the minimal RW mount requirements.
6910 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6912 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6913 struct btrfs_device *failing_dev)
6915 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6916 struct extent_map *em;
6920 read_lock(&map_tree->map_tree.lock);
6921 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6922 read_unlock(&map_tree->map_tree.lock);
6923 /* No chunk at all? Return false anyway */
6929 struct map_lookup *map;
6934 map = em->map_lookup;
6936 btrfs_get_num_tolerated_disk_barrier_failures(
6938 for (i = 0; i < map->num_stripes; i++) {
6939 struct btrfs_device *dev = map->stripes[i].dev;
6941 if (!dev || !dev->bdev ||
6942 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6943 dev->last_flush_error)
6945 else if (failing_dev && failing_dev == dev)
6948 if (missing > max_tolerated) {
6951 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6952 em->start, missing, max_tolerated);
6953 free_extent_map(em);
6957 next_start = extent_map_end(em);
6958 free_extent_map(em);
6960 read_lock(&map_tree->map_tree.lock);
6961 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6962 (u64)(-1) - next_start);
6963 read_unlock(&map_tree->map_tree.lock);
6969 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6971 struct btrfs_root *root = fs_info->chunk_root;
6972 struct btrfs_path *path;
6973 struct extent_buffer *leaf;
6974 struct btrfs_key key;
6975 struct btrfs_key found_key;
6980 path = btrfs_alloc_path();
6984 mutex_lock(&uuid_mutex);
6985 mutex_lock(&fs_info->chunk_mutex);
6988 * Read all device items, and then all the chunk items. All
6989 * device items are found before any chunk item (their object id
6990 * is smaller than the lowest possible object id for a chunk
6991 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6993 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6996 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7000 leaf = path->nodes[0];
7001 slot = path->slots[0];
7002 if (slot >= btrfs_header_nritems(leaf)) {
7003 ret = btrfs_next_leaf(root, path);
7010 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7011 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7012 struct btrfs_dev_item *dev_item;
7013 dev_item = btrfs_item_ptr(leaf, slot,
7014 struct btrfs_dev_item);
7015 ret = read_one_dev(fs_info, leaf, dev_item);
7019 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7020 struct btrfs_chunk *chunk;
7021 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7022 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7030 * After loading chunk tree, we've got all device information,
7031 * do another round of validation checks.
7033 if (total_dev != fs_info->fs_devices->total_devices) {
7035 "super_num_devices %llu mismatch with num_devices %llu found here",
7036 btrfs_super_num_devices(fs_info->super_copy),
7041 if (btrfs_super_total_bytes(fs_info->super_copy) <
7042 fs_info->fs_devices->total_rw_bytes) {
7044 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7045 btrfs_super_total_bytes(fs_info->super_copy),
7046 fs_info->fs_devices->total_rw_bytes);
7052 mutex_unlock(&fs_info->chunk_mutex);
7053 mutex_unlock(&uuid_mutex);
7055 btrfs_free_path(path);
7059 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7061 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7062 struct btrfs_device *device;
7064 while (fs_devices) {
7065 mutex_lock(&fs_devices->device_list_mutex);
7066 list_for_each_entry(device, &fs_devices->devices, dev_list)
7067 device->fs_info = fs_info;
7068 mutex_unlock(&fs_devices->device_list_mutex);
7070 fs_devices = fs_devices->seed;
7074 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7078 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7079 btrfs_dev_stat_reset(dev, i);
7082 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7084 struct btrfs_key key;
7085 struct btrfs_key found_key;
7086 struct btrfs_root *dev_root = fs_info->dev_root;
7087 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7088 struct extent_buffer *eb;
7091 struct btrfs_device *device;
7092 struct btrfs_path *path = NULL;
7095 path = btrfs_alloc_path();
7101 mutex_lock(&fs_devices->device_list_mutex);
7102 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7104 struct btrfs_dev_stats_item *ptr;
7106 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7107 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7108 key.offset = device->devid;
7109 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7111 __btrfs_reset_dev_stats(device);
7112 device->dev_stats_valid = 1;
7113 btrfs_release_path(path);
7116 slot = path->slots[0];
7117 eb = path->nodes[0];
7118 btrfs_item_key_to_cpu(eb, &found_key, slot);
7119 item_size = btrfs_item_size_nr(eb, slot);
7121 ptr = btrfs_item_ptr(eb, slot,
7122 struct btrfs_dev_stats_item);
7124 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7125 if (item_size >= (1 + i) * sizeof(__le64))
7126 btrfs_dev_stat_set(device, i,
7127 btrfs_dev_stats_value(eb, ptr, i));
7129 btrfs_dev_stat_reset(device, i);
7132 device->dev_stats_valid = 1;
7133 btrfs_dev_stat_print_on_load(device);
7134 btrfs_release_path(path);
7136 mutex_unlock(&fs_devices->device_list_mutex);
7139 btrfs_free_path(path);
7140 return ret < 0 ? ret : 0;
7143 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7144 struct btrfs_fs_info *fs_info,
7145 struct btrfs_device *device)
7147 struct btrfs_root *dev_root = fs_info->dev_root;
7148 struct btrfs_path *path;
7149 struct btrfs_key key;
7150 struct extent_buffer *eb;
7151 struct btrfs_dev_stats_item *ptr;
7155 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7156 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7157 key.offset = device->devid;
7159 path = btrfs_alloc_path();
7162 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7164 btrfs_warn_in_rcu(fs_info,
7165 "error %d while searching for dev_stats item for device %s",
7166 ret, rcu_str_deref(device->name));
7171 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7172 /* need to delete old one and insert a new one */
7173 ret = btrfs_del_item(trans, dev_root, path);
7175 btrfs_warn_in_rcu(fs_info,
7176 "delete too small dev_stats item for device %s failed %d",
7177 rcu_str_deref(device->name), ret);
7184 /* need to insert a new item */
7185 btrfs_release_path(path);
7186 ret = btrfs_insert_empty_item(trans, dev_root, path,
7187 &key, sizeof(*ptr));
7189 btrfs_warn_in_rcu(fs_info,
7190 "insert dev_stats item for device %s failed %d",
7191 rcu_str_deref(device->name), ret);
7196 eb = path->nodes[0];
7197 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7198 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7199 btrfs_set_dev_stats_value(eb, ptr, i,
7200 btrfs_dev_stat_read(device, i));
7201 btrfs_mark_buffer_dirty(eb);
7204 btrfs_free_path(path);
7209 * called from commit_transaction. Writes all changed device stats to disk.
7211 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7212 struct btrfs_fs_info *fs_info)
7214 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7215 struct btrfs_device *device;
7219 mutex_lock(&fs_devices->device_list_mutex);
7220 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7221 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7222 if (!device->dev_stats_valid || stats_cnt == 0)
7227 * There is a LOAD-LOAD control dependency between the value of
7228 * dev_stats_ccnt and updating the on-disk values which requires
7229 * reading the in-memory counters. Such control dependencies
7230 * require explicit read memory barriers.
7232 * This memory barriers pairs with smp_mb__before_atomic in
7233 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7234 * barrier implied by atomic_xchg in
7235 * btrfs_dev_stats_read_and_reset
7239 ret = update_dev_stat_item(trans, fs_info, device);
7241 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7243 mutex_unlock(&fs_devices->device_list_mutex);
7248 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7250 btrfs_dev_stat_inc(dev, index);
7251 btrfs_dev_stat_print_on_error(dev);
7254 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7256 if (!dev->dev_stats_valid)
7258 btrfs_err_rl_in_rcu(dev->fs_info,
7259 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7260 rcu_str_deref(dev->name),
7261 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7262 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7263 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7264 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7265 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7268 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7272 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7273 if (btrfs_dev_stat_read(dev, i) != 0)
7275 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7276 return; /* all values == 0, suppress message */
7278 btrfs_info_in_rcu(dev->fs_info,
7279 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7280 rcu_str_deref(dev->name),
7281 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7282 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7283 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7284 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7285 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7288 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7289 struct btrfs_ioctl_get_dev_stats *stats)
7291 struct btrfs_device *dev;
7292 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7295 mutex_lock(&fs_devices->device_list_mutex);
7296 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7297 mutex_unlock(&fs_devices->device_list_mutex);
7300 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7302 } else if (!dev->dev_stats_valid) {
7303 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7305 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7306 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7307 if (stats->nr_items > i)
7309 btrfs_dev_stat_read_and_reset(dev, i);
7311 btrfs_dev_stat_reset(dev, i);
7314 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7315 if (stats->nr_items > i)
7316 stats->values[i] = btrfs_dev_stat_read(dev, i);
7318 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7319 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7323 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7325 struct buffer_head *bh;
7326 struct btrfs_super_block *disk_super;
7332 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7335 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7338 disk_super = (struct btrfs_super_block *)bh->b_data;
7340 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7341 set_buffer_dirty(bh);
7342 sync_dirty_buffer(bh);
7346 /* Notify udev that device has changed */
7347 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7349 /* Update ctime/mtime for device path for libblkid */
7350 update_dev_time(device_path);
7354 * Update the size of all devices, which is used for writing out the
7357 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7359 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7360 struct btrfs_device *curr, *next;
7362 if (list_empty(&fs_devices->resized_devices))
7365 mutex_lock(&fs_devices->device_list_mutex);
7366 mutex_lock(&fs_info->chunk_mutex);
7367 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7369 list_del_init(&curr->resized_list);
7370 curr->commit_total_bytes = curr->disk_total_bytes;
7372 mutex_unlock(&fs_info->chunk_mutex);
7373 mutex_unlock(&fs_devices->device_list_mutex);
7376 /* Must be invoked during the transaction commit */
7377 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7379 struct btrfs_fs_info *fs_info = trans->fs_info;
7380 struct extent_map *em;
7381 struct map_lookup *map;
7382 struct btrfs_device *dev;
7385 if (list_empty(&trans->pending_chunks))
7388 /* In order to kick the device replace finish process */
7389 mutex_lock(&fs_info->chunk_mutex);
7390 list_for_each_entry(em, &trans->pending_chunks, list) {
7391 map = em->map_lookup;
7393 for (i = 0; i < map->num_stripes; i++) {
7394 dev = map->stripes[i].dev;
7395 dev->commit_bytes_used = dev->bytes_used;
7398 mutex_unlock(&fs_info->chunk_mutex);
7401 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7403 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7404 while (fs_devices) {
7405 fs_devices->fs_info = fs_info;
7406 fs_devices = fs_devices->seed;
7410 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7412 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7413 while (fs_devices) {
7414 fs_devices->fs_info = NULL;
7415 fs_devices = fs_devices->seed;