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 <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
49 .devs_max = 0, /* 0 == as many as possible */
51 .tolerated_failures = 1,
55 [BTRFS_RAID_RAID1] = {
60 .tolerated_failures = 1,
69 .tolerated_failures = 0,
73 [BTRFS_RAID_RAID0] = {
78 .tolerated_failures = 0,
82 [BTRFS_RAID_SINGLE] = {
87 .tolerated_failures = 0,
91 [BTRFS_RAID_RAID5] = {
96 .tolerated_failures = 1,
100 [BTRFS_RAID_RAID6] = {
105 .tolerated_failures = 2,
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_fs_info *fs_info);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
143 enum btrfs_map_op op,
144 u64 logical, u64 *length,
145 struct btrfs_bio **bbio_ret,
146 int mirror_num, int need_raid_map);
152 * There are several mutexes that protect manipulation of devices and low-level
153 * structures like chunks but not block groups, extents or files
155 * uuid_mutex (global lock)
156 * ------------------------
157 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
158 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
159 * device) or requested by the device= mount option
161 * the mutex can be very coarse and can cover long-running operations
163 * protects: updates to fs_devices counters like missing devices, rw devices,
164 * seeding, structure cloning, openning/closing devices at mount/umount time
166 * global::fs_devs - add, remove, updates to the global list
168 * does not protect: manipulation of the fs_devices::devices list!
170 * btrfs_device::name - renames (write side), read is RCU
172 * fs_devices::device_list_mutex (per-fs, with RCU)
173 * ------------------------------------------------
174 * protects updates to fs_devices::devices, ie. adding and deleting
176 * simple list traversal with read-only actions can be done with RCU protection
178 * may be used to exclude some operations from running concurrently without any
179 * modifications to the list (see write_all_supers)
183 * coarse lock owned by a mounted filesystem; used to exclude some operations
184 * that cannot run in parallel and affect the higher-level properties of the
185 * filesystem like: device add/deleting/resize/replace, or balance
189 * protects balance structures (status, state) and context accessed from
190 * several places (internally, ioctl)
194 * protects chunks, adding or removing during allocation, trim or when a new
195 * device is added/removed
199 * a big lock that is held by the cleaner thread and prevents running subvolume
200 * cleaning together with relocation or delayed iputs
213 DEFINE_MUTEX(uuid_mutex);
214 static LIST_HEAD(fs_uuids);
215 struct list_head *btrfs_get_fs_uuids(void)
221 * alloc_fs_devices - allocate struct btrfs_fs_devices
222 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
224 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
225 * The returned struct is not linked onto any lists and can be destroyed with
226 * kfree() right away.
228 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
230 struct btrfs_fs_devices *fs_devs;
232 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
234 return ERR_PTR(-ENOMEM);
236 mutex_init(&fs_devs->device_list_mutex);
238 INIT_LIST_HEAD(&fs_devs->devices);
239 INIT_LIST_HEAD(&fs_devs->resized_devices);
240 INIT_LIST_HEAD(&fs_devs->alloc_list);
241 INIT_LIST_HEAD(&fs_devs->list);
243 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
248 static void free_device(struct btrfs_device *device)
250 rcu_string_free(device->name);
251 bio_put(device->flush_bio);
255 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
257 struct btrfs_device *device;
258 WARN_ON(fs_devices->opened);
259 while (!list_empty(&fs_devices->devices)) {
260 device = list_entry(fs_devices->devices.next,
261 struct btrfs_device, dev_list);
262 list_del(&device->dev_list);
268 static void btrfs_kobject_uevent(struct block_device *bdev,
269 enum kobject_action action)
273 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
275 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
277 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
278 &disk_to_dev(bdev->bd_disk)->kobj);
281 void btrfs_cleanup_fs_uuids(void)
283 struct btrfs_fs_devices *fs_devices;
285 while (!list_empty(&fs_uuids)) {
286 fs_devices = list_entry(fs_uuids.next,
287 struct btrfs_fs_devices, list);
288 list_del(&fs_devices->list);
289 free_fs_devices(fs_devices);
294 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
295 * Returned struct is not linked onto any lists and must be destroyed using
298 static struct btrfs_device *__alloc_device(void)
300 struct btrfs_device *dev;
302 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
304 return ERR_PTR(-ENOMEM);
307 * Preallocate a bio that's always going to be used for flushing device
308 * barriers and matches the device lifespan
310 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
311 if (!dev->flush_bio) {
313 return ERR_PTR(-ENOMEM);
316 INIT_LIST_HEAD(&dev->dev_list);
317 INIT_LIST_HEAD(&dev->dev_alloc_list);
318 INIT_LIST_HEAD(&dev->resized_list);
320 spin_lock_init(&dev->io_lock);
322 atomic_set(&dev->reada_in_flight, 0);
323 atomic_set(&dev->dev_stats_ccnt, 0);
324 btrfs_device_data_ordered_init(dev);
325 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
326 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
332 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
335 * If devid and uuid are both specified, the match must be exact, otherwise
336 * only devid is used.
338 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
339 u64 devid, const u8 *uuid)
341 struct list_head *head = &fs_devices->devices;
342 struct btrfs_device *dev;
344 list_for_each_entry(dev, head, dev_list) {
345 if (dev->devid == devid &&
346 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
353 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
355 struct btrfs_fs_devices *fs_devices;
357 list_for_each_entry(fs_devices, &fs_uuids, list) {
358 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
365 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
366 int flush, struct block_device **bdev,
367 struct buffer_head **bh)
371 *bdev = blkdev_get_by_path(device_path, flags, holder);
374 ret = PTR_ERR(*bdev);
379 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
380 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
382 blkdev_put(*bdev, flags);
385 invalidate_bdev(*bdev);
386 *bh = btrfs_read_dev_super(*bdev);
389 blkdev_put(*bdev, flags);
401 static void requeue_list(struct btrfs_pending_bios *pending_bios,
402 struct bio *head, struct bio *tail)
405 struct bio *old_head;
407 old_head = pending_bios->head;
408 pending_bios->head = head;
409 if (pending_bios->tail)
410 tail->bi_next = old_head;
412 pending_bios->tail = tail;
416 * we try to collect pending bios for a device so we don't get a large
417 * number of procs sending bios down to the same device. This greatly
418 * improves the schedulers ability to collect and merge the bios.
420 * But, it also turns into a long list of bios to process and that is sure
421 * to eventually make the worker thread block. The solution here is to
422 * make some progress and then put this work struct back at the end of
423 * the list if the block device is congested. This way, multiple devices
424 * can make progress from a single worker thread.
426 static noinline void run_scheduled_bios(struct btrfs_device *device)
428 struct btrfs_fs_info *fs_info = device->fs_info;
430 struct backing_dev_info *bdi;
431 struct btrfs_pending_bios *pending_bios;
435 unsigned long num_run;
436 unsigned long batch_run = 0;
437 unsigned long last_waited = 0;
439 int sync_pending = 0;
440 struct blk_plug plug;
443 * this function runs all the bios we've collected for
444 * a particular device. We don't want to wander off to
445 * another device without first sending all of these down.
446 * So, setup a plug here and finish it off before we return
448 blk_start_plug(&plug);
450 bdi = device->bdev->bd_bdi;
453 spin_lock(&device->io_lock);
458 /* take all the bios off the list at once and process them
459 * later on (without the lock held). But, remember the
460 * tail and other pointers so the bios can be properly reinserted
461 * into the list if we hit congestion
463 if (!force_reg && device->pending_sync_bios.head) {
464 pending_bios = &device->pending_sync_bios;
467 pending_bios = &device->pending_bios;
471 pending = pending_bios->head;
472 tail = pending_bios->tail;
473 WARN_ON(pending && !tail);
476 * if pending was null this time around, no bios need processing
477 * at all and we can stop. Otherwise it'll loop back up again
478 * and do an additional check so no bios are missed.
480 * device->running_pending is used to synchronize with the
483 if (device->pending_sync_bios.head == NULL &&
484 device->pending_bios.head == NULL) {
486 device->running_pending = 0;
489 device->running_pending = 1;
492 pending_bios->head = NULL;
493 pending_bios->tail = NULL;
495 spin_unlock(&device->io_lock);
500 /* we want to work on both lists, but do more bios on the
501 * sync list than the regular list
504 pending_bios != &device->pending_sync_bios &&
505 device->pending_sync_bios.head) ||
506 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
507 device->pending_bios.head)) {
508 spin_lock(&device->io_lock);
509 requeue_list(pending_bios, pending, tail);
514 pending = pending->bi_next;
517 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
520 * if we're doing the sync list, record that our
521 * plug has some sync requests on it
523 * If we're doing the regular list and there are
524 * sync requests sitting around, unplug before
527 if (pending_bios == &device->pending_sync_bios) {
529 } else if (sync_pending) {
530 blk_finish_plug(&plug);
531 blk_start_plug(&plug);
535 btrfsic_submit_bio(cur);
542 * we made progress, there is more work to do and the bdi
543 * is now congested. Back off and let other work structs
546 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
547 fs_info->fs_devices->open_devices > 1) {
548 struct io_context *ioc;
550 ioc = current->io_context;
553 * the main goal here is that we don't want to
554 * block if we're going to be able to submit
555 * more requests without blocking.
557 * This code does two great things, it pokes into
558 * the elevator code from a filesystem _and_
559 * it makes assumptions about how batching works.
561 if (ioc && ioc->nr_batch_requests > 0 &&
562 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
564 ioc->last_waited == last_waited)) {
566 * we want to go through our batch of
567 * requests and stop. So, we copy out
568 * the ioc->last_waited time and test
569 * against it before looping
571 last_waited = ioc->last_waited;
575 spin_lock(&device->io_lock);
576 requeue_list(pending_bios, pending, tail);
577 device->running_pending = 1;
579 spin_unlock(&device->io_lock);
580 btrfs_queue_work(fs_info->submit_workers,
590 spin_lock(&device->io_lock);
591 if (device->pending_bios.head || device->pending_sync_bios.head)
593 spin_unlock(&device->io_lock);
596 blk_finish_plug(&plug);
599 static void pending_bios_fn(struct btrfs_work *work)
601 struct btrfs_device *device;
603 device = container_of(work, struct btrfs_device, work);
604 run_scheduled_bios(device);
608 * Search and remove all stale (devices which are not mounted) devices.
609 * When both inputs are NULL, it will search and release all stale devices.
610 * path: Optional. When provided will it release all unmounted devices
611 * matching this path only.
612 * skip_dev: Optional. Will skip this device when searching for the stale
615 static void btrfs_free_stale_devices(const char *path,
616 struct btrfs_device *skip_dev)
618 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
619 struct btrfs_device *dev, *tmp_dev;
621 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) {
626 list_for_each_entry_safe(dev, tmp_dev,
627 &fs_devs->devices, dev_list) {
630 if (skip_dev && skip_dev == dev)
632 if (path && !dev->name)
637 not_found = strcmp(rcu_str_deref(dev->name),
643 /* delete the stale device */
644 if (fs_devs->num_devices == 1) {
645 btrfs_sysfs_remove_fsid(fs_devs);
646 list_del(&fs_devs->list);
647 free_fs_devices(fs_devs);
650 fs_devs->num_devices--;
651 list_del(&dev->dev_list);
658 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
659 struct btrfs_device *device, fmode_t flags,
662 struct request_queue *q;
663 struct block_device *bdev;
664 struct buffer_head *bh;
665 struct btrfs_super_block *disk_super;
674 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
679 disk_super = (struct btrfs_super_block *)bh->b_data;
680 devid = btrfs_stack_device_id(&disk_super->dev_item);
681 if (devid != device->devid)
684 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
687 device->generation = btrfs_super_generation(disk_super);
689 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
690 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
691 fs_devices->seeding = 1;
693 if (bdev_read_only(bdev))
694 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
696 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
699 q = bdev_get_queue(bdev);
700 if (!blk_queue_nonrot(q))
701 fs_devices->rotating = 1;
704 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
705 device->mode = flags;
707 fs_devices->open_devices++;
708 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
709 device->devid != BTRFS_DEV_REPLACE_DEVID) {
710 fs_devices->rw_devices++;
711 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
719 blkdev_put(bdev, flags);
725 * Add new device to list of registered devices
728 * device pointer which was just added or updated when successful
729 * error pointer when failed
731 static noinline struct btrfs_device *device_list_add(const char *path,
732 struct btrfs_super_block *disk_super)
734 struct btrfs_device *device;
735 struct btrfs_fs_devices *fs_devices;
736 struct rcu_string *name;
737 u64 found_transid = btrfs_super_generation(disk_super);
738 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
740 fs_devices = find_fsid(disk_super->fsid);
742 fs_devices = alloc_fs_devices(disk_super->fsid);
743 if (IS_ERR(fs_devices))
744 return ERR_CAST(fs_devices);
746 list_add(&fs_devices->list, &fs_uuids);
750 device = find_device(fs_devices, devid,
751 disk_super->dev_item.uuid);
755 if (fs_devices->opened)
756 return ERR_PTR(-EBUSY);
758 device = btrfs_alloc_device(NULL, &devid,
759 disk_super->dev_item.uuid);
760 if (IS_ERR(device)) {
761 /* we can safely leave the fs_devices entry around */
765 name = rcu_string_strdup(path, GFP_NOFS);
768 return ERR_PTR(-ENOMEM);
770 rcu_assign_pointer(device->name, name);
772 mutex_lock(&fs_devices->device_list_mutex);
773 list_add_rcu(&device->dev_list, &fs_devices->devices);
774 fs_devices->num_devices++;
775 mutex_unlock(&fs_devices->device_list_mutex);
777 device->fs_devices = fs_devices;
778 btrfs_free_stale_devices(path, device);
780 if (disk_super->label[0])
781 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
782 disk_super->label, devid, found_transid, path);
784 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
785 disk_super->fsid, devid, found_transid, path);
787 } else if (!device->name || strcmp(device->name->str, path)) {
789 * When FS is already mounted.
790 * 1. If you are here and if the device->name is NULL that
791 * means this device was missing at time of FS mount.
792 * 2. If you are here and if the device->name is different
793 * from 'path' that means either
794 * a. The same device disappeared and reappeared with
796 * b. The missing-disk-which-was-replaced, has
799 * We must allow 1 and 2a above. But 2b would be a spurious
802 * Further in case of 1 and 2a above, the disk at 'path'
803 * would have missed some transaction when it was away and
804 * in case of 2a the stale bdev has to be updated as well.
805 * 2b must not be allowed at all time.
809 * For now, we do allow update to btrfs_fs_device through the
810 * btrfs dev scan cli after FS has been mounted. We're still
811 * tracking a problem where systems fail mount by subvolume id
812 * when we reject replacement on a mounted FS.
814 if (!fs_devices->opened && found_transid < device->generation) {
816 * That is if the FS is _not_ mounted and if you
817 * are here, that means there is more than one
818 * disk with same uuid and devid.We keep the one
819 * with larger generation number or the last-in if
820 * generation are equal.
822 return ERR_PTR(-EEXIST);
825 name = rcu_string_strdup(path, GFP_NOFS);
827 return ERR_PTR(-ENOMEM);
828 rcu_string_free(device->name);
829 rcu_assign_pointer(device->name, name);
830 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
831 fs_devices->missing_devices--;
832 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
837 * Unmount does not free the btrfs_device struct but would zero
838 * generation along with most of the other members. So just update
839 * it back. We need it to pick the disk with largest generation
842 if (!fs_devices->opened)
843 device->generation = found_transid;
845 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
850 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
852 struct btrfs_fs_devices *fs_devices;
853 struct btrfs_device *device;
854 struct btrfs_device *orig_dev;
856 fs_devices = alloc_fs_devices(orig->fsid);
857 if (IS_ERR(fs_devices))
860 mutex_lock(&orig->device_list_mutex);
861 fs_devices->total_devices = orig->total_devices;
863 /* We have held the volume lock, it is safe to get the devices. */
864 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
865 struct rcu_string *name;
867 device = btrfs_alloc_device(NULL, &orig_dev->devid,
873 * This is ok to do without rcu read locked because we hold the
874 * uuid mutex so nothing we touch in here is going to disappear.
876 if (orig_dev->name) {
877 name = rcu_string_strdup(orig_dev->name->str,
883 rcu_assign_pointer(device->name, name);
886 list_add(&device->dev_list, &fs_devices->devices);
887 device->fs_devices = fs_devices;
888 fs_devices->num_devices++;
890 mutex_unlock(&orig->device_list_mutex);
893 mutex_unlock(&orig->device_list_mutex);
894 free_fs_devices(fs_devices);
895 return ERR_PTR(-ENOMEM);
898 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
900 struct btrfs_device *device, *next;
901 struct btrfs_device *latest_dev = NULL;
903 mutex_lock(&uuid_mutex);
905 /* This is the initialized path, it is safe to release the devices. */
906 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
907 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
908 &device->dev_state)) {
909 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
910 &device->dev_state) &&
912 device->generation > latest_dev->generation)) {
918 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
920 * In the first step, keep the device which has
921 * the correct fsid and the devid that is used
922 * for the dev_replace procedure.
923 * In the second step, the dev_replace state is
924 * read from the device tree and it is known
925 * whether the procedure is really active or
926 * not, which means whether this device is
927 * used or whether it should be removed.
929 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
930 &device->dev_state)) {
935 blkdev_put(device->bdev, device->mode);
937 fs_devices->open_devices--;
939 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
940 list_del_init(&device->dev_alloc_list);
941 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
942 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
944 fs_devices->rw_devices--;
946 list_del_init(&device->dev_list);
947 fs_devices->num_devices--;
951 if (fs_devices->seed) {
952 fs_devices = fs_devices->seed;
956 fs_devices->latest_bdev = latest_dev->bdev;
958 mutex_unlock(&uuid_mutex);
961 static void free_device_rcu(struct rcu_head *head)
963 struct btrfs_device *device;
965 device = container_of(head, struct btrfs_device, rcu);
969 static void btrfs_close_bdev(struct btrfs_device *device)
974 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
975 sync_blockdev(device->bdev);
976 invalidate_bdev(device->bdev);
979 blkdev_put(device->bdev, device->mode);
982 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
984 struct btrfs_fs_devices *fs_devices = device->fs_devices;
985 struct btrfs_device *new_device;
986 struct rcu_string *name;
989 fs_devices->open_devices--;
991 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
992 device->devid != BTRFS_DEV_REPLACE_DEVID) {
993 list_del_init(&device->dev_alloc_list);
994 fs_devices->rw_devices--;
997 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
998 fs_devices->missing_devices--;
1000 new_device = btrfs_alloc_device(NULL, &device->devid,
1002 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1004 /* Safe because we are under uuid_mutex */
1006 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1007 BUG_ON(!name); /* -ENOMEM */
1008 rcu_assign_pointer(new_device->name, name);
1011 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1012 new_device->fs_devices = device->fs_devices;
1015 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1017 struct btrfs_device *device, *tmp;
1018 struct list_head pending_put;
1020 INIT_LIST_HEAD(&pending_put);
1022 if (--fs_devices->opened > 0)
1025 mutex_lock(&fs_devices->device_list_mutex);
1026 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1027 btrfs_prepare_close_one_device(device);
1028 list_add(&device->dev_list, &pending_put);
1030 mutex_unlock(&fs_devices->device_list_mutex);
1033 * btrfs_show_devname() is using the device_list_mutex,
1034 * sometimes call to blkdev_put() leads vfs calling
1035 * into this func. So do put outside of device_list_mutex,
1038 while (!list_empty(&pending_put)) {
1039 device = list_first_entry(&pending_put,
1040 struct btrfs_device, dev_list);
1041 list_del(&device->dev_list);
1042 btrfs_close_bdev(device);
1043 call_rcu(&device->rcu, free_device_rcu);
1046 WARN_ON(fs_devices->open_devices);
1047 WARN_ON(fs_devices->rw_devices);
1048 fs_devices->opened = 0;
1049 fs_devices->seeding = 0;
1054 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1056 struct btrfs_fs_devices *seed_devices = NULL;
1059 mutex_lock(&uuid_mutex);
1060 ret = __btrfs_close_devices(fs_devices);
1061 if (!fs_devices->opened) {
1062 seed_devices = fs_devices->seed;
1063 fs_devices->seed = NULL;
1065 mutex_unlock(&uuid_mutex);
1067 while (seed_devices) {
1068 fs_devices = seed_devices;
1069 seed_devices = fs_devices->seed;
1070 __btrfs_close_devices(fs_devices);
1071 free_fs_devices(fs_devices);
1076 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1077 fmode_t flags, void *holder)
1079 struct list_head *head = &fs_devices->devices;
1080 struct btrfs_device *device;
1081 struct btrfs_device *latest_dev = NULL;
1084 flags |= FMODE_EXCL;
1086 list_for_each_entry(device, head, dev_list) {
1087 /* Just open everything we can; ignore failures here */
1088 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1092 device->generation > latest_dev->generation)
1093 latest_dev = device;
1095 if (fs_devices->open_devices == 0) {
1099 fs_devices->opened = 1;
1100 fs_devices->latest_bdev = latest_dev->bdev;
1101 fs_devices->total_rw_bytes = 0;
1106 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1107 fmode_t flags, void *holder)
1111 mutex_lock(&uuid_mutex);
1112 if (fs_devices->opened) {
1113 fs_devices->opened++;
1116 ret = __btrfs_open_devices(fs_devices, flags, holder);
1118 mutex_unlock(&uuid_mutex);
1122 static void btrfs_release_disk_super(struct page *page)
1128 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1130 struct btrfs_super_block **disk_super)
1135 /* make sure our super fits in the device */
1136 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1139 /* make sure our super fits in the page */
1140 if (sizeof(**disk_super) > PAGE_SIZE)
1143 /* make sure our super doesn't straddle pages on disk */
1144 index = bytenr >> PAGE_SHIFT;
1145 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1148 /* pull in the page with our super */
1149 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1152 if (IS_ERR_OR_NULL(*page))
1157 /* align our pointer to the offset of the super block */
1158 *disk_super = p + (bytenr & ~PAGE_MASK);
1160 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1161 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1162 btrfs_release_disk_super(*page);
1166 if ((*disk_super)->label[0] &&
1167 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1168 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1174 * Look for a btrfs signature on a device. This may be called out of the mount path
1175 * and we are not allowed to call set_blocksize during the scan. The superblock
1176 * is read via pagecache
1178 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1179 struct btrfs_fs_devices **fs_devices_ret)
1181 struct btrfs_super_block *disk_super;
1182 struct btrfs_device *device;
1183 struct block_device *bdev;
1189 * we would like to check all the supers, but that would make
1190 * a btrfs mount succeed after a mkfs from a different FS.
1191 * So, we need to add a special mount option to scan for
1192 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1194 bytenr = btrfs_sb_offset(0);
1195 flags |= FMODE_EXCL;
1196 mutex_lock(&uuid_mutex);
1198 bdev = blkdev_get_by_path(path, flags, holder);
1200 ret = PTR_ERR(bdev);
1204 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1206 goto error_bdev_put;
1209 device = device_list_add(path, disk_super);
1211 ret = PTR_ERR(device);
1213 *fs_devices_ret = device->fs_devices;
1215 btrfs_release_disk_super(page);
1218 blkdev_put(bdev, flags);
1220 mutex_unlock(&uuid_mutex);
1224 /* helper to account the used device space in the range */
1225 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1226 u64 end, u64 *length)
1228 struct btrfs_key key;
1229 struct btrfs_root *root = device->fs_info->dev_root;
1230 struct btrfs_dev_extent *dev_extent;
1231 struct btrfs_path *path;
1235 struct extent_buffer *l;
1239 if (start >= device->total_bytes ||
1240 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1243 path = btrfs_alloc_path();
1246 path->reada = READA_FORWARD;
1248 key.objectid = device->devid;
1250 key.type = BTRFS_DEV_EXTENT_KEY;
1252 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1256 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1263 slot = path->slots[0];
1264 if (slot >= btrfs_header_nritems(l)) {
1265 ret = btrfs_next_leaf(root, path);
1273 btrfs_item_key_to_cpu(l, &key, slot);
1275 if (key.objectid < device->devid)
1278 if (key.objectid > device->devid)
1281 if (key.type != BTRFS_DEV_EXTENT_KEY)
1284 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1285 extent_end = key.offset + btrfs_dev_extent_length(l,
1287 if (key.offset <= start && extent_end > end) {
1288 *length = end - start + 1;
1290 } else if (key.offset <= start && extent_end > start)
1291 *length += extent_end - start;
1292 else if (key.offset > start && extent_end <= end)
1293 *length += extent_end - key.offset;
1294 else if (key.offset > start && key.offset <= end) {
1295 *length += end - key.offset + 1;
1297 } else if (key.offset > end)
1305 btrfs_free_path(path);
1309 static int contains_pending_extent(struct btrfs_transaction *transaction,
1310 struct btrfs_device *device,
1311 u64 *start, u64 len)
1313 struct btrfs_fs_info *fs_info = device->fs_info;
1314 struct extent_map *em;
1315 struct list_head *search_list = &fs_info->pinned_chunks;
1317 u64 physical_start = *start;
1320 search_list = &transaction->pending_chunks;
1322 list_for_each_entry(em, search_list, list) {
1323 struct map_lookup *map;
1326 map = em->map_lookup;
1327 for (i = 0; i < map->num_stripes; i++) {
1330 if (map->stripes[i].dev != device)
1332 if (map->stripes[i].physical >= physical_start + len ||
1333 map->stripes[i].physical + em->orig_block_len <=
1337 * Make sure that while processing the pinned list we do
1338 * not override our *start with a lower value, because
1339 * we can have pinned chunks that fall within this
1340 * device hole and that have lower physical addresses
1341 * than the pending chunks we processed before. If we
1342 * do not take this special care we can end up getting
1343 * 2 pending chunks that start at the same physical
1344 * device offsets because the end offset of a pinned
1345 * chunk can be equal to the start offset of some
1348 end = map->stripes[i].physical + em->orig_block_len;
1355 if (search_list != &fs_info->pinned_chunks) {
1356 search_list = &fs_info->pinned_chunks;
1365 * find_free_dev_extent_start - find free space in the specified device
1366 * @device: the device which we search the free space in
1367 * @num_bytes: the size of the free space that we need
1368 * @search_start: the position from which to begin the search
1369 * @start: store the start of the free space.
1370 * @len: the size of the free space. that we find, or the size
1371 * of the max free space if we don't find suitable free space
1373 * this uses a pretty simple search, the expectation is that it is
1374 * called very infrequently and that a given device has a small number
1377 * @start is used to store the start of the free space if we find. But if we
1378 * don't find suitable free space, it will be used to store the start position
1379 * of the max free space.
1381 * @len is used to store the size of the free space that we find.
1382 * But if we don't find suitable free space, it is used to store the size of
1383 * the max free space.
1385 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1386 struct btrfs_device *device, u64 num_bytes,
1387 u64 search_start, u64 *start, u64 *len)
1389 struct btrfs_fs_info *fs_info = device->fs_info;
1390 struct btrfs_root *root = fs_info->dev_root;
1391 struct btrfs_key key;
1392 struct btrfs_dev_extent *dev_extent;
1393 struct btrfs_path *path;
1398 u64 search_end = device->total_bytes;
1401 struct extent_buffer *l;
1404 * We don't want to overwrite the superblock on the drive nor any area
1405 * used by the boot loader (grub for example), so we make sure to start
1406 * at an offset of at least 1MB.
1408 search_start = max_t(u64, search_start, SZ_1M);
1410 path = btrfs_alloc_path();
1414 max_hole_start = search_start;
1418 if (search_start >= search_end ||
1419 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1424 path->reada = READA_FORWARD;
1425 path->search_commit_root = 1;
1426 path->skip_locking = 1;
1428 key.objectid = device->devid;
1429 key.offset = search_start;
1430 key.type = BTRFS_DEV_EXTENT_KEY;
1432 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1436 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1443 slot = path->slots[0];
1444 if (slot >= btrfs_header_nritems(l)) {
1445 ret = btrfs_next_leaf(root, path);
1453 btrfs_item_key_to_cpu(l, &key, slot);
1455 if (key.objectid < device->devid)
1458 if (key.objectid > device->devid)
1461 if (key.type != BTRFS_DEV_EXTENT_KEY)
1464 if (key.offset > search_start) {
1465 hole_size = key.offset - search_start;
1468 * Have to check before we set max_hole_start, otherwise
1469 * we could end up sending back this offset anyway.
1471 if (contains_pending_extent(transaction, device,
1474 if (key.offset >= search_start) {
1475 hole_size = key.offset - search_start;
1482 if (hole_size > max_hole_size) {
1483 max_hole_start = search_start;
1484 max_hole_size = hole_size;
1488 * If this free space is greater than which we need,
1489 * it must be the max free space that we have found
1490 * until now, so max_hole_start must point to the start
1491 * of this free space and the length of this free space
1492 * is stored in max_hole_size. Thus, we return
1493 * max_hole_start and max_hole_size and go back to the
1496 if (hole_size >= num_bytes) {
1502 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1503 extent_end = key.offset + btrfs_dev_extent_length(l,
1505 if (extent_end > search_start)
1506 search_start = extent_end;
1513 * At this point, search_start should be the end of
1514 * allocated dev extents, and when shrinking the device,
1515 * search_end may be smaller than search_start.
1517 if (search_end > search_start) {
1518 hole_size = search_end - search_start;
1520 if (contains_pending_extent(transaction, device, &search_start,
1522 btrfs_release_path(path);
1526 if (hole_size > max_hole_size) {
1527 max_hole_start = search_start;
1528 max_hole_size = hole_size;
1533 if (max_hole_size < num_bytes)
1539 btrfs_free_path(path);
1540 *start = max_hole_start;
1542 *len = max_hole_size;
1546 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1547 struct btrfs_device *device, u64 num_bytes,
1548 u64 *start, u64 *len)
1550 /* FIXME use last free of some kind */
1551 return find_free_dev_extent_start(trans->transaction, device,
1552 num_bytes, 0, start, len);
1555 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1556 struct btrfs_device *device,
1557 u64 start, u64 *dev_extent_len)
1559 struct btrfs_fs_info *fs_info = device->fs_info;
1560 struct btrfs_root *root = fs_info->dev_root;
1562 struct btrfs_path *path;
1563 struct btrfs_key key;
1564 struct btrfs_key found_key;
1565 struct extent_buffer *leaf = NULL;
1566 struct btrfs_dev_extent *extent = NULL;
1568 path = btrfs_alloc_path();
1572 key.objectid = device->devid;
1574 key.type = BTRFS_DEV_EXTENT_KEY;
1576 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1578 ret = btrfs_previous_item(root, path, key.objectid,
1579 BTRFS_DEV_EXTENT_KEY);
1582 leaf = path->nodes[0];
1583 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1584 extent = btrfs_item_ptr(leaf, path->slots[0],
1585 struct btrfs_dev_extent);
1586 BUG_ON(found_key.offset > start || found_key.offset +
1587 btrfs_dev_extent_length(leaf, extent) < start);
1589 btrfs_release_path(path);
1591 } else if (ret == 0) {
1592 leaf = path->nodes[0];
1593 extent = btrfs_item_ptr(leaf, path->slots[0],
1594 struct btrfs_dev_extent);
1596 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1600 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1602 ret = btrfs_del_item(trans, root, path);
1604 btrfs_handle_fs_error(fs_info, ret,
1605 "Failed to remove dev extent item");
1607 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1610 btrfs_free_path(path);
1614 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1615 struct btrfs_device *device,
1616 u64 chunk_offset, u64 start, u64 num_bytes)
1619 struct btrfs_path *path;
1620 struct btrfs_fs_info *fs_info = device->fs_info;
1621 struct btrfs_root *root = fs_info->dev_root;
1622 struct btrfs_dev_extent *extent;
1623 struct extent_buffer *leaf;
1624 struct btrfs_key key;
1626 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1627 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1628 path = btrfs_alloc_path();
1632 key.objectid = device->devid;
1634 key.type = BTRFS_DEV_EXTENT_KEY;
1635 ret = btrfs_insert_empty_item(trans, root, path, &key,
1640 leaf = path->nodes[0];
1641 extent = btrfs_item_ptr(leaf, path->slots[0],
1642 struct btrfs_dev_extent);
1643 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1644 BTRFS_CHUNK_TREE_OBJECTID);
1645 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1646 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1647 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1649 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1650 btrfs_mark_buffer_dirty(leaf);
1652 btrfs_free_path(path);
1656 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1658 struct extent_map_tree *em_tree;
1659 struct extent_map *em;
1663 em_tree = &fs_info->mapping_tree.map_tree;
1664 read_lock(&em_tree->lock);
1665 n = rb_last(&em_tree->map);
1667 em = rb_entry(n, struct extent_map, rb_node);
1668 ret = em->start + em->len;
1670 read_unlock(&em_tree->lock);
1675 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1679 struct btrfs_key key;
1680 struct btrfs_key found_key;
1681 struct btrfs_path *path;
1683 path = btrfs_alloc_path();
1687 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1688 key.type = BTRFS_DEV_ITEM_KEY;
1689 key.offset = (u64)-1;
1691 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1695 BUG_ON(ret == 0); /* Corruption */
1697 ret = btrfs_previous_item(fs_info->chunk_root, path,
1698 BTRFS_DEV_ITEMS_OBJECTID,
1699 BTRFS_DEV_ITEM_KEY);
1703 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1705 *devid_ret = found_key.offset + 1;
1709 btrfs_free_path(path);
1714 * the device information is stored in the chunk root
1715 * the btrfs_device struct should be fully filled in
1717 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1718 struct btrfs_fs_info *fs_info,
1719 struct btrfs_device *device)
1721 struct btrfs_root *root = fs_info->chunk_root;
1723 struct btrfs_path *path;
1724 struct btrfs_dev_item *dev_item;
1725 struct extent_buffer *leaf;
1726 struct btrfs_key key;
1729 path = btrfs_alloc_path();
1733 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1734 key.type = BTRFS_DEV_ITEM_KEY;
1735 key.offset = device->devid;
1737 ret = btrfs_insert_empty_item(trans, root, path, &key,
1742 leaf = path->nodes[0];
1743 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1745 btrfs_set_device_id(leaf, dev_item, device->devid);
1746 btrfs_set_device_generation(leaf, dev_item, 0);
1747 btrfs_set_device_type(leaf, dev_item, device->type);
1748 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1749 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1750 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1751 btrfs_set_device_total_bytes(leaf, dev_item,
1752 btrfs_device_get_disk_total_bytes(device));
1753 btrfs_set_device_bytes_used(leaf, dev_item,
1754 btrfs_device_get_bytes_used(device));
1755 btrfs_set_device_group(leaf, dev_item, 0);
1756 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1757 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1758 btrfs_set_device_start_offset(leaf, dev_item, 0);
1760 ptr = btrfs_device_uuid(dev_item);
1761 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1762 ptr = btrfs_device_fsid(dev_item);
1763 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1764 btrfs_mark_buffer_dirty(leaf);
1768 btrfs_free_path(path);
1773 * Function to update ctime/mtime for a given device path.
1774 * Mainly used for ctime/mtime based probe like libblkid.
1776 static void update_dev_time(const char *path_name)
1780 filp = filp_open(path_name, O_RDWR, 0);
1783 file_update_time(filp);
1784 filp_close(filp, NULL);
1787 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1788 struct btrfs_device *device)
1790 struct btrfs_root *root = fs_info->chunk_root;
1792 struct btrfs_path *path;
1793 struct btrfs_key key;
1794 struct btrfs_trans_handle *trans;
1796 path = btrfs_alloc_path();
1800 trans = btrfs_start_transaction(root, 0);
1801 if (IS_ERR(trans)) {
1802 btrfs_free_path(path);
1803 return PTR_ERR(trans);
1805 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1806 key.type = BTRFS_DEV_ITEM_KEY;
1807 key.offset = device->devid;
1809 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1813 btrfs_abort_transaction(trans, ret);
1814 btrfs_end_transaction(trans);
1818 ret = btrfs_del_item(trans, root, path);
1820 btrfs_abort_transaction(trans, ret);
1821 btrfs_end_transaction(trans);
1825 btrfs_free_path(path);
1827 ret = btrfs_commit_transaction(trans);
1832 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1833 * filesystem. It's up to the caller to adjust that number regarding eg. device
1836 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1844 seq = read_seqbegin(&fs_info->profiles_lock);
1846 all_avail = fs_info->avail_data_alloc_bits |
1847 fs_info->avail_system_alloc_bits |
1848 fs_info->avail_metadata_alloc_bits;
1849 } while (read_seqretry(&fs_info->profiles_lock, seq));
1851 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1852 if (!(all_avail & btrfs_raid_group[i]))
1855 if (num_devices < btrfs_raid_array[i].devs_min) {
1856 int ret = btrfs_raid_mindev_error[i];
1866 static struct btrfs_device * btrfs_find_next_active_device(
1867 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1869 struct btrfs_device *next_device;
1871 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1872 if (next_device != device &&
1873 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1874 && next_device->bdev)
1882 * Helper function to check if the given device is part of s_bdev / latest_bdev
1883 * and replace it with the provided or the next active device, in the context
1884 * where this function called, there should be always be another device (or
1885 * this_dev) which is active.
1887 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1888 struct btrfs_device *device, struct btrfs_device *this_dev)
1890 struct btrfs_device *next_device;
1893 next_device = this_dev;
1895 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1897 ASSERT(next_device);
1899 if (fs_info->sb->s_bdev &&
1900 (fs_info->sb->s_bdev == device->bdev))
1901 fs_info->sb->s_bdev = next_device->bdev;
1903 if (fs_info->fs_devices->latest_bdev == device->bdev)
1904 fs_info->fs_devices->latest_bdev = next_device->bdev;
1907 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1910 struct btrfs_device *device;
1911 struct btrfs_fs_devices *cur_devices;
1915 mutex_lock(&fs_info->volume_mutex);
1916 mutex_lock(&uuid_mutex);
1918 num_devices = fs_info->fs_devices->num_devices;
1919 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1920 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1921 WARN_ON(num_devices < 1);
1924 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1926 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1930 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1935 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1936 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1940 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1941 fs_info->fs_devices->rw_devices == 1) {
1942 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1946 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1947 mutex_lock(&fs_info->chunk_mutex);
1948 list_del_init(&device->dev_alloc_list);
1949 device->fs_devices->rw_devices--;
1950 mutex_unlock(&fs_info->chunk_mutex);
1953 mutex_unlock(&uuid_mutex);
1954 ret = btrfs_shrink_device(device, 0);
1955 mutex_lock(&uuid_mutex);
1960 * TODO: the superblock still includes this device in its num_devices
1961 * counter although write_all_supers() is not locked out. This
1962 * could give a filesystem state which requires a degraded mount.
1964 ret = btrfs_rm_dev_item(fs_info, device);
1968 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1969 btrfs_scrub_cancel_dev(fs_info, device);
1972 * the device list mutex makes sure that we don't change
1973 * the device list while someone else is writing out all
1974 * the device supers. Whoever is writing all supers, should
1975 * lock the device list mutex before getting the number of
1976 * devices in the super block (super_copy). Conversely,
1977 * whoever updates the number of devices in the super block
1978 * (super_copy) should hold the device list mutex.
1981 cur_devices = device->fs_devices;
1982 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1983 list_del_rcu(&device->dev_list);
1985 device->fs_devices->num_devices--;
1986 device->fs_devices->total_devices--;
1988 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1989 device->fs_devices->missing_devices--;
1991 btrfs_assign_next_active_device(fs_info, device, NULL);
1994 device->fs_devices->open_devices--;
1995 /* remove sysfs entry */
1996 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1999 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2000 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2001 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2004 * at this point, the device is zero sized and detached from
2005 * the devices list. All that's left is to zero out the old
2006 * supers and free the device.
2008 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2009 btrfs_scratch_superblocks(device->bdev, device->name->str);
2011 btrfs_close_bdev(device);
2012 call_rcu(&device->rcu, free_device_rcu);
2014 if (cur_devices->open_devices == 0) {
2015 struct btrfs_fs_devices *fs_devices;
2016 fs_devices = fs_info->fs_devices;
2017 while (fs_devices) {
2018 if (fs_devices->seed == cur_devices) {
2019 fs_devices->seed = cur_devices->seed;
2022 fs_devices = fs_devices->seed;
2024 cur_devices->seed = NULL;
2025 __btrfs_close_devices(cur_devices);
2026 free_fs_devices(cur_devices);
2030 mutex_unlock(&uuid_mutex);
2031 mutex_unlock(&fs_info->volume_mutex);
2035 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2036 mutex_lock(&fs_info->chunk_mutex);
2037 list_add(&device->dev_alloc_list,
2038 &fs_info->fs_devices->alloc_list);
2039 device->fs_devices->rw_devices++;
2040 mutex_unlock(&fs_info->chunk_mutex);
2045 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2046 struct btrfs_device *srcdev)
2048 struct btrfs_fs_devices *fs_devices;
2050 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2053 * in case of fs with no seed, srcdev->fs_devices will point
2054 * to fs_devices of fs_info. However when the dev being replaced is
2055 * a seed dev it will point to the seed's local fs_devices. In short
2056 * srcdev will have its correct fs_devices in both the cases.
2058 fs_devices = srcdev->fs_devices;
2060 list_del_rcu(&srcdev->dev_list);
2061 list_del(&srcdev->dev_alloc_list);
2062 fs_devices->num_devices--;
2063 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2064 fs_devices->missing_devices--;
2066 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2067 fs_devices->rw_devices--;
2070 fs_devices->open_devices--;
2073 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2074 struct btrfs_device *srcdev)
2076 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2078 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2079 /* zero out the old super if it is writable */
2080 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2083 btrfs_close_bdev(srcdev);
2084 call_rcu(&srcdev->rcu, free_device_rcu);
2086 /* if this is no devs we rather delete the fs_devices */
2087 if (!fs_devices->num_devices) {
2088 struct btrfs_fs_devices *tmp_fs_devices;
2091 * On a mounted FS, num_devices can't be zero unless it's a
2092 * seed. In case of a seed device being replaced, the replace
2093 * target added to the sprout FS, so there will be no more
2094 * device left under the seed FS.
2096 ASSERT(fs_devices->seeding);
2098 tmp_fs_devices = fs_info->fs_devices;
2099 while (tmp_fs_devices) {
2100 if (tmp_fs_devices->seed == fs_devices) {
2101 tmp_fs_devices->seed = fs_devices->seed;
2104 tmp_fs_devices = tmp_fs_devices->seed;
2106 fs_devices->seed = NULL;
2107 __btrfs_close_devices(fs_devices);
2108 free_fs_devices(fs_devices);
2112 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2113 struct btrfs_device *tgtdev)
2115 mutex_lock(&uuid_mutex);
2117 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2119 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2122 fs_info->fs_devices->open_devices--;
2124 fs_info->fs_devices->num_devices--;
2126 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2128 list_del_rcu(&tgtdev->dev_list);
2130 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2131 mutex_unlock(&uuid_mutex);
2134 * The update_dev_time() with in btrfs_scratch_superblocks()
2135 * may lead to a call to btrfs_show_devname() which will try
2136 * to hold device_list_mutex. And here this device
2137 * is already out of device list, so we don't have to hold
2138 * the device_list_mutex lock.
2140 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2142 btrfs_close_bdev(tgtdev);
2143 call_rcu(&tgtdev->rcu, free_device_rcu);
2146 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2147 const char *device_path,
2148 struct btrfs_device **device)
2151 struct btrfs_super_block *disk_super;
2154 struct block_device *bdev;
2155 struct buffer_head *bh;
2158 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2159 fs_info->bdev_holder, 0, &bdev, &bh);
2162 disk_super = (struct btrfs_super_block *)bh->b_data;
2163 devid = btrfs_stack_device_id(&disk_super->dev_item);
2164 dev_uuid = disk_super->dev_item.uuid;
2165 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2169 blkdev_put(bdev, FMODE_READ);
2173 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2174 const char *device_path,
2175 struct btrfs_device **device)
2178 if (strcmp(device_path, "missing") == 0) {
2179 struct list_head *devices;
2180 struct btrfs_device *tmp;
2182 devices = &fs_info->fs_devices->devices;
2184 * It is safe to read the devices since the volume_mutex
2185 * is held by the caller.
2187 list_for_each_entry(tmp, devices, dev_list) {
2188 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2189 &tmp->dev_state) && !tmp->bdev) {
2196 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2200 return btrfs_find_device_by_path(fs_info, device_path, device);
2205 * Lookup a device given by device id, or the path if the id is 0.
2207 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2208 const char *devpath,
2209 struct btrfs_device **device)
2215 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2219 if (!devpath || !devpath[0])
2222 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2229 * does all the dirty work required for changing file system's UUID.
2231 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2233 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2234 struct btrfs_fs_devices *old_devices;
2235 struct btrfs_fs_devices *seed_devices;
2236 struct btrfs_super_block *disk_super = fs_info->super_copy;
2237 struct btrfs_device *device;
2240 BUG_ON(!mutex_is_locked(&uuid_mutex));
2241 if (!fs_devices->seeding)
2244 seed_devices = alloc_fs_devices(NULL);
2245 if (IS_ERR(seed_devices))
2246 return PTR_ERR(seed_devices);
2248 old_devices = clone_fs_devices(fs_devices);
2249 if (IS_ERR(old_devices)) {
2250 kfree(seed_devices);
2251 return PTR_ERR(old_devices);
2254 list_add(&old_devices->list, &fs_uuids);
2256 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2257 seed_devices->opened = 1;
2258 INIT_LIST_HEAD(&seed_devices->devices);
2259 INIT_LIST_HEAD(&seed_devices->alloc_list);
2260 mutex_init(&seed_devices->device_list_mutex);
2262 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2263 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2265 list_for_each_entry(device, &seed_devices->devices, dev_list)
2266 device->fs_devices = seed_devices;
2268 mutex_lock(&fs_info->chunk_mutex);
2269 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2270 mutex_unlock(&fs_info->chunk_mutex);
2272 fs_devices->seeding = 0;
2273 fs_devices->num_devices = 0;
2274 fs_devices->open_devices = 0;
2275 fs_devices->missing_devices = 0;
2276 fs_devices->rotating = 0;
2277 fs_devices->seed = seed_devices;
2279 generate_random_uuid(fs_devices->fsid);
2280 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2281 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2282 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2284 super_flags = btrfs_super_flags(disk_super) &
2285 ~BTRFS_SUPER_FLAG_SEEDING;
2286 btrfs_set_super_flags(disk_super, super_flags);
2292 * Store the expected generation for seed devices in device items.
2294 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2295 struct btrfs_fs_info *fs_info)
2297 struct btrfs_root *root = fs_info->chunk_root;
2298 struct btrfs_path *path;
2299 struct extent_buffer *leaf;
2300 struct btrfs_dev_item *dev_item;
2301 struct btrfs_device *device;
2302 struct btrfs_key key;
2303 u8 fs_uuid[BTRFS_FSID_SIZE];
2304 u8 dev_uuid[BTRFS_UUID_SIZE];
2308 path = btrfs_alloc_path();
2312 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2314 key.type = BTRFS_DEV_ITEM_KEY;
2317 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2321 leaf = path->nodes[0];
2323 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2324 ret = btrfs_next_leaf(root, path);
2329 leaf = path->nodes[0];
2330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2331 btrfs_release_path(path);
2335 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2336 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2337 key.type != BTRFS_DEV_ITEM_KEY)
2340 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2341 struct btrfs_dev_item);
2342 devid = btrfs_device_id(leaf, dev_item);
2343 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2345 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2347 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2348 BUG_ON(!device); /* Logic error */
2350 if (device->fs_devices->seeding) {
2351 btrfs_set_device_generation(leaf, dev_item,
2352 device->generation);
2353 btrfs_mark_buffer_dirty(leaf);
2361 btrfs_free_path(path);
2365 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2367 struct btrfs_root *root = fs_info->dev_root;
2368 struct request_queue *q;
2369 struct btrfs_trans_handle *trans;
2370 struct btrfs_device *device;
2371 struct block_device *bdev;
2372 struct list_head *devices;
2373 struct super_block *sb = fs_info->sb;
2374 struct rcu_string *name;
2376 int seeding_dev = 0;
2378 bool unlocked = false;
2380 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2383 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2384 fs_info->bdev_holder);
2386 return PTR_ERR(bdev);
2388 if (fs_info->fs_devices->seeding) {
2390 down_write(&sb->s_umount);
2391 mutex_lock(&uuid_mutex);
2394 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2396 devices = &fs_info->fs_devices->devices;
2398 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2399 list_for_each_entry(device, devices, dev_list) {
2400 if (device->bdev == bdev) {
2403 &fs_info->fs_devices->device_list_mutex);
2407 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2409 device = btrfs_alloc_device(fs_info, NULL, NULL);
2410 if (IS_ERR(device)) {
2411 /* we can safely leave the fs_devices entry around */
2412 ret = PTR_ERR(device);
2416 name = rcu_string_strdup(device_path, GFP_KERNEL);
2419 goto error_free_device;
2421 rcu_assign_pointer(device->name, name);
2423 trans = btrfs_start_transaction(root, 0);
2424 if (IS_ERR(trans)) {
2425 ret = PTR_ERR(trans);
2426 goto error_free_device;
2429 q = bdev_get_queue(bdev);
2430 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2431 device->generation = trans->transid;
2432 device->io_width = fs_info->sectorsize;
2433 device->io_align = fs_info->sectorsize;
2434 device->sector_size = fs_info->sectorsize;
2435 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2436 fs_info->sectorsize);
2437 device->disk_total_bytes = device->total_bytes;
2438 device->commit_total_bytes = device->total_bytes;
2439 device->fs_info = fs_info;
2440 device->bdev = bdev;
2441 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2442 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2443 device->mode = FMODE_EXCL;
2444 device->dev_stats_valid = 1;
2445 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2448 sb->s_flags &= ~SB_RDONLY;
2449 ret = btrfs_prepare_sprout(fs_info);
2451 btrfs_abort_transaction(trans, ret);
2456 device->fs_devices = fs_info->fs_devices;
2458 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2459 mutex_lock(&fs_info->chunk_mutex);
2460 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2461 list_add(&device->dev_alloc_list,
2462 &fs_info->fs_devices->alloc_list);
2463 fs_info->fs_devices->num_devices++;
2464 fs_info->fs_devices->open_devices++;
2465 fs_info->fs_devices->rw_devices++;
2466 fs_info->fs_devices->total_devices++;
2467 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2469 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2471 if (!blk_queue_nonrot(q))
2472 fs_info->fs_devices->rotating = 1;
2474 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2475 btrfs_set_super_total_bytes(fs_info->super_copy,
2476 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2478 tmp = btrfs_super_num_devices(fs_info->super_copy);
2479 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2481 /* add sysfs device entry */
2482 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2485 * we've got more storage, clear any full flags on the space
2488 btrfs_clear_space_info_full(fs_info);
2490 mutex_unlock(&fs_info->chunk_mutex);
2491 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2494 mutex_lock(&fs_info->chunk_mutex);
2495 ret = init_first_rw_device(trans, fs_info);
2496 mutex_unlock(&fs_info->chunk_mutex);
2498 btrfs_abort_transaction(trans, ret);
2503 ret = btrfs_add_dev_item(trans, fs_info, device);
2505 btrfs_abort_transaction(trans, ret);
2510 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2512 ret = btrfs_finish_sprout(trans, fs_info);
2514 btrfs_abort_transaction(trans, ret);
2518 /* Sprouting would change fsid of the mounted root,
2519 * so rename the fsid on the sysfs
2521 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2523 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2525 "sysfs: failed to create fsid for sprout");
2528 ret = btrfs_commit_transaction(trans);
2531 mutex_unlock(&uuid_mutex);
2532 up_write(&sb->s_umount);
2535 if (ret) /* transaction commit */
2538 ret = btrfs_relocate_sys_chunks(fs_info);
2540 btrfs_handle_fs_error(fs_info, ret,
2541 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2542 trans = btrfs_attach_transaction(root);
2543 if (IS_ERR(trans)) {
2544 if (PTR_ERR(trans) == -ENOENT)
2546 ret = PTR_ERR(trans);
2550 ret = btrfs_commit_transaction(trans);
2553 /* Update ctime/mtime for libblkid */
2554 update_dev_time(device_path);
2558 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2561 sb->s_flags |= SB_RDONLY;
2563 btrfs_end_transaction(trans);
2565 free_device(device);
2567 blkdev_put(bdev, FMODE_EXCL);
2568 if (seeding_dev && !unlocked) {
2569 mutex_unlock(&uuid_mutex);
2570 up_write(&sb->s_umount);
2575 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2576 const char *device_path,
2577 struct btrfs_device *srcdev,
2578 struct btrfs_device **device_out)
2580 struct btrfs_device *device;
2581 struct block_device *bdev;
2582 struct list_head *devices;
2583 struct rcu_string *name;
2584 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2588 if (fs_info->fs_devices->seeding) {
2589 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2593 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2594 fs_info->bdev_holder);
2596 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2597 return PTR_ERR(bdev);
2600 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2602 devices = &fs_info->fs_devices->devices;
2603 list_for_each_entry(device, devices, dev_list) {
2604 if (device->bdev == bdev) {
2606 "target device is in the filesystem!");
2613 if (i_size_read(bdev->bd_inode) <
2614 btrfs_device_get_total_bytes(srcdev)) {
2616 "target device is smaller than source device!");
2622 device = btrfs_alloc_device(NULL, &devid, NULL);
2623 if (IS_ERR(device)) {
2624 ret = PTR_ERR(device);
2628 name = rcu_string_strdup(device_path, GFP_KERNEL);
2630 free_device(device);
2634 rcu_assign_pointer(device->name, name);
2636 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2637 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2638 device->generation = 0;
2639 device->io_width = fs_info->sectorsize;
2640 device->io_align = fs_info->sectorsize;
2641 device->sector_size = fs_info->sectorsize;
2642 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2643 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2644 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2645 ASSERT(list_empty(&srcdev->resized_list));
2646 device->commit_total_bytes = srcdev->commit_total_bytes;
2647 device->commit_bytes_used = device->bytes_used;
2648 device->fs_info = fs_info;
2649 device->bdev = bdev;
2650 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2651 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2652 device->mode = FMODE_EXCL;
2653 device->dev_stats_valid = 1;
2654 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2655 device->fs_devices = fs_info->fs_devices;
2656 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2657 fs_info->fs_devices->num_devices++;
2658 fs_info->fs_devices->open_devices++;
2659 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2661 *device_out = device;
2665 blkdev_put(bdev, FMODE_EXCL);
2669 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2670 struct btrfs_device *tgtdev)
2672 u32 sectorsize = fs_info->sectorsize;
2674 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2675 tgtdev->io_width = sectorsize;
2676 tgtdev->io_align = sectorsize;
2677 tgtdev->sector_size = sectorsize;
2678 tgtdev->fs_info = fs_info;
2679 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &tgtdev->dev_state);
2682 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2683 struct btrfs_device *device)
2686 struct btrfs_path *path;
2687 struct btrfs_root *root = device->fs_info->chunk_root;
2688 struct btrfs_dev_item *dev_item;
2689 struct extent_buffer *leaf;
2690 struct btrfs_key key;
2692 path = btrfs_alloc_path();
2696 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2697 key.type = BTRFS_DEV_ITEM_KEY;
2698 key.offset = device->devid;
2700 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2709 leaf = path->nodes[0];
2710 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2712 btrfs_set_device_id(leaf, dev_item, device->devid);
2713 btrfs_set_device_type(leaf, dev_item, device->type);
2714 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2715 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2716 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2717 btrfs_set_device_total_bytes(leaf, dev_item,
2718 btrfs_device_get_disk_total_bytes(device));
2719 btrfs_set_device_bytes_used(leaf, dev_item,
2720 btrfs_device_get_bytes_used(device));
2721 btrfs_mark_buffer_dirty(leaf);
2724 btrfs_free_path(path);
2728 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2729 struct btrfs_device *device, u64 new_size)
2731 struct btrfs_fs_info *fs_info = device->fs_info;
2732 struct btrfs_super_block *super_copy = fs_info->super_copy;
2733 struct btrfs_fs_devices *fs_devices;
2737 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2740 new_size = round_down(new_size, fs_info->sectorsize);
2742 mutex_lock(&fs_info->chunk_mutex);
2743 old_total = btrfs_super_total_bytes(super_copy);
2744 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2746 if (new_size <= device->total_bytes ||
2747 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2748 mutex_unlock(&fs_info->chunk_mutex);
2752 fs_devices = fs_info->fs_devices;
2754 btrfs_set_super_total_bytes(super_copy,
2755 round_down(old_total + diff, fs_info->sectorsize));
2756 device->fs_devices->total_rw_bytes += diff;
2758 btrfs_device_set_total_bytes(device, new_size);
2759 btrfs_device_set_disk_total_bytes(device, new_size);
2760 btrfs_clear_space_info_full(device->fs_info);
2761 if (list_empty(&device->resized_list))
2762 list_add_tail(&device->resized_list,
2763 &fs_devices->resized_devices);
2764 mutex_unlock(&fs_info->chunk_mutex);
2766 return btrfs_update_device(trans, device);
2769 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2770 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2772 struct btrfs_root *root = fs_info->chunk_root;
2774 struct btrfs_path *path;
2775 struct btrfs_key key;
2777 path = btrfs_alloc_path();
2781 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2782 key.offset = chunk_offset;
2783 key.type = BTRFS_CHUNK_ITEM_KEY;
2785 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2788 else if (ret > 0) { /* Logic error or corruption */
2789 btrfs_handle_fs_error(fs_info, -ENOENT,
2790 "Failed lookup while freeing chunk.");
2795 ret = btrfs_del_item(trans, root, path);
2797 btrfs_handle_fs_error(fs_info, ret,
2798 "Failed to delete chunk item.");
2800 btrfs_free_path(path);
2804 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2806 struct btrfs_super_block *super_copy = fs_info->super_copy;
2807 struct btrfs_disk_key *disk_key;
2808 struct btrfs_chunk *chunk;
2815 struct btrfs_key key;
2817 mutex_lock(&fs_info->chunk_mutex);
2818 array_size = btrfs_super_sys_array_size(super_copy);
2820 ptr = super_copy->sys_chunk_array;
2823 while (cur < array_size) {
2824 disk_key = (struct btrfs_disk_key *)ptr;
2825 btrfs_disk_key_to_cpu(&key, disk_key);
2827 len = sizeof(*disk_key);
2829 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2830 chunk = (struct btrfs_chunk *)(ptr + len);
2831 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2832 len += btrfs_chunk_item_size(num_stripes);
2837 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2838 key.offset == chunk_offset) {
2839 memmove(ptr, ptr + len, array_size - (cur + len));
2841 btrfs_set_super_sys_array_size(super_copy, array_size);
2847 mutex_unlock(&fs_info->chunk_mutex);
2851 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2852 u64 logical, u64 length)
2854 struct extent_map_tree *em_tree;
2855 struct extent_map *em;
2857 em_tree = &fs_info->mapping_tree.map_tree;
2858 read_lock(&em_tree->lock);
2859 em = lookup_extent_mapping(em_tree, logical, length);
2860 read_unlock(&em_tree->lock);
2863 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2865 return ERR_PTR(-EINVAL);
2868 if (em->start > logical || em->start + em->len < logical) {
2870 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2871 logical, length, em->start, em->start + em->len);
2872 free_extent_map(em);
2873 return ERR_PTR(-EINVAL);
2876 /* callers are responsible for dropping em's ref. */
2880 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2881 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2883 struct extent_map *em;
2884 struct map_lookup *map;
2885 u64 dev_extent_len = 0;
2887 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2889 em = get_chunk_map(fs_info, chunk_offset, 1);
2892 * This is a logic error, but we don't want to just rely on the
2893 * user having built with ASSERT enabled, so if ASSERT doesn't
2894 * do anything we still error out.
2899 map = em->map_lookup;
2900 mutex_lock(&fs_info->chunk_mutex);
2901 check_system_chunk(trans, fs_info, map->type);
2902 mutex_unlock(&fs_info->chunk_mutex);
2905 * Take the device list mutex to prevent races with the final phase of
2906 * a device replace operation that replaces the device object associated
2907 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2909 mutex_lock(&fs_devices->device_list_mutex);
2910 for (i = 0; i < map->num_stripes; i++) {
2911 struct btrfs_device *device = map->stripes[i].dev;
2912 ret = btrfs_free_dev_extent(trans, device,
2913 map->stripes[i].physical,
2916 mutex_unlock(&fs_devices->device_list_mutex);
2917 btrfs_abort_transaction(trans, ret);
2921 if (device->bytes_used > 0) {
2922 mutex_lock(&fs_info->chunk_mutex);
2923 btrfs_device_set_bytes_used(device,
2924 device->bytes_used - dev_extent_len);
2925 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2926 btrfs_clear_space_info_full(fs_info);
2927 mutex_unlock(&fs_info->chunk_mutex);
2930 if (map->stripes[i].dev) {
2931 ret = btrfs_update_device(trans, map->stripes[i].dev);
2933 mutex_unlock(&fs_devices->device_list_mutex);
2934 btrfs_abort_transaction(trans, ret);
2939 mutex_unlock(&fs_devices->device_list_mutex);
2941 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2943 btrfs_abort_transaction(trans, ret);
2947 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2949 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2950 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2952 btrfs_abort_transaction(trans, ret);
2957 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2959 btrfs_abort_transaction(trans, ret);
2965 free_extent_map(em);
2969 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2971 struct btrfs_root *root = fs_info->chunk_root;
2972 struct btrfs_trans_handle *trans;
2976 * Prevent races with automatic removal of unused block groups.
2977 * After we relocate and before we remove the chunk with offset
2978 * chunk_offset, automatic removal of the block group can kick in,
2979 * resulting in a failure when calling btrfs_remove_chunk() below.
2981 * Make sure to acquire this mutex before doing a tree search (dev
2982 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2983 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2984 * we release the path used to search the chunk/dev tree and before
2985 * the current task acquires this mutex and calls us.
2987 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2989 ret = btrfs_can_relocate(fs_info, chunk_offset);
2993 /* step one, relocate all the extents inside this chunk */
2994 btrfs_scrub_pause(fs_info);
2995 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2996 btrfs_scrub_continue(fs_info);
3000 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3002 if (IS_ERR(trans)) {
3003 ret = PTR_ERR(trans);
3004 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3009 * step two, delete the device extents and the
3010 * chunk tree entries
3012 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3013 btrfs_end_transaction(trans);
3017 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3019 struct btrfs_root *chunk_root = fs_info->chunk_root;
3020 struct btrfs_path *path;
3021 struct extent_buffer *leaf;
3022 struct btrfs_chunk *chunk;
3023 struct btrfs_key key;
3024 struct btrfs_key found_key;
3026 bool retried = false;
3030 path = btrfs_alloc_path();
3035 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3036 key.offset = (u64)-1;
3037 key.type = BTRFS_CHUNK_ITEM_KEY;
3040 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3041 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3043 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3046 BUG_ON(ret == 0); /* Corruption */
3048 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3051 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3057 leaf = path->nodes[0];
3058 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3060 chunk = btrfs_item_ptr(leaf, path->slots[0],
3061 struct btrfs_chunk);
3062 chunk_type = btrfs_chunk_type(leaf, chunk);
3063 btrfs_release_path(path);
3065 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3066 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3072 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3074 if (found_key.offset == 0)
3076 key.offset = found_key.offset - 1;
3079 if (failed && !retried) {
3083 } else if (WARN_ON(failed && retried)) {
3087 btrfs_free_path(path);
3092 * return 1 : allocate a data chunk successfully,
3093 * return <0: errors during allocating a data chunk,
3094 * return 0 : no need to allocate a data chunk.
3096 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3099 struct btrfs_block_group_cache *cache;
3103 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3105 chunk_type = cache->flags;
3106 btrfs_put_block_group(cache);
3108 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3109 spin_lock(&fs_info->data_sinfo->lock);
3110 bytes_used = fs_info->data_sinfo->bytes_used;
3111 spin_unlock(&fs_info->data_sinfo->lock);
3114 struct btrfs_trans_handle *trans;
3117 trans = btrfs_join_transaction(fs_info->tree_root);
3119 return PTR_ERR(trans);
3121 ret = btrfs_force_chunk_alloc(trans, fs_info,
3122 BTRFS_BLOCK_GROUP_DATA);
3123 btrfs_end_transaction(trans);
3133 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3134 struct btrfs_balance_control *bctl)
3136 struct btrfs_root *root = fs_info->tree_root;
3137 struct btrfs_trans_handle *trans;
3138 struct btrfs_balance_item *item;
3139 struct btrfs_disk_balance_args disk_bargs;
3140 struct btrfs_path *path;
3141 struct extent_buffer *leaf;
3142 struct btrfs_key key;
3145 path = btrfs_alloc_path();
3149 trans = btrfs_start_transaction(root, 0);
3150 if (IS_ERR(trans)) {
3151 btrfs_free_path(path);
3152 return PTR_ERR(trans);
3155 key.objectid = BTRFS_BALANCE_OBJECTID;
3156 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3159 ret = btrfs_insert_empty_item(trans, root, path, &key,
3164 leaf = path->nodes[0];
3165 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3167 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3169 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3170 btrfs_set_balance_data(leaf, item, &disk_bargs);
3171 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3172 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3173 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3174 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3176 btrfs_set_balance_flags(leaf, item, bctl->flags);
3178 btrfs_mark_buffer_dirty(leaf);
3180 btrfs_free_path(path);
3181 err = btrfs_commit_transaction(trans);
3187 static int del_balance_item(struct btrfs_fs_info *fs_info)
3189 struct btrfs_root *root = fs_info->tree_root;
3190 struct btrfs_trans_handle *trans;
3191 struct btrfs_path *path;
3192 struct btrfs_key key;
3195 path = btrfs_alloc_path();
3199 trans = btrfs_start_transaction(root, 0);
3200 if (IS_ERR(trans)) {
3201 btrfs_free_path(path);
3202 return PTR_ERR(trans);
3205 key.objectid = BTRFS_BALANCE_OBJECTID;
3206 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3209 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3217 ret = btrfs_del_item(trans, root, path);
3219 btrfs_free_path(path);
3220 err = btrfs_commit_transaction(trans);
3227 * This is a heuristic used to reduce the number of chunks balanced on
3228 * resume after balance was interrupted.
3230 static void update_balance_args(struct btrfs_balance_control *bctl)
3233 * Turn on soft mode for chunk types that were being converted.
3235 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3236 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3237 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3238 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3239 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3240 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3243 * Turn on usage filter if is not already used. The idea is
3244 * that chunks that we have already balanced should be
3245 * reasonably full. Don't do it for chunks that are being
3246 * converted - that will keep us from relocating unconverted
3247 * (albeit full) chunks.
3249 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3250 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3251 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3252 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3253 bctl->data.usage = 90;
3255 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3256 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3257 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3258 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3259 bctl->sys.usage = 90;
3261 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3262 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3263 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3264 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3265 bctl->meta.usage = 90;
3270 * Should be called with both balance and volume mutexes held to
3271 * serialize other volume operations (add_dev/rm_dev/resize) with
3272 * restriper. Same goes for unset_balance_control.
3274 static void set_balance_control(struct btrfs_balance_control *bctl)
3276 struct btrfs_fs_info *fs_info = bctl->fs_info;
3278 BUG_ON(fs_info->balance_ctl);
3280 spin_lock(&fs_info->balance_lock);
3281 fs_info->balance_ctl = bctl;
3282 spin_unlock(&fs_info->balance_lock);
3285 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3287 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3289 BUG_ON(!fs_info->balance_ctl);
3291 spin_lock(&fs_info->balance_lock);
3292 fs_info->balance_ctl = NULL;
3293 spin_unlock(&fs_info->balance_lock);
3299 * Balance filters. Return 1 if chunk should be filtered out
3300 * (should not be balanced).
3302 static int chunk_profiles_filter(u64 chunk_type,
3303 struct btrfs_balance_args *bargs)
3305 chunk_type = chunk_to_extended(chunk_type) &
3306 BTRFS_EXTENDED_PROFILE_MASK;
3308 if (bargs->profiles & chunk_type)
3314 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3315 struct btrfs_balance_args *bargs)
3317 struct btrfs_block_group_cache *cache;
3319 u64 user_thresh_min;
3320 u64 user_thresh_max;
3323 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3324 chunk_used = btrfs_block_group_used(&cache->item);
3326 if (bargs->usage_min == 0)
3327 user_thresh_min = 0;
3329 user_thresh_min = div_factor_fine(cache->key.offset,
3332 if (bargs->usage_max == 0)
3333 user_thresh_max = 1;
3334 else if (bargs->usage_max > 100)
3335 user_thresh_max = cache->key.offset;
3337 user_thresh_max = div_factor_fine(cache->key.offset,
3340 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3343 btrfs_put_block_group(cache);
3347 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3348 u64 chunk_offset, struct btrfs_balance_args *bargs)
3350 struct btrfs_block_group_cache *cache;
3351 u64 chunk_used, user_thresh;
3354 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3355 chunk_used = btrfs_block_group_used(&cache->item);
3357 if (bargs->usage_min == 0)
3359 else if (bargs->usage > 100)
3360 user_thresh = cache->key.offset;
3362 user_thresh = div_factor_fine(cache->key.offset,
3365 if (chunk_used < user_thresh)
3368 btrfs_put_block_group(cache);
3372 static int chunk_devid_filter(struct extent_buffer *leaf,
3373 struct btrfs_chunk *chunk,
3374 struct btrfs_balance_args *bargs)
3376 struct btrfs_stripe *stripe;
3377 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3380 for (i = 0; i < num_stripes; i++) {
3381 stripe = btrfs_stripe_nr(chunk, i);
3382 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3389 /* [pstart, pend) */
3390 static int chunk_drange_filter(struct extent_buffer *leaf,
3391 struct btrfs_chunk *chunk,
3392 struct btrfs_balance_args *bargs)
3394 struct btrfs_stripe *stripe;
3395 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3401 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3404 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3405 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3406 factor = num_stripes / 2;
3407 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3408 factor = num_stripes - 1;
3409 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3410 factor = num_stripes - 2;
3412 factor = num_stripes;
3415 for (i = 0; i < num_stripes; i++) {
3416 stripe = btrfs_stripe_nr(chunk, i);
3417 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3420 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3421 stripe_length = btrfs_chunk_length(leaf, chunk);
3422 stripe_length = div_u64(stripe_length, factor);
3424 if (stripe_offset < bargs->pend &&
3425 stripe_offset + stripe_length > bargs->pstart)
3432 /* [vstart, vend) */
3433 static int chunk_vrange_filter(struct extent_buffer *leaf,
3434 struct btrfs_chunk *chunk,
3436 struct btrfs_balance_args *bargs)
3438 if (chunk_offset < bargs->vend &&
3439 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3440 /* at least part of the chunk is inside this vrange */
3446 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3447 struct btrfs_chunk *chunk,
3448 struct btrfs_balance_args *bargs)
3450 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3452 if (bargs->stripes_min <= num_stripes
3453 && num_stripes <= bargs->stripes_max)
3459 static int chunk_soft_convert_filter(u64 chunk_type,
3460 struct btrfs_balance_args *bargs)
3462 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3465 chunk_type = chunk_to_extended(chunk_type) &
3466 BTRFS_EXTENDED_PROFILE_MASK;
3468 if (bargs->target == chunk_type)
3474 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3475 struct extent_buffer *leaf,
3476 struct btrfs_chunk *chunk, u64 chunk_offset)
3478 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3479 struct btrfs_balance_args *bargs = NULL;
3480 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3483 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3484 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3488 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3489 bargs = &bctl->data;
3490 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3492 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3493 bargs = &bctl->meta;
3495 /* profiles filter */
3496 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3497 chunk_profiles_filter(chunk_type, bargs)) {
3502 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3503 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3505 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3506 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3511 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3512 chunk_devid_filter(leaf, chunk, bargs)) {
3516 /* drange filter, makes sense only with devid filter */
3517 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3518 chunk_drange_filter(leaf, chunk, bargs)) {
3523 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3524 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3528 /* stripes filter */
3529 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3530 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3534 /* soft profile changing mode */
3535 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3536 chunk_soft_convert_filter(chunk_type, bargs)) {
3541 * limited by count, must be the last filter
3543 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3544 if (bargs->limit == 0)
3548 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3550 * Same logic as the 'limit' filter; the minimum cannot be
3551 * determined here because we do not have the global information
3552 * about the count of all chunks that satisfy the filters.
3554 if (bargs->limit_max == 0)
3563 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3565 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3566 struct btrfs_root *chunk_root = fs_info->chunk_root;
3567 struct btrfs_root *dev_root = fs_info->dev_root;
3568 struct list_head *devices;
3569 struct btrfs_device *device;
3573 struct btrfs_chunk *chunk;
3574 struct btrfs_path *path = NULL;
3575 struct btrfs_key key;
3576 struct btrfs_key found_key;
3577 struct btrfs_trans_handle *trans;
3578 struct extent_buffer *leaf;
3581 int enospc_errors = 0;
3582 bool counting = true;
3583 /* The single value limit and min/max limits use the same bytes in the */
3584 u64 limit_data = bctl->data.limit;
3585 u64 limit_meta = bctl->meta.limit;
3586 u64 limit_sys = bctl->sys.limit;
3590 int chunk_reserved = 0;
3592 /* step one make some room on all the devices */
3593 devices = &fs_info->fs_devices->devices;
3594 list_for_each_entry(device, devices, dev_list) {
3595 old_size = btrfs_device_get_total_bytes(device);
3596 size_to_free = div_factor(old_size, 1);
3597 size_to_free = min_t(u64, size_to_free, SZ_1M);
3598 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3599 btrfs_device_get_total_bytes(device) -
3600 btrfs_device_get_bytes_used(device) > size_to_free ||
3601 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3604 ret = btrfs_shrink_device(device, old_size - size_to_free);
3608 /* btrfs_shrink_device never returns ret > 0 */
3613 trans = btrfs_start_transaction(dev_root, 0);
3614 if (IS_ERR(trans)) {
3615 ret = PTR_ERR(trans);
3616 btrfs_info_in_rcu(fs_info,
3617 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3618 rcu_str_deref(device->name), ret,
3619 old_size, old_size - size_to_free);
3623 ret = btrfs_grow_device(trans, device, old_size);
3625 btrfs_end_transaction(trans);
3626 /* btrfs_grow_device never returns ret > 0 */
3628 btrfs_info_in_rcu(fs_info,
3629 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3630 rcu_str_deref(device->name), ret,
3631 old_size, old_size - size_to_free);
3635 btrfs_end_transaction(trans);
3638 /* step two, relocate all the chunks */
3639 path = btrfs_alloc_path();
3645 /* zero out stat counters */
3646 spin_lock(&fs_info->balance_lock);
3647 memset(&bctl->stat, 0, sizeof(bctl->stat));
3648 spin_unlock(&fs_info->balance_lock);
3652 * The single value limit and min/max limits use the same bytes
3655 bctl->data.limit = limit_data;
3656 bctl->meta.limit = limit_meta;
3657 bctl->sys.limit = limit_sys;
3659 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3660 key.offset = (u64)-1;
3661 key.type = BTRFS_CHUNK_ITEM_KEY;
3664 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3665 atomic_read(&fs_info->balance_cancel_req)) {
3670 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3671 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3673 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3678 * this shouldn't happen, it means the last relocate
3682 BUG(); /* FIXME break ? */
3684 ret = btrfs_previous_item(chunk_root, path, 0,
3685 BTRFS_CHUNK_ITEM_KEY);
3687 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3692 leaf = path->nodes[0];
3693 slot = path->slots[0];
3694 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3696 if (found_key.objectid != key.objectid) {
3697 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3701 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3702 chunk_type = btrfs_chunk_type(leaf, chunk);
3705 spin_lock(&fs_info->balance_lock);
3706 bctl->stat.considered++;
3707 spin_unlock(&fs_info->balance_lock);
3710 ret = should_balance_chunk(fs_info, leaf, chunk,
3713 btrfs_release_path(path);
3715 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3720 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3721 spin_lock(&fs_info->balance_lock);
3722 bctl->stat.expected++;
3723 spin_unlock(&fs_info->balance_lock);
3725 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3727 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3729 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3736 * Apply limit_min filter, no need to check if the LIMITS
3737 * filter is used, limit_min is 0 by default
3739 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3740 count_data < bctl->data.limit_min)
3741 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3742 count_meta < bctl->meta.limit_min)
3743 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3744 count_sys < bctl->sys.limit_min)) {
3745 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3749 if (!chunk_reserved) {
3751 * We may be relocating the only data chunk we have,
3752 * which could potentially end up with losing data's
3753 * raid profile, so lets allocate an empty one in
3756 ret = btrfs_may_alloc_data_chunk(fs_info,
3759 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3761 } else if (ret == 1) {
3766 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3767 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3768 if (ret && ret != -ENOSPC)
3770 if (ret == -ENOSPC) {
3773 spin_lock(&fs_info->balance_lock);
3774 bctl->stat.completed++;
3775 spin_unlock(&fs_info->balance_lock);
3778 if (found_key.offset == 0)
3780 key.offset = found_key.offset - 1;
3784 btrfs_release_path(path);
3789 btrfs_free_path(path);
3790 if (enospc_errors) {
3791 btrfs_info(fs_info, "%d enospc errors during balance",
3801 * alloc_profile_is_valid - see if a given profile is valid and reduced
3802 * @flags: profile to validate
3803 * @extended: if true @flags is treated as an extended profile
3805 static int alloc_profile_is_valid(u64 flags, int extended)
3807 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3808 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3810 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3812 /* 1) check that all other bits are zeroed */
3816 /* 2) see if profile is reduced */
3818 return !extended; /* "0" is valid for usual profiles */
3820 /* true if exactly one bit set */
3821 return (flags & (flags - 1)) == 0;
3824 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3826 /* cancel requested || normal exit path */
3827 return atomic_read(&fs_info->balance_cancel_req) ||
3828 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3829 atomic_read(&fs_info->balance_cancel_req) == 0);
3832 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3836 unset_balance_control(fs_info);
3837 ret = del_balance_item(fs_info);
3839 btrfs_handle_fs_error(fs_info, ret, NULL);
3841 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3844 /* Non-zero return value signifies invalidity */
3845 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3848 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3849 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3850 (bctl_arg->target & ~allowed)));
3854 * Should be called with both balance and volume mutexes held
3856 int btrfs_balance(struct btrfs_balance_control *bctl,
3857 struct btrfs_ioctl_balance_args *bargs)
3859 struct btrfs_fs_info *fs_info = bctl->fs_info;
3860 u64 meta_target, data_target;
3867 if (btrfs_fs_closing(fs_info) ||
3868 atomic_read(&fs_info->balance_pause_req) ||
3869 atomic_read(&fs_info->balance_cancel_req)) {
3874 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3875 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3879 * In case of mixed groups both data and meta should be picked,
3880 * and identical options should be given for both of them.
3882 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3883 if (mixed && (bctl->flags & allowed)) {
3884 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3885 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3886 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3888 "with mixed groups data and metadata balance options must be the same");
3894 num_devices = fs_info->fs_devices->num_devices;
3895 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3896 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3897 BUG_ON(num_devices < 1);
3900 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3901 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3902 if (num_devices > 1)
3903 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3904 if (num_devices > 2)
3905 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3906 if (num_devices > 3)
3907 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3908 BTRFS_BLOCK_GROUP_RAID6);
3909 if (validate_convert_profile(&bctl->data, allowed)) {
3911 "unable to start balance with target data profile %llu",
3916 if (validate_convert_profile(&bctl->meta, allowed)) {
3918 "unable to start balance with target metadata profile %llu",
3923 if (validate_convert_profile(&bctl->sys, allowed)) {
3925 "unable to start balance with target system profile %llu",
3931 /* allow to reduce meta or sys integrity only if force set */
3932 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3933 BTRFS_BLOCK_GROUP_RAID10 |
3934 BTRFS_BLOCK_GROUP_RAID5 |
3935 BTRFS_BLOCK_GROUP_RAID6;
3937 seq = read_seqbegin(&fs_info->profiles_lock);
3939 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3940 (fs_info->avail_system_alloc_bits & allowed) &&
3941 !(bctl->sys.target & allowed)) ||
3942 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3943 (fs_info->avail_metadata_alloc_bits & allowed) &&
3944 !(bctl->meta.target & allowed))) {
3945 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3947 "force reducing metadata integrity");
3950 "balance will reduce metadata integrity, use force if you want this");
3955 } while (read_seqretry(&fs_info->profiles_lock, seq));
3957 /* if we're not converting, the target field is uninitialized */
3958 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3959 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3960 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3961 bctl->data.target : fs_info->avail_data_alloc_bits;
3962 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3963 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3965 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3966 meta_target, data_target);
3969 ret = insert_balance_item(fs_info, bctl);
3970 if (ret && ret != -EEXIST)
3973 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3974 BUG_ON(ret == -EEXIST);
3975 set_balance_control(bctl);
3977 BUG_ON(ret != -EEXIST);
3978 spin_lock(&fs_info->balance_lock);
3979 update_balance_args(bctl);
3980 spin_unlock(&fs_info->balance_lock);
3983 atomic_inc(&fs_info->balance_running);
3984 mutex_unlock(&fs_info->balance_mutex);
3986 ret = __btrfs_balance(fs_info);
3988 mutex_lock(&fs_info->balance_mutex);
3989 atomic_dec(&fs_info->balance_running);
3992 memset(bargs, 0, sizeof(*bargs));
3993 update_ioctl_balance_args(fs_info, 0, bargs);
3996 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3997 balance_need_close(fs_info)) {
3998 __cancel_balance(fs_info);
4001 wake_up(&fs_info->balance_wait_q);
4005 if (bctl->flags & BTRFS_BALANCE_RESUME)
4006 __cancel_balance(fs_info);
4009 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4014 static int balance_kthread(void *data)
4016 struct btrfs_fs_info *fs_info = data;
4019 mutex_lock(&fs_info->volume_mutex);
4020 mutex_lock(&fs_info->balance_mutex);
4022 if (fs_info->balance_ctl) {
4023 btrfs_info(fs_info, "continuing balance");
4024 ret = btrfs_balance(fs_info->balance_ctl, NULL);
4027 mutex_unlock(&fs_info->balance_mutex);
4028 mutex_unlock(&fs_info->volume_mutex);
4033 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4035 struct task_struct *tsk;
4037 spin_lock(&fs_info->balance_lock);
4038 if (!fs_info->balance_ctl) {
4039 spin_unlock(&fs_info->balance_lock);
4042 spin_unlock(&fs_info->balance_lock);
4044 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4045 btrfs_info(fs_info, "force skipping balance");
4049 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4050 return PTR_ERR_OR_ZERO(tsk);
4053 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4055 struct btrfs_balance_control *bctl;
4056 struct btrfs_balance_item *item;
4057 struct btrfs_disk_balance_args disk_bargs;
4058 struct btrfs_path *path;
4059 struct extent_buffer *leaf;
4060 struct btrfs_key key;
4063 path = btrfs_alloc_path();
4067 key.objectid = BTRFS_BALANCE_OBJECTID;
4068 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4071 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4074 if (ret > 0) { /* ret = -ENOENT; */
4079 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4085 leaf = path->nodes[0];
4086 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4088 bctl->fs_info = fs_info;
4089 bctl->flags = btrfs_balance_flags(leaf, item);
4090 bctl->flags |= BTRFS_BALANCE_RESUME;
4092 btrfs_balance_data(leaf, item, &disk_bargs);
4093 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4094 btrfs_balance_meta(leaf, item, &disk_bargs);
4095 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4096 btrfs_balance_sys(leaf, item, &disk_bargs);
4097 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4099 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4101 mutex_lock(&fs_info->volume_mutex);
4102 mutex_lock(&fs_info->balance_mutex);
4104 set_balance_control(bctl);
4106 mutex_unlock(&fs_info->balance_mutex);
4107 mutex_unlock(&fs_info->volume_mutex);
4109 btrfs_free_path(path);
4113 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4117 mutex_lock(&fs_info->balance_mutex);
4118 if (!fs_info->balance_ctl) {
4119 mutex_unlock(&fs_info->balance_mutex);
4123 if (atomic_read(&fs_info->balance_running)) {
4124 atomic_inc(&fs_info->balance_pause_req);
4125 mutex_unlock(&fs_info->balance_mutex);
4127 wait_event(fs_info->balance_wait_q,
4128 atomic_read(&fs_info->balance_running) == 0);
4130 mutex_lock(&fs_info->balance_mutex);
4131 /* we are good with balance_ctl ripped off from under us */
4132 BUG_ON(atomic_read(&fs_info->balance_running));
4133 atomic_dec(&fs_info->balance_pause_req);
4138 mutex_unlock(&fs_info->balance_mutex);
4142 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4144 if (sb_rdonly(fs_info->sb))
4147 mutex_lock(&fs_info->balance_mutex);
4148 if (!fs_info->balance_ctl) {
4149 mutex_unlock(&fs_info->balance_mutex);
4153 atomic_inc(&fs_info->balance_cancel_req);
4155 * if we are running just wait and return, balance item is
4156 * deleted in btrfs_balance in this case
4158 if (atomic_read(&fs_info->balance_running)) {
4159 mutex_unlock(&fs_info->balance_mutex);
4160 wait_event(fs_info->balance_wait_q,
4161 atomic_read(&fs_info->balance_running) == 0);
4162 mutex_lock(&fs_info->balance_mutex);
4164 /* __cancel_balance needs volume_mutex */
4165 mutex_unlock(&fs_info->balance_mutex);
4166 mutex_lock(&fs_info->volume_mutex);
4167 mutex_lock(&fs_info->balance_mutex);
4169 if (fs_info->balance_ctl)
4170 __cancel_balance(fs_info);
4172 mutex_unlock(&fs_info->volume_mutex);
4175 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4176 atomic_dec(&fs_info->balance_cancel_req);
4177 mutex_unlock(&fs_info->balance_mutex);
4181 static int btrfs_uuid_scan_kthread(void *data)
4183 struct btrfs_fs_info *fs_info = data;
4184 struct btrfs_root *root = fs_info->tree_root;
4185 struct btrfs_key key;
4186 struct btrfs_path *path = NULL;
4188 struct extent_buffer *eb;
4190 struct btrfs_root_item root_item;
4192 struct btrfs_trans_handle *trans = NULL;
4194 path = btrfs_alloc_path();
4201 key.type = BTRFS_ROOT_ITEM_KEY;
4205 ret = btrfs_search_forward(root, &key, path, 0);
4212 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4213 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4214 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4215 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4218 eb = path->nodes[0];
4219 slot = path->slots[0];
4220 item_size = btrfs_item_size_nr(eb, slot);
4221 if (item_size < sizeof(root_item))
4224 read_extent_buffer(eb, &root_item,
4225 btrfs_item_ptr_offset(eb, slot),
4226 (int)sizeof(root_item));
4227 if (btrfs_root_refs(&root_item) == 0)
4230 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4231 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4235 btrfs_release_path(path);
4237 * 1 - subvol uuid item
4238 * 1 - received_subvol uuid item
4240 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4241 if (IS_ERR(trans)) {
4242 ret = PTR_ERR(trans);
4250 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4251 ret = btrfs_uuid_tree_add(trans, fs_info,
4253 BTRFS_UUID_KEY_SUBVOL,
4256 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4262 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4263 ret = btrfs_uuid_tree_add(trans, fs_info,
4264 root_item.received_uuid,
4265 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4268 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4276 ret = btrfs_end_transaction(trans);
4282 btrfs_release_path(path);
4283 if (key.offset < (u64)-1) {
4285 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4287 key.type = BTRFS_ROOT_ITEM_KEY;
4288 } else if (key.objectid < (u64)-1) {
4290 key.type = BTRFS_ROOT_ITEM_KEY;
4299 btrfs_free_path(path);
4300 if (trans && !IS_ERR(trans))
4301 btrfs_end_transaction(trans);
4303 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4305 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4306 up(&fs_info->uuid_tree_rescan_sem);
4311 * Callback for btrfs_uuid_tree_iterate().
4313 * 0 check succeeded, the entry is not outdated.
4314 * < 0 if an error occurred.
4315 * > 0 if the check failed, which means the caller shall remove the entry.
4317 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4318 u8 *uuid, u8 type, u64 subid)
4320 struct btrfs_key key;
4322 struct btrfs_root *subvol_root;
4324 if (type != BTRFS_UUID_KEY_SUBVOL &&
4325 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4328 key.objectid = subid;
4329 key.type = BTRFS_ROOT_ITEM_KEY;
4330 key.offset = (u64)-1;
4331 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4332 if (IS_ERR(subvol_root)) {
4333 ret = PTR_ERR(subvol_root);
4340 case BTRFS_UUID_KEY_SUBVOL:
4341 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4344 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4345 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4355 static int btrfs_uuid_rescan_kthread(void *data)
4357 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4361 * 1st step is to iterate through the existing UUID tree and
4362 * to delete all entries that contain outdated data.
4363 * 2nd step is to add all missing entries to the UUID tree.
4365 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4367 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4368 up(&fs_info->uuid_tree_rescan_sem);
4371 return btrfs_uuid_scan_kthread(data);
4374 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4376 struct btrfs_trans_handle *trans;
4377 struct btrfs_root *tree_root = fs_info->tree_root;
4378 struct btrfs_root *uuid_root;
4379 struct task_struct *task;
4386 trans = btrfs_start_transaction(tree_root, 2);
4388 return PTR_ERR(trans);
4390 uuid_root = btrfs_create_tree(trans, fs_info,
4391 BTRFS_UUID_TREE_OBJECTID);
4392 if (IS_ERR(uuid_root)) {
4393 ret = PTR_ERR(uuid_root);
4394 btrfs_abort_transaction(trans, ret);
4395 btrfs_end_transaction(trans);
4399 fs_info->uuid_root = uuid_root;
4401 ret = btrfs_commit_transaction(trans);
4405 down(&fs_info->uuid_tree_rescan_sem);
4406 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4408 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4409 btrfs_warn(fs_info, "failed to start uuid_scan task");
4410 up(&fs_info->uuid_tree_rescan_sem);
4411 return PTR_ERR(task);
4417 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4419 struct task_struct *task;
4421 down(&fs_info->uuid_tree_rescan_sem);
4422 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4424 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4425 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4426 up(&fs_info->uuid_tree_rescan_sem);
4427 return PTR_ERR(task);
4434 * shrinking a device means finding all of the device extents past
4435 * the new size, and then following the back refs to the chunks.
4436 * The chunk relocation code actually frees the device extent
4438 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4440 struct btrfs_fs_info *fs_info = device->fs_info;
4441 struct btrfs_root *root = fs_info->dev_root;
4442 struct btrfs_trans_handle *trans;
4443 struct btrfs_dev_extent *dev_extent = NULL;
4444 struct btrfs_path *path;
4450 bool retried = false;
4451 bool checked_pending_chunks = false;
4452 struct extent_buffer *l;
4453 struct btrfs_key key;
4454 struct btrfs_super_block *super_copy = fs_info->super_copy;
4455 u64 old_total = btrfs_super_total_bytes(super_copy);
4456 u64 old_size = btrfs_device_get_total_bytes(device);
4459 new_size = round_down(new_size, fs_info->sectorsize);
4460 diff = round_down(old_size - new_size, fs_info->sectorsize);
4462 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4465 path = btrfs_alloc_path();
4469 path->reada = READA_FORWARD;
4471 mutex_lock(&fs_info->chunk_mutex);
4473 btrfs_device_set_total_bytes(device, new_size);
4474 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4475 device->fs_devices->total_rw_bytes -= diff;
4476 atomic64_sub(diff, &fs_info->free_chunk_space);
4478 mutex_unlock(&fs_info->chunk_mutex);
4481 key.objectid = device->devid;
4482 key.offset = (u64)-1;
4483 key.type = BTRFS_DEV_EXTENT_KEY;
4486 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4487 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4489 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4493 ret = btrfs_previous_item(root, path, 0, key.type);
4495 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4500 btrfs_release_path(path);
4505 slot = path->slots[0];
4506 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4508 if (key.objectid != device->devid) {
4509 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4510 btrfs_release_path(path);
4514 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4515 length = btrfs_dev_extent_length(l, dev_extent);
4517 if (key.offset + length <= new_size) {
4518 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4519 btrfs_release_path(path);
4523 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4524 btrfs_release_path(path);
4527 * We may be relocating the only data chunk we have,
4528 * which could potentially end up with losing data's
4529 * raid profile, so lets allocate an empty one in
4532 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4534 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4538 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4539 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4540 if (ret && ret != -ENOSPC)
4544 } while (key.offset-- > 0);
4546 if (failed && !retried) {
4550 } else if (failed && retried) {
4555 /* Shrinking succeeded, else we would be at "done". */
4556 trans = btrfs_start_transaction(root, 0);
4557 if (IS_ERR(trans)) {
4558 ret = PTR_ERR(trans);
4562 mutex_lock(&fs_info->chunk_mutex);
4565 * We checked in the above loop all device extents that were already in
4566 * the device tree. However before we have updated the device's
4567 * total_bytes to the new size, we might have had chunk allocations that
4568 * have not complete yet (new block groups attached to transaction
4569 * handles), and therefore their device extents were not yet in the
4570 * device tree and we missed them in the loop above. So if we have any
4571 * pending chunk using a device extent that overlaps the device range
4572 * that we can not use anymore, commit the current transaction and
4573 * repeat the search on the device tree - this way we guarantee we will
4574 * not have chunks using device extents that end beyond 'new_size'.
4576 if (!checked_pending_chunks) {
4577 u64 start = new_size;
4578 u64 len = old_size - new_size;
4580 if (contains_pending_extent(trans->transaction, device,
4582 mutex_unlock(&fs_info->chunk_mutex);
4583 checked_pending_chunks = true;
4586 ret = btrfs_commit_transaction(trans);
4593 btrfs_device_set_disk_total_bytes(device, new_size);
4594 if (list_empty(&device->resized_list))
4595 list_add_tail(&device->resized_list,
4596 &fs_info->fs_devices->resized_devices);
4598 WARN_ON(diff > old_total);
4599 btrfs_set_super_total_bytes(super_copy,
4600 round_down(old_total - diff, fs_info->sectorsize));
4601 mutex_unlock(&fs_info->chunk_mutex);
4603 /* Now btrfs_update_device() will change the on-disk size. */
4604 ret = btrfs_update_device(trans, device);
4605 btrfs_end_transaction(trans);
4607 btrfs_free_path(path);
4609 mutex_lock(&fs_info->chunk_mutex);
4610 btrfs_device_set_total_bytes(device, old_size);
4611 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4612 device->fs_devices->total_rw_bytes += diff;
4613 atomic64_add(diff, &fs_info->free_chunk_space);
4614 mutex_unlock(&fs_info->chunk_mutex);
4619 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4620 struct btrfs_key *key,
4621 struct btrfs_chunk *chunk, int item_size)
4623 struct btrfs_super_block *super_copy = fs_info->super_copy;
4624 struct btrfs_disk_key disk_key;
4628 mutex_lock(&fs_info->chunk_mutex);
4629 array_size = btrfs_super_sys_array_size(super_copy);
4630 if (array_size + item_size + sizeof(disk_key)
4631 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4632 mutex_unlock(&fs_info->chunk_mutex);
4636 ptr = super_copy->sys_chunk_array + array_size;
4637 btrfs_cpu_key_to_disk(&disk_key, key);
4638 memcpy(ptr, &disk_key, sizeof(disk_key));
4639 ptr += sizeof(disk_key);
4640 memcpy(ptr, chunk, item_size);
4641 item_size += sizeof(disk_key);
4642 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4643 mutex_unlock(&fs_info->chunk_mutex);
4649 * sort the devices in descending order by max_avail, total_avail
4651 static int btrfs_cmp_device_info(const void *a, const void *b)
4653 const struct btrfs_device_info *di_a = a;
4654 const struct btrfs_device_info *di_b = b;
4656 if (di_a->max_avail > di_b->max_avail)
4658 if (di_a->max_avail < di_b->max_avail)
4660 if (di_a->total_avail > di_b->total_avail)
4662 if (di_a->total_avail < di_b->total_avail)
4667 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4669 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4672 btrfs_set_fs_incompat(info, RAID56);
4675 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4676 - sizeof(struct btrfs_chunk)) \
4677 / sizeof(struct btrfs_stripe) + 1)
4679 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4680 - 2 * sizeof(struct btrfs_disk_key) \
4681 - 2 * sizeof(struct btrfs_chunk)) \
4682 / sizeof(struct btrfs_stripe) + 1)
4684 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4685 u64 start, u64 type)
4687 struct btrfs_fs_info *info = trans->fs_info;
4688 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4689 struct btrfs_device *device;
4690 struct map_lookup *map = NULL;
4691 struct extent_map_tree *em_tree;
4692 struct extent_map *em;
4693 struct btrfs_device_info *devices_info = NULL;
4695 int num_stripes; /* total number of stripes to allocate */
4696 int data_stripes; /* number of stripes that count for
4698 int sub_stripes; /* sub_stripes info for map */
4699 int dev_stripes; /* stripes per dev */
4700 int devs_max; /* max devs to use */
4701 int devs_min; /* min devs needed */
4702 int devs_increment; /* ndevs has to be a multiple of this */
4703 int ncopies; /* how many copies to data has */
4705 u64 max_stripe_size;
4714 BUG_ON(!alloc_profile_is_valid(type, 0));
4716 if (list_empty(&fs_devices->alloc_list))
4719 index = __get_raid_index(type);
4721 sub_stripes = btrfs_raid_array[index].sub_stripes;
4722 dev_stripes = btrfs_raid_array[index].dev_stripes;
4723 devs_max = btrfs_raid_array[index].devs_max;
4724 devs_min = btrfs_raid_array[index].devs_min;
4725 devs_increment = btrfs_raid_array[index].devs_increment;
4726 ncopies = btrfs_raid_array[index].ncopies;
4728 if (type & BTRFS_BLOCK_GROUP_DATA) {
4729 max_stripe_size = SZ_1G;
4730 max_chunk_size = 10 * max_stripe_size;
4732 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4733 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4734 /* for larger filesystems, use larger metadata chunks */
4735 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4736 max_stripe_size = SZ_1G;
4738 max_stripe_size = SZ_256M;
4739 max_chunk_size = max_stripe_size;
4741 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4742 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4743 max_stripe_size = SZ_32M;
4744 max_chunk_size = 2 * max_stripe_size;
4746 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4748 btrfs_err(info, "invalid chunk type 0x%llx requested",
4753 /* we don't want a chunk larger than 10% of writeable space */
4754 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4757 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4763 * in the first pass through the devices list, we gather information
4764 * about the available holes on each device.
4767 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4771 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4773 "BTRFS: read-only device in alloc_list\n");
4777 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4778 &device->dev_state) ||
4779 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4782 if (device->total_bytes > device->bytes_used)
4783 total_avail = device->total_bytes - device->bytes_used;
4787 /* If there is no space on this device, skip it. */
4788 if (total_avail == 0)
4791 ret = find_free_dev_extent(trans, device,
4792 max_stripe_size * dev_stripes,
4793 &dev_offset, &max_avail);
4794 if (ret && ret != -ENOSPC)
4798 max_avail = max_stripe_size * dev_stripes;
4800 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4803 if (ndevs == fs_devices->rw_devices) {
4804 WARN(1, "%s: found more than %llu devices\n",
4805 __func__, fs_devices->rw_devices);
4808 devices_info[ndevs].dev_offset = dev_offset;
4809 devices_info[ndevs].max_avail = max_avail;
4810 devices_info[ndevs].total_avail = total_avail;
4811 devices_info[ndevs].dev = device;
4816 * now sort the devices by hole size / available space
4818 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4819 btrfs_cmp_device_info, NULL);
4821 /* round down to number of usable stripes */
4822 ndevs = round_down(ndevs, devs_increment);
4824 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4829 ndevs = min(ndevs, devs_max);
4832 * The primary goal is to maximize the number of stripes, so use as
4833 * many devices as possible, even if the stripes are not maximum sized.
4835 * The DUP profile stores more than one stripe per device, the
4836 * max_avail is the total size so we have to adjust.
4838 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4839 num_stripes = ndevs * dev_stripes;
4842 * this will have to be fixed for RAID1 and RAID10 over
4845 data_stripes = num_stripes / ncopies;
4847 if (type & BTRFS_BLOCK_GROUP_RAID5)
4848 data_stripes = num_stripes - 1;
4850 if (type & BTRFS_BLOCK_GROUP_RAID6)
4851 data_stripes = num_stripes - 2;
4854 * Use the number of data stripes to figure out how big this chunk
4855 * is really going to be in terms of logical address space,
4856 * and compare that answer with the max chunk size
4858 if (stripe_size * data_stripes > max_chunk_size) {
4859 u64 mask = (1ULL << 24) - 1;
4861 stripe_size = div_u64(max_chunk_size, data_stripes);
4863 /* bump the answer up to a 16MB boundary */
4864 stripe_size = (stripe_size + mask) & ~mask;
4866 /* but don't go higher than the limits we found
4867 * while searching for free extents
4869 if (stripe_size > devices_info[ndevs-1].max_avail)
4870 stripe_size = devices_info[ndevs-1].max_avail;
4873 /* align to BTRFS_STRIPE_LEN */
4874 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4876 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4881 map->num_stripes = num_stripes;
4883 for (i = 0; i < ndevs; ++i) {
4884 for (j = 0; j < dev_stripes; ++j) {
4885 int s = i * dev_stripes + j;
4886 map->stripes[s].dev = devices_info[i].dev;
4887 map->stripes[s].physical = devices_info[i].dev_offset +
4891 map->stripe_len = BTRFS_STRIPE_LEN;
4892 map->io_align = BTRFS_STRIPE_LEN;
4893 map->io_width = BTRFS_STRIPE_LEN;
4895 map->sub_stripes = sub_stripes;
4897 num_bytes = stripe_size * data_stripes;
4899 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4901 em = alloc_extent_map();
4907 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4908 em->map_lookup = map;
4910 em->len = num_bytes;
4911 em->block_start = 0;
4912 em->block_len = em->len;
4913 em->orig_block_len = stripe_size;
4915 em_tree = &info->mapping_tree.map_tree;
4916 write_lock(&em_tree->lock);
4917 ret = add_extent_mapping(em_tree, em, 0);
4919 write_unlock(&em_tree->lock);
4920 free_extent_map(em);
4924 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4925 refcount_inc(&em->refs);
4926 write_unlock(&em_tree->lock);
4928 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4930 goto error_del_extent;
4932 for (i = 0; i < map->num_stripes; i++) {
4933 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4934 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4937 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4939 free_extent_map(em);
4940 check_raid56_incompat_flag(info, type);
4942 kfree(devices_info);
4946 write_lock(&em_tree->lock);
4947 remove_extent_mapping(em_tree, em);
4948 write_unlock(&em_tree->lock);
4950 /* One for our allocation */
4951 free_extent_map(em);
4952 /* One for the tree reference */
4953 free_extent_map(em);
4954 /* One for the pending_chunks list reference */
4955 free_extent_map(em);
4957 kfree(devices_info);
4961 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4962 struct btrfs_fs_info *fs_info,
4963 u64 chunk_offset, u64 chunk_size)
4965 struct btrfs_root *extent_root = fs_info->extent_root;
4966 struct btrfs_root *chunk_root = fs_info->chunk_root;
4967 struct btrfs_key key;
4968 struct btrfs_device *device;
4969 struct btrfs_chunk *chunk;
4970 struct btrfs_stripe *stripe;
4971 struct extent_map *em;
4972 struct map_lookup *map;
4979 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4983 map = em->map_lookup;
4984 item_size = btrfs_chunk_item_size(map->num_stripes);
4985 stripe_size = em->orig_block_len;
4987 chunk = kzalloc(item_size, GFP_NOFS);
4994 * Take the device list mutex to prevent races with the final phase of
4995 * a device replace operation that replaces the device object associated
4996 * with the map's stripes, because the device object's id can change
4997 * at any time during that final phase of the device replace operation
4998 * (dev-replace.c:btrfs_dev_replace_finishing()).
5000 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5001 for (i = 0; i < map->num_stripes; i++) {
5002 device = map->stripes[i].dev;
5003 dev_offset = map->stripes[i].physical;
5005 ret = btrfs_update_device(trans, device);
5008 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5009 dev_offset, stripe_size);
5014 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5018 stripe = &chunk->stripe;
5019 for (i = 0; i < map->num_stripes; i++) {
5020 device = map->stripes[i].dev;
5021 dev_offset = map->stripes[i].physical;
5023 btrfs_set_stack_stripe_devid(stripe, device->devid);
5024 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5025 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5028 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5030 btrfs_set_stack_chunk_length(chunk, chunk_size);
5031 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5032 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5033 btrfs_set_stack_chunk_type(chunk, map->type);
5034 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5035 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5036 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5037 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5038 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5040 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5041 key.type = BTRFS_CHUNK_ITEM_KEY;
5042 key.offset = chunk_offset;
5044 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5045 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5047 * TODO: Cleanup of inserted chunk root in case of
5050 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5055 free_extent_map(em);
5060 * Chunk allocation falls into two parts. The first part does works
5061 * that make the new allocated chunk useable, but not do any operation
5062 * that modifies the chunk tree. The second part does the works that
5063 * require modifying the chunk tree. This division is important for the
5064 * bootstrap process of adding storage to a seed btrfs.
5066 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5067 struct btrfs_fs_info *fs_info, u64 type)
5071 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5072 chunk_offset = find_next_chunk(fs_info);
5073 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5076 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5077 struct btrfs_fs_info *fs_info)
5080 u64 sys_chunk_offset;
5084 chunk_offset = find_next_chunk(fs_info);
5085 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5086 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5090 sys_chunk_offset = find_next_chunk(fs_info);
5091 alloc_profile = btrfs_system_alloc_profile(fs_info);
5092 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5096 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5100 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5101 BTRFS_BLOCK_GROUP_RAID10 |
5102 BTRFS_BLOCK_GROUP_RAID5 |
5103 BTRFS_BLOCK_GROUP_DUP)) {
5105 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5114 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5116 struct extent_map *em;
5117 struct map_lookup *map;
5122 em = get_chunk_map(fs_info, chunk_offset, 1);
5126 map = em->map_lookup;
5127 for (i = 0; i < map->num_stripes; i++) {
5128 if (test_bit(BTRFS_DEV_STATE_MISSING,
5129 &map->stripes[i].dev->dev_state)) {
5133 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5134 &map->stripes[i].dev->dev_state)) {
5141 * If the number of missing devices is larger than max errors,
5142 * we can not write the data into that chunk successfully, so
5145 if (miss_ndevs > btrfs_chunk_max_errors(map))
5148 free_extent_map(em);
5152 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5154 extent_map_tree_init(&tree->map_tree);
5157 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5159 struct extent_map *em;
5162 write_lock(&tree->map_tree.lock);
5163 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5165 remove_extent_mapping(&tree->map_tree, em);
5166 write_unlock(&tree->map_tree.lock);
5170 free_extent_map(em);
5171 /* once for the tree */
5172 free_extent_map(em);
5176 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5178 struct extent_map *em;
5179 struct map_lookup *map;
5182 em = get_chunk_map(fs_info, logical, len);
5185 * We could return errors for these cases, but that could get
5186 * ugly and we'd probably do the same thing which is just not do
5187 * anything else and exit, so return 1 so the callers don't try
5188 * to use other copies.
5192 map = em->map_lookup;
5193 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5194 ret = map->num_stripes;
5195 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5196 ret = map->sub_stripes;
5197 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5199 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5201 * There could be two corrupted data stripes, we need
5202 * to loop retry in order to rebuild the correct data.
5204 * Fail a stripe at a time on every retry except the
5205 * stripe under reconstruction.
5207 ret = map->num_stripes;
5210 free_extent_map(em);
5212 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5213 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5214 fs_info->dev_replace.tgtdev)
5216 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5221 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5224 struct extent_map *em;
5225 struct map_lookup *map;
5226 unsigned long len = fs_info->sectorsize;
5228 em = get_chunk_map(fs_info, logical, len);
5230 if (!WARN_ON(IS_ERR(em))) {
5231 map = em->map_lookup;
5232 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5233 len = map->stripe_len * nr_data_stripes(map);
5234 free_extent_map(em);
5239 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5241 struct extent_map *em;
5242 struct map_lookup *map;
5245 em = get_chunk_map(fs_info, logical, len);
5247 if(!WARN_ON(IS_ERR(em))) {
5248 map = em->map_lookup;
5249 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5251 free_extent_map(em);
5256 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5257 struct map_lookup *map, int first, int num,
5258 int optimal, int dev_replace_is_ongoing)
5262 struct btrfs_device *srcdev;
5264 if (dev_replace_is_ongoing &&
5265 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5266 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5267 srcdev = fs_info->dev_replace.srcdev;
5272 * try to avoid the drive that is the source drive for a
5273 * dev-replace procedure, only choose it if no other non-missing
5274 * mirror is available
5276 for (tolerance = 0; tolerance < 2; tolerance++) {
5277 if (map->stripes[optimal].dev->bdev &&
5278 (tolerance || map->stripes[optimal].dev != srcdev))
5280 for (i = first; i < first + num; i++) {
5281 if (map->stripes[i].dev->bdev &&
5282 (tolerance || map->stripes[i].dev != srcdev))
5287 /* we couldn't find one that doesn't fail. Just return something
5288 * and the io error handling code will clean up eventually
5293 static inline int parity_smaller(u64 a, u64 b)
5298 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5299 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5301 struct btrfs_bio_stripe s;
5308 for (i = 0; i < num_stripes - 1; i++) {
5309 if (parity_smaller(bbio->raid_map[i],
5310 bbio->raid_map[i+1])) {
5311 s = bbio->stripes[i];
5312 l = bbio->raid_map[i];
5313 bbio->stripes[i] = bbio->stripes[i+1];
5314 bbio->raid_map[i] = bbio->raid_map[i+1];
5315 bbio->stripes[i+1] = s;
5316 bbio->raid_map[i+1] = l;
5324 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5326 struct btrfs_bio *bbio = kzalloc(
5327 /* the size of the btrfs_bio */
5328 sizeof(struct btrfs_bio) +
5329 /* plus the variable array for the stripes */
5330 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5331 /* plus the variable array for the tgt dev */
5332 sizeof(int) * (real_stripes) +
5334 * plus the raid_map, which includes both the tgt dev
5337 sizeof(u64) * (total_stripes),
5338 GFP_NOFS|__GFP_NOFAIL);
5340 atomic_set(&bbio->error, 0);
5341 refcount_set(&bbio->refs, 1);
5346 void btrfs_get_bbio(struct btrfs_bio *bbio)
5348 WARN_ON(!refcount_read(&bbio->refs));
5349 refcount_inc(&bbio->refs);
5352 void btrfs_put_bbio(struct btrfs_bio *bbio)
5356 if (refcount_dec_and_test(&bbio->refs))
5360 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5362 * Please note that, discard won't be sent to target device of device
5365 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5366 u64 logical, u64 length,
5367 struct btrfs_bio **bbio_ret)
5369 struct extent_map *em;
5370 struct map_lookup *map;
5371 struct btrfs_bio *bbio;
5375 u64 stripe_end_offset;
5382 u32 sub_stripes = 0;
5383 u64 stripes_per_dev = 0;
5384 u32 remaining_stripes = 0;
5385 u32 last_stripe = 0;
5389 /* discard always return a bbio */
5392 em = get_chunk_map(fs_info, logical, length);
5396 map = em->map_lookup;
5397 /* we don't discard raid56 yet */
5398 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5403 offset = logical - em->start;
5404 length = min_t(u64, em->len - offset, length);
5406 stripe_len = map->stripe_len;
5408 * stripe_nr counts the total number of stripes we have to stride
5409 * to get to this block
5411 stripe_nr = div64_u64(offset, stripe_len);
5413 /* stripe_offset is the offset of this block in its stripe */
5414 stripe_offset = offset - stripe_nr * stripe_len;
5416 stripe_nr_end = round_up(offset + length, map->stripe_len);
5417 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5418 stripe_cnt = stripe_nr_end - stripe_nr;
5419 stripe_end_offset = stripe_nr_end * map->stripe_len -
5422 * after this, stripe_nr is the number of stripes on this
5423 * device we have to walk to find the data, and stripe_index is
5424 * the number of our device in the stripe array
5428 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5429 BTRFS_BLOCK_GROUP_RAID10)) {
5430 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5433 sub_stripes = map->sub_stripes;
5435 factor = map->num_stripes / sub_stripes;
5436 num_stripes = min_t(u64, map->num_stripes,
5437 sub_stripes * stripe_cnt);
5438 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5439 stripe_index *= sub_stripes;
5440 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5441 &remaining_stripes);
5442 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5443 last_stripe *= sub_stripes;
5444 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5445 BTRFS_BLOCK_GROUP_DUP)) {
5446 num_stripes = map->num_stripes;
5448 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5452 bbio = alloc_btrfs_bio(num_stripes, 0);
5458 for (i = 0; i < num_stripes; i++) {
5459 bbio->stripes[i].physical =
5460 map->stripes[stripe_index].physical +
5461 stripe_offset + stripe_nr * map->stripe_len;
5462 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5464 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5465 BTRFS_BLOCK_GROUP_RAID10)) {
5466 bbio->stripes[i].length = stripes_per_dev *
5469 if (i / sub_stripes < remaining_stripes)
5470 bbio->stripes[i].length +=
5474 * Special for the first stripe and
5477 * |-------|...|-------|
5481 if (i < sub_stripes)
5482 bbio->stripes[i].length -=
5485 if (stripe_index >= last_stripe &&
5486 stripe_index <= (last_stripe +
5488 bbio->stripes[i].length -=
5491 if (i == sub_stripes - 1)
5494 bbio->stripes[i].length = length;
5498 if (stripe_index == map->num_stripes) {
5505 bbio->map_type = map->type;
5506 bbio->num_stripes = num_stripes;
5508 free_extent_map(em);
5513 * In dev-replace case, for repair case (that's the only case where the mirror
5514 * is selected explicitly when calling btrfs_map_block), blocks left of the
5515 * left cursor can also be read from the target drive.
5517 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5519 * For READ, it also needs to be supported using the same mirror number.
5521 * If the requested block is not left of the left cursor, EIO is returned. This
5522 * can happen because btrfs_num_copies() returns one more in the dev-replace
5525 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5526 u64 logical, u64 length,
5527 u64 srcdev_devid, int *mirror_num,
5530 struct btrfs_bio *bbio = NULL;
5532 int index_srcdev = 0;
5534 u64 physical_of_found = 0;
5538 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5539 logical, &length, &bbio, 0, 0);
5541 ASSERT(bbio == NULL);
5545 num_stripes = bbio->num_stripes;
5546 if (*mirror_num > num_stripes) {
5548 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5549 * that means that the requested area is not left of the left
5552 btrfs_put_bbio(bbio);
5557 * process the rest of the function using the mirror_num of the source
5558 * drive. Therefore look it up first. At the end, patch the device
5559 * pointer to the one of the target drive.
5561 for (i = 0; i < num_stripes; i++) {
5562 if (bbio->stripes[i].dev->devid != srcdev_devid)
5566 * In case of DUP, in order to keep it simple, only add the
5567 * mirror with the lowest physical address
5570 physical_of_found <= bbio->stripes[i].physical)
5575 physical_of_found = bbio->stripes[i].physical;
5578 btrfs_put_bbio(bbio);
5584 *mirror_num = index_srcdev + 1;
5585 *physical = physical_of_found;
5589 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5590 struct btrfs_bio **bbio_ret,
5591 struct btrfs_dev_replace *dev_replace,
5592 int *num_stripes_ret, int *max_errors_ret)
5594 struct btrfs_bio *bbio = *bbio_ret;
5595 u64 srcdev_devid = dev_replace->srcdev->devid;
5596 int tgtdev_indexes = 0;
5597 int num_stripes = *num_stripes_ret;
5598 int max_errors = *max_errors_ret;
5601 if (op == BTRFS_MAP_WRITE) {
5602 int index_where_to_add;
5605 * duplicate the write operations while the dev replace
5606 * procedure is running. Since the copying of the old disk to
5607 * the new disk takes place at run time while the filesystem is
5608 * mounted writable, the regular write operations to the old
5609 * disk have to be duplicated to go to the new disk as well.
5611 * Note that device->missing is handled by the caller, and that
5612 * the write to the old disk is already set up in the stripes
5615 index_where_to_add = num_stripes;
5616 for (i = 0; i < num_stripes; i++) {
5617 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5618 /* write to new disk, too */
5619 struct btrfs_bio_stripe *new =
5620 bbio->stripes + index_where_to_add;
5621 struct btrfs_bio_stripe *old =
5624 new->physical = old->physical;
5625 new->length = old->length;
5626 new->dev = dev_replace->tgtdev;
5627 bbio->tgtdev_map[i] = index_where_to_add;
5628 index_where_to_add++;
5633 num_stripes = index_where_to_add;
5634 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5635 int index_srcdev = 0;
5637 u64 physical_of_found = 0;
5640 * During the dev-replace procedure, the target drive can also
5641 * be used to read data in case it is needed to repair a corrupt
5642 * block elsewhere. This is possible if the requested area is
5643 * left of the left cursor. In this area, the target drive is a
5644 * full copy of the source drive.
5646 for (i = 0; i < num_stripes; i++) {
5647 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5649 * In case of DUP, in order to keep it simple,
5650 * only add the mirror with the lowest physical
5654 physical_of_found <=
5655 bbio->stripes[i].physical)
5659 physical_of_found = bbio->stripes[i].physical;
5663 struct btrfs_bio_stripe *tgtdev_stripe =
5664 bbio->stripes + num_stripes;
5666 tgtdev_stripe->physical = physical_of_found;
5667 tgtdev_stripe->length =
5668 bbio->stripes[index_srcdev].length;
5669 tgtdev_stripe->dev = dev_replace->tgtdev;
5670 bbio->tgtdev_map[index_srcdev] = num_stripes;
5677 *num_stripes_ret = num_stripes;
5678 *max_errors_ret = max_errors;
5679 bbio->num_tgtdevs = tgtdev_indexes;
5683 static bool need_full_stripe(enum btrfs_map_op op)
5685 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5688 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5689 enum btrfs_map_op op,
5690 u64 logical, u64 *length,
5691 struct btrfs_bio **bbio_ret,
5692 int mirror_num, int need_raid_map)
5694 struct extent_map *em;
5695 struct map_lookup *map;
5705 int tgtdev_indexes = 0;
5706 struct btrfs_bio *bbio = NULL;
5707 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5708 int dev_replace_is_ongoing = 0;
5709 int num_alloc_stripes;
5710 int patch_the_first_stripe_for_dev_replace = 0;
5711 u64 physical_to_patch_in_first_stripe = 0;
5712 u64 raid56_full_stripe_start = (u64)-1;
5714 if (op == BTRFS_MAP_DISCARD)
5715 return __btrfs_map_block_for_discard(fs_info, logical,
5718 em = get_chunk_map(fs_info, logical, *length);
5722 map = em->map_lookup;
5723 offset = logical - em->start;
5725 stripe_len = map->stripe_len;
5728 * stripe_nr counts the total number of stripes we have to stride
5729 * to get to this block
5731 stripe_nr = div64_u64(stripe_nr, stripe_len);
5733 stripe_offset = stripe_nr * stripe_len;
5734 if (offset < stripe_offset) {
5736 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5737 stripe_offset, offset, em->start, logical,
5739 free_extent_map(em);
5743 /* stripe_offset is the offset of this block in its stripe*/
5744 stripe_offset = offset - stripe_offset;
5746 /* if we're here for raid56, we need to know the stripe aligned start */
5747 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5748 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5749 raid56_full_stripe_start = offset;
5751 /* allow a write of a full stripe, but make sure we don't
5752 * allow straddling of stripes
5754 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5756 raid56_full_stripe_start *= full_stripe_len;
5759 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5761 /* For writes to RAID[56], allow a full stripeset across all disks.
5762 For other RAID types and for RAID[56] reads, just allow a single
5763 stripe (on a single disk). */
5764 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5765 (op == BTRFS_MAP_WRITE)) {
5766 max_len = stripe_len * nr_data_stripes(map) -
5767 (offset - raid56_full_stripe_start);
5769 /* we limit the length of each bio to what fits in a stripe */
5770 max_len = stripe_len - stripe_offset;
5772 *length = min_t(u64, em->len - offset, max_len);
5774 *length = em->len - offset;
5777 /* This is for when we're called from btrfs_merge_bio_hook() and all
5778 it cares about is the length */
5782 btrfs_dev_replace_lock(dev_replace, 0);
5783 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5784 if (!dev_replace_is_ongoing)
5785 btrfs_dev_replace_unlock(dev_replace, 0);
5787 btrfs_dev_replace_set_lock_blocking(dev_replace);
5789 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5790 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5791 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5792 dev_replace->srcdev->devid,
5794 &physical_to_patch_in_first_stripe);
5798 patch_the_first_stripe_for_dev_replace = 1;
5799 } else if (mirror_num > map->num_stripes) {
5805 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5806 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5808 if (!need_full_stripe(op))
5810 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5811 if (need_full_stripe(op))
5812 num_stripes = map->num_stripes;
5813 else if (mirror_num)
5814 stripe_index = mirror_num - 1;
5816 stripe_index = find_live_mirror(fs_info, map, 0,
5818 current->pid % map->num_stripes,
5819 dev_replace_is_ongoing);
5820 mirror_num = stripe_index + 1;
5823 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5824 if (need_full_stripe(op)) {
5825 num_stripes = map->num_stripes;
5826 } else if (mirror_num) {
5827 stripe_index = mirror_num - 1;
5832 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5833 u32 factor = map->num_stripes / map->sub_stripes;
5835 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5836 stripe_index *= map->sub_stripes;
5838 if (need_full_stripe(op))
5839 num_stripes = map->sub_stripes;
5840 else if (mirror_num)
5841 stripe_index += mirror_num - 1;
5843 int old_stripe_index = stripe_index;
5844 stripe_index = find_live_mirror(fs_info, map,
5846 map->sub_stripes, stripe_index +
5847 current->pid % map->sub_stripes,
5848 dev_replace_is_ongoing);
5849 mirror_num = stripe_index - old_stripe_index + 1;
5852 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5853 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5854 /* push stripe_nr back to the start of the full stripe */
5855 stripe_nr = div64_u64(raid56_full_stripe_start,
5856 stripe_len * nr_data_stripes(map));
5858 /* RAID[56] write or recovery. Return all stripes */
5859 num_stripes = map->num_stripes;
5860 max_errors = nr_parity_stripes(map);
5862 *length = map->stripe_len;
5867 * Mirror #0 or #1 means the original data block.
5868 * Mirror #2 is RAID5 parity block.
5869 * Mirror #3 is RAID6 Q block.
5871 stripe_nr = div_u64_rem(stripe_nr,
5872 nr_data_stripes(map), &stripe_index);
5874 stripe_index = nr_data_stripes(map) +
5877 /* We distribute the parity blocks across stripes */
5878 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5880 if (!need_full_stripe(op) && mirror_num <= 1)
5885 * after this, stripe_nr is the number of stripes on this
5886 * device we have to walk to find the data, and stripe_index is
5887 * the number of our device in the stripe array
5889 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5891 mirror_num = stripe_index + 1;
5893 if (stripe_index >= map->num_stripes) {
5895 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5896 stripe_index, map->num_stripes);
5901 num_alloc_stripes = num_stripes;
5902 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5903 if (op == BTRFS_MAP_WRITE)
5904 num_alloc_stripes <<= 1;
5905 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5906 num_alloc_stripes++;
5907 tgtdev_indexes = num_stripes;
5910 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5915 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5916 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5918 /* build raid_map */
5919 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5920 (need_full_stripe(op) || mirror_num > 1)) {
5924 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5925 sizeof(struct btrfs_bio_stripe) *
5927 sizeof(int) * tgtdev_indexes);
5929 /* Work out the disk rotation on this stripe-set */
5930 div_u64_rem(stripe_nr, num_stripes, &rot);
5932 /* Fill in the logical address of each stripe */
5933 tmp = stripe_nr * nr_data_stripes(map);
5934 for (i = 0; i < nr_data_stripes(map); i++)
5935 bbio->raid_map[(i+rot) % num_stripes] =
5936 em->start + (tmp + i) * map->stripe_len;
5938 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5939 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5940 bbio->raid_map[(i+rot+1) % num_stripes] =
5945 for (i = 0; i < num_stripes; i++) {
5946 bbio->stripes[i].physical =
5947 map->stripes[stripe_index].physical +
5949 stripe_nr * map->stripe_len;
5950 bbio->stripes[i].dev =
5951 map->stripes[stripe_index].dev;
5955 if (need_full_stripe(op))
5956 max_errors = btrfs_chunk_max_errors(map);
5959 sort_parity_stripes(bbio, num_stripes);
5961 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5962 need_full_stripe(op)) {
5963 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5968 bbio->map_type = map->type;
5969 bbio->num_stripes = num_stripes;
5970 bbio->max_errors = max_errors;
5971 bbio->mirror_num = mirror_num;
5974 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5975 * mirror_num == num_stripes + 1 && dev_replace target drive is
5976 * available as a mirror
5978 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5979 WARN_ON(num_stripes > 1);
5980 bbio->stripes[0].dev = dev_replace->tgtdev;
5981 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5982 bbio->mirror_num = map->num_stripes + 1;
5985 if (dev_replace_is_ongoing) {
5986 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5987 btrfs_dev_replace_unlock(dev_replace, 0);
5989 free_extent_map(em);
5993 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5994 u64 logical, u64 *length,
5995 struct btrfs_bio **bbio_ret, int mirror_num)
5997 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6001 /* For Scrub/replace */
6002 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6003 u64 logical, u64 *length,
6004 struct btrfs_bio **bbio_ret)
6006 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6009 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
6010 u64 chunk_start, u64 physical, u64 devid,
6011 u64 **logical, int *naddrs, int *stripe_len)
6013 struct extent_map *em;
6014 struct map_lookup *map;
6022 em = get_chunk_map(fs_info, chunk_start, 1);
6026 map = em->map_lookup;
6028 rmap_len = map->stripe_len;
6030 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6031 length = div_u64(length, map->num_stripes / map->sub_stripes);
6032 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6033 length = div_u64(length, map->num_stripes);
6034 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6035 length = div_u64(length, nr_data_stripes(map));
6036 rmap_len = map->stripe_len * nr_data_stripes(map);
6039 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6040 BUG_ON(!buf); /* -ENOMEM */
6042 for (i = 0; i < map->num_stripes; i++) {
6043 if (devid && map->stripes[i].dev->devid != devid)
6045 if (map->stripes[i].physical > physical ||
6046 map->stripes[i].physical + length <= physical)
6049 stripe_nr = physical - map->stripes[i].physical;
6050 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6052 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6053 stripe_nr = stripe_nr * map->num_stripes + i;
6054 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6055 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6056 stripe_nr = stripe_nr * map->num_stripes + i;
6057 } /* else if RAID[56], multiply by nr_data_stripes().
6058 * Alternatively, just use rmap_len below instead of
6059 * map->stripe_len */
6061 bytenr = chunk_start + stripe_nr * rmap_len;
6062 WARN_ON(nr >= map->num_stripes);
6063 for (j = 0; j < nr; j++) {
6064 if (buf[j] == bytenr)
6068 WARN_ON(nr >= map->num_stripes);
6075 *stripe_len = rmap_len;
6077 free_extent_map(em);
6081 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6083 bio->bi_private = bbio->private;
6084 bio->bi_end_io = bbio->end_io;
6087 btrfs_put_bbio(bbio);
6090 static void btrfs_end_bio(struct bio *bio)
6092 struct btrfs_bio *bbio = bio->bi_private;
6093 int is_orig_bio = 0;
6095 if (bio->bi_status) {
6096 atomic_inc(&bbio->error);
6097 if (bio->bi_status == BLK_STS_IOERR ||
6098 bio->bi_status == BLK_STS_TARGET) {
6099 unsigned int stripe_index =
6100 btrfs_io_bio(bio)->stripe_index;
6101 struct btrfs_device *dev;
6103 BUG_ON(stripe_index >= bbio->num_stripes);
6104 dev = bbio->stripes[stripe_index].dev;
6106 if (bio_op(bio) == REQ_OP_WRITE)
6107 btrfs_dev_stat_inc_and_print(dev,
6108 BTRFS_DEV_STAT_WRITE_ERRS);
6110 btrfs_dev_stat_inc_and_print(dev,
6111 BTRFS_DEV_STAT_READ_ERRS);
6112 if (bio->bi_opf & REQ_PREFLUSH)
6113 btrfs_dev_stat_inc_and_print(dev,
6114 BTRFS_DEV_STAT_FLUSH_ERRS);
6119 if (bio == bbio->orig_bio)
6122 btrfs_bio_counter_dec(bbio->fs_info);
6124 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6127 bio = bbio->orig_bio;
6130 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6131 /* only send an error to the higher layers if it is
6132 * beyond the tolerance of the btrfs bio
6134 if (atomic_read(&bbio->error) > bbio->max_errors) {
6135 bio->bi_status = BLK_STS_IOERR;
6138 * this bio is actually up to date, we didn't
6139 * go over the max number of errors
6141 bio->bi_status = BLK_STS_OK;
6144 btrfs_end_bbio(bbio, bio);
6145 } else if (!is_orig_bio) {
6151 * see run_scheduled_bios for a description of why bios are collected for
6154 * This will add one bio to the pending list for a device and make sure
6155 * the work struct is scheduled.
6157 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6160 struct btrfs_fs_info *fs_info = device->fs_info;
6161 int should_queue = 1;
6162 struct btrfs_pending_bios *pending_bios;
6164 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6170 /* don't bother with additional async steps for reads, right now */
6171 if (bio_op(bio) == REQ_OP_READ) {
6172 btrfsic_submit_bio(bio);
6176 WARN_ON(bio->bi_next);
6177 bio->bi_next = NULL;
6179 spin_lock(&device->io_lock);
6180 if (op_is_sync(bio->bi_opf))
6181 pending_bios = &device->pending_sync_bios;
6183 pending_bios = &device->pending_bios;
6185 if (pending_bios->tail)
6186 pending_bios->tail->bi_next = bio;
6188 pending_bios->tail = bio;
6189 if (!pending_bios->head)
6190 pending_bios->head = bio;
6191 if (device->running_pending)
6194 spin_unlock(&device->io_lock);
6197 btrfs_queue_work(fs_info->submit_workers, &device->work);
6200 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6201 u64 physical, int dev_nr, int async)
6203 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6204 struct btrfs_fs_info *fs_info = bbio->fs_info;
6206 bio->bi_private = bbio;
6207 btrfs_io_bio(bio)->stripe_index = dev_nr;
6208 bio->bi_end_io = btrfs_end_bio;
6209 bio->bi_iter.bi_sector = physical >> 9;
6212 struct rcu_string *name;
6215 name = rcu_dereference(dev->name);
6216 btrfs_debug(fs_info,
6217 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6218 bio_op(bio), bio->bi_opf,
6219 (u64)bio->bi_iter.bi_sector,
6220 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6221 bio->bi_iter.bi_size);
6225 bio_set_dev(bio, dev->bdev);
6227 btrfs_bio_counter_inc_noblocked(fs_info);
6230 btrfs_schedule_bio(dev, bio);
6232 btrfsic_submit_bio(bio);
6235 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6237 atomic_inc(&bbio->error);
6238 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6239 /* Should be the original bio. */
6240 WARN_ON(bio != bbio->orig_bio);
6242 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6243 bio->bi_iter.bi_sector = logical >> 9;
6244 if (atomic_read(&bbio->error) > bbio->max_errors)
6245 bio->bi_status = BLK_STS_IOERR;
6247 bio->bi_status = BLK_STS_OK;
6248 btrfs_end_bbio(bbio, bio);
6252 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6253 int mirror_num, int async_submit)
6255 struct btrfs_device *dev;
6256 struct bio *first_bio = bio;
6257 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6263 struct btrfs_bio *bbio = NULL;
6265 length = bio->bi_iter.bi_size;
6266 map_length = length;
6268 btrfs_bio_counter_inc_blocked(fs_info);
6269 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6270 &map_length, &bbio, mirror_num, 1);
6272 btrfs_bio_counter_dec(fs_info);
6273 return errno_to_blk_status(ret);
6276 total_devs = bbio->num_stripes;
6277 bbio->orig_bio = first_bio;
6278 bbio->private = first_bio->bi_private;
6279 bbio->end_io = first_bio->bi_end_io;
6280 bbio->fs_info = fs_info;
6281 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6283 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6284 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6285 /* In this case, map_length has been set to the length of
6286 a single stripe; not the whole write */
6287 if (bio_op(bio) == REQ_OP_WRITE) {
6288 ret = raid56_parity_write(fs_info, bio, bbio,
6291 ret = raid56_parity_recover(fs_info, bio, bbio,
6292 map_length, mirror_num, 1);
6295 btrfs_bio_counter_dec(fs_info);
6296 return errno_to_blk_status(ret);
6299 if (map_length < length) {
6301 "mapping failed logical %llu bio len %llu len %llu",
6302 logical, length, map_length);
6306 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6307 dev = bbio->stripes[dev_nr].dev;
6308 if (!dev || !dev->bdev ||
6309 (bio_op(first_bio) == REQ_OP_WRITE &&
6310 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6311 bbio_error(bbio, first_bio, logical);
6315 if (dev_nr < total_devs - 1)
6316 bio = btrfs_bio_clone(first_bio);
6320 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6321 dev_nr, async_submit);
6323 btrfs_bio_counter_dec(fs_info);
6327 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6330 struct btrfs_device *device;
6331 struct btrfs_fs_devices *cur_devices;
6333 cur_devices = fs_info->fs_devices;
6334 while (cur_devices) {
6336 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6337 device = find_device(cur_devices, devid, uuid);
6341 cur_devices = cur_devices->seed;
6346 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6347 u64 devid, u8 *dev_uuid)
6349 struct btrfs_device *device;
6351 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6355 list_add(&device->dev_list, &fs_devices->devices);
6356 device->fs_devices = fs_devices;
6357 fs_devices->num_devices++;
6359 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6360 fs_devices->missing_devices++;
6366 * btrfs_alloc_device - allocate struct btrfs_device
6367 * @fs_info: used only for generating a new devid, can be NULL if
6368 * devid is provided (i.e. @devid != NULL).
6369 * @devid: a pointer to devid for this device. If NULL a new devid
6371 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6374 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6375 * on error. Returned struct is not linked onto any lists and must be
6376 * destroyed with free_device.
6378 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6382 struct btrfs_device *dev;
6385 if (WARN_ON(!devid && !fs_info))
6386 return ERR_PTR(-EINVAL);
6388 dev = __alloc_device();
6397 ret = find_next_devid(fs_info, &tmp);
6400 return ERR_PTR(ret);
6406 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6408 generate_random_uuid(dev->uuid);
6410 btrfs_init_work(&dev->work, btrfs_submit_helper,
6411 pending_bios_fn, NULL, NULL);
6416 /* Return -EIO if any error, otherwise return 0. */
6417 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6418 struct extent_buffer *leaf,
6419 struct btrfs_chunk *chunk, u64 logical)
6427 length = btrfs_chunk_length(leaf, chunk);
6428 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6429 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6430 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6431 type = btrfs_chunk_type(leaf, chunk);
6434 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6438 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6439 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6442 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6443 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6444 btrfs_chunk_sector_size(leaf, chunk));
6447 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6448 btrfs_err(fs_info, "invalid chunk length %llu", length);
6451 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6452 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6456 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6458 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6459 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6460 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6461 btrfs_chunk_type(leaf, chunk));
6464 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6465 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6466 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6467 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6468 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6469 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6470 num_stripes != 1)) {
6472 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6473 num_stripes, sub_stripes,
6474 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6481 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6482 u64 devid, u8 *uuid, bool error)
6485 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6488 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6492 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6493 struct extent_buffer *leaf,
6494 struct btrfs_chunk *chunk)
6496 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6497 struct map_lookup *map;
6498 struct extent_map *em;
6502 u8 uuid[BTRFS_UUID_SIZE];
6507 logical = key->offset;
6508 length = btrfs_chunk_length(leaf, chunk);
6509 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6511 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6515 read_lock(&map_tree->map_tree.lock);
6516 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6517 read_unlock(&map_tree->map_tree.lock);
6519 /* already mapped? */
6520 if (em && em->start <= logical && em->start + em->len > logical) {
6521 free_extent_map(em);
6524 free_extent_map(em);
6527 em = alloc_extent_map();
6530 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6532 free_extent_map(em);
6536 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6537 em->map_lookup = map;
6538 em->start = logical;
6541 em->block_start = 0;
6542 em->block_len = em->len;
6544 map->num_stripes = num_stripes;
6545 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6546 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6547 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6548 map->type = btrfs_chunk_type(leaf, chunk);
6549 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6550 for (i = 0; i < num_stripes; i++) {
6551 map->stripes[i].physical =
6552 btrfs_stripe_offset_nr(leaf, chunk, i);
6553 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6554 read_extent_buffer(leaf, uuid, (unsigned long)
6555 btrfs_stripe_dev_uuid_nr(chunk, i),
6557 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6559 if (!map->stripes[i].dev &&
6560 !btrfs_test_opt(fs_info, DEGRADED)) {
6561 free_extent_map(em);
6562 btrfs_report_missing_device(fs_info, devid, uuid, true);
6565 if (!map->stripes[i].dev) {
6566 map->stripes[i].dev =
6567 add_missing_dev(fs_info->fs_devices, devid,
6569 if (IS_ERR(map->stripes[i].dev)) {
6570 free_extent_map(em);
6572 "failed to init missing dev %llu: %ld",
6573 devid, PTR_ERR(map->stripes[i].dev));
6574 return PTR_ERR(map->stripes[i].dev);
6576 btrfs_report_missing_device(fs_info, devid, uuid, false);
6578 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6579 &(map->stripes[i].dev->dev_state));
6583 write_lock(&map_tree->map_tree.lock);
6584 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6585 write_unlock(&map_tree->map_tree.lock);
6586 BUG_ON(ret); /* Tree corruption */
6587 free_extent_map(em);
6592 static void fill_device_from_item(struct extent_buffer *leaf,
6593 struct btrfs_dev_item *dev_item,
6594 struct btrfs_device *device)
6598 device->devid = btrfs_device_id(leaf, dev_item);
6599 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6600 device->total_bytes = device->disk_total_bytes;
6601 device->commit_total_bytes = device->disk_total_bytes;
6602 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6603 device->commit_bytes_used = device->bytes_used;
6604 device->type = btrfs_device_type(leaf, dev_item);
6605 device->io_align = btrfs_device_io_align(leaf, dev_item);
6606 device->io_width = btrfs_device_io_width(leaf, dev_item);
6607 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6608 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6609 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6611 ptr = btrfs_device_uuid(dev_item);
6612 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6615 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6618 struct btrfs_fs_devices *fs_devices;
6621 BUG_ON(!mutex_is_locked(&uuid_mutex));
6624 fs_devices = fs_info->fs_devices->seed;
6625 while (fs_devices) {
6626 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6629 fs_devices = fs_devices->seed;
6632 fs_devices = find_fsid(fsid);
6634 if (!btrfs_test_opt(fs_info, DEGRADED))
6635 return ERR_PTR(-ENOENT);
6637 fs_devices = alloc_fs_devices(fsid);
6638 if (IS_ERR(fs_devices))
6641 fs_devices->seeding = 1;
6642 fs_devices->opened = 1;
6646 fs_devices = clone_fs_devices(fs_devices);
6647 if (IS_ERR(fs_devices))
6650 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6651 fs_info->bdev_holder);
6653 free_fs_devices(fs_devices);
6654 fs_devices = ERR_PTR(ret);
6658 if (!fs_devices->seeding) {
6659 __btrfs_close_devices(fs_devices);
6660 free_fs_devices(fs_devices);
6661 fs_devices = ERR_PTR(-EINVAL);
6665 fs_devices->seed = fs_info->fs_devices->seed;
6666 fs_info->fs_devices->seed = fs_devices;
6671 static int read_one_dev(struct btrfs_fs_info *fs_info,
6672 struct extent_buffer *leaf,
6673 struct btrfs_dev_item *dev_item)
6675 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6676 struct btrfs_device *device;
6679 u8 fs_uuid[BTRFS_FSID_SIZE];
6680 u8 dev_uuid[BTRFS_UUID_SIZE];
6682 devid = btrfs_device_id(leaf, dev_item);
6683 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6685 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6688 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6689 fs_devices = open_seed_devices(fs_info, fs_uuid);
6690 if (IS_ERR(fs_devices))
6691 return PTR_ERR(fs_devices);
6694 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6696 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6697 btrfs_report_missing_device(fs_info, devid,
6702 device = add_missing_dev(fs_devices, devid, dev_uuid);
6703 if (IS_ERR(device)) {
6705 "failed to add missing dev %llu: %ld",
6706 devid, PTR_ERR(device));
6707 return PTR_ERR(device);
6709 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6711 if (!device->bdev) {
6712 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6713 btrfs_report_missing_device(fs_info,
6714 devid, dev_uuid, true);
6717 btrfs_report_missing_device(fs_info, devid,
6721 if (!device->bdev &&
6722 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6724 * this happens when a device that was properly setup
6725 * in the device info lists suddenly goes bad.
6726 * device->bdev is NULL, and so we have to set
6727 * device->missing to one here
6729 device->fs_devices->missing_devices++;
6730 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6733 /* Move the device to its own fs_devices */
6734 if (device->fs_devices != fs_devices) {
6735 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6736 &device->dev_state));
6738 list_move(&device->dev_list, &fs_devices->devices);
6739 device->fs_devices->num_devices--;
6740 fs_devices->num_devices++;
6742 device->fs_devices->missing_devices--;
6743 fs_devices->missing_devices++;
6745 device->fs_devices = fs_devices;
6749 if (device->fs_devices != fs_info->fs_devices) {
6750 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6751 if (device->generation !=
6752 btrfs_device_generation(leaf, dev_item))
6756 fill_device_from_item(leaf, dev_item, device);
6757 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6758 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6759 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6760 device->fs_devices->total_rw_bytes += device->total_bytes;
6761 atomic64_add(device->total_bytes - device->bytes_used,
6762 &fs_info->free_chunk_space);
6768 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6770 struct btrfs_root *root = fs_info->tree_root;
6771 struct btrfs_super_block *super_copy = fs_info->super_copy;
6772 struct extent_buffer *sb;
6773 struct btrfs_disk_key *disk_key;
6774 struct btrfs_chunk *chunk;
6776 unsigned long sb_array_offset;
6783 struct btrfs_key key;
6785 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6787 * This will create extent buffer of nodesize, superblock size is
6788 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6789 * overallocate but we can keep it as-is, only the first page is used.
6791 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6794 set_extent_buffer_uptodate(sb);
6795 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6797 * The sb extent buffer is artificial and just used to read the system array.
6798 * set_extent_buffer_uptodate() call does not properly mark all it's
6799 * pages up-to-date when the page is larger: extent does not cover the
6800 * whole page and consequently check_page_uptodate does not find all
6801 * the page's extents up-to-date (the hole beyond sb),
6802 * write_extent_buffer then triggers a WARN_ON.
6804 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6805 * but sb spans only this function. Add an explicit SetPageUptodate call
6806 * to silence the warning eg. on PowerPC 64.
6808 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6809 SetPageUptodate(sb->pages[0]);
6811 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6812 array_size = btrfs_super_sys_array_size(super_copy);
6814 array_ptr = super_copy->sys_chunk_array;
6815 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6818 while (cur_offset < array_size) {
6819 disk_key = (struct btrfs_disk_key *)array_ptr;
6820 len = sizeof(*disk_key);
6821 if (cur_offset + len > array_size)
6822 goto out_short_read;
6824 btrfs_disk_key_to_cpu(&key, disk_key);
6827 sb_array_offset += len;
6830 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6831 chunk = (struct btrfs_chunk *)sb_array_offset;
6833 * At least one btrfs_chunk with one stripe must be
6834 * present, exact stripe count check comes afterwards
6836 len = btrfs_chunk_item_size(1);
6837 if (cur_offset + len > array_size)
6838 goto out_short_read;
6840 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6843 "invalid number of stripes %u in sys_array at offset %u",
6844 num_stripes, cur_offset);
6849 type = btrfs_chunk_type(sb, chunk);
6850 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6852 "invalid chunk type %llu in sys_array at offset %u",
6858 len = btrfs_chunk_item_size(num_stripes);
6859 if (cur_offset + len > array_size)
6860 goto out_short_read;
6862 ret = read_one_chunk(fs_info, &key, sb, chunk);
6867 "unexpected item type %u in sys_array at offset %u",
6868 (u32)key.type, cur_offset);
6873 sb_array_offset += len;
6876 clear_extent_buffer_uptodate(sb);
6877 free_extent_buffer_stale(sb);
6881 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6883 clear_extent_buffer_uptodate(sb);
6884 free_extent_buffer_stale(sb);
6889 * Check if all chunks in the fs are OK for read-write degraded mount
6891 * If the @failing_dev is specified, it's accounted as missing.
6893 * Return true if all chunks meet the minimal RW mount requirements.
6894 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6896 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6897 struct btrfs_device *failing_dev)
6899 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6900 struct extent_map *em;
6904 read_lock(&map_tree->map_tree.lock);
6905 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6906 read_unlock(&map_tree->map_tree.lock);
6907 /* No chunk at all? Return false anyway */
6913 struct map_lookup *map;
6918 map = em->map_lookup;
6920 btrfs_get_num_tolerated_disk_barrier_failures(
6922 for (i = 0; i < map->num_stripes; i++) {
6923 struct btrfs_device *dev = map->stripes[i].dev;
6925 if (!dev || !dev->bdev ||
6926 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6927 dev->last_flush_error)
6929 else if (failing_dev && failing_dev == dev)
6932 if (missing > max_tolerated) {
6935 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6936 em->start, missing, max_tolerated);
6937 free_extent_map(em);
6941 next_start = extent_map_end(em);
6942 free_extent_map(em);
6944 read_lock(&map_tree->map_tree.lock);
6945 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6946 (u64)(-1) - next_start);
6947 read_unlock(&map_tree->map_tree.lock);
6953 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6955 struct btrfs_root *root = fs_info->chunk_root;
6956 struct btrfs_path *path;
6957 struct extent_buffer *leaf;
6958 struct btrfs_key key;
6959 struct btrfs_key found_key;
6964 path = btrfs_alloc_path();
6968 mutex_lock(&uuid_mutex);
6969 mutex_lock(&fs_info->chunk_mutex);
6972 * Read all device items, and then all the chunk items. All
6973 * device items are found before any chunk item (their object id
6974 * is smaller than the lowest possible object id for a chunk
6975 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6977 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6980 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6984 leaf = path->nodes[0];
6985 slot = path->slots[0];
6986 if (slot >= btrfs_header_nritems(leaf)) {
6987 ret = btrfs_next_leaf(root, path);
6994 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6995 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6996 struct btrfs_dev_item *dev_item;
6997 dev_item = btrfs_item_ptr(leaf, slot,
6998 struct btrfs_dev_item);
6999 ret = read_one_dev(fs_info, leaf, dev_item);
7003 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7004 struct btrfs_chunk *chunk;
7005 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7006 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7014 * After loading chunk tree, we've got all device information,
7015 * do another round of validation checks.
7017 if (total_dev != fs_info->fs_devices->total_devices) {
7019 "super_num_devices %llu mismatch with num_devices %llu found here",
7020 btrfs_super_num_devices(fs_info->super_copy),
7025 if (btrfs_super_total_bytes(fs_info->super_copy) <
7026 fs_info->fs_devices->total_rw_bytes) {
7028 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7029 btrfs_super_total_bytes(fs_info->super_copy),
7030 fs_info->fs_devices->total_rw_bytes);
7036 mutex_unlock(&fs_info->chunk_mutex);
7037 mutex_unlock(&uuid_mutex);
7039 btrfs_free_path(path);
7043 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7045 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7046 struct btrfs_device *device;
7048 while (fs_devices) {
7049 mutex_lock(&fs_devices->device_list_mutex);
7050 list_for_each_entry(device, &fs_devices->devices, dev_list)
7051 device->fs_info = fs_info;
7052 mutex_unlock(&fs_devices->device_list_mutex);
7054 fs_devices = fs_devices->seed;
7058 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7062 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7063 btrfs_dev_stat_reset(dev, i);
7066 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7068 struct btrfs_key key;
7069 struct btrfs_key found_key;
7070 struct btrfs_root *dev_root = fs_info->dev_root;
7071 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7072 struct extent_buffer *eb;
7075 struct btrfs_device *device;
7076 struct btrfs_path *path = NULL;
7079 path = btrfs_alloc_path();
7085 mutex_lock(&fs_devices->device_list_mutex);
7086 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7088 struct btrfs_dev_stats_item *ptr;
7090 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7091 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7092 key.offset = device->devid;
7093 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7095 __btrfs_reset_dev_stats(device);
7096 device->dev_stats_valid = 1;
7097 btrfs_release_path(path);
7100 slot = path->slots[0];
7101 eb = path->nodes[0];
7102 btrfs_item_key_to_cpu(eb, &found_key, slot);
7103 item_size = btrfs_item_size_nr(eb, slot);
7105 ptr = btrfs_item_ptr(eb, slot,
7106 struct btrfs_dev_stats_item);
7108 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7109 if (item_size >= (1 + i) * sizeof(__le64))
7110 btrfs_dev_stat_set(device, i,
7111 btrfs_dev_stats_value(eb, ptr, i));
7113 btrfs_dev_stat_reset(device, i);
7116 device->dev_stats_valid = 1;
7117 btrfs_dev_stat_print_on_load(device);
7118 btrfs_release_path(path);
7120 mutex_unlock(&fs_devices->device_list_mutex);
7123 btrfs_free_path(path);
7124 return ret < 0 ? ret : 0;
7127 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7128 struct btrfs_fs_info *fs_info,
7129 struct btrfs_device *device)
7131 struct btrfs_root *dev_root = fs_info->dev_root;
7132 struct btrfs_path *path;
7133 struct btrfs_key key;
7134 struct extent_buffer *eb;
7135 struct btrfs_dev_stats_item *ptr;
7139 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7140 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7141 key.offset = device->devid;
7143 path = btrfs_alloc_path();
7146 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7148 btrfs_warn_in_rcu(fs_info,
7149 "error %d while searching for dev_stats item for device %s",
7150 ret, rcu_str_deref(device->name));
7155 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7156 /* need to delete old one and insert a new one */
7157 ret = btrfs_del_item(trans, dev_root, path);
7159 btrfs_warn_in_rcu(fs_info,
7160 "delete too small dev_stats item for device %s failed %d",
7161 rcu_str_deref(device->name), ret);
7168 /* need to insert a new item */
7169 btrfs_release_path(path);
7170 ret = btrfs_insert_empty_item(trans, dev_root, path,
7171 &key, sizeof(*ptr));
7173 btrfs_warn_in_rcu(fs_info,
7174 "insert dev_stats item for device %s failed %d",
7175 rcu_str_deref(device->name), ret);
7180 eb = path->nodes[0];
7181 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7182 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7183 btrfs_set_dev_stats_value(eb, ptr, i,
7184 btrfs_dev_stat_read(device, i));
7185 btrfs_mark_buffer_dirty(eb);
7188 btrfs_free_path(path);
7193 * called from commit_transaction. Writes all changed device stats to disk.
7195 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7196 struct btrfs_fs_info *fs_info)
7198 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7199 struct btrfs_device *device;
7203 mutex_lock(&fs_devices->device_list_mutex);
7204 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7205 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7206 if (!device->dev_stats_valid || stats_cnt == 0)
7211 * There is a LOAD-LOAD control dependency between the value of
7212 * dev_stats_ccnt and updating the on-disk values which requires
7213 * reading the in-memory counters. Such control dependencies
7214 * require explicit read memory barriers.
7216 * This memory barriers pairs with smp_mb__before_atomic in
7217 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7218 * barrier implied by atomic_xchg in
7219 * btrfs_dev_stats_read_and_reset
7223 ret = update_dev_stat_item(trans, fs_info, device);
7225 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7227 mutex_unlock(&fs_devices->device_list_mutex);
7232 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7234 btrfs_dev_stat_inc(dev, index);
7235 btrfs_dev_stat_print_on_error(dev);
7238 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7240 if (!dev->dev_stats_valid)
7242 btrfs_err_rl_in_rcu(dev->fs_info,
7243 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7244 rcu_str_deref(dev->name),
7245 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7246 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7247 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7248 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7249 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7252 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7256 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7257 if (btrfs_dev_stat_read(dev, i) != 0)
7259 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7260 return; /* all values == 0, suppress message */
7262 btrfs_info_in_rcu(dev->fs_info,
7263 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7264 rcu_str_deref(dev->name),
7265 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7266 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7267 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7268 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7269 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7272 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7273 struct btrfs_ioctl_get_dev_stats *stats)
7275 struct btrfs_device *dev;
7276 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7279 mutex_lock(&fs_devices->device_list_mutex);
7280 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7281 mutex_unlock(&fs_devices->device_list_mutex);
7284 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7286 } else if (!dev->dev_stats_valid) {
7287 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7289 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7290 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7291 if (stats->nr_items > i)
7293 btrfs_dev_stat_read_and_reset(dev, i);
7295 btrfs_dev_stat_reset(dev, i);
7298 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7299 if (stats->nr_items > i)
7300 stats->values[i] = btrfs_dev_stat_read(dev, i);
7302 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7303 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7307 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7309 struct buffer_head *bh;
7310 struct btrfs_super_block *disk_super;
7316 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7319 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7322 disk_super = (struct btrfs_super_block *)bh->b_data;
7324 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7325 set_buffer_dirty(bh);
7326 sync_dirty_buffer(bh);
7330 /* Notify udev that device has changed */
7331 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7333 /* Update ctime/mtime for device path for libblkid */
7334 update_dev_time(device_path);
7338 * Update the size of all devices, which is used for writing out the
7341 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7343 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7344 struct btrfs_device *curr, *next;
7346 if (list_empty(&fs_devices->resized_devices))
7349 mutex_lock(&fs_devices->device_list_mutex);
7350 mutex_lock(&fs_info->chunk_mutex);
7351 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7353 list_del_init(&curr->resized_list);
7354 curr->commit_total_bytes = curr->disk_total_bytes;
7356 mutex_unlock(&fs_info->chunk_mutex);
7357 mutex_unlock(&fs_devices->device_list_mutex);
7360 /* Must be invoked during the transaction commit */
7361 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7362 struct btrfs_transaction *transaction)
7364 struct extent_map *em;
7365 struct map_lookup *map;
7366 struct btrfs_device *dev;
7369 if (list_empty(&transaction->pending_chunks))
7372 /* In order to kick the device replace finish process */
7373 mutex_lock(&fs_info->chunk_mutex);
7374 list_for_each_entry(em, &transaction->pending_chunks, list) {
7375 map = em->map_lookup;
7377 for (i = 0; i < map->num_stripes; i++) {
7378 dev = map->stripes[i].dev;
7379 dev->commit_bytes_used = dev->bytes_used;
7382 mutex_unlock(&fs_info->chunk_mutex);
7385 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7387 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7388 while (fs_devices) {
7389 fs_devices->fs_info = fs_info;
7390 fs_devices = fs_devices->seed;
7394 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7396 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7397 while (fs_devices) {
7398 fs_devices->fs_info = NULL;
7399 fs_devices = fs_devices->seed;