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 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2670 struct btrfs_device *device)
2673 struct btrfs_path *path;
2674 struct btrfs_root *root = device->fs_info->chunk_root;
2675 struct btrfs_dev_item *dev_item;
2676 struct extent_buffer *leaf;
2677 struct btrfs_key key;
2679 path = btrfs_alloc_path();
2683 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2684 key.type = BTRFS_DEV_ITEM_KEY;
2685 key.offset = device->devid;
2687 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2696 leaf = path->nodes[0];
2697 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2699 btrfs_set_device_id(leaf, dev_item, device->devid);
2700 btrfs_set_device_type(leaf, dev_item, device->type);
2701 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2702 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2703 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2704 btrfs_set_device_total_bytes(leaf, dev_item,
2705 btrfs_device_get_disk_total_bytes(device));
2706 btrfs_set_device_bytes_used(leaf, dev_item,
2707 btrfs_device_get_bytes_used(device));
2708 btrfs_mark_buffer_dirty(leaf);
2711 btrfs_free_path(path);
2715 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2716 struct btrfs_device *device, u64 new_size)
2718 struct btrfs_fs_info *fs_info = device->fs_info;
2719 struct btrfs_super_block *super_copy = fs_info->super_copy;
2720 struct btrfs_fs_devices *fs_devices;
2724 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2727 new_size = round_down(new_size, fs_info->sectorsize);
2729 mutex_lock(&fs_info->chunk_mutex);
2730 old_total = btrfs_super_total_bytes(super_copy);
2731 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2733 if (new_size <= device->total_bytes ||
2734 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2735 mutex_unlock(&fs_info->chunk_mutex);
2739 fs_devices = fs_info->fs_devices;
2741 btrfs_set_super_total_bytes(super_copy,
2742 round_down(old_total + diff, fs_info->sectorsize));
2743 device->fs_devices->total_rw_bytes += diff;
2745 btrfs_device_set_total_bytes(device, new_size);
2746 btrfs_device_set_disk_total_bytes(device, new_size);
2747 btrfs_clear_space_info_full(device->fs_info);
2748 if (list_empty(&device->resized_list))
2749 list_add_tail(&device->resized_list,
2750 &fs_devices->resized_devices);
2751 mutex_unlock(&fs_info->chunk_mutex);
2753 return btrfs_update_device(trans, device);
2756 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2757 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2759 struct btrfs_root *root = fs_info->chunk_root;
2761 struct btrfs_path *path;
2762 struct btrfs_key key;
2764 path = btrfs_alloc_path();
2768 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2769 key.offset = chunk_offset;
2770 key.type = BTRFS_CHUNK_ITEM_KEY;
2772 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2775 else if (ret > 0) { /* Logic error or corruption */
2776 btrfs_handle_fs_error(fs_info, -ENOENT,
2777 "Failed lookup while freeing chunk.");
2782 ret = btrfs_del_item(trans, root, path);
2784 btrfs_handle_fs_error(fs_info, ret,
2785 "Failed to delete chunk item.");
2787 btrfs_free_path(path);
2791 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2793 struct btrfs_super_block *super_copy = fs_info->super_copy;
2794 struct btrfs_disk_key *disk_key;
2795 struct btrfs_chunk *chunk;
2802 struct btrfs_key key;
2804 mutex_lock(&fs_info->chunk_mutex);
2805 array_size = btrfs_super_sys_array_size(super_copy);
2807 ptr = super_copy->sys_chunk_array;
2810 while (cur < array_size) {
2811 disk_key = (struct btrfs_disk_key *)ptr;
2812 btrfs_disk_key_to_cpu(&key, disk_key);
2814 len = sizeof(*disk_key);
2816 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2817 chunk = (struct btrfs_chunk *)(ptr + len);
2818 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2819 len += btrfs_chunk_item_size(num_stripes);
2824 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2825 key.offset == chunk_offset) {
2826 memmove(ptr, ptr + len, array_size - (cur + len));
2828 btrfs_set_super_sys_array_size(super_copy, array_size);
2834 mutex_unlock(&fs_info->chunk_mutex);
2838 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2839 u64 logical, u64 length)
2841 struct extent_map_tree *em_tree;
2842 struct extent_map *em;
2844 em_tree = &fs_info->mapping_tree.map_tree;
2845 read_lock(&em_tree->lock);
2846 em = lookup_extent_mapping(em_tree, logical, length);
2847 read_unlock(&em_tree->lock);
2850 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2852 return ERR_PTR(-EINVAL);
2855 if (em->start > logical || em->start + em->len < logical) {
2857 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2858 logical, length, em->start, em->start + em->len);
2859 free_extent_map(em);
2860 return ERR_PTR(-EINVAL);
2863 /* callers are responsible for dropping em's ref. */
2867 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2868 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2870 struct extent_map *em;
2871 struct map_lookup *map;
2872 u64 dev_extent_len = 0;
2874 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2876 em = get_chunk_map(fs_info, chunk_offset, 1);
2879 * This is a logic error, but we don't want to just rely on the
2880 * user having built with ASSERT enabled, so if ASSERT doesn't
2881 * do anything we still error out.
2886 map = em->map_lookup;
2887 mutex_lock(&fs_info->chunk_mutex);
2888 check_system_chunk(trans, fs_info, map->type);
2889 mutex_unlock(&fs_info->chunk_mutex);
2892 * Take the device list mutex to prevent races with the final phase of
2893 * a device replace operation that replaces the device object associated
2894 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2896 mutex_lock(&fs_devices->device_list_mutex);
2897 for (i = 0; i < map->num_stripes; i++) {
2898 struct btrfs_device *device = map->stripes[i].dev;
2899 ret = btrfs_free_dev_extent(trans, device,
2900 map->stripes[i].physical,
2903 mutex_unlock(&fs_devices->device_list_mutex);
2904 btrfs_abort_transaction(trans, ret);
2908 if (device->bytes_used > 0) {
2909 mutex_lock(&fs_info->chunk_mutex);
2910 btrfs_device_set_bytes_used(device,
2911 device->bytes_used - dev_extent_len);
2912 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2913 btrfs_clear_space_info_full(fs_info);
2914 mutex_unlock(&fs_info->chunk_mutex);
2917 if (map->stripes[i].dev) {
2918 ret = btrfs_update_device(trans, map->stripes[i].dev);
2920 mutex_unlock(&fs_devices->device_list_mutex);
2921 btrfs_abort_transaction(trans, ret);
2926 mutex_unlock(&fs_devices->device_list_mutex);
2928 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2930 btrfs_abort_transaction(trans, ret);
2934 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2936 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2937 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2939 btrfs_abort_transaction(trans, ret);
2944 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2946 btrfs_abort_transaction(trans, ret);
2952 free_extent_map(em);
2956 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2958 struct btrfs_root *root = fs_info->chunk_root;
2959 struct btrfs_trans_handle *trans;
2963 * Prevent races with automatic removal of unused block groups.
2964 * After we relocate and before we remove the chunk with offset
2965 * chunk_offset, automatic removal of the block group can kick in,
2966 * resulting in a failure when calling btrfs_remove_chunk() below.
2968 * Make sure to acquire this mutex before doing a tree search (dev
2969 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2970 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2971 * we release the path used to search the chunk/dev tree and before
2972 * the current task acquires this mutex and calls us.
2974 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2976 ret = btrfs_can_relocate(fs_info, chunk_offset);
2980 /* step one, relocate all the extents inside this chunk */
2981 btrfs_scrub_pause(fs_info);
2982 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2983 btrfs_scrub_continue(fs_info);
2987 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2989 if (IS_ERR(trans)) {
2990 ret = PTR_ERR(trans);
2991 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2996 * step two, delete the device extents and the
2997 * chunk tree entries
2999 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3000 btrfs_end_transaction(trans);
3004 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3006 struct btrfs_root *chunk_root = fs_info->chunk_root;
3007 struct btrfs_path *path;
3008 struct extent_buffer *leaf;
3009 struct btrfs_chunk *chunk;
3010 struct btrfs_key key;
3011 struct btrfs_key found_key;
3013 bool retried = false;
3017 path = btrfs_alloc_path();
3022 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3023 key.offset = (u64)-1;
3024 key.type = BTRFS_CHUNK_ITEM_KEY;
3027 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3028 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3030 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3033 BUG_ON(ret == 0); /* Corruption */
3035 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3038 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3044 leaf = path->nodes[0];
3045 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3047 chunk = btrfs_item_ptr(leaf, path->slots[0],
3048 struct btrfs_chunk);
3049 chunk_type = btrfs_chunk_type(leaf, chunk);
3050 btrfs_release_path(path);
3052 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3053 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3059 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3061 if (found_key.offset == 0)
3063 key.offset = found_key.offset - 1;
3066 if (failed && !retried) {
3070 } else if (WARN_ON(failed && retried)) {
3074 btrfs_free_path(path);
3079 * return 1 : allocate a data chunk successfully,
3080 * return <0: errors during allocating a data chunk,
3081 * return 0 : no need to allocate a data chunk.
3083 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3086 struct btrfs_block_group_cache *cache;
3090 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3092 chunk_type = cache->flags;
3093 btrfs_put_block_group(cache);
3095 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3096 spin_lock(&fs_info->data_sinfo->lock);
3097 bytes_used = fs_info->data_sinfo->bytes_used;
3098 spin_unlock(&fs_info->data_sinfo->lock);
3101 struct btrfs_trans_handle *trans;
3104 trans = btrfs_join_transaction(fs_info->tree_root);
3106 return PTR_ERR(trans);
3108 ret = btrfs_force_chunk_alloc(trans, fs_info,
3109 BTRFS_BLOCK_GROUP_DATA);
3110 btrfs_end_transaction(trans);
3120 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3121 struct btrfs_balance_control *bctl)
3123 struct btrfs_root *root = fs_info->tree_root;
3124 struct btrfs_trans_handle *trans;
3125 struct btrfs_balance_item *item;
3126 struct btrfs_disk_balance_args disk_bargs;
3127 struct btrfs_path *path;
3128 struct extent_buffer *leaf;
3129 struct btrfs_key key;
3132 path = btrfs_alloc_path();
3136 trans = btrfs_start_transaction(root, 0);
3137 if (IS_ERR(trans)) {
3138 btrfs_free_path(path);
3139 return PTR_ERR(trans);
3142 key.objectid = BTRFS_BALANCE_OBJECTID;
3143 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3146 ret = btrfs_insert_empty_item(trans, root, path, &key,
3151 leaf = path->nodes[0];
3152 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3154 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3156 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3157 btrfs_set_balance_data(leaf, item, &disk_bargs);
3158 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3159 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3160 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3161 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3163 btrfs_set_balance_flags(leaf, item, bctl->flags);
3165 btrfs_mark_buffer_dirty(leaf);
3167 btrfs_free_path(path);
3168 err = btrfs_commit_transaction(trans);
3174 static int del_balance_item(struct btrfs_fs_info *fs_info)
3176 struct btrfs_root *root = fs_info->tree_root;
3177 struct btrfs_trans_handle *trans;
3178 struct btrfs_path *path;
3179 struct btrfs_key key;
3182 path = btrfs_alloc_path();
3186 trans = btrfs_start_transaction(root, 0);
3187 if (IS_ERR(trans)) {
3188 btrfs_free_path(path);
3189 return PTR_ERR(trans);
3192 key.objectid = BTRFS_BALANCE_OBJECTID;
3193 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3196 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3204 ret = btrfs_del_item(trans, root, path);
3206 btrfs_free_path(path);
3207 err = btrfs_commit_transaction(trans);
3214 * This is a heuristic used to reduce the number of chunks balanced on
3215 * resume after balance was interrupted.
3217 static void update_balance_args(struct btrfs_balance_control *bctl)
3220 * Turn on soft mode for chunk types that were being converted.
3222 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3223 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3224 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3225 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3226 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3227 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3230 * Turn on usage filter if is not already used. The idea is
3231 * that chunks that we have already balanced should be
3232 * reasonably full. Don't do it for chunks that are being
3233 * converted - that will keep us from relocating unconverted
3234 * (albeit full) chunks.
3236 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3237 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3238 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3239 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3240 bctl->data.usage = 90;
3242 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3243 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3244 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3245 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3246 bctl->sys.usage = 90;
3248 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3249 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3250 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3251 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3252 bctl->meta.usage = 90;
3257 * Should be called with both balance and volume mutexes held to
3258 * serialize other volume operations (add_dev/rm_dev/resize) with
3259 * restriper. Same goes for unset_balance_control.
3261 static void set_balance_control(struct btrfs_balance_control *bctl)
3263 struct btrfs_fs_info *fs_info = bctl->fs_info;
3265 BUG_ON(fs_info->balance_ctl);
3267 spin_lock(&fs_info->balance_lock);
3268 fs_info->balance_ctl = bctl;
3269 spin_unlock(&fs_info->balance_lock);
3272 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3274 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3276 BUG_ON(!fs_info->balance_ctl);
3278 spin_lock(&fs_info->balance_lock);
3279 fs_info->balance_ctl = NULL;
3280 spin_unlock(&fs_info->balance_lock);
3286 * Balance filters. Return 1 if chunk should be filtered out
3287 * (should not be balanced).
3289 static int chunk_profiles_filter(u64 chunk_type,
3290 struct btrfs_balance_args *bargs)
3292 chunk_type = chunk_to_extended(chunk_type) &
3293 BTRFS_EXTENDED_PROFILE_MASK;
3295 if (bargs->profiles & chunk_type)
3301 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3302 struct btrfs_balance_args *bargs)
3304 struct btrfs_block_group_cache *cache;
3306 u64 user_thresh_min;
3307 u64 user_thresh_max;
3310 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3311 chunk_used = btrfs_block_group_used(&cache->item);
3313 if (bargs->usage_min == 0)
3314 user_thresh_min = 0;
3316 user_thresh_min = div_factor_fine(cache->key.offset,
3319 if (bargs->usage_max == 0)
3320 user_thresh_max = 1;
3321 else if (bargs->usage_max > 100)
3322 user_thresh_max = cache->key.offset;
3324 user_thresh_max = div_factor_fine(cache->key.offset,
3327 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3330 btrfs_put_block_group(cache);
3334 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3335 u64 chunk_offset, struct btrfs_balance_args *bargs)
3337 struct btrfs_block_group_cache *cache;
3338 u64 chunk_used, user_thresh;
3341 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3342 chunk_used = btrfs_block_group_used(&cache->item);
3344 if (bargs->usage_min == 0)
3346 else if (bargs->usage > 100)
3347 user_thresh = cache->key.offset;
3349 user_thresh = div_factor_fine(cache->key.offset,
3352 if (chunk_used < user_thresh)
3355 btrfs_put_block_group(cache);
3359 static int chunk_devid_filter(struct extent_buffer *leaf,
3360 struct btrfs_chunk *chunk,
3361 struct btrfs_balance_args *bargs)
3363 struct btrfs_stripe *stripe;
3364 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3367 for (i = 0; i < num_stripes; i++) {
3368 stripe = btrfs_stripe_nr(chunk, i);
3369 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3376 /* [pstart, pend) */
3377 static int chunk_drange_filter(struct extent_buffer *leaf,
3378 struct btrfs_chunk *chunk,
3379 struct btrfs_balance_args *bargs)
3381 struct btrfs_stripe *stripe;
3382 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3388 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3391 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3392 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3393 factor = num_stripes / 2;
3394 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3395 factor = num_stripes - 1;
3396 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3397 factor = num_stripes - 2;
3399 factor = num_stripes;
3402 for (i = 0; i < num_stripes; i++) {
3403 stripe = btrfs_stripe_nr(chunk, i);
3404 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3407 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3408 stripe_length = btrfs_chunk_length(leaf, chunk);
3409 stripe_length = div_u64(stripe_length, factor);
3411 if (stripe_offset < bargs->pend &&
3412 stripe_offset + stripe_length > bargs->pstart)
3419 /* [vstart, vend) */
3420 static int chunk_vrange_filter(struct extent_buffer *leaf,
3421 struct btrfs_chunk *chunk,
3423 struct btrfs_balance_args *bargs)
3425 if (chunk_offset < bargs->vend &&
3426 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3427 /* at least part of the chunk is inside this vrange */
3433 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3434 struct btrfs_chunk *chunk,
3435 struct btrfs_balance_args *bargs)
3437 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3439 if (bargs->stripes_min <= num_stripes
3440 && num_stripes <= bargs->stripes_max)
3446 static int chunk_soft_convert_filter(u64 chunk_type,
3447 struct btrfs_balance_args *bargs)
3449 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3452 chunk_type = chunk_to_extended(chunk_type) &
3453 BTRFS_EXTENDED_PROFILE_MASK;
3455 if (bargs->target == chunk_type)
3461 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3462 struct extent_buffer *leaf,
3463 struct btrfs_chunk *chunk, u64 chunk_offset)
3465 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3466 struct btrfs_balance_args *bargs = NULL;
3467 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3470 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3471 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3475 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3476 bargs = &bctl->data;
3477 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3479 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3480 bargs = &bctl->meta;
3482 /* profiles filter */
3483 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3484 chunk_profiles_filter(chunk_type, bargs)) {
3489 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3490 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3492 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3493 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3498 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3499 chunk_devid_filter(leaf, chunk, bargs)) {
3503 /* drange filter, makes sense only with devid filter */
3504 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3505 chunk_drange_filter(leaf, chunk, bargs)) {
3510 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3511 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3515 /* stripes filter */
3516 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3517 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3521 /* soft profile changing mode */
3522 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3523 chunk_soft_convert_filter(chunk_type, bargs)) {
3528 * limited by count, must be the last filter
3530 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3531 if (bargs->limit == 0)
3535 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3537 * Same logic as the 'limit' filter; the minimum cannot be
3538 * determined here because we do not have the global information
3539 * about the count of all chunks that satisfy the filters.
3541 if (bargs->limit_max == 0)
3550 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3552 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3553 struct btrfs_root *chunk_root = fs_info->chunk_root;
3554 struct btrfs_root *dev_root = fs_info->dev_root;
3555 struct list_head *devices;
3556 struct btrfs_device *device;
3560 struct btrfs_chunk *chunk;
3561 struct btrfs_path *path = NULL;
3562 struct btrfs_key key;
3563 struct btrfs_key found_key;
3564 struct btrfs_trans_handle *trans;
3565 struct extent_buffer *leaf;
3568 int enospc_errors = 0;
3569 bool counting = true;
3570 /* The single value limit and min/max limits use the same bytes in the */
3571 u64 limit_data = bctl->data.limit;
3572 u64 limit_meta = bctl->meta.limit;
3573 u64 limit_sys = bctl->sys.limit;
3577 int chunk_reserved = 0;
3579 /* step one make some room on all the devices */
3580 devices = &fs_info->fs_devices->devices;
3581 list_for_each_entry(device, devices, dev_list) {
3582 old_size = btrfs_device_get_total_bytes(device);
3583 size_to_free = div_factor(old_size, 1);
3584 size_to_free = min_t(u64, size_to_free, SZ_1M);
3585 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3586 btrfs_device_get_total_bytes(device) -
3587 btrfs_device_get_bytes_used(device) > size_to_free ||
3588 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3591 ret = btrfs_shrink_device(device, old_size - size_to_free);
3595 /* btrfs_shrink_device never returns ret > 0 */
3600 trans = btrfs_start_transaction(dev_root, 0);
3601 if (IS_ERR(trans)) {
3602 ret = PTR_ERR(trans);
3603 btrfs_info_in_rcu(fs_info,
3604 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3605 rcu_str_deref(device->name), ret,
3606 old_size, old_size - size_to_free);
3610 ret = btrfs_grow_device(trans, device, old_size);
3612 btrfs_end_transaction(trans);
3613 /* btrfs_grow_device never returns ret > 0 */
3615 btrfs_info_in_rcu(fs_info,
3616 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3617 rcu_str_deref(device->name), ret,
3618 old_size, old_size - size_to_free);
3622 btrfs_end_transaction(trans);
3625 /* step two, relocate all the chunks */
3626 path = btrfs_alloc_path();
3632 /* zero out stat counters */
3633 spin_lock(&fs_info->balance_lock);
3634 memset(&bctl->stat, 0, sizeof(bctl->stat));
3635 spin_unlock(&fs_info->balance_lock);
3639 * The single value limit and min/max limits use the same bytes
3642 bctl->data.limit = limit_data;
3643 bctl->meta.limit = limit_meta;
3644 bctl->sys.limit = limit_sys;
3646 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3647 key.offset = (u64)-1;
3648 key.type = BTRFS_CHUNK_ITEM_KEY;
3651 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3652 atomic_read(&fs_info->balance_cancel_req)) {
3657 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3658 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3660 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3665 * this shouldn't happen, it means the last relocate
3669 BUG(); /* FIXME break ? */
3671 ret = btrfs_previous_item(chunk_root, path, 0,
3672 BTRFS_CHUNK_ITEM_KEY);
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3679 leaf = path->nodes[0];
3680 slot = path->slots[0];
3681 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3683 if (found_key.objectid != key.objectid) {
3684 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3688 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3689 chunk_type = btrfs_chunk_type(leaf, chunk);
3692 spin_lock(&fs_info->balance_lock);
3693 bctl->stat.considered++;
3694 spin_unlock(&fs_info->balance_lock);
3697 ret = should_balance_chunk(fs_info, leaf, chunk,
3700 btrfs_release_path(path);
3702 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3707 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3708 spin_lock(&fs_info->balance_lock);
3709 bctl->stat.expected++;
3710 spin_unlock(&fs_info->balance_lock);
3712 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3714 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3716 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3723 * Apply limit_min filter, no need to check if the LIMITS
3724 * filter is used, limit_min is 0 by default
3726 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3727 count_data < bctl->data.limit_min)
3728 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3729 count_meta < bctl->meta.limit_min)
3730 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3731 count_sys < bctl->sys.limit_min)) {
3732 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3736 if (!chunk_reserved) {
3738 * We may be relocating the only data chunk we have,
3739 * which could potentially end up with losing data's
3740 * raid profile, so lets allocate an empty one in
3743 ret = btrfs_may_alloc_data_chunk(fs_info,
3746 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3748 } else if (ret == 1) {
3753 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3754 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3755 if (ret && ret != -ENOSPC)
3757 if (ret == -ENOSPC) {
3760 spin_lock(&fs_info->balance_lock);
3761 bctl->stat.completed++;
3762 spin_unlock(&fs_info->balance_lock);
3765 if (found_key.offset == 0)
3767 key.offset = found_key.offset - 1;
3771 btrfs_release_path(path);
3776 btrfs_free_path(path);
3777 if (enospc_errors) {
3778 btrfs_info(fs_info, "%d enospc errors during balance",
3788 * alloc_profile_is_valid - see if a given profile is valid and reduced
3789 * @flags: profile to validate
3790 * @extended: if true @flags is treated as an extended profile
3792 static int alloc_profile_is_valid(u64 flags, int extended)
3794 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3795 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3797 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3799 /* 1) check that all other bits are zeroed */
3803 /* 2) see if profile is reduced */
3805 return !extended; /* "0" is valid for usual profiles */
3807 /* true if exactly one bit set */
3808 return (flags & (flags - 1)) == 0;
3811 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3813 /* cancel requested || normal exit path */
3814 return atomic_read(&fs_info->balance_cancel_req) ||
3815 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3816 atomic_read(&fs_info->balance_cancel_req) == 0);
3819 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3823 unset_balance_control(fs_info);
3824 ret = del_balance_item(fs_info);
3826 btrfs_handle_fs_error(fs_info, ret, NULL);
3828 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3831 /* Non-zero return value signifies invalidity */
3832 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3835 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3836 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3837 (bctl_arg->target & ~allowed)));
3841 * Should be called with both balance and volume mutexes held
3843 int btrfs_balance(struct btrfs_balance_control *bctl,
3844 struct btrfs_ioctl_balance_args *bargs)
3846 struct btrfs_fs_info *fs_info = bctl->fs_info;
3847 u64 meta_target, data_target;
3854 if (btrfs_fs_closing(fs_info) ||
3855 atomic_read(&fs_info->balance_pause_req) ||
3856 atomic_read(&fs_info->balance_cancel_req)) {
3861 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3862 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3866 * In case of mixed groups both data and meta should be picked,
3867 * and identical options should be given for both of them.
3869 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3870 if (mixed && (bctl->flags & allowed)) {
3871 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3872 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3873 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3875 "with mixed groups data and metadata balance options must be the same");
3881 num_devices = fs_info->fs_devices->num_devices;
3882 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3883 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3884 BUG_ON(num_devices < 1);
3887 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3888 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3889 if (num_devices > 1)
3890 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3891 if (num_devices > 2)
3892 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3893 if (num_devices > 3)
3894 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3895 BTRFS_BLOCK_GROUP_RAID6);
3896 if (validate_convert_profile(&bctl->data, allowed)) {
3898 "unable to start balance with target data profile %llu",
3903 if (validate_convert_profile(&bctl->meta, allowed)) {
3905 "unable to start balance with target metadata profile %llu",
3910 if (validate_convert_profile(&bctl->sys, allowed)) {
3912 "unable to start balance with target system profile %llu",
3918 /* allow to reduce meta or sys integrity only if force set */
3919 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3920 BTRFS_BLOCK_GROUP_RAID10 |
3921 BTRFS_BLOCK_GROUP_RAID5 |
3922 BTRFS_BLOCK_GROUP_RAID6;
3924 seq = read_seqbegin(&fs_info->profiles_lock);
3926 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3927 (fs_info->avail_system_alloc_bits & allowed) &&
3928 !(bctl->sys.target & allowed)) ||
3929 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3930 (fs_info->avail_metadata_alloc_bits & allowed) &&
3931 !(bctl->meta.target & allowed))) {
3932 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3934 "force reducing metadata integrity");
3937 "balance will reduce metadata integrity, use force if you want this");
3942 } while (read_seqretry(&fs_info->profiles_lock, seq));
3944 /* if we're not converting, the target field is uninitialized */
3945 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3946 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3947 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3948 bctl->data.target : fs_info->avail_data_alloc_bits;
3949 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3950 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3952 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3953 meta_target, data_target);
3956 ret = insert_balance_item(fs_info, bctl);
3957 if (ret && ret != -EEXIST)
3960 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3961 BUG_ON(ret == -EEXIST);
3962 set_balance_control(bctl);
3964 BUG_ON(ret != -EEXIST);
3965 spin_lock(&fs_info->balance_lock);
3966 update_balance_args(bctl);
3967 spin_unlock(&fs_info->balance_lock);
3970 atomic_inc(&fs_info->balance_running);
3971 mutex_unlock(&fs_info->balance_mutex);
3973 ret = __btrfs_balance(fs_info);
3975 mutex_lock(&fs_info->balance_mutex);
3976 atomic_dec(&fs_info->balance_running);
3979 memset(bargs, 0, sizeof(*bargs));
3980 update_ioctl_balance_args(fs_info, 0, bargs);
3983 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3984 balance_need_close(fs_info)) {
3985 __cancel_balance(fs_info);
3988 wake_up(&fs_info->balance_wait_q);
3992 if (bctl->flags & BTRFS_BALANCE_RESUME)
3993 __cancel_balance(fs_info);
3996 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4001 static int balance_kthread(void *data)
4003 struct btrfs_fs_info *fs_info = data;
4006 mutex_lock(&fs_info->volume_mutex);
4007 mutex_lock(&fs_info->balance_mutex);
4009 if (fs_info->balance_ctl) {
4010 btrfs_info(fs_info, "continuing balance");
4011 ret = btrfs_balance(fs_info->balance_ctl, NULL);
4014 mutex_unlock(&fs_info->balance_mutex);
4015 mutex_unlock(&fs_info->volume_mutex);
4020 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4022 struct task_struct *tsk;
4024 spin_lock(&fs_info->balance_lock);
4025 if (!fs_info->balance_ctl) {
4026 spin_unlock(&fs_info->balance_lock);
4029 spin_unlock(&fs_info->balance_lock);
4031 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4032 btrfs_info(fs_info, "force skipping balance");
4036 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4037 return PTR_ERR_OR_ZERO(tsk);
4040 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4042 struct btrfs_balance_control *bctl;
4043 struct btrfs_balance_item *item;
4044 struct btrfs_disk_balance_args disk_bargs;
4045 struct btrfs_path *path;
4046 struct extent_buffer *leaf;
4047 struct btrfs_key key;
4050 path = btrfs_alloc_path();
4054 key.objectid = BTRFS_BALANCE_OBJECTID;
4055 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4058 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4061 if (ret > 0) { /* ret = -ENOENT; */
4066 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4072 leaf = path->nodes[0];
4073 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4075 bctl->fs_info = fs_info;
4076 bctl->flags = btrfs_balance_flags(leaf, item);
4077 bctl->flags |= BTRFS_BALANCE_RESUME;
4079 btrfs_balance_data(leaf, item, &disk_bargs);
4080 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4081 btrfs_balance_meta(leaf, item, &disk_bargs);
4082 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4083 btrfs_balance_sys(leaf, item, &disk_bargs);
4084 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4086 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4088 mutex_lock(&fs_info->volume_mutex);
4089 mutex_lock(&fs_info->balance_mutex);
4091 set_balance_control(bctl);
4093 mutex_unlock(&fs_info->balance_mutex);
4094 mutex_unlock(&fs_info->volume_mutex);
4096 btrfs_free_path(path);
4100 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4104 mutex_lock(&fs_info->balance_mutex);
4105 if (!fs_info->balance_ctl) {
4106 mutex_unlock(&fs_info->balance_mutex);
4110 if (atomic_read(&fs_info->balance_running)) {
4111 atomic_inc(&fs_info->balance_pause_req);
4112 mutex_unlock(&fs_info->balance_mutex);
4114 wait_event(fs_info->balance_wait_q,
4115 atomic_read(&fs_info->balance_running) == 0);
4117 mutex_lock(&fs_info->balance_mutex);
4118 /* we are good with balance_ctl ripped off from under us */
4119 BUG_ON(atomic_read(&fs_info->balance_running));
4120 atomic_dec(&fs_info->balance_pause_req);
4125 mutex_unlock(&fs_info->balance_mutex);
4129 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4131 if (sb_rdonly(fs_info->sb))
4134 mutex_lock(&fs_info->balance_mutex);
4135 if (!fs_info->balance_ctl) {
4136 mutex_unlock(&fs_info->balance_mutex);
4140 atomic_inc(&fs_info->balance_cancel_req);
4142 * if we are running just wait and return, balance item is
4143 * deleted in btrfs_balance in this case
4145 if (atomic_read(&fs_info->balance_running)) {
4146 mutex_unlock(&fs_info->balance_mutex);
4147 wait_event(fs_info->balance_wait_q,
4148 atomic_read(&fs_info->balance_running) == 0);
4149 mutex_lock(&fs_info->balance_mutex);
4151 /* __cancel_balance needs volume_mutex */
4152 mutex_unlock(&fs_info->balance_mutex);
4153 mutex_lock(&fs_info->volume_mutex);
4154 mutex_lock(&fs_info->balance_mutex);
4156 if (fs_info->balance_ctl)
4157 __cancel_balance(fs_info);
4159 mutex_unlock(&fs_info->volume_mutex);
4162 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4163 atomic_dec(&fs_info->balance_cancel_req);
4164 mutex_unlock(&fs_info->balance_mutex);
4168 static int btrfs_uuid_scan_kthread(void *data)
4170 struct btrfs_fs_info *fs_info = data;
4171 struct btrfs_root *root = fs_info->tree_root;
4172 struct btrfs_key key;
4173 struct btrfs_path *path = NULL;
4175 struct extent_buffer *eb;
4177 struct btrfs_root_item root_item;
4179 struct btrfs_trans_handle *trans = NULL;
4181 path = btrfs_alloc_path();
4188 key.type = BTRFS_ROOT_ITEM_KEY;
4192 ret = btrfs_search_forward(root, &key, path, 0);
4199 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4200 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4201 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4202 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4205 eb = path->nodes[0];
4206 slot = path->slots[0];
4207 item_size = btrfs_item_size_nr(eb, slot);
4208 if (item_size < sizeof(root_item))
4211 read_extent_buffer(eb, &root_item,
4212 btrfs_item_ptr_offset(eb, slot),
4213 (int)sizeof(root_item));
4214 if (btrfs_root_refs(&root_item) == 0)
4217 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4218 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4222 btrfs_release_path(path);
4224 * 1 - subvol uuid item
4225 * 1 - received_subvol uuid item
4227 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4228 if (IS_ERR(trans)) {
4229 ret = PTR_ERR(trans);
4237 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4238 ret = btrfs_uuid_tree_add(trans, fs_info,
4240 BTRFS_UUID_KEY_SUBVOL,
4243 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4249 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4250 ret = btrfs_uuid_tree_add(trans, fs_info,
4251 root_item.received_uuid,
4252 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4255 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4263 ret = btrfs_end_transaction(trans);
4269 btrfs_release_path(path);
4270 if (key.offset < (u64)-1) {
4272 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4274 key.type = BTRFS_ROOT_ITEM_KEY;
4275 } else if (key.objectid < (u64)-1) {
4277 key.type = BTRFS_ROOT_ITEM_KEY;
4286 btrfs_free_path(path);
4287 if (trans && !IS_ERR(trans))
4288 btrfs_end_transaction(trans);
4290 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4292 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4293 up(&fs_info->uuid_tree_rescan_sem);
4298 * Callback for btrfs_uuid_tree_iterate().
4300 * 0 check succeeded, the entry is not outdated.
4301 * < 0 if an error occurred.
4302 * > 0 if the check failed, which means the caller shall remove the entry.
4304 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4305 u8 *uuid, u8 type, u64 subid)
4307 struct btrfs_key key;
4309 struct btrfs_root *subvol_root;
4311 if (type != BTRFS_UUID_KEY_SUBVOL &&
4312 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4315 key.objectid = subid;
4316 key.type = BTRFS_ROOT_ITEM_KEY;
4317 key.offset = (u64)-1;
4318 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4319 if (IS_ERR(subvol_root)) {
4320 ret = PTR_ERR(subvol_root);
4327 case BTRFS_UUID_KEY_SUBVOL:
4328 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4331 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4332 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4342 static int btrfs_uuid_rescan_kthread(void *data)
4344 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4348 * 1st step is to iterate through the existing UUID tree and
4349 * to delete all entries that contain outdated data.
4350 * 2nd step is to add all missing entries to the UUID tree.
4352 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4354 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4355 up(&fs_info->uuid_tree_rescan_sem);
4358 return btrfs_uuid_scan_kthread(data);
4361 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4363 struct btrfs_trans_handle *trans;
4364 struct btrfs_root *tree_root = fs_info->tree_root;
4365 struct btrfs_root *uuid_root;
4366 struct task_struct *task;
4373 trans = btrfs_start_transaction(tree_root, 2);
4375 return PTR_ERR(trans);
4377 uuid_root = btrfs_create_tree(trans, fs_info,
4378 BTRFS_UUID_TREE_OBJECTID);
4379 if (IS_ERR(uuid_root)) {
4380 ret = PTR_ERR(uuid_root);
4381 btrfs_abort_transaction(trans, ret);
4382 btrfs_end_transaction(trans);
4386 fs_info->uuid_root = uuid_root;
4388 ret = btrfs_commit_transaction(trans);
4392 down(&fs_info->uuid_tree_rescan_sem);
4393 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4395 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4396 btrfs_warn(fs_info, "failed to start uuid_scan task");
4397 up(&fs_info->uuid_tree_rescan_sem);
4398 return PTR_ERR(task);
4404 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4406 struct task_struct *task;
4408 down(&fs_info->uuid_tree_rescan_sem);
4409 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4411 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4412 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4413 up(&fs_info->uuid_tree_rescan_sem);
4414 return PTR_ERR(task);
4421 * shrinking a device means finding all of the device extents past
4422 * the new size, and then following the back refs to the chunks.
4423 * The chunk relocation code actually frees the device extent
4425 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4427 struct btrfs_fs_info *fs_info = device->fs_info;
4428 struct btrfs_root *root = fs_info->dev_root;
4429 struct btrfs_trans_handle *trans;
4430 struct btrfs_dev_extent *dev_extent = NULL;
4431 struct btrfs_path *path;
4437 bool retried = false;
4438 bool checked_pending_chunks = false;
4439 struct extent_buffer *l;
4440 struct btrfs_key key;
4441 struct btrfs_super_block *super_copy = fs_info->super_copy;
4442 u64 old_total = btrfs_super_total_bytes(super_copy);
4443 u64 old_size = btrfs_device_get_total_bytes(device);
4446 new_size = round_down(new_size, fs_info->sectorsize);
4447 diff = round_down(old_size - new_size, fs_info->sectorsize);
4449 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4452 path = btrfs_alloc_path();
4456 path->reada = READA_FORWARD;
4458 mutex_lock(&fs_info->chunk_mutex);
4460 btrfs_device_set_total_bytes(device, new_size);
4461 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4462 device->fs_devices->total_rw_bytes -= diff;
4463 atomic64_sub(diff, &fs_info->free_chunk_space);
4465 mutex_unlock(&fs_info->chunk_mutex);
4468 key.objectid = device->devid;
4469 key.offset = (u64)-1;
4470 key.type = BTRFS_DEV_EXTENT_KEY;
4473 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4474 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4476 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4480 ret = btrfs_previous_item(root, path, 0, key.type);
4482 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4487 btrfs_release_path(path);
4492 slot = path->slots[0];
4493 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4495 if (key.objectid != device->devid) {
4496 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4497 btrfs_release_path(path);
4501 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4502 length = btrfs_dev_extent_length(l, dev_extent);
4504 if (key.offset + length <= new_size) {
4505 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4506 btrfs_release_path(path);
4510 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4511 btrfs_release_path(path);
4514 * We may be relocating the only data chunk we have,
4515 * which could potentially end up with losing data's
4516 * raid profile, so lets allocate an empty one in
4519 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4521 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4525 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4526 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4527 if (ret && ret != -ENOSPC)
4531 } while (key.offset-- > 0);
4533 if (failed && !retried) {
4537 } else if (failed && retried) {
4542 /* Shrinking succeeded, else we would be at "done". */
4543 trans = btrfs_start_transaction(root, 0);
4544 if (IS_ERR(trans)) {
4545 ret = PTR_ERR(trans);
4549 mutex_lock(&fs_info->chunk_mutex);
4552 * We checked in the above loop all device extents that were already in
4553 * the device tree. However before we have updated the device's
4554 * total_bytes to the new size, we might have had chunk allocations that
4555 * have not complete yet (new block groups attached to transaction
4556 * handles), and therefore their device extents were not yet in the
4557 * device tree and we missed them in the loop above. So if we have any
4558 * pending chunk using a device extent that overlaps the device range
4559 * that we can not use anymore, commit the current transaction and
4560 * repeat the search on the device tree - this way we guarantee we will
4561 * not have chunks using device extents that end beyond 'new_size'.
4563 if (!checked_pending_chunks) {
4564 u64 start = new_size;
4565 u64 len = old_size - new_size;
4567 if (contains_pending_extent(trans->transaction, device,
4569 mutex_unlock(&fs_info->chunk_mutex);
4570 checked_pending_chunks = true;
4573 ret = btrfs_commit_transaction(trans);
4580 btrfs_device_set_disk_total_bytes(device, new_size);
4581 if (list_empty(&device->resized_list))
4582 list_add_tail(&device->resized_list,
4583 &fs_info->fs_devices->resized_devices);
4585 WARN_ON(diff > old_total);
4586 btrfs_set_super_total_bytes(super_copy,
4587 round_down(old_total - diff, fs_info->sectorsize));
4588 mutex_unlock(&fs_info->chunk_mutex);
4590 /* Now btrfs_update_device() will change the on-disk size. */
4591 ret = btrfs_update_device(trans, device);
4592 btrfs_end_transaction(trans);
4594 btrfs_free_path(path);
4596 mutex_lock(&fs_info->chunk_mutex);
4597 btrfs_device_set_total_bytes(device, old_size);
4598 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4599 device->fs_devices->total_rw_bytes += diff;
4600 atomic64_add(diff, &fs_info->free_chunk_space);
4601 mutex_unlock(&fs_info->chunk_mutex);
4606 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4607 struct btrfs_key *key,
4608 struct btrfs_chunk *chunk, int item_size)
4610 struct btrfs_super_block *super_copy = fs_info->super_copy;
4611 struct btrfs_disk_key disk_key;
4615 mutex_lock(&fs_info->chunk_mutex);
4616 array_size = btrfs_super_sys_array_size(super_copy);
4617 if (array_size + item_size + sizeof(disk_key)
4618 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4619 mutex_unlock(&fs_info->chunk_mutex);
4623 ptr = super_copy->sys_chunk_array + array_size;
4624 btrfs_cpu_key_to_disk(&disk_key, key);
4625 memcpy(ptr, &disk_key, sizeof(disk_key));
4626 ptr += sizeof(disk_key);
4627 memcpy(ptr, chunk, item_size);
4628 item_size += sizeof(disk_key);
4629 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4630 mutex_unlock(&fs_info->chunk_mutex);
4636 * sort the devices in descending order by max_avail, total_avail
4638 static int btrfs_cmp_device_info(const void *a, const void *b)
4640 const struct btrfs_device_info *di_a = a;
4641 const struct btrfs_device_info *di_b = b;
4643 if (di_a->max_avail > di_b->max_avail)
4645 if (di_a->max_avail < di_b->max_avail)
4647 if (di_a->total_avail > di_b->total_avail)
4649 if (di_a->total_avail < di_b->total_avail)
4654 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4656 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4659 btrfs_set_fs_incompat(info, RAID56);
4662 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4663 - sizeof(struct btrfs_chunk)) \
4664 / sizeof(struct btrfs_stripe) + 1)
4666 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4667 - 2 * sizeof(struct btrfs_disk_key) \
4668 - 2 * sizeof(struct btrfs_chunk)) \
4669 / sizeof(struct btrfs_stripe) + 1)
4671 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4672 u64 start, u64 type)
4674 struct btrfs_fs_info *info = trans->fs_info;
4675 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4676 struct btrfs_device *device;
4677 struct map_lookup *map = NULL;
4678 struct extent_map_tree *em_tree;
4679 struct extent_map *em;
4680 struct btrfs_device_info *devices_info = NULL;
4682 int num_stripes; /* total number of stripes to allocate */
4683 int data_stripes; /* number of stripes that count for
4685 int sub_stripes; /* sub_stripes info for map */
4686 int dev_stripes; /* stripes per dev */
4687 int devs_max; /* max devs to use */
4688 int devs_min; /* min devs needed */
4689 int devs_increment; /* ndevs has to be a multiple of this */
4690 int ncopies; /* how many copies to data has */
4692 u64 max_stripe_size;
4701 BUG_ON(!alloc_profile_is_valid(type, 0));
4703 if (list_empty(&fs_devices->alloc_list)) {
4704 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4705 btrfs_debug(info, "%s: no writable device", __func__);
4709 index = __get_raid_index(type);
4711 sub_stripes = btrfs_raid_array[index].sub_stripes;
4712 dev_stripes = btrfs_raid_array[index].dev_stripes;
4713 devs_max = btrfs_raid_array[index].devs_max;
4714 devs_min = btrfs_raid_array[index].devs_min;
4715 devs_increment = btrfs_raid_array[index].devs_increment;
4716 ncopies = btrfs_raid_array[index].ncopies;
4718 if (type & BTRFS_BLOCK_GROUP_DATA) {
4719 max_stripe_size = SZ_1G;
4720 max_chunk_size = 10 * max_stripe_size;
4722 devs_max = BTRFS_MAX_DEVS(info);
4723 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4724 /* for larger filesystems, use larger metadata chunks */
4725 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4726 max_stripe_size = SZ_1G;
4728 max_stripe_size = SZ_256M;
4729 max_chunk_size = max_stripe_size;
4731 devs_max = BTRFS_MAX_DEVS(info);
4732 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4733 max_stripe_size = SZ_32M;
4734 max_chunk_size = 2 * max_stripe_size;
4736 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4738 btrfs_err(info, "invalid chunk type 0x%llx requested",
4743 /* we don't want a chunk larger than 10% of writeable space */
4744 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4747 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4753 * in the first pass through the devices list, we gather information
4754 * about the available holes on each device.
4757 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4761 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4763 "BTRFS: read-only device in alloc_list\n");
4767 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4768 &device->dev_state) ||
4769 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4772 if (device->total_bytes > device->bytes_used)
4773 total_avail = device->total_bytes - device->bytes_used;
4777 /* If there is no space on this device, skip it. */
4778 if (total_avail == 0)
4781 ret = find_free_dev_extent(trans, device,
4782 max_stripe_size * dev_stripes,
4783 &dev_offset, &max_avail);
4784 if (ret && ret != -ENOSPC)
4788 max_avail = max_stripe_size * dev_stripes;
4790 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4791 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4793 "%s: devid %llu has no free space, have=%llu want=%u",
4794 __func__, device->devid, max_avail,
4795 BTRFS_STRIPE_LEN * dev_stripes);
4799 if (ndevs == fs_devices->rw_devices) {
4800 WARN(1, "%s: found more than %llu devices\n",
4801 __func__, fs_devices->rw_devices);
4804 devices_info[ndevs].dev_offset = dev_offset;
4805 devices_info[ndevs].max_avail = max_avail;
4806 devices_info[ndevs].total_avail = total_avail;
4807 devices_info[ndevs].dev = device;
4812 * now sort the devices by hole size / available space
4814 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4815 btrfs_cmp_device_info, NULL);
4817 /* round down to number of usable stripes */
4818 ndevs = round_down(ndevs, devs_increment);
4820 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4822 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4824 "%s: not enough devices with free space: have=%d minimum required=%d",
4825 __func__, ndevs, min(devs_min,
4826 devs_increment * sub_stripes));
4831 ndevs = min(ndevs, devs_max);
4834 * The primary goal is to maximize the number of stripes, so use as
4835 * many devices as possible, even if the stripes are not maximum sized.
4837 * The DUP profile stores more than one stripe per device, the
4838 * max_avail is the total size so we have to adjust.
4840 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4841 num_stripes = ndevs * dev_stripes;
4844 * this will have to be fixed for RAID1 and RAID10 over
4847 data_stripes = num_stripes / ncopies;
4849 if (type & BTRFS_BLOCK_GROUP_RAID5)
4850 data_stripes = num_stripes - 1;
4852 if (type & BTRFS_BLOCK_GROUP_RAID6)
4853 data_stripes = num_stripes - 2;
4856 * Use the number of data stripes to figure out how big this chunk
4857 * is really going to be in terms of logical address space,
4858 * and compare that answer with the max chunk size
4860 if (stripe_size * data_stripes > max_chunk_size) {
4861 u64 mask = (1ULL << 24) - 1;
4863 stripe_size = div_u64(max_chunk_size, data_stripes);
4865 /* bump the answer up to a 16MB boundary */
4866 stripe_size = (stripe_size + mask) & ~mask;
4868 /* but don't go higher than the limits we found
4869 * while searching for free extents
4871 if (stripe_size > devices_info[ndevs-1].max_avail)
4872 stripe_size = devices_info[ndevs-1].max_avail;
4875 /* align to BTRFS_STRIPE_LEN */
4876 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4878 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4883 map->num_stripes = num_stripes;
4885 for (i = 0; i < ndevs; ++i) {
4886 for (j = 0; j < dev_stripes; ++j) {
4887 int s = i * dev_stripes + j;
4888 map->stripes[s].dev = devices_info[i].dev;
4889 map->stripes[s].physical = devices_info[i].dev_offset +
4893 map->stripe_len = BTRFS_STRIPE_LEN;
4894 map->io_align = BTRFS_STRIPE_LEN;
4895 map->io_width = BTRFS_STRIPE_LEN;
4897 map->sub_stripes = sub_stripes;
4899 num_bytes = stripe_size * data_stripes;
4901 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4903 em = alloc_extent_map();
4909 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4910 em->map_lookup = map;
4912 em->len = num_bytes;
4913 em->block_start = 0;
4914 em->block_len = em->len;
4915 em->orig_block_len = stripe_size;
4917 em_tree = &info->mapping_tree.map_tree;
4918 write_lock(&em_tree->lock);
4919 ret = add_extent_mapping(em_tree, em, 0);
4921 write_unlock(&em_tree->lock);
4922 free_extent_map(em);
4926 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4927 refcount_inc(&em->refs);
4928 write_unlock(&em_tree->lock);
4930 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4932 goto error_del_extent;
4934 for (i = 0; i < map->num_stripes; i++) {
4935 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4936 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4939 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4941 free_extent_map(em);
4942 check_raid56_incompat_flag(info, type);
4944 kfree(devices_info);
4948 write_lock(&em_tree->lock);
4949 remove_extent_mapping(em_tree, em);
4950 write_unlock(&em_tree->lock);
4952 /* One for our allocation */
4953 free_extent_map(em);
4954 /* One for the tree reference */
4955 free_extent_map(em);
4956 /* One for the pending_chunks list reference */
4957 free_extent_map(em);
4959 kfree(devices_info);
4963 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4964 struct btrfs_fs_info *fs_info,
4965 u64 chunk_offset, u64 chunk_size)
4967 struct btrfs_root *extent_root = fs_info->extent_root;
4968 struct btrfs_root *chunk_root = fs_info->chunk_root;
4969 struct btrfs_key key;
4970 struct btrfs_device *device;
4971 struct btrfs_chunk *chunk;
4972 struct btrfs_stripe *stripe;
4973 struct extent_map *em;
4974 struct map_lookup *map;
4981 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4985 map = em->map_lookup;
4986 item_size = btrfs_chunk_item_size(map->num_stripes);
4987 stripe_size = em->orig_block_len;
4989 chunk = kzalloc(item_size, GFP_NOFS);
4996 * Take the device list mutex to prevent races with the final phase of
4997 * a device replace operation that replaces the device object associated
4998 * with the map's stripes, because the device object's id can change
4999 * at any time during that final phase of the device replace operation
5000 * (dev-replace.c:btrfs_dev_replace_finishing()).
5002 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5003 for (i = 0; i < map->num_stripes; i++) {
5004 device = map->stripes[i].dev;
5005 dev_offset = map->stripes[i].physical;
5007 ret = btrfs_update_device(trans, device);
5010 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5011 dev_offset, stripe_size);
5016 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5020 stripe = &chunk->stripe;
5021 for (i = 0; i < map->num_stripes; i++) {
5022 device = map->stripes[i].dev;
5023 dev_offset = map->stripes[i].physical;
5025 btrfs_set_stack_stripe_devid(stripe, device->devid);
5026 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5027 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5030 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5032 btrfs_set_stack_chunk_length(chunk, chunk_size);
5033 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5034 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5035 btrfs_set_stack_chunk_type(chunk, map->type);
5036 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5037 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5038 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5039 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5040 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5042 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5043 key.type = BTRFS_CHUNK_ITEM_KEY;
5044 key.offset = chunk_offset;
5046 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5047 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5049 * TODO: Cleanup of inserted chunk root in case of
5052 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5057 free_extent_map(em);
5062 * Chunk allocation falls into two parts. The first part does works
5063 * that make the new allocated chunk useable, but not do any operation
5064 * that modifies the chunk tree. The second part does the works that
5065 * require modifying the chunk tree. This division is important for the
5066 * bootstrap process of adding storage to a seed btrfs.
5068 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5069 struct btrfs_fs_info *fs_info, u64 type)
5073 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5074 chunk_offset = find_next_chunk(fs_info);
5075 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5078 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5079 struct btrfs_fs_info *fs_info)
5082 u64 sys_chunk_offset;
5086 chunk_offset = find_next_chunk(fs_info);
5087 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5088 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5092 sys_chunk_offset = find_next_chunk(fs_info);
5093 alloc_profile = btrfs_system_alloc_profile(fs_info);
5094 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5098 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5102 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5103 BTRFS_BLOCK_GROUP_RAID10 |
5104 BTRFS_BLOCK_GROUP_RAID5 |
5105 BTRFS_BLOCK_GROUP_DUP)) {
5107 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5116 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5118 struct extent_map *em;
5119 struct map_lookup *map;
5124 em = get_chunk_map(fs_info, chunk_offset, 1);
5128 map = em->map_lookup;
5129 for (i = 0; i < map->num_stripes; i++) {
5130 if (test_bit(BTRFS_DEV_STATE_MISSING,
5131 &map->stripes[i].dev->dev_state)) {
5135 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5136 &map->stripes[i].dev->dev_state)) {
5143 * If the number of missing devices is larger than max errors,
5144 * we can not write the data into that chunk successfully, so
5147 if (miss_ndevs > btrfs_chunk_max_errors(map))
5150 free_extent_map(em);
5154 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5156 extent_map_tree_init(&tree->map_tree);
5159 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5161 struct extent_map *em;
5164 write_lock(&tree->map_tree.lock);
5165 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5167 remove_extent_mapping(&tree->map_tree, em);
5168 write_unlock(&tree->map_tree.lock);
5172 free_extent_map(em);
5173 /* once for the tree */
5174 free_extent_map(em);
5178 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5180 struct extent_map *em;
5181 struct map_lookup *map;
5184 em = get_chunk_map(fs_info, logical, len);
5187 * We could return errors for these cases, but that could get
5188 * ugly and we'd probably do the same thing which is just not do
5189 * anything else and exit, so return 1 so the callers don't try
5190 * to use other copies.
5194 map = em->map_lookup;
5195 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5196 ret = map->num_stripes;
5197 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5198 ret = map->sub_stripes;
5199 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5201 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5203 * There could be two corrupted data stripes, we need
5204 * to loop retry in order to rebuild the correct data.
5206 * Fail a stripe at a time on every retry except the
5207 * stripe under reconstruction.
5209 ret = map->num_stripes;
5212 free_extent_map(em);
5214 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5215 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5216 fs_info->dev_replace.tgtdev)
5218 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5223 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5226 struct extent_map *em;
5227 struct map_lookup *map;
5228 unsigned long len = fs_info->sectorsize;
5230 em = get_chunk_map(fs_info, logical, len);
5232 if (!WARN_ON(IS_ERR(em))) {
5233 map = em->map_lookup;
5234 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5235 len = map->stripe_len * nr_data_stripes(map);
5236 free_extent_map(em);
5241 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5243 struct extent_map *em;
5244 struct map_lookup *map;
5247 em = get_chunk_map(fs_info, logical, len);
5249 if(!WARN_ON(IS_ERR(em))) {
5250 map = em->map_lookup;
5251 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5253 free_extent_map(em);
5258 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5259 struct map_lookup *map, int first, int num,
5260 int optimal, int dev_replace_is_ongoing)
5264 struct btrfs_device *srcdev;
5266 if (dev_replace_is_ongoing &&
5267 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5268 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5269 srcdev = fs_info->dev_replace.srcdev;
5274 * try to avoid the drive that is the source drive for a
5275 * dev-replace procedure, only choose it if no other non-missing
5276 * mirror is available
5278 for (tolerance = 0; tolerance < 2; tolerance++) {
5279 if (map->stripes[optimal].dev->bdev &&
5280 (tolerance || map->stripes[optimal].dev != srcdev))
5282 for (i = first; i < first + num; i++) {
5283 if (map->stripes[i].dev->bdev &&
5284 (tolerance || map->stripes[i].dev != srcdev))
5289 /* we couldn't find one that doesn't fail. Just return something
5290 * and the io error handling code will clean up eventually
5295 static inline int parity_smaller(u64 a, u64 b)
5300 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5301 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5303 struct btrfs_bio_stripe s;
5310 for (i = 0; i < num_stripes - 1; i++) {
5311 if (parity_smaller(bbio->raid_map[i],
5312 bbio->raid_map[i+1])) {
5313 s = bbio->stripes[i];
5314 l = bbio->raid_map[i];
5315 bbio->stripes[i] = bbio->stripes[i+1];
5316 bbio->raid_map[i] = bbio->raid_map[i+1];
5317 bbio->stripes[i+1] = s;
5318 bbio->raid_map[i+1] = l;
5326 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5328 struct btrfs_bio *bbio = kzalloc(
5329 /* the size of the btrfs_bio */
5330 sizeof(struct btrfs_bio) +
5331 /* plus the variable array for the stripes */
5332 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5333 /* plus the variable array for the tgt dev */
5334 sizeof(int) * (real_stripes) +
5336 * plus the raid_map, which includes both the tgt dev
5339 sizeof(u64) * (total_stripes),
5340 GFP_NOFS|__GFP_NOFAIL);
5342 atomic_set(&bbio->error, 0);
5343 refcount_set(&bbio->refs, 1);
5348 void btrfs_get_bbio(struct btrfs_bio *bbio)
5350 WARN_ON(!refcount_read(&bbio->refs));
5351 refcount_inc(&bbio->refs);
5354 void btrfs_put_bbio(struct btrfs_bio *bbio)
5358 if (refcount_dec_and_test(&bbio->refs))
5362 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5364 * Please note that, discard won't be sent to target device of device
5367 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5368 u64 logical, u64 length,
5369 struct btrfs_bio **bbio_ret)
5371 struct extent_map *em;
5372 struct map_lookup *map;
5373 struct btrfs_bio *bbio;
5377 u64 stripe_end_offset;
5384 u32 sub_stripes = 0;
5385 u64 stripes_per_dev = 0;
5386 u32 remaining_stripes = 0;
5387 u32 last_stripe = 0;
5391 /* discard always return a bbio */
5394 em = get_chunk_map(fs_info, logical, length);
5398 map = em->map_lookup;
5399 /* we don't discard raid56 yet */
5400 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5405 offset = logical - em->start;
5406 length = min_t(u64, em->len - offset, length);
5408 stripe_len = map->stripe_len;
5410 * stripe_nr counts the total number of stripes we have to stride
5411 * to get to this block
5413 stripe_nr = div64_u64(offset, stripe_len);
5415 /* stripe_offset is the offset of this block in its stripe */
5416 stripe_offset = offset - stripe_nr * stripe_len;
5418 stripe_nr_end = round_up(offset + length, map->stripe_len);
5419 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5420 stripe_cnt = stripe_nr_end - stripe_nr;
5421 stripe_end_offset = stripe_nr_end * map->stripe_len -
5424 * after this, stripe_nr is the number of stripes on this
5425 * device we have to walk to find the data, and stripe_index is
5426 * the number of our device in the stripe array
5430 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5431 BTRFS_BLOCK_GROUP_RAID10)) {
5432 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5435 sub_stripes = map->sub_stripes;
5437 factor = map->num_stripes / sub_stripes;
5438 num_stripes = min_t(u64, map->num_stripes,
5439 sub_stripes * stripe_cnt);
5440 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5441 stripe_index *= sub_stripes;
5442 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5443 &remaining_stripes);
5444 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5445 last_stripe *= sub_stripes;
5446 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5447 BTRFS_BLOCK_GROUP_DUP)) {
5448 num_stripes = map->num_stripes;
5450 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5454 bbio = alloc_btrfs_bio(num_stripes, 0);
5460 for (i = 0; i < num_stripes; i++) {
5461 bbio->stripes[i].physical =
5462 map->stripes[stripe_index].physical +
5463 stripe_offset + stripe_nr * map->stripe_len;
5464 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5466 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5467 BTRFS_BLOCK_GROUP_RAID10)) {
5468 bbio->stripes[i].length = stripes_per_dev *
5471 if (i / sub_stripes < remaining_stripes)
5472 bbio->stripes[i].length +=
5476 * Special for the first stripe and
5479 * |-------|...|-------|
5483 if (i < sub_stripes)
5484 bbio->stripes[i].length -=
5487 if (stripe_index >= last_stripe &&
5488 stripe_index <= (last_stripe +
5490 bbio->stripes[i].length -=
5493 if (i == sub_stripes - 1)
5496 bbio->stripes[i].length = length;
5500 if (stripe_index == map->num_stripes) {
5507 bbio->map_type = map->type;
5508 bbio->num_stripes = num_stripes;
5510 free_extent_map(em);
5515 * In dev-replace case, for repair case (that's the only case where the mirror
5516 * is selected explicitly when calling btrfs_map_block), blocks left of the
5517 * left cursor can also be read from the target drive.
5519 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5521 * For READ, it also needs to be supported using the same mirror number.
5523 * If the requested block is not left of the left cursor, EIO is returned. This
5524 * can happen because btrfs_num_copies() returns one more in the dev-replace
5527 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5528 u64 logical, u64 length,
5529 u64 srcdev_devid, int *mirror_num,
5532 struct btrfs_bio *bbio = NULL;
5534 int index_srcdev = 0;
5536 u64 physical_of_found = 0;
5540 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5541 logical, &length, &bbio, 0, 0);
5543 ASSERT(bbio == NULL);
5547 num_stripes = bbio->num_stripes;
5548 if (*mirror_num > num_stripes) {
5550 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5551 * that means that the requested area is not left of the left
5554 btrfs_put_bbio(bbio);
5559 * process the rest of the function using the mirror_num of the source
5560 * drive. Therefore look it up first. At the end, patch the device
5561 * pointer to the one of the target drive.
5563 for (i = 0; i < num_stripes; i++) {
5564 if (bbio->stripes[i].dev->devid != srcdev_devid)
5568 * In case of DUP, in order to keep it simple, only add the
5569 * mirror with the lowest physical address
5572 physical_of_found <= bbio->stripes[i].physical)
5577 physical_of_found = bbio->stripes[i].physical;
5580 btrfs_put_bbio(bbio);
5586 *mirror_num = index_srcdev + 1;
5587 *physical = physical_of_found;
5591 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5592 struct btrfs_bio **bbio_ret,
5593 struct btrfs_dev_replace *dev_replace,
5594 int *num_stripes_ret, int *max_errors_ret)
5596 struct btrfs_bio *bbio = *bbio_ret;
5597 u64 srcdev_devid = dev_replace->srcdev->devid;
5598 int tgtdev_indexes = 0;
5599 int num_stripes = *num_stripes_ret;
5600 int max_errors = *max_errors_ret;
5603 if (op == BTRFS_MAP_WRITE) {
5604 int index_where_to_add;
5607 * duplicate the write operations while the dev replace
5608 * procedure is running. Since the copying of the old disk to
5609 * the new disk takes place at run time while the filesystem is
5610 * mounted writable, the regular write operations to the old
5611 * disk have to be duplicated to go to the new disk as well.
5613 * Note that device->missing is handled by the caller, and that
5614 * the write to the old disk is already set up in the stripes
5617 index_where_to_add = num_stripes;
5618 for (i = 0; i < num_stripes; i++) {
5619 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5620 /* write to new disk, too */
5621 struct btrfs_bio_stripe *new =
5622 bbio->stripes + index_where_to_add;
5623 struct btrfs_bio_stripe *old =
5626 new->physical = old->physical;
5627 new->length = old->length;
5628 new->dev = dev_replace->tgtdev;
5629 bbio->tgtdev_map[i] = index_where_to_add;
5630 index_where_to_add++;
5635 num_stripes = index_where_to_add;
5636 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5637 int index_srcdev = 0;
5639 u64 physical_of_found = 0;
5642 * During the dev-replace procedure, the target drive can also
5643 * be used to read data in case it is needed to repair a corrupt
5644 * block elsewhere. This is possible if the requested area is
5645 * left of the left cursor. In this area, the target drive is a
5646 * full copy of the source drive.
5648 for (i = 0; i < num_stripes; i++) {
5649 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5651 * In case of DUP, in order to keep it simple,
5652 * only add the mirror with the lowest physical
5656 physical_of_found <=
5657 bbio->stripes[i].physical)
5661 physical_of_found = bbio->stripes[i].physical;
5665 struct btrfs_bio_stripe *tgtdev_stripe =
5666 bbio->stripes + num_stripes;
5668 tgtdev_stripe->physical = physical_of_found;
5669 tgtdev_stripe->length =
5670 bbio->stripes[index_srcdev].length;
5671 tgtdev_stripe->dev = dev_replace->tgtdev;
5672 bbio->tgtdev_map[index_srcdev] = num_stripes;
5679 *num_stripes_ret = num_stripes;
5680 *max_errors_ret = max_errors;
5681 bbio->num_tgtdevs = tgtdev_indexes;
5685 static bool need_full_stripe(enum btrfs_map_op op)
5687 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5690 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5691 enum btrfs_map_op op,
5692 u64 logical, u64 *length,
5693 struct btrfs_bio **bbio_ret,
5694 int mirror_num, int need_raid_map)
5696 struct extent_map *em;
5697 struct map_lookup *map;
5707 int tgtdev_indexes = 0;
5708 struct btrfs_bio *bbio = NULL;
5709 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5710 int dev_replace_is_ongoing = 0;
5711 int num_alloc_stripes;
5712 int patch_the_first_stripe_for_dev_replace = 0;
5713 u64 physical_to_patch_in_first_stripe = 0;
5714 u64 raid56_full_stripe_start = (u64)-1;
5716 if (op == BTRFS_MAP_DISCARD)
5717 return __btrfs_map_block_for_discard(fs_info, logical,
5720 em = get_chunk_map(fs_info, logical, *length);
5724 map = em->map_lookup;
5725 offset = logical - em->start;
5727 stripe_len = map->stripe_len;
5730 * stripe_nr counts the total number of stripes we have to stride
5731 * to get to this block
5733 stripe_nr = div64_u64(stripe_nr, stripe_len);
5735 stripe_offset = stripe_nr * stripe_len;
5736 if (offset < stripe_offset) {
5738 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5739 stripe_offset, offset, em->start, logical,
5741 free_extent_map(em);
5745 /* stripe_offset is the offset of this block in its stripe*/
5746 stripe_offset = offset - stripe_offset;
5748 /* if we're here for raid56, we need to know the stripe aligned start */
5749 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5750 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5751 raid56_full_stripe_start = offset;
5753 /* allow a write of a full stripe, but make sure we don't
5754 * allow straddling of stripes
5756 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5758 raid56_full_stripe_start *= full_stripe_len;
5761 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5763 /* For writes to RAID[56], allow a full stripeset across all disks.
5764 For other RAID types and for RAID[56] reads, just allow a single
5765 stripe (on a single disk). */
5766 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5767 (op == BTRFS_MAP_WRITE)) {
5768 max_len = stripe_len * nr_data_stripes(map) -
5769 (offset - raid56_full_stripe_start);
5771 /* we limit the length of each bio to what fits in a stripe */
5772 max_len = stripe_len - stripe_offset;
5774 *length = min_t(u64, em->len - offset, max_len);
5776 *length = em->len - offset;
5779 /* This is for when we're called from btrfs_merge_bio_hook() and all
5780 it cares about is the length */
5784 btrfs_dev_replace_lock(dev_replace, 0);
5785 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5786 if (!dev_replace_is_ongoing)
5787 btrfs_dev_replace_unlock(dev_replace, 0);
5789 btrfs_dev_replace_set_lock_blocking(dev_replace);
5791 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5792 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5793 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5794 dev_replace->srcdev->devid,
5796 &physical_to_patch_in_first_stripe);
5800 patch_the_first_stripe_for_dev_replace = 1;
5801 } else if (mirror_num > map->num_stripes) {
5807 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5808 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5810 if (!need_full_stripe(op))
5812 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5813 if (need_full_stripe(op))
5814 num_stripes = map->num_stripes;
5815 else if (mirror_num)
5816 stripe_index = mirror_num - 1;
5818 stripe_index = find_live_mirror(fs_info, map, 0,
5820 current->pid % map->num_stripes,
5821 dev_replace_is_ongoing);
5822 mirror_num = stripe_index + 1;
5825 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5826 if (need_full_stripe(op)) {
5827 num_stripes = map->num_stripes;
5828 } else if (mirror_num) {
5829 stripe_index = mirror_num - 1;
5834 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5835 u32 factor = map->num_stripes / map->sub_stripes;
5837 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5838 stripe_index *= map->sub_stripes;
5840 if (need_full_stripe(op))
5841 num_stripes = map->sub_stripes;
5842 else if (mirror_num)
5843 stripe_index += mirror_num - 1;
5845 int old_stripe_index = stripe_index;
5846 stripe_index = find_live_mirror(fs_info, map,
5848 map->sub_stripes, stripe_index +
5849 current->pid % map->sub_stripes,
5850 dev_replace_is_ongoing);
5851 mirror_num = stripe_index - old_stripe_index + 1;
5854 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5855 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5856 /* push stripe_nr back to the start of the full stripe */
5857 stripe_nr = div64_u64(raid56_full_stripe_start,
5858 stripe_len * nr_data_stripes(map));
5860 /* RAID[56] write or recovery. Return all stripes */
5861 num_stripes = map->num_stripes;
5862 max_errors = nr_parity_stripes(map);
5864 *length = map->stripe_len;
5869 * Mirror #0 or #1 means the original data block.
5870 * Mirror #2 is RAID5 parity block.
5871 * Mirror #3 is RAID6 Q block.
5873 stripe_nr = div_u64_rem(stripe_nr,
5874 nr_data_stripes(map), &stripe_index);
5876 stripe_index = nr_data_stripes(map) +
5879 /* We distribute the parity blocks across stripes */
5880 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5882 if (!need_full_stripe(op) && mirror_num <= 1)
5887 * after this, stripe_nr is the number of stripes on this
5888 * device we have to walk to find the data, and stripe_index is
5889 * the number of our device in the stripe array
5891 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5893 mirror_num = stripe_index + 1;
5895 if (stripe_index >= map->num_stripes) {
5897 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5898 stripe_index, map->num_stripes);
5903 num_alloc_stripes = num_stripes;
5904 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5905 if (op == BTRFS_MAP_WRITE)
5906 num_alloc_stripes <<= 1;
5907 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5908 num_alloc_stripes++;
5909 tgtdev_indexes = num_stripes;
5912 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5917 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5918 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5920 /* build raid_map */
5921 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5922 (need_full_stripe(op) || mirror_num > 1)) {
5926 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5927 sizeof(struct btrfs_bio_stripe) *
5929 sizeof(int) * tgtdev_indexes);
5931 /* Work out the disk rotation on this stripe-set */
5932 div_u64_rem(stripe_nr, num_stripes, &rot);
5934 /* Fill in the logical address of each stripe */
5935 tmp = stripe_nr * nr_data_stripes(map);
5936 for (i = 0; i < nr_data_stripes(map); i++)
5937 bbio->raid_map[(i+rot) % num_stripes] =
5938 em->start + (tmp + i) * map->stripe_len;
5940 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5941 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5942 bbio->raid_map[(i+rot+1) % num_stripes] =
5947 for (i = 0; i < num_stripes; i++) {
5948 bbio->stripes[i].physical =
5949 map->stripes[stripe_index].physical +
5951 stripe_nr * map->stripe_len;
5952 bbio->stripes[i].dev =
5953 map->stripes[stripe_index].dev;
5957 if (need_full_stripe(op))
5958 max_errors = btrfs_chunk_max_errors(map);
5961 sort_parity_stripes(bbio, num_stripes);
5963 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5964 need_full_stripe(op)) {
5965 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5970 bbio->map_type = map->type;
5971 bbio->num_stripes = num_stripes;
5972 bbio->max_errors = max_errors;
5973 bbio->mirror_num = mirror_num;
5976 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5977 * mirror_num == num_stripes + 1 && dev_replace target drive is
5978 * available as a mirror
5980 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5981 WARN_ON(num_stripes > 1);
5982 bbio->stripes[0].dev = dev_replace->tgtdev;
5983 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5984 bbio->mirror_num = map->num_stripes + 1;
5987 if (dev_replace_is_ongoing) {
5988 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5989 btrfs_dev_replace_unlock(dev_replace, 0);
5991 free_extent_map(em);
5995 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5996 u64 logical, u64 *length,
5997 struct btrfs_bio **bbio_ret, int mirror_num)
5999 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6003 /* For Scrub/replace */
6004 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6005 u64 logical, u64 *length,
6006 struct btrfs_bio **bbio_ret)
6008 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6011 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
6012 u64 chunk_start, u64 physical, u64 devid,
6013 u64 **logical, int *naddrs, int *stripe_len)
6015 struct extent_map *em;
6016 struct map_lookup *map;
6024 em = get_chunk_map(fs_info, chunk_start, 1);
6028 map = em->map_lookup;
6030 rmap_len = map->stripe_len;
6032 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6033 length = div_u64(length, map->num_stripes / map->sub_stripes);
6034 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6035 length = div_u64(length, map->num_stripes);
6036 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6037 length = div_u64(length, nr_data_stripes(map));
6038 rmap_len = map->stripe_len * nr_data_stripes(map);
6041 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6042 BUG_ON(!buf); /* -ENOMEM */
6044 for (i = 0; i < map->num_stripes; i++) {
6045 if (devid && map->stripes[i].dev->devid != devid)
6047 if (map->stripes[i].physical > physical ||
6048 map->stripes[i].physical + length <= physical)
6051 stripe_nr = physical - map->stripes[i].physical;
6052 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6054 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6055 stripe_nr = stripe_nr * map->num_stripes + i;
6056 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6057 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6058 stripe_nr = stripe_nr * map->num_stripes + i;
6059 } /* else if RAID[56], multiply by nr_data_stripes().
6060 * Alternatively, just use rmap_len below instead of
6061 * map->stripe_len */
6063 bytenr = chunk_start + stripe_nr * rmap_len;
6064 WARN_ON(nr >= map->num_stripes);
6065 for (j = 0; j < nr; j++) {
6066 if (buf[j] == bytenr)
6070 WARN_ON(nr >= map->num_stripes);
6077 *stripe_len = rmap_len;
6079 free_extent_map(em);
6083 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6085 bio->bi_private = bbio->private;
6086 bio->bi_end_io = bbio->end_io;
6089 btrfs_put_bbio(bbio);
6092 static void btrfs_end_bio(struct bio *bio)
6094 struct btrfs_bio *bbio = bio->bi_private;
6095 int is_orig_bio = 0;
6097 if (bio->bi_status) {
6098 atomic_inc(&bbio->error);
6099 if (bio->bi_status == BLK_STS_IOERR ||
6100 bio->bi_status == BLK_STS_TARGET) {
6101 unsigned int stripe_index =
6102 btrfs_io_bio(bio)->stripe_index;
6103 struct btrfs_device *dev;
6105 BUG_ON(stripe_index >= bbio->num_stripes);
6106 dev = bbio->stripes[stripe_index].dev;
6108 if (bio_op(bio) == REQ_OP_WRITE)
6109 btrfs_dev_stat_inc_and_print(dev,
6110 BTRFS_DEV_STAT_WRITE_ERRS);
6112 btrfs_dev_stat_inc_and_print(dev,
6113 BTRFS_DEV_STAT_READ_ERRS);
6114 if (bio->bi_opf & REQ_PREFLUSH)
6115 btrfs_dev_stat_inc_and_print(dev,
6116 BTRFS_DEV_STAT_FLUSH_ERRS);
6121 if (bio == bbio->orig_bio)
6124 btrfs_bio_counter_dec(bbio->fs_info);
6126 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6129 bio = bbio->orig_bio;
6132 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6133 /* only send an error to the higher layers if it is
6134 * beyond the tolerance of the btrfs bio
6136 if (atomic_read(&bbio->error) > bbio->max_errors) {
6137 bio->bi_status = BLK_STS_IOERR;
6140 * this bio is actually up to date, we didn't
6141 * go over the max number of errors
6143 bio->bi_status = BLK_STS_OK;
6146 btrfs_end_bbio(bbio, bio);
6147 } else if (!is_orig_bio) {
6153 * see run_scheduled_bios for a description of why bios are collected for
6156 * This will add one bio to the pending list for a device and make sure
6157 * the work struct is scheduled.
6159 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6162 struct btrfs_fs_info *fs_info = device->fs_info;
6163 int should_queue = 1;
6164 struct btrfs_pending_bios *pending_bios;
6166 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6172 /* don't bother with additional async steps for reads, right now */
6173 if (bio_op(bio) == REQ_OP_READ) {
6174 btrfsic_submit_bio(bio);
6178 WARN_ON(bio->bi_next);
6179 bio->bi_next = NULL;
6181 spin_lock(&device->io_lock);
6182 if (op_is_sync(bio->bi_opf))
6183 pending_bios = &device->pending_sync_bios;
6185 pending_bios = &device->pending_bios;
6187 if (pending_bios->tail)
6188 pending_bios->tail->bi_next = bio;
6190 pending_bios->tail = bio;
6191 if (!pending_bios->head)
6192 pending_bios->head = bio;
6193 if (device->running_pending)
6196 spin_unlock(&device->io_lock);
6199 btrfs_queue_work(fs_info->submit_workers, &device->work);
6202 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6203 u64 physical, int dev_nr, int async)
6205 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6206 struct btrfs_fs_info *fs_info = bbio->fs_info;
6208 bio->bi_private = bbio;
6209 btrfs_io_bio(bio)->stripe_index = dev_nr;
6210 bio->bi_end_io = btrfs_end_bio;
6211 bio->bi_iter.bi_sector = physical >> 9;
6214 struct rcu_string *name;
6217 name = rcu_dereference(dev->name);
6218 btrfs_debug(fs_info,
6219 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6220 bio_op(bio), bio->bi_opf,
6221 (u64)bio->bi_iter.bi_sector,
6222 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6223 bio->bi_iter.bi_size);
6227 bio_set_dev(bio, dev->bdev);
6229 btrfs_bio_counter_inc_noblocked(fs_info);
6232 btrfs_schedule_bio(dev, bio);
6234 btrfsic_submit_bio(bio);
6237 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6239 atomic_inc(&bbio->error);
6240 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6241 /* Should be the original bio. */
6242 WARN_ON(bio != bbio->orig_bio);
6244 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6245 bio->bi_iter.bi_sector = logical >> 9;
6246 if (atomic_read(&bbio->error) > bbio->max_errors)
6247 bio->bi_status = BLK_STS_IOERR;
6249 bio->bi_status = BLK_STS_OK;
6250 btrfs_end_bbio(bbio, bio);
6254 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6255 int mirror_num, int async_submit)
6257 struct btrfs_device *dev;
6258 struct bio *first_bio = bio;
6259 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6265 struct btrfs_bio *bbio = NULL;
6267 length = bio->bi_iter.bi_size;
6268 map_length = length;
6270 btrfs_bio_counter_inc_blocked(fs_info);
6271 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6272 &map_length, &bbio, mirror_num, 1);
6274 btrfs_bio_counter_dec(fs_info);
6275 return errno_to_blk_status(ret);
6278 total_devs = bbio->num_stripes;
6279 bbio->orig_bio = first_bio;
6280 bbio->private = first_bio->bi_private;
6281 bbio->end_io = first_bio->bi_end_io;
6282 bbio->fs_info = fs_info;
6283 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6285 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6286 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6287 /* In this case, map_length has been set to the length of
6288 a single stripe; not the whole write */
6289 if (bio_op(bio) == REQ_OP_WRITE) {
6290 ret = raid56_parity_write(fs_info, bio, bbio,
6293 ret = raid56_parity_recover(fs_info, bio, bbio,
6294 map_length, mirror_num, 1);
6297 btrfs_bio_counter_dec(fs_info);
6298 return errno_to_blk_status(ret);
6301 if (map_length < length) {
6303 "mapping failed logical %llu bio len %llu len %llu",
6304 logical, length, map_length);
6308 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6309 dev = bbio->stripes[dev_nr].dev;
6310 if (!dev || !dev->bdev ||
6311 (bio_op(first_bio) == REQ_OP_WRITE &&
6312 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6313 bbio_error(bbio, first_bio, logical);
6317 if (dev_nr < total_devs - 1)
6318 bio = btrfs_bio_clone(first_bio);
6322 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6323 dev_nr, async_submit);
6325 btrfs_bio_counter_dec(fs_info);
6329 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6332 struct btrfs_device *device;
6333 struct btrfs_fs_devices *cur_devices;
6335 cur_devices = fs_info->fs_devices;
6336 while (cur_devices) {
6338 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6339 device = find_device(cur_devices, devid, uuid);
6343 cur_devices = cur_devices->seed;
6348 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6349 u64 devid, u8 *dev_uuid)
6351 struct btrfs_device *device;
6353 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6357 list_add(&device->dev_list, &fs_devices->devices);
6358 device->fs_devices = fs_devices;
6359 fs_devices->num_devices++;
6361 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6362 fs_devices->missing_devices++;
6368 * btrfs_alloc_device - allocate struct btrfs_device
6369 * @fs_info: used only for generating a new devid, can be NULL if
6370 * devid is provided (i.e. @devid != NULL).
6371 * @devid: a pointer to devid for this device. If NULL a new devid
6373 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6376 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6377 * on error. Returned struct is not linked onto any lists and must be
6378 * destroyed with free_device.
6380 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6384 struct btrfs_device *dev;
6387 if (WARN_ON(!devid && !fs_info))
6388 return ERR_PTR(-EINVAL);
6390 dev = __alloc_device();
6399 ret = find_next_devid(fs_info, &tmp);
6402 return ERR_PTR(ret);
6408 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6410 generate_random_uuid(dev->uuid);
6412 btrfs_init_work(&dev->work, btrfs_submit_helper,
6413 pending_bios_fn, NULL, NULL);
6418 /* Return -EIO if any error, otherwise return 0. */
6419 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6420 struct extent_buffer *leaf,
6421 struct btrfs_chunk *chunk, u64 logical)
6429 length = btrfs_chunk_length(leaf, chunk);
6430 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6431 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6432 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6433 type = btrfs_chunk_type(leaf, chunk);
6436 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6440 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6441 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6444 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6445 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6446 btrfs_chunk_sector_size(leaf, chunk));
6449 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6450 btrfs_err(fs_info, "invalid chunk length %llu", length);
6453 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6454 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6458 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6460 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6461 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6462 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6463 btrfs_chunk_type(leaf, chunk));
6466 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6467 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6468 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6469 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6470 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6471 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6472 num_stripes != 1)) {
6474 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6475 num_stripes, sub_stripes,
6476 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6483 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6484 u64 devid, u8 *uuid, bool error)
6487 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6490 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6494 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6495 struct extent_buffer *leaf,
6496 struct btrfs_chunk *chunk)
6498 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6499 struct map_lookup *map;
6500 struct extent_map *em;
6504 u8 uuid[BTRFS_UUID_SIZE];
6509 logical = key->offset;
6510 length = btrfs_chunk_length(leaf, chunk);
6511 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6513 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6517 read_lock(&map_tree->map_tree.lock);
6518 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6519 read_unlock(&map_tree->map_tree.lock);
6521 /* already mapped? */
6522 if (em && em->start <= logical && em->start + em->len > logical) {
6523 free_extent_map(em);
6526 free_extent_map(em);
6529 em = alloc_extent_map();
6532 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6534 free_extent_map(em);
6538 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6539 em->map_lookup = map;
6540 em->start = logical;
6543 em->block_start = 0;
6544 em->block_len = em->len;
6546 map->num_stripes = num_stripes;
6547 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6548 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6549 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6550 map->type = btrfs_chunk_type(leaf, chunk);
6551 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6552 for (i = 0; i < num_stripes; i++) {
6553 map->stripes[i].physical =
6554 btrfs_stripe_offset_nr(leaf, chunk, i);
6555 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6556 read_extent_buffer(leaf, uuid, (unsigned long)
6557 btrfs_stripe_dev_uuid_nr(chunk, i),
6559 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6561 if (!map->stripes[i].dev &&
6562 !btrfs_test_opt(fs_info, DEGRADED)) {
6563 free_extent_map(em);
6564 btrfs_report_missing_device(fs_info, devid, uuid, true);
6567 if (!map->stripes[i].dev) {
6568 map->stripes[i].dev =
6569 add_missing_dev(fs_info->fs_devices, devid,
6571 if (IS_ERR(map->stripes[i].dev)) {
6572 free_extent_map(em);
6574 "failed to init missing dev %llu: %ld",
6575 devid, PTR_ERR(map->stripes[i].dev));
6576 return PTR_ERR(map->stripes[i].dev);
6578 btrfs_report_missing_device(fs_info, devid, uuid, false);
6580 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6581 &(map->stripes[i].dev->dev_state));
6585 write_lock(&map_tree->map_tree.lock);
6586 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6587 write_unlock(&map_tree->map_tree.lock);
6588 BUG_ON(ret); /* Tree corruption */
6589 free_extent_map(em);
6594 static void fill_device_from_item(struct extent_buffer *leaf,
6595 struct btrfs_dev_item *dev_item,
6596 struct btrfs_device *device)
6600 device->devid = btrfs_device_id(leaf, dev_item);
6601 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6602 device->total_bytes = device->disk_total_bytes;
6603 device->commit_total_bytes = device->disk_total_bytes;
6604 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6605 device->commit_bytes_used = device->bytes_used;
6606 device->type = btrfs_device_type(leaf, dev_item);
6607 device->io_align = btrfs_device_io_align(leaf, dev_item);
6608 device->io_width = btrfs_device_io_width(leaf, dev_item);
6609 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6610 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6611 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6613 ptr = btrfs_device_uuid(dev_item);
6614 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6617 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6620 struct btrfs_fs_devices *fs_devices;
6623 BUG_ON(!mutex_is_locked(&uuid_mutex));
6626 fs_devices = fs_info->fs_devices->seed;
6627 while (fs_devices) {
6628 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6631 fs_devices = fs_devices->seed;
6634 fs_devices = find_fsid(fsid);
6636 if (!btrfs_test_opt(fs_info, DEGRADED))
6637 return ERR_PTR(-ENOENT);
6639 fs_devices = alloc_fs_devices(fsid);
6640 if (IS_ERR(fs_devices))
6643 fs_devices->seeding = 1;
6644 fs_devices->opened = 1;
6648 fs_devices = clone_fs_devices(fs_devices);
6649 if (IS_ERR(fs_devices))
6652 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6653 fs_info->bdev_holder);
6655 free_fs_devices(fs_devices);
6656 fs_devices = ERR_PTR(ret);
6660 if (!fs_devices->seeding) {
6661 __btrfs_close_devices(fs_devices);
6662 free_fs_devices(fs_devices);
6663 fs_devices = ERR_PTR(-EINVAL);
6667 fs_devices->seed = fs_info->fs_devices->seed;
6668 fs_info->fs_devices->seed = fs_devices;
6673 static int read_one_dev(struct btrfs_fs_info *fs_info,
6674 struct extent_buffer *leaf,
6675 struct btrfs_dev_item *dev_item)
6677 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6678 struct btrfs_device *device;
6681 u8 fs_uuid[BTRFS_FSID_SIZE];
6682 u8 dev_uuid[BTRFS_UUID_SIZE];
6684 devid = btrfs_device_id(leaf, dev_item);
6685 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6687 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6690 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6691 fs_devices = open_seed_devices(fs_info, fs_uuid);
6692 if (IS_ERR(fs_devices))
6693 return PTR_ERR(fs_devices);
6696 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6698 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6699 btrfs_report_missing_device(fs_info, devid,
6704 device = add_missing_dev(fs_devices, devid, dev_uuid);
6705 if (IS_ERR(device)) {
6707 "failed to add missing dev %llu: %ld",
6708 devid, PTR_ERR(device));
6709 return PTR_ERR(device);
6711 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6713 if (!device->bdev) {
6714 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6715 btrfs_report_missing_device(fs_info,
6716 devid, dev_uuid, true);
6719 btrfs_report_missing_device(fs_info, devid,
6723 if (!device->bdev &&
6724 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6726 * this happens when a device that was properly setup
6727 * in the device info lists suddenly goes bad.
6728 * device->bdev is NULL, and so we have to set
6729 * device->missing to one here
6731 device->fs_devices->missing_devices++;
6732 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6735 /* Move the device to its own fs_devices */
6736 if (device->fs_devices != fs_devices) {
6737 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6738 &device->dev_state));
6740 list_move(&device->dev_list, &fs_devices->devices);
6741 device->fs_devices->num_devices--;
6742 fs_devices->num_devices++;
6744 device->fs_devices->missing_devices--;
6745 fs_devices->missing_devices++;
6747 device->fs_devices = fs_devices;
6751 if (device->fs_devices != fs_info->fs_devices) {
6752 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6753 if (device->generation !=
6754 btrfs_device_generation(leaf, dev_item))
6758 fill_device_from_item(leaf, dev_item, device);
6759 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6760 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6761 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6762 device->fs_devices->total_rw_bytes += device->total_bytes;
6763 atomic64_add(device->total_bytes - device->bytes_used,
6764 &fs_info->free_chunk_space);
6770 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6772 struct btrfs_root *root = fs_info->tree_root;
6773 struct btrfs_super_block *super_copy = fs_info->super_copy;
6774 struct extent_buffer *sb;
6775 struct btrfs_disk_key *disk_key;
6776 struct btrfs_chunk *chunk;
6778 unsigned long sb_array_offset;
6785 struct btrfs_key key;
6787 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6789 * This will create extent buffer of nodesize, superblock size is
6790 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6791 * overallocate but we can keep it as-is, only the first page is used.
6793 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6796 set_extent_buffer_uptodate(sb);
6797 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6799 * The sb extent buffer is artificial and just used to read the system array.
6800 * set_extent_buffer_uptodate() call does not properly mark all it's
6801 * pages up-to-date when the page is larger: extent does not cover the
6802 * whole page and consequently check_page_uptodate does not find all
6803 * the page's extents up-to-date (the hole beyond sb),
6804 * write_extent_buffer then triggers a WARN_ON.
6806 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6807 * but sb spans only this function. Add an explicit SetPageUptodate call
6808 * to silence the warning eg. on PowerPC 64.
6810 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6811 SetPageUptodate(sb->pages[0]);
6813 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6814 array_size = btrfs_super_sys_array_size(super_copy);
6816 array_ptr = super_copy->sys_chunk_array;
6817 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6820 while (cur_offset < array_size) {
6821 disk_key = (struct btrfs_disk_key *)array_ptr;
6822 len = sizeof(*disk_key);
6823 if (cur_offset + len > array_size)
6824 goto out_short_read;
6826 btrfs_disk_key_to_cpu(&key, disk_key);
6829 sb_array_offset += len;
6832 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6833 chunk = (struct btrfs_chunk *)sb_array_offset;
6835 * At least one btrfs_chunk with one stripe must be
6836 * present, exact stripe count check comes afterwards
6838 len = btrfs_chunk_item_size(1);
6839 if (cur_offset + len > array_size)
6840 goto out_short_read;
6842 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6845 "invalid number of stripes %u in sys_array at offset %u",
6846 num_stripes, cur_offset);
6851 type = btrfs_chunk_type(sb, chunk);
6852 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6854 "invalid chunk type %llu in sys_array at offset %u",
6860 len = btrfs_chunk_item_size(num_stripes);
6861 if (cur_offset + len > array_size)
6862 goto out_short_read;
6864 ret = read_one_chunk(fs_info, &key, sb, chunk);
6869 "unexpected item type %u in sys_array at offset %u",
6870 (u32)key.type, cur_offset);
6875 sb_array_offset += len;
6878 clear_extent_buffer_uptodate(sb);
6879 free_extent_buffer_stale(sb);
6883 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6885 clear_extent_buffer_uptodate(sb);
6886 free_extent_buffer_stale(sb);
6891 * Check if all chunks in the fs are OK for read-write degraded mount
6893 * If the @failing_dev is specified, it's accounted as missing.
6895 * Return true if all chunks meet the minimal RW mount requirements.
6896 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6898 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6899 struct btrfs_device *failing_dev)
6901 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6902 struct extent_map *em;
6906 read_lock(&map_tree->map_tree.lock);
6907 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6908 read_unlock(&map_tree->map_tree.lock);
6909 /* No chunk at all? Return false anyway */
6915 struct map_lookup *map;
6920 map = em->map_lookup;
6922 btrfs_get_num_tolerated_disk_barrier_failures(
6924 for (i = 0; i < map->num_stripes; i++) {
6925 struct btrfs_device *dev = map->stripes[i].dev;
6927 if (!dev || !dev->bdev ||
6928 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6929 dev->last_flush_error)
6931 else if (failing_dev && failing_dev == dev)
6934 if (missing > max_tolerated) {
6937 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6938 em->start, missing, max_tolerated);
6939 free_extent_map(em);
6943 next_start = extent_map_end(em);
6944 free_extent_map(em);
6946 read_lock(&map_tree->map_tree.lock);
6947 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6948 (u64)(-1) - next_start);
6949 read_unlock(&map_tree->map_tree.lock);
6955 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6957 struct btrfs_root *root = fs_info->chunk_root;
6958 struct btrfs_path *path;
6959 struct extent_buffer *leaf;
6960 struct btrfs_key key;
6961 struct btrfs_key found_key;
6966 path = btrfs_alloc_path();
6970 mutex_lock(&uuid_mutex);
6971 mutex_lock(&fs_info->chunk_mutex);
6974 * Read all device items, and then all the chunk items. All
6975 * device items are found before any chunk item (their object id
6976 * is smaller than the lowest possible object id for a chunk
6977 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6979 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6982 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6986 leaf = path->nodes[0];
6987 slot = path->slots[0];
6988 if (slot >= btrfs_header_nritems(leaf)) {
6989 ret = btrfs_next_leaf(root, path);
6996 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6997 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6998 struct btrfs_dev_item *dev_item;
6999 dev_item = btrfs_item_ptr(leaf, slot,
7000 struct btrfs_dev_item);
7001 ret = read_one_dev(fs_info, leaf, dev_item);
7005 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7006 struct btrfs_chunk *chunk;
7007 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7008 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7016 * After loading chunk tree, we've got all device information,
7017 * do another round of validation checks.
7019 if (total_dev != fs_info->fs_devices->total_devices) {
7021 "super_num_devices %llu mismatch with num_devices %llu found here",
7022 btrfs_super_num_devices(fs_info->super_copy),
7027 if (btrfs_super_total_bytes(fs_info->super_copy) <
7028 fs_info->fs_devices->total_rw_bytes) {
7030 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7031 btrfs_super_total_bytes(fs_info->super_copy),
7032 fs_info->fs_devices->total_rw_bytes);
7038 mutex_unlock(&fs_info->chunk_mutex);
7039 mutex_unlock(&uuid_mutex);
7041 btrfs_free_path(path);
7045 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7047 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7048 struct btrfs_device *device;
7050 while (fs_devices) {
7051 mutex_lock(&fs_devices->device_list_mutex);
7052 list_for_each_entry(device, &fs_devices->devices, dev_list)
7053 device->fs_info = fs_info;
7054 mutex_unlock(&fs_devices->device_list_mutex);
7056 fs_devices = fs_devices->seed;
7060 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7064 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7065 btrfs_dev_stat_reset(dev, i);
7068 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7070 struct btrfs_key key;
7071 struct btrfs_key found_key;
7072 struct btrfs_root *dev_root = fs_info->dev_root;
7073 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7074 struct extent_buffer *eb;
7077 struct btrfs_device *device;
7078 struct btrfs_path *path = NULL;
7081 path = btrfs_alloc_path();
7087 mutex_lock(&fs_devices->device_list_mutex);
7088 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7090 struct btrfs_dev_stats_item *ptr;
7092 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7093 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7094 key.offset = device->devid;
7095 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7097 __btrfs_reset_dev_stats(device);
7098 device->dev_stats_valid = 1;
7099 btrfs_release_path(path);
7102 slot = path->slots[0];
7103 eb = path->nodes[0];
7104 btrfs_item_key_to_cpu(eb, &found_key, slot);
7105 item_size = btrfs_item_size_nr(eb, slot);
7107 ptr = btrfs_item_ptr(eb, slot,
7108 struct btrfs_dev_stats_item);
7110 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7111 if (item_size >= (1 + i) * sizeof(__le64))
7112 btrfs_dev_stat_set(device, i,
7113 btrfs_dev_stats_value(eb, ptr, i));
7115 btrfs_dev_stat_reset(device, i);
7118 device->dev_stats_valid = 1;
7119 btrfs_dev_stat_print_on_load(device);
7120 btrfs_release_path(path);
7122 mutex_unlock(&fs_devices->device_list_mutex);
7125 btrfs_free_path(path);
7126 return ret < 0 ? ret : 0;
7129 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7130 struct btrfs_fs_info *fs_info,
7131 struct btrfs_device *device)
7133 struct btrfs_root *dev_root = fs_info->dev_root;
7134 struct btrfs_path *path;
7135 struct btrfs_key key;
7136 struct extent_buffer *eb;
7137 struct btrfs_dev_stats_item *ptr;
7141 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7142 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7143 key.offset = device->devid;
7145 path = btrfs_alloc_path();
7148 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7150 btrfs_warn_in_rcu(fs_info,
7151 "error %d while searching for dev_stats item for device %s",
7152 ret, rcu_str_deref(device->name));
7157 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7158 /* need to delete old one and insert a new one */
7159 ret = btrfs_del_item(trans, dev_root, path);
7161 btrfs_warn_in_rcu(fs_info,
7162 "delete too small dev_stats item for device %s failed %d",
7163 rcu_str_deref(device->name), ret);
7170 /* need to insert a new item */
7171 btrfs_release_path(path);
7172 ret = btrfs_insert_empty_item(trans, dev_root, path,
7173 &key, sizeof(*ptr));
7175 btrfs_warn_in_rcu(fs_info,
7176 "insert dev_stats item for device %s failed %d",
7177 rcu_str_deref(device->name), ret);
7182 eb = path->nodes[0];
7183 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7184 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7185 btrfs_set_dev_stats_value(eb, ptr, i,
7186 btrfs_dev_stat_read(device, i));
7187 btrfs_mark_buffer_dirty(eb);
7190 btrfs_free_path(path);
7195 * called from commit_transaction. Writes all changed device stats to disk.
7197 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7198 struct btrfs_fs_info *fs_info)
7200 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7201 struct btrfs_device *device;
7205 mutex_lock(&fs_devices->device_list_mutex);
7206 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7207 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7208 if (!device->dev_stats_valid || stats_cnt == 0)
7213 * There is a LOAD-LOAD control dependency between the value of
7214 * dev_stats_ccnt and updating the on-disk values which requires
7215 * reading the in-memory counters. Such control dependencies
7216 * require explicit read memory barriers.
7218 * This memory barriers pairs with smp_mb__before_atomic in
7219 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7220 * barrier implied by atomic_xchg in
7221 * btrfs_dev_stats_read_and_reset
7225 ret = update_dev_stat_item(trans, fs_info, device);
7227 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7229 mutex_unlock(&fs_devices->device_list_mutex);
7234 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7236 btrfs_dev_stat_inc(dev, index);
7237 btrfs_dev_stat_print_on_error(dev);
7240 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7242 if (!dev->dev_stats_valid)
7244 btrfs_err_rl_in_rcu(dev->fs_info,
7245 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7246 rcu_str_deref(dev->name),
7247 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7248 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7249 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7250 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7251 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7254 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7258 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7259 if (btrfs_dev_stat_read(dev, i) != 0)
7261 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7262 return; /* all values == 0, suppress message */
7264 btrfs_info_in_rcu(dev->fs_info,
7265 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7266 rcu_str_deref(dev->name),
7267 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7268 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7269 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7270 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7271 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7274 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7275 struct btrfs_ioctl_get_dev_stats *stats)
7277 struct btrfs_device *dev;
7278 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7281 mutex_lock(&fs_devices->device_list_mutex);
7282 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7283 mutex_unlock(&fs_devices->device_list_mutex);
7286 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7288 } else if (!dev->dev_stats_valid) {
7289 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7291 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7292 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7293 if (stats->nr_items > i)
7295 btrfs_dev_stat_read_and_reset(dev, i);
7297 btrfs_dev_stat_reset(dev, i);
7300 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7301 if (stats->nr_items > i)
7302 stats->values[i] = btrfs_dev_stat_read(dev, i);
7304 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7305 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7309 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7311 struct buffer_head *bh;
7312 struct btrfs_super_block *disk_super;
7318 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7321 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7324 disk_super = (struct btrfs_super_block *)bh->b_data;
7326 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7327 set_buffer_dirty(bh);
7328 sync_dirty_buffer(bh);
7332 /* Notify udev that device has changed */
7333 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7335 /* Update ctime/mtime for device path for libblkid */
7336 update_dev_time(device_path);
7340 * Update the size of all devices, which is used for writing out the
7343 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7345 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7346 struct btrfs_device *curr, *next;
7348 if (list_empty(&fs_devices->resized_devices))
7351 mutex_lock(&fs_devices->device_list_mutex);
7352 mutex_lock(&fs_info->chunk_mutex);
7353 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7355 list_del_init(&curr->resized_list);
7356 curr->commit_total_bytes = curr->disk_total_bytes;
7358 mutex_unlock(&fs_info->chunk_mutex);
7359 mutex_unlock(&fs_devices->device_list_mutex);
7362 /* Must be invoked during the transaction commit */
7363 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7365 struct btrfs_fs_info *fs_info = trans->fs_info;
7366 struct extent_map *em;
7367 struct map_lookup *map;
7368 struct btrfs_device *dev;
7371 if (list_empty(&trans->pending_chunks))
7374 /* In order to kick the device replace finish process */
7375 mutex_lock(&fs_info->chunk_mutex);
7376 list_for_each_entry(em, &trans->pending_chunks, list) {
7377 map = em->map_lookup;
7379 for (i = 0; i < map->num_stripes; i++) {
7380 dev = map->stripes[i].dev;
7381 dev->commit_bytes_used = dev->bytes_used;
7384 mutex_unlock(&fs_info->chunk_mutex);
7387 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7389 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7390 while (fs_devices) {
7391 fs_devices->fs_info = fs_info;
7392 fs_devices = fs_devices->seed;
7396 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7398 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7399 while (fs_devices) {
7400 fs_devices->fs_info = NULL;
7401 fs_devices = fs_devices->seed;