2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <linux/list_sort.h>
31 #include <asm/div64.h>
33 #include "extent_map.h"
35 #include "transaction.h"
36 #include "print-tree.h"
39 #include "async-thread.h"
40 #include "check-integrity.h"
41 #include "rcu-string.h"
43 #include "dev-replace.h"
46 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
47 [BTRFS_RAID_RAID10] = {
50 .devs_max = 0, /* 0 == as many as possible */
52 .tolerated_failures = 1,
56 [BTRFS_RAID_RAID1] = {
61 .tolerated_failures = 1,
70 .tolerated_failures = 0,
74 [BTRFS_RAID_RAID0] = {
79 .tolerated_failures = 0,
83 [BTRFS_RAID_SINGLE] = {
88 .tolerated_failures = 0,
92 [BTRFS_RAID_RAID5] = {
97 .tolerated_failures = 1,
101 [BTRFS_RAID_RAID6] = {
106 .tolerated_failures = 2,
112 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
113 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
114 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
115 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
116 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
117 [BTRFS_RAID_SINGLE] = 0,
118 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
119 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
123 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
124 * condition is not met. Zero means there's no corresponding
125 * BTRFS_ERROR_DEV_*_NOT_MET value.
127 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
128 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
129 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
130 [BTRFS_RAID_DUP] = 0,
131 [BTRFS_RAID_RAID0] = 0,
132 [BTRFS_RAID_SINGLE] = 0,
133 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
134 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
137 static int init_first_rw_device(struct btrfs_trans_handle *trans,
138 struct btrfs_fs_info *fs_info);
139 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
144 enum btrfs_map_op op,
145 u64 logical, u64 *length,
146 struct btrfs_bio **bbio_ret,
147 int mirror_num, int need_raid_map);
153 * There are several mutexes that protect manipulation of devices and low-level
154 * structures like chunks but not block groups, extents or files
156 * uuid_mutex (global lock)
157 * ------------------------
158 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
159 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
160 * device) or requested by the device= mount option
162 * the mutex can be very coarse and can cover long-running operations
164 * protects: updates to fs_devices counters like missing devices, rw devices,
165 * seeding, structure cloning, openning/closing devices at mount/umount time
167 * global::fs_devs - add, remove, updates to the global list
169 * does not protect: manipulation of the fs_devices::devices list!
171 * btrfs_device::name - renames (write side), read is RCU
173 * fs_devices::device_list_mutex (per-fs, with RCU)
174 * ------------------------------------------------
175 * protects updates to fs_devices::devices, ie. adding and deleting
177 * simple list traversal with read-only actions can be done with RCU protection
179 * may be used to exclude some operations from running concurrently without any
180 * modifications to the list (see write_all_supers)
184 * coarse lock owned by a mounted filesystem; used to exclude some operations
185 * that cannot run in parallel and affect the higher-level properties of the
186 * filesystem like: device add/deleting/resize/replace, or balance
190 * protects balance structures (status, state) and context accessed from
191 * several places (internally, ioctl)
195 * protects chunks, adding or removing during allocation, trim or when a new
196 * device is added/removed
200 * a big lock that is held by the cleaner thread and prevents running subvolume
201 * cleaning together with relocation or delayed iputs
214 DEFINE_MUTEX(uuid_mutex);
215 static LIST_HEAD(fs_uuids);
216 struct list_head *btrfs_get_fs_uuids(void)
222 * alloc_fs_devices - allocate struct btrfs_fs_devices
223 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
225 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
226 * The returned struct is not linked onto any lists and can be destroyed with
227 * kfree() right away.
229 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
231 struct btrfs_fs_devices *fs_devs;
233 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
235 return ERR_PTR(-ENOMEM);
237 mutex_init(&fs_devs->device_list_mutex);
239 INIT_LIST_HEAD(&fs_devs->devices);
240 INIT_LIST_HEAD(&fs_devs->resized_devices);
241 INIT_LIST_HEAD(&fs_devs->alloc_list);
242 INIT_LIST_HEAD(&fs_devs->list);
244 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
249 static void free_device(struct btrfs_device *device)
251 rcu_string_free(device->name);
252 bio_put(device->flush_bio);
256 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
258 struct btrfs_device *device;
259 WARN_ON(fs_devices->opened);
260 while (!list_empty(&fs_devices->devices)) {
261 device = list_entry(fs_devices->devices.next,
262 struct btrfs_device, dev_list);
263 list_del(&device->dev_list);
269 static void btrfs_kobject_uevent(struct block_device *bdev,
270 enum kobject_action action)
274 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
276 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
278 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
279 &disk_to_dev(bdev->bd_disk)->kobj);
282 void __exit btrfs_cleanup_fs_uuids(void)
284 struct btrfs_fs_devices *fs_devices;
286 while (!list_empty(&fs_uuids)) {
287 fs_devices = list_entry(fs_uuids.next,
288 struct btrfs_fs_devices, list);
289 list_del(&fs_devices->list);
290 free_fs_devices(fs_devices);
295 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
296 * Returned struct is not linked onto any lists and must be destroyed using
299 static struct btrfs_device *__alloc_device(void)
301 struct btrfs_device *dev;
303 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
305 return ERR_PTR(-ENOMEM);
308 * Preallocate a bio that's always going to be used for flushing device
309 * barriers and matches the device lifespan
311 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
312 if (!dev->flush_bio) {
314 return ERR_PTR(-ENOMEM);
317 INIT_LIST_HEAD(&dev->dev_list);
318 INIT_LIST_HEAD(&dev->dev_alloc_list);
319 INIT_LIST_HEAD(&dev->resized_list);
321 spin_lock_init(&dev->io_lock);
323 atomic_set(&dev->reada_in_flight, 0);
324 atomic_set(&dev->dev_stats_ccnt, 0);
325 btrfs_device_data_ordered_init(dev);
326 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
327 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
333 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
336 * If devid and uuid are both specified, the match must be exact, otherwise
337 * only devid is used.
339 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
340 u64 devid, const u8 *uuid)
342 struct list_head *head = &fs_devices->devices;
343 struct btrfs_device *dev;
345 list_for_each_entry(dev, head, dev_list) {
346 if (dev->devid == devid &&
347 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
354 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
356 struct btrfs_fs_devices *fs_devices;
358 list_for_each_entry(fs_devices, &fs_uuids, list) {
359 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
366 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
367 int flush, struct block_device **bdev,
368 struct buffer_head **bh)
372 *bdev = blkdev_get_by_path(device_path, flags, holder);
375 ret = PTR_ERR(*bdev);
380 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
381 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
383 blkdev_put(*bdev, flags);
386 invalidate_bdev(*bdev);
387 *bh = btrfs_read_dev_super(*bdev);
390 blkdev_put(*bdev, flags);
402 static void requeue_list(struct btrfs_pending_bios *pending_bios,
403 struct bio *head, struct bio *tail)
406 struct bio *old_head;
408 old_head = pending_bios->head;
409 pending_bios->head = head;
410 if (pending_bios->tail)
411 tail->bi_next = old_head;
413 pending_bios->tail = tail;
417 * we try to collect pending bios for a device so we don't get a large
418 * number of procs sending bios down to the same device. This greatly
419 * improves the schedulers ability to collect and merge the bios.
421 * But, it also turns into a long list of bios to process and that is sure
422 * to eventually make the worker thread block. The solution here is to
423 * make some progress and then put this work struct back at the end of
424 * the list if the block device is congested. This way, multiple devices
425 * can make progress from a single worker thread.
427 static noinline void run_scheduled_bios(struct btrfs_device *device)
429 struct btrfs_fs_info *fs_info = device->fs_info;
431 struct backing_dev_info *bdi;
432 struct btrfs_pending_bios *pending_bios;
436 unsigned long num_run;
437 unsigned long batch_run = 0;
438 unsigned long last_waited = 0;
440 int sync_pending = 0;
441 struct blk_plug plug;
444 * this function runs all the bios we've collected for
445 * a particular device. We don't want to wander off to
446 * another device without first sending all of these down.
447 * So, setup a plug here and finish it off before we return
449 blk_start_plug(&plug);
451 bdi = device->bdev->bd_bdi;
454 spin_lock(&device->io_lock);
459 /* take all the bios off the list at once and process them
460 * later on (without the lock held). But, remember the
461 * tail and other pointers so the bios can be properly reinserted
462 * into the list if we hit congestion
464 if (!force_reg && device->pending_sync_bios.head) {
465 pending_bios = &device->pending_sync_bios;
468 pending_bios = &device->pending_bios;
472 pending = pending_bios->head;
473 tail = pending_bios->tail;
474 WARN_ON(pending && !tail);
477 * if pending was null this time around, no bios need processing
478 * at all and we can stop. Otherwise it'll loop back up again
479 * and do an additional check so no bios are missed.
481 * device->running_pending is used to synchronize with the
484 if (device->pending_sync_bios.head == NULL &&
485 device->pending_bios.head == NULL) {
487 device->running_pending = 0;
490 device->running_pending = 1;
493 pending_bios->head = NULL;
494 pending_bios->tail = NULL;
496 spin_unlock(&device->io_lock);
501 /* we want to work on both lists, but do more bios on the
502 * sync list than the regular list
505 pending_bios != &device->pending_sync_bios &&
506 device->pending_sync_bios.head) ||
507 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
508 device->pending_bios.head)) {
509 spin_lock(&device->io_lock);
510 requeue_list(pending_bios, pending, tail);
515 pending = pending->bi_next;
518 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
521 * if we're doing the sync list, record that our
522 * plug has some sync requests on it
524 * If we're doing the regular list and there are
525 * sync requests sitting around, unplug before
528 if (pending_bios == &device->pending_sync_bios) {
530 } else if (sync_pending) {
531 blk_finish_plug(&plug);
532 blk_start_plug(&plug);
536 btrfsic_submit_bio(cur);
543 * we made progress, there is more work to do and the bdi
544 * is now congested. Back off and let other work structs
547 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
548 fs_info->fs_devices->open_devices > 1) {
549 struct io_context *ioc;
551 ioc = current->io_context;
554 * the main goal here is that we don't want to
555 * block if we're going to be able to submit
556 * more requests without blocking.
558 * This code does two great things, it pokes into
559 * the elevator code from a filesystem _and_
560 * it makes assumptions about how batching works.
562 if (ioc && ioc->nr_batch_requests > 0 &&
563 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
565 ioc->last_waited == last_waited)) {
567 * we want to go through our batch of
568 * requests and stop. So, we copy out
569 * the ioc->last_waited time and test
570 * against it before looping
572 last_waited = ioc->last_waited;
576 spin_lock(&device->io_lock);
577 requeue_list(pending_bios, pending, tail);
578 device->running_pending = 1;
580 spin_unlock(&device->io_lock);
581 btrfs_queue_work(fs_info->submit_workers,
591 spin_lock(&device->io_lock);
592 if (device->pending_bios.head || device->pending_sync_bios.head)
594 spin_unlock(&device->io_lock);
597 blk_finish_plug(&plug);
600 static void pending_bios_fn(struct btrfs_work *work)
602 struct btrfs_device *device;
604 device = container_of(work, struct btrfs_device, work);
605 run_scheduled_bios(device);
609 * Search and remove all stale (devices which are not mounted) devices.
610 * When both inputs are NULL, it will search and release all stale devices.
611 * path: Optional. When provided will it release all unmounted devices
612 * matching this path only.
613 * skip_dev: Optional. Will skip this device when searching for the stale
616 static void btrfs_free_stale_devices(const char *path,
617 struct btrfs_device *skip_dev)
619 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
620 struct btrfs_device *dev, *tmp_dev;
622 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) {
627 list_for_each_entry_safe(dev, tmp_dev,
628 &fs_devs->devices, dev_list) {
631 if (skip_dev && skip_dev == dev)
633 if (path && !dev->name)
638 not_found = strcmp(rcu_str_deref(dev->name),
644 /* delete the stale device */
645 if (fs_devs->num_devices == 1) {
646 btrfs_sysfs_remove_fsid(fs_devs);
647 list_del(&fs_devs->list);
648 free_fs_devices(fs_devs);
651 fs_devs->num_devices--;
652 list_del(&dev->dev_list);
659 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
660 struct btrfs_device *device, fmode_t flags,
663 struct request_queue *q;
664 struct block_device *bdev;
665 struct buffer_head *bh;
666 struct btrfs_super_block *disk_super;
675 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
680 disk_super = (struct btrfs_super_block *)bh->b_data;
681 devid = btrfs_stack_device_id(&disk_super->dev_item);
682 if (devid != device->devid)
685 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
688 device->generation = btrfs_super_generation(disk_super);
690 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
691 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
692 fs_devices->seeding = 1;
694 if (bdev_read_only(bdev))
695 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
697 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
700 q = bdev_get_queue(bdev);
701 if (!blk_queue_nonrot(q))
702 fs_devices->rotating = 1;
705 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
706 device->mode = flags;
708 fs_devices->open_devices++;
709 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
710 device->devid != BTRFS_DEV_REPLACE_DEVID) {
711 fs_devices->rw_devices++;
712 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
720 blkdev_put(bdev, flags);
726 * Add new device to list of registered devices
729 * device pointer which was just added or updated when successful
730 * error pointer when failed
732 static noinline struct btrfs_device *device_list_add(const char *path,
733 struct btrfs_super_block *disk_super)
735 struct btrfs_device *device;
736 struct btrfs_fs_devices *fs_devices;
737 struct rcu_string *name;
738 u64 found_transid = btrfs_super_generation(disk_super);
739 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
741 fs_devices = find_fsid(disk_super->fsid);
743 fs_devices = alloc_fs_devices(disk_super->fsid);
744 if (IS_ERR(fs_devices))
745 return ERR_CAST(fs_devices);
747 list_add(&fs_devices->list, &fs_uuids);
751 device = find_device(fs_devices, devid,
752 disk_super->dev_item.uuid);
756 if (fs_devices->opened)
757 return ERR_PTR(-EBUSY);
759 device = btrfs_alloc_device(NULL, &devid,
760 disk_super->dev_item.uuid);
761 if (IS_ERR(device)) {
762 /* we can safely leave the fs_devices entry around */
766 name = rcu_string_strdup(path, GFP_NOFS);
769 return ERR_PTR(-ENOMEM);
771 rcu_assign_pointer(device->name, name);
773 mutex_lock(&fs_devices->device_list_mutex);
774 list_add_rcu(&device->dev_list, &fs_devices->devices);
775 fs_devices->num_devices++;
776 mutex_unlock(&fs_devices->device_list_mutex);
778 device->fs_devices = fs_devices;
779 btrfs_free_stale_devices(path, device);
781 if (disk_super->label[0])
782 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
783 disk_super->label, devid, found_transid, path);
785 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
786 disk_super->fsid, devid, found_transid, path);
788 } else if (!device->name || strcmp(device->name->str, path)) {
790 * When FS is already mounted.
791 * 1. If you are here and if the device->name is NULL that
792 * means this device was missing at time of FS mount.
793 * 2. If you are here and if the device->name is different
794 * from 'path' that means either
795 * a. The same device disappeared and reappeared with
797 * b. The missing-disk-which-was-replaced, has
800 * We must allow 1 and 2a above. But 2b would be a spurious
803 * Further in case of 1 and 2a above, the disk at 'path'
804 * would have missed some transaction when it was away and
805 * in case of 2a the stale bdev has to be updated as well.
806 * 2b must not be allowed at all time.
810 * For now, we do allow update to btrfs_fs_device through the
811 * btrfs dev scan cli after FS has been mounted. We're still
812 * tracking a problem where systems fail mount by subvolume id
813 * when we reject replacement on a mounted FS.
815 if (!fs_devices->opened && found_transid < device->generation) {
817 * That is if the FS is _not_ mounted and if you
818 * are here, that means there is more than one
819 * disk with same uuid and devid.We keep the one
820 * with larger generation number or the last-in if
821 * generation are equal.
823 return ERR_PTR(-EEXIST);
826 name = rcu_string_strdup(path, GFP_NOFS);
828 return ERR_PTR(-ENOMEM);
829 rcu_string_free(device->name);
830 rcu_assign_pointer(device->name, name);
831 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
832 fs_devices->missing_devices--;
833 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
838 * Unmount does not free the btrfs_device struct but would zero
839 * generation along with most of the other members. So just update
840 * it back. We need it to pick the disk with largest generation
843 if (!fs_devices->opened)
844 device->generation = found_transid;
846 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
851 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
853 struct btrfs_fs_devices *fs_devices;
854 struct btrfs_device *device;
855 struct btrfs_device *orig_dev;
857 fs_devices = alloc_fs_devices(orig->fsid);
858 if (IS_ERR(fs_devices))
861 mutex_lock(&orig->device_list_mutex);
862 fs_devices->total_devices = orig->total_devices;
864 /* We have held the volume lock, it is safe to get the devices. */
865 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
866 struct rcu_string *name;
868 device = btrfs_alloc_device(NULL, &orig_dev->devid,
874 * This is ok to do without rcu read locked because we hold the
875 * uuid mutex so nothing we touch in here is going to disappear.
877 if (orig_dev->name) {
878 name = rcu_string_strdup(orig_dev->name->str,
884 rcu_assign_pointer(device->name, name);
887 list_add(&device->dev_list, &fs_devices->devices);
888 device->fs_devices = fs_devices;
889 fs_devices->num_devices++;
891 mutex_unlock(&orig->device_list_mutex);
894 mutex_unlock(&orig->device_list_mutex);
895 free_fs_devices(fs_devices);
896 return ERR_PTR(-ENOMEM);
900 * After we have read the system tree and know devids belonging to
901 * this filesystem, remove the device which does not belong there.
903 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
905 struct btrfs_device *device, *next;
906 struct btrfs_device *latest_dev = NULL;
908 mutex_lock(&uuid_mutex);
910 /* This is the initialized path, it is safe to release the devices. */
911 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
912 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
913 &device->dev_state)) {
914 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
915 &device->dev_state) &&
917 device->generation > latest_dev->generation)) {
923 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
925 * In the first step, keep the device which has
926 * the correct fsid and the devid that is used
927 * for the dev_replace procedure.
928 * In the second step, the dev_replace state is
929 * read from the device tree and it is known
930 * whether the procedure is really active or
931 * not, which means whether this device is
932 * used or whether it should be removed.
934 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
935 &device->dev_state)) {
940 blkdev_put(device->bdev, device->mode);
942 fs_devices->open_devices--;
944 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
945 list_del_init(&device->dev_alloc_list);
946 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
947 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
949 fs_devices->rw_devices--;
951 list_del_init(&device->dev_list);
952 fs_devices->num_devices--;
956 if (fs_devices->seed) {
957 fs_devices = fs_devices->seed;
961 fs_devices->latest_bdev = latest_dev->bdev;
963 mutex_unlock(&uuid_mutex);
966 static void free_device_rcu(struct rcu_head *head)
968 struct btrfs_device *device;
970 device = container_of(head, struct btrfs_device, rcu);
974 static void btrfs_close_bdev(struct btrfs_device *device)
979 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
980 sync_blockdev(device->bdev);
981 invalidate_bdev(device->bdev);
984 blkdev_put(device->bdev, device->mode);
987 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
989 struct btrfs_fs_devices *fs_devices = device->fs_devices;
990 struct btrfs_device *new_device;
991 struct rcu_string *name;
994 fs_devices->open_devices--;
996 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
997 device->devid != BTRFS_DEV_REPLACE_DEVID) {
998 list_del_init(&device->dev_alloc_list);
999 fs_devices->rw_devices--;
1002 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1003 fs_devices->missing_devices--;
1005 new_device = btrfs_alloc_device(NULL, &device->devid,
1007 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1009 /* Safe because we are under uuid_mutex */
1011 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1012 BUG_ON(!name); /* -ENOMEM */
1013 rcu_assign_pointer(new_device->name, name);
1016 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1017 new_device->fs_devices = device->fs_devices;
1020 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1022 struct btrfs_device *device, *tmp;
1023 struct list_head pending_put;
1025 INIT_LIST_HEAD(&pending_put);
1027 if (--fs_devices->opened > 0)
1030 mutex_lock(&fs_devices->device_list_mutex);
1031 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1032 btrfs_prepare_close_one_device(device);
1033 list_add(&device->dev_list, &pending_put);
1035 mutex_unlock(&fs_devices->device_list_mutex);
1038 * btrfs_show_devname() is using the device_list_mutex,
1039 * sometimes call to blkdev_put() leads vfs calling
1040 * into this func. So do put outside of device_list_mutex,
1043 while (!list_empty(&pending_put)) {
1044 device = list_first_entry(&pending_put,
1045 struct btrfs_device, dev_list);
1046 list_del(&device->dev_list);
1047 btrfs_close_bdev(device);
1048 call_rcu(&device->rcu, free_device_rcu);
1051 WARN_ON(fs_devices->open_devices);
1052 WARN_ON(fs_devices->rw_devices);
1053 fs_devices->opened = 0;
1054 fs_devices->seeding = 0;
1059 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1061 struct btrfs_fs_devices *seed_devices = NULL;
1064 mutex_lock(&uuid_mutex);
1065 ret = __btrfs_close_devices(fs_devices);
1066 if (!fs_devices->opened) {
1067 seed_devices = fs_devices->seed;
1068 fs_devices->seed = NULL;
1070 mutex_unlock(&uuid_mutex);
1072 while (seed_devices) {
1073 fs_devices = seed_devices;
1074 seed_devices = fs_devices->seed;
1075 __btrfs_close_devices(fs_devices);
1076 free_fs_devices(fs_devices);
1081 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1082 fmode_t flags, void *holder)
1084 struct list_head *head = &fs_devices->devices;
1085 struct btrfs_device *device;
1086 struct btrfs_device *latest_dev = NULL;
1089 flags |= FMODE_EXCL;
1091 list_for_each_entry(device, head, dev_list) {
1092 /* Just open everything we can; ignore failures here */
1093 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1097 device->generation > latest_dev->generation)
1098 latest_dev = device;
1100 if (fs_devices->open_devices == 0) {
1104 fs_devices->opened = 1;
1105 fs_devices->latest_bdev = latest_dev->bdev;
1106 fs_devices->total_rw_bytes = 0;
1111 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1113 struct btrfs_device *dev1, *dev2;
1115 dev1 = list_entry(a, struct btrfs_device, dev_list);
1116 dev2 = list_entry(b, struct btrfs_device, dev_list);
1118 if (dev1->devid < dev2->devid)
1120 else if (dev1->devid > dev2->devid)
1125 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1126 fmode_t flags, void *holder)
1130 mutex_lock(&uuid_mutex);
1131 if (fs_devices->opened) {
1132 fs_devices->opened++;
1135 list_sort(NULL, &fs_devices->devices, devid_cmp);
1136 ret = __btrfs_open_devices(fs_devices, flags, holder);
1138 mutex_unlock(&uuid_mutex);
1142 static void btrfs_release_disk_super(struct page *page)
1148 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1150 struct btrfs_super_block **disk_super)
1155 /* make sure our super fits in the device */
1156 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1159 /* make sure our super fits in the page */
1160 if (sizeof(**disk_super) > PAGE_SIZE)
1163 /* make sure our super doesn't straddle pages on disk */
1164 index = bytenr >> PAGE_SHIFT;
1165 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1168 /* pull in the page with our super */
1169 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1172 if (IS_ERR_OR_NULL(*page))
1177 /* align our pointer to the offset of the super block */
1178 *disk_super = p + (bytenr & ~PAGE_MASK);
1180 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1181 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1182 btrfs_release_disk_super(*page);
1186 if ((*disk_super)->label[0] &&
1187 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1188 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1194 * Look for a btrfs signature on a device. This may be called out of the mount path
1195 * and we are not allowed to call set_blocksize during the scan. The superblock
1196 * is read via pagecache
1198 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1199 struct btrfs_fs_devices **fs_devices_ret)
1201 struct btrfs_super_block *disk_super;
1202 struct btrfs_device *device;
1203 struct block_device *bdev;
1209 * we would like to check all the supers, but that would make
1210 * a btrfs mount succeed after a mkfs from a different FS.
1211 * So, we need to add a special mount option to scan for
1212 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1214 bytenr = btrfs_sb_offset(0);
1215 flags |= FMODE_EXCL;
1216 mutex_lock(&uuid_mutex);
1218 bdev = blkdev_get_by_path(path, flags, holder);
1220 ret = PTR_ERR(bdev);
1224 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1226 goto error_bdev_put;
1229 device = device_list_add(path, disk_super);
1231 ret = PTR_ERR(device);
1233 *fs_devices_ret = device->fs_devices;
1235 btrfs_release_disk_super(page);
1238 blkdev_put(bdev, flags);
1240 mutex_unlock(&uuid_mutex);
1244 /* helper to account the used device space in the range */
1245 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1246 u64 end, u64 *length)
1248 struct btrfs_key key;
1249 struct btrfs_root *root = device->fs_info->dev_root;
1250 struct btrfs_dev_extent *dev_extent;
1251 struct btrfs_path *path;
1255 struct extent_buffer *l;
1259 if (start >= device->total_bytes ||
1260 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1263 path = btrfs_alloc_path();
1266 path->reada = READA_FORWARD;
1268 key.objectid = device->devid;
1270 key.type = BTRFS_DEV_EXTENT_KEY;
1272 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1276 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1283 slot = path->slots[0];
1284 if (slot >= btrfs_header_nritems(l)) {
1285 ret = btrfs_next_leaf(root, path);
1293 btrfs_item_key_to_cpu(l, &key, slot);
1295 if (key.objectid < device->devid)
1298 if (key.objectid > device->devid)
1301 if (key.type != BTRFS_DEV_EXTENT_KEY)
1304 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1305 extent_end = key.offset + btrfs_dev_extent_length(l,
1307 if (key.offset <= start && extent_end > end) {
1308 *length = end - start + 1;
1310 } else if (key.offset <= start && extent_end > start)
1311 *length += extent_end - start;
1312 else if (key.offset > start && extent_end <= end)
1313 *length += extent_end - key.offset;
1314 else if (key.offset > start && key.offset <= end) {
1315 *length += end - key.offset + 1;
1317 } else if (key.offset > end)
1325 btrfs_free_path(path);
1329 static int contains_pending_extent(struct btrfs_transaction *transaction,
1330 struct btrfs_device *device,
1331 u64 *start, u64 len)
1333 struct btrfs_fs_info *fs_info = device->fs_info;
1334 struct extent_map *em;
1335 struct list_head *search_list = &fs_info->pinned_chunks;
1337 u64 physical_start = *start;
1340 search_list = &transaction->pending_chunks;
1342 list_for_each_entry(em, search_list, list) {
1343 struct map_lookup *map;
1346 map = em->map_lookup;
1347 for (i = 0; i < map->num_stripes; i++) {
1350 if (map->stripes[i].dev != device)
1352 if (map->stripes[i].physical >= physical_start + len ||
1353 map->stripes[i].physical + em->orig_block_len <=
1357 * Make sure that while processing the pinned list we do
1358 * not override our *start with a lower value, because
1359 * we can have pinned chunks that fall within this
1360 * device hole and that have lower physical addresses
1361 * than the pending chunks we processed before. If we
1362 * do not take this special care we can end up getting
1363 * 2 pending chunks that start at the same physical
1364 * device offsets because the end offset of a pinned
1365 * chunk can be equal to the start offset of some
1368 end = map->stripes[i].physical + em->orig_block_len;
1375 if (search_list != &fs_info->pinned_chunks) {
1376 search_list = &fs_info->pinned_chunks;
1385 * find_free_dev_extent_start - find free space in the specified device
1386 * @device: the device which we search the free space in
1387 * @num_bytes: the size of the free space that we need
1388 * @search_start: the position from which to begin the search
1389 * @start: store the start of the free space.
1390 * @len: the size of the free space. that we find, or the size
1391 * of the max free space if we don't find suitable free space
1393 * this uses a pretty simple search, the expectation is that it is
1394 * called very infrequently and that a given device has a small number
1397 * @start is used to store the start of the free space if we find. But if we
1398 * don't find suitable free space, it will be used to store the start position
1399 * of the max free space.
1401 * @len is used to store the size of the free space that we find.
1402 * But if we don't find suitable free space, it is used to store the size of
1403 * the max free space.
1405 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1406 struct btrfs_device *device, u64 num_bytes,
1407 u64 search_start, u64 *start, u64 *len)
1409 struct btrfs_fs_info *fs_info = device->fs_info;
1410 struct btrfs_root *root = fs_info->dev_root;
1411 struct btrfs_key key;
1412 struct btrfs_dev_extent *dev_extent;
1413 struct btrfs_path *path;
1418 u64 search_end = device->total_bytes;
1421 struct extent_buffer *l;
1424 * We don't want to overwrite the superblock on the drive nor any area
1425 * used by the boot loader (grub for example), so we make sure to start
1426 * at an offset of at least 1MB.
1428 search_start = max_t(u64, search_start, SZ_1M);
1430 path = btrfs_alloc_path();
1434 max_hole_start = search_start;
1438 if (search_start >= search_end ||
1439 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1444 path->reada = READA_FORWARD;
1445 path->search_commit_root = 1;
1446 path->skip_locking = 1;
1448 key.objectid = device->devid;
1449 key.offset = search_start;
1450 key.type = BTRFS_DEV_EXTENT_KEY;
1452 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1456 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1463 slot = path->slots[0];
1464 if (slot >= btrfs_header_nritems(l)) {
1465 ret = btrfs_next_leaf(root, path);
1473 btrfs_item_key_to_cpu(l, &key, slot);
1475 if (key.objectid < device->devid)
1478 if (key.objectid > device->devid)
1481 if (key.type != BTRFS_DEV_EXTENT_KEY)
1484 if (key.offset > search_start) {
1485 hole_size = key.offset - search_start;
1488 * Have to check before we set max_hole_start, otherwise
1489 * we could end up sending back this offset anyway.
1491 if (contains_pending_extent(transaction, device,
1494 if (key.offset >= search_start) {
1495 hole_size = key.offset - search_start;
1502 if (hole_size > max_hole_size) {
1503 max_hole_start = search_start;
1504 max_hole_size = hole_size;
1508 * If this free space is greater than which we need,
1509 * it must be the max free space that we have found
1510 * until now, so max_hole_start must point to the start
1511 * of this free space and the length of this free space
1512 * is stored in max_hole_size. Thus, we return
1513 * max_hole_start and max_hole_size and go back to the
1516 if (hole_size >= num_bytes) {
1522 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1523 extent_end = key.offset + btrfs_dev_extent_length(l,
1525 if (extent_end > search_start)
1526 search_start = extent_end;
1533 * At this point, search_start should be the end of
1534 * allocated dev extents, and when shrinking the device,
1535 * search_end may be smaller than search_start.
1537 if (search_end > search_start) {
1538 hole_size = search_end - search_start;
1540 if (contains_pending_extent(transaction, device, &search_start,
1542 btrfs_release_path(path);
1546 if (hole_size > max_hole_size) {
1547 max_hole_start = search_start;
1548 max_hole_size = hole_size;
1553 if (max_hole_size < num_bytes)
1559 btrfs_free_path(path);
1560 *start = max_hole_start;
1562 *len = max_hole_size;
1566 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1567 struct btrfs_device *device, u64 num_bytes,
1568 u64 *start, u64 *len)
1570 /* FIXME use last free of some kind */
1571 return find_free_dev_extent_start(trans->transaction, device,
1572 num_bytes, 0, start, len);
1575 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1576 struct btrfs_device *device,
1577 u64 start, u64 *dev_extent_len)
1579 struct btrfs_fs_info *fs_info = device->fs_info;
1580 struct btrfs_root *root = fs_info->dev_root;
1582 struct btrfs_path *path;
1583 struct btrfs_key key;
1584 struct btrfs_key found_key;
1585 struct extent_buffer *leaf = NULL;
1586 struct btrfs_dev_extent *extent = NULL;
1588 path = btrfs_alloc_path();
1592 key.objectid = device->devid;
1594 key.type = BTRFS_DEV_EXTENT_KEY;
1596 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1598 ret = btrfs_previous_item(root, path, key.objectid,
1599 BTRFS_DEV_EXTENT_KEY);
1602 leaf = path->nodes[0];
1603 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1604 extent = btrfs_item_ptr(leaf, path->slots[0],
1605 struct btrfs_dev_extent);
1606 BUG_ON(found_key.offset > start || found_key.offset +
1607 btrfs_dev_extent_length(leaf, extent) < start);
1609 btrfs_release_path(path);
1611 } else if (ret == 0) {
1612 leaf = path->nodes[0];
1613 extent = btrfs_item_ptr(leaf, path->slots[0],
1614 struct btrfs_dev_extent);
1616 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1620 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1622 ret = btrfs_del_item(trans, root, path);
1624 btrfs_handle_fs_error(fs_info, ret,
1625 "Failed to remove dev extent item");
1627 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1630 btrfs_free_path(path);
1634 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1635 struct btrfs_device *device,
1636 u64 chunk_offset, u64 start, u64 num_bytes)
1639 struct btrfs_path *path;
1640 struct btrfs_fs_info *fs_info = device->fs_info;
1641 struct btrfs_root *root = fs_info->dev_root;
1642 struct btrfs_dev_extent *extent;
1643 struct extent_buffer *leaf;
1644 struct btrfs_key key;
1646 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1647 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1648 path = btrfs_alloc_path();
1652 key.objectid = device->devid;
1654 key.type = BTRFS_DEV_EXTENT_KEY;
1655 ret = btrfs_insert_empty_item(trans, root, path, &key,
1660 leaf = path->nodes[0];
1661 extent = btrfs_item_ptr(leaf, path->slots[0],
1662 struct btrfs_dev_extent);
1663 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1664 BTRFS_CHUNK_TREE_OBJECTID);
1665 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1666 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1667 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1669 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1670 btrfs_mark_buffer_dirty(leaf);
1672 btrfs_free_path(path);
1676 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1678 struct extent_map_tree *em_tree;
1679 struct extent_map *em;
1683 em_tree = &fs_info->mapping_tree.map_tree;
1684 read_lock(&em_tree->lock);
1685 n = rb_last(&em_tree->map);
1687 em = rb_entry(n, struct extent_map, rb_node);
1688 ret = em->start + em->len;
1690 read_unlock(&em_tree->lock);
1695 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1699 struct btrfs_key key;
1700 struct btrfs_key found_key;
1701 struct btrfs_path *path;
1703 path = btrfs_alloc_path();
1707 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1708 key.type = BTRFS_DEV_ITEM_KEY;
1709 key.offset = (u64)-1;
1711 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1715 BUG_ON(ret == 0); /* Corruption */
1717 ret = btrfs_previous_item(fs_info->chunk_root, path,
1718 BTRFS_DEV_ITEMS_OBJECTID,
1719 BTRFS_DEV_ITEM_KEY);
1723 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1725 *devid_ret = found_key.offset + 1;
1729 btrfs_free_path(path);
1734 * the device information is stored in the chunk root
1735 * the btrfs_device struct should be fully filled in
1737 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1738 struct btrfs_fs_info *fs_info,
1739 struct btrfs_device *device)
1741 struct btrfs_root *root = fs_info->chunk_root;
1743 struct btrfs_path *path;
1744 struct btrfs_dev_item *dev_item;
1745 struct extent_buffer *leaf;
1746 struct btrfs_key key;
1749 path = btrfs_alloc_path();
1753 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1754 key.type = BTRFS_DEV_ITEM_KEY;
1755 key.offset = device->devid;
1757 ret = btrfs_insert_empty_item(trans, root, path, &key,
1762 leaf = path->nodes[0];
1763 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1765 btrfs_set_device_id(leaf, dev_item, device->devid);
1766 btrfs_set_device_generation(leaf, dev_item, 0);
1767 btrfs_set_device_type(leaf, dev_item, device->type);
1768 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1769 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1770 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1771 btrfs_set_device_total_bytes(leaf, dev_item,
1772 btrfs_device_get_disk_total_bytes(device));
1773 btrfs_set_device_bytes_used(leaf, dev_item,
1774 btrfs_device_get_bytes_used(device));
1775 btrfs_set_device_group(leaf, dev_item, 0);
1776 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1777 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1778 btrfs_set_device_start_offset(leaf, dev_item, 0);
1780 ptr = btrfs_device_uuid(dev_item);
1781 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1782 ptr = btrfs_device_fsid(dev_item);
1783 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1784 btrfs_mark_buffer_dirty(leaf);
1788 btrfs_free_path(path);
1793 * Function to update ctime/mtime for a given device path.
1794 * Mainly used for ctime/mtime based probe like libblkid.
1796 static void update_dev_time(const char *path_name)
1800 filp = filp_open(path_name, O_RDWR, 0);
1803 file_update_time(filp);
1804 filp_close(filp, NULL);
1807 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1808 struct btrfs_device *device)
1810 struct btrfs_root *root = fs_info->chunk_root;
1812 struct btrfs_path *path;
1813 struct btrfs_key key;
1814 struct btrfs_trans_handle *trans;
1816 path = btrfs_alloc_path();
1820 trans = btrfs_start_transaction(root, 0);
1821 if (IS_ERR(trans)) {
1822 btrfs_free_path(path);
1823 return PTR_ERR(trans);
1825 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1826 key.type = BTRFS_DEV_ITEM_KEY;
1827 key.offset = device->devid;
1829 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1833 btrfs_abort_transaction(trans, ret);
1834 btrfs_end_transaction(trans);
1838 ret = btrfs_del_item(trans, root, path);
1840 btrfs_abort_transaction(trans, ret);
1841 btrfs_end_transaction(trans);
1845 btrfs_free_path(path);
1847 ret = btrfs_commit_transaction(trans);
1852 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1853 * filesystem. It's up to the caller to adjust that number regarding eg. device
1856 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1864 seq = read_seqbegin(&fs_info->profiles_lock);
1866 all_avail = fs_info->avail_data_alloc_bits |
1867 fs_info->avail_system_alloc_bits |
1868 fs_info->avail_metadata_alloc_bits;
1869 } while (read_seqretry(&fs_info->profiles_lock, seq));
1871 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1872 if (!(all_avail & btrfs_raid_group[i]))
1875 if (num_devices < btrfs_raid_array[i].devs_min) {
1876 int ret = btrfs_raid_mindev_error[i];
1886 static struct btrfs_device * btrfs_find_next_active_device(
1887 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1889 struct btrfs_device *next_device;
1891 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1892 if (next_device != device &&
1893 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1894 && next_device->bdev)
1902 * Helper function to check if the given device is part of s_bdev / latest_bdev
1903 * and replace it with the provided or the next active device, in the context
1904 * where this function called, there should be always be another device (or
1905 * this_dev) which is active.
1907 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1908 struct btrfs_device *device, struct btrfs_device *this_dev)
1910 struct btrfs_device *next_device;
1913 next_device = this_dev;
1915 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1917 ASSERT(next_device);
1919 if (fs_info->sb->s_bdev &&
1920 (fs_info->sb->s_bdev == device->bdev))
1921 fs_info->sb->s_bdev = next_device->bdev;
1923 if (fs_info->fs_devices->latest_bdev == device->bdev)
1924 fs_info->fs_devices->latest_bdev = next_device->bdev;
1927 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1930 struct btrfs_device *device;
1931 struct btrfs_fs_devices *cur_devices;
1935 mutex_lock(&fs_info->volume_mutex);
1936 mutex_lock(&uuid_mutex);
1938 num_devices = fs_info->fs_devices->num_devices;
1939 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1940 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1941 WARN_ON(num_devices < 1);
1944 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1946 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1950 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1955 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1956 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1960 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1961 fs_info->fs_devices->rw_devices == 1) {
1962 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1966 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1967 mutex_lock(&fs_info->chunk_mutex);
1968 list_del_init(&device->dev_alloc_list);
1969 device->fs_devices->rw_devices--;
1970 mutex_unlock(&fs_info->chunk_mutex);
1973 mutex_unlock(&uuid_mutex);
1974 ret = btrfs_shrink_device(device, 0);
1975 mutex_lock(&uuid_mutex);
1980 * TODO: the superblock still includes this device in its num_devices
1981 * counter although write_all_supers() is not locked out. This
1982 * could give a filesystem state which requires a degraded mount.
1984 ret = btrfs_rm_dev_item(fs_info, device);
1988 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1989 btrfs_scrub_cancel_dev(fs_info, device);
1992 * the device list mutex makes sure that we don't change
1993 * the device list while someone else is writing out all
1994 * the device supers. Whoever is writing all supers, should
1995 * lock the device list mutex before getting the number of
1996 * devices in the super block (super_copy). Conversely,
1997 * whoever updates the number of devices in the super block
1998 * (super_copy) should hold the device list mutex.
2001 cur_devices = device->fs_devices;
2002 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2003 list_del_rcu(&device->dev_list);
2005 device->fs_devices->num_devices--;
2006 device->fs_devices->total_devices--;
2008 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2009 device->fs_devices->missing_devices--;
2011 btrfs_assign_next_active_device(fs_info, device, NULL);
2014 device->fs_devices->open_devices--;
2015 /* remove sysfs entry */
2016 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2019 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2020 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2021 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2024 * at this point, the device is zero sized and detached from
2025 * the devices list. All that's left is to zero out the old
2026 * supers and free the device.
2028 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2029 btrfs_scratch_superblocks(device->bdev, device->name->str);
2031 btrfs_close_bdev(device);
2032 call_rcu(&device->rcu, free_device_rcu);
2034 if (cur_devices->open_devices == 0) {
2035 struct btrfs_fs_devices *fs_devices;
2036 fs_devices = fs_info->fs_devices;
2037 while (fs_devices) {
2038 if (fs_devices->seed == cur_devices) {
2039 fs_devices->seed = cur_devices->seed;
2042 fs_devices = fs_devices->seed;
2044 cur_devices->seed = NULL;
2045 __btrfs_close_devices(cur_devices);
2046 free_fs_devices(cur_devices);
2050 mutex_unlock(&uuid_mutex);
2051 mutex_unlock(&fs_info->volume_mutex);
2055 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2056 mutex_lock(&fs_info->chunk_mutex);
2057 list_add(&device->dev_alloc_list,
2058 &fs_info->fs_devices->alloc_list);
2059 device->fs_devices->rw_devices++;
2060 mutex_unlock(&fs_info->chunk_mutex);
2065 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2066 struct btrfs_device *srcdev)
2068 struct btrfs_fs_devices *fs_devices;
2070 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2073 * in case of fs with no seed, srcdev->fs_devices will point
2074 * to fs_devices of fs_info. However when the dev being replaced is
2075 * a seed dev it will point to the seed's local fs_devices. In short
2076 * srcdev will have its correct fs_devices in both the cases.
2078 fs_devices = srcdev->fs_devices;
2080 list_del_rcu(&srcdev->dev_list);
2081 list_del(&srcdev->dev_alloc_list);
2082 fs_devices->num_devices--;
2083 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2084 fs_devices->missing_devices--;
2086 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2087 fs_devices->rw_devices--;
2090 fs_devices->open_devices--;
2093 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2094 struct btrfs_device *srcdev)
2096 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2098 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2099 /* zero out the old super if it is writable */
2100 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2103 btrfs_close_bdev(srcdev);
2104 call_rcu(&srcdev->rcu, free_device_rcu);
2106 /* if this is no devs we rather delete the fs_devices */
2107 if (!fs_devices->num_devices) {
2108 struct btrfs_fs_devices *tmp_fs_devices;
2111 * On a mounted FS, num_devices can't be zero unless it's a
2112 * seed. In case of a seed device being replaced, the replace
2113 * target added to the sprout FS, so there will be no more
2114 * device left under the seed FS.
2116 ASSERT(fs_devices->seeding);
2118 tmp_fs_devices = fs_info->fs_devices;
2119 while (tmp_fs_devices) {
2120 if (tmp_fs_devices->seed == fs_devices) {
2121 tmp_fs_devices->seed = fs_devices->seed;
2124 tmp_fs_devices = tmp_fs_devices->seed;
2126 fs_devices->seed = NULL;
2127 __btrfs_close_devices(fs_devices);
2128 free_fs_devices(fs_devices);
2132 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2133 struct btrfs_device *tgtdev)
2135 mutex_lock(&uuid_mutex);
2137 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2139 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2142 fs_info->fs_devices->open_devices--;
2144 fs_info->fs_devices->num_devices--;
2146 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2148 list_del_rcu(&tgtdev->dev_list);
2150 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2151 mutex_unlock(&uuid_mutex);
2154 * The update_dev_time() with in btrfs_scratch_superblocks()
2155 * may lead to a call to btrfs_show_devname() which will try
2156 * to hold device_list_mutex. And here this device
2157 * is already out of device list, so we don't have to hold
2158 * the device_list_mutex lock.
2160 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2162 btrfs_close_bdev(tgtdev);
2163 call_rcu(&tgtdev->rcu, free_device_rcu);
2166 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2167 const char *device_path,
2168 struct btrfs_device **device)
2171 struct btrfs_super_block *disk_super;
2174 struct block_device *bdev;
2175 struct buffer_head *bh;
2178 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2179 fs_info->bdev_holder, 0, &bdev, &bh);
2182 disk_super = (struct btrfs_super_block *)bh->b_data;
2183 devid = btrfs_stack_device_id(&disk_super->dev_item);
2184 dev_uuid = disk_super->dev_item.uuid;
2185 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2189 blkdev_put(bdev, FMODE_READ);
2193 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2194 const char *device_path,
2195 struct btrfs_device **device)
2198 if (strcmp(device_path, "missing") == 0) {
2199 struct list_head *devices;
2200 struct btrfs_device *tmp;
2202 devices = &fs_info->fs_devices->devices;
2204 * It is safe to read the devices since the volume_mutex
2205 * is held by the caller.
2207 list_for_each_entry(tmp, devices, dev_list) {
2208 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2209 &tmp->dev_state) && !tmp->bdev) {
2216 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2220 return btrfs_find_device_by_path(fs_info, device_path, device);
2225 * Lookup a device given by device id, or the path if the id is 0.
2227 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2228 const char *devpath,
2229 struct btrfs_device **device)
2235 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2239 if (!devpath || !devpath[0])
2242 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2249 * does all the dirty work required for changing file system's UUID.
2251 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2253 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2254 struct btrfs_fs_devices *old_devices;
2255 struct btrfs_fs_devices *seed_devices;
2256 struct btrfs_super_block *disk_super = fs_info->super_copy;
2257 struct btrfs_device *device;
2260 lockdep_assert_held(&uuid_mutex);
2261 if (!fs_devices->seeding)
2264 seed_devices = alloc_fs_devices(NULL);
2265 if (IS_ERR(seed_devices))
2266 return PTR_ERR(seed_devices);
2268 old_devices = clone_fs_devices(fs_devices);
2269 if (IS_ERR(old_devices)) {
2270 kfree(seed_devices);
2271 return PTR_ERR(old_devices);
2274 list_add(&old_devices->list, &fs_uuids);
2276 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2277 seed_devices->opened = 1;
2278 INIT_LIST_HEAD(&seed_devices->devices);
2279 INIT_LIST_HEAD(&seed_devices->alloc_list);
2280 mutex_init(&seed_devices->device_list_mutex);
2282 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2283 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2285 list_for_each_entry(device, &seed_devices->devices, dev_list)
2286 device->fs_devices = seed_devices;
2288 mutex_lock(&fs_info->chunk_mutex);
2289 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2290 mutex_unlock(&fs_info->chunk_mutex);
2292 fs_devices->seeding = 0;
2293 fs_devices->num_devices = 0;
2294 fs_devices->open_devices = 0;
2295 fs_devices->missing_devices = 0;
2296 fs_devices->rotating = 0;
2297 fs_devices->seed = seed_devices;
2299 generate_random_uuid(fs_devices->fsid);
2300 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2301 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2302 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2304 super_flags = btrfs_super_flags(disk_super) &
2305 ~BTRFS_SUPER_FLAG_SEEDING;
2306 btrfs_set_super_flags(disk_super, super_flags);
2312 * Store the expected generation for seed devices in device items.
2314 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2315 struct btrfs_fs_info *fs_info)
2317 struct btrfs_root *root = fs_info->chunk_root;
2318 struct btrfs_path *path;
2319 struct extent_buffer *leaf;
2320 struct btrfs_dev_item *dev_item;
2321 struct btrfs_device *device;
2322 struct btrfs_key key;
2323 u8 fs_uuid[BTRFS_FSID_SIZE];
2324 u8 dev_uuid[BTRFS_UUID_SIZE];
2328 path = btrfs_alloc_path();
2332 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2334 key.type = BTRFS_DEV_ITEM_KEY;
2337 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2341 leaf = path->nodes[0];
2343 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2344 ret = btrfs_next_leaf(root, path);
2349 leaf = path->nodes[0];
2350 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2351 btrfs_release_path(path);
2355 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2356 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2357 key.type != BTRFS_DEV_ITEM_KEY)
2360 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2361 struct btrfs_dev_item);
2362 devid = btrfs_device_id(leaf, dev_item);
2363 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2365 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2367 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2368 BUG_ON(!device); /* Logic error */
2370 if (device->fs_devices->seeding) {
2371 btrfs_set_device_generation(leaf, dev_item,
2372 device->generation);
2373 btrfs_mark_buffer_dirty(leaf);
2381 btrfs_free_path(path);
2385 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2387 struct btrfs_root *root = fs_info->dev_root;
2388 struct request_queue *q;
2389 struct btrfs_trans_handle *trans;
2390 struct btrfs_device *device;
2391 struct block_device *bdev;
2392 struct list_head *devices;
2393 struct super_block *sb = fs_info->sb;
2394 struct rcu_string *name;
2396 int seeding_dev = 0;
2398 bool unlocked = false;
2400 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2403 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2404 fs_info->bdev_holder);
2406 return PTR_ERR(bdev);
2408 if (fs_info->fs_devices->seeding) {
2410 down_write(&sb->s_umount);
2411 mutex_lock(&uuid_mutex);
2414 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2416 devices = &fs_info->fs_devices->devices;
2418 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2419 list_for_each_entry(device, devices, dev_list) {
2420 if (device->bdev == bdev) {
2423 &fs_info->fs_devices->device_list_mutex);
2427 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2429 device = btrfs_alloc_device(fs_info, NULL, NULL);
2430 if (IS_ERR(device)) {
2431 /* we can safely leave the fs_devices entry around */
2432 ret = PTR_ERR(device);
2436 name = rcu_string_strdup(device_path, GFP_KERNEL);
2439 goto error_free_device;
2441 rcu_assign_pointer(device->name, name);
2443 trans = btrfs_start_transaction(root, 0);
2444 if (IS_ERR(trans)) {
2445 ret = PTR_ERR(trans);
2446 goto error_free_device;
2449 q = bdev_get_queue(bdev);
2450 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2451 device->generation = trans->transid;
2452 device->io_width = fs_info->sectorsize;
2453 device->io_align = fs_info->sectorsize;
2454 device->sector_size = fs_info->sectorsize;
2455 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2456 fs_info->sectorsize);
2457 device->disk_total_bytes = device->total_bytes;
2458 device->commit_total_bytes = device->total_bytes;
2459 device->fs_info = fs_info;
2460 device->bdev = bdev;
2461 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2462 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2463 device->mode = FMODE_EXCL;
2464 device->dev_stats_valid = 1;
2465 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2468 sb->s_flags &= ~SB_RDONLY;
2469 ret = btrfs_prepare_sprout(fs_info);
2471 btrfs_abort_transaction(trans, ret);
2476 device->fs_devices = fs_info->fs_devices;
2478 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2479 mutex_lock(&fs_info->chunk_mutex);
2480 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2481 list_add(&device->dev_alloc_list,
2482 &fs_info->fs_devices->alloc_list);
2483 fs_info->fs_devices->num_devices++;
2484 fs_info->fs_devices->open_devices++;
2485 fs_info->fs_devices->rw_devices++;
2486 fs_info->fs_devices->total_devices++;
2487 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2489 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2491 if (!blk_queue_nonrot(q))
2492 fs_info->fs_devices->rotating = 1;
2494 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2495 btrfs_set_super_total_bytes(fs_info->super_copy,
2496 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2498 tmp = btrfs_super_num_devices(fs_info->super_copy);
2499 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2501 /* add sysfs device entry */
2502 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2505 * we've got more storage, clear any full flags on the space
2508 btrfs_clear_space_info_full(fs_info);
2510 mutex_unlock(&fs_info->chunk_mutex);
2511 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2514 mutex_lock(&fs_info->chunk_mutex);
2515 ret = init_first_rw_device(trans, fs_info);
2516 mutex_unlock(&fs_info->chunk_mutex);
2518 btrfs_abort_transaction(trans, ret);
2523 ret = btrfs_add_dev_item(trans, fs_info, device);
2525 btrfs_abort_transaction(trans, ret);
2530 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2532 ret = btrfs_finish_sprout(trans, fs_info);
2534 btrfs_abort_transaction(trans, ret);
2538 /* Sprouting would change fsid of the mounted root,
2539 * so rename the fsid on the sysfs
2541 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2543 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2545 "sysfs: failed to create fsid for sprout");
2548 ret = btrfs_commit_transaction(trans);
2551 mutex_unlock(&uuid_mutex);
2552 up_write(&sb->s_umount);
2555 if (ret) /* transaction commit */
2558 ret = btrfs_relocate_sys_chunks(fs_info);
2560 btrfs_handle_fs_error(fs_info, ret,
2561 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2562 trans = btrfs_attach_transaction(root);
2563 if (IS_ERR(trans)) {
2564 if (PTR_ERR(trans) == -ENOENT)
2566 ret = PTR_ERR(trans);
2570 ret = btrfs_commit_transaction(trans);
2573 /* Update ctime/mtime for libblkid */
2574 update_dev_time(device_path);
2578 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2581 sb->s_flags |= SB_RDONLY;
2583 btrfs_end_transaction(trans);
2585 free_device(device);
2587 blkdev_put(bdev, FMODE_EXCL);
2588 if (seeding_dev && !unlocked) {
2589 mutex_unlock(&uuid_mutex);
2590 up_write(&sb->s_umount);
2595 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2596 const char *device_path,
2597 struct btrfs_device *srcdev,
2598 struct btrfs_device **device_out)
2600 struct btrfs_device *device;
2601 struct block_device *bdev;
2602 struct list_head *devices;
2603 struct rcu_string *name;
2604 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2608 if (fs_info->fs_devices->seeding) {
2609 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2613 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2614 fs_info->bdev_holder);
2616 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2617 return PTR_ERR(bdev);
2620 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2622 devices = &fs_info->fs_devices->devices;
2623 list_for_each_entry(device, devices, dev_list) {
2624 if (device->bdev == bdev) {
2626 "target device is in the filesystem!");
2633 if (i_size_read(bdev->bd_inode) <
2634 btrfs_device_get_total_bytes(srcdev)) {
2636 "target device is smaller than source device!");
2642 device = btrfs_alloc_device(NULL, &devid, NULL);
2643 if (IS_ERR(device)) {
2644 ret = PTR_ERR(device);
2648 name = rcu_string_strdup(device_path, GFP_KERNEL);
2650 free_device(device);
2654 rcu_assign_pointer(device->name, name);
2656 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2657 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2658 device->generation = 0;
2659 device->io_width = fs_info->sectorsize;
2660 device->io_align = fs_info->sectorsize;
2661 device->sector_size = fs_info->sectorsize;
2662 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2663 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2664 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2665 device->commit_total_bytes = srcdev->commit_total_bytes;
2666 device->commit_bytes_used = device->bytes_used;
2667 device->fs_info = fs_info;
2668 device->bdev = bdev;
2669 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2670 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2671 device->mode = FMODE_EXCL;
2672 device->dev_stats_valid = 1;
2673 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2674 device->fs_devices = fs_info->fs_devices;
2675 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2676 fs_info->fs_devices->num_devices++;
2677 fs_info->fs_devices->open_devices++;
2678 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2680 *device_out = device;
2684 blkdev_put(bdev, FMODE_EXCL);
2688 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2689 struct btrfs_device *device)
2692 struct btrfs_path *path;
2693 struct btrfs_root *root = device->fs_info->chunk_root;
2694 struct btrfs_dev_item *dev_item;
2695 struct extent_buffer *leaf;
2696 struct btrfs_key key;
2698 path = btrfs_alloc_path();
2702 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2703 key.type = BTRFS_DEV_ITEM_KEY;
2704 key.offset = device->devid;
2706 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2715 leaf = path->nodes[0];
2716 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2718 btrfs_set_device_id(leaf, dev_item, device->devid);
2719 btrfs_set_device_type(leaf, dev_item, device->type);
2720 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2721 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2722 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2723 btrfs_set_device_total_bytes(leaf, dev_item,
2724 btrfs_device_get_disk_total_bytes(device));
2725 btrfs_set_device_bytes_used(leaf, dev_item,
2726 btrfs_device_get_bytes_used(device));
2727 btrfs_mark_buffer_dirty(leaf);
2730 btrfs_free_path(path);
2734 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2735 struct btrfs_device *device, u64 new_size)
2737 struct btrfs_fs_info *fs_info = device->fs_info;
2738 struct btrfs_super_block *super_copy = fs_info->super_copy;
2739 struct btrfs_fs_devices *fs_devices;
2743 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2746 new_size = round_down(new_size, fs_info->sectorsize);
2748 mutex_lock(&fs_info->chunk_mutex);
2749 old_total = btrfs_super_total_bytes(super_copy);
2750 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2752 if (new_size <= device->total_bytes ||
2753 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2754 mutex_unlock(&fs_info->chunk_mutex);
2758 fs_devices = fs_info->fs_devices;
2760 btrfs_set_super_total_bytes(super_copy,
2761 round_down(old_total + diff, fs_info->sectorsize));
2762 device->fs_devices->total_rw_bytes += diff;
2764 btrfs_device_set_total_bytes(device, new_size);
2765 btrfs_device_set_disk_total_bytes(device, new_size);
2766 btrfs_clear_space_info_full(device->fs_info);
2767 if (list_empty(&device->resized_list))
2768 list_add_tail(&device->resized_list,
2769 &fs_devices->resized_devices);
2770 mutex_unlock(&fs_info->chunk_mutex);
2772 return btrfs_update_device(trans, device);
2775 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2776 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2778 struct btrfs_root *root = fs_info->chunk_root;
2780 struct btrfs_path *path;
2781 struct btrfs_key key;
2783 path = btrfs_alloc_path();
2787 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2788 key.offset = chunk_offset;
2789 key.type = BTRFS_CHUNK_ITEM_KEY;
2791 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2794 else if (ret > 0) { /* Logic error or corruption */
2795 btrfs_handle_fs_error(fs_info, -ENOENT,
2796 "Failed lookup while freeing chunk.");
2801 ret = btrfs_del_item(trans, root, path);
2803 btrfs_handle_fs_error(fs_info, ret,
2804 "Failed to delete chunk item.");
2806 btrfs_free_path(path);
2810 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2812 struct btrfs_super_block *super_copy = fs_info->super_copy;
2813 struct btrfs_disk_key *disk_key;
2814 struct btrfs_chunk *chunk;
2821 struct btrfs_key key;
2823 mutex_lock(&fs_info->chunk_mutex);
2824 array_size = btrfs_super_sys_array_size(super_copy);
2826 ptr = super_copy->sys_chunk_array;
2829 while (cur < array_size) {
2830 disk_key = (struct btrfs_disk_key *)ptr;
2831 btrfs_disk_key_to_cpu(&key, disk_key);
2833 len = sizeof(*disk_key);
2835 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2836 chunk = (struct btrfs_chunk *)(ptr + len);
2837 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2838 len += btrfs_chunk_item_size(num_stripes);
2843 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2844 key.offset == chunk_offset) {
2845 memmove(ptr, ptr + len, array_size - (cur + len));
2847 btrfs_set_super_sys_array_size(super_copy, array_size);
2853 mutex_unlock(&fs_info->chunk_mutex);
2857 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2858 u64 logical, u64 length)
2860 struct extent_map_tree *em_tree;
2861 struct extent_map *em;
2863 em_tree = &fs_info->mapping_tree.map_tree;
2864 read_lock(&em_tree->lock);
2865 em = lookup_extent_mapping(em_tree, logical, length);
2866 read_unlock(&em_tree->lock);
2869 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2871 return ERR_PTR(-EINVAL);
2874 if (em->start > logical || em->start + em->len < logical) {
2876 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2877 logical, length, em->start, em->start + em->len);
2878 free_extent_map(em);
2879 return ERR_PTR(-EINVAL);
2882 /* callers are responsible for dropping em's ref. */
2886 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2887 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2889 struct extent_map *em;
2890 struct map_lookup *map;
2891 u64 dev_extent_len = 0;
2893 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2895 em = get_chunk_map(fs_info, chunk_offset, 1);
2898 * This is a logic error, but we don't want to just rely on the
2899 * user having built with ASSERT enabled, so if ASSERT doesn't
2900 * do anything we still error out.
2905 map = em->map_lookup;
2906 mutex_lock(&fs_info->chunk_mutex);
2907 check_system_chunk(trans, fs_info, map->type);
2908 mutex_unlock(&fs_info->chunk_mutex);
2911 * Take the device list mutex to prevent races with the final phase of
2912 * a device replace operation that replaces the device object associated
2913 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2915 mutex_lock(&fs_devices->device_list_mutex);
2916 for (i = 0; i < map->num_stripes; i++) {
2917 struct btrfs_device *device = map->stripes[i].dev;
2918 ret = btrfs_free_dev_extent(trans, device,
2919 map->stripes[i].physical,
2922 mutex_unlock(&fs_devices->device_list_mutex);
2923 btrfs_abort_transaction(trans, ret);
2927 if (device->bytes_used > 0) {
2928 mutex_lock(&fs_info->chunk_mutex);
2929 btrfs_device_set_bytes_used(device,
2930 device->bytes_used - dev_extent_len);
2931 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2932 btrfs_clear_space_info_full(fs_info);
2933 mutex_unlock(&fs_info->chunk_mutex);
2936 if (map->stripes[i].dev) {
2937 ret = btrfs_update_device(trans, map->stripes[i].dev);
2939 mutex_unlock(&fs_devices->device_list_mutex);
2940 btrfs_abort_transaction(trans, ret);
2945 mutex_unlock(&fs_devices->device_list_mutex);
2947 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2949 btrfs_abort_transaction(trans, ret);
2953 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2955 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2956 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2958 btrfs_abort_transaction(trans, ret);
2963 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2965 btrfs_abort_transaction(trans, ret);
2971 free_extent_map(em);
2975 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2977 struct btrfs_root *root = fs_info->chunk_root;
2978 struct btrfs_trans_handle *trans;
2982 * Prevent races with automatic removal of unused block groups.
2983 * After we relocate and before we remove the chunk with offset
2984 * chunk_offset, automatic removal of the block group can kick in,
2985 * resulting in a failure when calling btrfs_remove_chunk() below.
2987 * Make sure to acquire this mutex before doing a tree search (dev
2988 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2989 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2990 * we release the path used to search the chunk/dev tree and before
2991 * the current task acquires this mutex and calls us.
2993 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2995 ret = btrfs_can_relocate(fs_info, chunk_offset);
2999 /* step one, relocate all the extents inside this chunk */
3000 btrfs_scrub_pause(fs_info);
3001 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3002 btrfs_scrub_continue(fs_info);
3007 * We add the kobjects here (and after forcing data chunk creation)
3008 * since relocation is the only place we'll create chunks of a new
3009 * type at runtime. The only place where we'll remove the last
3010 * chunk of a type is the call immediately below this one. Even
3011 * so, we're protected against races with the cleaner thread since
3012 * we're covered by the delete_unused_bgs_mutex.
3014 btrfs_add_raid_kobjects(fs_info);
3016 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3018 if (IS_ERR(trans)) {
3019 ret = PTR_ERR(trans);
3020 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3025 * step two, delete the device extents and the
3026 * chunk tree entries
3028 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3029 btrfs_end_transaction(trans);
3033 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3035 struct btrfs_root *chunk_root = fs_info->chunk_root;
3036 struct btrfs_path *path;
3037 struct extent_buffer *leaf;
3038 struct btrfs_chunk *chunk;
3039 struct btrfs_key key;
3040 struct btrfs_key found_key;
3042 bool retried = false;
3046 path = btrfs_alloc_path();
3051 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3052 key.offset = (u64)-1;
3053 key.type = BTRFS_CHUNK_ITEM_KEY;
3056 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3057 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3059 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3062 BUG_ON(ret == 0); /* Corruption */
3064 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3067 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3073 leaf = path->nodes[0];
3074 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3076 chunk = btrfs_item_ptr(leaf, path->slots[0],
3077 struct btrfs_chunk);
3078 chunk_type = btrfs_chunk_type(leaf, chunk);
3079 btrfs_release_path(path);
3081 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3082 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3088 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3090 if (found_key.offset == 0)
3092 key.offset = found_key.offset - 1;
3095 if (failed && !retried) {
3099 } else if (WARN_ON(failed && retried)) {
3103 btrfs_free_path(path);
3108 * return 1 : allocate a data chunk successfully,
3109 * return <0: errors during allocating a data chunk,
3110 * return 0 : no need to allocate a data chunk.
3112 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3115 struct btrfs_block_group_cache *cache;
3119 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3121 chunk_type = cache->flags;
3122 btrfs_put_block_group(cache);
3124 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3125 spin_lock(&fs_info->data_sinfo->lock);
3126 bytes_used = fs_info->data_sinfo->bytes_used;
3127 spin_unlock(&fs_info->data_sinfo->lock);
3130 struct btrfs_trans_handle *trans;
3133 trans = btrfs_join_transaction(fs_info->tree_root);
3135 return PTR_ERR(trans);
3137 ret = btrfs_force_chunk_alloc(trans, fs_info,
3138 BTRFS_BLOCK_GROUP_DATA);
3139 btrfs_end_transaction(trans);
3143 btrfs_add_raid_kobjects(fs_info);
3151 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3152 struct btrfs_balance_control *bctl)
3154 struct btrfs_root *root = fs_info->tree_root;
3155 struct btrfs_trans_handle *trans;
3156 struct btrfs_balance_item *item;
3157 struct btrfs_disk_balance_args disk_bargs;
3158 struct btrfs_path *path;
3159 struct extent_buffer *leaf;
3160 struct btrfs_key key;
3163 path = btrfs_alloc_path();
3167 trans = btrfs_start_transaction(root, 0);
3168 if (IS_ERR(trans)) {
3169 btrfs_free_path(path);
3170 return PTR_ERR(trans);
3173 key.objectid = BTRFS_BALANCE_OBJECTID;
3174 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3177 ret = btrfs_insert_empty_item(trans, root, path, &key,
3182 leaf = path->nodes[0];
3183 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3185 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3187 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3188 btrfs_set_balance_data(leaf, item, &disk_bargs);
3189 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3190 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3191 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3192 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3194 btrfs_set_balance_flags(leaf, item, bctl->flags);
3196 btrfs_mark_buffer_dirty(leaf);
3198 btrfs_free_path(path);
3199 err = btrfs_commit_transaction(trans);
3205 static int del_balance_item(struct btrfs_fs_info *fs_info)
3207 struct btrfs_root *root = fs_info->tree_root;
3208 struct btrfs_trans_handle *trans;
3209 struct btrfs_path *path;
3210 struct btrfs_key key;
3213 path = btrfs_alloc_path();
3217 trans = btrfs_start_transaction(root, 0);
3218 if (IS_ERR(trans)) {
3219 btrfs_free_path(path);
3220 return PTR_ERR(trans);
3223 key.objectid = BTRFS_BALANCE_OBJECTID;
3224 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3227 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3235 ret = btrfs_del_item(trans, root, path);
3237 btrfs_free_path(path);
3238 err = btrfs_commit_transaction(trans);
3245 * This is a heuristic used to reduce the number of chunks balanced on
3246 * resume after balance was interrupted.
3248 static void update_balance_args(struct btrfs_balance_control *bctl)
3251 * Turn on soft mode for chunk types that were being converted.
3253 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3254 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3255 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3256 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3257 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3258 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3261 * Turn on usage filter if is not already used. The idea is
3262 * that chunks that we have already balanced should be
3263 * reasonably full. Don't do it for chunks that are being
3264 * converted - that will keep us from relocating unconverted
3265 * (albeit full) chunks.
3267 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3268 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3269 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3270 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3271 bctl->data.usage = 90;
3273 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3274 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3275 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3276 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3277 bctl->sys.usage = 90;
3279 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3280 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3281 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3282 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3283 bctl->meta.usage = 90;
3288 * Should be called with both balance and volume mutexes held to
3289 * serialize other volume operations (add_dev/rm_dev/resize) with
3290 * restriper. Same goes for unset_balance_control.
3292 static void set_balance_control(struct btrfs_balance_control *bctl)
3294 struct btrfs_fs_info *fs_info = bctl->fs_info;
3296 BUG_ON(fs_info->balance_ctl);
3298 spin_lock(&fs_info->balance_lock);
3299 fs_info->balance_ctl = bctl;
3300 spin_unlock(&fs_info->balance_lock);
3303 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3305 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3307 BUG_ON(!fs_info->balance_ctl);
3309 spin_lock(&fs_info->balance_lock);
3310 fs_info->balance_ctl = NULL;
3311 spin_unlock(&fs_info->balance_lock);
3317 * Balance filters. Return 1 if chunk should be filtered out
3318 * (should not be balanced).
3320 static int chunk_profiles_filter(u64 chunk_type,
3321 struct btrfs_balance_args *bargs)
3323 chunk_type = chunk_to_extended(chunk_type) &
3324 BTRFS_EXTENDED_PROFILE_MASK;
3326 if (bargs->profiles & chunk_type)
3332 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3333 struct btrfs_balance_args *bargs)
3335 struct btrfs_block_group_cache *cache;
3337 u64 user_thresh_min;
3338 u64 user_thresh_max;
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)
3345 user_thresh_min = 0;
3347 user_thresh_min = div_factor_fine(cache->key.offset,
3350 if (bargs->usage_max == 0)
3351 user_thresh_max = 1;
3352 else if (bargs->usage_max > 100)
3353 user_thresh_max = cache->key.offset;
3355 user_thresh_max = div_factor_fine(cache->key.offset,
3358 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3361 btrfs_put_block_group(cache);
3365 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3366 u64 chunk_offset, struct btrfs_balance_args *bargs)
3368 struct btrfs_block_group_cache *cache;
3369 u64 chunk_used, user_thresh;
3372 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3373 chunk_used = btrfs_block_group_used(&cache->item);
3375 if (bargs->usage_min == 0)
3377 else if (bargs->usage > 100)
3378 user_thresh = cache->key.offset;
3380 user_thresh = div_factor_fine(cache->key.offset,
3383 if (chunk_used < user_thresh)
3386 btrfs_put_block_group(cache);
3390 static int chunk_devid_filter(struct extent_buffer *leaf,
3391 struct btrfs_chunk *chunk,
3392 struct btrfs_balance_args *bargs)
3394 struct btrfs_stripe *stripe;
3395 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3398 for (i = 0; i < num_stripes; i++) {
3399 stripe = btrfs_stripe_nr(chunk, i);
3400 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3407 /* [pstart, pend) */
3408 static int chunk_drange_filter(struct extent_buffer *leaf,
3409 struct btrfs_chunk *chunk,
3410 struct btrfs_balance_args *bargs)
3412 struct btrfs_stripe *stripe;
3413 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3419 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3422 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3423 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3424 factor = num_stripes / 2;
3425 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3426 factor = num_stripes - 1;
3427 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3428 factor = num_stripes - 2;
3430 factor = num_stripes;
3433 for (i = 0; i < num_stripes; i++) {
3434 stripe = btrfs_stripe_nr(chunk, i);
3435 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3438 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3439 stripe_length = btrfs_chunk_length(leaf, chunk);
3440 stripe_length = div_u64(stripe_length, factor);
3442 if (stripe_offset < bargs->pend &&
3443 stripe_offset + stripe_length > bargs->pstart)
3450 /* [vstart, vend) */
3451 static int chunk_vrange_filter(struct extent_buffer *leaf,
3452 struct btrfs_chunk *chunk,
3454 struct btrfs_balance_args *bargs)
3456 if (chunk_offset < bargs->vend &&
3457 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3458 /* at least part of the chunk is inside this vrange */
3464 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3465 struct btrfs_chunk *chunk,
3466 struct btrfs_balance_args *bargs)
3468 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3470 if (bargs->stripes_min <= num_stripes
3471 && num_stripes <= bargs->stripes_max)
3477 static int chunk_soft_convert_filter(u64 chunk_type,
3478 struct btrfs_balance_args *bargs)
3480 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3483 chunk_type = chunk_to_extended(chunk_type) &
3484 BTRFS_EXTENDED_PROFILE_MASK;
3486 if (bargs->target == chunk_type)
3492 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3493 struct extent_buffer *leaf,
3494 struct btrfs_chunk *chunk, u64 chunk_offset)
3496 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3497 struct btrfs_balance_args *bargs = NULL;
3498 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3501 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3502 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3506 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3507 bargs = &bctl->data;
3508 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3510 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3511 bargs = &bctl->meta;
3513 /* profiles filter */
3514 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3515 chunk_profiles_filter(chunk_type, bargs)) {
3520 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3521 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3523 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3524 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3529 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3530 chunk_devid_filter(leaf, chunk, bargs)) {
3534 /* drange filter, makes sense only with devid filter */
3535 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3536 chunk_drange_filter(leaf, chunk, bargs)) {
3541 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3542 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3546 /* stripes filter */
3547 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3548 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3552 /* soft profile changing mode */
3553 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3554 chunk_soft_convert_filter(chunk_type, bargs)) {
3559 * limited by count, must be the last filter
3561 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3562 if (bargs->limit == 0)
3566 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3568 * Same logic as the 'limit' filter; the minimum cannot be
3569 * determined here because we do not have the global information
3570 * about the count of all chunks that satisfy the filters.
3572 if (bargs->limit_max == 0)
3581 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3583 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3584 struct btrfs_root *chunk_root = fs_info->chunk_root;
3585 struct btrfs_root *dev_root = fs_info->dev_root;
3586 struct list_head *devices;
3587 struct btrfs_device *device;
3591 struct btrfs_chunk *chunk;
3592 struct btrfs_path *path = NULL;
3593 struct btrfs_key key;
3594 struct btrfs_key found_key;
3595 struct btrfs_trans_handle *trans;
3596 struct extent_buffer *leaf;
3599 int enospc_errors = 0;
3600 bool counting = true;
3601 /* The single value limit and min/max limits use the same bytes in the */
3602 u64 limit_data = bctl->data.limit;
3603 u64 limit_meta = bctl->meta.limit;
3604 u64 limit_sys = bctl->sys.limit;
3608 int chunk_reserved = 0;
3610 /* step one make some room on all the devices */
3611 devices = &fs_info->fs_devices->devices;
3612 list_for_each_entry(device, devices, dev_list) {
3613 old_size = btrfs_device_get_total_bytes(device);
3614 size_to_free = div_factor(old_size, 1);
3615 size_to_free = min_t(u64, size_to_free, SZ_1M);
3616 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3617 btrfs_device_get_total_bytes(device) -
3618 btrfs_device_get_bytes_used(device) > size_to_free ||
3619 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3622 ret = btrfs_shrink_device(device, old_size - size_to_free);
3626 /* btrfs_shrink_device never returns ret > 0 */
3631 trans = btrfs_start_transaction(dev_root, 0);
3632 if (IS_ERR(trans)) {
3633 ret = PTR_ERR(trans);
3634 btrfs_info_in_rcu(fs_info,
3635 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3636 rcu_str_deref(device->name), ret,
3637 old_size, old_size - size_to_free);
3641 ret = btrfs_grow_device(trans, device, old_size);
3643 btrfs_end_transaction(trans);
3644 /* btrfs_grow_device never returns ret > 0 */
3646 btrfs_info_in_rcu(fs_info,
3647 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3648 rcu_str_deref(device->name), ret,
3649 old_size, old_size - size_to_free);
3653 btrfs_end_transaction(trans);
3656 /* step two, relocate all the chunks */
3657 path = btrfs_alloc_path();
3663 /* zero out stat counters */
3664 spin_lock(&fs_info->balance_lock);
3665 memset(&bctl->stat, 0, sizeof(bctl->stat));
3666 spin_unlock(&fs_info->balance_lock);
3670 * The single value limit and min/max limits use the same bytes
3673 bctl->data.limit = limit_data;
3674 bctl->meta.limit = limit_meta;
3675 bctl->sys.limit = limit_sys;
3677 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3678 key.offset = (u64)-1;
3679 key.type = BTRFS_CHUNK_ITEM_KEY;
3682 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3683 atomic_read(&fs_info->balance_cancel_req)) {
3688 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3689 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3691 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3696 * this shouldn't happen, it means the last relocate
3700 BUG(); /* FIXME break ? */
3702 ret = btrfs_previous_item(chunk_root, path, 0,
3703 BTRFS_CHUNK_ITEM_KEY);
3705 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3710 leaf = path->nodes[0];
3711 slot = path->slots[0];
3712 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3714 if (found_key.objectid != key.objectid) {
3715 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3719 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3720 chunk_type = btrfs_chunk_type(leaf, chunk);
3723 spin_lock(&fs_info->balance_lock);
3724 bctl->stat.considered++;
3725 spin_unlock(&fs_info->balance_lock);
3728 ret = should_balance_chunk(fs_info, leaf, chunk,
3731 btrfs_release_path(path);
3733 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3738 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3739 spin_lock(&fs_info->balance_lock);
3740 bctl->stat.expected++;
3741 spin_unlock(&fs_info->balance_lock);
3743 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3745 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3747 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3754 * Apply limit_min filter, no need to check if the LIMITS
3755 * filter is used, limit_min is 0 by default
3757 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3758 count_data < bctl->data.limit_min)
3759 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3760 count_meta < bctl->meta.limit_min)
3761 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3762 count_sys < bctl->sys.limit_min)) {
3763 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3767 if (!chunk_reserved) {
3769 * We may be relocating the only data chunk we have,
3770 * which could potentially end up with losing data's
3771 * raid profile, so lets allocate an empty one in
3774 ret = btrfs_may_alloc_data_chunk(fs_info,
3777 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3779 } else if (ret == 1) {
3784 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3785 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3786 if (ret && ret != -ENOSPC)
3788 if (ret == -ENOSPC) {
3791 spin_lock(&fs_info->balance_lock);
3792 bctl->stat.completed++;
3793 spin_unlock(&fs_info->balance_lock);
3796 if (found_key.offset == 0)
3798 key.offset = found_key.offset - 1;
3802 btrfs_release_path(path);
3807 btrfs_free_path(path);
3808 if (enospc_errors) {
3809 btrfs_info(fs_info, "%d enospc errors during balance",
3819 * alloc_profile_is_valid - see if a given profile is valid and reduced
3820 * @flags: profile to validate
3821 * @extended: if true @flags is treated as an extended profile
3823 static int alloc_profile_is_valid(u64 flags, int extended)
3825 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3826 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3828 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3830 /* 1) check that all other bits are zeroed */
3834 /* 2) see if profile is reduced */
3836 return !extended; /* "0" is valid for usual profiles */
3838 /* true if exactly one bit set */
3839 return (flags & (flags - 1)) == 0;
3842 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3844 /* cancel requested || normal exit path */
3845 return atomic_read(&fs_info->balance_cancel_req) ||
3846 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3847 atomic_read(&fs_info->balance_cancel_req) == 0);
3850 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3854 unset_balance_control(fs_info);
3855 ret = del_balance_item(fs_info);
3857 btrfs_handle_fs_error(fs_info, ret, NULL);
3859 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3862 /* Non-zero return value signifies invalidity */
3863 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3866 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3867 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3868 (bctl_arg->target & ~allowed)));
3872 * Should be called with both balance and volume mutexes held
3874 int btrfs_balance(struct btrfs_balance_control *bctl,
3875 struct btrfs_ioctl_balance_args *bargs)
3877 struct btrfs_fs_info *fs_info = bctl->fs_info;
3878 u64 meta_target, data_target;
3885 if (btrfs_fs_closing(fs_info) ||
3886 atomic_read(&fs_info->balance_pause_req) ||
3887 atomic_read(&fs_info->balance_cancel_req)) {
3892 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3893 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3897 * In case of mixed groups both data and meta should be picked,
3898 * and identical options should be given for both of them.
3900 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3901 if (mixed && (bctl->flags & allowed)) {
3902 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3903 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3904 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3906 "with mixed groups data and metadata balance options must be the same");
3912 num_devices = fs_info->fs_devices->num_devices;
3913 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3914 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3915 BUG_ON(num_devices < 1);
3918 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3919 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3920 if (num_devices > 1)
3921 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3922 if (num_devices > 2)
3923 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3924 if (num_devices > 3)
3925 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3926 BTRFS_BLOCK_GROUP_RAID6);
3927 if (validate_convert_profile(&bctl->data, allowed)) {
3929 "unable to start balance with target data profile %llu",
3934 if (validate_convert_profile(&bctl->meta, allowed)) {
3936 "unable to start balance with target metadata profile %llu",
3941 if (validate_convert_profile(&bctl->sys, allowed)) {
3943 "unable to start balance with target system profile %llu",
3949 /* allow to reduce meta or sys integrity only if force set */
3950 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3951 BTRFS_BLOCK_GROUP_RAID10 |
3952 BTRFS_BLOCK_GROUP_RAID5 |
3953 BTRFS_BLOCK_GROUP_RAID6;
3955 seq = read_seqbegin(&fs_info->profiles_lock);
3957 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3958 (fs_info->avail_system_alloc_bits & allowed) &&
3959 !(bctl->sys.target & allowed)) ||
3960 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3961 (fs_info->avail_metadata_alloc_bits & allowed) &&
3962 !(bctl->meta.target & allowed))) {
3963 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3965 "force reducing metadata integrity");
3968 "balance will reduce metadata integrity, use force if you want this");
3973 } while (read_seqretry(&fs_info->profiles_lock, seq));
3975 /* if we're not converting, the target field is uninitialized */
3976 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3977 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3978 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3979 bctl->data.target : fs_info->avail_data_alloc_bits;
3980 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3981 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3983 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3984 meta_target, data_target);
3987 ret = insert_balance_item(fs_info, bctl);
3988 if (ret && ret != -EEXIST)
3991 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3992 BUG_ON(ret == -EEXIST);
3993 set_balance_control(bctl);
3995 BUG_ON(ret != -EEXIST);
3996 spin_lock(&fs_info->balance_lock);
3997 update_balance_args(bctl);
3998 spin_unlock(&fs_info->balance_lock);
4001 atomic_inc(&fs_info->balance_running);
4002 mutex_unlock(&fs_info->balance_mutex);
4004 ret = __btrfs_balance(fs_info);
4006 mutex_lock(&fs_info->balance_mutex);
4007 atomic_dec(&fs_info->balance_running);
4010 memset(bargs, 0, sizeof(*bargs));
4011 update_ioctl_balance_args(fs_info, 0, bargs);
4014 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4015 balance_need_close(fs_info)) {
4016 __cancel_balance(fs_info);
4019 wake_up(&fs_info->balance_wait_q);
4023 if (bctl->flags & BTRFS_BALANCE_RESUME)
4024 __cancel_balance(fs_info);
4027 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4032 static int balance_kthread(void *data)
4034 struct btrfs_fs_info *fs_info = data;
4037 mutex_lock(&fs_info->volume_mutex);
4038 mutex_lock(&fs_info->balance_mutex);
4040 if (fs_info->balance_ctl) {
4041 btrfs_info(fs_info, "continuing balance");
4042 ret = btrfs_balance(fs_info->balance_ctl, NULL);
4045 mutex_unlock(&fs_info->balance_mutex);
4046 mutex_unlock(&fs_info->volume_mutex);
4051 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4053 struct task_struct *tsk;
4055 spin_lock(&fs_info->balance_lock);
4056 if (!fs_info->balance_ctl) {
4057 spin_unlock(&fs_info->balance_lock);
4060 spin_unlock(&fs_info->balance_lock);
4062 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4063 btrfs_info(fs_info, "force skipping balance");
4067 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4068 return PTR_ERR_OR_ZERO(tsk);
4071 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4073 struct btrfs_balance_control *bctl;
4074 struct btrfs_balance_item *item;
4075 struct btrfs_disk_balance_args disk_bargs;
4076 struct btrfs_path *path;
4077 struct extent_buffer *leaf;
4078 struct btrfs_key key;
4081 path = btrfs_alloc_path();
4085 key.objectid = BTRFS_BALANCE_OBJECTID;
4086 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4089 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4092 if (ret > 0) { /* ret = -ENOENT; */
4097 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4103 leaf = path->nodes[0];
4104 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4106 bctl->fs_info = fs_info;
4107 bctl->flags = btrfs_balance_flags(leaf, item);
4108 bctl->flags |= BTRFS_BALANCE_RESUME;
4110 btrfs_balance_data(leaf, item, &disk_bargs);
4111 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4112 btrfs_balance_meta(leaf, item, &disk_bargs);
4113 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4114 btrfs_balance_sys(leaf, item, &disk_bargs);
4115 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4117 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4119 mutex_lock(&fs_info->volume_mutex);
4120 mutex_lock(&fs_info->balance_mutex);
4122 set_balance_control(bctl);
4124 mutex_unlock(&fs_info->balance_mutex);
4125 mutex_unlock(&fs_info->volume_mutex);
4127 btrfs_free_path(path);
4131 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4135 mutex_lock(&fs_info->balance_mutex);
4136 if (!fs_info->balance_ctl) {
4137 mutex_unlock(&fs_info->balance_mutex);
4141 if (atomic_read(&fs_info->balance_running)) {
4142 atomic_inc(&fs_info->balance_pause_req);
4143 mutex_unlock(&fs_info->balance_mutex);
4145 wait_event(fs_info->balance_wait_q,
4146 atomic_read(&fs_info->balance_running) == 0);
4148 mutex_lock(&fs_info->balance_mutex);
4149 /* we are good with balance_ctl ripped off from under us */
4150 BUG_ON(atomic_read(&fs_info->balance_running));
4151 atomic_dec(&fs_info->balance_pause_req);
4156 mutex_unlock(&fs_info->balance_mutex);
4160 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4162 if (sb_rdonly(fs_info->sb))
4165 mutex_lock(&fs_info->balance_mutex);
4166 if (!fs_info->balance_ctl) {
4167 mutex_unlock(&fs_info->balance_mutex);
4171 atomic_inc(&fs_info->balance_cancel_req);
4173 * if we are running just wait and return, balance item is
4174 * deleted in btrfs_balance in this case
4176 if (atomic_read(&fs_info->balance_running)) {
4177 mutex_unlock(&fs_info->balance_mutex);
4178 wait_event(fs_info->balance_wait_q,
4179 atomic_read(&fs_info->balance_running) == 0);
4180 mutex_lock(&fs_info->balance_mutex);
4182 /* __cancel_balance needs volume_mutex */
4183 mutex_unlock(&fs_info->balance_mutex);
4184 mutex_lock(&fs_info->volume_mutex);
4185 mutex_lock(&fs_info->balance_mutex);
4187 if (fs_info->balance_ctl)
4188 __cancel_balance(fs_info);
4190 mutex_unlock(&fs_info->volume_mutex);
4193 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4194 atomic_dec(&fs_info->balance_cancel_req);
4195 mutex_unlock(&fs_info->balance_mutex);
4199 static int btrfs_uuid_scan_kthread(void *data)
4201 struct btrfs_fs_info *fs_info = data;
4202 struct btrfs_root *root = fs_info->tree_root;
4203 struct btrfs_key key;
4204 struct btrfs_path *path = NULL;
4206 struct extent_buffer *eb;
4208 struct btrfs_root_item root_item;
4210 struct btrfs_trans_handle *trans = NULL;
4212 path = btrfs_alloc_path();
4219 key.type = BTRFS_ROOT_ITEM_KEY;
4223 ret = btrfs_search_forward(root, &key, path,
4224 BTRFS_OLDEST_GENERATION);
4231 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4232 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4233 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4234 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4237 eb = path->nodes[0];
4238 slot = path->slots[0];
4239 item_size = btrfs_item_size_nr(eb, slot);
4240 if (item_size < sizeof(root_item))
4243 read_extent_buffer(eb, &root_item,
4244 btrfs_item_ptr_offset(eb, slot),
4245 (int)sizeof(root_item));
4246 if (btrfs_root_refs(&root_item) == 0)
4249 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4250 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4254 btrfs_release_path(path);
4256 * 1 - subvol uuid item
4257 * 1 - received_subvol uuid item
4259 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4260 if (IS_ERR(trans)) {
4261 ret = PTR_ERR(trans);
4269 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4270 ret = btrfs_uuid_tree_add(trans, fs_info,
4272 BTRFS_UUID_KEY_SUBVOL,
4275 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4281 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4282 ret = btrfs_uuid_tree_add(trans, fs_info,
4283 root_item.received_uuid,
4284 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4287 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4295 ret = btrfs_end_transaction(trans);
4301 btrfs_release_path(path);
4302 if (key.offset < (u64)-1) {
4304 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4306 key.type = BTRFS_ROOT_ITEM_KEY;
4307 } else if (key.objectid < (u64)-1) {
4309 key.type = BTRFS_ROOT_ITEM_KEY;
4318 btrfs_free_path(path);
4319 if (trans && !IS_ERR(trans))
4320 btrfs_end_transaction(trans);
4322 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4324 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4325 up(&fs_info->uuid_tree_rescan_sem);
4330 * Callback for btrfs_uuid_tree_iterate().
4332 * 0 check succeeded, the entry is not outdated.
4333 * < 0 if an error occurred.
4334 * > 0 if the check failed, which means the caller shall remove the entry.
4336 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4337 u8 *uuid, u8 type, u64 subid)
4339 struct btrfs_key key;
4341 struct btrfs_root *subvol_root;
4343 if (type != BTRFS_UUID_KEY_SUBVOL &&
4344 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4347 key.objectid = subid;
4348 key.type = BTRFS_ROOT_ITEM_KEY;
4349 key.offset = (u64)-1;
4350 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4351 if (IS_ERR(subvol_root)) {
4352 ret = PTR_ERR(subvol_root);
4359 case BTRFS_UUID_KEY_SUBVOL:
4360 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4363 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4364 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4374 static int btrfs_uuid_rescan_kthread(void *data)
4376 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4380 * 1st step is to iterate through the existing UUID tree and
4381 * to delete all entries that contain outdated data.
4382 * 2nd step is to add all missing entries to the UUID tree.
4384 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4386 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4387 up(&fs_info->uuid_tree_rescan_sem);
4390 return btrfs_uuid_scan_kthread(data);
4393 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4395 struct btrfs_trans_handle *trans;
4396 struct btrfs_root *tree_root = fs_info->tree_root;
4397 struct btrfs_root *uuid_root;
4398 struct task_struct *task;
4405 trans = btrfs_start_transaction(tree_root, 2);
4407 return PTR_ERR(trans);
4409 uuid_root = btrfs_create_tree(trans, fs_info,
4410 BTRFS_UUID_TREE_OBJECTID);
4411 if (IS_ERR(uuid_root)) {
4412 ret = PTR_ERR(uuid_root);
4413 btrfs_abort_transaction(trans, ret);
4414 btrfs_end_transaction(trans);
4418 fs_info->uuid_root = uuid_root;
4420 ret = btrfs_commit_transaction(trans);
4424 down(&fs_info->uuid_tree_rescan_sem);
4425 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4427 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4428 btrfs_warn(fs_info, "failed to start uuid_scan task");
4429 up(&fs_info->uuid_tree_rescan_sem);
4430 return PTR_ERR(task);
4436 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4438 struct task_struct *task;
4440 down(&fs_info->uuid_tree_rescan_sem);
4441 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4443 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4444 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4445 up(&fs_info->uuid_tree_rescan_sem);
4446 return PTR_ERR(task);
4453 * shrinking a device means finding all of the device extents past
4454 * the new size, and then following the back refs to the chunks.
4455 * The chunk relocation code actually frees the device extent
4457 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4459 struct btrfs_fs_info *fs_info = device->fs_info;
4460 struct btrfs_root *root = fs_info->dev_root;
4461 struct btrfs_trans_handle *trans;
4462 struct btrfs_dev_extent *dev_extent = NULL;
4463 struct btrfs_path *path;
4469 bool retried = false;
4470 bool checked_pending_chunks = false;
4471 struct extent_buffer *l;
4472 struct btrfs_key key;
4473 struct btrfs_super_block *super_copy = fs_info->super_copy;
4474 u64 old_total = btrfs_super_total_bytes(super_copy);
4475 u64 old_size = btrfs_device_get_total_bytes(device);
4478 new_size = round_down(new_size, fs_info->sectorsize);
4479 diff = round_down(old_size - new_size, fs_info->sectorsize);
4481 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4484 path = btrfs_alloc_path();
4488 path->reada = READA_FORWARD;
4490 mutex_lock(&fs_info->chunk_mutex);
4492 btrfs_device_set_total_bytes(device, new_size);
4493 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4494 device->fs_devices->total_rw_bytes -= diff;
4495 atomic64_sub(diff, &fs_info->free_chunk_space);
4497 mutex_unlock(&fs_info->chunk_mutex);
4500 key.objectid = device->devid;
4501 key.offset = (u64)-1;
4502 key.type = BTRFS_DEV_EXTENT_KEY;
4505 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4508 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4512 ret = btrfs_previous_item(root, path, 0, key.type);
4514 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4519 btrfs_release_path(path);
4524 slot = path->slots[0];
4525 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4527 if (key.objectid != device->devid) {
4528 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4529 btrfs_release_path(path);
4533 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4534 length = btrfs_dev_extent_length(l, dev_extent);
4536 if (key.offset + length <= new_size) {
4537 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4538 btrfs_release_path(path);
4542 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4543 btrfs_release_path(path);
4546 * We may be relocating the only data chunk we have,
4547 * which could potentially end up with losing data's
4548 * raid profile, so lets allocate an empty one in
4551 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4553 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4557 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4558 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4559 if (ret && ret != -ENOSPC)
4563 } while (key.offset-- > 0);
4565 if (failed && !retried) {
4569 } else if (failed && retried) {
4574 /* Shrinking succeeded, else we would be at "done". */
4575 trans = btrfs_start_transaction(root, 0);
4576 if (IS_ERR(trans)) {
4577 ret = PTR_ERR(trans);
4581 mutex_lock(&fs_info->chunk_mutex);
4584 * We checked in the above loop all device extents that were already in
4585 * the device tree. However before we have updated the device's
4586 * total_bytes to the new size, we might have had chunk allocations that
4587 * have not complete yet (new block groups attached to transaction
4588 * handles), and therefore their device extents were not yet in the
4589 * device tree and we missed them in the loop above. So if we have any
4590 * pending chunk using a device extent that overlaps the device range
4591 * that we can not use anymore, commit the current transaction and
4592 * repeat the search on the device tree - this way we guarantee we will
4593 * not have chunks using device extents that end beyond 'new_size'.
4595 if (!checked_pending_chunks) {
4596 u64 start = new_size;
4597 u64 len = old_size - new_size;
4599 if (contains_pending_extent(trans->transaction, device,
4601 mutex_unlock(&fs_info->chunk_mutex);
4602 checked_pending_chunks = true;
4605 ret = btrfs_commit_transaction(trans);
4612 btrfs_device_set_disk_total_bytes(device, new_size);
4613 if (list_empty(&device->resized_list))
4614 list_add_tail(&device->resized_list,
4615 &fs_info->fs_devices->resized_devices);
4617 WARN_ON(diff > old_total);
4618 btrfs_set_super_total_bytes(super_copy,
4619 round_down(old_total - diff, fs_info->sectorsize));
4620 mutex_unlock(&fs_info->chunk_mutex);
4622 /* Now btrfs_update_device() will change the on-disk size. */
4623 ret = btrfs_update_device(trans, device);
4624 btrfs_end_transaction(trans);
4626 btrfs_free_path(path);
4628 mutex_lock(&fs_info->chunk_mutex);
4629 btrfs_device_set_total_bytes(device, old_size);
4630 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4631 device->fs_devices->total_rw_bytes += diff;
4632 atomic64_add(diff, &fs_info->free_chunk_space);
4633 mutex_unlock(&fs_info->chunk_mutex);
4638 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4639 struct btrfs_key *key,
4640 struct btrfs_chunk *chunk, int item_size)
4642 struct btrfs_super_block *super_copy = fs_info->super_copy;
4643 struct btrfs_disk_key disk_key;
4647 mutex_lock(&fs_info->chunk_mutex);
4648 array_size = btrfs_super_sys_array_size(super_copy);
4649 if (array_size + item_size + sizeof(disk_key)
4650 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4651 mutex_unlock(&fs_info->chunk_mutex);
4655 ptr = super_copy->sys_chunk_array + array_size;
4656 btrfs_cpu_key_to_disk(&disk_key, key);
4657 memcpy(ptr, &disk_key, sizeof(disk_key));
4658 ptr += sizeof(disk_key);
4659 memcpy(ptr, chunk, item_size);
4660 item_size += sizeof(disk_key);
4661 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4662 mutex_unlock(&fs_info->chunk_mutex);
4668 * sort the devices in descending order by max_avail, total_avail
4670 static int btrfs_cmp_device_info(const void *a, const void *b)
4672 const struct btrfs_device_info *di_a = a;
4673 const struct btrfs_device_info *di_b = b;
4675 if (di_a->max_avail > di_b->max_avail)
4677 if (di_a->max_avail < di_b->max_avail)
4679 if (di_a->total_avail > di_b->total_avail)
4681 if (di_a->total_avail < di_b->total_avail)
4686 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4688 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4691 btrfs_set_fs_incompat(info, RAID56);
4694 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4695 - sizeof(struct btrfs_chunk)) \
4696 / sizeof(struct btrfs_stripe) + 1)
4698 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4699 - 2 * sizeof(struct btrfs_disk_key) \
4700 - 2 * sizeof(struct btrfs_chunk)) \
4701 / sizeof(struct btrfs_stripe) + 1)
4703 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4704 u64 start, u64 type)
4706 struct btrfs_fs_info *info = trans->fs_info;
4707 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4708 struct btrfs_device *device;
4709 struct map_lookup *map = NULL;
4710 struct extent_map_tree *em_tree;
4711 struct extent_map *em;
4712 struct btrfs_device_info *devices_info = NULL;
4714 int num_stripes; /* total number of stripes to allocate */
4715 int data_stripes; /* number of stripes that count for
4717 int sub_stripes; /* sub_stripes info for map */
4718 int dev_stripes; /* stripes per dev */
4719 int devs_max; /* max devs to use */
4720 int devs_min; /* min devs needed */
4721 int devs_increment; /* ndevs has to be a multiple of this */
4722 int ncopies; /* how many copies to data has */
4724 u64 max_stripe_size;
4733 BUG_ON(!alloc_profile_is_valid(type, 0));
4735 if (list_empty(&fs_devices->alloc_list)) {
4736 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4737 btrfs_debug(info, "%s: no writable device", __func__);
4741 index = btrfs_bg_flags_to_raid_index(type);
4743 sub_stripes = btrfs_raid_array[index].sub_stripes;
4744 dev_stripes = btrfs_raid_array[index].dev_stripes;
4745 devs_max = btrfs_raid_array[index].devs_max;
4746 devs_min = btrfs_raid_array[index].devs_min;
4747 devs_increment = btrfs_raid_array[index].devs_increment;
4748 ncopies = btrfs_raid_array[index].ncopies;
4750 if (type & BTRFS_BLOCK_GROUP_DATA) {
4751 max_stripe_size = SZ_1G;
4752 max_chunk_size = 10 * max_stripe_size;
4754 devs_max = BTRFS_MAX_DEVS(info);
4755 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4756 /* for larger filesystems, use larger metadata chunks */
4757 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4758 max_stripe_size = SZ_1G;
4760 max_stripe_size = SZ_256M;
4761 max_chunk_size = max_stripe_size;
4763 devs_max = BTRFS_MAX_DEVS(info);
4764 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4765 max_stripe_size = SZ_32M;
4766 max_chunk_size = 2 * max_stripe_size;
4768 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4770 btrfs_err(info, "invalid chunk type 0x%llx requested",
4775 /* we don't want a chunk larger than 10% of writeable space */
4776 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4779 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4785 * in the first pass through the devices list, we gather information
4786 * about the available holes on each device.
4789 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4793 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4795 "BTRFS: read-only device in alloc_list\n");
4799 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4800 &device->dev_state) ||
4801 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4804 if (device->total_bytes > device->bytes_used)
4805 total_avail = device->total_bytes - device->bytes_used;
4809 /* If there is no space on this device, skip it. */
4810 if (total_avail == 0)
4813 ret = find_free_dev_extent(trans, device,
4814 max_stripe_size * dev_stripes,
4815 &dev_offset, &max_avail);
4816 if (ret && ret != -ENOSPC)
4820 max_avail = max_stripe_size * dev_stripes;
4822 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4823 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4825 "%s: devid %llu has no free space, have=%llu want=%u",
4826 __func__, device->devid, max_avail,
4827 BTRFS_STRIPE_LEN * dev_stripes);
4831 if (ndevs == fs_devices->rw_devices) {
4832 WARN(1, "%s: found more than %llu devices\n",
4833 __func__, fs_devices->rw_devices);
4836 devices_info[ndevs].dev_offset = dev_offset;
4837 devices_info[ndevs].max_avail = max_avail;
4838 devices_info[ndevs].total_avail = total_avail;
4839 devices_info[ndevs].dev = device;
4844 * now sort the devices by hole size / available space
4846 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4847 btrfs_cmp_device_info, NULL);
4849 /* round down to number of usable stripes */
4850 ndevs = round_down(ndevs, devs_increment);
4852 if (ndevs < devs_min) {
4854 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4856 "%s: not enough devices with free space: have=%d minimum required=%d",
4857 __func__, ndevs, devs_min);
4862 ndevs = min(ndevs, devs_max);
4865 * The primary goal is to maximize the number of stripes, so use as
4866 * many devices as possible, even if the stripes are not maximum sized.
4868 * The DUP profile stores more than one stripe per device, the
4869 * max_avail is the total size so we have to adjust.
4871 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4872 num_stripes = ndevs * dev_stripes;
4875 * this will have to be fixed for RAID1 and RAID10 over
4878 data_stripes = num_stripes / ncopies;
4880 if (type & BTRFS_BLOCK_GROUP_RAID5)
4881 data_stripes = num_stripes - 1;
4883 if (type & BTRFS_BLOCK_GROUP_RAID6)
4884 data_stripes = num_stripes - 2;
4887 * Use the number of data stripes to figure out how big this chunk
4888 * is really going to be in terms of logical address space,
4889 * and compare that answer with the max chunk size
4891 if (stripe_size * data_stripes > max_chunk_size) {
4892 stripe_size = div_u64(max_chunk_size, data_stripes);
4894 /* bump the answer up to a 16MB boundary */
4895 stripe_size = round_up(stripe_size, SZ_16M);
4898 * But don't go higher than the limits we found while searching
4901 stripe_size = min(devices_info[ndevs - 1].max_avail,
4905 /* align to BTRFS_STRIPE_LEN */
4906 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4908 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4913 map->num_stripes = num_stripes;
4915 for (i = 0; i < ndevs; ++i) {
4916 for (j = 0; j < dev_stripes; ++j) {
4917 int s = i * dev_stripes + j;
4918 map->stripes[s].dev = devices_info[i].dev;
4919 map->stripes[s].physical = devices_info[i].dev_offset +
4923 map->stripe_len = BTRFS_STRIPE_LEN;
4924 map->io_align = BTRFS_STRIPE_LEN;
4925 map->io_width = BTRFS_STRIPE_LEN;
4927 map->sub_stripes = sub_stripes;
4929 num_bytes = stripe_size * data_stripes;
4931 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4933 em = alloc_extent_map();
4939 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4940 em->map_lookup = map;
4942 em->len = num_bytes;
4943 em->block_start = 0;
4944 em->block_len = em->len;
4945 em->orig_block_len = stripe_size;
4947 em_tree = &info->mapping_tree.map_tree;
4948 write_lock(&em_tree->lock);
4949 ret = add_extent_mapping(em_tree, em, 0);
4951 write_unlock(&em_tree->lock);
4952 free_extent_map(em);
4956 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4957 refcount_inc(&em->refs);
4958 write_unlock(&em_tree->lock);
4960 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4962 goto error_del_extent;
4964 for (i = 0; i < map->num_stripes; i++) {
4965 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4966 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4969 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4971 free_extent_map(em);
4972 check_raid56_incompat_flag(info, type);
4974 kfree(devices_info);
4978 write_lock(&em_tree->lock);
4979 remove_extent_mapping(em_tree, em);
4980 write_unlock(&em_tree->lock);
4982 /* One for our allocation */
4983 free_extent_map(em);
4984 /* One for the tree reference */
4985 free_extent_map(em);
4986 /* One for the pending_chunks list reference */
4987 free_extent_map(em);
4989 kfree(devices_info);
4993 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4994 struct btrfs_fs_info *fs_info,
4995 u64 chunk_offset, u64 chunk_size)
4997 struct btrfs_root *extent_root = fs_info->extent_root;
4998 struct btrfs_root *chunk_root = fs_info->chunk_root;
4999 struct btrfs_key key;
5000 struct btrfs_device *device;
5001 struct btrfs_chunk *chunk;
5002 struct btrfs_stripe *stripe;
5003 struct extent_map *em;
5004 struct map_lookup *map;
5011 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
5015 map = em->map_lookup;
5016 item_size = btrfs_chunk_item_size(map->num_stripes);
5017 stripe_size = em->orig_block_len;
5019 chunk = kzalloc(item_size, GFP_NOFS);
5026 * Take the device list mutex to prevent races with the final phase of
5027 * a device replace operation that replaces the device object associated
5028 * with the map's stripes, because the device object's id can change
5029 * at any time during that final phase of the device replace operation
5030 * (dev-replace.c:btrfs_dev_replace_finishing()).
5032 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5033 for (i = 0; i < map->num_stripes; i++) {
5034 device = map->stripes[i].dev;
5035 dev_offset = map->stripes[i].physical;
5037 ret = btrfs_update_device(trans, device);
5040 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5041 dev_offset, stripe_size);
5046 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5050 stripe = &chunk->stripe;
5051 for (i = 0; i < map->num_stripes; i++) {
5052 device = map->stripes[i].dev;
5053 dev_offset = map->stripes[i].physical;
5055 btrfs_set_stack_stripe_devid(stripe, device->devid);
5056 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5057 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5060 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5062 btrfs_set_stack_chunk_length(chunk, chunk_size);
5063 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5064 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5065 btrfs_set_stack_chunk_type(chunk, map->type);
5066 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5067 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5068 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5069 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5070 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5072 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5073 key.type = BTRFS_CHUNK_ITEM_KEY;
5074 key.offset = chunk_offset;
5076 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5077 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5079 * TODO: Cleanup of inserted chunk root in case of
5082 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5087 free_extent_map(em);
5092 * Chunk allocation falls into two parts. The first part does works
5093 * that make the new allocated chunk useable, but not do any operation
5094 * that modifies the chunk tree. The second part does the works that
5095 * require modifying the chunk tree. This division is important for the
5096 * bootstrap process of adding storage to a seed btrfs.
5098 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5099 struct btrfs_fs_info *fs_info, u64 type)
5103 lockdep_assert_held(&fs_info->chunk_mutex);
5104 chunk_offset = find_next_chunk(fs_info);
5105 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5108 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5109 struct btrfs_fs_info *fs_info)
5112 u64 sys_chunk_offset;
5116 chunk_offset = find_next_chunk(fs_info);
5117 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5118 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5122 sys_chunk_offset = find_next_chunk(fs_info);
5123 alloc_profile = btrfs_system_alloc_profile(fs_info);
5124 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5128 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5132 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5133 BTRFS_BLOCK_GROUP_RAID10 |
5134 BTRFS_BLOCK_GROUP_RAID5 |
5135 BTRFS_BLOCK_GROUP_DUP)) {
5137 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5146 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5148 struct extent_map *em;
5149 struct map_lookup *map;
5154 em = get_chunk_map(fs_info, chunk_offset, 1);
5158 map = em->map_lookup;
5159 for (i = 0; i < map->num_stripes; i++) {
5160 if (test_bit(BTRFS_DEV_STATE_MISSING,
5161 &map->stripes[i].dev->dev_state)) {
5165 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5166 &map->stripes[i].dev->dev_state)) {
5173 * If the number of missing devices is larger than max errors,
5174 * we can not write the data into that chunk successfully, so
5177 if (miss_ndevs > btrfs_chunk_max_errors(map))
5180 free_extent_map(em);
5184 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5186 extent_map_tree_init(&tree->map_tree);
5189 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5191 struct extent_map *em;
5194 write_lock(&tree->map_tree.lock);
5195 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5197 remove_extent_mapping(&tree->map_tree, em);
5198 write_unlock(&tree->map_tree.lock);
5202 free_extent_map(em);
5203 /* once for the tree */
5204 free_extent_map(em);
5208 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5210 struct extent_map *em;
5211 struct map_lookup *map;
5214 em = get_chunk_map(fs_info, logical, len);
5217 * We could return errors for these cases, but that could get
5218 * ugly and we'd probably do the same thing which is just not do
5219 * anything else and exit, so return 1 so the callers don't try
5220 * to use other copies.
5224 map = em->map_lookup;
5225 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5226 ret = map->num_stripes;
5227 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5228 ret = map->sub_stripes;
5229 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5231 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5233 * There could be two corrupted data stripes, we need
5234 * to loop retry in order to rebuild the correct data.
5236 * Fail a stripe at a time on every retry except the
5237 * stripe under reconstruction.
5239 ret = map->num_stripes;
5242 free_extent_map(em);
5244 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5245 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5246 fs_info->dev_replace.tgtdev)
5248 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5253 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5256 struct extent_map *em;
5257 struct map_lookup *map;
5258 unsigned long len = fs_info->sectorsize;
5260 em = get_chunk_map(fs_info, logical, len);
5262 if (!WARN_ON(IS_ERR(em))) {
5263 map = em->map_lookup;
5264 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5265 len = map->stripe_len * nr_data_stripes(map);
5266 free_extent_map(em);
5271 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5273 struct extent_map *em;
5274 struct map_lookup *map;
5277 em = get_chunk_map(fs_info, logical, len);
5279 if(!WARN_ON(IS_ERR(em))) {
5280 map = em->map_lookup;
5281 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5283 free_extent_map(em);
5288 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5289 struct map_lookup *map, int first,
5290 int dev_replace_is_ongoing)
5294 int preferred_mirror;
5296 struct btrfs_device *srcdev;
5299 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5301 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5302 num_stripes = map->sub_stripes;
5304 num_stripes = map->num_stripes;
5306 preferred_mirror = first + current->pid % num_stripes;
5308 if (dev_replace_is_ongoing &&
5309 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5310 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5311 srcdev = fs_info->dev_replace.srcdev;
5316 * try to avoid the drive that is the source drive for a
5317 * dev-replace procedure, only choose it if no other non-missing
5318 * mirror is available
5320 for (tolerance = 0; tolerance < 2; tolerance++) {
5321 if (map->stripes[preferred_mirror].dev->bdev &&
5322 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5323 return preferred_mirror;
5324 for (i = first; i < first + num_stripes; i++) {
5325 if (map->stripes[i].dev->bdev &&
5326 (tolerance || map->stripes[i].dev != srcdev))
5331 /* we couldn't find one that doesn't fail. Just return something
5332 * and the io error handling code will clean up eventually
5334 return preferred_mirror;
5337 static inline int parity_smaller(u64 a, u64 b)
5342 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5343 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5345 struct btrfs_bio_stripe s;
5352 for (i = 0; i < num_stripes - 1; i++) {
5353 if (parity_smaller(bbio->raid_map[i],
5354 bbio->raid_map[i+1])) {
5355 s = bbio->stripes[i];
5356 l = bbio->raid_map[i];
5357 bbio->stripes[i] = bbio->stripes[i+1];
5358 bbio->raid_map[i] = bbio->raid_map[i+1];
5359 bbio->stripes[i+1] = s;
5360 bbio->raid_map[i+1] = l;
5368 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5370 struct btrfs_bio *bbio = kzalloc(
5371 /* the size of the btrfs_bio */
5372 sizeof(struct btrfs_bio) +
5373 /* plus the variable array for the stripes */
5374 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5375 /* plus the variable array for the tgt dev */
5376 sizeof(int) * (real_stripes) +
5378 * plus the raid_map, which includes both the tgt dev
5381 sizeof(u64) * (total_stripes),
5382 GFP_NOFS|__GFP_NOFAIL);
5384 atomic_set(&bbio->error, 0);
5385 refcount_set(&bbio->refs, 1);
5390 void btrfs_get_bbio(struct btrfs_bio *bbio)
5392 WARN_ON(!refcount_read(&bbio->refs));
5393 refcount_inc(&bbio->refs);
5396 void btrfs_put_bbio(struct btrfs_bio *bbio)
5400 if (refcount_dec_and_test(&bbio->refs))
5404 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5406 * Please note that, discard won't be sent to target device of device
5409 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5410 u64 logical, u64 length,
5411 struct btrfs_bio **bbio_ret)
5413 struct extent_map *em;
5414 struct map_lookup *map;
5415 struct btrfs_bio *bbio;
5419 u64 stripe_end_offset;
5426 u32 sub_stripes = 0;
5427 u64 stripes_per_dev = 0;
5428 u32 remaining_stripes = 0;
5429 u32 last_stripe = 0;
5433 /* discard always return a bbio */
5436 em = get_chunk_map(fs_info, logical, length);
5440 map = em->map_lookup;
5441 /* we don't discard raid56 yet */
5442 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5447 offset = logical - em->start;
5448 length = min_t(u64, em->len - offset, length);
5450 stripe_len = map->stripe_len;
5452 * stripe_nr counts the total number of stripes we have to stride
5453 * to get to this block
5455 stripe_nr = div64_u64(offset, stripe_len);
5457 /* stripe_offset is the offset of this block in its stripe */
5458 stripe_offset = offset - stripe_nr * stripe_len;
5460 stripe_nr_end = round_up(offset + length, map->stripe_len);
5461 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5462 stripe_cnt = stripe_nr_end - stripe_nr;
5463 stripe_end_offset = stripe_nr_end * map->stripe_len -
5466 * after this, stripe_nr is the number of stripes on this
5467 * device we have to walk to find the data, and stripe_index is
5468 * the number of our device in the stripe array
5472 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5473 BTRFS_BLOCK_GROUP_RAID10)) {
5474 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5477 sub_stripes = map->sub_stripes;
5479 factor = map->num_stripes / sub_stripes;
5480 num_stripes = min_t(u64, map->num_stripes,
5481 sub_stripes * stripe_cnt);
5482 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5483 stripe_index *= sub_stripes;
5484 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5485 &remaining_stripes);
5486 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5487 last_stripe *= sub_stripes;
5488 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5489 BTRFS_BLOCK_GROUP_DUP)) {
5490 num_stripes = map->num_stripes;
5492 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5496 bbio = alloc_btrfs_bio(num_stripes, 0);
5502 for (i = 0; i < num_stripes; i++) {
5503 bbio->stripes[i].physical =
5504 map->stripes[stripe_index].physical +
5505 stripe_offset + stripe_nr * map->stripe_len;
5506 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5508 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5509 BTRFS_BLOCK_GROUP_RAID10)) {
5510 bbio->stripes[i].length = stripes_per_dev *
5513 if (i / sub_stripes < remaining_stripes)
5514 bbio->stripes[i].length +=
5518 * Special for the first stripe and
5521 * |-------|...|-------|
5525 if (i < sub_stripes)
5526 bbio->stripes[i].length -=
5529 if (stripe_index >= last_stripe &&
5530 stripe_index <= (last_stripe +
5532 bbio->stripes[i].length -=
5535 if (i == sub_stripes - 1)
5538 bbio->stripes[i].length = length;
5542 if (stripe_index == map->num_stripes) {
5549 bbio->map_type = map->type;
5550 bbio->num_stripes = num_stripes;
5552 free_extent_map(em);
5557 * In dev-replace case, for repair case (that's the only case where the mirror
5558 * is selected explicitly when calling btrfs_map_block), blocks left of the
5559 * left cursor can also be read from the target drive.
5561 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5563 * For READ, it also needs to be supported using the same mirror number.
5565 * If the requested block is not left of the left cursor, EIO is returned. This
5566 * can happen because btrfs_num_copies() returns one more in the dev-replace
5569 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5570 u64 logical, u64 length,
5571 u64 srcdev_devid, int *mirror_num,
5574 struct btrfs_bio *bbio = NULL;
5576 int index_srcdev = 0;
5578 u64 physical_of_found = 0;
5582 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5583 logical, &length, &bbio, 0, 0);
5585 ASSERT(bbio == NULL);
5589 num_stripes = bbio->num_stripes;
5590 if (*mirror_num > num_stripes) {
5592 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5593 * that means that the requested area is not left of the left
5596 btrfs_put_bbio(bbio);
5601 * process the rest of the function using the mirror_num of the source
5602 * drive. Therefore look it up first. At the end, patch the device
5603 * pointer to the one of the target drive.
5605 for (i = 0; i < num_stripes; i++) {
5606 if (bbio->stripes[i].dev->devid != srcdev_devid)
5610 * In case of DUP, in order to keep it simple, only add the
5611 * mirror with the lowest physical address
5614 physical_of_found <= bbio->stripes[i].physical)
5619 physical_of_found = bbio->stripes[i].physical;
5622 btrfs_put_bbio(bbio);
5628 *mirror_num = index_srcdev + 1;
5629 *physical = physical_of_found;
5633 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5634 struct btrfs_bio **bbio_ret,
5635 struct btrfs_dev_replace *dev_replace,
5636 int *num_stripes_ret, int *max_errors_ret)
5638 struct btrfs_bio *bbio = *bbio_ret;
5639 u64 srcdev_devid = dev_replace->srcdev->devid;
5640 int tgtdev_indexes = 0;
5641 int num_stripes = *num_stripes_ret;
5642 int max_errors = *max_errors_ret;
5645 if (op == BTRFS_MAP_WRITE) {
5646 int index_where_to_add;
5649 * duplicate the write operations while the dev replace
5650 * procedure is running. Since the copying of the old disk to
5651 * the new disk takes place at run time while the filesystem is
5652 * mounted writable, the regular write operations to the old
5653 * disk have to be duplicated to go to the new disk as well.
5655 * Note that device->missing is handled by the caller, and that
5656 * the write to the old disk is already set up in the stripes
5659 index_where_to_add = num_stripes;
5660 for (i = 0; i < num_stripes; i++) {
5661 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5662 /* write to new disk, too */
5663 struct btrfs_bio_stripe *new =
5664 bbio->stripes + index_where_to_add;
5665 struct btrfs_bio_stripe *old =
5668 new->physical = old->physical;
5669 new->length = old->length;
5670 new->dev = dev_replace->tgtdev;
5671 bbio->tgtdev_map[i] = index_where_to_add;
5672 index_where_to_add++;
5677 num_stripes = index_where_to_add;
5678 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5679 int index_srcdev = 0;
5681 u64 physical_of_found = 0;
5684 * During the dev-replace procedure, the target drive can also
5685 * be used to read data in case it is needed to repair a corrupt
5686 * block elsewhere. This is possible if the requested area is
5687 * left of the left cursor. In this area, the target drive is a
5688 * full copy of the source drive.
5690 for (i = 0; i < num_stripes; i++) {
5691 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5693 * In case of DUP, in order to keep it simple,
5694 * only add the mirror with the lowest physical
5698 physical_of_found <=
5699 bbio->stripes[i].physical)
5703 physical_of_found = bbio->stripes[i].physical;
5707 struct btrfs_bio_stripe *tgtdev_stripe =
5708 bbio->stripes + num_stripes;
5710 tgtdev_stripe->physical = physical_of_found;
5711 tgtdev_stripe->length =
5712 bbio->stripes[index_srcdev].length;
5713 tgtdev_stripe->dev = dev_replace->tgtdev;
5714 bbio->tgtdev_map[index_srcdev] = num_stripes;
5721 *num_stripes_ret = num_stripes;
5722 *max_errors_ret = max_errors;
5723 bbio->num_tgtdevs = tgtdev_indexes;
5727 static bool need_full_stripe(enum btrfs_map_op op)
5729 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5732 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5733 enum btrfs_map_op op,
5734 u64 logical, u64 *length,
5735 struct btrfs_bio **bbio_ret,
5736 int mirror_num, int need_raid_map)
5738 struct extent_map *em;
5739 struct map_lookup *map;
5749 int tgtdev_indexes = 0;
5750 struct btrfs_bio *bbio = NULL;
5751 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5752 int dev_replace_is_ongoing = 0;
5753 int num_alloc_stripes;
5754 int patch_the_first_stripe_for_dev_replace = 0;
5755 u64 physical_to_patch_in_first_stripe = 0;
5756 u64 raid56_full_stripe_start = (u64)-1;
5758 if (op == BTRFS_MAP_DISCARD)
5759 return __btrfs_map_block_for_discard(fs_info, logical,
5762 em = get_chunk_map(fs_info, logical, *length);
5766 map = em->map_lookup;
5767 offset = logical - em->start;
5769 stripe_len = map->stripe_len;
5772 * stripe_nr counts the total number of stripes we have to stride
5773 * to get to this block
5775 stripe_nr = div64_u64(stripe_nr, stripe_len);
5777 stripe_offset = stripe_nr * stripe_len;
5778 if (offset < stripe_offset) {
5780 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5781 stripe_offset, offset, em->start, logical,
5783 free_extent_map(em);
5787 /* stripe_offset is the offset of this block in its stripe*/
5788 stripe_offset = offset - stripe_offset;
5790 /* if we're here for raid56, we need to know the stripe aligned start */
5791 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5792 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5793 raid56_full_stripe_start = offset;
5795 /* allow a write of a full stripe, but make sure we don't
5796 * allow straddling of stripes
5798 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5800 raid56_full_stripe_start *= full_stripe_len;
5803 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5805 /* For writes to RAID[56], allow a full stripeset across all disks.
5806 For other RAID types and for RAID[56] reads, just allow a single
5807 stripe (on a single disk). */
5808 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5809 (op == BTRFS_MAP_WRITE)) {
5810 max_len = stripe_len * nr_data_stripes(map) -
5811 (offset - raid56_full_stripe_start);
5813 /* we limit the length of each bio to what fits in a stripe */
5814 max_len = stripe_len - stripe_offset;
5816 *length = min_t(u64, em->len - offset, max_len);
5818 *length = em->len - offset;
5821 /* This is for when we're called from btrfs_merge_bio_hook() and all
5822 it cares about is the length */
5826 btrfs_dev_replace_read_lock(dev_replace);
5827 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5828 if (!dev_replace_is_ongoing)
5829 btrfs_dev_replace_read_unlock(dev_replace);
5831 btrfs_dev_replace_set_lock_blocking(dev_replace);
5833 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5834 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5835 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5836 dev_replace->srcdev->devid,
5838 &physical_to_patch_in_first_stripe);
5842 patch_the_first_stripe_for_dev_replace = 1;
5843 } else if (mirror_num > map->num_stripes) {
5849 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5850 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5852 if (!need_full_stripe(op))
5854 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5855 if (need_full_stripe(op))
5856 num_stripes = map->num_stripes;
5857 else if (mirror_num)
5858 stripe_index = mirror_num - 1;
5860 stripe_index = find_live_mirror(fs_info, map, 0,
5861 dev_replace_is_ongoing);
5862 mirror_num = stripe_index + 1;
5865 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5866 if (need_full_stripe(op)) {
5867 num_stripes = map->num_stripes;
5868 } else if (mirror_num) {
5869 stripe_index = mirror_num - 1;
5874 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5875 u32 factor = map->num_stripes / map->sub_stripes;
5877 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5878 stripe_index *= map->sub_stripes;
5880 if (need_full_stripe(op))
5881 num_stripes = map->sub_stripes;
5882 else if (mirror_num)
5883 stripe_index += mirror_num - 1;
5885 int old_stripe_index = stripe_index;
5886 stripe_index = find_live_mirror(fs_info, map,
5888 dev_replace_is_ongoing);
5889 mirror_num = stripe_index - old_stripe_index + 1;
5892 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5893 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5894 /* push stripe_nr back to the start of the full stripe */
5895 stripe_nr = div64_u64(raid56_full_stripe_start,
5896 stripe_len * nr_data_stripes(map));
5898 /* RAID[56] write or recovery. Return all stripes */
5899 num_stripes = map->num_stripes;
5900 max_errors = nr_parity_stripes(map);
5902 *length = map->stripe_len;
5907 * Mirror #0 or #1 means the original data block.
5908 * Mirror #2 is RAID5 parity block.
5909 * Mirror #3 is RAID6 Q block.
5911 stripe_nr = div_u64_rem(stripe_nr,
5912 nr_data_stripes(map), &stripe_index);
5914 stripe_index = nr_data_stripes(map) +
5917 /* We distribute the parity blocks across stripes */
5918 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5920 if (!need_full_stripe(op) && mirror_num <= 1)
5925 * after this, stripe_nr is the number of stripes on this
5926 * device we have to walk to find the data, and stripe_index is
5927 * the number of our device in the stripe array
5929 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5931 mirror_num = stripe_index + 1;
5933 if (stripe_index >= map->num_stripes) {
5935 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5936 stripe_index, map->num_stripes);
5941 num_alloc_stripes = num_stripes;
5942 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5943 if (op == BTRFS_MAP_WRITE)
5944 num_alloc_stripes <<= 1;
5945 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5946 num_alloc_stripes++;
5947 tgtdev_indexes = num_stripes;
5950 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5955 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5956 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5958 /* build raid_map */
5959 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5960 (need_full_stripe(op) || mirror_num > 1)) {
5964 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5965 sizeof(struct btrfs_bio_stripe) *
5967 sizeof(int) * tgtdev_indexes);
5969 /* Work out the disk rotation on this stripe-set */
5970 div_u64_rem(stripe_nr, num_stripes, &rot);
5972 /* Fill in the logical address of each stripe */
5973 tmp = stripe_nr * nr_data_stripes(map);
5974 for (i = 0; i < nr_data_stripes(map); i++)
5975 bbio->raid_map[(i+rot) % num_stripes] =
5976 em->start + (tmp + i) * map->stripe_len;
5978 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5979 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5980 bbio->raid_map[(i+rot+1) % num_stripes] =
5985 for (i = 0; i < num_stripes; i++) {
5986 bbio->stripes[i].physical =
5987 map->stripes[stripe_index].physical +
5989 stripe_nr * map->stripe_len;
5990 bbio->stripes[i].dev =
5991 map->stripes[stripe_index].dev;
5995 if (need_full_stripe(op))
5996 max_errors = btrfs_chunk_max_errors(map);
5999 sort_parity_stripes(bbio, num_stripes);
6001 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6002 need_full_stripe(op)) {
6003 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6008 bbio->map_type = map->type;
6009 bbio->num_stripes = num_stripes;
6010 bbio->max_errors = max_errors;
6011 bbio->mirror_num = mirror_num;
6014 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6015 * mirror_num == num_stripes + 1 && dev_replace target drive is
6016 * available as a mirror
6018 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6019 WARN_ON(num_stripes > 1);
6020 bbio->stripes[0].dev = dev_replace->tgtdev;
6021 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6022 bbio->mirror_num = map->num_stripes + 1;
6025 if (dev_replace_is_ongoing) {
6026 btrfs_dev_replace_clear_lock_blocking(dev_replace);
6027 btrfs_dev_replace_read_unlock(dev_replace);
6029 free_extent_map(em);
6033 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6034 u64 logical, u64 *length,
6035 struct btrfs_bio **bbio_ret, int mirror_num)
6037 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6041 /* For Scrub/replace */
6042 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6043 u64 logical, u64 *length,
6044 struct btrfs_bio **bbio_ret)
6046 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6049 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
6050 u64 chunk_start, u64 physical, u64 devid,
6051 u64 **logical, int *naddrs, int *stripe_len)
6053 struct extent_map *em;
6054 struct map_lookup *map;
6062 em = get_chunk_map(fs_info, chunk_start, 1);
6066 map = em->map_lookup;
6068 rmap_len = map->stripe_len;
6070 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6071 length = div_u64(length, map->num_stripes / map->sub_stripes);
6072 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6073 length = div_u64(length, map->num_stripes);
6074 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6075 length = div_u64(length, nr_data_stripes(map));
6076 rmap_len = map->stripe_len * nr_data_stripes(map);
6079 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6080 BUG_ON(!buf); /* -ENOMEM */
6082 for (i = 0; i < map->num_stripes; i++) {
6083 if (devid && map->stripes[i].dev->devid != devid)
6085 if (map->stripes[i].physical > physical ||
6086 map->stripes[i].physical + length <= physical)
6089 stripe_nr = physical - map->stripes[i].physical;
6090 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6092 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6093 stripe_nr = stripe_nr * map->num_stripes + i;
6094 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6095 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6096 stripe_nr = stripe_nr * map->num_stripes + i;
6097 } /* else if RAID[56], multiply by nr_data_stripes().
6098 * Alternatively, just use rmap_len below instead of
6099 * map->stripe_len */
6101 bytenr = chunk_start + stripe_nr * rmap_len;
6102 WARN_ON(nr >= map->num_stripes);
6103 for (j = 0; j < nr; j++) {
6104 if (buf[j] == bytenr)
6108 WARN_ON(nr >= map->num_stripes);
6115 *stripe_len = rmap_len;
6117 free_extent_map(em);
6121 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6123 bio->bi_private = bbio->private;
6124 bio->bi_end_io = bbio->end_io;
6127 btrfs_put_bbio(bbio);
6130 static void btrfs_end_bio(struct bio *bio)
6132 struct btrfs_bio *bbio = bio->bi_private;
6133 int is_orig_bio = 0;
6135 if (bio->bi_status) {
6136 atomic_inc(&bbio->error);
6137 if (bio->bi_status == BLK_STS_IOERR ||
6138 bio->bi_status == BLK_STS_TARGET) {
6139 unsigned int stripe_index =
6140 btrfs_io_bio(bio)->stripe_index;
6141 struct btrfs_device *dev;
6143 BUG_ON(stripe_index >= bbio->num_stripes);
6144 dev = bbio->stripes[stripe_index].dev;
6146 if (bio_op(bio) == REQ_OP_WRITE)
6147 btrfs_dev_stat_inc_and_print(dev,
6148 BTRFS_DEV_STAT_WRITE_ERRS);
6150 btrfs_dev_stat_inc_and_print(dev,
6151 BTRFS_DEV_STAT_READ_ERRS);
6152 if (bio->bi_opf & REQ_PREFLUSH)
6153 btrfs_dev_stat_inc_and_print(dev,
6154 BTRFS_DEV_STAT_FLUSH_ERRS);
6159 if (bio == bbio->orig_bio)
6162 btrfs_bio_counter_dec(bbio->fs_info);
6164 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6167 bio = bbio->orig_bio;
6170 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6171 /* only send an error to the higher layers if it is
6172 * beyond the tolerance of the btrfs bio
6174 if (atomic_read(&bbio->error) > bbio->max_errors) {
6175 bio->bi_status = BLK_STS_IOERR;
6178 * this bio is actually up to date, we didn't
6179 * go over the max number of errors
6181 bio->bi_status = BLK_STS_OK;
6184 btrfs_end_bbio(bbio, bio);
6185 } else if (!is_orig_bio) {
6191 * see run_scheduled_bios for a description of why bios are collected for
6194 * This will add one bio to the pending list for a device and make sure
6195 * the work struct is scheduled.
6197 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6200 struct btrfs_fs_info *fs_info = device->fs_info;
6201 int should_queue = 1;
6202 struct btrfs_pending_bios *pending_bios;
6204 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6210 /* don't bother with additional async steps for reads, right now */
6211 if (bio_op(bio) == REQ_OP_READ) {
6212 btrfsic_submit_bio(bio);
6216 WARN_ON(bio->bi_next);
6217 bio->bi_next = NULL;
6219 spin_lock(&device->io_lock);
6220 if (op_is_sync(bio->bi_opf))
6221 pending_bios = &device->pending_sync_bios;
6223 pending_bios = &device->pending_bios;
6225 if (pending_bios->tail)
6226 pending_bios->tail->bi_next = bio;
6228 pending_bios->tail = bio;
6229 if (!pending_bios->head)
6230 pending_bios->head = bio;
6231 if (device->running_pending)
6234 spin_unlock(&device->io_lock);
6237 btrfs_queue_work(fs_info->submit_workers, &device->work);
6240 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6241 u64 physical, int dev_nr, int async)
6243 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6244 struct btrfs_fs_info *fs_info = bbio->fs_info;
6246 bio->bi_private = bbio;
6247 btrfs_io_bio(bio)->stripe_index = dev_nr;
6248 bio->bi_end_io = btrfs_end_bio;
6249 bio->bi_iter.bi_sector = physical >> 9;
6252 struct rcu_string *name;
6255 name = rcu_dereference(dev->name);
6256 btrfs_debug(fs_info,
6257 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6258 bio_op(bio), bio->bi_opf,
6259 (u64)bio->bi_iter.bi_sector,
6260 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6261 bio->bi_iter.bi_size);
6265 bio_set_dev(bio, dev->bdev);
6267 btrfs_bio_counter_inc_noblocked(fs_info);
6270 btrfs_schedule_bio(dev, bio);
6272 btrfsic_submit_bio(bio);
6275 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6277 atomic_inc(&bbio->error);
6278 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6279 /* Should be the original bio. */
6280 WARN_ON(bio != bbio->orig_bio);
6282 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6283 bio->bi_iter.bi_sector = logical >> 9;
6284 if (atomic_read(&bbio->error) > bbio->max_errors)
6285 bio->bi_status = BLK_STS_IOERR;
6287 bio->bi_status = BLK_STS_OK;
6288 btrfs_end_bbio(bbio, bio);
6292 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6293 int mirror_num, int async_submit)
6295 struct btrfs_device *dev;
6296 struct bio *first_bio = bio;
6297 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6303 struct btrfs_bio *bbio = NULL;
6305 length = bio->bi_iter.bi_size;
6306 map_length = length;
6308 btrfs_bio_counter_inc_blocked(fs_info);
6309 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6310 &map_length, &bbio, mirror_num, 1);
6312 btrfs_bio_counter_dec(fs_info);
6313 return errno_to_blk_status(ret);
6316 total_devs = bbio->num_stripes;
6317 bbio->orig_bio = first_bio;
6318 bbio->private = first_bio->bi_private;
6319 bbio->end_io = first_bio->bi_end_io;
6320 bbio->fs_info = fs_info;
6321 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6323 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6324 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6325 /* In this case, map_length has been set to the length of
6326 a single stripe; not the whole write */
6327 if (bio_op(bio) == REQ_OP_WRITE) {
6328 ret = raid56_parity_write(fs_info, bio, bbio,
6331 ret = raid56_parity_recover(fs_info, bio, bbio,
6332 map_length, mirror_num, 1);
6335 btrfs_bio_counter_dec(fs_info);
6336 return errno_to_blk_status(ret);
6339 if (map_length < length) {
6341 "mapping failed logical %llu bio len %llu len %llu",
6342 logical, length, map_length);
6346 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6347 dev = bbio->stripes[dev_nr].dev;
6348 if (!dev || !dev->bdev ||
6349 (bio_op(first_bio) == REQ_OP_WRITE &&
6350 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6351 bbio_error(bbio, first_bio, logical);
6355 if (dev_nr < total_devs - 1)
6356 bio = btrfs_bio_clone(first_bio);
6360 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6361 dev_nr, async_submit);
6363 btrfs_bio_counter_dec(fs_info);
6367 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6370 struct btrfs_device *device;
6371 struct btrfs_fs_devices *cur_devices;
6373 cur_devices = fs_info->fs_devices;
6374 while (cur_devices) {
6376 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6377 device = find_device(cur_devices, devid, uuid);
6381 cur_devices = cur_devices->seed;
6386 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6387 u64 devid, u8 *dev_uuid)
6389 struct btrfs_device *device;
6391 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6395 list_add(&device->dev_list, &fs_devices->devices);
6396 device->fs_devices = fs_devices;
6397 fs_devices->num_devices++;
6399 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6400 fs_devices->missing_devices++;
6406 * btrfs_alloc_device - allocate struct btrfs_device
6407 * @fs_info: used only for generating a new devid, can be NULL if
6408 * devid is provided (i.e. @devid != NULL).
6409 * @devid: a pointer to devid for this device. If NULL a new devid
6411 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6414 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6415 * on error. Returned struct is not linked onto any lists and must be
6416 * destroyed with free_device.
6418 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6422 struct btrfs_device *dev;
6425 if (WARN_ON(!devid && !fs_info))
6426 return ERR_PTR(-EINVAL);
6428 dev = __alloc_device();
6437 ret = find_next_devid(fs_info, &tmp);
6440 return ERR_PTR(ret);
6446 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6448 generate_random_uuid(dev->uuid);
6450 btrfs_init_work(&dev->work, btrfs_submit_helper,
6451 pending_bios_fn, NULL, NULL);
6456 /* Return -EIO if any error, otherwise return 0. */
6457 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6458 struct extent_buffer *leaf,
6459 struct btrfs_chunk *chunk, u64 logical)
6467 length = btrfs_chunk_length(leaf, chunk);
6468 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6469 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6470 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6471 type = btrfs_chunk_type(leaf, chunk);
6474 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6478 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6479 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6482 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6483 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6484 btrfs_chunk_sector_size(leaf, chunk));
6487 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6488 btrfs_err(fs_info, "invalid chunk length %llu", length);
6491 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6492 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6496 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6498 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6499 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6500 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6501 btrfs_chunk_type(leaf, chunk));
6504 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6505 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6506 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6507 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6508 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6509 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6510 num_stripes != 1)) {
6512 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6513 num_stripes, sub_stripes,
6514 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6521 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6522 u64 devid, u8 *uuid, bool error)
6525 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6528 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6532 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6533 struct extent_buffer *leaf,
6534 struct btrfs_chunk *chunk)
6536 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6537 struct map_lookup *map;
6538 struct extent_map *em;
6542 u8 uuid[BTRFS_UUID_SIZE];
6547 logical = key->offset;
6548 length = btrfs_chunk_length(leaf, chunk);
6549 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6551 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6555 read_lock(&map_tree->map_tree.lock);
6556 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6557 read_unlock(&map_tree->map_tree.lock);
6559 /* already mapped? */
6560 if (em && em->start <= logical && em->start + em->len > logical) {
6561 free_extent_map(em);
6564 free_extent_map(em);
6567 em = alloc_extent_map();
6570 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6572 free_extent_map(em);
6576 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6577 em->map_lookup = map;
6578 em->start = logical;
6581 em->block_start = 0;
6582 em->block_len = em->len;
6584 map->num_stripes = num_stripes;
6585 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6586 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6587 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6588 map->type = btrfs_chunk_type(leaf, chunk);
6589 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6590 for (i = 0; i < num_stripes; i++) {
6591 map->stripes[i].physical =
6592 btrfs_stripe_offset_nr(leaf, chunk, i);
6593 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6594 read_extent_buffer(leaf, uuid, (unsigned long)
6595 btrfs_stripe_dev_uuid_nr(chunk, i),
6597 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6599 if (!map->stripes[i].dev &&
6600 !btrfs_test_opt(fs_info, DEGRADED)) {
6601 free_extent_map(em);
6602 btrfs_report_missing_device(fs_info, devid, uuid, true);
6605 if (!map->stripes[i].dev) {
6606 map->stripes[i].dev =
6607 add_missing_dev(fs_info->fs_devices, devid,
6609 if (IS_ERR(map->stripes[i].dev)) {
6610 free_extent_map(em);
6612 "failed to init missing dev %llu: %ld",
6613 devid, PTR_ERR(map->stripes[i].dev));
6614 return PTR_ERR(map->stripes[i].dev);
6616 btrfs_report_missing_device(fs_info, devid, uuid, false);
6618 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6619 &(map->stripes[i].dev->dev_state));
6623 write_lock(&map_tree->map_tree.lock);
6624 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6625 write_unlock(&map_tree->map_tree.lock);
6626 BUG_ON(ret); /* Tree corruption */
6627 free_extent_map(em);
6632 static void fill_device_from_item(struct extent_buffer *leaf,
6633 struct btrfs_dev_item *dev_item,
6634 struct btrfs_device *device)
6638 device->devid = btrfs_device_id(leaf, dev_item);
6639 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6640 device->total_bytes = device->disk_total_bytes;
6641 device->commit_total_bytes = device->disk_total_bytes;
6642 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6643 device->commit_bytes_used = device->bytes_used;
6644 device->type = btrfs_device_type(leaf, dev_item);
6645 device->io_align = btrfs_device_io_align(leaf, dev_item);
6646 device->io_width = btrfs_device_io_width(leaf, dev_item);
6647 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6648 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6649 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6651 ptr = btrfs_device_uuid(dev_item);
6652 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6655 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6658 struct btrfs_fs_devices *fs_devices;
6661 lockdep_assert_held(&uuid_mutex);
6664 fs_devices = fs_info->fs_devices->seed;
6665 while (fs_devices) {
6666 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6669 fs_devices = fs_devices->seed;
6672 fs_devices = find_fsid(fsid);
6674 if (!btrfs_test_opt(fs_info, DEGRADED))
6675 return ERR_PTR(-ENOENT);
6677 fs_devices = alloc_fs_devices(fsid);
6678 if (IS_ERR(fs_devices))
6681 fs_devices->seeding = 1;
6682 fs_devices->opened = 1;
6686 fs_devices = clone_fs_devices(fs_devices);
6687 if (IS_ERR(fs_devices))
6690 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6691 fs_info->bdev_holder);
6693 free_fs_devices(fs_devices);
6694 fs_devices = ERR_PTR(ret);
6698 if (!fs_devices->seeding) {
6699 __btrfs_close_devices(fs_devices);
6700 free_fs_devices(fs_devices);
6701 fs_devices = ERR_PTR(-EINVAL);
6705 fs_devices->seed = fs_info->fs_devices->seed;
6706 fs_info->fs_devices->seed = fs_devices;
6711 static int read_one_dev(struct btrfs_fs_info *fs_info,
6712 struct extent_buffer *leaf,
6713 struct btrfs_dev_item *dev_item)
6715 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6716 struct btrfs_device *device;
6719 u8 fs_uuid[BTRFS_FSID_SIZE];
6720 u8 dev_uuid[BTRFS_UUID_SIZE];
6722 devid = btrfs_device_id(leaf, dev_item);
6723 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6725 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6728 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6729 fs_devices = open_seed_devices(fs_info, fs_uuid);
6730 if (IS_ERR(fs_devices))
6731 return PTR_ERR(fs_devices);
6734 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6736 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6737 btrfs_report_missing_device(fs_info, devid,
6742 device = add_missing_dev(fs_devices, devid, dev_uuid);
6743 if (IS_ERR(device)) {
6745 "failed to add missing dev %llu: %ld",
6746 devid, PTR_ERR(device));
6747 return PTR_ERR(device);
6749 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6751 if (!device->bdev) {
6752 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6753 btrfs_report_missing_device(fs_info,
6754 devid, dev_uuid, true);
6757 btrfs_report_missing_device(fs_info, devid,
6761 if (!device->bdev &&
6762 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6764 * this happens when a device that was properly setup
6765 * in the device info lists suddenly goes bad.
6766 * device->bdev is NULL, and so we have to set
6767 * device->missing to one here
6769 device->fs_devices->missing_devices++;
6770 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6773 /* Move the device to its own fs_devices */
6774 if (device->fs_devices != fs_devices) {
6775 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6776 &device->dev_state));
6778 list_move(&device->dev_list, &fs_devices->devices);
6779 device->fs_devices->num_devices--;
6780 fs_devices->num_devices++;
6782 device->fs_devices->missing_devices--;
6783 fs_devices->missing_devices++;
6785 device->fs_devices = fs_devices;
6789 if (device->fs_devices != fs_info->fs_devices) {
6790 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6791 if (device->generation !=
6792 btrfs_device_generation(leaf, dev_item))
6796 fill_device_from_item(leaf, dev_item, device);
6797 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6798 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6799 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6800 device->fs_devices->total_rw_bytes += device->total_bytes;
6801 atomic64_add(device->total_bytes - device->bytes_used,
6802 &fs_info->free_chunk_space);
6808 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6810 struct btrfs_root *root = fs_info->tree_root;
6811 struct btrfs_super_block *super_copy = fs_info->super_copy;
6812 struct extent_buffer *sb;
6813 struct btrfs_disk_key *disk_key;
6814 struct btrfs_chunk *chunk;
6816 unsigned long sb_array_offset;
6823 struct btrfs_key key;
6825 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6827 * This will create extent buffer of nodesize, superblock size is
6828 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6829 * overallocate but we can keep it as-is, only the first page is used.
6831 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6834 set_extent_buffer_uptodate(sb);
6835 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6837 * The sb extent buffer is artificial and just used to read the system array.
6838 * set_extent_buffer_uptodate() call does not properly mark all it's
6839 * pages up-to-date when the page is larger: extent does not cover the
6840 * whole page and consequently check_page_uptodate does not find all
6841 * the page's extents up-to-date (the hole beyond sb),
6842 * write_extent_buffer then triggers a WARN_ON.
6844 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6845 * but sb spans only this function. Add an explicit SetPageUptodate call
6846 * to silence the warning eg. on PowerPC 64.
6848 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6849 SetPageUptodate(sb->pages[0]);
6851 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6852 array_size = btrfs_super_sys_array_size(super_copy);
6854 array_ptr = super_copy->sys_chunk_array;
6855 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6858 while (cur_offset < array_size) {
6859 disk_key = (struct btrfs_disk_key *)array_ptr;
6860 len = sizeof(*disk_key);
6861 if (cur_offset + len > array_size)
6862 goto out_short_read;
6864 btrfs_disk_key_to_cpu(&key, disk_key);
6867 sb_array_offset += len;
6870 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6871 chunk = (struct btrfs_chunk *)sb_array_offset;
6873 * At least one btrfs_chunk with one stripe must be
6874 * present, exact stripe count check comes afterwards
6876 len = btrfs_chunk_item_size(1);
6877 if (cur_offset + len > array_size)
6878 goto out_short_read;
6880 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6883 "invalid number of stripes %u in sys_array at offset %u",
6884 num_stripes, cur_offset);
6889 type = btrfs_chunk_type(sb, chunk);
6890 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6892 "invalid chunk type %llu in sys_array at offset %u",
6898 len = btrfs_chunk_item_size(num_stripes);
6899 if (cur_offset + len > array_size)
6900 goto out_short_read;
6902 ret = read_one_chunk(fs_info, &key, sb, chunk);
6907 "unexpected item type %u in sys_array at offset %u",
6908 (u32)key.type, cur_offset);
6913 sb_array_offset += len;
6916 clear_extent_buffer_uptodate(sb);
6917 free_extent_buffer_stale(sb);
6921 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6923 clear_extent_buffer_uptodate(sb);
6924 free_extent_buffer_stale(sb);
6929 * Check if all chunks in the fs are OK for read-write degraded mount
6931 * If the @failing_dev is specified, it's accounted as missing.
6933 * Return true if all chunks meet the minimal RW mount requirements.
6934 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6936 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6937 struct btrfs_device *failing_dev)
6939 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6940 struct extent_map *em;
6944 read_lock(&map_tree->map_tree.lock);
6945 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6946 read_unlock(&map_tree->map_tree.lock);
6947 /* No chunk at all? Return false anyway */
6953 struct map_lookup *map;
6958 map = em->map_lookup;
6960 btrfs_get_num_tolerated_disk_barrier_failures(
6962 for (i = 0; i < map->num_stripes; i++) {
6963 struct btrfs_device *dev = map->stripes[i].dev;
6965 if (!dev || !dev->bdev ||
6966 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6967 dev->last_flush_error)
6969 else if (failing_dev && failing_dev == dev)
6972 if (missing > max_tolerated) {
6975 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6976 em->start, missing, max_tolerated);
6977 free_extent_map(em);
6981 next_start = extent_map_end(em);
6982 free_extent_map(em);
6984 read_lock(&map_tree->map_tree.lock);
6985 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6986 (u64)(-1) - next_start);
6987 read_unlock(&map_tree->map_tree.lock);
6993 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6995 struct btrfs_root *root = fs_info->chunk_root;
6996 struct btrfs_path *path;
6997 struct extent_buffer *leaf;
6998 struct btrfs_key key;
6999 struct btrfs_key found_key;
7004 path = btrfs_alloc_path();
7008 mutex_lock(&uuid_mutex);
7009 mutex_lock(&fs_info->chunk_mutex);
7012 * Read all device items, and then all the chunk items. All
7013 * device items are found before any chunk item (their object id
7014 * is smaller than the lowest possible object id for a chunk
7015 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7017 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7020 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7024 leaf = path->nodes[0];
7025 slot = path->slots[0];
7026 if (slot >= btrfs_header_nritems(leaf)) {
7027 ret = btrfs_next_leaf(root, path);
7034 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7035 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7036 struct btrfs_dev_item *dev_item;
7037 dev_item = btrfs_item_ptr(leaf, slot,
7038 struct btrfs_dev_item);
7039 ret = read_one_dev(fs_info, leaf, dev_item);
7043 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7044 struct btrfs_chunk *chunk;
7045 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7046 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7054 * After loading chunk tree, we've got all device information,
7055 * do another round of validation checks.
7057 if (total_dev != fs_info->fs_devices->total_devices) {
7059 "super_num_devices %llu mismatch with num_devices %llu found here",
7060 btrfs_super_num_devices(fs_info->super_copy),
7065 if (btrfs_super_total_bytes(fs_info->super_copy) <
7066 fs_info->fs_devices->total_rw_bytes) {
7068 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7069 btrfs_super_total_bytes(fs_info->super_copy),
7070 fs_info->fs_devices->total_rw_bytes);
7076 mutex_unlock(&fs_info->chunk_mutex);
7077 mutex_unlock(&uuid_mutex);
7079 btrfs_free_path(path);
7083 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7085 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7086 struct btrfs_device *device;
7088 while (fs_devices) {
7089 mutex_lock(&fs_devices->device_list_mutex);
7090 list_for_each_entry(device, &fs_devices->devices, dev_list)
7091 device->fs_info = fs_info;
7092 mutex_unlock(&fs_devices->device_list_mutex);
7094 fs_devices = fs_devices->seed;
7098 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7102 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7103 btrfs_dev_stat_reset(dev, i);
7106 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7108 struct btrfs_key key;
7109 struct btrfs_key found_key;
7110 struct btrfs_root *dev_root = fs_info->dev_root;
7111 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7112 struct extent_buffer *eb;
7115 struct btrfs_device *device;
7116 struct btrfs_path *path = NULL;
7119 path = btrfs_alloc_path();
7125 mutex_lock(&fs_devices->device_list_mutex);
7126 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7128 struct btrfs_dev_stats_item *ptr;
7130 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7131 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7132 key.offset = device->devid;
7133 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7135 __btrfs_reset_dev_stats(device);
7136 device->dev_stats_valid = 1;
7137 btrfs_release_path(path);
7140 slot = path->slots[0];
7141 eb = path->nodes[0];
7142 btrfs_item_key_to_cpu(eb, &found_key, slot);
7143 item_size = btrfs_item_size_nr(eb, slot);
7145 ptr = btrfs_item_ptr(eb, slot,
7146 struct btrfs_dev_stats_item);
7148 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7149 if (item_size >= (1 + i) * sizeof(__le64))
7150 btrfs_dev_stat_set(device, i,
7151 btrfs_dev_stats_value(eb, ptr, i));
7153 btrfs_dev_stat_reset(device, i);
7156 device->dev_stats_valid = 1;
7157 btrfs_dev_stat_print_on_load(device);
7158 btrfs_release_path(path);
7160 mutex_unlock(&fs_devices->device_list_mutex);
7163 btrfs_free_path(path);
7164 return ret < 0 ? ret : 0;
7167 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7168 struct btrfs_fs_info *fs_info,
7169 struct btrfs_device *device)
7171 struct btrfs_root *dev_root = fs_info->dev_root;
7172 struct btrfs_path *path;
7173 struct btrfs_key key;
7174 struct extent_buffer *eb;
7175 struct btrfs_dev_stats_item *ptr;
7179 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7180 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7181 key.offset = device->devid;
7183 path = btrfs_alloc_path();
7186 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7188 btrfs_warn_in_rcu(fs_info,
7189 "error %d while searching for dev_stats item for device %s",
7190 ret, rcu_str_deref(device->name));
7195 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7196 /* need to delete old one and insert a new one */
7197 ret = btrfs_del_item(trans, dev_root, path);
7199 btrfs_warn_in_rcu(fs_info,
7200 "delete too small dev_stats item for device %s failed %d",
7201 rcu_str_deref(device->name), ret);
7208 /* need to insert a new item */
7209 btrfs_release_path(path);
7210 ret = btrfs_insert_empty_item(trans, dev_root, path,
7211 &key, sizeof(*ptr));
7213 btrfs_warn_in_rcu(fs_info,
7214 "insert dev_stats item for device %s failed %d",
7215 rcu_str_deref(device->name), ret);
7220 eb = path->nodes[0];
7221 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7222 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7223 btrfs_set_dev_stats_value(eb, ptr, i,
7224 btrfs_dev_stat_read(device, i));
7225 btrfs_mark_buffer_dirty(eb);
7228 btrfs_free_path(path);
7233 * called from commit_transaction. Writes all changed device stats to disk.
7235 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7236 struct btrfs_fs_info *fs_info)
7238 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7239 struct btrfs_device *device;
7243 mutex_lock(&fs_devices->device_list_mutex);
7244 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7245 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7246 if (!device->dev_stats_valid || stats_cnt == 0)
7251 * There is a LOAD-LOAD control dependency between the value of
7252 * dev_stats_ccnt and updating the on-disk values which requires
7253 * reading the in-memory counters. Such control dependencies
7254 * require explicit read memory barriers.
7256 * This memory barriers pairs with smp_mb__before_atomic in
7257 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7258 * barrier implied by atomic_xchg in
7259 * btrfs_dev_stats_read_and_reset
7263 ret = update_dev_stat_item(trans, fs_info, device);
7265 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7267 mutex_unlock(&fs_devices->device_list_mutex);
7272 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7274 btrfs_dev_stat_inc(dev, index);
7275 btrfs_dev_stat_print_on_error(dev);
7278 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7280 if (!dev->dev_stats_valid)
7282 btrfs_err_rl_in_rcu(dev->fs_info,
7283 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7284 rcu_str_deref(dev->name),
7285 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7286 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7287 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7288 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7289 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7292 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7296 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7297 if (btrfs_dev_stat_read(dev, i) != 0)
7299 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7300 return; /* all values == 0, suppress message */
7302 btrfs_info_in_rcu(dev->fs_info,
7303 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7304 rcu_str_deref(dev->name),
7305 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7306 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7307 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7308 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7309 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7312 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7313 struct btrfs_ioctl_get_dev_stats *stats)
7315 struct btrfs_device *dev;
7316 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7319 mutex_lock(&fs_devices->device_list_mutex);
7320 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7321 mutex_unlock(&fs_devices->device_list_mutex);
7324 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7326 } else if (!dev->dev_stats_valid) {
7327 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7329 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7330 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7331 if (stats->nr_items > i)
7333 btrfs_dev_stat_read_and_reset(dev, i);
7335 btrfs_dev_stat_reset(dev, i);
7338 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7339 if (stats->nr_items > i)
7340 stats->values[i] = btrfs_dev_stat_read(dev, i);
7342 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7343 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7347 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7349 struct buffer_head *bh;
7350 struct btrfs_super_block *disk_super;
7356 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7359 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7362 disk_super = (struct btrfs_super_block *)bh->b_data;
7364 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7365 set_buffer_dirty(bh);
7366 sync_dirty_buffer(bh);
7370 /* Notify udev that device has changed */
7371 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7373 /* Update ctime/mtime for device path for libblkid */
7374 update_dev_time(device_path);
7378 * Update the size of all devices, which is used for writing out the
7381 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7383 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7384 struct btrfs_device *curr, *next;
7386 if (list_empty(&fs_devices->resized_devices))
7389 mutex_lock(&fs_devices->device_list_mutex);
7390 mutex_lock(&fs_info->chunk_mutex);
7391 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7393 list_del_init(&curr->resized_list);
7394 curr->commit_total_bytes = curr->disk_total_bytes;
7396 mutex_unlock(&fs_info->chunk_mutex);
7397 mutex_unlock(&fs_devices->device_list_mutex);
7400 /* Must be invoked during the transaction commit */
7401 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7403 struct btrfs_fs_info *fs_info = trans->fs_info;
7404 struct extent_map *em;
7405 struct map_lookup *map;
7406 struct btrfs_device *dev;
7409 if (list_empty(&trans->pending_chunks))
7412 /* In order to kick the device replace finish process */
7413 mutex_lock(&fs_info->chunk_mutex);
7414 list_for_each_entry(em, &trans->pending_chunks, list) {
7415 map = em->map_lookup;
7417 for (i = 0; i < map->num_stripes; i++) {
7418 dev = map->stripes[i].dev;
7419 dev->commit_bytes_used = dev->bytes_used;
7422 mutex_unlock(&fs_info->chunk_mutex);
7425 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7427 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7428 while (fs_devices) {
7429 fs_devices->fs_info = fs_info;
7430 fs_devices = fs_devices->seed;
7434 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7436 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7437 while (fs_devices) {
7438 fs_devices->fs_info = NULL;
7439 fs_devices = fs_devices->seed;