1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/iocontext.h>
12 #include <linux/capability.h>
13 #include <linux/ratelimit.h>
14 #include <linux/kthread.h>
15 #include <linux/raid/pq.h>
16 #include <linux/semaphore.h>
17 #include <linux/uuid.h>
18 #include <linux/list_sort.h>
19 #include <asm/div64.h>
21 #include "extent_map.h"
23 #include "transaction.h"
24 #include "print-tree.h"
27 #include "async-thread.h"
28 #include "check-integrity.h"
29 #include "rcu-string.h"
31 #include "dev-replace.h"
34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
35 [BTRFS_RAID_RAID10] = {
38 .devs_max = 0, /* 0 == as many as possible */
40 .tolerated_failures = 1,
44 [BTRFS_RAID_RAID1] = {
49 .tolerated_failures = 1,
58 .tolerated_failures = 0,
62 [BTRFS_RAID_RAID0] = {
67 .tolerated_failures = 0,
71 [BTRFS_RAID_SINGLE] = {
76 .tolerated_failures = 0,
80 [BTRFS_RAID_RAID5] = {
85 .tolerated_failures = 1,
89 [BTRFS_RAID_RAID6] = {
94 .tolerated_failures = 2,
100 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
101 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
102 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
103 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
104 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
105 [BTRFS_RAID_SINGLE] = 0,
106 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
107 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
111 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
112 * condition is not met. Zero means there's no corresponding
113 * BTRFS_ERROR_DEV_*_NOT_MET value.
115 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
116 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
117 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
118 [BTRFS_RAID_DUP] = 0,
119 [BTRFS_RAID_RAID0] = 0,
120 [BTRFS_RAID_SINGLE] = 0,
121 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
122 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
125 static int init_first_rw_device(struct btrfs_trans_handle *trans,
126 struct btrfs_fs_info *fs_info);
127 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
128 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
129 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
130 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
131 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
132 enum btrfs_map_op op,
133 u64 logical, u64 *length,
134 struct btrfs_bio **bbio_ret,
135 int mirror_num, int need_raid_map);
141 * There are several mutexes that protect manipulation of devices and low-level
142 * structures like chunks but not block groups, extents or files
144 * uuid_mutex (global lock)
145 * ------------------------
146 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
147 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
148 * device) or requested by the device= mount option
150 * the mutex can be very coarse and can cover long-running operations
152 * protects: updates to fs_devices counters like missing devices, rw devices,
153 * seeding, structure cloning, openning/closing devices at mount/umount time
155 * global::fs_devs - add, remove, updates to the global list
157 * does not protect: manipulation of the fs_devices::devices list!
159 * btrfs_device::name - renames (write side), read is RCU
161 * fs_devices::device_list_mutex (per-fs, with RCU)
162 * ------------------------------------------------
163 * protects updates to fs_devices::devices, ie. adding and deleting
165 * simple list traversal with read-only actions can be done with RCU protection
167 * may be used to exclude some operations from running concurrently without any
168 * modifications to the list (see write_all_supers)
172 * protects balance structures (status, state) and context accessed from
173 * several places (internally, ioctl)
177 * protects chunks, adding or removing during allocation, trim or when a new
178 * device is added/removed
182 * a big lock that is held by the cleaner thread and prevents running subvolume
183 * cleaning together with relocation or delayed iputs
196 * Exclusive operations, BTRFS_FS_EXCL_OP
197 * ======================================
199 * Maintains the exclusivity of the following operations that apply to the
200 * whole filesystem and cannot run in parallel.
205 * - Device replace (*)
208 * The device operations (as above) can be in one of the following states:
214 * Only device operations marked with (*) can go into the Paused state for the
217 * - ioctl (only Balance can be Paused through ioctl)
218 * - filesystem remounted as read-only
219 * - filesystem unmounted and mounted as read-only
220 * - system power-cycle and filesystem mounted as read-only
221 * - filesystem or device errors leading to forced read-only
223 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
224 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
225 * A device operation in Paused or Running state can be canceled or resumed
226 * either by ioctl (Balance only) or when remounted as read-write.
227 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
231 DEFINE_MUTEX(uuid_mutex);
232 static LIST_HEAD(fs_uuids);
233 struct list_head *btrfs_get_fs_uuids(void)
239 * alloc_fs_devices - allocate struct btrfs_fs_devices
240 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
242 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
243 * The returned struct is not linked onto any lists and can be destroyed with
244 * kfree() right away.
246 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
248 struct btrfs_fs_devices *fs_devs;
250 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
252 return ERR_PTR(-ENOMEM);
254 mutex_init(&fs_devs->device_list_mutex);
256 INIT_LIST_HEAD(&fs_devs->devices);
257 INIT_LIST_HEAD(&fs_devs->resized_devices);
258 INIT_LIST_HEAD(&fs_devs->alloc_list);
259 INIT_LIST_HEAD(&fs_devs->fs_list);
261 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
266 void btrfs_free_device(struct btrfs_device *device)
268 rcu_string_free(device->name);
269 bio_put(device->flush_bio);
273 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
275 struct btrfs_device *device;
276 WARN_ON(fs_devices->opened);
277 while (!list_empty(&fs_devices->devices)) {
278 device = list_entry(fs_devices->devices.next,
279 struct btrfs_device, dev_list);
280 list_del(&device->dev_list);
281 btrfs_free_device(device);
286 static void btrfs_kobject_uevent(struct block_device *bdev,
287 enum kobject_action action)
291 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
293 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
295 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
296 &disk_to_dev(bdev->bd_disk)->kobj);
299 void __exit btrfs_cleanup_fs_uuids(void)
301 struct btrfs_fs_devices *fs_devices;
303 while (!list_empty(&fs_uuids)) {
304 fs_devices = list_entry(fs_uuids.next,
305 struct btrfs_fs_devices, fs_list);
306 list_del(&fs_devices->fs_list);
307 free_fs_devices(fs_devices);
312 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
313 * Returned struct is not linked onto any lists and must be destroyed using
316 static struct btrfs_device *__alloc_device(void)
318 struct btrfs_device *dev;
320 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
322 return ERR_PTR(-ENOMEM);
325 * Preallocate a bio that's always going to be used for flushing device
326 * barriers and matches the device lifespan
328 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
329 if (!dev->flush_bio) {
331 return ERR_PTR(-ENOMEM);
334 INIT_LIST_HEAD(&dev->dev_list);
335 INIT_LIST_HEAD(&dev->dev_alloc_list);
336 INIT_LIST_HEAD(&dev->resized_list);
338 spin_lock_init(&dev->io_lock);
340 atomic_set(&dev->reada_in_flight, 0);
341 atomic_set(&dev->dev_stats_ccnt, 0);
342 btrfs_device_data_ordered_init(dev);
343 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
344 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
350 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
353 * If devid and uuid are both specified, the match must be exact, otherwise
354 * only devid is used.
356 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
357 u64 devid, const u8 *uuid)
359 struct btrfs_device *dev;
361 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
362 if (dev->devid == devid &&
363 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
370 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
372 struct btrfs_fs_devices *fs_devices;
374 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
375 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
382 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
383 int flush, struct block_device **bdev,
384 struct buffer_head **bh)
388 *bdev = blkdev_get_by_path(device_path, flags, holder);
391 ret = PTR_ERR(*bdev);
396 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
397 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
399 blkdev_put(*bdev, flags);
402 invalidate_bdev(*bdev);
403 *bh = btrfs_read_dev_super(*bdev);
406 blkdev_put(*bdev, flags);
418 static void requeue_list(struct btrfs_pending_bios *pending_bios,
419 struct bio *head, struct bio *tail)
422 struct bio *old_head;
424 old_head = pending_bios->head;
425 pending_bios->head = head;
426 if (pending_bios->tail)
427 tail->bi_next = old_head;
429 pending_bios->tail = tail;
433 * we try to collect pending bios for a device so we don't get a large
434 * number of procs sending bios down to the same device. This greatly
435 * improves the schedulers ability to collect and merge the bios.
437 * But, it also turns into a long list of bios to process and that is sure
438 * to eventually make the worker thread block. The solution here is to
439 * make some progress and then put this work struct back at the end of
440 * the list if the block device is congested. This way, multiple devices
441 * can make progress from a single worker thread.
443 static noinline void run_scheduled_bios(struct btrfs_device *device)
445 struct btrfs_fs_info *fs_info = device->fs_info;
447 struct backing_dev_info *bdi;
448 struct btrfs_pending_bios *pending_bios;
452 unsigned long num_run;
453 unsigned long batch_run = 0;
454 unsigned long last_waited = 0;
456 int sync_pending = 0;
457 struct blk_plug plug;
460 * this function runs all the bios we've collected for
461 * a particular device. We don't want to wander off to
462 * another device without first sending all of these down.
463 * So, setup a plug here and finish it off before we return
465 blk_start_plug(&plug);
467 bdi = device->bdev->bd_bdi;
470 spin_lock(&device->io_lock);
475 /* take all the bios off the list at once and process them
476 * later on (without the lock held). But, remember the
477 * tail and other pointers so the bios can be properly reinserted
478 * into the list if we hit congestion
480 if (!force_reg && device->pending_sync_bios.head) {
481 pending_bios = &device->pending_sync_bios;
484 pending_bios = &device->pending_bios;
488 pending = pending_bios->head;
489 tail = pending_bios->tail;
490 WARN_ON(pending && !tail);
493 * if pending was null this time around, no bios need processing
494 * at all and we can stop. Otherwise it'll loop back up again
495 * and do an additional check so no bios are missed.
497 * device->running_pending is used to synchronize with the
500 if (device->pending_sync_bios.head == NULL &&
501 device->pending_bios.head == NULL) {
503 device->running_pending = 0;
506 device->running_pending = 1;
509 pending_bios->head = NULL;
510 pending_bios->tail = NULL;
512 spin_unlock(&device->io_lock);
517 /* we want to work on both lists, but do more bios on the
518 * sync list than the regular list
521 pending_bios != &device->pending_sync_bios &&
522 device->pending_sync_bios.head) ||
523 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
524 device->pending_bios.head)) {
525 spin_lock(&device->io_lock);
526 requeue_list(pending_bios, pending, tail);
531 pending = pending->bi_next;
534 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
537 * if we're doing the sync list, record that our
538 * plug has some sync requests on it
540 * If we're doing the regular list and there are
541 * sync requests sitting around, unplug before
544 if (pending_bios == &device->pending_sync_bios) {
546 } else if (sync_pending) {
547 blk_finish_plug(&plug);
548 blk_start_plug(&plug);
552 btrfsic_submit_bio(cur);
559 * we made progress, there is more work to do and the bdi
560 * is now congested. Back off and let other work structs
563 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
564 fs_info->fs_devices->open_devices > 1) {
565 struct io_context *ioc;
567 ioc = current->io_context;
570 * the main goal here is that we don't want to
571 * block if we're going to be able to submit
572 * more requests without blocking.
574 * This code does two great things, it pokes into
575 * the elevator code from a filesystem _and_
576 * it makes assumptions about how batching works.
578 if (ioc && ioc->nr_batch_requests > 0 &&
579 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
581 ioc->last_waited == last_waited)) {
583 * we want to go through our batch of
584 * requests and stop. So, we copy out
585 * the ioc->last_waited time and test
586 * against it before looping
588 last_waited = ioc->last_waited;
592 spin_lock(&device->io_lock);
593 requeue_list(pending_bios, pending, tail);
594 device->running_pending = 1;
596 spin_unlock(&device->io_lock);
597 btrfs_queue_work(fs_info->submit_workers,
607 spin_lock(&device->io_lock);
608 if (device->pending_bios.head || device->pending_sync_bios.head)
610 spin_unlock(&device->io_lock);
613 blk_finish_plug(&plug);
616 static void pending_bios_fn(struct btrfs_work *work)
618 struct btrfs_device *device;
620 device = container_of(work, struct btrfs_device, work);
621 run_scheduled_bios(device);
625 * Search and remove all stale (devices which are not mounted) devices.
626 * When both inputs are NULL, it will search and release all stale devices.
627 * path: Optional. When provided will it release all unmounted devices
628 * matching this path only.
629 * skip_dev: Optional. Will skip this device when searching for the stale
632 static void btrfs_free_stale_devices(const char *path,
633 struct btrfs_device *skip_dev)
635 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
636 struct btrfs_device *dev, *tmp_dev;
638 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, fs_list) {
643 list_for_each_entry_safe(dev, tmp_dev,
644 &fs_devs->devices, dev_list) {
647 if (skip_dev && skip_dev == dev)
649 if (path && !dev->name)
654 not_found = strcmp(rcu_str_deref(dev->name),
660 /* delete the stale device */
661 if (fs_devs->num_devices == 1) {
662 btrfs_sysfs_remove_fsid(fs_devs);
663 list_del(&fs_devs->fs_list);
664 free_fs_devices(fs_devs);
667 fs_devs->num_devices--;
668 list_del(&dev->dev_list);
669 btrfs_free_device(dev);
675 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
676 struct btrfs_device *device, fmode_t flags,
679 struct request_queue *q;
680 struct block_device *bdev;
681 struct buffer_head *bh;
682 struct btrfs_super_block *disk_super;
691 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
696 disk_super = (struct btrfs_super_block *)bh->b_data;
697 devid = btrfs_stack_device_id(&disk_super->dev_item);
698 if (devid != device->devid)
701 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
704 device->generation = btrfs_super_generation(disk_super);
706 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
707 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
708 fs_devices->seeding = 1;
710 if (bdev_read_only(bdev))
711 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
713 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
716 q = bdev_get_queue(bdev);
717 if (!blk_queue_nonrot(q))
718 fs_devices->rotating = 1;
721 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
722 device->mode = flags;
724 fs_devices->open_devices++;
725 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
726 device->devid != BTRFS_DEV_REPLACE_DEVID) {
727 fs_devices->rw_devices++;
728 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
736 blkdev_put(bdev, flags);
742 * Add new device to list of registered devices
745 * device pointer which was just added or updated when successful
746 * error pointer when failed
748 static noinline struct btrfs_device *device_list_add(const char *path,
749 struct btrfs_super_block *disk_super)
751 struct btrfs_device *device;
752 struct btrfs_fs_devices *fs_devices;
753 struct rcu_string *name;
754 u64 found_transid = btrfs_super_generation(disk_super);
755 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
757 fs_devices = find_fsid(disk_super->fsid);
759 fs_devices = alloc_fs_devices(disk_super->fsid);
760 if (IS_ERR(fs_devices))
761 return ERR_CAST(fs_devices);
763 list_add(&fs_devices->fs_list, &fs_uuids);
767 device = find_device(fs_devices, devid,
768 disk_super->dev_item.uuid);
772 if (fs_devices->opened)
773 return ERR_PTR(-EBUSY);
775 device = btrfs_alloc_device(NULL, &devid,
776 disk_super->dev_item.uuid);
777 if (IS_ERR(device)) {
778 /* we can safely leave the fs_devices entry around */
782 name = rcu_string_strdup(path, GFP_NOFS);
784 btrfs_free_device(device);
785 return ERR_PTR(-ENOMEM);
787 rcu_assign_pointer(device->name, name);
789 mutex_lock(&fs_devices->device_list_mutex);
790 list_add_rcu(&device->dev_list, &fs_devices->devices);
791 fs_devices->num_devices++;
792 mutex_unlock(&fs_devices->device_list_mutex);
794 device->fs_devices = fs_devices;
795 btrfs_free_stale_devices(path, device);
797 if (disk_super->label[0])
798 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
799 disk_super->label, devid, found_transid, path);
801 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
802 disk_super->fsid, devid, found_transid, path);
804 } else if (!device->name || strcmp(device->name->str, path)) {
806 * When FS is already mounted.
807 * 1. If you are here and if the device->name is NULL that
808 * means this device was missing at time of FS mount.
809 * 2. If you are here and if the device->name is different
810 * from 'path' that means either
811 * a. The same device disappeared and reappeared with
813 * b. The missing-disk-which-was-replaced, has
816 * We must allow 1 and 2a above. But 2b would be a spurious
819 * Further in case of 1 and 2a above, the disk at 'path'
820 * would have missed some transaction when it was away and
821 * in case of 2a the stale bdev has to be updated as well.
822 * 2b must not be allowed at all time.
826 * For now, we do allow update to btrfs_fs_device through the
827 * btrfs dev scan cli after FS has been mounted. We're still
828 * tracking a problem where systems fail mount by subvolume id
829 * when we reject replacement on a mounted FS.
831 if (!fs_devices->opened && found_transid < device->generation) {
833 * That is if the FS is _not_ mounted and if you
834 * are here, that means there is more than one
835 * disk with same uuid and devid.We keep the one
836 * with larger generation number or the last-in if
837 * generation are equal.
839 return ERR_PTR(-EEXIST);
842 name = rcu_string_strdup(path, GFP_NOFS);
844 return ERR_PTR(-ENOMEM);
845 rcu_string_free(device->name);
846 rcu_assign_pointer(device->name, name);
847 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
848 fs_devices->missing_devices--;
849 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
854 * Unmount does not free the btrfs_device struct but would zero
855 * generation along with most of the other members. So just update
856 * it back. We need it to pick the disk with largest generation
859 if (!fs_devices->opened)
860 device->generation = found_transid;
862 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
867 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
869 struct btrfs_fs_devices *fs_devices;
870 struct btrfs_device *device;
871 struct btrfs_device *orig_dev;
873 fs_devices = alloc_fs_devices(orig->fsid);
874 if (IS_ERR(fs_devices))
877 mutex_lock(&orig->device_list_mutex);
878 fs_devices->total_devices = orig->total_devices;
880 /* We have held the volume lock, it is safe to get the devices. */
881 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
882 struct rcu_string *name;
884 device = btrfs_alloc_device(NULL, &orig_dev->devid,
890 * This is ok to do without rcu read locked because we hold the
891 * uuid mutex so nothing we touch in here is going to disappear.
893 if (orig_dev->name) {
894 name = rcu_string_strdup(orig_dev->name->str,
897 btrfs_free_device(device);
900 rcu_assign_pointer(device->name, name);
903 list_add(&device->dev_list, &fs_devices->devices);
904 device->fs_devices = fs_devices;
905 fs_devices->num_devices++;
907 mutex_unlock(&orig->device_list_mutex);
910 mutex_unlock(&orig->device_list_mutex);
911 free_fs_devices(fs_devices);
912 return ERR_PTR(-ENOMEM);
916 * After we have read the system tree and know devids belonging to
917 * this filesystem, remove the device which does not belong there.
919 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
921 struct btrfs_device *device, *next;
922 struct btrfs_device *latest_dev = NULL;
924 mutex_lock(&uuid_mutex);
926 /* This is the initialized path, it is safe to release the devices. */
927 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
928 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
929 &device->dev_state)) {
930 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
931 &device->dev_state) &&
933 device->generation > latest_dev->generation)) {
939 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
941 * In the first step, keep the device which has
942 * the correct fsid and the devid that is used
943 * for the dev_replace procedure.
944 * In the second step, the dev_replace state is
945 * read from the device tree and it is known
946 * whether the procedure is really active or
947 * not, which means whether this device is
948 * used or whether it should be removed.
950 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
951 &device->dev_state)) {
956 blkdev_put(device->bdev, device->mode);
958 fs_devices->open_devices--;
960 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
961 list_del_init(&device->dev_alloc_list);
962 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
963 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
965 fs_devices->rw_devices--;
967 list_del_init(&device->dev_list);
968 fs_devices->num_devices--;
969 btrfs_free_device(device);
972 if (fs_devices->seed) {
973 fs_devices = fs_devices->seed;
977 fs_devices->latest_bdev = latest_dev->bdev;
979 mutex_unlock(&uuid_mutex);
982 static void free_device_rcu(struct rcu_head *head)
984 struct btrfs_device *device;
986 device = container_of(head, struct btrfs_device, rcu);
987 btrfs_free_device(device);
990 static void btrfs_close_bdev(struct btrfs_device *device)
995 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
996 sync_blockdev(device->bdev);
997 invalidate_bdev(device->bdev);
1000 blkdev_put(device->bdev, device->mode);
1003 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
1005 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1006 struct btrfs_device *new_device;
1007 struct rcu_string *name;
1010 fs_devices->open_devices--;
1012 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1013 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1014 list_del_init(&device->dev_alloc_list);
1015 fs_devices->rw_devices--;
1018 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1019 fs_devices->missing_devices--;
1021 new_device = btrfs_alloc_device(NULL, &device->devid,
1023 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1025 /* Safe because we are under uuid_mutex */
1027 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1028 BUG_ON(!name); /* -ENOMEM */
1029 rcu_assign_pointer(new_device->name, name);
1032 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1033 new_device->fs_devices = device->fs_devices;
1036 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1038 struct btrfs_device *device, *tmp;
1039 struct list_head pending_put;
1041 INIT_LIST_HEAD(&pending_put);
1043 if (--fs_devices->opened > 0)
1046 mutex_lock(&fs_devices->device_list_mutex);
1047 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1048 btrfs_prepare_close_one_device(device);
1049 list_add(&device->dev_list, &pending_put);
1051 mutex_unlock(&fs_devices->device_list_mutex);
1054 * btrfs_show_devname() is using the device_list_mutex,
1055 * sometimes call to blkdev_put() leads vfs calling
1056 * into this func. So do put outside of device_list_mutex,
1059 while (!list_empty(&pending_put)) {
1060 device = list_first_entry(&pending_put,
1061 struct btrfs_device, dev_list);
1062 list_del(&device->dev_list);
1063 btrfs_close_bdev(device);
1064 call_rcu(&device->rcu, free_device_rcu);
1067 WARN_ON(fs_devices->open_devices);
1068 WARN_ON(fs_devices->rw_devices);
1069 fs_devices->opened = 0;
1070 fs_devices->seeding = 0;
1075 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1077 struct btrfs_fs_devices *seed_devices = NULL;
1080 mutex_lock(&uuid_mutex);
1081 ret = close_fs_devices(fs_devices);
1082 if (!fs_devices->opened) {
1083 seed_devices = fs_devices->seed;
1084 fs_devices->seed = NULL;
1086 mutex_unlock(&uuid_mutex);
1088 while (seed_devices) {
1089 fs_devices = seed_devices;
1090 seed_devices = fs_devices->seed;
1091 close_fs_devices(fs_devices);
1092 free_fs_devices(fs_devices);
1097 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1098 fmode_t flags, void *holder)
1100 struct btrfs_device *device;
1101 struct btrfs_device *latest_dev = NULL;
1104 flags |= FMODE_EXCL;
1106 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1107 /* Just open everything we can; ignore failures here */
1108 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1112 device->generation > latest_dev->generation)
1113 latest_dev = device;
1115 if (fs_devices->open_devices == 0) {
1119 fs_devices->opened = 1;
1120 fs_devices->latest_bdev = latest_dev->bdev;
1121 fs_devices->total_rw_bytes = 0;
1126 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1128 struct btrfs_device *dev1, *dev2;
1130 dev1 = list_entry(a, struct btrfs_device, dev_list);
1131 dev2 = list_entry(b, struct btrfs_device, dev_list);
1133 if (dev1->devid < dev2->devid)
1135 else if (dev1->devid > dev2->devid)
1140 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1141 fmode_t flags, void *holder)
1145 mutex_lock(&uuid_mutex);
1146 if (fs_devices->opened) {
1147 fs_devices->opened++;
1150 list_sort(NULL, &fs_devices->devices, devid_cmp);
1151 ret = open_fs_devices(fs_devices, flags, holder);
1153 mutex_unlock(&uuid_mutex);
1157 static void btrfs_release_disk_super(struct page *page)
1163 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1165 struct btrfs_super_block **disk_super)
1170 /* make sure our super fits in the device */
1171 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1174 /* make sure our super fits in the page */
1175 if (sizeof(**disk_super) > PAGE_SIZE)
1178 /* make sure our super doesn't straddle pages on disk */
1179 index = bytenr >> PAGE_SHIFT;
1180 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1183 /* pull in the page with our super */
1184 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1187 if (IS_ERR_OR_NULL(*page))
1192 /* align our pointer to the offset of the super block */
1193 *disk_super = p + (bytenr & ~PAGE_MASK);
1195 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1196 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1197 btrfs_release_disk_super(*page);
1201 if ((*disk_super)->label[0] &&
1202 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1203 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1209 * Look for a btrfs signature on a device. This may be called out of the mount path
1210 * and we are not allowed to call set_blocksize during the scan. The superblock
1211 * is read via pagecache
1213 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1214 struct btrfs_fs_devices **fs_devices_ret)
1216 struct btrfs_super_block *disk_super;
1217 struct btrfs_device *device;
1218 struct block_device *bdev;
1224 * we would like to check all the supers, but that would make
1225 * a btrfs mount succeed after a mkfs from a different FS.
1226 * So, we need to add a special mount option to scan for
1227 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1229 bytenr = btrfs_sb_offset(0);
1230 flags |= FMODE_EXCL;
1231 mutex_lock(&uuid_mutex);
1233 bdev = blkdev_get_by_path(path, flags, holder);
1235 ret = PTR_ERR(bdev);
1239 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1241 goto error_bdev_put;
1244 device = device_list_add(path, disk_super);
1246 ret = PTR_ERR(device);
1248 *fs_devices_ret = device->fs_devices;
1250 btrfs_release_disk_super(page);
1253 blkdev_put(bdev, flags);
1255 mutex_unlock(&uuid_mutex);
1259 /* helper to account the used device space in the range */
1260 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1261 u64 end, u64 *length)
1263 struct btrfs_key key;
1264 struct btrfs_root *root = device->fs_info->dev_root;
1265 struct btrfs_dev_extent *dev_extent;
1266 struct btrfs_path *path;
1270 struct extent_buffer *l;
1274 if (start >= device->total_bytes ||
1275 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1278 path = btrfs_alloc_path();
1281 path->reada = READA_FORWARD;
1283 key.objectid = device->devid;
1285 key.type = BTRFS_DEV_EXTENT_KEY;
1287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1291 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1298 slot = path->slots[0];
1299 if (slot >= btrfs_header_nritems(l)) {
1300 ret = btrfs_next_leaf(root, path);
1308 btrfs_item_key_to_cpu(l, &key, slot);
1310 if (key.objectid < device->devid)
1313 if (key.objectid > device->devid)
1316 if (key.type != BTRFS_DEV_EXTENT_KEY)
1319 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1320 extent_end = key.offset + btrfs_dev_extent_length(l,
1322 if (key.offset <= start && extent_end > end) {
1323 *length = end - start + 1;
1325 } else if (key.offset <= start && extent_end > start)
1326 *length += extent_end - start;
1327 else if (key.offset > start && extent_end <= end)
1328 *length += extent_end - key.offset;
1329 else if (key.offset > start && key.offset <= end) {
1330 *length += end - key.offset + 1;
1332 } else if (key.offset > end)
1340 btrfs_free_path(path);
1344 static int contains_pending_extent(struct btrfs_transaction *transaction,
1345 struct btrfs_device *device,
1346 u64 *start, u64 len)
1348 struct btrfs_fs_info *fs_info = device->fs_info;
1349 struct extent_map *em;
1350 struct list_head *search_list = &fs_info->pinned_chunks;
1352 u64 physical_start = *start;
1355 search_list = &transaction->pending_chunks;
1357 list_for_each_entry(em, search_list, list) {
1358 struct map_lookup *map;
1361 map = em->map_lookup;
1362 for (i = 0; i < map->num_stripes; i++) {
1365 if (map->stripes[i].dev != device)
1367 if (map->stripes[i].physical >= physical_start + len ||
1368 map->stripes[i].physical + em->orig_block_len <=
1372 * Make sure that while processing the pinned list we do
1373 * not override our *start with a lower value, because
1374 * we can have pinned chunks that fall within this
1375 * device hole and that have lower physical addresses
1376 * than the pending chunks we processed before. If we
1377 * do not take this special care we can end up getting
1378 * 2 pending chunks that start at the same physical
1379 * device offsets because the end offset of a pinned
1380 * chunk can be equal to the start offset of some
1383 end = map->stripes[i].physical + em->orig_block_len;
1390 if (search_list != &fs_info->pinned_chunks) {
1391 search_list = &fs_info->pinned_chunks;
1400 * find_free_dev_extent_start - find free space in the specified device
1401 * @device: the device which we search the free space in
1402 * @num_bytes: the size of the free space that we need
1403 * @search_start: the position from which to begin the search
1404 * @start: store the start of the free space.
1405 * @len: the size of the free space. that we find, or the size
1406 * of the max free space if we don't find suitable free space
1408 * this uses a pretty simple search, the expectation is that it is
1409 * called very infrequently and that a given device has a small number
1412 * @start is used to store the start of the free space if we find. But if we
1413 * don't find suitable free space, it will be used to store the start position
1414 * of the max free space.
1416 * @len is used to store the size of the free space that we find.
1417 * But if we don't find suitable free space, it is used to store the size of
1418 * the max free space.
1420 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1421 struct btrfs_device *device, u64 num_bytes,
1422 u64 search_start, u64 *start, u64 *len)
1424 struct btrfs_fs_info *fs_info = device->fs_info;
1425 struct btrfs_root *root = fs_info->dev_root;
1426 struct btrfs_key key;
1427 struct btrfs_dev_extent *dev_extent;
1428 struct btrfs_path *path;
1433 u64 search_end = device->total_bytes;
1436 struct extent_buffer *l;
1439 * We don't want to overwrite the superblock on the drive nor any area
1440 * used by the boot loader (grub for example), so we make sure to start
1441 * at an offset of at least 1MB.
1443 search_start = max_t(u64, search_start, SZ_1M);
1445 path = btrfs_alloc_path();
1449 max_hole_start = search_start;
1453 if (search_start >= search_end ||
1454 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1459 path->reada = READA_FORWARD;
1460 path->search_commit_root = 1;
1461 path->skip_locking = 1;
1463 key.objectid = device->devid;
1464 key.offset = search_start;
1465 key.type = BTRFS_DEV_EXTENT_KEY;
1467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1471 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1478 slot = path->slots[0];
1479 if (slot >= btrfs_header_nritems(l)) {
1480 ret = btrfs_next_leaf(root, path);
1488 btrfs_item_key_to_cpu(l, &key, slot);
1490 if (key.objectid < device->devid)
1493 if (key.objectid > device->devid)
1496 if (key.type != BTRFS_DEV_EXTENT_KEY)
1499 if (key.offset > search_start) {
1500 hole_size = key.offset - search_start;
1503 * Have to check before we set max_hole_start, otherwise
1504 * we could end up sending back this offset anyway.
1506 if (contains_pending_extent(transaction, device,
1509 if (key.offset >= search_start) {
1510 hole_size = key.offset - search_start;
1517 if (hole_size > max_hole_size) {
1518 max_hole_start = search_start;
1519 max_hole_size = hole_size;
1523 * If this free space is greater than which we need,
1524 * it must be the max free space that we have found
1525 * until now, so max_hole_start must point to the start
1526 * of this free space and the length of this free space
1527 * is stored in max_hole_size. Thus, we return
1528 * max_hole_start and max_hole_size and go back to the
1531 if (hole_size >= num_bytes) {
1537 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1538 extent_end = key.offset + btrfs_dev_extent_length(l,
1540 if (extent_end > search_start)
1541 search_start = extent_end;
1548 * At this point, search_start should be the end of
1549 * allocated dev extents, and when shrinking the device,
1550 * search_end may be smaller than search_start.
1552 if (search_end > search_start) {
1553 hole_size = search_end - search_start;
1555 if (contains_pending_extent(transaction, device, &search_start,
1557 btrfs_release_path(path);
1561 if (hole_size > max_hole_size) {
1562 max_hole_start = search_start;
1563 max_hole_size = hole_size;
1568 if (max_hole_size < num_bytes)
1574 btrfs_free_path(path);
1575 *start = max_hole_start;
1577 *len = max_hole_size;
1581 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1582 struct btrfs_device *device, u64 num_bytes,
1583 u64 *start, u64 *len)
1585 /* FIXME use last free of some kind */
1586 return find_free_dev_extent_start(trans->transaction, device,
1587 num_bytes, 0, start, len);
1590 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1591 struct btrfs_device *device,
1592 u64 start, u64 *dev_extent_len)
1594 struct btrfs_fs_info *fs_info = device->fs_info;
1595 struct btrfs_root *root = fs_info->dev_root;
1597 struct btrfs_path *path;
1598 struct btrfs_key key;
1599 struct btrfs_key found_key;
1600 struct extent_buffer *leaf = NULL;
1601 struct btrfs_dev_extent *extent = NULL;
1603 path = btrfs_alloc_path();
1607 key.objectid = device->devid;
1609 key.type = BTRFS_DEV_EXTENT_KEY;
1611 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1613 ret = btrfs_previous_item(root, path, key.objectid,
1614 BTRFS_DEV_EXTENT_KEY);
1617 leaf = path->nodes[0];
1618 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1619 extent = btrfs_item_ptr(leaf, path->slots[0],
1620 struct btrfs_dev_extent);
1621 BUG_ON(found_key.offset > start || found_key.offset +
1622 btrfs_dev_extent_length(leaf, extent) < start);
1624 btrfs_release_path(path);
1626 } else if (ret == 0) {
1627 leaf = path->nodes[0];
1628 extent = btrfs_item_ptr(leaf, path->slots[0],
1629 struct btrfs_dev_extent);
1631 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1635 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1637 ret = btrfs_del_item(trans, root, path);
1639 btrfs_handle_fs_error(fs_info, ret,
1640 "Failed to remove dev extent item");
1642 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1645 btrfs_free_path(path);
1649 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1650 struct btrfs_device *device,
1651 u64 chunk_offset, u64 start, u64 num_bytes)
1654 struct btrfs_path *path;
1655 struct btrfs_fs_info *fs_info = device->fs_info;
1656 struct btrfs_root *root = fs_info->dev_root;
1657 struct btrfs_dev_extent *extent;
1658 struct extent_buffer *leaf;
1659 struct btrfs_key key;
1661 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1662 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1663 path = btrfs_alloc_path();
1667 key.objectid = device->devid;
1669 key.type = BTRFS_DEV_EXTENT_KEY;
1670 ret = btrfs_insert_empty_item(trans, root, path, &key,
1675 leaf = path->nodes[0];
1676 extent = btrfs_item_ptr(leaf, path->slots[0],
1677 struct btrfs_dev_extent);
1678 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1679 BTRFS_CHUNK_TREE_OBJECTID);
1680 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1681 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1682 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1684 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1685 btrfs_mark_buffer_dirty(leaf);
1687 btrfs_free_path(path);
1691 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1693 struct extent_map_tree *em_tree;
1694 struct extent_map *em;
1698 em_tree = &fs_info->mapping_tree.map_tree;
1699 read_lock(&em_tree->lock);
1700 n = rb_last(&em_tree->map);
1702 em = rb_entry(n, struct extent_map, rb_node);
1703 ret = em->start + em->len;
1705 read_unlock(&em_tree->lock);
1710 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1714 struct btrfs_key key;
1715 struct btrfs_key found_key;
1716 struct btrfs_path *path;
1718 path = btrfs_alloc_path();
1722 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1723 key.type = BTRFS_DEV_ITEM_KEY;
1724 key.offset = (u64)-1;
1726 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1730 BUG_ON(ret == 0); /* Corruption */
1732 ret = btrfs_previous_item(fs_info->chunk_root, path,
1733 BTRFS_DEV_ITEMS_OBJECTID,
1734 BTRFS_DEV_ITEM_KEY);
1738 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1740 *devid_ret = found_key.offset + 1;
1744 btrfs_free_path(path);
1749 * the device information is stored in the chunk root
1750 * the btrfs_device struct should be fully filled in
1752 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1753 struct btrfs_fs_info *fs_info,
1754 struct btrfs_device *device)
1756 struct btrfs_root *root = fs_info->chunk_root;
1758 struct btrfs_path *path;
1759 struct btrfs_dev_item *dev_item;
1760 struct extent_buffer *leaf;
1761 struct btrfs_key key;
1764 path = btrfs_alloc_path();
1768 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1769 key.type = BTRFS_DEV_ITEM_KEY;
1770 key.offset = device->devid;
1772 ret = btrfs_insert_empty_item(trans, root, path, &key,
1777 leaf = path->nodes[0];
1778 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1780 btrfs_set_device_id(leaf, dev_item, device->devid);
1781 btrfs_set_device_generation(leaf, dev_item, 0);
1782 btrfs_set_device_type(leaf, dev_item, device->type);
1783 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1784 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1785 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1786 btrfs_set_device_total_bytes(leaf, dev_item,
1787 btrfs_device_get_disk_total_bytes(device));
1788 btrfs_set_device_bytes_used(leaf, dev_item,
1789 btrfs_device_get_bytes_used(device));
1790 btrfs_set_device_group(leaf, dev_item, 0);
1791 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1792 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1793 btrfs_set_device_start_offset(leaf, dev_item, 0);
1795 ptr = btrfs_device_uuid(dev_item);
1796 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1797 ptr = btrfs_device_fsid(dev_item);
1798 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1799 btrfs_mark_buffer_dirty(leaf);
1803 btrfs_free_path(path);
1808 * Function to update ctime/mtime for a given device path.
1809 * Mainly used for ctime/mtime based probe like libblkid.
1811 static void update_dev_time(const char *path_name)
1815 filp = filp_open(path_name, O_RDWR, 0);
1818 file_update_time(filp);
1819 filp_close(filp, NULL);
1822 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1823 struct btrfs_device *device)
1825 struct btrfs_root *root = fs_info->chunk_root;
1827 struct btrfs_path *path;
1828 struct btrfs_key key;
1829 struct btrfs_trans_handle *trans;
1831 path = btrfs_alloc_path();
1835 trans = btrfs_start_transaction(root, 0);
1836 if (IS_ERR(trans)) {
1837 btrfs_free_path(path);
1838 return PTR_ERR(trans);
1840 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1841 key.type = BTRFS_DEV_ITEM_KEY;
1842 key.offset = device->devid;
1844 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1848 btrfs_abort_transaction(trans, ret);
1849 btrfs_end_transaction(trans);
1853 ret = btrfs_del_item(trans, root, path);
1855 btrfs_abort_transaction(trans, ret);
1856 btrfs_end_transaction(trans);
1860 btrfs_free_path(path);
1862 ret = btrfs_commit_transaction(trans);
1867 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1868 * filesystem. It's up to the caller to adjust that number regarding eg. device
1871 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1879 seq = read_seqbegin(&fs_info->profiles_lock);
1881 all_avail = fs_info->avail_data_alloc_bits |
1882 fs_info->avail_system_alloc_bits |
1883 fs_info->avail_metadata_alloc_bits;
1884 } while (read_seqretry(&fs_info->profiles_lock, seq));
1886 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1887 if (!(all_avail & btrfs_raid_group[i]))
1890 if (num_devices < btrfs_raid_array[i].devs_min) {
1891 int ret = btrfs_raid_mindev_error[i];
1901 static struct btrfs_device * btrfs_find_next_active_device(
1902 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1904 struct btrfs_device *next_device;
1906 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1907 if (next_device != device &&
1908 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1909 && next_device->bdev)
1917 * Helper function to check if the given device is part of s_bdev / latest_bdev
1918 * and replace it with the provided or the next active device, in the context
1919 * where this function called, there should be always be another device (or
1920 * this_dev) which is active.
1922 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1923 struct btrfs_device *device, struct btrfs_device *this_dev)
1925 struct btrfs_device *next_device;
1928 next_device = this_dev;
1930 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1932 ASSERT(next_device);
1934 if (fs_info->sb->s_bdev &&
1935 (fs_info->sb->s_bdev == device->bdev))
1936 fs_info->sb->s_bdev = next_device->bdev;
1938 if (fs_info->fs_devices->latest_bdev == device->bdev)
1939 fs_info->fs_devices->latest_bdev = next_device->bdev;
1942 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1945 struct btrfs_device *device;
1946 struct btrfs_fs_devices *cur_devices;
1947 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1951 mutex_lock(&uuid_mutex);
1953 num_devices = fs_devices->num_devices;
1954 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1955 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1956 WARN_ON(num_devices < 1);
1959 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1961 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1965 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1970 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1971 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1975 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1976 fs_info->fs_devices->rw_devices == 1) {
1977 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1981 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1982 mutex_lock(&fs_info->chunk_mutex);
1983 list_del_init(&device->dev_alloc_list);
1984 device->fs_devices->rw_devices--;
1985 mutex_unlock(&fs_info->chunk_mutex);
1988 mutex_unlock(&uuid_mutex);
1989 ret = btrfs_shrink_device(device, 0);
1990 mutex_lock(&uuid_mutex);
1995 * TODO: the superblock still includes this device in its num_devices
1996 * counter although write_all_supers() is not locked out. This
1997 * could give a filesystem state which requires a degraded mount.
1999 ret = btrfs_rm_dev_item(fs_info, device);
2003 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2004 btrfs_scrub_cancel_dev(fs_info, device);
2007 * the device list mutex makes sure that we don't change
2008 * the device list while someone else is writing out all
2009 * the device supers. Whoever is writing all supers, should
2010 * lock the device list mutex before getting the number of
2011 * devices in the super block (super_copy). Conversely,
2012 * whoever updates the number of devices in the super block
2013 * (super_copy) should hold the device list mutex.
2016 cur_devices = device->fs_devices;
2017 mutex_lock(&fs_devices->device_list_mutex);
2018 list_del_rcu(&device->dev_list);
2020 device->fs_devices->num_devices--;
2021 device->fs_devices->total_devices--;
2023 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2024 device->fs_devices->missing_devices--;
2026 btrfs_assign_next_active_device(fs_info, device, NULL);
2029 device->fs_devices->open_devices--;
2030 /* remove sysfs entry */
2031 btrfs_sysfs_rm_device_link(fs_devices, device);
2034 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2035 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2036 mutex_unlock(&fs_devices->device_list_mutex);
2039 * at this point, the device is zero sized and detached from
2040 * the devices list. All that's left is to zero out the old
2041 * supers and free the device.
2043 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2044 btrfs_scratch_superblocks(device->bdev, device->name->str);
2046 btrfs_close_bdev(device);
2047 call_rcu(&device->rcu, free_device_rcu);
2049 if (cur_devices->open_devices == 0) {
2050 while (fs_devices) {
2051 if (fs_devices->seed == cur_devices) {
2052 fs_devices->seed = cur_devices->seed;
2055 fs_devices = fs_devices->seed;
2057 cur_devices->seed = NULL;
2058 close_fs_devices(cur_devices);
2059 free_fs_devices(cur_devices);
2063 mutex_unlock(&uuid_mutex);
2067 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2068 mutex_lock(&fs_info->chunk_mutex);
2069 list_add(&device->dev_alloc_list,
2070 &fs_devices->alloc_list);
2071 device->fs_devices->rw_devices++;
2072 mutex_unlock(&fs_info->chunk_mutex);
2077 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2078 struct btrfs_device *srcdev)
2080 struct btrfs_fs_devices *fs_devices;
2082 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2085 * in case of fs with no seed, srcdev->fs_devices will point
2086 * to fs_devices of fs_info. However when the dev being replaced is
2087 * a seed dev it will point to the seed's local fs_devices. In short
2088 * srcdev will have its correct fs_devices in both the cases.
2090 fs_devices = srcdev->fs_devices;
2092 list_del_rcu(&srcdev->dev_list);
2093 list_del(&srcdev->dev_alloc_list);
2094 fs_devices->num_devices--;
2095 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2096 fs_devices->missing_devices--;
2098 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2099 fs_devices->rw_devices--;
2102 fs_devices->open_devices--;
2105 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2106 struct btrfs_device *srcdev)
2108 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2110 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2111 /* zero out the old super if it is writable */
2112 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2115 btrfs_close_bdev(srcdev);
2116 call_rcu(&srcdev->rcu, free_device_rcu);
2118 /* if this is no devs we rather delete the fs_devices */
2119 if (!fs_devices->num_devices) {
2120 struct btrfs_fs_devices *tmp_fs_devices;
2123 * On a mounted FS, num_devices can't be zero unless it's a
2124 * seed. In case of a seed device being replaced, the replace
2125 * target added to the sprout FS, so there will be no more
2126 * device left under the seed FS.
2128 ASSERT(fs_devices->seeding);
2130 tmp_fs_devices = fs_info->fs_devices;
2131 while (tmp_fs_devices) {
2132 if (tmp_fs_devices->seed == fs_devices) {
2133 tmp_fs_devices->seed = fs_devices->seed;
2136 tmp_fs_devices = tmp_fs_devices->seed;
2138 fs_devices->seed = NULL;
2139 close_fs_devices(fs_devices);
2140 free_fs_devices(fs_devices);
2144 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2145 struct btrfs_device *tgtdev)
2147 mutex_lock(&uuid_mutex);
2149 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2151 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2154 fs_info->fs_devices->open_devices--;
2156 fs_info->fs_devices->num_devices--;
2158 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2160 list_del_rcu(&tgtdev->dev_list);
2162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2163 mutex_unlock(&uuid_mutex);
2166 * The update_dev_time() with in btrfs_scratch_superblocks()
2167 * may lead to a call to btrfs_show_devname() which will try
2168 * to hold device_list_mutex. And here this device
2169 * is already out of device list, so we don't have to hold
2170 * the device_list_mutex lock.
2172 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2174 btrfs_close_bdev(tgtdev);
2175 call_rcu(&tgtdev->rcu, free_device_rcu);
2178 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2179 const char *device_path,
2180 struct btrfs_device **device)
2183 struct btrfs_super_block *disk_super;
2186 struct block_device *bdev;
2187 struct buffer_head *bh;
2190 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2191 fs_info->bdev_holder, 0, &bdev, &bh);
2194 disk_super = (struct btrfs_super_block *)bh->b_data;
2195 devid = btrfs_stack_device_id(&disk_super->dev_item);
2196 dev_uuid = disk_super->dev_item.uuid;
2197 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2201 blkdev_put(bdev, FMODE_READ);
2205 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2206 const char *device_path,
2207 struct btrfs_device **device)
2210 if (strcmp(device_path, "missing") == 0) {
2211 struct list_head *devices;
2212 struct btrfs_device *tmp;
2214 devices = &fs_info->fs_devices->devices;
2215 list_for_each_entry(tmp, devices, dev_list) {
2216 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2217 &tmp->dev_state) && !tmp->bdev) {
2224 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2228 return btrfs_find_device_by_path(fs_info, device_path, device);
2233 * Lookup a device given by device id, or the path if the id is 0.
2235 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2236 const char *devpath,
2237 struct btrfs_device **device)
2243 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2247 if (!devpath || !devpath[0])
2250 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2257 * does all the dirty work required for changing file system's UUID.
2259 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2261 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2262 struct btrfs_fs_devices *old_devices;
2263 struct btrfs_fs_devices *seed_devices;
2264 struct btrfs_super_block *disk_super = fs_info->super_copy;
2265 struct btrfs_device *device;
2268 lockdep_assert_held(&uuid_mutex);
2269 if (!fs_devices->seeding)
2272 seed_devices = alloc_fs_devices(NULL);
2273 if (IS_ERR(seed_devices))
2274 return PTR_ERR(seed_devices);
2276 old_devices = clone_fs_devices(fs_devices);
2277 if (IS_ERR(old_devices)) {
2278 kfree(seed_devices);
2279 return PTR_ERR(old_devices);
2282 list_add(&old_devices->fs_list, &fs_uuids);
2284 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2285 seed_devices->opened = 1;
2286 INIT_LIST_HEAD(&seed_devices->devices);
2287 INIT_LIST_HEAD(&seed_devices->alloc_list);
2288 mutex_init(&seed_devices->device_list_mutex);
2290 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2291 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2293 list_for_each_entry(device, &seed_devices->devices, dev_list)
2294 device->fs_devices = seed_devices;
2296 mutex_lock(&fs_info->chunk_mutex);
2297 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2298 mutex_unlock(&fs_info->chunk_mutex);
2300 fs_devices->seeding = 0;
2301 fs_devices->num_devices = 0;
2302 fs_devices->open_devices = 0;
2303 fs_devices->missing_devices = 0;
2304 fs_devices->rotating = 0;
2305 fs_devices->seed = seed_devices;
2307 generate_random_uuid(fs_devices->fsid);
2308 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2309 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2310 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2312 super_flags = btrfs_super_flags(disk_super) &
2313 ~BTRFS_SUPER_FLAG_SEEDING;
2314 btrfs_set_super_flags(disk_super, super_flags);
2320 * Store the expected generation for seed devices in device items.
2322 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2323 struct btrfs_fs_info *fs_info)
2325 struct btrfs_root *root = fs_info->chunk_root;
2326 struct btrfs_path *path;
2327 struct extent_buffer *leaf;
2328 struct btrfs_dev_item *dev_item;
2329 struct btrfs_device *device;
2330 struct btrfs_key key;
2331 u8 fs_uuid[BTRFS_FSID_SIZE];
2332 u8 dev_uuid[BTRFS_UUID_SIZE];
2336 path = btrfs_alloc_path();
2340 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2342 key.type = BTRFS_DEV_ITEM_KEY;
2345 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2349 leaf = path->nodes[0];
2351 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2352 ret = btrfs_next_leaf(root, path);
2357 leaf = path->nodes[0];
2358 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2359 btrfs_release_path(path);
2363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2364 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2365 key.type != BTRFS_DEV_ITEM_KEY)
2368 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2369 struct btrfs_dev_item);
2370 devid = btrfs_device_id(leaf, dev_item);
2371 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2373 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2375 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2376 BUG_ON(!device); /* Logic error */
2378 if (device->fs_devices->seeding) {
2379 btrfs_set_device_generation(leaf, dev_item,
2380 device->generation);
2381 btrfs_mark_buffer_dirty(leaf);
2389 btrfs_free_path(path);
2393 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2395 struct btrfs_root *root = fs_info->dev_root;
2396 struct request_queue *q;
2397 struct btrfs_trans_handle *trans;
2398 struct btrfs_device *device;
2399 struct block_device *bdev;
2400 struct list_head *devices;
2401 struct super_block *sb = fs_info->sb;
2402 struct rcu_string *name;
2404 int seeding_dev = 0;
2406 bool unlocked = false;
2408 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2411 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2412 fs_info->bdev_holder);
2414 return PTR_ERR(bdev);
2416 if (fs_info->fs_devices->seeding) {
2418 down_write(&sb->s_umount);
2419 mutex_lock(&uuid_mutex);
2422 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2424 devices = &fs_info->fs_devices->devices;
2426 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2427 list_for_each_entry(device, devices, dev_list) {
2428 if (device->bdev == bdev) {
2431 &fs_info->fs_devices->device_list_mutex);
2435 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2437 device = btrfs_alloc_device(fs_info, NULL, NULL);
2438 if (IS_ERR(device)) {
2439 /* we can safely leave the fs_devices entry around */
2440 ret = PTR_ERR(device);
2444 name = rcu_string_strdup(device_path, GFP_KERNEL);
2447 goto error_free_device;
2449 rcu_assign_pointer(device->name, name);
2451 trans = btrfs_start_transaction(root, 0);
2452 if (IS_ERR(trans)) {
2453 ret = PTR_ERR(trans);
2454 goto error_free_device;
2457 q = bdev_get_queue(bdev);
2458 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2459 device->generation = trans->transid;
2460 device->io_width = fs_info->sectorsize;
2461 device->io_align = fs_info->sectorsize;
2462 device->sector_size = fs_info->sectorsize;
2463 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2464 fs_info->sectorsize);
2465 device->disk_total_bytes = device->total_bytes;
2466 device->commit_total_bytes = device->total_bytes;
2467 device->fs_info = fs_info;
2468 device->bdev = bdev;
2469 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2470 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2471 device->mode = FMODE_EXCL;
2472 device->dev_stats_valid = 1;
2473 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2476 sb->s_flags &= ~SB_RDONLY;
2477 ret = btrfs_prepare_sprout(fs_info);
2479 btrfs_abort_transaction(trans, ret);
2484 device->fs_devices = fs_info->fs_devices;
2486 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2487 mutex_lock(&fs_info->chunk_mutex);
2488 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2489 list_add(&device->dev_alloc_list,
2490 &fs_info->fs_devices->alloc_list);
2491 fs_info->fs_devices->num_devices++;
2492 fs_info->fs_devices->open_devices++;
2493 fs_info->fs_devices->rw_devices++;
2494 fs_info->fs_devices->total_devices++;
2495 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2497 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2499 if (!blk_queue_nonrot(q))
2500 fs_info->fs_devices->rotating = 1;
2502 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2503 btrfs_set_super_total_bytes(fs_info->super_copy,
2504 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2506 tmp = btrfs_super_num_devices(fs_info->super_copy);
2507 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2509 /* add sysfs device entry */
2510 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2513 * we've got more storage, clear any full flags on the space
2516 btrfs_clear_space_info_full(fs_info);
2518 mutex_unlock(&fs_info->chunk_mutex);
2519 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2522 mutex_lock(&fs_info->chunk_mutex);
2523 ret = init_first_rw_device(trans, fs_info);
2524 mutex_unlock(&fs_info->chunk_mutex);
2526 btrfs_abort_transaction(trans, ret);
2531 ret = btrfs_add_dev_item(trans, fs_info, device);
2533 btrfs_abort_transaction(trans, ret);
2538 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2540 ret = btrfs_finish_sprout(trans, fs_info);
2542 btrfs_abort_transaction(trans, ret);
2546 /* Sprouting would change fsid of the mounted root,
2547 * so rename the fsid on the sysfs
2549 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2551 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2553 "sysfs: failed to create fsid for sprout");
2556 ret = btrfs_commit_transaction(trans);
2559 mutex_unlock(&uuid_mutex);
2560 up_write(&sb->s_umount);
2563 if (ret) /* transaction commit */
2566 ret = btrfs_relocate_sys_chunks(fs_info);
2568 btrfs_handle_fs_error(fs_info, ret,
2569 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2570 trans = btrfs_attach_transaction(root);
2571 if (IS_ERR(trans)) {
2572 if (PTR_ERR(trans) == -ENOENT)
2574 ret = PTR_ERR(trans);
2578 ret = btrfs_commit_transaction(trans);
2581 /* Update ctime/mtime for libblkid */
2582 update_dev_time(device_path);
2586 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2589 sb->s_flags |= SB_RDONLY;
2591 btrfs_end_transaction(trans);
2593 btrfs_free_device(device);
2595 blkdev_put(bdev, FMODE_EXCL);
2596 if (seeding_dev && !unlocked) {
2597 mutex_unlock(&uuid_mutex);
2598 up_write(&sb->s_umount);
2603 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2604 struct btrfs_device *device)
2607 struct btrfs_path *path;
2608 struct btrfs_root *root = device->fs_info->chunk_root;
2609 struct btrfs_dev_item *dev_item;
2610 struct extent_buffer *leaf;
2611 struct btrfs_key key;
2613 path = btrfs_alloc_path();
2617 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2618 key.type = BTRFS_DEV_ITEM_KEY;
2619 key.offset = device->devid;
2621 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2630 leaf = path->nodes[0];
2631 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2633 btrfs_set_device_id(leaf, dev_item, device->devid);
2634 btrfs_set_device_type(leaf, dev_item, device->type);
2635 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2636 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2637 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2638 btrfs_set_device_total_bytes(leaf, dev_item,
2639 btrfs_device_get_disk_total_bytes(device));
2640 btrfs_set_device_bytes_used(leaf, dev_item,
2641 btrfs_device_get_bytes_used(device));
2642 btrfs_mark_buffer_dirty(leaf);
2645 btrfs_free_path(path);
2649 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2650 struct btrfs_device *device, u64 new_size)
2652 struct btrfs_fs_info *fs_info = device->fs_info;
2653 struct btrfs_super_block *super_copy = fs_info->super_copy;
2654 struct btrfs_fs_devices *fs_devices;
2658 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2661 new_size = round_down(new_size, fs_info->sectorsize);
2663 mutex_lock(&fs_info->chunk_mutex);
2664 old_total = btrfs_super_total_bytes(super_copy);
2665 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2667 if (new_size <= device->total_bytes ||
2668 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2669 mutex_unlock(&fs_info->chunk_mutex);
2673 fs_devices = fs_info->fs_devices;
2675 btrfs_set_super_total_bytes(super_copy,
2676 round_down(old_total + diff, fs_info->sectorsize));
2677 device->fs_devices->total_rw_bytes += diff;
2679 btrfs_device_set_total_bytes(device, new_size);
2680 btrfs_device_set_disk_total_bytes(device, new_size);
2681 btrfs_clear_space_info_full(device->fs_info);
2682 if (list_empty(&device->resized_list))
2683 list_add_tail(&device->resized_list,
2684 &fs_devices->resized_devices);
2685 mutex_unlock(&fs_info->chunk_mutex);
2687 return btrfs_update_device(trans, device);
2690 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2691 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2693 struct btrfs_root *root = fs_info->chunk_root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
2698 path = btrfs_alloc_path();
2702 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2703 key.offset = chunk_offset;
2704 key.type = BTRFS_CHUNK_ITEM_KEY;
2706 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2709 else if (ret > 0) { /* Logic error or corruption */
2710 btrfs_handle_fs_error(fs_info, -ENOENT,
2711 "Failed lookup while freeing chunk.");
2716 ret = btrfs_del_item(trans, root, path);
2718 btrfs_handle_fs_error(fs_info, ret,
2719 "Failed to delete chunk item.");
2721 btrfs_free_path(path);
2725 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2727 struct btrfs_super_block *super_copy = fs_info->super_copy;
2728 struct btrfs_disk_key *disk_key;
2729 struct btrfs_chunk *chunk;
2736 struct btrfs_key key;
2738 mutex_lock(&fs_info->chunk_mutex);
2739 array_size = btrfs_super_sys_array_size(super_copy);
2741 ptr = super_copy->sys_chunk_array;
2744 while (cur < array_size) {
2745 disk_key = (struct btrfs_disk_key *)ptr;
2746 btrfs_disk_key_to_cpu(&key, disk_key);
2748 len = sizeof(*disk_key);
2750 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2751 chunk = (struct btrfs_chunk *)(ptr + len);
2752 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2753 len += btrfs_chunk_item_size(num_stripes);
2758 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2759 key.offset == chunk_offset) {
2760 memmove(ptr, ptr + len, array_size - (cur + len));
2762 btrfs_set_super_sys_array_size(super_copy, array_size);
2768 mutex_unlock(&fs_info->chunk_mutex);
2772 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2773 u64 logical, u64 length)
2775 struct extent_map_tree *em_tree;
2776 struct extent_map *em;
2778 em_tree = &fs_info->mapping_tree.map_tree;
2779 read_lock(&em_tree->lock);
2780 em = lookup_extent_mapping(em_tree, logical, length);
2781 read_unlock(&em_tree->lock);
2784 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2786 return ERR_PTR(-EINVAL);
2789 if (em->start > logical || em->start + em->len < logical) {
2791 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2792 logical, length, em->start, em->start + em->len);
2793 free_extent_map(em);
2794 return ERR_PTR(-EINVAL);
2797 /* callers are responsible for dropping em's ref. */
2801 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2802 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2804 struct extent_map *em;
2805 struct map_lookup *map;
2806 u64 dev_extent_len = 0;
2808 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2810 em = get_chunk_map(fs_info, chunk_offset, 1);
2813 * This is a logic error, but we don't want to just rely on the
2814 * user having built with ASSERT enabled, so if ASSERT doesn't
2815 * do anything we still error out.
2820 map = em->map_lookup;
2821 mutex_lock(&fs_info->chunk_mutex);
2822 check_system_chunk(trans, fs_info, map->type);
2823 mutex_unlock(&fs_info->chunk_mutex);
2826 * Take the device list mutex to prevent races with the final phase of
2827 * a device replace operation that replaces the device object associated
2828 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2830 mutex_lock(&fs_devices->device_list_mutex);
2831 for (i = 0; i < map->num_stripes; i++) {
2832 struct btrfs_device *device = map->stripes[i].dev;
2833 ret = btrfs_free_dev_extent(trans, device,
2834 map->stripes[i].physical,
2837 mutex_unlock(&fs_devices->device_list_mutex);
2838 btrfs_abort_transaction(trans, ret);
2842 if (device->bytes_used > 0) {
2843 mutex_lock(&fs_info->chunk_mutex);
2844 btrfs_device_set_bytes_used(device,
2845 device->bytes_used - dev_extent_len);
2846 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2847 btrfs_clear_space_info_full(fs_info);
2848 mutex_unlock(&fs_info->chunk_mutex);
2851 if (map->stripes[i].dev) {
2852 ret = btrfs_update_device(trans, map->stripes[i].dev);
2854 mutex_unlock(&fs_devices->device_list_mutex);
2855 btrfs_abort_transaction(trans, ret);
2860 mutex_unlock(&fs_devices->device_list_mutex);
2862 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2864 btrfs_abort_transaction(trans, ret);
2868 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2870 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2871 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2873 btrfs_abort_transaction(trans, ret);
2878 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2880 btrfs_abort_transaction(trans, ret);
2886 free_extent_map(em);
2890 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2892 struct btrfs_root *root = fs_info->chunk_root;
2893 struct btrfs_trans_handle *trans;
2897 * Prevent races with automatic removal of unused block groups.
2898 * After we relocate and before we remove the chunk with offset
2899 * chunk_offset, automatic removal of the block group can kick in,
2900 * resulting in a failure when calling btrfs_remove_chunk() below.
2902 * Make sure to acquire this mutex before doing a tree search (dev
2903 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2904 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2905 * we release the path used to search the chunk/dev tree and before
2906 * the current task acquires this mutex and calls us.
2908 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2910 ret = btrfs_can_relocate(fs_info, chunk_offset);
2914 /* step one, relocate all the extents inside this chunk */
2915 btrfs_scrub_pause(fs_info);
2916 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2917 btrfs_scrub_continue(fs_info);
2922 * We add the kobjects here (and after forcing data chunk creation)
2923 * since relocation is the only place we'll create chunks of a new
2924 * type at runtime. The only place where we'll remove the last
2925 * chunk of a type is the call immediately below this one. Even
2926 * so, we're protected against races with the cleaner thread since
2927 * we're covered by the delete_unused_bgs_mutex.
2929 btrfs_add_raid_kobjects(fs_info);
2931 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2933 if (IS_ERR(trans)) {
2934 ret = PTR_ERR(trans);
2935 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2940 * step two, delete the device extents and the
2941 * chunk tree entries
2943 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2944 btrfs_end_transaction(trans);
2948 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2950 struct btrfs_root *chunk_root = fs_info->chunk_root;
2951 struct btrfs_path *path;
2952 struct extent_buffer *leaf;
2953 struct btrfs_chunk *chunk;
2954 struct btrfs_key key;
2955 struct btrfs_key found_key;
2957 bool retried = false;
2961 path = btrfs_alloc_path();
2966 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967 key.offset = (u64)-1;
2968 key.type = BTRFS_CHUNK_ITEM_KEY;
2971 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2972 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2974 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2977 BUG_ON(ret == 0); /* Corruption */
2979 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2982 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2988 leaf = path->nodes[0];
2989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2991 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992 struct btrfs_chunk);
2993 chunk_type = btrfs_chunk_type(leaf, chunk);
2994 btrfs_release_path(path);
2996 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3003 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3005 if (found_key.offset == 0)
3007 key.offset = found_key.offset - 1;
3010 if (failed && !retried) {
3014 } else if (WARN_ON(failed && retried)) {
3018 btrfs_free_path(path);
3023 * return 1 : allocate a data chunk successfully,
3024 * return <0: errors during allocating a data chunk,
3025 * return 0 : no need to allocate a data chunk.
3027 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3030 struct btrfs_block_group_cache *cache;
3034 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3036 chunk_type = cache->flags;
3037 btrfs_put_block_group(cache);
3039 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3040 spin_lock(&fs_info->data_sinfo->lock);
3041 bytes_used = fs_info->data_sinfo->bytes_used;
3042 spin_unlock(&fs_info->data_sinfo->lock);
3045 struct btrfs_trans_handle *trans;
3048 trans = btrfs_join_transaction(fs_info->tree_root);
3050 return PTR_ERR(trans);
3052 ret = btrfs_force_chunk_alloc(trans, fs_info,
3053 BTRFS_BLOCK_GROUP_DATA);
3054 btrfs_end_transaction(trans);
3058 btrfs_add_raid_kobjects(fs_info);
3066 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3067 struct btrfs_balance_control *bctl)
3069 struct btrfs_root *root = fs_info->tree_root;
3070 struct btrfs_trans_handle *trans;
3071 struct btrfs_balance_item *item;
3072 struct btrfs_disk_balance_args disk_bargs;
3073 struct btrfs_path *path;
3074 struct extent_buffer *leaf;
3075 struct btrfs_key key;
3078 path = btrfs_alloc_path();
3082 trans = btrfs_start_transaction(root, 0);
3083 if (IS_ERR(trans)) {
3084 btrfs_free_path(path);
3085 return PTR_ERR(trans);
3088 key.objectid = BTRFS_BALANCE_OBJECTID;
3089 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3092 ret = btrfs_insert_empty_item(trans, root, path, &key,
3097 leaf = path->nodes[0];
3098 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3100 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3102 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3103 btrfs_set_balance_data(leaf, item, &disk_bargs);
3104 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3105 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3106 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3107 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3109 btrfs_set_balance_flags(leaf, item, bctl->flags);
3111 btrfs_mark_buffer_dirty(leaf);
3113 btrfs_free_path(path);
3114 err = btrfs_commit_transaction(trans);
3120 static int del_balance_item(struct btrfs_fs_info *fs_info)
3122 struct btrfs_root *root = fs_info->tree_root;
3123 struct btrfs_trans_handle *trans;
3124 struct btrfs_path *path;
3125 struct btrfs_key key;
3128 path = btrfs_alloc_path();
3132 trans = btrfs_start_transaction(root, 0);
3133 if (IS_ERR(trans)) {
3134 btrfs_free_path(path);
3135 return PTR_ERR(trans);
3138 key.objectid = BTRFS_BALANCE_OBJECTID;
3139 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3142 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3150 ret = btrfs_del_item(trans, root, path);
3152 btrfs_free_path(path);
3153 err = btrfs_commit_transaction(trans);
3160 * This is a heuristic used to reduce the number of chunks balanced on
3161 * resume after balance was interrupted.
3163 static void update_balance_args(struct btrfs_balance_control *bctl)
3166 * Turn on soft mode for chunk types that were being converted.
3168 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3169 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3170 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3171 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3172 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3173 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3176 * Turn on usage filter if is not already used. The idea is
3177 * that chunks that we have already balanced should be
3178 * reasonably full. Don't do it for chunks that are being
3179 * converted - that will keep us from relocating unconverted
3180 * (albeit full) chunks.
3182 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3183 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3184 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3185 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3186 bctl->data.usage = 90;
3188 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3189 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3190 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3191 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3192 bctl->sys.usage = 90;
3194 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3195 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3196 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3197 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3198 bctl->meta.usage = 90;
3203 * Clear the balance status in fs_info and delete the balance item from disk.
3205 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3207 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3210 BUG_ON(!fs_info->balance_ctl);
3212 spin_lock(&fs_info->balance_lock);
3213 fs_info->balance_ctl = NULL;
3214 spin_unlock(&fs_info->balance_lock);
3217 ret = del_balance_item(fs_info);
3219 btrfs_handle_fs_error(fs_info, ret, NULL);
3223 * Balance filters. Return 1 if chunk should be filtered out
3224 * (should not be balanced).
3226 static int chunk_profiles_filter(u64 chunk_type,
3227 struct btrfs_balance_args *bargs)
3229 chunk_type = chunk_to_extended(chunk_type) &
3230 BTRFS_EXTENDED_PROFILE_MASK;
3232 if (bargs->profiles & chunk_type)
3238 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3239 struct btrfs_balance_args *bargs)
3241 struct btrfs_block_group_cache *cache;
3243 u64 user_thresh_min;
3244 u64 user_thresh_max;
3247 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3248 chunk_used = btrfs_block_group_used(&cache->item);