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/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
31 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
32 [BTRFS_RAID_RAID10] = {
35 .devs_max = 0, /* 0 == as many as possible */
37 .tolerated_failures = 1,
40 .raid_name = "raid10",
41 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
42 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
44 [BTRFS_RAID_RAID1] = {
49 .tolerated_failures = 1,
53 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
54 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
61 .tolerated_failures = 0,
65 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
68 [BTRFS_RAID_RAID0] = {
73 .tolerated_failures = 0,
77 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
80 [BTRFS_RAID_SINGLE] = {
85 .tolerated_failures = 0,
88 .raid_name = "single",
92 [BTRFS_RAID_RAID5] = {
97 .tolerated_failures = 1,
100 .raid_name = "raid5",
101 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
102 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
104 [BTRFS_RAID_RAID6] = {
109 .tolerated_failures = 2,
112 .raid_name = "raid6",
113 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
114 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
118 const char *get_raid_name(enum btrfs_raid_types type)
120 if (type >= BTRFS_NR_RAID_TYPES)
123 return btrfs_raid_array[type].raid_name;
126 static int init_first_rw_device(struct btrfs_trans_handle *trans,
127 struct btrfs_fs_info *fs_info);
128 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
129 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
130 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
131 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
132 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
133 enum btrfs_map_op op,
134 u64 logical, u64 *length,
135 struct btrfs_bio **bbio_ret,
136 int mirror_num, int need_raid_map);
142 * There are several mutexes that protect manipulation of devices and low-level
143 * structures like chunks but not block groups, extents or files
145 * uuid_mutex (global lock)
146 * ------------------------
147 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
148 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
149 * device) or requested by the device= mount option
151 * the mutex can be very coarse and can cover long-running operations
153 * protects: updates to fs_devices counters like missing devices, rw devices,
154 * seeding, structure cloning, openning/closing devices at mount/umount time
156 * global::fs_devs - add, remove, updates to the global list
158 * does not protect: manipulation of the fs_devices::devices list!
160 * btrfs_device::name - renames (write side), read is RCU
162 * fs_devices::device_list_mutex (per-fs, with RCU)
163 * ------------------------------------------------
164 * protects updates to fs_devices::devices, ie. adding and deleting
166 * simple list traversal with read-only actions can be done with RCU protection
168 * may be used to exclude some operations from running concurrently without any
169 * modifications to the list (see write_all_supers)
173 * protects balance structures (status, state) and context accessed from
174 * several places (internally, ioctl)
178 * protects chunks, adding or removing during allocation, trim or when a new
179 * device is added/removed
183 * a big lock that is held by the cleaner thread and prevents running subvolume
184 * cleaning together with relocation or delayed iputs
197 * Exclusive operations, BTRFS_FS_EXCL_OP
198 * ======================================
200 * Maintains the exclusivity of the following operations that apply to the
201 * whole filesystem and cannot run in parallel.
206 * - Device replace (*)
209 * The device operations (as above) can be in one of the following states:
215 * Only device operations marked with (*) can go into the Paused state for the
218 * - ioctl (only Balance can be Paused through ioctl)
219 * - filesystem remounted as read-only
220 * - filesystem unmounted and mounted as read-only
221 * - system power-cycle and filesystem mounted as read-only
222 * - filesystem or device errors leading to forced read-only
224 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
225 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
226 * A device operation in Paused or Running state can be canceled or resumed
227 * either by ioctl (Balance only) or when remounted as read-write.
228 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
232 DEFINE_MUTEX(uuid_mutex);
233 static LIST_HEAD(fs_uuids);
234 struct list_head *btrfs_get_fs_uuids(void)
240 * alloc_fs_devices - allocate struct btrfs_fs_devices
241 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
243 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
244 * The returned struct is not linked onto any lists and can be destroyed with
245 * kfree() right away.
247 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
249 struct btrfs_fs_devices *fs_devs;
251 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
253 return ERR_PTR(-ENOMEM);
255 mutex_init(&fs_devs->device_list_mutex);
257 INIT_LIST_HEAD(&fs_devs->devices);
258 INIT_LIST_HEAD(&fs_devs->resized_devices);
259 INIT_LIST_HEAD(&fs_devs->alloc_list);
260 INIT_LIST_HEAD(&fs_devs->fs_list);
262 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
267 void btrfs_free_device(struct btrfs_device *device)
269 rcu_string_free(device->name);
270 bio_put(device->flush_bio);
274 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
276 struct btrfs_device *device;
277 WARN_ON(fs_devices->opened);
278 while (!list_empty(&fs_devices->devices)) {
279 device = list_entry(fs_devices->devices.next,
280 struct btrfs_device, dev_list);
281 list_del(&device->dev_list);
282 btrfs_free_device(device);
287 static void btrfs_kobject_uevent(struct block_device *bdev,
288 enum kobject_action action)
292 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
294 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
296 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
297 &disk_to_dev(bdev->bd_disk)->kobj);
300 void __exit btrfs_cleanup_fs_uuids(void)
302 struct btrfs_fs_devices *fs_devices;
304 while (!list_empty(&fs_uuids)) {
305 fs_devices = list_entry(fs_uuids.next,
306 struct btrfs_fs_devices, fs_list);
307 list_del(&fs_devices->fs_list);
308 free_fs_devices(fs_devices);
313 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
314 * Returned struct is not linked onto any lists and must be destroyed using
317 static struct btrfs_device *__alloc_device(void)
319 struct btrfs_device *dev;
321 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
323 return ERR_PTR(-ENOMEM);
326 * Preallocate a bio that's always going to be used for flushing device
327 * barriers and matches the device lifespan
329 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
330 if (!dev->flush_bio) {
332 return ERR_PTR(-ENOMEM);
335 INIT_LIST_HEAD(&dev->dev_list);
336 INIT_LIST_HEAD(&dev->dev_alloc_list);
337 INIT_LIST_HEAD(&dev->resized_list);
339 spin_lock_init(&dev->io_lock);
341 atomic_set(&dev->reada_in_flight, 0);
342 atomic_set(&dev->dev_stats_ccnt, 0);
343 btrfs_device_data_ordered_init(dev);
344 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
345 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
351 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
354 * If devid and uuid are both specified, the match must be exact, otherwise
355 * only devid is used.
357 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
358 u64 devid, const u8 *uuid)
360 struct btrfs_device *dev;
362 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
363 if (dev->devid == devid &&
364 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
371 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
373 struct btrfs_fs_devices *fs_devices;
375 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
376 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
383 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
384 int flush, struct block_device **bdev,
385 struct buffer_head **bh)
389 *bdev = blkdev_get_by_path(device_path, flags, holder);
392 ret = PTR_ERR(*bdev);
397 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
398 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
400 blkdev_put(*bdev, flags);
403 invalidate_bdev(*bdev);
404 *bh = btrfs_read_dev_super(*bdev);
407 blkdev_put(*bdev, flags);
419 static void requeue_list(struct btrfs_pending_bios *pending_bios,
420 struct bio *head, struct bio *tail)
423 struct bio *old_head;
425 old_head = pending_bios->head;
426 pending_bios->head = head;
427 if (pending_bios->tail)
428 tail->bi_next = old_head;
430 pending_bios->tail = tail;
434 * we try to collect pending bios for a device so we don't get a large
435 * number of procs sending bios down to the same device. This greatly
436 * improves the schedulers ability to collect and merge the bios.
438 * But, it also turns into a long list of bios to process and that is sure
439 * to eventually make the worker thread block. The solution here is to
440 * make some progress and then put this work struct back at the end of
441 * the list if the block device is congested. This way, multiple devices
442 * can make progress from a single worker thread.
444 static noinline void run_scheduled_bios(struct btrfs_device *device)
446 struct btrfs_fs_info *fs_info = device->fs_info;
448 struct backing_dev_info *bdi;
449 struct btrfs_pending_bios *pending_bios;
453 unsigned long num_run;
454 unsigned long batch_run = 0;
455 unsigned long last_waited = 0;
457 int sync_pending = 0;
458 struct blk_plug plug;
461 * this function runs all the bios we've collected for
462 * a particular device. We don't want to wander off to
463 * another device without first sending all of these down.
464 * So, setup a plug here and finish it off before we return
466 blk_start_plug(&plug);
468 bdi = device->bdev->bd_bdi;
471 spin_lock(&device->io_lock);
476 /* take all the bios off the list at once and process them
477 * later on (without the lock held). But, remember the
478 * tail and other pointers so the bios can be properly reinserted
479 * into the list if we hit congestion
481 if (!force_reg && device->pending_sync_bios.head) {
482 pending_bios = &device->pending_sync_bios;
485 pending_bios = &device->pending_bios;
489 pending = pending_bios->head;
490 tail = pending_bios->tail;
491 WARN_ON(pending && !tail);
494 * if pending was null this time around, no bios need processing
495 * at all and we can stop. Otherwise it'll loop back up again
496 * and do an additional check so no bios are missed.
498 * device->running_pending is used to synchronize with the
501 if (device->pending_sync_bios.head == NULL &&
502 device->pending_bios.head == NULL) {
504 device->running_pending = 0;
507 device->running_pending = 1;
510 pending_bios->head = NULL;
511 pending_bios->tail = NULL;
513 spin_unlock(&device->io_lock);
518 /* we want to work on both lists, but do more bios on the
519 * sync list than the regular list
522 pending_bios != &device->pending_sync_bios &&
523 device->pending_sync_bios.head) ||
524 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
525 device->pending_bios.head)) {
526 spin_lock(&device->io_lock);
527 requeue_list(pending_bios, pending, tail);
532 pending = pending->bi_next;
535 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
538 * if we're doing the sync list, record that our
539 * plug has some sync requests on it
541 * If we're doing the regular list and there are
542 * sync requests sitting around, unplug before
545 if (pending_bios == &device->pending_sync_bios) {
547 } else if (sync_pending) {
548 blk_finish_plug(&plug);
549 blk_start_plug(&plug);
553 btrfsic_submit_bio(cur);
560 * we made progress, there is more work to do and the bdi
561 * is now congested. Back off and let other work structs
564 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
565 fs_info->fs_devices->open_devices > 1) {
566 struct io_context *ioc;
568 ioc = current->io_context;
571 * the main goal here is that we don't want to
572 * block if we're going to be able to submit
573 * more requests without blocking.
575 * This code does two great things, it pokes into
576 * the elevator code from a filesystem _and_
577 * it makes assumptions about how batching works.
579 if (ioc && ioc->nr_batch_requests > 0 &&
580 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
582 ioc->last_waited == last_waited)) {
584 * we want to go through our batch of
585 * requests and stop. So, we copy out
586 * the ioc->last_waited time and test
587 * against it before looping
589 last_waited = ioc->last_waited;
593 spin_lock(&device->io_lock);
594 requeue_list(pending_bios, pending, tail);
595 device->running_pending = 1;
597 spin_unlock(&device->io_lock);
598 btrfs_queue_work(fs_info->submit_workers,
608 spin_lock(&device->io_lock);
609 if (device->pending_bios.head || device->pending_sync_bios.head)
611 spin_unlock(&device->io_lock);
614 blk_finish_plug(&plug);
617 static void pending_bios_fn(struct btrfs_work *work)
619 struct btrfs_device *device;
621 device = container_of(work, struct btrfs_device, work);
622 run_scheduled_bios(device);
626 * Search and remove all stale (devices which are not mounted) devices.
627 * When both inputs are NULL, it will search and release all stale devices.
628 * path: Optional. When provided will it release all unmounted devices
629 * matching this path only.
630 * skip_dev: Optional. Will skip this device when searching for the stale
633 static void btrfs_free_stale_devices(const char *path,
634 struct btrfs_device *skip_device)
636 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
637 struct btrfs_device *device, *tmp_device;
639 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
640 mutex_lock(&fs_devices->device_list_mutex);
641 if (fs_devices->opened) {
642 mutex_unlock(&fs_devices->device_list_mutex);
646 list_for_each_entry_safe(device, tmp_device,
647 &fs_devices->devices, dev_list) {
650 if (skip_device && skip_device == device)
652 if (path && !device->name)
657 not_found = strcmp(rcu_str_deref(device->name),
663 /* delete the stale device */
664 fs_devices->num_devices--;
665 list_del(&device->dev_list);
666 btrfs_free_device(device);
668 if (fs_devices->num_devices == 0)
671 mutex_unlock(&fs_devices->device_list_mutex);
672 if (fs_devices->num_devices == 0) {
673 btrfs_sysfs_remove_fsid(fs_devices);
674 list_del(&fs_devices->fs_list);
675 free_fs_devices(fs_devices);
680 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
681 struct btrfs_device *device, fmode_t flags,
684 struct request_queue *q;
685 struct block_device *bdev;
686 struct buffer_head *bh;
687 struct btrfs_super_block *disk_super;
696 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
701 disk_super = (struct btrfs_super_block *)bh->b_data;
702 devid = btrfs_stack_device_id(&disk_super->dev_item);
703 if (devid != device->devid)
706 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
709 device->generation = btrfs_super_generation(disk_super);
711 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
712 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
713 fs_devices->seeding = 1;
715 if (bdev_read_only(bdev))
716 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
718 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
721 q = bdev_get_queue(bdev);
722 if (!blk_queue_nonrot(q))
723 fs_devices->rotating = 1;
726 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
727 device->mode = flags;
729 fs_devices->open_devices++;
730 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
731 device->devid != BTRFS_DEV_REPLACE_DEVID) {
732 fs_devices->rw_devices++;
733 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
741 blkdev_put(bdev, flags);
747 * Add new device to list of registered devices
750 * device pointer which was just added or updated when successful
751 * error pointer when failed
753 static noinline struct btrfs_device *device_list_add(const char *path,
754 struct btrfs_super_block *disk_super,
755 bool *new_device_added)
757 struct btrfs_device *device;
758 struct btrfs_fs_devices *fs_devices;
759 struct rcu_string *name;
760 u64 found_transid = btrfs_super_generation(disk_super);
761 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
763 fs_devices = find_fsid(disk_super->fsid);
765 fs_devices = alloc_fs_devices(disk_super->fsid);
766 if (IS_ERR(fs_devices))
767 return ERR_CAST(fs_devices);
769 mutex_lock(&fs_devices->device_list_mutex);
770 list_add(&fs_devices->fs_list, &fs_uuids);
774 mutex_lock(&fs_devices->device_list_mutex);
775 device = find_device(fs_devices, devid,
776 disk_super->dev_item.uuid);
780 if (fs_devices->opened) {
781 mutex_unlock(&fs_devices->device_list_mutex);
782 return ERR_PTR(-EBUSY);
785 device = btrfs_alloc_device(NULL, &devid,
786 disk_super->dev_item.uuid);
787 if (IS_ERR(device)) {
788 mutex_unlock(&fs_devices->device_list_mutex);
789 /* we can safely leave the fs_devices entry around */
793 name = rcu_string_strdup(path, GFP_NOFS);
795 btrfs_free_device(device);
796 mutex_unlock(&fs_devices->device_list_mutex);
797 return ERR_PTR(-ENOMEM);
799 rcu_assign_pointer(device->name, name);
801 list_add_rcu(&device->dev_list, &fs_devices->devices);
802 fs_devices->num_devices++;
804 device->fs_devices = fs_devices;
805 *new_device_added = true;
807 if (disk_super->label[0])
808 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
809 disk_super->label, devid, found_transid, path);
811 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
812 disk_super->fsid, devid, found_transid, path);
814 } else if (!device->name || strcmp(device->name->str, path)) {
816 * When FS is already mounted.
817 * 1. If you are here and if the device->name is NULL that
818 * means this device was missing at time of FS mount.
819 * 2. If you are here and if the device->name is different
820 * from 'path' that means either
821 * a. The same device disappeared and reappeared with
823 * b. The missing-disk-which-was-replaced, has
826 * We must allow 1 and 2a above. But 2b would be a spurious
829 * Further in case of 1 and 2a above, the disk at 'path'
830 * would have missed some transaction when it was away and
831 * in case of 2a the stale bdev has to be updated as well.
832 * 2b must not be allowed at all time.
836 * For now, we do allow update to btrfs_fs_device through the
837 * btrfs dev scan cli after FS has been mounted. We're still
838 * tracking a problem where systems fail mount by subvolume id
839 * when we reject replacement on a mounted FS.
841 if (!fs_devices->opened && found_transid < device->generation) {
843 * That is if the FS is _not_ mounted and if you
844 * are here, that means there is more than one
845 * disk with same uuid and devid.We keep the one
846 * with larger generation number or the last-in if
847 * generation are equal.
849 mutex_unlock(&fs_devices->device_list_mutex);
850 return ERR_PTR(-EEXIST);
853 name = rcu_string_strdup(path, GFP_NOFS);
855 mutex_unlock(&fs_devices->device_list_mutex);
856 return ERR_PTR(-ENOMEM);
858 rcu_string_free(device->name);
859 rcu_assign_pointer(device->name, name);
860 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
861 fs_devices->missing_devices--;
862 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
867 * Unmount does not free the btrfs_device struct but would zero
868 * generation along with most of the other members. So just update
869 * it back. We need it to pick the disk with largest generation
872 if (!fs_devices->opened)
873 device->generation = found_transid;
875 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
877 mutex_unlock(&fs_devices->device_list_mutex);
881 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
883 struct btrfs_fs_devices *fs_devices;
884 struct btrfs_device *device;
885 struct btrfs_device *orig_dev;
887 fs_devices = alloc_fs_devices(orig->fsid);
888 if (IS_ERR(fs_devices))
891 mutex_lock(&orig->device_list_mutex);
892 fs_devices->total_devices = orig->total_devices;
894 /* We have held the volume lock, it is safe to get the devices. */
895 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
896 struct rcu_string *name;
898 device = btrfs_alloc_device(NULL, &orig_dev->devid,
904 * This is ok to do without rcu read locked because we hold the
905 * uuid mutex so nothing we touch in here is going to disappear.
907 if (orig_dev->name) {
908 name = rcu_string_strdup(orig_dev->name->str,
911 btrfs_free_device(device);
914 rcu_assign_pointer(device->name, name);
917 list_add(&device->dev_list, &fs_devices->devices);
918 device->fs_devices = fs_devices;
919 fs_devices->num_devices++;
921 mutex_unlock(&orig->device_list_mutex);
924 mutex_unlock(&orig->device_list_mutex);
925 free_fs_devices(fs_devices);
926 return ERR_PTR(-ENOMEM);
930 * After we have read the system tree and know devids belonging to
931 * this filesystem, remove the device which does not belong there.
933 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
935 struct btrfs_device *device, *next;
936 struct btrfs_device *latest_dev = NULL;
938 mutex_lock(&uuid_mutex);
940 /* This is the initialized path, it is safe to release the devices. */
941 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
942 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
943 &device->dev_state)) {
944 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
945 &device->dev_state) &&
947 device->generation > latest_dev->generation)) {
953 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
955 * In the first step, keep the device which has
956 * the correct fsid and the devid that is used
957 * for the dev_replace procedure.
958 * In the second step, the dev_replace state is
959 * read from the device tree and it is known
960 * whether the procedure is really active or
961 * not, which means whether this device is
962 * used or whether it should be removed.
964 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
965 &device->dev_state)) {
970 blkdev_put(device->bdev, device->mode);
972 fs_devices->open_devices--;
974 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
975 list_del_init(&device->dev_alloc_list);
976 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
977 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
979 fs_devices->rw_devices--;
981 list_del_init(&device->dev_list);
982 fs_devices->num_devices--;
983 btrfs_free_device(device);
986 if (fs_devices->seed) {
987 fs_devices = fs_devices->seed;
991 fs_devices->latest_bdev = latest_dev->bdev;
993 mutex_unlock(&uuid_mutex);
996 static void free_device_rcu(struct rcu_head *head)
998 struct btrfs_device *device;
1000 device = container_of(head, struct btrfs_device, rcu);
1001 btrfs_free_device(device);
1004 static void btrfs_close_bdev(struct btrfs_device *device)
1009 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1010 sync_blockdev(device->bdev);
1011 invalidate_bdev(device->bdev);
1014 blkdev_put(device->bdev, device->mode);
1017 static void btrfs_close_one_device(struct btrfs_device *device)
1019 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1020 struct btrfs_device *new_device;
1021 struct rcu_string *name;
1024 fs_devices->open_devices--;
1026 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1027 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1028 list_del_init(&device->dev_alloc_list);
1029 fs_devices->rw_devices--;
1032 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1033 fs_devices->missing_devices--;
1035 btrfs_close_bdev(device);
1037 new_device = btrfs_alloc_device(NULL, &device->devid,
1039 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1041 /* Safe because we are under uuid_mutex */
1043 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1044 BUG_ON(!name); /* -ENOMEM */
1045 rcu_assign_pointer(new_device->name, name);
1048 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1049 new_device->fs_devices = device->fs_devices;
1051 call_rcu(&device->rcu, free_device_rcu);
1054 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1056 struct btrfs_device *device, *tmp;
1058 if (--fs_devices->opened > 0)
1061 mutex_lock(&fs_devices->device_list_mutex);
1062 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1063 btrfs_close_one_device(device);
1065 mutex_unlock(&fs_devices->device_list_mutex);
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 lockdep_assert_held(&uuid_mutex);
1147 mutex_lock(&fs_devices->device_list_mutex);
1148 if (fs_devices->opened) {
1149 fs_devices->opened++;
1152 list_sort(NULL, &fs_devices->devices, devid_cmp);
1153 ret = open_fs_devices(fs_devices, flags, holder);
1155 mutex_unlock(&fs_devices->device_list_mutex);
1160 static void btrfs_release_disk_super(struct page *page)
1166 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1168 struct btrfs_super_block **disk_super)
1173 /* make sure our super fits in the device */
1174 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1177 /* make sure our super fits in the page */
1178 if (sizeof(**disk_super) > PAGE_SIZE)
1181 /* make sure our super doesn't straddle pages on disk */
1182 index = bytenr >> PAGE_SHIFT;
1183 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1186 /* pull in the page with our super */
1187 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1190 if (IS_ERR_OR_NULL(*page))
1195 /* align our pointer to the offset of the super block */
1196 *disk_super = p + (bytenr & ~PAGE_MASK);
1198 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1199 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1200 btrfs_release_disk_super(*page);
1204 if ((*disk_super)->label[0] &&
1205 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1206 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1212 * Look for a btrfs signature on a device. This may be called out of the mount path
1213 * and we are not allowed to call set_blocksize during the scan. The superblock
1214 * is read via pagecache
1216 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1219 struct btrfs_super_block *disk_super;
1220 bool new_device_added = false;
1221 struct btrfs_device *device = NULL;
1222 struct block_device *bdev;
1226 lockdep_assert_held(&uuid_mutex);
1229 * we would like to check all the supers, but that would make
1230 * a btrfs mount succeed after a mkfs from a different FS.
1231 * So, we need to add a special mount option to scan for
1232 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1234 bytenr = btrfs_sb_offset(0);
1235 flags |= FMODE_EXCL;
1237 bdev = blkdev_get_by_path(path, flags, holder);
1239 return ERR_CAST(bdev);
1241 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1242 device = ERR_PTR(-EINVAL);
1243 goto error_bdev_put;
1246 device = device_list_add(path, disk_super, &new_device_added);
1247 if (!IS_ERR(device)) {
1248 if (new_device_added)
1249 btrfs_free_stale_devices(path, device);
1252 btrfs_release_disk_super(page);
1255 blkdev_put(bdev, flags);
1260 static int contains_pending_extent(struct btrfs_transaction *transaction,
1261 struct btrfs_device *device,
1262 u64 *start, u64 len)
1264 struct btrfs_fs_info *fs_info = device->fs_info;
1265 struct extent_map *em;
1266 struct list_head *search_list = &fs_info->pinned_chunks;
1268 u64 physical_start = *start;
1271 search_list = &transaction->pending_chunks;
1273 list_for_each_entry(em, search_list, list) {
1274 struct map_lookup *map;
1277 map = em->map_lookup;
1278 for (i = 0; i < map->num_stripes; i++) {
1281 if (map->stripes[i].dev != device)
1283 if (map->stripes[i].physical >= physical_start + len ||
1284 map->stripes[i].physical + em->orig_block_len <=
1288 * Make sure that while processing the pinned list we do
1289 * not override our *start with a lower value, because
1290 * we can have pinned chunks that fall within this
1291 * device hole and that have lower physical addresses
1292 * than the pending chunks we processed before. If we
1293 * do not take this special care we can end up getting
1294 * 2 pending chunks that start at the same physical
1295 * device offsets because the end offset of a pinned
1296 * chunk can be equal to the start offset of some
1299 end = map->stripes[i].physical + em->orig_block_len;
1306 if (search_list != &fs_info->pinned_chunks) {
1307 search_list = &fs_info->pinned_chunks;
1316 * find_free_dev_extent_start - find free space in the specified device
1317 * @device: the device which we search the free space in
1318 * @num_bytes: the size of the free space that we need
1319 * @search_start: the position from which to begin the search
1320 * @start: store the start of the free space.
1321 * @len: the size of the free space. that we find, or the size
1322 * of the max free space if we don't find suitable free space
1324 * this uses a pretty simple search, the expectation is that it is
1325 * called very infrequently and that a given device has a small number
1328 * @start is used to store the start of the free space if we find. But if we
1329 * don't find suitable free space, it will be used to store the start position
1330 * of the max free space.
1332 * @len is used to store the size of the free space that we find.
1333 * But if we don't find suitable free space, it is used to store the size of
1334 * the max free space.
1336 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1337 struct btrfs_device *device, u64 num_bytes,
1338 u64 search_start, u64 *start, u64 *len)
1340 struct btrfs_fs_info *fs_info = device->fs_info;
1341 struct btrfs_root *root = fs_info->dev_root;
1342 struct btrfs_key key;
1343 struct btrfs_dev_extent *dev_extent;
1344 struct btrfs_path *path;
1349 u64 search_end = device->total_bytes;
1352 struct extent_buffer *l;
1355 * We don't want to overwrite the superblock on the drive nor any area
1356 * used by the boot loader (grub for example), so we make sure to start
1357 * at an offset of at least 1MB.
1359 search_start = max_t(u64, search_start, SZ_1M);
1361 path = btrfs_alloc_path();
1365 max_hole_start = search_start;
1369 if (search_start >= search_end ||
1370 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1375 path->reada = READA_FORWARD;
1376 path->search_commit_root = 1;
1377 path->skip_locking = 1;
1379 key.objectid = device->devid;
1380 key.offset = search_start;
1381 key.type = BTRFS_DEV_EXTENT_KEY;
1383 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1387 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1394 slot = path->slots[0];
1395 if (slot >= btrfs_header_nritems(l)) {
1396 ret = btrfs_next_leaf(root, path);
1404 btrfs_item_key_to_cpu(l, &key, slot);
1406 if (key.objectid < device->devid)
1409 if (key.objectid > device->devid)
1412 if (key.type != BTRFS_DEV_EXTENT_KEY)
1415 if (key.offset > search_start) {
1416 hole_size = key.offset - search_start;
1419 * Have to check before we set max_hole_start, otherwise
1420 * we could end up sending back this offset anyway.
1422 if (contains_pending_extent(transaction, device,
1425 if (key.offset >= search_start) {
1426 hole_size = key.offset - search_start;
1433 if (hole_size > max_hole_size) {
1434 max_hole_start = search_start;
1435 max_hole_size = hole_size;
1439 * If this free space is greater than which we need,
1440 * it must be the max free space that we have found
1441 * until now, so max_hole_start must point to the start
1442 * of this free space and the length of this free space
1443 * is stored in max_hole_size. Thus, we return
1444 * max_hole_start and max_hole_size and go back to the
1447 if (hole_size >= num_bytes) {
1453 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1454 extent_end = key.offset + btrfs_dev_extent_length(l,
1456 if (extent_end > search_start)
1457 search_start = extent_end;
1464 * At this point, search_start should be the end of
1465 * allocated dev extents, and when shrinking the device,
1466 * search_end may be smaller than search_start.
1468 if (search_end > search_start) {
1469 hole_size = search_end - search_start;
1471 if (contains_pending_extent(transaction, device, &search_start,
1473 btrfs_release_path(path);
1477 if (hole_size > max_hole_size) {
1478 max_hole_start = search_start;
1479 max_hole_size = hole_size;
1484 if (max_hole_size < num_bytes)
1490 btrfs_free_path(path);
1491 *start = max_hole_start;
1493 *len = max_hole_size;
1497 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1498 struct btrfs_device *device, u64 num_bytes,
1499 u64 *start, u64 *len)
1501 /* FIXME use last free of some kind */
1502 return find_free_dev_extent_start(trans->transaction, device,
1503 num_bytes, 0, start, len);
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1507 struct btrfs_device *device,
1508 u64 start, u64 *dev_extent_len)
1510 struct btrfs_fs_info *fs_info = device->fs_info;
1511 struct btrfs_root *root = fs_info->dev_root;
1513 struct btrfs_path *path;
1514 struct btrfs_key key;
1515 struct btrfs_key found_key;
1516 struct extent_buffer *leaf = NULL;
1517 struct btrfs_dev_extent *extent = NULL;
1519 path = btrfs_alloc_path();
1523 key.objectid = device->devid;
1525 key.type = BTRFS_DEV_EXTENT_KEY;
1527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1529 ret = btrfs_previous_item(root, path, key.objectid,
1530 BTRFS_DEV_EXTENT_KEY);
1533 leaf = path->nodes[0];
1534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 BUG_ON(found_key.offset > start || found_key.offset +
1538 btrfs_dev_extent_length(leaf, extent) < start);
1540 btrfs_release_path(path);
1542 } else if (ret == 0) {
1543 leaf = path->nodes[0];
1544 extent = btrfs_item_ptr(leaf, path->slots[0],
1545 struct btrfs_dev_extent);
1547 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1551 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1553 ret = btrfs_del_item(trans, root, path);
1555 btrfs_handle_fs_error(fs_info, ret,
1556 "Failed to remove dev extent item");
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1561 btrfs_free_path(path);
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566 struct btrfs_device *device,
1567 u64 chunk_offset, u64 start, u64 num_bytes)
1570 struct btrfs_path *path;
1571 struct btrfs_fs_info *fs_info = device->fs_info;
1572 struct btrfs_root *root = fs_info->dev_root;
1573 struct btrfs_dev_extent *extent;
1574 struct extent_buffer *leaf;
1575 struct btrfs_key key;
1577 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1578 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1579 path = btrfs_alloc_path();
1583 key.objectid = device->devid;
1585 key.type = BTRFS_DEV_EXTENT_KEY;
1586 ret = btrfs_insert_empty_item(trans, root, path, &key,
1591 leaf = path->nodes[0];
1592 extent = btrfs_item_ptr(leaf, path->slots[0],
1593 struct btrfs_dev_extent);
1594 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1595 BTRFS_CHUNK_TREE_OBJECTID);
1596 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1597 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1598 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1600 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1601 btrfs_mark_buffer_dirty(leaf);
1603 btrfs_free_path(path);
1607 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1609 struct extent_map_tree *em_tree;
1610 struct extent_map *em;
1614 em_tree = &fs_info->mapping_tree.map_tree;
1615 read_lock(&em_tree->lock);
1616 n = rb_last(&em_tree->map);
1618 em = rb_entry(n, struct extent_map, rb_node);
1619 ret = em->start + em->len;
1621 read_unlock(&em_tree->lock);
1626 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1630 struct btrfs_key key;
1631 struct btrfs_key found_key;
1632 struct btrfs_path *path;
1634 path = btrfs_alloc_path();
1638 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1639 key.type = BTRFS_DEV_ITEM_KEY;
1640 key.offset = (u64)-1;
1642 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1646 BUG_ON(ret == 0); /* Corruption */
1648 ret = btrfs_previous_item(fs_info->chunk_root, path,
1649 BTRFS_DEV_ITEMS_OBJECTID,
1650 BTRFS_DEV_ITEM_KEY);
1654 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656 *devid_ret = found_key.offset + 1;
1660 btrfs_free_path(path);
1665 * the device information is stored in the chunk root
1666 * the btrfs_device struct should be fully filled in
1668 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1669 struct btrfs_fs_info *fs_info,
1670 struct btrfs_device *device)
1672 struct btrfs_root *root = fs_info->chunk_root;
1674 struct btrfs_path *path;
1675 struct btrfs_dev_item *dev_item;
1676 struct extent_buffer *leaf;
1677 struct btrfs_key key;
1680 path = btrfs_alloc_path();
1684 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1685 key.type = BTRFS_DEV_ITEM_KEY;
1686 key.offset = device->devid;
1688 ret = btrfs_insert_empty_item(trans, root, path, &key,
1693 leaf = path->nodes[0];
1694 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1696 btrfs_set_device_id(leaf, dev_item, device->devid);
1697 btrfs_set_device_generation(leaf, dev_item, 0);
1698 btrfs_set_device_type(leaf, dev_item, device->type);
1699 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1700 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1701 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1702 btrfs_set_device_total_bytes(leaf, dev_item,
1703 btrfs_device_get_disk_total_bytes(device));
1704 btrfs_set_device_bytes_used(leaf, dev_item,
1705 btrfs_device_get_bytes_used(device));
1706 btrfs_set_device_group(leaf, dev_item, 0);
1707 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1708 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1709 btrfs_set_device_start_offset(leaf, dev_item, 0);
1711 ptr = btrfs_device_uuid(dev_item);
1712 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1713 ptr = btrfs_device_fsid(dev_item);
1714 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1715 btrfs_mark_buffer_dirty(leaf);
1719 btrfs_free_path(path);
1724 * Function to update ctime/mtime for a given device path.
1725 * Mainly used for ctime/mtime based probe like libblkid.
1727 static void update_dev_time(const char *path_name)
1731 filp = filp_open(path_name, O_RDWR, 0);
1734 file_update_time(filp);
1735 filp_close(filp, NULL);
1738 static int btrfs_rm_dev_item(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_key key;
1745 struct btrfs_trans_handle *trans;
1747 path = btrfs_alloc_path();
1751 trans = btrfs_start_transaction(root, 0);
1752 if (IS_ERR(trans)) {
1753 btrfs_free_path(path);
1754 return PTR_ERR(trans);
1756 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1757 key.type = BTRFS_DEV_ITEM_KEY;
1758 key.offset = device->devid;
1760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1764 btrfs_abort_transaction(trans, ret);
1765 btrfs_end_transaction(trans);
1769 ret = btrfs_del_item(trans, root, path);
1771 btrfs_abort_transaction(trans, ret);
1772 btrfs_end_transaction(trans);
1776 btrfs_free_path(path);
1778 ret = btrfs_commit_transaction(trans);
1783 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1784 * filesystem. It's up to the caller to adjust that number regarding eg. device
1787 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1795 seq = read_seqbegin(&fs_info->profiles_lock);
1797 all_avail = fs_info->avail_data_alloc_bits |
1798 fs_info->avail_system_alloc_bits |
1799 fs_info->avail_metadata_alloc_bits;
1800 } while (read_seqretry(&fs_info->profiles_lock, seq));
1802 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1803 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1806 if (num_devices < btrfs_raid_array[i].devs_min) {
1807 int ret = btrfs_raid_array[i].mindev_error;
1817 static struct btrfs_device * btrfs_find_next_active_device(
1818 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1820 struct btrfs_device *next_device;
1822 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1823 if (next_device != device &&
1824 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1825 && next_device->bdev)
1833 * Helper function to check if the given device is part of s_bdev / latest_bdev
1834 * and replace it with the provided or the next active device, in the context
1835 * where this function called, there should be always be another device (or
1836 * this_dev) which is active.
1838 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1839 struct btrfs_device *device, struct btrfs_device *this_dev)
1841 struct btrfs_device *next_device;
1844 next_device = this_dev;
1846 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1848 ASSERT(next_device);
1850 if (fs_info->sb->s_bdev &&
1851 (fs_info->sb->s_bdev == device->bdev))
1852 fs_info->sb->s_bdev = next_device->bdev;
1854 if (fs_info->fs_devices->latest_bdev == device->bdev)
1855 fs_info->fs_devices->latest_bdev = next_device->bdev;
1858 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1861 struct btrfs_device *device;
1862 struct btrfs_fs_devices *cur_devices;
1863 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1867 mutex_lock(&uuid_mutex);
1869 num_devices = fs_devices->num_devices;
1870 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1871 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1872 WARN_ON(num_devices < 1);
1875 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1877 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1881 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1886 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1887 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1891 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1892 fs_info->fs_devices->rw_devices == 1) {
1893 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1897 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1898 mutex_lock(&fs_info->chunk_mutex);
1899 list_del_init(&device->dev_alloc_list);
1900 device->fs_devices->rw_devices--;
1901 mutex_unlock(&fs_info->chunk_mutex);
1904 mutex_unlock(&uuid_mutex);
1905 ret = btrfs_shrink_device(device, 0);
1906 mutex_lock(&uuid_mutex);
1911 * TODO: the superblock still includes this device in its num_devices
1912 * counter although write_all_supers() is not locked out. This
1913 * could give a filesystem state which requires a degraded mount.
1915 ret = btrfs_rm_dev_item(fs_info, device);
1919 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1920 btrfs_scrub_cancel_dev(fs_info, device);
1923 * the device list mutex makes sure that we don't change
1924 * the device list while someone else is writing out all
1925 * the device supers. Whoever is writing all supers, should
1926 * lock the device list mutex before getting the number of
1927 * devices in the super block (super_copy). Conversely,
1928 * whoever updates the number of devices in the super block
1929 * (super_copy) should hold the device list mutex.
1933 * In normal cases the cur_devices == fs_devices. But in case
1934 * of deleting a seed device, the cur_devices should point to
1935 * its own fs_devices listed under the fs_devices->seed.
1937 cur_devices = device->fs_devices;
1938 mutex_lock(&fs_devices->device_list_mutex);
1939 list_del_rcu(&device->dev_list);
1941 cur_devices->num_devices--;
1942 cur_devices->total_devices--;
1943 /* Update total_devices of the parent fs_devices if it's seed */
1944 if (cur_devices != fs_devices)
1945 fs_devices->total_devices--;
1947 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1948 cur_devices->missing_devices--;
1950 btrfs_assign_next_active_device(fs_info, device, NULL);
1953 cur_devices->open_devices--;
1954 /* remove sysfs entry */
1955 btrfs_sysfs_rm_device_link(fs_devices, device);
1958 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1959 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1960 mutex_unlock(&fs_devices->device_list_mutex);
1963 * at this point, the device is zero sized and detached from
1964 * the devices list. All that's left is to zero out the old
1965 * supers and free the device.
1967 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1968 btrfs_scratch_superblocks(device->bdev, device->name->str);
1970 btrfs_close_bdev(device);
1971 call_rcu(&device->rcu, free_device_rcu);
1973 if (cur_devices->open_devices == 0) {
1974 while (fs_devices) {
1975 if (fs_devices->seed == cur_devices) {
1976 fs_devices->seed = cur_devices->seed;
1979 fs_devices = fs_devices->seed;
1981 cur_devices->seed = NULL;
1982 close_fs_devices(cur_devices);
1983 free_fs_devices(cur_devices);
1987 mutex_unlock(&uuid_mutex);
1991 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1992 mutex_lock(&fs_info->chunk_mutex);
1993 list_add(&device->dev_alloc_list,
1994 &fs_devices->alloc_list);
1995 device->fs_devices->rw_devices++;
1996 mutex_unlock(&fs_info->chunk_mutex);
2001 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2002 struct btrfs_device *srcdev)
2004 struct btrfs_fs_devices *fs_devices;
2006 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2009 * in case of fs with no seed, srcdev->fs_devices will point
2010 * to fs_devices of fs_info. However when the dev being replaced is
2011 * a seed dev it will point to the seed's local fs_devices. In short
2012 * srcdev will have its correct fs_devices in both the cases.
2014 fs_devices = srcdev->fs_devices;
2016 list_del_rcu(&srcdev->dev_list);
2017 list_del(&srcdev->dev_alloc_list);
2018 fs_devices->num_devices--;
2019 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2020 fs_devices->missing_devices--;
2022 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2023 fs_devices->rw_devices--;
2026 fs_devices->open_devices--;
2029 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2030 struct btrfs_device *srcdev)
2032 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2034 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2035 /* zero out the old super if it is writable */
2036 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2039 btrfs_close_bdev(srcdev);
2040 call_rcu(&srcdev->rcu, free_device_rcu);
2042 /* if this is no devs we rather delete the fs_devices */
2043 if (!fs_devices->num_devices) {
2044 struct btrfs_fs_devices *tmp_fs_devices;
2047 * On a mounted FS, num_devices can't be zero unless it's a
2048 * seed. In case of a seed device being replaced, the replace
2049 * target added to the sprout FS, so there will be no more
2050 * device left under the seed FS.
2052 ASSERT(fs_devices->seeding);
2054 tmp_fs_devices = fs_info->fs_devices;
2055 while (tmp_fs_devices) {
2056 if (tmp_fs_devices->seed == fs_devices) {
2057 tmp_fs_devices->seed = fs_devices->seed;
2060 tmp_fs_devices = tmp_fs_devices->seed;
2062 fs_devices->seed = NULL;
2063 close_fs_devices(fs_devices);
2064 free_fs_devices(fs_devices);
2068 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2069 struct btrfs_device *tgtdev)
2071 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2074 mutex_lock(&fs_devices->device_list_mutex);
2076 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2079 fs_devices->open_devices--;
2081 fs_devices->num_devices--;
2083 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2085 list_del_rcu(&tgtdev->dev_list);
2087 mutex_unlock(&fs_devices->device_list_mutex);
2090 * The update_dev_time() with in btrfs_scratch_superblocks()
2091 * may lead to a call to btrfs_show_devname() which will try
2092 * to hold device_list_mutex. And here this device
2093 * is already out of device list, so we don't have to hold
2094 * the device_list_mutex lock.
2096 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2098 btrfs_close_bdev(tgtdev);
2099 call_rcu(&tgtdev->rcu, free_device_rcu);
2102 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2103 const char *device_path,
2104 struct btrfs_device **device)
2107 struct btrfs_super_block *disk_super;
2110 struct block_device *bdev;
2111 struct buffer_head *bh;
2114 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2115 fs_info->bdev_holder, 0, &bdev, &bh);
2118 disk_super = (struct btrfs_super_block *)bh->b_data;
2119 devid = btrfs_stack_device_id(&disk_super->dev_item);
2120 dev_uuid = disk_super->dev_item.uuid;
2121 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2125 blkdev_put(bdev, FMODE_READ);
2129 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2130 const char *device_path,
2131 struct btrfs_device **device)
2134 if (strcmp(device_path, "missing") == 0) {
2135 struct list_head *devices;
2136 struct btrfs_device *tmp;
2138 devices = &fs_info->fs_devices->devices;
2139 list_for_each_entry(tmp, devices, dev_list) {
2140 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2141 &tmp->dev_state) && !tmp->bdev) {
2148 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2152 return btrfs_find_device_by_path(fs_info, device_path, device);
2157 * Lookup a device given by device id, or the path if the id is 0.
2159 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2160 const char *devpath,
2161 struct btrfs_device **device)
2167 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2171 if (!devpath || !devpath[0])
2174 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2181 * does all the dirty work required for changing file system's UUID.
2183 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2185 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2186 struct btrfs_fs_devices *old_devices;
2187 struct btrfs_fs_devices *seed_devices;
2188 struct btrfs_super_block *disk_super = fs_info->super_copy;
2189 struct btrfs_device *device;
2192 lockdep_assert_held(&uuid_mutex);
2193 if (!fs_devices->seeding)
2196 seed_devices = alloc_fs_devices(NULL);
2197 if (IS_ERR(seed_devices))
2198 return PTR_ERR(seed_devices);
2200 old_devices = clone_fs_devices(fs_devices);
2201 if (IS_ERR(old_devices)) {
2202 kfree(seed_devices);
2203 return PTR_ERR(old_devices);
2206 list_add(&old_devices->fs_list, &fs_uuids);
2208 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2209 seed_devices->opened = 1;
2210 INIT_LIST_HEAD(&seed_devices->devices);
2211 INIT_LIST_HEAD(&seed_devices->alloc_list);
2212 mutex_init(&seed_devices->device_list_mutex);
2214 mutex_lock(&fs_devices->device_list_mutex);
2215 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2217 list_for_each_entry(device, &seed_devices->devices, dev_list)
2218 device->fs_devices = seed_devices;
2220 mutex_lock(&fs_info->chunk_mutex);
2221 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2222 mutex_unlock(&fs_info->chunk_mutex);
2224 fs_devices->seeding = 0;
2225 fs_devices->num_devices = 0;
2226 fs_devices->open_devices = 0;
2227 fs_devices->missing_devices = 0;
2228 fs_devices->rotating = 0;
2229 fs_devices->seed = seed_devices;
2231 generate_random_uuid(fs_devices->fsid);
2232 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2234 mutex_unlock(&fs_devices->device_list_mutex);
2236 super_flags = btrfs_super_flags(disk_super) &
2237 ~BTRFS_SUPER_FLAG_SEEDING;
2238 btrfs_set_super_flags(disk_super, super_flags);
2244 * Store the expected generation for seed devices in device items.
2246 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2247 struct btrfs_fs_info *fs_info)
2249 struct btrfs_root *root = fs_info->chunk_root;
2250 struct btrfs_path *path;
2251 struct extent_buffer *leaf;
2252 struct btrfs_dev_item *dev_item;
2253 struct btrfs_device *device;
2254 struct btrfs_key key;
2255 u8 fs_uuid[BTRFS_FSID_SIZE];
2256 u8 dev_uuid[BTRFS_UUID_SIZE];
2260 path = btrfs_alloc_path();
2264 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2266 key.type = BTRFS_DEV_ITEM_KEY;
2269 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2273 leaf = path->nodes[0];
2275 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2276 ret = btrfs_next_leaf(root, path);
2281 leaf = path->nodes[0];
2282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283 btrfs_release_path(path);
2287 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2288 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2289 key.type != BTRFS_DEV_ITEM_KEY)
2292 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2293 struct btrfs_dev_item);
2294 devid = btrfs_device_id(leaf, dev_item);
2295 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2297 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2299 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2300 BUG_ON(!device); /* Logic error */
2302 if (device->fs_devices->seeding) {
2303 btrfs_set_device_generation(leaf, dev_item,
2304 device->generation);
2305 btrfs_mark_buffer_dirty(leaf);
2313 btrfs_free_path(path);
2317 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2319 struct btrfs_root *root = fs_info->dev_root;
2320 struct request_queue *q;
2321 struct btrfs_trans_handle *trans;
2322 struct btrfs_device *device;
2323 struct block_device *bdev;
2324 struct super_block *sb = fs_info->sb;
2325 struct rcu_string *name;
2326 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2328 int seeding_dev = 0;
2330 bool unlocked = false;
2332 if (sb_rdonly(sb) && !fs_devices->seeding)
2335 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2336 fs_info->bdev_holder);
2338 return PTR_ERR(bdev);
2340 if (fs_devices->seeding) {
2342 down_write(&sb->s_umount);
2343 mutex_lock(&uuid_mutex);
2346 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2348 mutex_lock(&fs_devices->device_list_mutex);
2349 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2350 if (device->bdev == bdev) {
2353 &fs_devices->device_list_mutex);
2357 mutex_unlock(&fs_devices->device_list_mutex);
2359 device = btrfs_alloc_device(fs_info, NULL, NULL);
2360 if (IS_ERR(device)) {
2361 /* we can safely leave the fs_devices entry around */
2362 ret = PTR_ERR(device);
2366 name = rcu_string_strdup(device_path, GFP_KERNEL);
2369 goto error_free_device;
2371 rcu_assign_pointer(device->name, name);
2373 trans = btrfs_start_transaction(root, 0);
2374 if (IS_ERR(trans)) {
2375 ret = PTR_ERR(trans);
2376 goto error_free_device;
2379 q = bdev_get_queue(bdev);
2380 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2381 device->generation = trans->transid;
2382 device->io_width = fs_info->sectorsize;
2383 device->io_align = fs_info->sectorsize;
2384 device->sector_size = fs_info->sectorsize;
2385 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2386 fs_info->sectorsize);
2387 device->disk_total_bytes = device->total_bytes;
2388 device->commit_total_bytes = device->total_bytes;
2389 device->fs_info = fs_info;
2390 device->bdev = bdev;
2391 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2392 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2393 device->mode = FMODE_EXCL;
2394 device->dev_stats_valid = 1;
2395 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2398 sb->s_flags &= ~SB_RDONLY;
2399 ret = btrfs_prepare_sprout(fs_info);
2401 btrfs_abort_transaction(trans, ret);
2406 device->fs_devices = fs_devices;
2408 mutex_lock(&fs_devices->device_list_mutex);
2409 mutex_lock(&fs_info->chunk_mutex);
2410 list_add_rcu(&device->dev_list, &fs_devices->devices);
2411 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2412 fs_devices->num_devices++;
2413 fs_devices->open_devices++;
2414 fs_devices->rw_devices++;
2415 fs_devices->total_devices++;
2416 fs_devices->total_rw_bytes += device->total_bytes;
2418 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2420 if (!blk_queue_nonrot(q))
2421 fs_devices->rotating = 1;
2423 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2424 btrfs_set_super_total_bytes(fs_info->super_copy,
2425 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2427 tmp = btrfs_super_num_devices(fs_info->super_copy);
2428 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2430 /* add sysfs device entry */
2431 btrfs_sysfs_add_device_link(fs_devices, device);
2434 * we've got more storage, clear any full flags on the space
2437 btrfs_clear_space_info_full(fs_info);
2439 mutex_unlock(&fs_info->chunk_mutex);
2440 mutex_unlock(&fs_devices->device_list_mutex);
2443 mutex_lock(&fs_info->chunk_mutex);
2444 ret = init_first_rw_device(trans, fs_info);
2445 mutex_unlock(&fs_info->chunk_mutex);
2447 btrfs_abort_transaction(trans, ret);
2452 ret = btrfs_add_dev_item(trans, fs_info, device);
2454 btrfs_abort_transaction(trans, ret);
2459 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2461 ret = btrfs_finish_sprout(trans, fs_info);
2463 btrfs_abort_transaction(trans, ret);
2467 /* Sprouting would change fsid of the mounted root,
2468 * so rename the fsid on the sysfs
2470 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2472 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2474 "sysfs: failed to create fsid for sprout");
2477 ret = btrfs_commit_transaction(trans);
2480 mutex_unlock(&uuid_mutex);
2481 up_write(&sb->s_umount);
2484 if (ret) /* transaction commit */
2487 ret = btrfs_relocate_sys_chunks(fs_info);
2489 btrfs_handle_fs_error(fs_info, ret,
2490 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2491 trans = btrfs_attach_transaction(root);
2492 if (IS_ERR(trans)) {
2493 if (PTR_ERR(trans) == -ENOENT)
2495 ret = PTR_ERR(trans);
2499 ret = btrfs_commit_transaction(trans);
2502 /* Update ctime/mtime for libblkid */
2503 update_dev_time(device_path);
2507 btrfs_sysfs_rm_device_link(fs_devices, device);
2510 sb->s_flags |= SB_RDONLY;
2512 btrfs_end_transaction(trans);
2514 btrfs_free_device(device);
2516 blkdev_put(bdev, FMODE_EXCL);
2517 if (seeding_dev && !unlocked) {
2518 mutex_unlock(&uuid_mutex);
2519 up_write(&sb->s_umount);
2524 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2525 struct btrfs_device *device)
2528 struct btrfs_path *path;
2529 struct btrfs_root *root = device->fs_info->chunk_root;
2530 struct btrfs_dev_item *dev_item;
2531 struct extent_buffer *leaf;
2532 struct btrfs_key key;
2534 path = btrfs_alloc_path();
2538 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2539 key.type = BTRFS_DEV_ITEM_KEY;
2540 key.offset = device->devid;
2542 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2551 leaf = path->nodes[0];
2552 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2554 btrfs_set_device_id(leaf, dev_item, device->devid);
2555 btrfs_set_device_type(leaf, dev_item, device->type);
2556 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2557 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2558 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2559 btrfs_set_device_total_bytes(leaf, dev_item,
2560 btrfs_device_get_disk_total_bytes(device));
2561 btrfs_set_device_bytes_used(leaf, dev_item,
2562 btrfs_device_get_bytes_used(device));
2563 btrfs_mark_buffer_dirty(leaf);
2566 btrfs_free_path(path);
2570 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2571 struct btrfs_device *device, u64 new_size)
2573 struct btrfs_fs_info *fs_info = device->fs_info;
2574 struct btrfs_super_block *super_copy = fs_info->super_copy;
2575 struct btrfs_fs_devices *fs_devices;
2579 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2582 new_size = round_down(new_size, fs_info->sectorsize);
2584 mutex_lock(&fs_info->chunk_mutex);
2585 old_total = btrfs_super_total_bytes(super_copy);
2586 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2588 if (new_size <= device->total_bytes ||
2589 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2590 mutex_unlock(&fs_info->chunk_mutex);
2594 fs_devices = fs_info->fs_devices;
2596 btrfs_set_super_total_bytes(super_copy,
2597 round_down(old_total + diff, fs_info->sectorsize));
2598 device->fs_devices->total_rw_bytes += diff;
2600 btrfs_device_set_total_bytes(device, new_size);
2601 btrfs_device_set_disk_total_bytes(device, new_size);
2602 btrfs_clear_space_info_full(device->fs_info);
2603 if (list_empty(&device->resized_list))
2604 list_add_tail(&device->resized_list,
2605 &fs_devices->resized_devices);
2606 mutex_unlock(&fs_info->chunk_mutex);
2608 return btrfs_update_device(trans, device);
2611 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2612 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2614 struct btrfs_root *root = fs_info->chunk_root;
2616 struct btrfs_path *path;
2617 struct btrfs_key key;
2619 path = btrfs_alloc_path();
2623 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2624 key.offset = chunk_offset;
2625 key.type = BTRFS_CHUNK_ITEM_KEY;
2627 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2630 else if (ret > 0) { /* Logic error or corruption */
2631 btrfs_handle_fs_error(fs_info, -ENOENT,
2632 "Failed lookup while freeing chunk.");
2637 ret = btrfs_del_item(trans, root, path);
2639 btrfs_handle_fs_error(fs_info, ret,
2640 "Failed to delete chunk item.");
2642 btrfs_free_path(path);
2646 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2648 struct btrfs_super_block *super_copy = fs_info->super_copy;
2649 struct btrfs_disk_key *disk_key;
2650 struct btrfs_chunk *chunk;
2657 struct btrfs_key key;
2659 mutex_lock(&fs_info->chunk_mutex);
2660 array_size = btrfs_super_sys_array_size(super_copy);
2662 ptr = super_copy->sys_chunk_array;
2665 while (cur < array_size) {
2666 disk_key = (struct btrfs_disk_key *)ptr;
2667 btrfs_disk_key_to_cpu(&key, disk_key);
2669 len = sizeof(*disk_key);
2671 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2672 chunk = (struct btrfs_chunk *)(ptr + len);
2673 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2674 len += btrfs_chunk_item_size(num_stripes);
2679 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2680 key.offset == chunk_offset) {
2681 memmove(ptr, ptr + len, array_size - (cur + len));
2683 btrfs_set_super_sys_array_size(super_copy, array_size);
2689 mutex_unlock(&fs_info->chunk_mutex);
2693 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2694 u64 logical, u64 length)
2696 struct extent_map_tree *em_tree;
2697 struct extent_map *em;
2699 em_tree = &fs_info->mapping_tree.map_tree;
2700 read_lock(&em_tree->lock);
2701 em = lookup_extent_mapping(em_tree, logical, length);
2702 read_unlock(&em_tree->lock);
2705 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2707 return ERR_PTR(-EINVAL);
2710 if (em->start > logical || em->start + em->len < logical) {
2712 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2713 logical, length, em->start, em->start + em->len);
2714 free_extent_map(em);
2715 return ERR_PTR(-EINVAL);
2718 /* callers are responsible for dropping em's ref. */
2722 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2723 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2725 struct extent_map *em;
2726 struct map_lookup *map;
2727 u64 dev_extent_len = 0;
2729 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2731 em = get_chunk_map(fs_info, chunk_offset, 1);
2734 * This is a logic error, but we don't want to just rely on the
2735 * user having built with ASSERT enabled, so if ASSERT doesn't
2736 * do anything we still error out.
2741 map = em->map_lookup;
2742 mutex_lock(&fs_info->chunk_mutex);
2743 check_system_chunk(trans, map->type);
2744 mutex_unlock(&fs_info->chunk_mutex);
2747 * Take the device list mutex to prevent races with the final phase of
2748 * a device replace operation that replaces the device object associated
2749 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2751 mutex_lock(&fs_devices->device_list_mutex);
2752 for (i = 0; i < map->num_stripes; i++) {
2753 struct btrfs_device *device = map->stripes[i].dev;
2754 ret = btrfs_free_dev_extent(trans, device,
2755 map->stripes[i].physical,
2758 mutex_unlock(&fs_devices->device_list_mutex);
2759 btrfs_abort_transaction(trans, ret);
2763 if (device->bytes_used > 0) {
2764 mutex_lock(&fs_info->chunk_mutex);
2765 btrfs_device_set_bytes_used(device,
2766 device->bytes_used - dev_extent_len);
2767 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2768 btrfs_clear_space_info_full(fs_info);
2769 mutex_unlock(&fs_info->chunk_mutex);
2772 if (map->stripes[i].dev) {
2773 ret = btrfs_update_device(trans, map->stripes[i].dev);
2775 mutex_unlock(&fs_devices->device_list_mutex);
2776 btrfs_abort_transaction(trans, ret);
2781 mutex_unlock(&fs_devices->device_list_mutex);
2783 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2785 btrfs_abort_transaction(trans, ret);
2789 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2791 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2792 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2794 btrfs_abort_transaction(trans, ret);
2799 ret = btrfs_remove_block_group(trans, chunk_offset, em);
2801 btrfs_abort_transaction(trans, ret);
2807 free_extent_map(em);
2811 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2813 struct btrfs_root *root = fs_info->chunk_root;
2814 struct btrfs_trans_handle *trans;
2818 * Prevent races with automatic removal of unused block groups.
2819 * After we relocate and before we remove the chunk with offset
2820 * chunk_offset, automatic removal of the block group can kick in,
2821 * resulting in a failure when calling btrfs_remove_chunk() below.
2823 * Make sure to acquire this mutex before doing a tree search (dev
2824 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2825 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2826 * we release the path used to search the chunk/dev tree and before
2827 * the current task acquires this mutex and calls us.
2829 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2831 ret = btrfs_can_relocate(fs_info, chunk_offset);
2835 /* step one, relocate all the extents inside this chunk */
2836 btrfs_scrub_pause(fs_info);
2837 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2838 btrfs_scrub_continue(fs_info);
2843 * We add the kobjects here (and after forcing data chunk creation)
2844 * since relocation is the only place we'll create chunks of a new
2845 * type at runtime. The only place where we'll remove the last
2846 * chunk of a type is the call immediately below this one. Even
2847 * so, we're protected against races with the cleaner thread since
2848 * we're covered by the delete_unused_bgs_mutex.
2850 btrfs_add_raid_kobjects(fs_info);
2852 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2854 if (IS_ERR(trans)) {
2855 ret = PTR_ERR(trans);
2856 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2861 * step two, delete the device extents and the
2862 * chunk tree entries
2864 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2865 btrfs_end_transaction(trans);
2869 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2871 struct btrfs_root *chunk_root = fs_info->chunk_root;
2872 struct btrfs_path *path;
2873 struct extent_buffer *leaf;
2874 struct btrfs_chunk *chunk;
2875 struct btrfs_key key;
2876 struct btrfs_key found_key;
2878 bool retried = false;
2882 path = btrfs_alloc_path();
2887 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2888 key.offset = (u64)-1;
2889 key.type = BTRFS_CHUNK_ITEM_KEY;
2892 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2893 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2895 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2898 BUG_ON(ret == 0); /* Corruption */
2900 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2903 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2909 leaf = path->nodes[0];
2910 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2912 chunk = btrfs_item_ptr(leaf, path->slots[0],
2913 struct btrfs_chunk);
2914 chunk_type = btrfs_chunk_type(leaf, chunk);
2915 btrfs_release_path(path);
2917 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2918 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2924 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2926 if (found_key.offset == 0)
2928 key.offset = found_key.offset - 1;
2931 if (failed && !retried) {
2935 } else if (WARN_ON(failed && retried)) {
2939 btrfs_free_path(path);
2944 * return 1 : allocate a data chunk successfully,
2945 * return <0: errors during allocating a data chunk,
2946 * return 0 : no need to allocate a data chunk.
2948 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
2951 struct btrfs_block_group_cache *cache;
2955 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2957 chunk_type = cache->flags;
2958 btrfs_put_block_group(cache);
2960 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
2961 spin_lock(&fs_info->data_sinfo->lock);
2962 bytes_used = fs_info->data_sinfo->bytes_used;
2963 spin_unlock(&fs_info->data_sinfo->lock);
2966 struct btrfs_trans_handle *trans;
2969 trans = btrfs_join_transaction(fs_info->tree_root);
2971 return PTR_ERR(trans);
2973 ret = btrfs_force_chunk_alloc(trans,
2974 BTRFS_BLOCK_GROUP_DATA);
2975 btrfs_end_transaction(trans);
2979 btrfs_add_raid_kobjects(fs_info);
2987 static int insert_balance_item(struct btrfs_fs_info *fs_info,
2988 struct btrfs_balance_control *bctl)
2990 struct btrfs_root *root = fs_info->tree_root;
2991 struct btrfs_trans_handle *trans;
2992 struct btrfs_balance_item *item;
2993 struct btrfs_disk_balance_args disk_bargs;
2994 struct btrfs_path *path;
2995 struct extent_buffer *leaf;
2996 struct btrfs_key key;
2999 path = btrfs_alloc_path();
3003 trans = btrfs_start_transaction(root, 0);
3004 if (IS_ERR(trans)) {
3005 btrfs_free_path(path);
3006 return PTR_ERR(trans);
3009 key.objectid = BTRFS_BALANCE_OBJECTID;
3010 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3013 ret = btrfs_insert_empty_item(trans, root, path, &key,
3018 leaf = path->nodes[0];
3019 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3021 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3023 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3024 btrfs_set_balance_data(leaf, item, &disk_bargs);
3025 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3026 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3027 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3028 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3030 btrfs_set_balance_flags(leaf, item, bctl->flags);
3032 btrfs_mark_buffer_dirty(leaf);
3034 btrfs_free_path(path);
3035 err = btrfs_commit_transaction(trans);
3041 static int del_balance_item(struct btrfs_fs_info *fs_info)
3043 struct btrfs_root *root = fs_info->tree_root;
3044 struct btrfs_trans_handle *trans;
3045 struct btrfs_path *path;
3046 struct btrfs_key key;
3049 path = btrfs_alloc_path();
3053 trans = btrfs_start_transaction(root, 0);
3054 if (IS_ERR(trans)) {
3055 btrfs_free_path(path);
3056 return PTR_ERR(trans);
3059 key.objectid = BTRFS_BALANCE_OBJECTID;
3060 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3063 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3071 ret = btrfs_del_item(trans, root, path);
3073 btrfs_free_path(path);
3074 err = btrfs_commit_transaction(trans);
3081 * This is a heuristic used to reduce the number of chunks balanced on
3082 * resume after balance was interrupted.
3084 static void update_balance_args(struct btrfs_balance_control *bctl)
3087 * Turn on soft mode for chunk types that were being converted.
3089 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3090 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3091 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3092 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3093 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3094 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3097 * Turn on usage filter if is not already used. The idea is
3098 * that chunks that we have already balanced should be
3099 * reasonably full. Don't do it for chunks that are being
3100 * converted - that will keep us from relocating unconverted
3101 * (albeit full) chunks.
3103 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3104 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3105 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3106 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3107 bctl->data.usage = 90;
3109 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3110 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3111 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3112 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3113 bctl->sys.usage = 90;
3115 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3116 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3117 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3118 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3119 bctl->meta.usage = 90;
3124 * Clear the balance status in fs_info and delete the balance item from disk.
3126 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3128 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3131 BUG_ON(!fs_info->balance_ctl);
3133 spin_lock(&fs_info->balance_lock);
3134 fs_info->balance_ctl = NULL;
3135 spin_unlock(&fs_info->balance_lock);
3138 ret = del_balance_item(fs_info);
3140 btrfs_handle_fs_error(fs_info, ret, NULL);
3144 * Balance filters. Return 1 if chunk should be filtered out
3145 * (should not be balanced).
3147 static int chunk_profiles_filter(u64 chunk_type,
3148 struct btrfs_balance_args *bargs)
3150 chunk_type = chunk_to_extended(chunk_type) &
3151 BTRFS_EXTENDED_PROFILE_MASK;
3153 if (bargs->profiles & chunk_type)
3159 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3160 struct btrfs_balance_args *bargs)
3162 struct btrfs_block_group_cache *cache;
3164 u64 user_thresh_min;
3165 u64 user_thresh_max;
3168 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3169 chunk_used = btrfs_block_group_used(&cache->item);
3171 if (bargs->usage_min == 0)
3172 user_thresh_min = 0;
3174 user_thresh_min = div_factor_fine(cache->key.offset,
3177 if (bargs->usage_max == 0)
3178 user_thresh_max = 1;
3179 else if (bargs->usage_max > 100)
3180 user_thresh_max = cache->key.offset;
3182 user_thresh_max = div_factor_fine(cache->key.offset,
3185 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3188 btrfs_put_block_group(cache);
3192 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3193 u64 chunk_offset, struct btrfs_balance_args *bargs)
3195 struct btrfs_block_group_cache *cache;
3196 u64 chunk_used, user_thresh;
3199 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3200 chunk_used = btrfs_block_group_used(&cache->item);
3202 if (bargs->usage_min == 0)
3204 else if (bargs->usage > 100)
3205 user_thresh = cache->key.offset;
3207 user_thresh = div_factor_fine(cache->key.offset,
3210 if (chunk_used < user_thresh)
3213 btrfs_put_block_group(cache);
3217 static int chunk_devid_filter(struct extent_buffer *leaf,
3218 struct btrfs_chunk *chunk,
3219 struct btrfs_balance_args *bargs)
3221 struct btrfs_stripe *stripe;
3222 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3225 for (i = 0; i < num_stripes; i++) {
3226 stripe = btrfs_stripe_nr(chunk, i);
3227 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3234 /* [pstart, pend) */
3235 static int chunk_drange_filter(struct extent_buffer *leaf,
3236 struct btrfs_chunk *chunk,
3237 struct btrfs_balance_args *bargs)
3239 struct btrfs_stripe *stripe;
3240 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3246 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3249 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3250 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3251 factor = num_stripes / 2;
3252 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3253 factor = num_stripes - 1;
3254 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3255 factor = num_stripes - 2;
3257 factor = num_stripes;
3260 for (i = 0; i < num_stripes; i++) {
3261 stripe = btrfs_stripe_nr(chunk, i);
3262 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3265 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3266 stripe_length = btrfs_chunk_length(leaf, chunk);
3267 stripe_length = div_u64(stripe_length, factor);
3269 if (stripe_offset < bargs->pend &&
3270 stripe_offset + stripe_length > bargs->pstart)
3277 /* [vstart, vend) */
3278 static int chunk_vrange_filter(struct extent_buffer *leaf,
3279 struct btrfs_chunk *chunk,
3281 struct btrfs_balance_args *bargs)
3283 if (chunk_offset < bargs->vend &&
3284 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3285 /* at least part of the chunk is inside this vrange */
3291 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3292 struct btrfs_chunk *chunk,
3293 struct btrfs_balance_args *bargs)
3295 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3297 if (bargs->stripes_min <= num_stripes
3298 && num_stripes <= bargs->stripes_max)
3304 static int chunk_soft_convert_filter(u64 chunk_type,
3305 struct btrfs_balance_args *bargs)
3307 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3310 chunk_type = chunk_to_extended(chunk_type) &
3311 BTRFS_EXTENDED_PROFILE_MASK;
3313 if (bargs->target == chunk_type)
3319 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3320 struct extent_buffer *leaf,
3321 struct btrfs_chunk *chunk, u64 chunk_offset)
3323 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3324 struct btrfs_balance_args *bargs = NULL;
3325 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3328 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3329 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3333 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3334 bargs = &bctl->data;
3335 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3337 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3338 bargs = &bctl->meta;
3340 /* profiles filter */
3341 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3342 chunk_profiles_filter(chunk_type, bargs)) {
3347 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3348 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3350 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3351 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3356 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3357 chunk_devid_filter(leaf, chunk, bargs)) {
3361 /* drange filter, makes sense only with devid filter */
3362 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3363 chunk_drange_filter(leaf, chunk, bargs)) {
3368 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3369 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3373 /* stripes filter */
3374 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3375 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3379 /* soft profile changing mode */
3380 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3381 chunk_soft_convert_filter(chunk_type, bargs)) {
3386 * limited by count, must be the last filter
3388 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3389 if (bargs->limit == 0)
3393 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3395 * Same logic as the 'limit' filter; the minimum cannot be
3396 * determined here because we do not have the global information
3397 * about the count of all chunks that satisfy the filters.
3399 if (bargs->limit_max == 0)
3408 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3410 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3411 struct btrfs_root *chunk_root = fs_info->chunk_root;
3412 struct btrfs_root *dev_root = fs_info->dev_root;
3413 struct list_head *devices;
3414 struct btrfs_device *device;
3418 struct btrfs_chunk *chunk;
3419 struct btrfs_path *path = NULL;
3420 struct btrfs_key key;
3421 struct btrfs_key found_key;
3422 struct btrfs_trans_handle *trans;
3423 struct extent_buffer *leaf;
3426 int enospc_errors = 0;
3427 bool counting = true;
3428 /* The single value limit and min/max limits use the same bytes in the */
3429 u64 limit_data = bctl->data.limit;
3430 u64 limit_meta = bctl->meta.limit;
3431 u64 limit_sys = bctl->sys.limit;
3435 int chunk_reserved = 0;
3437 /* step one make some room on all the devices */
3438 devices = &fs_info->fs_devices->devices;
3439 list_for_each_entry(device, devices, dev_list) {
3440 old_size = btrfs_device_get_total_bytes(device);
3441 size_to_free = div_factor(old_size, 1);
3442 size_to_free = min_t(u64, size_to_free, SZ_1M);
3443 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3444 btrfs_device_get_total_bytes(device) -
3445 btrfs_device_get_bytes_used(device) > size_to_free ||
3446 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3449 ret = btrfs_shrink_device(device, old_size - size_to_free);
3453 /* btrfs_shrink_device never returns ret > 0 */
3458 trans = btrfs_start_transaction(dev_root, 0);
3459 if (IS_ERR(trans)) {
3460 ret = PTR_ERR(trans);
3461 btrfs_info_in_rcu(fs_info,
3462 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3463 rcu_str_deref(device->name), ret,
3464 old_size, old_size - size_to_free);
3468 ret = btrfs_grow_device(trans, device, old_size);
3470 btrfs_end_transaction(trans);
3471 /* btrfs_grow_device never returns ret > 0 */
3473 btrfs_info_in_rcu(fs_info,
3474 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3475 rcu_str_deref(device->name), ret,
3476 old_size, old_size - size_to_free);
3480 btrfs_end_transaction(trans);
3483 /* step two, relocate all the chunks */
3484 path = btrfs_alloc_path();
3490 /* zero out stat counters */
3491 spin_lock(&fs_info->balance_lock);
3492 memset(&bctl->stat, 0, sizeof(bctl->stat));
3493 spin_unlock(&fs_info->balance_lock);
3497 * The single value limit and min/max limits use the same bytes
3500 bctl->data.limit = limit_data;
3501 bctl->meta.limit = limit_meta;
3502 bctl->sys.limit = limit_sys;
3504 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3505 key.offset = (u64)-1;
3506 key.type = BTRFS_CHUNK_ITEM_KEY;
3509 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3510 atomic_read(&fs_info->balance_cancel_req)) {
3515 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3516 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3518 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3523 * this shouldn't happen, it means the last relocate
3527 BUG(); /* FIXME break ? */
3529 ret = btrfs_previous_item(chunk_root, path, 0,
3530 BTRFS_CHUNK_ITEM_KEY);
3532 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3537 leaf = path->nodes[0];
3538 slot = path->slots[0];
3539 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3541 if (found_key.objectid != key.objectid) {
3542 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3546 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3547 chunk_type = btrfs_chunk_type(leaf, chunk);
3550 spin_lock(&fs_info->balance_lock);
3551 bctl->stat.considered++;
3552 spin_unlock(&fs_info->balance_lock);
3555 ret = should_balance_chunk(fs_info, leaf, chunk,
3558 btrfs_release_path(path);
3560 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3565 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3566 spin_lock(&fs_info->balance_lock);
3567 bctl->stat.expected++;
3568 spin_unlock(&fs_info->balance_lock);
3570 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3572 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3574 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3581 * Apply limit_min filter, no need to check if the LIMITS
3582 * filter is used, limit_min is 0 by default
3584 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3585 count_data < bctl->data.limit_min)
3586 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3587 count_meta < bctl->meta.limit_min)
3588 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3589 count_sys < bctl->sys.limit_min)) {
3590 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3594 if (!chunk_reserved) {
3596 * We may be relocating the only data chunk we have,
3597 * which could potentially end up with losing data's
3598 * raid profile, so lets allocate an empty one in
3601 ret = btrfs_may_alloc_data_chunk(fs_info,
3604 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3606 } else if (ret == 1) {
3611 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3612 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3613 if (ret && ret != -ENOSPC)
3615 if (ret == -ENOSPC) {
3618 spin_lock(&fs_info->balance_lock);
3619 bctl->stat.completed++;
3620 spin_unlock(&fs_info->balance_lock);
3623 if (found_key.offset == 0)
3625 key.offset = found_key.offset - 1;
3629 btrfs_release_path(path);
3634 btrfs_free_path(path);
3635 if (enospc_errors) {
3636 btrfs_info(fs_info, "%d enospc errors during balance",
3646 * alloc_profile_is_valid - see if a given profile is valid and reduced
3647 * @flags: profile to validate
3648 * @extended: if true @flags is treated as an extended profile
3650 static int alloc_profile_is_valid(u64 flags, int extended)
3652 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3653 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3655 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3657 /* 1) check that all other bits are zeroed */
3661 /* 2) see if profile is reduced */
3663 return !extended; /* "0" is valid for usual profiles */
3665 /* true if exactly one bit set */
3666 return (flags & (flags - 1)) == 0;
3669 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3671 /* cancel requested || normal exit path */
3672 return atomic_read(&fs_info->balance_cancel_req) ||
3673 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3674 atomic_read(&fs_info->balance_cancel_req) == 0);
3677 /* Non-zero return value signifies invalidity */
3678 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3681 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3682 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3683 (bctl_arg->target & ~allowed)));
3687 * Should be called with balance mutexe held
3689 int btrfs_balance(struct btrfs_fs_info *fs_info,
3690 struct btrfs_balance_control *bctl,
3691 struct btrfs_ioctl_balance_args *bargs)
3693 u64 meta_target, data_target;
3700 if (btrfs_fs_closing(fs_info) ||
3701 atomic_read(&fs_info->balance_pause_req) ||
3702 atomic_read(&fs_info->balance_cancel_req)) {
3707 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3708 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3712 * In case of mixed groups both data and meta should be picked,
3713 * and identical options should be given for both of them.
3715 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3716 if (mixed && (bctl->flags & allowed)) {
3717 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3718 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3719 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3721 "balance: mixed groups data and metadata options must be the same");
3727 num_devices = fs_info->fs_devices->num_devices;
3728 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3729 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3730 BUG_ON(num_devices < 1);
3733 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3734 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3735 if (num_devices > 1)
3736 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3737 if (num_devices > 2)
3738 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3739 if (num_devices > 3)
3740 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3741 BTRFS_BLOCK_GROUP_RAID6);
3742 if (validate_convert_profile(&bctl->data, allowed)) {
3743 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
3746 "balance: invalid convert data profile %s",
3747 get_raid_name(index));
3751 if (validate_convert_profile(&bctl->meta, allowed)) {
3752 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
3755 "balance: invalid convert metadata profile %s",
3756 get_raid_name(index));
3760 if (validate_convert_profile(&bctl->sys, allowed)) {
3761 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
3764 "balance: invalid convert system profile %s",
3765 get_raid_name(index));
3770 /* allow to reduce meta or sys integrity only if force set */
3771 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3772 BTRFS_BLOCK_GROUP_RAID10 |
3773 BTRFS_BLOCK_GROUP_RAID5 |
3774 BTRFS_BLOCK_GROUP_RAID6;
3776 seq = read_seqbegin(&fs_info->profiles_lock);
3778 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3779 (fs_info->avail_system_alloc_bits & allowed) &&
3780 !(bctl->sys.target & allowed)) ||
3781 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3782 (fs_info->avail_metadata_alloc_bits & allowed) &&
3783 !(bctl->meta.target & allowed))) {
3784 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3786 "balance: force reducing metadata integrity");
3789 "balance: reduces metadata integrity, use --force if you want this");
3794 } while (read_seqretry(&fs_info->profiles_lock, seq));
3796 /* if we're not converting, the target field is uninitialized */
3797 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3798 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3799 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3800 bctl->data.target : fs_info->avail_data_alloc_bits;
3801 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3802 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3803 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
3804 int data_index = btrfs_bg_flags_to_raid_index(data_target);
3807 "balance: metadata profile %s has lower redundancy than data profile %s",
3808 get_raid_name(meta_index), get_raid_name(data_index));
3811 ret = insert_balance_item(fs_info, bctl);
3812 if (ret && ret != -EEXIST)
3815 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3816 BUG_ON(ret == -EEXIST);
3817 BUG_ON(fs_info->balance_ctl);
3818 spin_lock(&fs_info->balance_lock);
3819 fs_info->balance_ctl = bctl;
3820 spin_unlock(&fs_info->balance_lock);
3822 BUG_ON(ret != -EEXIST);
3823 spin_lock(&fs_info->balance_lock);
3824 update_balance_args(bctl);
3825 spin_unlock(&fs_info->balance_lock);
3828 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3829 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3830 mutex_unlock(&fs_info->balance_mutex);
3832 ret = __btrfs_balance(fs_info);
3834 mutex_lock(&fs_info->balance_mutex);
3835 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3838 memset(bargs, 0, sizeof(*bargs));
3839 btrfs_update_ioctl_balance_args(fs_info, bargs);
3842 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3843 balance_need_close(fs_info)) {
3844 reset_balance_state(fs_info);
3845 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3848 wake_up(&fs_info->balance_wait_q);
3852 if (bctl->flags & BTRFS_BALANCE_RESUME)
3853 reset_balance_state(fs_info);
3856 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3861 static int balance_kthread(void *data)
3863 struct btrfs_fs_info *fs_info = data;
3866 mutex_lock(&fs_info->balance_mutex);
3867 if (fs_info->balance_ctl) {
3868 btrfs_info(fs_info, "balance: resuming");
3869 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
3871 mutex_unlock(&fs_info->balance_mutex);
3876 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3878 struct task_struct *tsk;
3880 mutex_lock(&fs_info->balance_mutex);
3881 if (!fs_info->balance_ctl) {
3882 mutex_unlock(&fs_info->balance_mutex);
3885 mutex_unlock(&fs_info->balance_mutex);
3887 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3888 btrfs_info(fs_info, "balance: resume skipped");
3893 * A ro->rw remount sequence should continue with the paused balance
3894 * regardless of who pauses it, system or the user as of now, so set
3897 spin_lock(&fs_info->balance_lock);
3898 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3899 spin_unlock(&fs_info->balance_lock);
3901 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3902 return PTR_ERR_OR_ZERO(tsk);
3905 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3907 struct btrfs_balance_control *bctl;
3908 struct btrfs_balance_item *item;
3909 struct btrfs_disk_balance_args disk_bargs;
3910 struct btrfs_path *path;
3911 struct extent_buffer *leaf;
3912 struct btrfs_key key;
3915 path = btrfs_alloc_path();
3919 key.objectid = BTRFS_BALANCE_OBJECTID;
3920 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3923 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3926 if (ret > 0) { /* ret = -ENOENT; */
3931 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3937 leaf = path->nodes[0];
3938 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3940 bctl->flags = btrfs_balance_flags(leaf, item);
3941 bctl->flags |= BTRFS_BALANCE_RESUME;
3943 btrfs_balance_data(leaf, item, &disk_bargs);
3944 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3945 btrfs_balance_meta(leaf, item, &disk_bargs);
3946 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3947 btrfs_balance_sys(leaf, item, &disk_bargs);
3948 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3951 * This should never happen, as the paused balance state is recovered
3952 * during mount without any chance of other exclusive ops to collide.
3954 * This gives the exclusive op status to balance and keeps in paused
3955 * state until user intervention (cancel or umount). If the ownership
3956 * cannot be assigned, show a message but do not fail. The balance
3957 * is in a paused state and must have fs_info::balance_ctl properly
3960 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3962 "balance: cannot set exclusive op status, resume manually");
3964 mutex_lock(&fs_info->balance_mutex);
3965 BUG_ON(fs_info->balance_ctl);
3966 spin_lock(&fs_info->balance_lock);
3967 fs_info->balance_ctl = bctl;
3968 spin_unlock(&fs_info->balance_lock);
3969 mutex_unlock(&fs_info->balance_mutex);
3971 btrfs_free_path(path);
3975 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3979 mutex_lock(&fs_info->balance_mutex);
3980 if (!fs_info->balance_ctl) {
3981 mutex_unlock(&fs_info->balance_mutex);
3985 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3986 atomic_inc(&fs_info->balance_pause_req);
3987 mutex_unlock(&fs_info->balance_mutex);
3989 wait_event(fs_info->balance_wait_q,
3990 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3992 mutex_lock(&fs_info->balance_mutex);
3993 /* we are good with balance_ctl ripped off from under us */
3994 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3995 atomic_dec(&fs_info->balance_pause_req);
4000 mutex_unlock(&fs_info->balance_mutex);
4004 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4006 mutex_lock(&fs_info->balance_mutex);
4007 if (!fs_info->balance_ctl) {
4008 mutex_unlock(&fs_info->balance_mutex);
4013 * A paused balance with the item stored on disk can be resumed at
4014 * mount time if the mount is read-write. Otherwise it's still paused
4015 * and we must not allow cancelling as it deletes the item.
4017 if (sb_rdonly(fs_info->sb)) {
4018 mutex_unlock(&fs_info->balance_mutex);
4022 atomic_inc(&fs_info->balance_cancel_req);
4024 * if we are running just wait and return, balance item is
4025 * deleted in btrfs_balance in this case
4027 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4028 mutex_unlock(&fs_info->balance_mutex);
4029 wait_event(fs_info->balance_wait_q,
4030 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4031 mutex_lock(&fs_info->balance_mutex);
4033 mutex_unlock(&fs_info->balance_mutex);
4035 * Lock released to allow other waiters to continue, we'll
4036 * reexamine the status again.
4038 mutex_lock(&fs_info->balance_mutex);
4040 if (fs_info->balance_ctl) {
4041 reset_balance_state(fs_info);
4042 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4043 btrfs_info(fs_info, "balance: canceled");
4047 BUG_ON(fs_info->balance_ctl ||
4048 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4049 atomic_dec(&fs_info->balance_cancel_req);
4050 mutex_unlock(&fs_info->balance_mutex);
4054 static int btrfs_uuid_scan_kthread(void *data)
4056 struct btrfs_fs_info *fs_info = data;
4057 struct btrfs_root *root = fs_info->tree_root;
4058 struct btrfs_key key;
4059 struct btrfs_path *path = NULL;
4061 struct extent_buffer *eb;
4063 struct btrfs_root_item root_item;
4065 struct btrfs_trans_handle *trans = NULL;
4067 path = btrfs_alloc_path();
4074 key.type = BTRFS_ROOT_ITEM_KEY;
4078 ret = btrfs_search_forward(root, &key, path,
4079 BTRFS_OLDEST_GENERATION);
4086 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4087 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4088 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4089 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4092 eb = path->nodes[0];
4093 slot = path->slots[0];
4094 item_size = btrfs_item_size_nr(eb, slot);
4095 if (item_size < sizeof(root_item))
4098 read_extent_buffer(eb, &root_item,
4099 btrfs_item_ptr_offset(eb, slot),
4100 (int)sizeof(root_item));
4101 if (btrfs_root_refs(&root_item) == 0)
4104 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4105 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4109 btrfs_release_path(path);
4111 * 1 - subvol uuid item
4112 * 1 - received_subvol uuid item
4114 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4115 if (IS_ERR(trans)) {
4116 ret = PTR_ERR(trans);
4124 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4125 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4126 BTRFS_UUID_KEY_SUBVOL,
4129 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4135 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4136 ret = btrfs_uuid_tree_add(trans,
4137 root_item.received_uuid,
4138 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4141 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4149 ret = btrfs_end_transaction(trans);
4155 btrfs_release_path(path);
4156 if (key.offset < (u64)-1) {
4158 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4160 key.type = BTRFS_ROOT_ITEM_KEY;
4161 } else if (key.objectid < (u64)-1) {
4163 key.type = BTRFS_ROOT_ITEM_KEY;
4172 btrfs_free_path(path);
4173 if (trans && !IS_ERR(trans))
4174 btrfs_end_transaction(trans);
4176 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4178 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4179 up(&fs_info->uuid_tree_rescan_sem);
4184 * Callback for btrfs_uuid_tree_iterate().
4186 * 0 check succeeded, the entry is not outdated.
4187 * < 0 if an error occurred.
4188 * > 0 if the check failed, which means the caller shall remove the entry.
4190 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4191 u8 *uuid, u8 type, u64 subid)
4193 struct btrfs_key key;
4195 struct btrfs_root *subvol_root;
4197 if (type != BTRFS_UUID_KEY_SUBVOL &&
4198 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4201 key.objectid = subid;
4202 key.type = BTRFS_ROOT_ITEM_KEY;
4203 key.offset = (u64)-1;
4204 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4205 if (IS_ERR(subvol_root)) {
4206 ret = PTR_ERR(subvol_root);
4213 case BTRFS_UUID_KEY_SUBVOL:
4214 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4217 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4218 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4228 static int btrfs_uuid_rescan_kthread(void *data)
4230 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4234 * 1st step is to iterate through the existing UUID tree and
4235 * to delete all entries that contain outdated data.
4236 * 2nd step is to add all missing entries to the UUID tree.
4238 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4240 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4241 up(&fs_info->uuid_tree_rescan_sem);
4244 return btrfs_uuid_scan_kthread(data);
4247 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4249 struct btrfs_trans_handle *trans;
4250 struct btrfs_root *tree_root = fs_info->tree_root;
4251 struct btrfs_root *uuid_root;
4252 struct task_struct *task;
4259 trans = btrfs_start_transaction(tree_root, 2);
4261 return PTR_ERR(trans);
4263 uuid_root = btrfs_create_tree(trans, fs_info,
4264 BTRFS_UUID_TREE_OBJECTID);
4265 if (IS_ERR(uuid_root)) {
4266 ret = PTR_ERR(uuid_root);
4267 btrfs_abort_transaction(trans, ret);
4268 btrfs_end_transaction(trans);
4272 fs_info->uuid_root = uuid_root;
4274 ret = btrfs_commit_transaction(trans);
4278 down(&fs_info->uuid_tree_rescan_sem);
4279 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4281 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4282 btrfs_warn(fs_info, "failed to start uuid_scan task");
4283 up(&fs_info->uuid_tree_rescan_sem);
4284 return PTR_ERR(task);
4290 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4292 struct task_struct *task;
4294 down(&fs_info->uuid_tree_rescan_sem);
4295 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4297 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4298 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4299 up(&fs_info->uuid_tree_rescan_sem);
4300 return PTR_ERR(task);
4307 * shrinking a device means finding all of the device extents past
4308 * the new size, and then following the back refs to the chunks.
4309 * The chunk relocation code actually frees the device extent
4311 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4313 struct btrfs_fs_info *fs_info = device->fs_info;
4314 struct btrfs_root *root = fs_info->dev_root;
4315 struct btrfs_trans_handle *trans;
4316 struct btrfs_dev_extent *dev_extent = NULL;
4317 struct btrfs_path *path;
4323 bool retried = false;
4324 bool checked_pending_chunks = false;
4325 struct extent_buffer *l;
4326 struct btrfs_key key;
4327 struct btrfs_super_block *super_copy = fs_info->super_copy;
4328 u64 old_total = btrfs_super_total_bytes(super_copy);
4329 u64 old_size = btrfs_device_get_total_bytes(device);
4332 new_size = round_down(new_size, fs_info->sectorsize);
4333 diff = round_down(old_size - new_size, fs_info->sectorsize);
4335 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4338 path = btrfs_alloc_path();
4342 path->reada = READA_BACK;
4344 mutex_lock(&fs_info->chunk_mutex);
4346 btrfs_device_set_total_bytes(device, new_size);
4347 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4348 device->fs_devices->total_rw_bytes -= diff;
4349 atomic64_sub(diff, &fs_info->free_chunk_space);
4351 mutex_unlock(&fs_info->chunk_mutex);
4354 key.objectid = device->devid;
4355 key.offset = (u64)-1;
4356 key.type = BTRFS_DEV_EXTENT_KEY;
4359 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4360 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4362 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4366 ret = btrfs_previous_item(root, path, 0, key.type);
4368 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4373 btrfs_release_path(path);
4378 slot = path->slots[0];
4379 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4381 if (key.objectid != device->devid) {
4382 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4383 btrfs_release_path(path);
4387 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4388 length = btrfs_dev_extent_length(l, dev_extent);
4390 if (key.offset + length <= new_size) {
4391 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4392 btrfs_release_path(path);
4396 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4397 btrfs_release_path(path);
4400 * We may be relocating the only data chunk we have,
4401 * which could potentially end up with losing data's
4402 * raid profile, so lets allocate an empty one in
4405 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4407 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4411 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4412 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4413 if (ret && ret != -ENOSPC)
4417 } while (key.offset-- > 0);
4419 if (failed && !retried) {
4423 } else if (failed && retried) {
4428 /* Shrinking succeeded, else we would be at "done". */
4429 trans = btrfs_start_transaction(root, 0);
4430 if (IS_ERR(trans)) {
4431 ret = PTR_ERR(trans);
4435 mutex_lock(&fs_info->chunk_mutex);
4438 * We checked in the above loop all device extents that were already in
4439 * the device tree. However before we have updated the device's
4440 * total_bytes to the new size, we might have had chunk allocations that
4441 * have not complete yet (new block groups attached to transaction
4442 * handles), and therefore their device extents were not yet in the
4443 * device tree and we missed them in the loop above. So if we have any
4444 * pending chunk using a device extent that overlaps the device range
4445 * that we can not use anymore, commit the current transaction and
4446 * repeat the search on the device tree - this way we guarantee we will
4447 * not have chunks using device extents that end beyond 'new_size'.
4449 if (!checked_pending_chunks) {
4450 u64 start = new_size;
4451 u64 len = old_size - new_size;
4453 if (contains_pending_extent(trans->transaction, device,
4455 mutex_unlock(&fs_info->chunk_mutex);
4456 checked_pending_chunks = true;
4459 ret = btrfs_commit_transaction(trans);
4466 btrfs_device_set_disk_total_bytes(device, new_size);
4467 if (list_empty(&device->resized_list))
4468 list_add_tail(&device->resized_list,
4469 &fs_info->fs_devices->resized_devices);
4471 WARN_ON(diff > old_total);
4472 btrfs_set_super_total_bytes(super_copy,
4473 round_down(old_total - diff, fs_info->sectorsize));
4474 mutex_unlock(&fs_info->chunk_mutex);
4476 /* Now btrfs_update_device() will change the on-disk size. */
4477 ret = btrfs_update_device(trans, device);
4478 btrfs_end_transaction(trans);
4480 btrfs_free_path(path);
4482 mutex_lock(&fs_info->chunk_mutex);
4483 btrfs_device_set_total_bytes(device, old_size);
4484 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4485 device->fs_devices->total_rw_bytes += diff;
4486 atomic64_add(diff, &fs_info->free_chunk_space);
4487 mutex_unlock(&fs_info->chunk_mutex);
4492 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4493 struct btrfs_key *key,
4494 struct btrfs_chunk *chunk, int item_size)
4496 struct btrfs_super_block *super_copy = fs_info->super_copy;
4497 struct btrfs_disk_key disk_key;
4501 mutex_lock(&fs_info->chunk_mutex);
4502 array_size = btrfs_super_sys_array_size(super_copy);
4503 if (array_size + item_size + sizeof(disk_key)
4504 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4505 mutex_unlock(&fs_info->chunk_mutex);
4509 ptr = super_copy->sys_chunk_array + array_size;
4510 btrfs_cpu_key_to_disk(&disk_key, key);
4511 memcpy(ptr, &disk_key, sizeof(disk_key));
4512 ptr += sizeof(disk_key);
4513 memcpy(ptr, chunk, item_size);
4514 item_size += sizeof(disk_key);
4515 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4516 mutex_unlock(&fs_info->chunk_mutex);
4522 * sort the devices in descending order by max_avail, total_avail
4524 static int btrfs_cmp_device_info(const void *a, const void *b)
4526 const struct btrfs_device_info *di_a = a;
4527 const struct btrfs_device_info *di_b = b;
4529 if (di_a->max_avail > di_b->max_avail)
4531 if (di_a->max_avail < di_b->max_avail)
4533 if (di_a->total_avail > di_b->total_avail)
4535 if (di_a->total_avail < di_b->total_avail)
4540 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4542 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4545 btrfs_set_fs_incompat(info, RAID56);
4548 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4549 - sizeof(struct btrfs_chunk)) \
4550 / sizeof(struct btrfs_stripe) + 1)
4552 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4553 - 2 * sizeof(struct btrfs_disk_key) \
4554 - 2 * sizeof(struct btrfs_chunk)) \
4555 / sizeof(struct btrfs_stripe) + 1)
4557 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4558 u64 start, u64 type)
4560 struct btrfs_fs_info *info = trans->fs_info;
4561 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4562 struct btrfs_device *device;
4563 struct map_lookup *map = NULL;
4564 struct extent_map_tree *em_tree;
4565 struct extent_map *em;
4566 struct btrfs_device_info *devices_info = NULL;
4568 int num_stripes; /* total number of stripes to allocate */
4569 int data_stripes; /* number of stripes that count for
4571 int sub_stripes; /* sub_stripes info for map */
4572 int dev_stripes; /* stripes per dev */
4573 int devs_max; /* max devs to use */
4574 int devs_min; /* min devs needed */
4575 int devs_increment; /* ndevs has to be a multiple of this */
4576 int ncopies; /* how many copies to data has */
4578 u64 max_stripe_size;
4587 BUG_ON(!alloc_profile_is_valid(type, 0));
4589 if (list_empty(&fs_devices->alloc_list)) {
4590 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4591 btrfs_debug(info, "%s: no writable device", __func__);
4595 index = btrfs_bg_flags_to_raid_index(type);
4597 sub_stripes = btrfs_raid_array[index].sub_stripes;
4598 dev_stripes = btrfs_raid_array[index].dev_stripes;
4599 devs_max = btrfs_raid_array[index].devs_max;
4600 devs_min = btrfs_raid_array[index].devs_min;
4601 devs_increment = btrfs_raid_array[index].devs_increment;
4602 ncopies = btrfs_raid_array[index].ncopies;
4604 if (type & BTRFS_BLOCK_GROUP_DATA) {
4605 max_stripe_size = SZ_1G;
4606 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4608 devs_max = BTRFS_MAX_DEVS(info);
4609 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4610 /* for larger filesystems, use larger metadata chunks */
4611 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4612 max_stripe_size = SZ_1G;
4614 max_stripe_size = SZ_256M;
4615 max_chunk_size = max_stripe_size;
4617 devs_max = BTRFS_MAX_DEVS(info);
4618 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4619 max_stripe_size = SZ_32M;
4620 max_chunk_size = 2 * max_stripe_size;
4622 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4624 btrfs_err(info, "invalid chunk type 0x%llx requested",
4629 /* we don't want a chunk larger than 10% of writeable space */
4630 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4633 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4639 * in the first pass through the devices list, we gather information
4640 * about the available holes on each device.
4643 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4647 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4649 "BTRFS: read-only device in alloc_list\n");
4653 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4654 &device->dev_state) ||
4655 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4658 if (device->total_bytes > device->bytes_used)
4659 total_avail = device->total_bytes - device->bytes_used;
4663 /* If there is no space on this device, skip it. */
4664 if (total_avail == 0)
4667 ret = find_free_dev_extent(trans, device,
4668 max_stripe_size * dev_stripes,
4669 &dev_offset, &max_avail);
4670 if (ret && ret != -ENOSPC)
4674 max_avail = max_stripe_size * dev_stripes;
4676 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4677 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4679 "%s: devid %llu has no free space, have=%llu want=%u",
4680 __func__, device->devid, max_avail,
4681 BTRFS_STRIPE_LEN * dev_stripes);
4685 if (ndevs == fs_devices->rw_devices) {
4686 WARN(1, "%s: found more than %llu devices\n",
4687 __func__, fs_devices->rw_devices);
4690 devices_info[ndevs].dev_offset = dev_offset;
4691 devices_info[ndevs].max_avail = max_avail;
4692 devices_info[ndevs].total_avail = total_avail;
4693 devices_info[ndevs].dev = device;
4698 * now sort the devices by hole size / available space
4700 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4701 btrfs_cmp_device_info, NULL);
4703 /* round down to number of usable stripes */
4704 ndevs = round_down(ndevs, devs_increment);
4706 if (ndevs < devs_min) {
4708 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4710 "%s: not enough devices with free space: have=%d minimum required=%d",
4711 __func__, ndevs, devs_min);
4716 ndevs = min(ndevs, devs_max);
4719 * The primary goal is to maximize the number of stripes, so use as
4720 * many devices as possible, even if the stripes are not maximum sized.
4722 * The DUP profile stores more than one stripe per device, the
4723 * max_avail is the total size so we have to adjust.
4725 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4726 num_stripes = ndevs * dev_stripes;
4729 * this will have to be fixed for RAID1 and RAID10 over
4732 data_stripes = num_stripes / ncopies;
4734 if (type & BTRFS_BLOCK_GROUP_RAID5)
4735 data_stripes = num_stripes - 1;
4737 if (type & BTRFS_BLOCK_GROUP_RAID6)
4738 data_stripes = num_stripes - 2;
4741 * Use the number of data stripes to figure out how big this chunk
4742 * is really going to be in terms of logical address space,
4743 * and compare that answer with the max chunk size
4745 if (stripe_size * data_stripes > max_chunk_size) {
4746 stripe_size = div_u64(max_chunk_size, data_stripes);
4748 /* bump the answer up to a 16MB boundary */
4749 stripe_size = round_up(stripe_size, SZ_16M);
4752 * But don't go higher than the limits we found while searching
4755 stripe_size = min(devices_info[ndevs - 1].max_avail,
4759 /* align to BTRFS_STRIPE_LEN */
4760 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4762 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4767 map->num_stripes = num_stripes;
4769 for (i = 0; i < ndevs; ++i) {
4770 for (j = 0; j < dev_stripes; ++j) {
4771 int s = i * dev_stripes + j;
4772 map->stripes[s].dev = devices_info[i].dev;
4773 map->stripes[s].physical = devices_info[i].dev_offset +
4777 map->stripe_len = BTRFS_STRIPE_LEN;
4778 map->io_align = BTRFS_STRIPE_LEN;
4779 map->io_width = BTRFS_STRIPE_LEN;
4781 map->sub_stripes = sub_stripes;
4783 num_bytes = stripe_size * data_stripes;
4785 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4787 em = alloc_extent_map();
4793 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4794 em->map_lookup = map;
4796 em->len = num_bytes;
4797 em->block_start = 0;
4798 em->block_len = em->len;
4799 em->orig_block_len = stripe_size;
4801 em_tree = &info->mapping_tree.map_tree;
4802 write_lock(&em_tree->lock);
4803 ret = add_extent_mapping(em_tree, em, 0);
4805 write_unlock(&em_tree->lock);
4806 free_extent_map(em);
4810 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4811 refcount_inc(&em->refs);
4812 write_unlock(&em_tree->lock);
4814 ret = btrfs_make_block_group(trans, 0, type, start, num_bytes);
4816 goto error_del_extent;
4818 for (i = 0; i < map->num_stripes; i++) {
4819 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4820 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4823 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4825 free_extent_map(em);
4826 check_raid56_incompat_flag(info, type);
4828 kfree(devices_info);
4832 write_lock(&em_tree->lock);
4833 remove_extent_mapping(em_tree, em);
4834 write_unlock(&em_tree->lock);
4836 /* One for our allocation */
4837 free_extent_map(em);
4838 /* One for the tree reference */
4839 free_extent_map(em);
4840 /* One for the pending_chunks list reference */
4841 free_extent_map(em);
4843 kfree(devices_info);
4847 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4848 struct btrfs_fs_info *fs_info,
4849 u64 chunk_offset, u64 chunk_size)
4851 struct btrfs_root *extent_root = fs_info->extent_root;
4852 struct btrfs_root *chunk_root = fs_info->chunk_root;
4853 struct btrfs_key key;
4854 struct btrfs_device *device;
4855 struct btrfs_chunk *chunk;
4856 struct btrfs_stripe *stripe;
4857 struct extent_map *em;
4858 struct map_lookup *map;
4865 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4869 map = em->map_lookup;
4870 item_size = btrfs_chunk_item_size(map->num_stripes);
4871 stripe_size = em->orig_block_len;
4873 chunk = kzalloc(item_size, GFP_NOFS);
4880 * Take the device list mutex to prevent races with the final phase of
4881 * a device replace operation that replaces the device object associated
4882 * with the map's stripes, because the device object's id can change
4883 * at any time during that final phase of the device replace operation
4884 * (dev-replace.c:btrfs_dev_replace_finishing()).
4886 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4887 for (i = 0; i < map->num_stripes; i++) {
4888 device = map->stripes[i].dev;
4889 dev_offset = map->stripes[i].physical;
4891 ret = btrfs_update_device(trans, device);
4894 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4895 dev_offset, stripe_size);
4900 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4904 stripe = &chunk->stripe;
4905 for (i = 0; i < map->num_stripes; i++) {
4906 device = map->stripes[i].dev;
4907 dev_offset = map->stripes[i].physical;
4909 btrfs_set_stack_stripe_devid(stripe, device->devid);
4910 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4911 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4914 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4916 btrfs_set_stack_chunk_length(chunk, chunk_size);
4917 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4918 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4919 btrfs_set_stack_chunk_type(chunk, map->type);
4920 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4921 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4922 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4923 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4924 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4926 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4927 key.type = BTRFS_CHUNK_ITEM_KEY;
4928 key.offset = chunk_offset;
4930 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4931 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4933 * TODO: Cleanup of inserted chunk root in case of
4936 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4941 free_extent_map(em);
4946 * Chunk allocation falls into two parts. The first part does works
4947 * that make the new allocated chunk useable, but not do any operation
4948 * that modifies the chunk tree. The second part does the works that
4949 * require modifying the chunk tree. This division is important for the
4950 * bootstrap process of adding storage to a seed btrfs.
4952 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
4956 lockdep_assert_held(&trans->fs_info->chunk_mutex);
4957 chunk_offset = find_next_chunk(trans->fs_info);
4958 return __btrfs_alloc_chunk(trans, chunk_offset, type);
4961 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4962 struct btrfs_fs_info *fs_info)
4965 u64 sys_chunk_offset;
4969 chunk_offset = find_next_chunk(fs_info);
4970 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4971 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4975 sys_chunk_offset = find_next_chunk(fs_info);
4976 alloc_profile = btrfs_system_alloc_profile(fs_info);
4977 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
4981 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4985 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4986 BTRFS_BLOCK_GROUP_RAID10 |
4987 BTRFS_BLOCK_GROUP_RAID5 |
4988 BTRFS_BLOCK_GROUP_DUP)) {
4990 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4999 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5001 struct extent_map *em;
5002 struct map_lookup *map;
5007 em = get_chunk_map(fs_info, chunk_offset, 1);
5011 map = em->map_lookup;
5012 for (i = 0; i < map->num_stripes; i++) {
5013 if (test_bit(BTRFS_DEV_STATE_MISSING,
5014 &map->stripes[i].dev->dev_state)) {
5018 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5019 &map->stripes[i].dev->dev_state)) {
5026 * If the number of missing devices is larger than max errors,
5027 * we can not write the data into that chunk successfully, so
5030 if (miss_ndevs > btrfs_chunk_max_errors(map))
5033 free_extent_map(em);
5037 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5039 extent_map_tree_init(&tree->map_tree);
5042 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5044 struct extent_map *em;
5047 write_lock(&tree->map_tree.lock);
5048 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5050 remove_extent_mapping(&tree->map_tree, em);
5051 write_unlock(&tree->map_tree.lock);
5055 free_extent_map(em);
5056 /* once for the tree */
5057 free_extent_map(em);
5061 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5063 struct extent_map *em;
5064 struct map_lookup *map;
5067 em = get_chunk_map(fs_info, logical, len);
5070 * We could return errors for these cases, but that could get
5071 * ugly and we'd probably do the same thing which is just not do
5072 * anything else and exit, so return 1 so the callers don't try
5073 * to use other copies.
5077 map = em->map_lookup;
5078 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5079 ret = map->num_stripes;
5080 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5081 ret = map->sub_stripes;
5082 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5084 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5086 * There could be two corrupted data stripes, we need
5087 * to loop retry in order to rebuild the correct data.
5089 * Fail a stripe at a time on every retry except the
5090 * stripe under reconstruction.
5092 ret = map->num_stripes;
5095 free_extent_map(em);
5097 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5098 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5099 fs_info->dev_replace.tgtdev)
5101 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5106 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5109 struct extent_map *em;
5110 struct map_lookup *map;
5111 unsigned long len = fs_info->sectorsize;
5113 em = get_chunk_map(fs_info, logical, len);
5115 if (!WARN_ON(IS_ERR(em))) {
5116 map = em->map_lookup;
5117 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5118 len = map->stripe_len * nr_data_stripes(map);
5119 free_extent_map(em);
5124 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5126 struct extent_map *em;
5127 struct map_lookup *map;
5130 em = get_chunk_map(fs_info, logical, len);
5132 if(!WARN_ON(IS_ERR(em))) {
5133 map = em->map_lookup;
5134 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5136 free_extent_map(em);
5141 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5142 struct map_lookup *map, int first,
5143 int dev_replace_is_ongoing)
5147 int preferred_mirror;
5149 struct btrfs_device *srcdev;
5152 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5154 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5155 num_stripes = map->sub_stripes;
5157 num_stripes = map->num_stripes;
5159 preferred_mirror = first + current->pid % num_stripes;
5161 if (dev_replace_is_ongoing &&
5162 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5163 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5164 srcdev = fs_info->dev_replace.srcdev;
5169 * try to avoid the drive that is the source drive for a
5170 * dev-replace procedure, only choose it if no other non-missing
5171 * mirror is available
5173 for (tolerance = 0; tolerance < 2; tolerance++) {
5174 if (map->stripes[preferred_mirror].dev->bdev &&
5175 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5176 return preferred_mirror;
5177 for (i = first; i < first + num_stripes; i++) {
5178 if (map->stripes[i].dev->bdev &&
5179 (tolerance || map->stripes[i].dev != srcdev))
5184 /* we couldn't find one that doesn't fail. Just return something
5185 * and the io error handling code will clean up eventually
5187 return preferred_mirror;
5190 static inline int parity_smaller(u64 a, u64 b)
5195 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5196 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5198 struct btrfs_bio_stripe s;
5205 for (i = 0; i < num_stripes - 1; i++) {
5206 if (parity_smaller(bbio->raid_map[i],
5207 bbio->raid_map[i+1])) {
5208 s = bbio->stripes[i];
5209 l = bbio->raid_map[i];
5210 bbio->stripes[i] = bbio->stripes[i+1];
5211 bbio->raid_map[i] = bbio->raid_map[i+1];
5212 bbio->stripes[i+1] = s;
5213 bbio->raid_map[i+1] = l;
5221 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5223 struct btrfs_bio *bbio = kzalloc(
5224 /* the size of the btrfs_bio */
5225 sizeof(struct btrfs_bio) +
5226 /* plus the variable array for the stripes */
5227 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5228 /* plus the variable array for the tgt dev */
5229 sizeof(int) * (real_stripes) +
5231 * plus the raid_map, which includes both the tgt dev
5234 sizeof(u64) * (total_stripes),
5235 GFP_NOFS|__GFP_NOFAIL);
5237 atomic_set(&bbio->error, 0);
5238 refcount_set(&bbio->refs, 1);
5243 void btrfs_get_bbio(struct btrfs_bio *bbio)
5245 WARN_ON(!refcount_read(&bbio->refs));
5246 refcount_inc(&bbio->refs);
5249 void btrfs_put_bbio(struct btrfs_bio *bbio)
5253 if (refcount_dec_and_test(&bbio->refs))
5257 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5259 * Please note that, discard won't be sent to target device of device
5262 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5263 u64 logical, u64 length,
5264 struct btrfs_bio **bbio_ret)
5266 struct extent_map *em;
5267 struct map_lookup *map;
5268 struct btrfs_bio *bbio;
5272 u64 stripe_end_offset;
5279 u32 sub_stripes = 0;
5280 u64 stripes_per_dev = 0;
5281 u32 remaining_stripes = 0;
5282 u32 last_stripe = 0;
5286 /* discard always return a bbio */
5289 em = get_chunk_map(fs_info, logical, length);
5293 map = em->map_lookup;
5294 /* we don't discard raid56 yet */
5295 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5300 offset = logical - em->start;
5301 length = min_t(u64, em->len - offset, length);
5303 stripe_len = map->stripe_len;
5305 * stripe_nr counts the total number of stripes we have to stride
5306 * to get to this block
5308 stripe_nr = div64_u64(offset, stripe_len);
5310 /* stripe_offset is the offset of this block in its stripe */
5311 stripe_offset = offset - stripe_nr * stripe_len;
5313 stripe_nr_end = round_up(offset + length, map->stripe_len);
5314 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5315 stripe_cnt = stripe_nr_end - stripe_nr;
5316 stripe_end_offset = stripe_nr_end * map->stripe_len -
5319 * after this, stripe_nr is the number of stripes on this
5320 * device we have to walk to find the data, and stripe_index is
5321 * the number of our device in the stripe array
5325 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5326 BTRFS_BLOCK_GROUP_RAID10)) {
5327 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5330 sub_stripes = map->sub_stripes;
5332 factor = map->num_stripes / sub_stripes;
5333 num_stripes = min_t(u64, map->num_stripes,
5334 sub_stripes * stripe_cnt);
5335 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5336 stripe_index *= sub_stripes;
5337 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5338 &remaining_stripes);
5339 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5340 last_stripe *= sub_stripes;
5341 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5342 BTRFS_BLOCK_GROUP_DUP)) {
5343 num_stripes = map->num_stripes;
5345 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5349 bbio = alloc_btrfs_bio(num_stripes, 0);
5355 for (i = 0; i < num_stripes; i++) {
5356 bbio->stripes[i].physical =
5357 map->stripes[stripe_index].physical +
5358 stripe_offset + stripe_nr * map->stripe_len;
5359 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5361 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5362 BTRFS_BLOCK_GROUP_RAID10)) {
5363 bbio->stripes[i].length = stripes_per_dev *
5366 if (i / sub_stripes < remaining_stripes)
5367 bbio->stripes[i].length +=
5371 * Special for the first stripe and
5374 * |-------|...|-------|
5378 if (i < sub_stripes)
5379 bbio->stripes[i].length -=
5382 if (stripe_index >= last_stripe &&
5383 stripe_index <= (last_stripe +
5385 bbio->stripes[i].length -=
5388 if (i == sub_stripes - 1)
5391 bbio->stripes[i].length = length;
5395 if (stripe_index == map->num_stripes) {
5402 bbio->map_type = map->type;
5403 bbio->num_stripes = num_stripes;
5405 free_extent_map(em);
5410 * In dev-replace case, for repair case (that's the only case where the mirror
5411 * is selected explicitly when calling btrfs_map_block), blocks left of the
5412 * left cursor can also be read from the target drive.
5414 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5416 * For READ, it also needs to be supported using the same mirror number.
5418 * If the requested block is not left of the left cursor, EIO is returned. This
5419 * can happen because btrfs_num_copies() returns one more in the dev-replace
5422 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5423 u64 logical, u64 length,
5424 u64 srcdev_devid, int *mirror_num,
5427 struct btrfs_bio *bbio = NULL;
5429 int index_srcdev = 0;
5431 u64 physical_of_found = 0;
5435 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5436 logical, &length, &bbio, 0, 0);
5438 ASSERT(bbio == NULL);
5442 num_stripes = bbio->num_stripes;
5443 if (*mirror_num > num_stripes) {
5445 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5446 * that means that the requested area is not left of the left
5449 btrfs_put_bbio(bbio);
5454 * process the rest of the function using the mirror_num of the source
5455 * drive. Therefore look it up first. At the end, patch the device
5456 * pointer to the one of the target drive.
5458 for (i = 0; i < num_stripes; i++) {
5459 if (bbio->stripes[i].dev->devid != srcdev_devid)
5463 * In case of DUP, in order to keep it simple, only add the
5464 * mirror with the lowest physical address
5467 physical_of_found <= bbio->stripes[i].physical)
5472 physical_of_found = bbio->stripes[i].physical;
5475 btrfs_put_bbio(bbio);
5481 *mirror_num = index_srcdev + 1;
5482 *physical = physical_of_found;
5486 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5487 struct btrfs_bio **bbio_ret,
5488 struct btrfs_dev_replace *dev_replace,
5489 int *num_stripes_ret, int *max_errors_ret)
5491 struct btrfs_bio *bbio = *bbio_ret;
5492 u64 srcdev_devid = dev_replace->srcdev->devid;
5493 int tgtdev_indexes = 0;
5494 int num_stripes = *num_stripes_ret;
5495 int max_errors = *max_errors_ret;
5498 if (op == BTRFS_MAP_WRITE) {
5499 int index_where_to_add;
5502 * duplicate the write operations while the dev replace
5503 * procedure is running. Since the copying of the old disk to
5504 * the new disk takes place at run time while the filesystem is
5505 * mounted writable, the regular write operations to the old
5506 * disk have to be duplicated to go to the new disk as well.
5508 * Note that device->missing is handled by the caller, and that
5509 * the write to the old disk is already set up in the stripes
5512 index_where_to_add = num_stripes;
5513 for (i = 0; i < num_stripes; i++) {
5514 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5515 /* write to new disk, too */
5516 struct btrfs_bio_stripe *new =
5517 bbio->stripes + index_where_to_add;
5518 struct btrfs_bio_stripe *old =
5521 new->physical = old->physical;
5522 new->length = old->length;
5523 new->dev = dev_replace->tgtdev;
5524 bbio->tgtdev_map[i] = index_where_to_add;
5525 index_where_to_add++;
5530 num_stripes = index_where_to_add;
5531 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5532 int index_srcdev = 0;
5534 u64 physical_of_found = 0;
5537 * During the dev-replace procedure, the target drive can also
5538 * be used to read data in case it is needed to repair a corrupt
5539 * block elsewhere. This is possible if the requested area is
5540 * left of the left cursor. In this area, the target drive is a
5541 * full copy of the source drive.
5543 for (i = 0; i < num_stripes; i++) {
5544 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5546 * In case of DUP, in order to keep it simple,
5547 * only add the mirror with the lowest physical
5551 physical_of_found <=
5552 bbio->stripes[i].physical)
5556 physical_of_found = bbio->stripes[i].physical;
5560 struct btrfs_bio_stripe *tgtdev_stripe =
5561 bbio->stripes + num_stripes;
5563 tgtdev_stripe->physical = physical_of_found;
5564 tgtdev_stripe->length =
5565 bbio->stripes[index_srcdev].length;
5566 tgtdev_stripe->dev = dev_replace->tgtdev;
5567 bbio->tgtdev_map[index_srcdev] = num_stripes;
5574 *num_stripes_ret = num_stripes;
5575 *max_errors_ret = max_errors;
5576 bbio->num_tgtdevs = tgtdev_indexes;
5580 static bool need_full_stripe(enum btrfs_map_op op)
5582 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5585 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5586 enum btrfs_map_op op,
5587 u64 logical, u64 *length,
5588 struct btrfs_bio **bbio_ret,
5589 int mirror_num, int need_raid_map)
5591 struct extent_map *em;
5592 struct map_lookup *map;
5602 int tgtdev_indexes = 0;
5603 struct btrfs_bio *bbio = NULL;
5604 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5605 int dev_replace_is_ongoing = 0;
5606 int num_alloc_stripes;
5607 int patch_the_first_stripe_for_dev_replace = 0;
5608 u64 physical_to_patch_in_first_stripe = 0;
5609 u64 raid56_full_stripe_start = (u64)-1;
5611 if (op == BTRFS_MAP_DISCARD)
5612 return __btrfs_map_block_for_discard(fs_info, logical,
5615 em = get_chunk_map(fs_info, logical, *length);
5619 map = em->map_lookup;
5620 offset = logical - em->start;
5622 stripe_len = map->stripe_len;
5625 * stripe_nr counts the total number of stripes we have to stride
5626 * to get to this block
5628 stripe_nr = div64_u64(stripe_nr, stripe_len);
5630 stripe_offset = stripe_nr * stripe_len;
5631 if (offset < stripe_offset) {
5633 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5634 stripe_offset, offset, em->start, logical,
5636 free_extent_map(em);
5640 /* stripe_offset is the offset of this block in its stripe*/
5641 stripe_offset = offset - stripe_offset;
5643 /* if we're here for raid56, we need to know the stripe aligned start */
5644 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5645 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5646 raid56_full_stripe_start = offset;
5648 /* allow a write of a full stripe, but make sure we don't
5649 * allow straddling of stripes
5651 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5653 raid56_full_stripe_start *= full_stripe_len;
5656 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5658 /* For writes to RAID[56], allow a full stripeset across all disks.
5659 For other RAID types and for RAID[56] reads, just allow a single
5660 stripe (on a single disk). */
5661 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5662 (op == BTRFS_MAP_WRITE)) {
5663 max_len = stripe_len * nr_data_stripes(map) -
5664 (offset - raid56_full_stripe_start);
5666 /* we limit the length of each bio to what fits in a stripe */
5667 max_len = stripe_len - stripe_offset;
5669 *length = min_t(u64, em->len - offset, max_len);
5671 *length = em->len - offset;
5674 /* This is for when we're called from btrfs_merge_bio_hook() and all
5675 it cares about is the length */
5679 btrfs_dev_replace_read_lock(dev_replace);
5680 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5681 if (!dev_replace_is_ongoing)
5682 btrfs_dev_replace_read_unlock(dev_replace);
5684 btrfs_dev_replace_set_lock_blocking(dev_replace);
5686 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5687 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5688 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5689 dev_replace->srcdev->devid,
5691 &physical_to_patch_in_first_stripe);
5695 patch_the_first_stripe_for_dev_replace = 1;
5696 } else if (mirror_num > map->num_stripes) {
5702 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5703 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5705 if (!need_full_stripe(op))
5707 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5708 if (need_full_stripe(op))
5709 num_stripes = map->num_stripes;
5710 else if (mirror_num)
5711 stripe_index = mirror_num - 1;
5713 stripe_index = find_live_mirror(fs_info, map, 0,
5714 dev_replace_is_ongoing);
5715 mirror_num = stripe_index + 1;
5718 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5719 if (need_full_stripe(op)) {
5720 num_stripes = map->num_stripes;
5721 } else if (mirror_num) {
5722 stripe_index = mirror_num - 1;
5727 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5728 u32 factor = map->num_stripes / map->sub_stripes;
5730 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5731 stripe_index *= map->sub_stripes;
5733 if (need_full_stripe(op))
5734 num_stripes = map->sub_stripes;
5735 else if (mirror_num)
5736 stripe_index += mirror_num - 1;
5738 int old_stripe_index = stripe_index;
5739 stripe_index = find_live_mirror(fs_info, map,
5741 dev_replace_is_ongoing);
5742 mirror_num = stripe_index - old_stripe_index + 1;
5745 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5746 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5747 /* push stripe_nr back to the start of the full stripe */
5748 stripe_nr = div64_u64(raid56_full_stripe_start,
5749 stripe_len * nr_data_stripes(map));
5751 /* RAID[56] write or recovery. Return all stripes */
5752 num_stripes = map->num_stripes;
5753 max_errors = nr_parity_stripes(map);
5755 *length = map->stripe_len;
5760 * Mirror #0 or #1 means the original data block.
5761 * Mirror #2 is RAID5 parity block.
5762 * Mirror #3 is RAID6 Q block.
5764 stripe_nr = div_u64_rem(stripe_nr,
5765 nr_data_stripes(map), &stripe_index);
5767 stripe_index = nr_data_stripes(map) +
5770 /* We distribute the parity blocks across stripes */
5771 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5773 if (!need_full_stripe(op) && mirror_num <= 1)
5778 * after this, stripe_nr is the number of stripes on this
5779 * device we have to walk to find the data, and stripe_index is
5780 * the number of our device in the stripe array
5782 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5784 mirror_num = stripe_index + 1;
5786 if (stripe_index >= map->num_stripes) {
5788 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5789 stripe_index, map->num_stripes);
5794 num_alloc_stripes = num_stripes;
5795 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5796 if (op == BTRFS_MAP_WRITE)
5797 num_alloc_stripes <<= 1;
5798 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5799 num_alloc_stripes++;
5800 tgtdev_indexes = num_stripes;
5803 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5808 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5809 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5811 /* build raid_map */
5812 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5813 (need_full_stripe(op) || mirror_num > 1)) {
5817 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5818 sizeof(struct btrfs_bio_stripe) *
5820 sizeof(int) * tgtdev_indexes);
5822 /* Work out the disk rotation on this stripe-set */
5823 div_u64_rem(stripe_nr, num_stripes, &rot);
5825 /* Fill in the logical address of each stripe */
5826 tmp = stripe_nr * nr_data_stripes(map);
5827 for (i = 0; i < nr_data_stripes(map); i++)
5828 bbio->raid_map[(i+rot) % num_stripes] =
5829 em->start + (tmp + i) * map->stripe_len;
5831 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5832 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5833 bbio->raid_map[(i+rot+1) % num_stripes] =
5838 for (i = 0; i < num_stripes; i++) {
5839 bbio->stripes[i].physical =
5840 map->stripes[stripe_index].physical +
5842 stripe_nr * map->stripe_len;
5843 bbio->stripes[i].dev =
5844 map->stripes[stripe_index].dev;
5848 if (need_full_stripe(op))
5849 max_errors = btrfs_chunk_max_errors(map);
5852 sort_parity_stripes(bbio, num_stripes);
5854 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5855 need_full_stripe(op)) {
5856 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5861 bbio->map_type = map->type;
5862 bbio->num_stripes = num_stripes;
5863 bbio->max_errors = max_errors;
5864 bbio->mirror_num = mirror_num;
5867 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5868 * mirror_num == num_stripes + 1 && dev_replace target drive is
5869 * available as a mirror
5871 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5872 WARN_ON(num_stripes > 1);
5873 bbio->stripes[0].dev = dev_replace->tgtdev;
5874 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5875 bbio->mirror_num = map->num_stripes + 1;
5878 if (dev_replace_is_ongoing) {
5879 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5880 btrfs_dev_replace_read_unlock(dev_replace);
5882 free_extent_map(em);
5886 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5887 u64 logical, u64 *length,
5888 struct btrfs_bio **bbio_ret, int mirror_num)
5890 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5894 /* For Scrub/replace */
5895 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5896 u64 logical, u64 *length,
5897 struct btrfs_bio **bbio_ret)
5899 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5902 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
5903 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
5905 struct extent_map *em;
5906 struct map_lookup *map;
5914 em = get_chunk_map(fs_info, chunk_start, 1);
5918 map = em->map_lookup;
5920 rmap_len = map->stripe_len;
5922 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5923 length = div_u64(length, map->num_stripes / map->sub_stripes);
5924 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5925 length = div_u64(length, map->num_stripes);
5926 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5927 length = div_u64(length, nr_data_stripes(map));
5928 rmap_len = map->stripe_len * nr_data_stripes(map);
5931 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5932 BUG_ON(!buf); /* -ENOMEM */
5934 for (i = 0; i < map->num_stripes; i++) {
5935 if (map->stripes[i].physical > physical ||
5936 map->stripes[i].physical + length <= physical)
5939 stripe_nr = physical - map->stripes[i].physical;
5940 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5942 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5943 stripe_nr = stripe_nr * map->num_stripes + i;
5944 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5945 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5946 stripe_nr = stripe_nr * map->num_stripes + i;
5947 } /* else if RAID[56], multiply by nr_data_stripes().
5948 * Alternatively, just use rmap_len below instead of
5949 * map->stripe_len */
5951 bytenr = chunk_start + stripe_nr * rmap_len;
5952 WARN_ON(nr >= map->num_stripes);
5953 for (j = 0; j < nr; j++) {
5954 if (buf[j] == bytenr)
5958 WARN_ON(nr >= map->num_stripes);
5965 *stripe_len = rmap_len;
5967 free_extent_map(em);
5971 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5973 bio->bi_private = bbio->private;
5974 bio->bi_end_io = bbio->end_io;
5977 btrfs_put_bbio(bbio);
5980 static void btrfs_end_bio(struct bio *bio)
5982 struct btrfs_bio *bbio = bio->bi_private;
5983 int is_orig_bio = 0;
5985 if (bio->bi_status) {
5986 atomic_inc(&bbio->error);
5987 if (bio->bi_status == BLK_STS_IOERR ||
5988 bio->bi_status == BLK_STS_TARGET) {
5989 unsigned int stripe_index =
5990 btrfs_io_bio(bio)->stripe_index;
5991 struct btrfs_device *dev;
5993 BUG_ON(stripe_index >= bbio->num_stripes);
5994 dev = bbio->stripes[stripe_index].dev;
5996 if (bio_op(bio) == REQ_OP_WRITE)
5997 btrfs_dev_stat_inc_and_print(dev,
5998 BTRFS_DEV_STAT_WRITE_ERRS);
6000 btrfs_dev_stat_inc_and_print(dev,
6001 BTRFS_DEV_STAT_READ_ERRS);
6002 if (bio->bi_opf & REQ_PREFLUSH)
6003 btrfs_dev_stat_inc_and_print(dev,
6004 BTRFS_DEV_STAT_FLUSH_ERRS);
6009 if (bio == bbio->orig_bio)
6012 btrfs_bio_counter_dec(bbio->fs_info);
6014 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6017 bio = bbio->orig_bio;
6020 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6021 /* only send an error to the higher layers if it is
6022 * beyond the tolerance of the btrfs bio
6024 if (atomic_read(&bbio->error) > bbio->max_errors) {
6025 bio->bi_status = BLK_STS_IOERR;
6028 * this bio is actually up to date, we didn't
6029 * go over the max number of errors
6031 bio->bi_status = BLK_STS_OK;
6034 btrfs_end_bbio(bbio, bio);
6035 } else if (!is_orig_bio) {
6041 * see run_scheduled_bios for a description of why bios are collected for
6044 * This will add one bio to the pending list for a device and make sure
6045 * the work struct is scheduled.
6047 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6050 struct btrfs_fs_info *fs_info = device->fs_info;
6051 int should_queue = 1;
6052 struct btrfs_pending_bios *pending_bios;
6054 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6060 /* don't bother with additional async steps for reads, right now */
6061 if (bio_op(bio) == REQ_OP_READ) {
6062 btrfsic_submit_bio(bio);
6066 WARN_ON(bio->bi_next);
6067 bio->bi_next = NULL;
6069 spin_lock(&device->io_lock);
6070 if (op_is_sync(bio->bi_opf))
6071 pending_bios = &device->pending_sync_bios;
6073 pending_bios = &device->pending_bios;
6075 if (pending_bios->tail)
6076 pending_bios->tail->bi_next = bio;
6078 pending_bios->tail = bio;
6079 if (!pending_bios->head)
6080 pending_bios->head = bio;
6081 if (device->running_pending)
6084 spin_unlock(&device->io_lock);
6087 btrfs_queue_work(fs_info->submit_workers, &device->work);
6090 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6091 u64 physical, int dev_nr, int async)
6093 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6094 struct btrfs_fs_info *fs_info = bbio->fs_info;
6096 bio->bi_private = bbio;
6097 btrfs_io_bio(bio)->stripe_index = dev_nr;
6098 bio->bi_end_io = btrfs_end_bio;
6099 bio->bi_iter.bi_sector = physical >> 9;
6102 struct rcu_string *name;
6105 name = rcu_dereference(dev->name);
6106 btrfs_debug(fs_info,
6107 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6108 bio_op(bio), bio->bi_opf,
6109 (u64)bio->bi_iter.bi_sector,
6110 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6111 bio->bi_iter.bi_size);
6115 bio_set_dev(bio, dev->bdev);
6117 btrfs_bio_counter_inc_noblocked(fs_info);
6120 btrfs_schedule_bio(dev, bio);
6122 btrfsic_submit_bio(bio);
6125 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6127 atomic_inc(&bbio->error);
6128 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6129 /* Should be the original bio. */
6130 WARN_ON(bio != bbio->orig_bio);
6132 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6133 bio->bi_iter.bi_sector = logical >> 9;
6134 if (atomic_read(&bbio->error) > bbio->max_errors)
6135 bio->bi_status = BLK_STS_IOERR;
6137 bio->bi_status = BLK_STS_OK;
6138 btrfs_end_bbio(bbio, bio);
6142 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6143 int mirror_num, int async_submit)
6145 struct btrfs_device *dev;
6146 struct bio *first_bio = bio;
6147 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6153 struct btrfs_bio *bbio = NULL;
6155 length = bio->bi_iter.bi_size;
6156 map_length = length;
6158 btrfs_bio_counter_inc_blocked(fs_info);
6159 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6160 &map_length, &bbio, mirror_num, 1);
6162 btrfs_bio_counter_dec(fs_info);
6163 return errno_to_blk_status(ret);
6166 total_devs = bbio->num_stripes;
6167 bbio->orig_bio = first_bio;
6168 bbio->private = first_bio->bi_private;
6169 bbio->end_io = first_bio->bi_end_io;
6170 bbio->fs_info = fs_info;
6171 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6173 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6174 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6175 /* In this case, map_length has been set to the length of
6176 a single stripe; not the whole write */
6177 if (bio_op(bio) == REQ_OP_WRITE) {
6178 ret = raid56_parity_write(fs_info, bio, bbio,
6181 ret = raid56_parity_recover(fs_info, bio, bbio,
6182 map_length, mirror_num, 1);
6185 btrfs_bio_counter_dec(fs_info);
6186 return errno_to_blk_status(ret);
6189 if (map_length < length) {
6191 "mapping failed logical %llu bio len %llu len %llu",
6192 logical, length, map_length);
6196 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6197 dev = bbio->stripes[dev_nr].dev;
6198 if (!dev || !dev->bdev ||
6199 (bio_op(first_bio) == REQ_OP_WRITE &&
6200 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6201 bbio_error(bbio, first_bio, logical);
6205 if (dev_nr < total_devs - 1)
6206 bio = btrfs_bio_clone(first_bio);
6210 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6211 dev_nr, async_submit);
6213 btrfs_bio_counter_dec(fs_info);
6217 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6220 struct btrfs_device *device;
6221 struct btrfs_fs_devices *cur_devices;
6223 cur_devices = fs_info->fs_devices;
6224 while (cur_devices) {
6226 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6227 device = find_device(cur_devices, devid, uuid);
6231 cur_devices = cur_devices->seed;
6236 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6237 u64 devid, u8 *dev_uuid)
6239 struct btrfs_device *device;
6241 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6245 list_add(&device->dev_list, &fs_devices->devices);
6246 device->fs_devices = fs_devices;
6247 fs_devices->num_devices++;
6249 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6250 fs_devices->missing_devices++;
6256 * btrfs_alloc_device - allocate struct btrfs_device
6257 * @fs_info: used only for generating a new devid, can be NULL if
6258 * devid is provided (i.e. @devid != NULL).
6259 * @devid: a pointer to devid for this device. If NULL a new devid
6261 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6264 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6265 * on error. Returned struct is not linked onto any lists and must be
6266 * destroyed with btrfs_free_device.
6268 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6272 struct btrfs_device *dev;
6275 if (WARN_ON(!devid && !fs_info))
6276 return ERR_PTR(-EINVAL);
6278 dev = __alloc_device();
6287 ret = find_next_devid(fs_info, &tmp);
6289 btrfs_free_device(dev);
6290 return ERR_PTR(ret);
6296 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6298 generate_random_uuid(dev->uuid);
6300 btrfs_init_work(&dev->work, btrfs_submit_helper,
6301 pending_bios_fn, NULL, NULL);
6306 /* Return -EIO if any error, otherwise return 0. */
6307 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6308 struct extent_buffer *leaf,
6309 struct btrfs_chunk *chunk, u64 logical)
6317 length = btrfs_chunk_length(leaf, chunk);
6318 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6319 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6320 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6321 type = btrfs_chunk_type(leaf, chunk);
6324 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6328 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6329 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6332 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6333 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6334 btrfs_chunk_sector_size(leaf, chunk));
6337 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6338 btrfs_err(fs_info, "invalid chunk length %llu", length);
6341 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6342 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6346 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6348 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6349 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6350 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6351 btrfs_chunk_type(leaf, chunk));
6354 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6355 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6356 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6357 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6358 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6359 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6360 num_stripes != 1)) {
6362 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6363 num_stripes, sub_stripes,
6364 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6371 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6372 u64 devid, u8 *uuid, bool error)
6375 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6378 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6382 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6383 struct extent_buffer *leaf,
6384 struct btrfs_chunk *chunk)
6386 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6387 struct map_lookup *map;
6388 struct extent_map *em;
6392 u8 uuid[BTRFS_UUID_SIZE];
6397 logical = key->offset;
6398 length = btrfs_chunk_length(leaf, chunk);
6399 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6401 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6405 read_lock(&map_tree->map_tree.lock);
6406 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6407 read_unlock(&map_tree->map_tree.lock);
6409 /* already mapped? */
6410 if (em && em->start <= logical && em->start + em->len > logical) {
6411 free_extent_map(em);
6414 free_extent_map(em);
6417 em = alloc_extent_map();
6420 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6422 free_extent_map(em);
6426 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6427 em->map_lookup = map;
6428 em->start = logical;
6431 em->block_start = 0;
6432 em->block_len = em->len;
6434 map->num_stripes = num_stripes;
6435 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6436 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6437 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6438 map->type = btrfs_chunk_type(leaf, chunk);
6439 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6440 for (i = 0; i < num_stripes; i++) {
6441 map->stripes[i].physical =
6442 btrfs_stripe_offset_nr(leaf, chunk, i);
6443 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6444 read_extent_buffer(leaf, uuid, (unsigned long)
6445 btrfs_stripe_dev_uuid_nr(chunk, i),
6447 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6449 if (!map->stripes[i].dev &&
6450 !btrfs_test_opt(fs_info, DEGRADED)) {
6451 free_extent_map(em);
6452 btrfs_report_missing_device(fs_info, devid, uuid, true);
6455 if (!map->stripes[i].dev) {
6456 map->stripes[i].dev =
6457 add_missing_dev(fs_info->fs_devices, devid,
6459 if (IS_ERR(map->stripes[i].dev)) {
6460 free_extent_map(em);
6462 "failed to init missing dev %llu: %ld",
6463 devid, PTR_ERR(map->stripes[i].dev));
6464 return PTR_ERR(map->stripes[i].dev);
6466 btrfs_report_missing_device(fs_info, devid, uuid, false);
6468 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6469 &(map->stripes[i].dev->dev_state));
6473 write_lock(&map_tree->map_tree.lock);
6474 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6475 write_unlock(&map_tree->map_tree.lock);
6476 BUG_ON(ret); /* Tree corruption */
6477 free_extent_map(em);
6482 static void fill_device_from_item(struct extent_buffer *leaf,
6483 struct btrfs_dev_item *dev_item,
6484 struct btrfs_device *device)
6488 device->devid = btrfs_device_id(leaf, dev_item);
6489 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6490 device->total_bytes = device->disk_total_bytes;
6491 device->commit_total_bytes = device->disk_total_bytes;
6492 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6493 device->commit_bytes_used = device->bytes_used;
6494 device->type = btrfs_device_type(leaf, dev_item);
6495 device->io_align = btrfs_device_io_align(leaf, dev_item);
6496 device->io_width = btrfs_device_io_width(leaf, dev_item);
6497 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6498 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6499 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6501 ptr = btrfs_device_uuid(dev_item);
6502 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6505 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6508 struct btrfs_fs_devices *fs_devices;
6511 lockdep_assert_held(&uuid_mutex);
6514 fs_devices = fs_info->fs_devices->seed;
6515 while (fs_devices) {
6516 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6519 fs_devices = fs_devices->seed;
6522 fs_devices = find_fsid(fsid);
6524 if (!btrfs_test_opt(fs_info, DEGRADED))
6525 return ERR_PTR(-ENOENT);
6527 fs_devices = alloc_fs_devices(fsid);
6528 if (IS_ERR(fs_devices))
6531 fs_devices->seeding = 1;
6532 fs_devices->opened = 1;
6536 fs_devices = clone_fs_devices(fs_devices);
6537 if (IS_ERR(fs_devices))
6540 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6542 free_fs_devices(fs_devices);
6543 fs_devices = ERR_PTR(ret);
6547 if (!fs_devices->seeding) {
6548 close_fs_devices(fs_devices);
6549 free_fs_devices(fs_devices);
6550 fs_devices = ERR_PTR(-EINVAL);
6554 fs_devices->seed = fs_info->fs_devices->seed;
6555 fs_info->fs_devices->seed = fs_devices;
6560 static int read_one_dev(struct btrfs_fs_info *fs_info,
6561 struct extent_buffer *leaf,
6562 struct btrfs_dev_item *dev_item)
6564 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6565 struct btrfs_device *device;
6568 u8 fs_uuid[BTRFS_FSID_SIZE];
6569 u8 dev_uuid[BTRFS_UUID_SIZE];
6571 devid = btrfs_device_id(leaf, dev_item);
6572 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6574 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6577 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6578 fs_devices = open_seed_devices(fs_info, fs_uuid);
6579 if (IS_ERR(fs_devices))
6580 return PTR_ERR(fs_devices);
6583 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6585 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6586 btrfs_report_missing_device(fs_info, devid,
6591 device = add_missing_dev(fs_devices, devid, dev_uuid);
6592 if (IS_ERR(device)) {
6594 "failed to add missing dev %llu: %ld",
6595 devid, PTR_ERR(device));
6596 return PTR_ERR(device);
6598 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6600 if (!device->bdev) {
6601 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6602 btrfs_report_missing_device(fs_info,
6603 devid, dev_uuid, true);
6606 btrfs_report_missing_device(fs_info, devid,
6610 if (!device->bdev &&
6611 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6613 * this happens when a device that was properly setup
6614 * in the device info lists suddenly goes bad.
6615 * device->bdev is NULL, and so we have to set
6616 * device->missing to one here
6618 device->fs_devices->missing_devices++;
6619 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6622 /* Move the device to its own fs_devices */
6623 if (device->fs_devices != fs_devices) {
6624 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6625 &device->dev_state));
6627 list_move(&device->dev_list, &fs_devices->devices);
6628 device->fs_devices->num_devices--;
6629 fs_devices->num_devices++;
6631 device->fs_devices->missing_devices--;
6632 fs_devices->missing_devices++;
6634 device->fs_devices = fs_devices;
6638 if (device->fs_devices != fs_info->fs_devices) {
6639 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6640 if (device->generation !=
6641 btrfs_device_generation(leaf, dev_item))
6645 fill_device_from_item(leaf, dev_item, device);
6646 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6647 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6648 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6649 device->fs_devices->total_rw_bytes += device->total_bytes;
6650 atomic64_add(device->total_bytes - device->bytes_used,
6651 &fs_info->free_chunk_space);
6657 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6659 struct btrfs_root *root = fs_info->tree_root;
6660 struct btrfs_super_block *super_copy = fs_info->super_copy;
6661 struct extent_buffer *sb;
6662 struct btrfs_disk_key *disk_key;
6663 struct btrfs_chunk *chunk;
6665 unsigned long sb_array_offset;
6672 struct btrfs_key key;
6674 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6676 * This will create extent buffer of nodesize, superblock size is
6677 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6678 * overallocate but we can keep it as-is, only the first page is used.
6680 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6683 set_extent_buffer_uptodate(sb);
6684 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6686 * The sb extent buffer is artificial and just used to read the system array.
6687 * set_extent_buffer_uptodate() call does not properly mark all it's
6688 * pages up-to-date when the page is larger: extent does not cover the
6689 * whole page and consequently check_page_uptodate does not find all
6690 * the page's extents up-to-date (the hole beyond sb),
6691 * write_extent_buffer then triggers a WARN_ON.
6693 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6694 * but sb spans only this function. Add an explicit SetPageUptodate call
6695 * to silence the warning eg. on PowerPC 64.
6697 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6698 SetPageUptodate(sb->pages[0]);
6700 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6701 array_size = btrfs_super_sys_array_size(super_copy);
6703 array_ptr = super_copy->sys_chunk_array;
6704 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6707 while (cur_offset < array_size) {
6708 disk_key = (struct btrfs_disk_key *)array_ptr;
6709 len = sizeof(*disk_key);
6710 if (cur_offset + len > array_size)
6711 goto out_short_read;
6713 btrfs_disk_key_to_cpu(&key, disk_key);
6716 sb_array_offset += len;
6719 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6720 chunk = (struct btrfs_chunk *)sb_array_offset;
6722 * At least one btrfs_chunk with one stripe must be
6723 * present, exact stripe count check comes afterwards
6725 len = btrfs_chunk_item_size(1);
6726 if (cur_offset + len > array_size)
6727 goto out_short_read;
6729 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6732 "invalid number of stripes %u in sys_array at offset %u",
6733 num_stripes, cur_offset);
6738 type = btrfs_chunk_type(sb, chunk);
6739 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6741 "invalid chunk type %llu in sys_array at offset %u",
6747 len = btrfs_chunk_item_size(num_stripes);
6748 if (cur_offset + len > array_size)
6749 goto out_short_read;
6751 ret = read_one_chunk(fs_info, &key, sb, chunk);
6756 "unexpected item type %u in sys_array at offset %u",
6757 (u32)key.type, cur_offset);
6762 sb_array_offset += len;
6765 clear_extent_buffer_uptodate(sb);
6766 free_extent_buffer_stale(sb);
6770 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6772 clear_extent_buffer_uptodate(sb);
6773 free_extent_buffer_stale(sb);
6778 * Check if all chunks in the fs are OK for read-write degraded mount
6780 * If the @failing_dev is specified, it's accounted as missing.
6782 * Return true if all chunks meet the minimal RW mount requirements.
6783 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6785 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6786 struct btrfs_device *failing_dev)
6788 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6789 struct extent_map *em;
6793 read_lock(&map_tree->map_tree.lock);
6794 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6795 read_unlock(&map_tree->map_tree.lock);
6796 /* No chunk at all? Return false anyway */
6802 struct map_lookup *map;
6807 map = em->map_lookup;
6809 btrfs_get_num_tolerated_disk_barrier_failures(
6811 for (i = 0; i < map->num_stripes; i++) {
6812 struct btrfs_device *dev = map->stripes[i].dev;
6814 if (!dev || !dev->bdev ||
6815 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6816 dev->last_flush_error)
6818 else if (failing_dev && failing_dev == dev)
6821 if (missing > max_tolerated) {
6824 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6825 em->start, missing, max_tolerated);
6826 free_extent_map(em);
6830 next_start = extent_map_end(em);
6831 free_extent_map(em);
6833 read_lock(&map_tree->map_tree.lock);
6834 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6835 (u64)(-1) - next_start);
6836 read_unlock(&map_tree->map_tree.lock);
6842 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6844 struct btrfs_root *root = fs_info->chunk_root;
6845 struct btrfs_path *path;
6846 struct extent_buffer *leaf;
6847 struct btrfs_key key;
6848 struct btrfs_key found_key;
6853 path = btrfs_alloc_path();
6858 * uuid_mutex is needed only if we are mounting a sprout FS
6859 * otherwise we don't need it.
6861 mutex_lock(&uuid_mutex);
6862 mutex_lock(&fs_info->chunk_mutex);
6865 * Read all device items, and then all the chunk items. All
6866 * device items are found before any chunk item (their object id
6867 * is smaller than the lowest possible object id for a chunk
6868 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6870 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6873 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6877 leaf = path->nodes[0];
6878 slot = path->slots[0];
6879 if (slot >= btrfs_header_nritems(leaf)) {
6880 ret = btrfs_next_leaf(root, path);
6887 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6888 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6889 struct btrfs_dev_item *dev_item;
6890 dev_item = btrfs_item_ptr(leaf, slot,
6891 struct btrfs_dev_item);
6892 ret = read_one_dev(fs_info, leaf, dev_item);
6896 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6897 struct btrfs_chunk *chunk;
6898 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6899 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6907 * After loading chunk tree, we've got all device information,
6908 * do another round of validation checks.
6910 if (total_dev != fs_info->fs_devices->total_devices) {
6912 "super_num_devices %llu mismatch with num_devices %llu found here",
6913 btrfs_super_num_devices(fs_info->super_copy),
6918 if (btrfs_super_total_bytes(fs_info->super_copy) <
6919 fs_info->fs_devices->total_rw_bytes) {
6921 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6922 btrfs_super_total_bytes(fs_info->super_copy),
6923 fs_info->fs_devices->total_rw_bytes);
6929 mutex_unlock(&fs_info->chunk_mutex);
6930 mutex_unlock(&uuid_mutex);
6932 btrfs_free_path(path);
6936 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6938 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6939 struct btrfs_device *device;
6941 while (fs_devices) {
6942 mutex_lock(&fs_devices->device_list_mutex);
6943 list_for_each_entry(device, &fs_devices->devices, dev_list)
6944 device->fs_info = fs_info;
6945 mutex_unlock(&fs_devices->device_list_mutex);
6947 fs_devices = fs_devices->seed;
6951 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6955 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6956 btrfs_dev_stat_reset(dev, i);
6959 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6961 struct btrfs_key key;
6962 struct btrfs_key found_key;
6963 struct btrfs_root *dev_root = fs_info->dev_root;
6964 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6965 struct extent_buffer *eb;
6968 struct btrfs_device *device;
6969 struct btrfs_path *path = NULL;
6972 path = btrfs_alloc_path();
6978 mutex_lock(&fs_devices->device_list_mutex);
6979 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6981 struct btrfs_dev_stats_item *ptr;
6983 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6984 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6985 key.offset = device->devid;
6986 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6988 __btrfs_reset_dev_stats(device);
6989 device->dev_stats_valid = 1;
6990 btrfs_release_path(path);
6993 slot = path->slots[0];
6994 eb = path->nodes[0];
6995 btrfs_item_key_to_cpu(eb, &found_key, slot);
6996 item_size = btrfs_item_size_nr(eb, slot);
6998 ptr = btrfs_item_ptr(eb, slot,
6999 struct btrfs_dev_stats_item);
7001 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7002 if (item_size >= (1 + i) * sizeof(__le64))
7003 btrfs_dev_stat_set(device, i,
7004 btrfs_dev_stats_value(eb, ptr, i));
7006 btrfs_dev_stat_reset(device, i);
7009 device->dev_stats_valid = 1;
7010 btrfs_dev_stat_print_on_load(device);
7011 btrfs_release_path(path);
7013 mutex_unlock(&fs_devices->device_list_mutex);
7016 btrfs_free_path(path);
7017 return ret < 0 ? ret : 0;
7020 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7021 struct btrfs_fs_info *fs_info,
7022 struct btrfs_device *device)
7024 struct btrfs_root *dev_root = fs_info->dev_root;
7025 struct btrfs_path *path;
7026 struct btrfs_key key;
7027 struct extent_buffer *eb;
7028 struct btrfs_dev_stats_item *ptr;
7032 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7033 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7034 key.offset = device->devid;
7036 path = btrfs_alloc_path();
7039 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7041 btrfs_warn_in_rcu(fs_info,
7042 "error %d while searching for dev_stats item for device %s",
7043 ret, rcu_str_deref(device->name));
7048 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7049 /* need to delete old one and insert a new one */
7050 ret = btrfs_del_item(trans, dev_root, path);
7052 btrfs_warn_in_rcu(fs_info,
7053 "delete too small dev_stats item for device %s failed %d",
7054 rcu_str_deref(device->name), ret);
7061 /* need to insert a new item */
7062 btrfs_release_path(path);
7063 ret = btrfs_insert_empty_item(trans, dev_root, path,
7064 &key, sizeof(*ptr));
7066 btrfs_warn_in_rcu(fs_info,
7067 "insert dev_stats item for device %s failed %d",
7068 rcu_str_deref(device->name), ret);
7073 eb = path->nodes[0];
7074 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7075 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7076 btrfs_set_dev_stats_value(eb, ptr, i,
7077 btrfs_dev_stat_read(device, i));
7078 btrfs_mark_buffer_dirty(eb);
7081 btrfs_free_path(path);
7086 * called from commit_transaction. Writes all changed device stats to disk.
7088 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7089 struct btrfs_fs_info *fs_info)
7091 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7092 struct btrfs_device *device;
7096 mutex_lock(&fs_devices->device_list_mutex);
7097 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7098 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7099 if (!device->dev_stats_valid || stats_cnt == 0)
7104 * There is a LOAD-LOAD control dependency between the value of
7105 * dev_stats_ccnt and updating the on-disk values which requires
7106 * reading the in-memory counters. Such control dependencies
7107 * require explicit read memory barriers.
7109 * This memory barriers pairs with smp_mb__before_atomic in
7110 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7111 * barrier implied by atomic_xchg in
7112 * btrfs_dev_stats_read_and_reset
7116 ret = update_dev_stat_item(trans, fs_info, device);
7118 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7120 mutex_unlock(&fs_devices->device_list_mutex);
7125 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7127 btrfs_dev_stat_inc(dev, index);
7128 btrfs_dev_stat_print_on_error(dev);
7131 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7133 if (!dev->dev_stats_valid)
7135 btrfs_err_rl_in_rcu(dev->fs_info,
7136 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7137 rcu_str_deref(dev->name),
7138 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7139 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7140 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7141 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7142 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7145 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7149 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7150 if (btrfs_dev_stat_read(dev, i) != 0)
7152 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7153 return; /* all values == 0, suppress message */
7155 btrfs_info_in_rcu(dev->fs_info,
7156 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7157 rcu_str_deref(dev->name),
7158 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7159 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7160 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7161 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7162 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7165 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7166 struct btrfs_ioctl_get_dev_stats *stats)
7168 struct btrfs_device *dev;
7169 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7172 mutex_lock(&fs_devices->device_list_mutex);
7173 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7174 mutex_unlock(&fs_devices->device_list_mutex);
7177 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7179 } else if (!dev->dev_stats_valid) {
7180 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7182 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7183 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7184 if (stats->nr_items > i)
7186 btrfs_dev_stat_read_and_reset(dev, i);
7188 btrfs_dev_stat_reset(dev, i);
7191 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7192 if (stats->nr_items > i)
7193 stats->values[i] = btrfs_dev_stat_read(dev, i);
7195 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7196 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7200 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7202 struct buffer_head *bh;
7203 struct btrfs_super_block *disk_super;
7209 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7212 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7215 disk_super = (struct btrfs_super_block *)bh->b_data;
7217 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7218 set_buffer_dirty(bh);
7219 sync_dirty_buffer(bh);
7223 /* Notify udev that device has changed */
7224 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7226 /* Update ctime/mtime for device path for libblkid */
7227 update_dev_time(device_path);
7231 * Update the size of all devices, which is used for writing out the
7234 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7236 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7237 struct btrfs_device *curr, *next;
7239 if (list_empty(&fs_devices->resized_devices))
7242 mutex_lock(&fs_devices->device_list_mutex);
7243 mutex_lock(&fs_info->chunk_mutex);
7244 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7246 list_del_init(&curr->resized_list);
7247 curr->commit_total_bytes = curr->disk_total_bytes;
7249 mutex_unlock(&fs_info->chunk_mutex);
7250 mutex_unlock(&fs_devices->device_list_mutex);
7253 /* Must be invoked during the transaction commit */
7254 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7256 struct btrfs_fs_info *fs_info = trans->fs_info;
7257 struct extent_map *em;
7258 struct map_lookup *map;
7259 struct btrfs_device *dev;
7262 if (list_empty(&trans->pending_chunks))
7265 /* In order to kick the device replace finish process */
7266 mutex_lock(&fs_info->chunk_mutex);
7267 list_for_each_entry(em, &trans->pending_chunks, list) {
7268 map = em->map_lookup;
7270 for (i = 0; i < map->num_stripes; i++) {
7271 dev = map->stripes[i].dev;
7272 dev->commit_bytes_used = dev->bytes_used;
7275 mutex_unlock(&fs_info->chunk_mutex);
7278 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7280 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7281 while (fs_devices) {
7282 fs_devices->fs_info = fs_info;
7283 fs_devices = fs_devices->seed;
7287 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7289 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7290 while (fs_devices) {
7291 fs_devices->fs_info = NULL;
7292 fs_devices = fs_devices->seed;