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.rb_root);
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_device *device)
1672 struct btrfs_path *path;
1673 struct btrfs_dev_item *dev_item;
1674 struct extent_buffer *leaf;
1675 struct btrfs_key key;
1678 path = btrfs_alloc_path();
1682 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1683 key.type = BTRFS_DEV_ITEM_KEY;
1684 key.offset = device->devid;
1686 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1687 &key, sizeof(*dev_item));
1691 leaf = path->nodes[0];
1692 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1694 btrfs_set_device_id(leaf, dev_item, device->devid);
1695 btrfs_set_device_generation(leaf, dev_item, 0);
1696 btrfs_set_device_type(leaf, dev_item, device->type);
1697 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1698 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1699 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1700 btrfs_set_device_total_bytes(leaf, dev_item,
1701 btrfs_device_get_disk_total_bytes(device));
1702 btrfs_set_device_bytes_used(leaf, dev_item,
1703 btrfs_device_get_bytes_used(device));
1704 btrfs_set_device_group(leaf, dev_item, 0);
1705 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1706 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1707 btrfs_set_device_start_offset(leaf, dev_item, 0);
1709 ptr = btrfs_device_uuid(dev_item);
1710 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1711 ptr = btrfs_device_fsid(dev_item);
1712 write_extent_buffer(leaf, trans->fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1713 btrfs_mark_buffer_dirty(leaf);
1717 btrfs_free_path(path);
1722 * Function to update ctime/mtime for a given device path.
1723 * Mainly used for ctime/mtime based probe like libblkid.
1725 static void update_dev_time(const char *path_name)
1729 filp = filp_open(path_name, O_RDWR, 0);
1732 file_update_time(filp);
1733 filp_close(filp, NULL);
1736 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1737 struct btrfs_device *device)
1739 struct btrfs_root *root = fs_info->chunk_root;
1741 struct btrfs_path *path;
1742 struct btrfs_key key;
1743 struct btrfs_trans_handle *trans;
1745 path = btrfs_alloc_path();
1749 trans = btrfs_start_transaction(root, 0);
1750 if (IS_ERR(trans)) {
1751 btrfs_free_path(path);
1752 return PTR_ERR(trans);
1754 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1755 key.type = BTRFS_DEV_ITEM_KEY;
1756 key.offset = device->devid;
1758 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762 btrfs_abort_transaction(trans, ret);
1763 btrfs_end_transaction(trans);
1767 ret = btrfs_del_item(trans, root, path);
1769 btrfs_abort_transaction(trans, ret);
1770 btrfs_end_transaction(trans);
1774 btrfs_free_path(path);
1776 ret = btrfs_commit_transaction(trans);
1781 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1782 * filesystem. It's up to the caller to adjust that number regarding eg. device
1785 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1793 seq = read_seqbegin(&fs_info->profiles_lock);
1795 all_avail = fs_info->avail_data_alloc_bits |
1796 fs_info->avail_system_alloc_bits |
1797 fs_info->avail_metadata_alloc_bits;
1798 } while (read_seqretry(&fs_info->profiles_lock, seq));
1800 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1801 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1804 if (num_devices < btrfs_raid_array[i].devs_min) {
1805 int ret = btrfs_raid_array[i].mindev_error;
1815 static struct btrfs_device * btrfs_find_next_active_device(
1816 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1818 struct btrfs_device *next_device;
1820 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1821 if (next_device != device &&
1822 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1823 && next_device->bdev)
1831 * Helper function to check if the given device is part of s_bdev / latest_bdev
1832 * and replace it with the provided or the next active device, in the context
1833 * where this function called, there should be always be another device (or
1834 * this_dev) which is active.
1836 void btrfs_assign_next_active_device(struct btrfs_device *device,
1837 struct btrfs_device *this_dev)
1839 struct btrfs_fs_info *fs_info = device->fs_info;
1840 struct btrfs_device *next_device;
1843 next_device = this_dev;
1845 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1847 ASSERT(next_device);
1849 if (fs_info->sb->s_bdev &&
1850 (fs_info->sb->s_bdev == device->bdev))
1851 fs_info->sb->s_bdev = next_device->bdev;
1853 if (fs_info->fs_devices->latest_bdev == device->bdev)
1854 fs_info->fs_devices->latest_bdev = next_device->bdev;
1858 * Return btrfs_fs_devices::num_devices excluding the device that's being
1859 * currently replaced.
1861 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1863 u64 num_devices = fs_info->fs_devices->num_devices;
1865 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1866 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1867 ASSERT(num_devices > 1);
1870 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1875 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1878 struct btrfs_device *device;
1879 struct btrfs_fs_devices *cur_devices;
1880 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1884 mutex_lock(&uuid_mutex);
1886 num_devices = btrfs_num_devices(fs_info);
1888 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1892 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
1894 if (IS_ERR(device)) {
1895 if (PTR_ERR(device) == -ENOENT &&
1896 strcmp(device_path, "missing") == 0)
1897 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1899 ret = PTR_ERR(device);
1903 if (btrfs_pinned_by_swapfile(fs_info, device)) {
1904 btrfs_warn_in_rcu(fs_info,
1905 "cannot remove device %s (devid %llu) due to active swapfile",
1906 rcu_str_deref(device->name), device->devid);
1911 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1912 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1916 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1917 fs_info->fs_devices->rw_devices == 1) {
1918 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1922 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1923 mutex_lock(&fs_info->chunk_mutex);
1924 list_del_init(&device->dev_alloc_list);
1925 device->fs_devices->rw_devices--;
1926 mutex_unlock(&fs_info->chunk_mutex);
1929 mutex_unlock(&uuid_mutex);
1930 ret = btrfs_shrink_device(device, 0);
1931 mutex_lock(&uuid_mutex);
1936 * TODO: the superblock still includes this device in its num_devices
1937 * counter although write_all_supers() is not locked out. This
1938 * could give a filesystem state which requires a degraded mount.
1940 ret = btrfs_rm_dev_item(fs_info, device);
1944 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1945 btrfs_scrub_cancel_dev(fs_info, device);
1948 * the device list mutex makes sure that we don't change
1949 * the device list while someone else is writing out all
1950 * the device supers. Whoever is writing all supers, should
1951 * lock the device list mutex before getting the number of
1952 * devices in the super block (super_copy). Conversely,
1953 * whoever updates the number of devices in the super block
1954 * (super_copy) should hold the device list mutex.
1958 * In normal cases the cur_devices == fs_devices. But in case
1959 * of deleting a seed device, the cur_devices should point to
1960 * its own fs_devices listed under the fs_devices->seed.
1962 cur_devices = device->fs_devices;
1963 mutex_lock(&fs_devices->device_list_mutex);
1964 list_del_rcu(&device->dev_list);
1966 cur_devices->num_devices--;
1967 cur_devices->total_devices--;
1968 /* Update total_devices of the parent fs_devices if it's seed */
1969 if (cur_devices != fs_devices)
1970 fs_devices->total_devices--;
1972 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1973 cur_devices->missing_devices--;
1975 btrfs_assign_next_active_device(device, NULL);
1978 cur_devices->open_devices--;
1979 /* remove sysfs entry */
1980 btrfs_sysfs_rm_device_link(fs_devices, device);
1983 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1984 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1985 mutex_unlock(&fs_devices->device_list_mutex);
1988 * at this point, the device is zero sized and detached from
1989 * the devices list. All that's left is to zero out the old
1990 * supers and free the device.
1992 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1993 btrfs_scratch_superblocks(device->bdev, device->name->str);
1995 btrfs_close_bdev(device);
1996 call_rcu(&device->rcu, free_device_rcu);
1998 if (cur_devices->open_devices == 0) {
1999 while (fs_devices) {
2000 if (fs_devices->seed == cur_devices) {
2001 fs_devices->seed = cur_devices->seed;
2004 fs_devices = fs_devices->seed;
2006 cur_devices->seed = NULL;
2007 close_fs_devices(cur_devices);
2008 free_fs_devices(cur_devices);
2012 mutex_unlock(&uuid_mutex);
2016 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2017 mutex_lock(&fs_info->chunk_mutex);
2018 list_add(&device->dev_alloc_list,
2019 &fs_devices->alloc_list);
2020 device->fs_devices->rw_devices++;
2021 mutex_unlock(&fs_info->chunk_mutex);
2026 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2028 struct btrfs_fs_devices *fs_devices;
2030 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2033 * in case of fs with no seed, srcdev->fs_devices will point
2034 * to fs_devices of fs_info. However when the dev being replaced is
2035 * a seed dev it will point to the seed's local fs_devices. In short
2036 * srcdev will have its correct fs_devices in both the cases.
2038 fs_devices = srcdev->fs_devices;
2040 list_del_rcu(&srcdev->dev_list);
2041 list_del(&srcdev->dev_alloc_list);
2042 fs_devices->num_devices--;
2043 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2044 fs_devices->missing_devices--;
2046 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2047 fs_devices->rw_devices--;
2050 fs_devices->open_devices--;
2053 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2054 struct btrfs_device *srcdev)
2056 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2058 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2059 /* zero out the old super if it is writable */
2060 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2063 btrfs_close_bdev(srcdev);
2064 call_rcu(&srcdev->rcu, free_device_rcu);
2066 /* if this is no devs we rather delete the fs_devices */
2067 if (!fs_devices->num_devices) {
2068 struct btrfs_fs_devices *tmp_fs_devices;
2071 * On a mounted FS, num_devices can't be zero unless it's a
2072 * seed. In case of a seed device being replaced, the replace
2073 * target added to the sprout FS, so there will be no more
2074 * device left under the seed FS.
2076 ASSERT(fs_devices->seeding);
2078 tmp_fs_devices = fs_info->fs_devices;
2079 while (tmp_fs_devices) {
2080 if (tmp_fs_devices->seed == fs_devices) {
2081 tmp_fs_devices->seed = fs_devices->seed;
2084 tmp_fs_devices = tmp_fs_devices->seed;
2086 fs_devices->seed = NULL;
2087 close_fs_devices(fs_devices);
2088 free_fs_devices(fs_devices);
2092 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2094 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2097 mutex_lock(&fs_devices->device_list_mutex);
2099 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2102 fs_devices->open_devices--;
2104 fs_devices->num_devices--;
2106 btrfs_assign_next_active_device(tgtdev, NULL);
2108 list_del_rcu(&tgtdev->dev_list);
2110 mutex_unlock(&fs_devices->device_list_mutex);
2113 * The update_dev_time() with in btrfs_scratch_superblocks()
2114 * may lead to a call to btrfs_show_devname() which will try
2115 * to hold device_list_mutex. And here this device
2116 * is already out of device list, so we don't have to hold
2117 * the device_list_mutex lock.
2119 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2121 btrfs_close_bdev(tgtdev);
2122 call_rcu(&tgtdev->rcu, free_device_rcu);
2125 static struct btrfs_device *btrfs_find_device_by_path(
2126 struct btrfs_fs_info *fs_info, const char *device_path)
2129 struct btrfs_super_block *disk_super;
2132 struct block_device *bdev;
2133 struct buffer_head *bh;
2134 struct btrfs_device *device;
2136 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2137 fs_info->bdev_holder, 0, &bdev, &bh);
2139 return ERR_PTR(ret);
2140 disk_super = (struct btrfs_super_block *)bh->b_data;
2141 devid = btrfs_stack_device_id(&disk_super->dev_item);
2142 dev_uuid = disk_super->dev_item.uuid;
2143 device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2146 device = ERR_PTR(-ENOENT);
2147 blkdev_put(bdev, FMODE_READ);
2151 static struct btrfs_device *btrfs_find_device_missing_or_by_path(
2152 struct btrfs_fs_info *fs_info, const char *device_path)
2154 struct btrfs_device *device = NULL;
2155 if (strcmp(device_path, "missing") == 0) {
2156 struct list_head *devices;
2157 struct btrfs_device *tmp;
2159 devices = &fs_info->fs_devices->devices;
2160 list_for_each_entry(tmp, devices, dev_list) {
2161 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2162 &tmp->dev_state) && !tmp->bdev) {
2169 return ERR_PTR(-ENOENT);
2171 device = btrfs_find_device_by_path(fs_info, device_path);
2178 * Lookup a device given by device id, or the path if the id is 0.
2180 struct btrfs_device *btrfs_find_device_by_devspec(
2181 struct btrfs_fs_info *fs_info, u64 devid, const char *devpath)
2183 struct btrfs_device *device;
2186 device = btrfs_find_device(fs_info, devid, NULL, NULL);
2188 return ERR_PTR(-ENOENT);
2190 if (!devpath || !devpath[0])
2191 return ERR_PTR(-EINVAL);
2192 device = btrfs_find_device_missing_or_by_path(fs_info, devpath);
2198 * does all the dirty work required for changing file system's UUID.
2200 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2202 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2203 struct btrfs_fs_devices *old_devices;
2204 struct btrfs_fs_devices *seed_devices;
2205 struct btrfs_super_block *disk_super = fs_info->super_copy;
2206 struct btrfs_device *device;
2209 lockdep_assert_held(&uuid_mutex);
2210 if (!fs_devices->seeding)
2213 seed_devices = alloc_fs_devices(NULL);
2214 if (IS_ERR(seed_devices))
2215 return PTR_ERR(seed_devices);
2217 old_devices = clone_fs_devices(fs_devices);
2218 if (IS_ERR(old_devices)) {
2219 kfree(seed_devices);
2220 return PTR_ERR(old_devices);
2223 list_add(&old_devices->fs_list, &fs_uuids);
2225 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2226 seed_devices->opened = 1;
2227 INIT_LIST_HEAD(&seed_devices->devices);
2228 INIT_LIST_HEAD(&seed_devices->alloc_list);
2229 mutex_init(&seed_devices->device_list_mutex);
2231 mutex_lock(&fs_devices->device_list_mutex);
2232 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2234 list_for_each_entry(device, &seed_devices->devices, dev_list)
2235 device->fs_devices = seed_devices;
2237 mutex_lock(&fs_info->chunk_mutex);
2238 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2239 mutex_unlock(&fs_info->chunk_mutex);
2241 fs_devices->seeding = 0;
2242 fs_devices->num_devices = 0;
2243 fs_devices->open_devices = 0;
2244 fs_devices->missing_devices = 0;
2245 fs_devices->rotating = 0;
2246 fs_devices->seed = seed_devices;
2248 generate_random_uuid(fs_devices->fsid);
2249 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2250 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2251 mutex_unlock(&fs_devices->device_list_mutex);
2253 super_flags = btrfs_super_flags(disk_super) &
2254 ~BTRFS_SUPER_FLAG_SEEDING;
2255 btrfs_set_super_flags(disk_super, super_flags);
2261 * Store the expected generation for seed devices in device items.
2263 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2264 struct btrfs_fs_info *fs_info)
2266 struct btrfs_root *root = fs_info->chunk_root;
2267 struct btrfs_path *path;
2268 struct extent_buffer *leaf;
2269 struct btrfs_dev_item *dev_item;
2270 struct btrfs_device *device;
2271 struct btrfs_key key;
2272 u8 fs_uuid[BTRFS_FSID_SIZE];
2273 u8 dev_uuid[BTRFS_UUID_SIZE];
2277 path = btrfs_alloc_path();
2281 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2283 key.type = BTRFS_DEV_ITEM_KEY;
2286 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2290 leaf = path->nodes[0];
2292 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2293 ret = btrfs_next_leaf(root, path);
2298 leaf = path->nodes[0];
2299 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2300 btrfs_release_path(path);
2304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2305 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2306 key.type != BTRFS_DEV_ITEM_KEY)
2309 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2310 struct btrfs_dev_item);
2311 devid = btrfs_device_id(leaf, dev_item);
2312 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2314 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2316 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2317 BUG_ON(!device); /* Logic error */
2319 if (device->fs_devices->seeding) {
2320 btrfs_set_device_generation(leaf, dev_item,
2321 device->generation);
2322 btrfs_mark_buffer_dirty(leaf);
2330 btrfs_free_path(path);
2334 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2336 struct btrfs_root *root = fs_info->dev_root;
2337 struct request_queue *q;
2338 struct btrfs_trans_handle *trans;
2339 struct btrfs_device *device;
2340 struct block_device *bdev;
2341 struct super_block *sb = fs_info->sb;
2342 struct rcu_string *name;
2343 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2344 u64 orig_super_total_bytes;
2345 u64 orig_super_num_devices;
2346 int seeding_dev = 0;
2348 bool unlocked = false;
2350 if (sb_rdonly(sb) && !fs_devices->seeding)
2353 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2354 fs_info->bdev_holder);
2356 return PTR_ERR(bdev);
2358 if (fs_devices->seeding) {
2360 down_write(&sb->s_umount);
2361 mutex_lock(&uuid_mutex);
2364 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2366 mutex_lock(&fs_devices->device_list_mutex);
2367 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2368 if (device->bdev == bdev) {
2371 &fs_devices->device_list_mutex);
2375 mutex_unlock(&fs_devices->device_list_mutex);
2377 device = btrfs_alloc_device(fs_info, NULL, NULL);
2378 if (IS_ERR(device)) {
2379 /* we can safely leave the fs_devices entry around */
2380 ret = PTR_ERR(device);
2384 name = rcu_string_strdup(device_path, GFP_KERNEL);
2387 goto error_free_device;
2389 rcu_assign_pointer(device->name, name);
2391 trans = btrfs_start_transaction(root, 0);
2392 if (IS_ERR(trans)) {
2393 ret = PTR_ERR(trans);
2394 goto error_free_device;
2397 q = bdev_get_queue(bdev);
2398 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2399 device->generation = trans->transid;
2400 device->io_width = fs_info->sectorsize;
2401 device->io_align = fs_info->sectorsize;
2402 device->sector_size = fs_info->sectorsize;
2403 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2404 fs_info->sectorsize);
2405 device->disk_total_bytes = device->total_bytes;
2406 device->commit_total_bytes = device->total_bytes;
2407 device->fs_info = fs_info;
2408 device->bdev = bdev;
2409 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2410 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2411 device->mode = FMODE_EXCL;
2412 device->dev_stats_valid = 1;
2413 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2416 sb->s_flags &= ~SB_RDONLY;
2417 ret = btrfs_prepare_sprout(fs_info);
2419 btrfs_abort_transaction(trans, ret);
2424 device->fs_devices = fs_devices;
2426 mutex_lock(&fs_devices->device_list_mutex);
2427 mutex_lock(&fs_info->chunk_mutex);
2428 list_add_rcu(&device->dev_list, &fs_devices->devices);
2429 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2430 fs_devices->num_devices++;
2431 fs_devices->open_devices++;
2432 fs_devices->rw_devices++;
2433 fs_devices->total_devices++;
2434 fs_devices->total_rw_bytes += device->total_bytes;
2436 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2438 if (!blk_queue_nonrot(q))
2439 fs_devices->rotating = 1;
2441 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2442 btrfs_set_super_total_bytes(fs_info->super_copy,
2443 round_down(orig_super_total_bytes + device->total_bytes,
2444 fs_info->sectorsize));
2446 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2447 btrfs_set_super_num_devices(fs_info->super_copy,
2448 orig_super_num_devices + 1);
2450 /* add sysfs device entry */
2451 btrfs_sysfs_add_device_link(fs_devices, device);
2454 * we've got more storage, clear any full flags on the space
2457 btrfs_clear_space_info_full(fs_info);
2459 mutex_unlock(&fs_info->chunk_mutex);
2460 mutex_unlock(&fs_devices->device_list_mutex);
2463 mutex_lock(&fs_info->chunk_mutex);
2464 ret = init_first_rw_device(trans, fs_info);
2465 mutex_unlock(&fs_info->chunk_mutex);
2467 btrfs_abort_transaction(trans, ret);
2472 ret = btrfs_add_dev_item(trans, device);
2474 btrfs_abort_transaction(trans, ret);
2479 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2481 ret = btrfs_finish_sprout(trans, fs_info);
2483 btrfs_abort_transaction(trans, ret);
2487 /* Sprouting would change fsid of the mounted root,
2488 * so rename the fsid on the sysfs
2490 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2492 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2494 "sysfs: failed to create fsid for sprout");
2497 ret = btrfs_commit_transaction(trans);
2500 mutex_unlock(&uuid_mutex);
2501 up_write(&sb->s_umount);
2504 if (ret) /* transaction commit */
2507 ret = btrfs_relocate_sys_chunks(fs_info);
2509 btrfs_handle_fs_error(fs_info, ret,
2510 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2511 trans = btrfs_attach_transaction(root);
2512 if (IS_ERR(trans)) {
2513 if (PTR_ERR(trans) == -ENOENT)
2515 ret = PTR_ERR(trans);
2519 ret = btrfs_commit_transaction(trans);
2522 /* Update ctime/mtime for libblkid */
2523 update_dev_time(device_path);
2527 btrfs_sysfs_rm_device_link(fs_devices, device);
2528 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2529 mutex_lock(&fs_info->chunk_mutex);
2530 list_del_rcu(&device->dev_list);
2531 list_del(&device->dev_alloc_list);
2532 fs_info->fs_devices->num_devices--;
2533 fs_info->fs_devices->open_devices--;
2534 fs_info->fs_devices->rw_devices--;
2535 fs_info->fs_devices->total_devices--;
2536 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2537 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2538 btrfs_set_super_total_bytes(fs_info->super_copy,
2539 orig_super_total_bytes);
2540 btrfs_set_super_num_devices(fs_info->super_copy,
2541 orig_super_num_devices);
2542 mutex_unlock(&fs_info->chunk_mutex);
2543 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2546 sb->s_flags |= SB_RDONLY;
2548 btrfs_end_transaction(trans);
2550 btrfs_free_device(device);
2552 blkdev_put(bdev, FMODE_EXCL);
2553 if (seeding_dev && !unlocked) {
2554 mutex_unlock(&uuid_mutex);
2555 up_write(&sb->s_umount);
2560 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2561 struct btrfs_device *device)
2564 struct btrfs_path *path;
2565 struct btrfs_root *root = device->fs_info->chunk_root;
2566 struct btrfs_dev_item *dev_item;
2567 struct extent_buffer *leaf;
2568 struct btrfs_key key;
2570 path = btrfs_alloc_path();
2574 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2575 key.type = BTRFS_DEV_ITEM_KEY;
2576 key.offset = device->devid;
2578 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2587 leaf = path->nodes[0];
2588 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2590 btrfs_set_device_id(leaf, dev_item, device->devid);
2591 btrfs_set_device_type(leaf, dev_item, device->type);
2592 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2593 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2594 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2595 btrfs_set_device_total_bytes(leaf, dev_item,
2596 btrfs_device_get_disk_total_bytes(device));
2597 btrfs_set_device_bytes_used(leaf, dev_item,
2598 btrfs_device_get_bytes_used(device));
2599 btrfs_mark_buffer_dirty(leaf);
2602 btrfs_free_path(path);
2606 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2607 struct btrfs_device *device, u64 new_size)
2609 struct btrfs_fs_info *fs_info = device->fs_info;
2610 struct btrfs_super_block *super_copy = fs_info->super_copy;
2611 struct btrfs_fs_devices *fs_devices;
2615 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2618 new_size = round_down(new_size, fs_info->sectorsize);
2620 mutex_lock(&fs_info->chunk_mutex);
2621 old_total = btrfs_super_total_bytes(super_copy);
2622 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2624 if (new_size <= device->total_bytes ||
2625 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2626 mutex_unlock(&fs_info->chunk_mutex);
2630 fs_devices = fs_info->fs_devices;
2632 btrfs_set_super_total_bytes(super_copy,
2633 round_down(old_total + diff, fs_info->sectorsize));
2634 device->fs_devices->total_rw_bytes += diff;
2636 btrfs_device_set_total_bytes(device, new_size);
2637 btrfs_device_set_disk_total_bytes(device, new_size);
2638 btrfs_clear_space_info_full(device->fs_info);
2639 if (list_empty(&device->resized_list))
2640 list_add_tail(&device->resized_list,
2641 &fs_devices->resized_devices);
2642 mutex_unlock(&fs_info->chunk_mutex);
2644 return btrfs_update_device(trans, device);
2647 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2649 struct btrfs_fs_info *fs_info = trans->fs_info;
2650 struct btrfs_root *root = fs_info->chunk_root;
2652 struct btrfs_path *path;
2653 struct btrfs_key key;
2655 path = btrfs_alloc_path();
2659 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2660 key.offset = chunk_offset;
2661 key.type = BTRFS_CHUNK_ITEM_KEY;
2663 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2666 else if (ret > 0) { /* Logic error or corruption */
2667 btrfs_handle_fs_error(fs_info, -ENOENT,
2668 "Failed lookup while freeing chunk.");
2673 ret = btrfs_del_item(trans, root, path);
2675 btrfs_handle_fs_error(fs_info, ret,
2676 "Failed to delete chunk item.");
2678 btrfs_free_path(path);
2682 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2684 struct btrfs_super_block *super_copy = fs_info->super_copy;
2685 struct btrfs_disk_key *disk_key;
2686 struct btrfs_chunk *chunk;
2693 struct btrfs_key key;
2695 mutex_lock(&fs_info->chunk_mutex);
2696 array_size = btrfs_super_sys_array_size(super_copy);
2698 ptr = super_copy->sys_chunk_array;
2701 while (cur < array_size) {
2702 disk_key = (struct btrfs_disk_key *)ptr;
2703 btrfs_disk_key_to_cpu(&key, disk_key);
2705 len = sizeof(*disk_key);
2707 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2708 chunk = (struct btrfs_chunk *)(ptr + len);
2709 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2710 len += btrfs_chunk_item_size(num_stripes);
2715 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2716 key.offset == chunk_offset) {
2717 memmove(ptr, ptr + len, array_size - (cur + len));
2719 btrfs_set_super_sys_array_size(super_copy, array_size);
2725 mutex_unlock(&fs_info->chunk_mutex);
2729 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2730 u64 logical, u64 length)
2732 struct extent_map_tree *em_tree;
2733 struct extent_map *em;
2735 em_tree = &fs_info->mapping_tree.map_tree;
2736 read_lock(&em_tree->lock);
2737 em = lookup_extent_mapping(em_tree, logical, length);
2738 read_unlock(&em_tree->lock);
2741 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2743 return ERR_PTR(-EINVAL);
2746 if (em->start > logical || em->start + em->len < logical) {
2748 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2749 logical, length, em->start, em->start + em->len);
2750 free_extent_map(em);
2751 return ERR_PTR(-EINVAL);
2754 /* callers are responsible for dropping em's ref. */
2758 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2760 struct btrfs_fs_info *fs_info = trans->fs_info;
2761 struct extent_map *em;
2762 struct map_lookup *map;
2763 u64 dev_extent_len = 0;
2765 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2767 em = get_chunk_map(fs_info, chunk_offset, 1);
2770 * This is a logic error, but we don't want to just rely on the
2771 * user having built with ASSERT enabled, so if ASSERT doesn't
2772 * do anything we still error out.
2777 map = em->map_lookup;
2778 mutex_lock(&fs_info->chunk_mutex);
2779 check_system_chunk(trans, map->type);
2780 mutex_unlock(&fs_info->chunk_mutex);
2783 * Take the device list mutex to prevent races with the final phase of
2784 * a device replace operation that replaces the device object associated
2785 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2787 mutex_lock(&fs_devices->device_list_mutex);
2788 for (i = 0; i < map->num_stripes; i++) {
2789 struct btrfs_device *device = map->stripes[i].dev;
2790 ret = btrfs_free_dev_extent(trans, device,
2791 map->stripes[i].physical,
2794 mutex_unlock(&fs_devices->device_list_mutex);
2795 btrfs_abort_transaction(trans, ret);
2799 if (device->bytes_used > 0) {
2800 mutex_lock(&fs_info->chunk_mutex);
2801 btrfs_device_set_bytes_used(device,
2802 device->bytes_used - dev_extent_len);
2803 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2804 btrfs_clear_space_info_full(fs_info);
2805 mutex_unlock(&fs_info->chunk_mutex);
2808 if (map->stripes[i].dev) {
2809 ret = btrfs_update_device(trans, map->stripes[i].dev);
2811 mutex_unlock(&fs_devices->device_list_mutex);
2812 btrfs_abort_transaction(trans, ret);
2817 mutex_unlock(&fs_devices->device_list_mutex);
2819 ret = btrfs_free_chunk(trans, chunk_offset);
2821 btrfs_abort_transaction(trans, ret);
2825 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2827 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2828 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2830 btrfs_abort_transaction(trans, ret);
2835 ret = btrfs_remove_block_group(trans, chunk_offset, em);
2837 btrfs_abort_transaction(trans, ret);
2843 free_extent_map(em);
2847 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2849 struct btrfs_root *root = fs_info->chunk_root;
2850 struct btrfs_trans_handle *trans;
2854 * Prevent races with automatic removal of unused block groups.
2855 * After we relocate and before we remove the chunk with offset
2856 * chunk_offset, automatic removal of the block group can kick in,
2857 * resulting in a failure when calling btrfs_remove_chunk() below.
2859 * Make sure to acquire this mutex before doing a tree search (dev
2860 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2861 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2862 * we release the path used to search the chunk/dev tree and before
2863 * the current task acquires this mutex and calls us.
2865 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2867 ret = btrfs_can_relocate(fs_info, chunk_offset);
2871 /* step one, relocate all the extents inside this chunk */
2872 btrfs_scrub_pause(fs_info);
2873 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2874 btrfs_scrub_continue(fs_info);
2879 * We add the kobjects here (and after forcing data chunk creation)
2880 * since relocation is the only place we'll create chunks of a new
2881 * type at runtime. The only place where we'll remove the last
2882 * chunk of a type is the call immediately below this one. Even
2883 * so, we're protected against races with the cleaner thread since
2884 * we're covered by the delete_unused_bgs_mutex.
2886 btrfs_add_raid_kobjects(fs_info);
2888 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2890 if (IS_ERR(trans)) {
2891 ret = PTR_ERR(trans);
2892 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2897 * step two, delete the device extents and the
2898 * chunk tree entries
2900 ret = btrfs_remove_chunk(trans, chunk_offset);
2901 btrfs_end_transaction(trans);
2905 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2907 struct btrfs_root *chunk_root = fs_info->chunk_root;
2908 struct btrfs_path *path;
2909 struct extent_buffer *leaf;
2910 struct btrfs_chunk *chunk;
2911 struct btrfs_key key;
2912 struct btrfs_key found_key;
2914 bool retried = false;
2918 path = btrfs_alloc_path();
2923 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2924 key.offset = (u64)-1;
2925 key.type = BTRFS_CHUNK_ITEM_KEY;
2928 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2929 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2931 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2934 BUG_ON(ret == 0); /* Corruption */
2936 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2939 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2945 leaf = path->nodes[0];
2946 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2948 chunk = btrfs_item_ptr(leaf, path->slots[0],
2949 struct btrfs_chunk);
2950 chunk_type = btrfs_chunk_type(leaf, chunk);
2951 btrfs_release_path(path);
2953 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2954 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2960 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2962 if (found_key.offset == 0)
2964 key.offset = found_key.offset - 1;
2967 if (failed && !retried) {
2971 } else if (WARN_ON(failed && retried)) {
2975 btrfs_free_path(path);
2980 * return 1 : allocate a data chunk successfully,
2981 * return <0: errors during allocating a data chunk,
2982 * return 0 : no need to allocate a data chunk.
2984 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
2987 struct btrfs_block_group_cache *cache;
2991 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2993 chunk_type = cache->flags;
2994 btrfs_put_block_group(cache);
2996 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
2997 spin_lock(&fs_info->data_sinfo->lock);
2998 bytes_used = fs_info->data_sinfo->bytes_used;
2999 spin_unlock(&fs_info->data_sinfo->lock);
3002 struct btrfs_trans_handle *trans;
3005 trans = btrfs_join_transaction(fs_info->tree_root);
3007 return PTR_ERR(trans);
3009 ret = btrfs_force_chunk_alloc(trans,
3010 BTRFS_BLOCK_GROUP_DATA);
3011 btrfs_end_transaction(trans);
3015 btrfs_add_raid_kobjects(fs_info);
3023 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3024 struct btrfs_balance_control *bctl)
3026 struct btrfs_root *root = fs_info->tree_root;
3027 struct btrfs_trans_handle *trans;
3028 struct btrfs_balance_item *item;
3029 struct btrfs_disk_balance_args disk_bargs;
3030 struct btrfs_path *path;
3031 struct extent_buffer *leaf;
3032 struct btrfs_key key;
3035 path = btrfs_alloc_path();
3039 trans = btrfs_start_transaction(root, 0);
3040 if (IS_ERR(trans)) {
3041 btrfs_free_path(path);
3042 return PTR_ERR(trans);
3045 key.objectid = BTRFS_BALANCE_OBJECTID;
3046 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3049 ret = btrfs_insert_empty_item(trans, root, path, &key,
3054 leaf = path->nodes[0];
3055 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3057 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3059 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3060 btrfs_set_balance_data(leaf, item, &disk_bargs);
3061 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3062 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3063 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3064 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3066 btrfs_set_balance_flags(leaf, item, bctl->flags);
3068 btrfs_mark_buffer_dirty(leaf);
3070 btrfs_free_path(path);
3071 err = btrfs_commit_transaction(trans);
3077 static int del_balance_item(struct btrfs_fs_info *fs_info)
3079 struct btrfs_root *root = fs_info->tree_root;
3080 struct btrfs_trans_handle *trans;
3081 struct btrfs_path *path;
3082 struct btrfs_key key;
3085 path = btrfs_alloc_path();
3089 trans = btrfs_start_transaction(root, 0);
3090 if (IS_ERR(trans)) {
3091 btrfs_free_path(path);
3092 return PTR_ERR(trans);
3095 key.objectid = BTRFS_BALANCE_OBJECTID;
3096 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3099 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3107 ret = btrfs_del_item(trans, root, path);
3109 btrfs_free_path(path);
3110 err = btrfs_commit_transaction(trans);
3117 * This is a heuristic used to reduce the number of chunks balanced on
3118 * resume after balance was interrupted.
3120 static void update_balance_args(struct btrfs_balance_control *bctl)
3123 * Turn on soft mode for chunk types that were being converted.
3125 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3126 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3127 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3128 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3129 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3130 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3133 * Turn on usage filter if is not already used. The idea is
3134 * that chunks that we have already balanced should be
3135 * reasonably full. Don't do it for chunks that are being
3136 * converted - that will keep us from relocating unconverted
3137 * (albeit full) chunks.
3139 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3140 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3141 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3142 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3143 bctl->data.usage = 90;
3145 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3146 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3147 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3148 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3149 bctl->sys.usage = 90;
3151 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3152 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3153 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3154 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3155 bctl->meta.usage = 90;
3160 * Clear the balance status in fs_info and delete the balance item from disk.
3162 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3164 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3167 BUG_ON(!fs_info->balance_ctl);
3169 spin_lock(&fs_info->balance_lock);
3170 fs_info->balance_ctl = NULL;
3171 spin_unlock(&fs_info->balance_lock);
3174 ret = del_balance_item(fs_info);
3176 btrfs_handle_fs_error(fs_info, ret, NULL);
3180 * Balance filters. Return 1 if chunk should be filtered out
3181 * (should not be balanced).
3183 static int chunk_profiles_filter(u64 chunk_type,
3184 struct btrfs_balance_args *bargs)
3186 chunk_type = chunk_to_extended(chunk_type) &
3187 BTRFS_EXTENDED_PROFILE_MASK;
3189 if (bargs->profiles & chunk_type)
3195 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3196 struct btrfs_balance_args *bargs)
3198 struct btrfs_block_group_cache *cache;
3200 u64 user_thresh_min;
3201 u64 user_thresh_max;
3204 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3205 chunk_used = btrfs_block_group_used(&cache->item);
3207 if (bargs->usage_min == 0)
3208 user_thresh_min = 0;
3210 user_thresh_min = div_factor_fine(cache->key.offset,
3213 if (bargs->usage_max == 0)
3214 user_thresh_max = 1;
3215 else if (bargs->usage_max > 100)
3216 user_thresh_max = cache->key.offset;
3218 user_thresh_max = div_factor_fine(cache->key.offset,
3221 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3224 btrfs_put_block_group(cache);
3228 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3229 u64 chunk_offset, struct btrfs_balance_args *bargs)
3231 struct btrfs_block_group_cache *cache;
3232 u64 chunk_used, user_thresh;
3235 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3236 chunk_used = btrfs_block_group_used(&cache->item);
3238 if (bargs->usage_min == 0)
3240 else if (bargs->usage > 100)
3241 user_thresh = cache->key.offset;
3243 user_thresh = div_factor_fine(cache->key.offset,
3246 if (chunk_used < user_thresh)
3249 btrfs_put_block_group(cache);
3253 static int chunk_devid_filter(struct extent_buffer *leaf,
3254 struct btrfs_chunk *chunk,
3255 struct btrfs_balance_args *bargs)
3257 struct btrfs_stripe *stripe;
3258 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3261 for (i = 0; i < num_stripes; i++) {
3262 stripe = btrfs_stripe_nr(chunk, i);
3263 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3270 /* [pstart, pend) */
3271 static int chunk_drange_filter(struct extent_buffer *leaf,
3272 struct btrfs_chunk *chunk,
3273 struct btrfs_balance_args *bargs)
3275 struct btrfs_stripe *stripe;
3276 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3282 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3285 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3286 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3287 factor = num_stripes / 2;
3288 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3289 factor = num_stripes - 1;
3290 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3291 factor = num_stripes - 2;
3293 factor = num_stripes;
3296 for (i = 0; i < num_stripes; i++) {
3297 stripe = btrfs_stripe_nr(chunk, i);
3298 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3301 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3302 stripe_length = btrfs_chunk_length(leaf, chunk);
3303 stripe_length = div_u64(stripe_length, factor);
3305 if (stripe_offset < bargs->pend &&
3306 stripe_offset + stripe_length > bargs->pstart)
3313 /* [vstart, vend) */
3314 static int chunk_vrange_filter(struct extent_buffer *leaf,
3315 struct btrfs_chunk *chunk,
3317 struct btrfs_balance_args *bargs)
3319 if (chunk_offset < bargs->vend &&
3320 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3321 /* at least part of the chunk is inside this vrange */
3327 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3328 struct btrfs_chunk *chunk,
3329 struct btrfs_balance_args *bargs)
3331 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3333 if (bargs->stripes_min <= num_stripes
3334 && num_stripes <= bargs->stripes_max)
3340 static int chunk_soft_convert_filter(u64 chunk_type,
3341 struct btrfs_balance_args *bargs)
3343 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3346 chunk_type = chunk_to_extended(chunk_type) &
3347 BTRFS_EXTENDED_PROFILE_MASK;
3349 if (bargs->target == chunk_type)
3355 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3356 struct extent_buffer *leaf,
3357 struct btrfs_chunk *chunk, u64 chunk_offset)
3359 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3360 struct btrfs_balance_args *bargs = NULL;
3361 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3364 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3365 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3369 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3370 bargs = &bctl->data;
3371 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3373 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3374 bargs = &bctl->meta;
3376 /* profiles filter */
3377 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3378 chunk_profiles_filter(chunk_type, bargs)) {
3383 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3384 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3386 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3387 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3392 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3393 chunk_devid_filter(leaf, chunk, bargs)) {
3397 /* drange filter, makes sense only with devid filter */
3398 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3399 chunk_drange_filter(leaf, chunk, bargs)) {
3404 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3405 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3409 /* stripes filter */
3410 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3411 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3415 /* soft profile changing mode */
3416 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3417 chunk_soft_convert_filter(chunk_type, bargs)) {
3422 * limited by count, must be the last filter
3424 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3425 if (bargs->limit == 0)
3429 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3431 * Same logic as the 'limit' filter; the minimum cannot be
3432 * determined here because we do not have the global information
3433 * about the count of all chunks that satisfy the filters.
3435 if (bargs->limit_max == 0)
3444 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3446 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3447 struct btrfs_root *chunk_root = fs_info->chunk_root;
3448 struct btrfs_root *dev_root = fs_info->dev_root;
3449 struct list_head *devices;
3450 struct btrfs_device *device;
3454 struct btrfs_chunk *chunk;
3455 struct btrfs_path *path = NULL;
3456 struct btrfs_key key;
3457 struct btrfs_key found_key;
3458 struct btrfs_trans_handle *trans;
3459 struct extent_buffer *leaf;
3462 int enospc_errors = 0;
3463 bool counting = true;
3464 /* The single value limit and min/max limits use the same bytes in the */
3465 u64 limit_data = bctl->data.limit;
3466 u64 limit_meta = bctl->meta.limit;
3467 u64 limit_sys = bctl->sys.limit;
3471 int chunk_reserved = 0;
3473 /* step one make some room on all the devices */
3474 devices = &fs_info->fs_devices->devices;
3475 list_for_each_entry(device, devices, dev_list) {
3476 old_size = btrfs_device_get_total_bytes(device);
3477 size_to_free = div_factor(old_size, 1);
3478 size_to_free = min_t(u64, size_to_free, SZ_1M);
3479 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3480 btrfs_device_get_total_bytes(device) -
3481 btrfs_device_get_bytes_used(device) > size_to_free ||
3482 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3485 ret = btrfs_shrink_device(device, old_size - size_to_free);
3489 /* btrfs_shrink_device never returns ret > 0 */
3494 trans = btrfs_start_transaction(dev_root, 0);
3495 if (IS_ERR(trans)) {
3496 ret = PTR_ERR(trans);
3497 btrfs_info_in_rcu(fs_info,
3498 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3499 rcu_str_deref(device->name), ret,
3500 old_size, old_size - size_to_free);
3504 ret = btrfs_grow_device(trans, device, old_size);
3506 btrfs_end_transaction(trans);
3507 /* btrfs_grow_device never returns ret > 0 */
3509 btrfs_info_in_rcu(fs_info,
3510 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3511 rcu_str_deref(device->name), ret,
3512 old_size, old_size - size_to_free);
3516 btrfs_end_transaction(trans);
3519 /* step two, relocate all the chunks */
3520 path = btrfs_alloc_path();
3526 /* zero out stat counters */
3527 spin_lock(&fs_info->balance_lock);
3528 memset(&bctl->stat, 0, sizeof(bctl->stat));
3529 spin_unlock(&fs_info->balance_lock);
3533 * The single value limit and min/max limits use the same bytes
3536 bctl->data.limit = limit_data;
3537 bctl->meta.limit = limit_meta;
3538 bctl->sys.limit = limit_sys;
3540 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3541 key.offset = (u64)-1;
3542 key.type = BTRFS_CHUNK_ITEM_KEY;
3545 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3546 atomic_read(&fs_info->balance_cancel_req)) {
3551 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3552 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3554 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3559 * this shouldn't happen, it means the last relocate
3563 BUG(); /* FIXME break ? */
3565 ret = btrfs_previous_item(chunk_root, path, 0,
3566 BTRFS_CHUNK_ITEM_KEY);
3568 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3573 leaf = path->nodes[0];
3574 slot = path->slots[0];
3575 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3577 if (found_key.objectid != key.objectid) {
3578 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3582 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3583 chunk_type = btrfs_chunk_type(leaf, chunk);
3586 spin_lock(&fs_info->balance_lock);
3587 bctl->stat.considered++;
3588 spin_unlock(&fs_info->balance_lock);
3591 ret = should_balance_chunk(fs_info, leaf, chunk,
3594 btrfs_release_path(path);
3596 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3601 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3602 spin_lock(&fs_info->balance_lock);
3603 bctl->stat.expected++;
3604 spin_unlock(&fs_info->balance_lock);
3606 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3608 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3610 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3617 * Apply limit_min filter, no need to check if the LIMITS
3618 * filter is used, limit_min is 0 by default
3620 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3621 count_data < bctl->data.limit_min)
3622 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3623 count_meta < bctl->meta.limit_min)
3624 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3625 count_sys < bctl->sys.limit_min)) {
3626 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3630 if (!chunk_reserved) {
3632 * We may be relocating the only data chunk we have,
3633 * which could potentially end up with losing data's
3634 * raid profile, so lets allocate an empty one in
3637 ret = btrfs_may_alloc_data_chunk(fs_info,
3640 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3642 } else if (ret == 1) {
3647 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3648 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3649 if (ret == -ENOSPC) {
3651 } else if (ret == -ETXTBSY) {
3653 "skipping relocation of block group %llu due to active swapfile",
3659 spin_lock(&fs_info->balance_lock);
3660 bctl->stat.completed++;
3661 spin_unlock(&fs_info->balance_lock);
3664 if (found_key.offset == 0)
3666 key.offset = found_key.offset - 1;
3670 btrfs_release_path(path);
3675 btrfs_free_path(path);
3676 if (enospc_errors) {
3677 btrfs_info(fs_info, "%d enospc errors during balance",
3687 * alloc_profile_is_valid - see if a given profile is valid and reduced
3688 * @flags: profile to validate
3689 * @extended: if true @flags is treated as an extended profile
3691 static int alloc_profile_is_valid(u64 flags, int extended)
3693 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3694 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3696 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3698 /* 1) check that all other bits are zeroed */
3702 /* 2) see if profile is reduced */
3704 return !extended; /* "0" is valid for usual profiles */
3706 /* true if exactly one bit set */
3707 return is_power_of_2(flags);
3710 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3712 /* cancel requested || normal exit path */
3713 return atomic_read(&fs_info->balance_cancel_req) ||
3714 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3715 atomic_read(&fs_info->balance_cancel_req) == 0);
3718 /* Non-zero return value signifies invalidity */
3719 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3722 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3723 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3724 (bctl_arg->target & ~allowed)));
3728 * Should be called with balance mutexe held
3730 int btrfs_balance(struct btrfs_fs_info *fs_info,
3731 struct btrfs_balance_control *bctl,
3732 struct btrfs_ioctl_balance_args *bargs)
3734 u64 meta_target, data_target;
3741 if (btrfs_fs_closing(fs_info) ||
3742 atomic_read(&fs_info->balance_pause_req) ||
3743 atomic_read(&fs_info->balance_cancel_req)) {
3748 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3749 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3753 * In case of mixed groups both data and meta should be picked,
3754 * and identical options should be given for both of them.
3756 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3757 if (mixed && (bctl->flags & allowed)) {
3758 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3759 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3760 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3762 "balance: mixed groups data and metadata options must be the same");
3768 num_devices = btrfs_num_devices(fs_info);
3770 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3771 if (num_devices > 1)
3772 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3773 if (num_devices > 2)
3774 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3775 if (num_devices > 3)
3776 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3777 BTRFS_BLOCK_GROUP_RAID6);
3778 if (validate_convert_profile(&bctl->data, allowed)) {
3779 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
3782 "balance: invalid convert data profile %s",
3783 get_raid_name(index));
3787 if (validate_convert_profile(&bctl->meta, allowed)) {
3788 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
3791 "balance: invalid convert metadata profile %s",
3792 get_raid_name(index));
3796 if (validate_convert_profile(&bctl->sys, allowed)) {
3797 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
3800 "balance: invalid convert system profile %s",
3801 get_raid_name(index));
3806 /* allow to reduce meta or sys integrity only if force set */
3807 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3808 BTRFS_BLOCK_GROUP_RAID10 |
3809 BTRFS_BLOCK_GROUP_RAID5 |
3810 BTRFS_BLOCK_GROUP_RAID6;
3812 seq = read_seqbegin(&fs_info->profiles_lock);
3814 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3815 (fs_info->avail_system_alloc_bits & allowed) &&
3816 !(bctl->sys.target & allowed)) ||
3817 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3818 (fs_info->avail_metadata_alloc_bits & allowed) &&
3819 !(bctl->meta.target & allowed))) {
3820 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3822 "balance: force reducing metadata integrity");
3825 "balance: reduces metadata integrity, use --force if you want this");
3830 } while (read_seqretry(&fs_info->profiles_lock, seq));
3832 /* if we're not converting, the target field is uninitialized */
3833 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3834 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3835 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3836 bctl->data.target : fs_info->avail_data_alloc_bits;
3837 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3838 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3839 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
3840 int data_index = btrfs_bg_flags_to_raid_index(data_target);
3843 "balance: metadata profile %s has lower redundancy than data profile %s",
3844 get_raid_name(meta_index), get_raid_name(data_index));
3847 ret = insert_balance_item(fs_info, bctl);
3848 if (ret && ret != -EEXIST)
3851 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3852 BUG_ON(ret == -EEXIST);
3853 BUG_ON(fs_info->balance_ctl);
3854 spin_lock(&fs_info->balance_lock);
3855 fs_info->balance_ctl = bctl;
3856 spin_unlock(&fs_info->balance_lock);
3858 BUG_ON(ret != -EEXIST);
3859 spin_lock(&fs_info->balance_lock);
3860 update_balance_args(bctl);
3861 spin_unlock(&fs_info->balance_lock);
3864 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3865 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3866 mutex_unlock(&fs_info->balance_mutex);
3868 ret = __btrfs_balance(fs_info);
3870 mutex_lock(&fs_info->balance_mutex);
3871 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3874 memset(bargs, 0, sizeof(*bargs));
3875 btrfs_update_ioctl_balance_args(fs_info, bargs);
3878 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3879 balance_need_close(fs_info)) {
3880 reset_balance_state(fs_info);
3881 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3884 wake_up(&fs_info->balance_wait_q);
3888 if (bctl->flags & BTRFS_BALANCE_RESUME)
3889 reset_balance_state(fs_info);
3892 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3897 static int balance_kthread(void *data)
3899 struct btrfs_fs_info *fs_info = data;
3902 mutex_lock(&fs_info->balance_mutex);
3903 if (fs_info->balance_ctl) {
3904 btrfs_info(fs_info, "balance: resuming");
3905 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
3907 mutex_unlock(&fs_info->balance_mutex);
3912 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3914 struct task_struct *tsk;
3916 mutex_lock(&fs_info->balance_mutex);
3917 if (!fs_info->balance_ctl) {
3918 mutex_unlock(&fs_info->balance_mutex);
3921 mutex_unlock(&fs_info->balance_mutex);
3923 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3924 btrfs_info(fs_info, "balance: resume skipped");
3929 * A ro->rw remount sequence should continue with the paused balance
3930 * regardless of who pauses it, system or the user as of now, so set
3933 spin_lock(&fs_info->balance_lock);
3934 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3935 spin_unlock(&fs_info->balance_lock);
3937 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3938 return PTR_ERR_OR_ZERO(tsk);
3941 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3943 struct btrfs_balance_control *bctl;
3944 struct btrfs_balance_item *item;
3945 struct btrfs_disk_balance_args disk_bargs;
3946 struct btrfs_path *path;
3947 struct extent_buffer *leaf;
3948 struct btrfs_key key;
3951 path = btrfs_alloc_path();
3955 key.objectid = BTRFS_BALANCE_OBJECTID;
3956 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3959 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3962 if (ret > 0) { /* ret = -ENOENT; */
3967 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3973 leaf = path->nodes[0];
3974 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3976 bctl->flags = btrfs_balance_flags(leaf, item);
3977 bctl->flags |= BTRFS_BALANCE_RESUME;
3979 btrfs_balance_data(leaf, item, &disk_bargs);
3980 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3981 btrfs_balance_meta(leaf, item, &disk_bargs);
3982 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3983 btrfs_balance_sys(leaf, item, &disk_bargs);
3984 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3987 * This should never happen, as the paused balance state is recovered
3988 * during mount without any chance of other exclusive ops to collide.
3990 * This gives the exclusive op status to balance and keeps in paused
3991 * state until user intervention (cancel or umount). If the ownership
3992 * cannot be assigned, show a message but do not fail. The balance
3993 * is in a paused state and must have fs_info::balance_ctl properly
3996 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3998 "balance: cannot set exclusive op status, resume manually");
4000 mutex_lock(&fs_info->balance_mutex);
4001 BUG_ON(fs_info->balance_ctl);
4002 spin_lock(&fs_info->balance_lock);
4003 fs_info->balance_ctl = bctl;
4004 spin_unlock(&fs_info->balance_lock);
4005 mutex_unlock(&fs_info->balance_mutex);
4007 btrfs_free_path(path);
4011 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4015 mutex_lock(&fs_info->balance_mutex);
4016 if (!fs_info->balance_ctl) {
4017 mutex_unlock(&fs_info->balance_mutex);
4021 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4022 atomic_inc(&fs_info->balance_pause_req);
4023 mutex_unlock(&fs_info->balance_mutex);
4025 wait_event(fs_info->balance_wait_q,
4026 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4028 mutex_lock(&fs_info->balance_mutex);
4029 /* we are good with balance_ctl ripped off from under us */
4030 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4031 atomic_dec(&fs_info->balance_pause_req);
4036 mutex_unlock(&fs_info->balance_mutex);
4040 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4042 mutex_lock(&fs_info->balance_mutex);
4043 if (!fs_info->balance_ctl) {
4044 mutex_unlock(&fs_info->balance_mutex);
4049 * A paused balance with the item stored on disk can be resumed at
4050 * mount time if the mount is read-write. Otherwise it's still paused
4051 * and we must not allow cancelling as it deletes the item.
4053 if (sb_rdonly(fs_info->sb)) {
4054 mutex_unlock(&fs_info->balance_mutex);
4058 atomic_inc(&fs_info->balance_cancel_req);
4060 * if we are running just wait and return, balance item is
4061 * deleted in btrfs_balance in this case
4063 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4064 mutex_unlock(&fs_info->balance_mutex);
4065 wait_event(fs_info->balance_wait_q,
4066 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4067 mutex_lock(&fs_info->balance_mutex);
4069 mutex_unlock(&fs_info->balance_mutex);
4071 * Lock released to allow other waiters to continue, we'll
4072 * reexamine the status again.
4074 mutex_lock(&fs_info->balance_mutex);
4076 if (fs_info->balance_ctl) {
4077 reset_balance_state(fs_info);
4078 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4079 btrfs_info(fs_info, "balance: canceled");
4083 BUG_ON(fs_info->balance_ctl ||
4084 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4085 atomic_dec(&fs_info->balance_cancel_req);
4086 mutex_unlock(&fs_info->balance_mutex);
4090 static int btrfs_uuid_scan_kthread(void *data)
4092 struct btrfs_fs_info *fs_info = data;
4093 struct btrfs_root *root = fs_info->tree_root;
4094 struct btrfs_key key;
4095 struct btrfs_path *path = NULL;
4097 struct extent_buffer *eb;
4099 struct btrfs_root_item root_item;
4101 struct btrfs_trans_handle *trans = NULL;
4103 path = btrfs_alloc_path();
4110 key.type = BTRFS_ROOT_ITEM_KEY;
4114 ret = btrfs_search_forward(root, &key, path,
4115 BTRFS_OLDEST_GENERATION);
4122 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4123 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4124 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4125 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4128 eb = path->nodes[0];
4129 slot = path->slots[0];
4130 item_size = btrfs_item_size_nr(eb, slot);
4131 if (item_size < sizeof(root_item))
4134 read_extent_buffer(eb, &root_item,
4135 btrfs_item_ptr_offset(eb, slot),
4136 (int)sizeof(root_item));
4137 if (btrfs_root_refs(&root_item) == 0)
4140 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4141 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4145 btrfs_release_path(path);
4147 * 1 - subvol uuid item
4148 * 1 - received_subvol uuid item
4150 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4151 if (IS_ERR(trans)) {
4152 ret = PTR_ERR(trans);
4160 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4161 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4162 BTRFS_UUID_KEY_SUBVOL,
4165 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4171 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4172 ret = btrfs_uuid_tree_add(trans,
4173 root_item.received_uuid,
4174 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4177 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4185 ret = btrfs_end_transaction(trans);
4191 btrfs_release_path(path);
4192 if (key.offset < (u64)-1) {
4194 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4196 key.type = BTRFS_ROOT_ITEM_KEY;
4197 } else if (key.objectid < (u64)-1) {
4199 key.type = BTRFS_ROOT_ITEM_KEY;
4208 btrfs_free_path(path);
4209 if (trans && !IS_ERR(trans))
4210 btrfs_end_transaction(trans);
4212 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4214 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4215 up(&fs_info->uuid_tree_rescan_sem);
4220 * Callback for btrfs_uuid_tree_iterate().
4222 * 0 check succeeded, the entry is not outdated.
4223 * < 0 if an error occurred.
4224 * > 0 if the check failed, which means the caller shall remove the entry.
4226 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4227 u8 *uuid, u8 type, u64 subid)
4229 struct btrfs_key key;
4231 struct btrfs_root *subvol_root;
4233 if (type != BTRFS_UUID_KEY_SUBVOL &&
4234 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4237 key.objectid = subid;
4238 key.type = BTRFS_ROOT_ITEM_KEY;
4239 key.offset = (u64)-1;
4240 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4241 if (IS_ERR(subvol_root)) {
4242 ret = PTR_ERR(subvol_root);
4249 case BTRFS_UUID_KEY_SUBVOL:
4250 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4253 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4254 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4264 static int btrfs_uuid_rescan_kthread(void *data)
4266 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4270 * 1st step is to iterate through the existing UUID tree and
4271 * to delete all entries that contain outdated data.
4272 * 2nd step is to add all missing entries to the UUID tree.
4274 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4276 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4277 up(&fs_info->uuid_tree_rescan_sem);
4280 return btrfs_uuid_scan_kthread(data);
4283 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4285 struct btrfs_trans_handle *trans;
4286 struct btrfs_root *tree_root = fs_info->tree_root;
4287 struct btrfs_root *uuid_root;
4288 struct task_struct *task;
4295 trans = btrfs_start_transaction(tree_root, 2);
4297 return PTR_ERR(trans);
4299 uuid_root = btrfs_create_tree(trans, fs_info,
4300 BTRFS_UUID_TREE_OBJECTID);
4301 if (IS_ERR(uuid_root)) {
4302 ret = PTR_ERR(uuid_root);
4303 btrfs_abort_transaction(trans, ret);
4304 btrfs_end_transaction(trans);
4308 fs_info->uuid_root = uuid_root;
4310 ret = btrfs_commit_transaction(trans);
4314 down(&fs_info->uuid_tree_rescan_sem);
4315 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4317 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4318 btrfs_warn(fs_info, "failed to start uuid_scan task");
4319 up(&fs_info->uuid_tree_rescan_sem);
4320 return PTR_ERR(task);
4326 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4328 struct task_struct *task;
4330 down(&fs_info->uuid_tree_rescan_sem);
4331 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4333 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4334 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4335 up(&fs_info->uuid_tree_rescan_sem);
4336 return PTR_ERR(task);
4343 * shrinking a device means finding all of the device extents past
4344 * the new size, and then following the back refs to the chunks.
4345 * The chunk relocation code actually frees the device extent
4347 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4349 struct btrfs_fs_info *fs_info = device->fs_info;
4350 struct btrfs_root *root = fs_info->dev_root;
4351 struct btrfs_trans_handle *trans;
4352 struct btrfs_dev_extent *dev_extent = NULL;
4353 struct btrfs_path *path;
4359 bool retried = false;
4360 bool checked_pending_chunks = false;
4361 struct extent_buffer *l;
4362 struct btrfs_key key;
4363 struct btrfs_super_block *super_copy = fs_info->super_copy;
4364 u64 old_total = btrfs_super_total_bytes(super_copy);
4365 u64 old_size = btrfs_device_get_total_bytes(device);
4368 new_size = round_down(new_size, fs_info->sectorsize);
4369 diff = round_down(old_size - new_size, fs_info->sectorsize);
4371 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4374 path = btrfs_alloc_path();
4378 path->reada = READA_BACK;
4380 mutex_lock(&fs_info->chunk_mutex);
4382 btrfs_device_set_total_bytes(device, new_size);
4383 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4384 device->fs_devices->total_rw_bytes -= diff;
4385 atomic64_sub(diff, &fs_info->free_chunk_space);
4387 mutex_unlock(&fs_info->chunk_mutex);
4390 key.objectid = device->devid;
4391 key.offset = (u64)-1;
4392 key.type = BTRFS_DEV_EXTENT_KEY;
4395 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4396 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4398 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4402 ret = btrfs_previous_item(root, path, 0, key.type);
4404 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4409 btrfs_release_path(path);
4414 slot = path->slots[0];
4415 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4417 if (key.objectid != device->devid) {
4418 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4419 btrfs_release_path(path);
4423 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4424 length = btrfs_dev_extent_length(l, dev_extent);
4426 if (key.offset + length <= new_size) {
4427 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4428 btrfs_release_path(path);
4432 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4433 btrfs_release_path(path);
4436 * We may be relocating the only data chunk we have,
4437 * which could potentially end up with losing data's
4438 * raid profile, so lets allocate an empty one in
4441 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4443 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4447 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4448 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4449 if (ret == -ENOSPC) {
4452 if (ret == -ETXTBSY) {
4454 "could not shrink block group %llu due to active swapfile",
4459 } while (key.offset-- > 0);
4461 if (failed && !retried) {
4465 } else if (failed && retried) {
4470 /* Shrinking succeeded, else we would be at "done". */
4471 trans = btrfs_start_transaction(root, 0);
4472 if (IS_ERR(trans)) {
4473 ret = PTR_ERR(trans);
4477 mutex_lock(&fs_info->chunk_mutex);
4480 * We checked in the above loop all device extents that were already in
4481 * the device tree. However before we have updated the device's
4482 * total_bytes to the new size, we might have had chunk allocations that
4483 * have not complete yet (new block groups attached to transaction
4484 * handles), and therefore their device extents were not yet in the
4485 * device tree and we missed them in the loop above. So if we have any
4486 * pending chunk using a device extent that overlaps the device range
4487 * that we can not use anymore, commit the current transaction and
4488 * repeat the search on the device tree - this way we guarantee we will
4489 * not have chunks using device extents that end beyond 'new_size'.
4491 if (!checked_pending_chunks) {
4492 u64 start = new_size;
4493 u64 len = old_size - new_size;
4495 if (contains_pending_extent(trans->transaction, device,
4497 mutex_unlock(&fs_info->chunk_mutex);
4498 checked_pending_chunks = true;
4501 ret = btrfs_commit_transaction(trans);
4508 btrfs_device_set_disk_total_bytes(device, new_size);
4509 if (list_empty(&device->resized_list))
4510 list_add_tail(&device->resized_list,
4511 &fs_info->fs_devices->resized_devices);
4513 WARN_ON(diff > old_total);
4514 btrfs_set_super_total_bytes(super_copy,
4515 round_down(old_total - diff, fs_info->sectorsize));
4516 mutex_unlock(&fs_info->chunk_mutex);
4518 /* Now btrfs_update_device() will change the on-disk size. */
4519 ret = btrfs_update_device(trans, device);
4521 btrfs_abort_transaction(trans, ret);
4522 btrfs_end_transaction(trans);
4524 ret = btrfs_commit_transaction(trans);
4527 btrfs_free_path(path);
4529 mutex_lock(&fs_info->chunk_mutex);
4530 btrfs_device_set_total_bytes(device, old_size);
4531 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4532 device->fs_devices->total_rw_bytes += diff;
4533 atomic64_add(diff, &fs_info->free_chunk_space);
4534 mutex_unlock(&fs_info->chunk_mutex);
4539 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4540 struct btrfs_key *key,
4541 struct btrfs_chunk *chunk, int item_size)
4543 struct btrfs_super_block *super_copy = fs_info->super_copy;
4544 struct btrfs_disk_key disk_key;
4548 mutex_lock(&fs_info->chunk_mutex);
4549 array_size = btrfs_super_sys_array_size(super_copy);
4550 if (array_size + item_size + sizeof(disk_key)
4551 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4552 mutex_unlock(&fs_info->chunk_mutex);
4556 ptr = super_copy->sys_chunk_array + array_size;
4557 btrfs_cpu_key_to_disk(&disk_key, key);
4558 memcpy(ptr, &disk_key, sizeof(disk_key));
4559 ptr += sizeof(disk_key);
4560 memcpy(ptr, chunk, item_size);
4561 item_size += sizeof(disk_key);
4562 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4563 mutex_unlock(&fs_info->chunk_mutex);
4569 * sort the devices in descending order by max_avail, total_avail
4571 static int btrfs_cmp_device_info(const void *a, const void *b)
4573 const struct btrfs_device_info *di_a = a;
4574 const struct btrfs_device_info *di_b = b;
4576 if (di_a->max_avail > di_b->max_avail)
4578 if (di_a->max_avail < di_b->max_avail)
4580 if (di_a->total_avail > di_b->total_avail)
4582 if (di_a->total_avail < di_b->total_avail)
4587 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4589 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4592 btrfs_set_fs_incompat(info, RAID56);
4595 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4596 - sizeof(struct btrfs_chunk)) \
4597 / sizeof(struct btrfs_stripe) + 1)
4599 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4600 - 2 * sizeof(struct btrfs_disk_key) \
4601 - 2 * sizeof(struct btrfs_chunk)) \
4602 / sizeof(struct btrfs_stripe) + 1)
4604 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4605 u64 start, u64 type)
4607 struct btrfs_fs_info *info = trans->fs_info;
4608 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4609 struct btrfs_device *device;
4610 struct map_lookup *map = NULL;
4611 struct extent_map_tree *em_tree;
4612 struct extent_map *em;
4613 struct btrfs_device_info *devices_info = NULL;
4615 int num_stripes; /* total number of stripes to allocate */
4616 int data_stripes; /* number of stripes that count for
4618 int sub_stripes; /* sub_stripes info for map */
4619 int dev_stripes; /* stripes per dev */
4620 int devs_max; /* max devs to use */
4621 int devs_min; /* min devs needed */
4622 int devs_increment; /* ndevs has to be a multiple of this */
4623 int ncopies; /* how many copies to data has */
4625 u64 max_stripe_size;
4634 BUG_ON(!alloc_profile_is_valid(type, 0));
4636 if (list_empty(&fs_devices->alloc_list)) {
4637 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4638 btrfs_debug(info, "%s: no writable device", __func__);
4642 index = btrfs_bg_flags_to_raid_index(type);
4644 sub_stripes = btrfs_raid_array[index].sub_stripes;
4645 dev_stripes = btrfs_raid_array[index].dev_stripes;
4646 devs_max = btrfs_raid_array[index].devs_max;
4647 devs_min = btrfs_raid_array[index].devs_min;
4648 devs_increment = btrfs_raid_array[index].devs_increment;
4649 ncopies = btrfs_raid_array[index].ncopies;
4651 if (type & BTRFS_BLOCK_GROUP_DATA) {
4652 max_stripe_size = SZ_1G;
4653 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4655 devs_max = BTRFS_MAX_DEVS(info);
4656 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4657 /* for larger filesystems, use larger metadata chunks */
4658 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4659 max_stripe_size = SZ_1G;
4661 max_stripe_size = SZ_256M;
4662 max_chunk_size = max_stripe_size;
4664 devs_max = BTRFS_MAX_DEVS(info);
4665 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4666 max_stripe_size = SZ_32M;
4667 max_chunk_size = 2 * max_stripe_size;
4669 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4671 btrfs_err(info, "invalid chunk type 0x%llx requested",
4676 /* we don't want a chunk larger than 10% of writeable space */
4677 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4680 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4686 * in the first pass through the devices list, we gather information
4687 * about the available holes on each device.
4690 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4694 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4696 "BTRFS: read-only device in alloc_list\n");
4700 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4701 &device->dev_state) ||
4702 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4705 if (device->total_bytes > device->bytes_used)
4706 total_avail = device->total_bytes - device->bytes_used;
4710 /* If there is no space on this device, skip it. */
4711 if (total_avail == 0)
4714 ret = find_free_dev_extent(trans, device,
4715 max_stripe_size * dev_stripes,
4716 &dev_offset, &max_avail);
4717 if (ret && ret != -ENOSPC)
4721 max_avail = max_stripe_size * dev_stripes;
4723 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4724 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4726 "%s: devid %llu has no free space, have=%llu want=%u",
4727 __func__, device->devid, max_avail,
4728 BTRFS_STRIPE_LEN * dev_stripes);
4732 if (ndevs == fs_devices->rw_devices) {
4733 WARN(1, "%s: found more than %llu devices\n",
4734 __func__, fs_devices->rw_devices);
4737 devices_info[ndevs].dev_offset = dev_offset;
4738 devices_info[ndevs].max_avail = max_avail;
4739 devices_info[ndevs].total_avail = total_avail;
4740 devices_info[ndevs].dev = device;
4745 * now sort the devices by hole size / available space
4747 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4748 btrfs_cmp_device_info, NULL);
4750 /* round down to number of usable stripes */
4751 ndevs = round_down(ndevs, devs_increment);
4753 if (ndevs < devs_min) {
4755 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4757 "%s: not enough devices with free space: have=%d minimum required=%d",
4758 __func__, ndevs, devs_min);
4763 ndevs = min(ndevs, devs_max);
4766 * The primary goal is to maximize the number of stripes, so use as
4767 * many devices as possible, even if the stripes are not maximum sized.
4769 * The DUP profile stores more than one stripe per device, the
4770 * max_avail is the total size so we have to adjust.
4772 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4773 num_stripes = ndevs * dev_stripes;
4776 * this will have to be fixed for RAID1 and RAID10 over
4779 data_stripes = num_stripes / ncopies;
4781 if (type & BTRFS_BLOCK_GROUP_RAID5)
4782 data_stripes = num_stripes - 1;
4784 if (type & BTRFS_BLOCK_GROUP_RAID6)
4785 data_stripes = num_stripes - 2;
4788 * Use the number of data stripes to figure out how big this chunk
4789 * is really going to be in terms of logical address space,
4790 * and compare that answer with the max chunk size
4792 if (stripe_size * data_stripes > max_chunk_size) {
4793 stripe_size = div_u64(max_chunk_size, data_stripes);
4795 /* bump the answer up to a 16MB boundary */
4796 stripe_size = round_up(stripe_size, SZ_16M);
4799 * But don't go higher than the limits we found while searching
4802 stripe_size = min(devices_info[ndevs - 1].max_avail,
4806 /* align to BTRFS_STRIPE_LEN */
4807 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4809 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4814 map->num_stripes = num_stripes;
4816 for (i = 0; i < ndevs; ++i) {
4817 for (j = 0; j < dev_stripes; ++j) {
4818 int s = i * dev_stripes + j;
4819 map->stripes[s].dev = devices_info[i].dev;
4820 map->stripes[s].physical = devices_info[i].dev_offset +
4824 map->stripe_len = BTRFS_STRIPE_LEN;
4825 map->io_align = BTRFS_STRIPE_LEN;
4826 map->io_width = BTRFS_STRIPE_LEN;
4828 map->sub_stripes = sub_stripes;
4830 num_bytes = stripe_size * data_stripes;
4832 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4834 em = alloc_extent_map();
4840 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4841 em->map_lookup = map;
4843 em->len = num_bytes;
4844 em->block_start = 0;
4845 em->block_len = em->len;
4846 em->orig_block_len = stripe_size;
4848 em_tree = &info->mapping_tree.map_tree;
4849 write_lock(&em_tree->lock);
4850 ret = add_extent_mapping(em_tree, em, 0);
4852 write_unlock(&em_tree->lock);
4853 free_extent_map(em);
4857 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4858 refcount_inc(&em->refs);
4859 write_unlock(&em_tree->lock);
4861 ret = btrfs_make_block_group(trans, 0, type, start, num_bytes);
4863 goto error_del_extent;
4865 for (i = 0; i < map->num_stripes; i++) {
4866 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4867 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4870 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4872 free_extent_map(em);
4873 check_raid56_incompat_flag(info, type);
4875 kfree(devices_info);
4879 write_lock(&em_tree->lock);
4880 remove_extent_mapping(em_tree, em);
4881 write_unlock(&em_tree->lock);
4883 /* One for our allocation */
4884 free_extent_map(em);
4885 /* One for the tree reference */
4886 free_extent_map(em);
4887 /* One for the pending_chunks list reference */
4888 free_extent_map(em);
4890 kfree(devices_info);
4894 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4895 u64 chunk_offset, u64 chunk_size)
4897 struct btrfs_fs_info *fs_info = trans->fs_info;
4898 struct btrfs_root *extent_root = fs_info->extent_root;
4899 struct btrfs_root *chunk_root = fs_info->chunk_root;
4900 struct btrfs_key key;
4901 struct btrfs_device *device;
4902 struct btrfs_chunk *chunk;
4903 struct btrfs_stripe *stripe;
4904 struct extent_map *em;
4905 struct map_lookup *map;
4912 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4916 map = em->map_lookup;
4917 item_size = btrfs_chunk_item_size(map->num_stripes);
4918 stripe_size = em->orig_block_len;
4920 chunk = kzalloc(item_size, GFP_NOFS);
4927 * Take the device list mutex to prevent races with the final phase of
4928 * a device replace operation that replaces the device object associated
4929 * with the map's stripes, because the device object's id can change
4930 * at any time during that final phase of the device replace operation
4931 * (dev-replace.c:btrfs_dev_replace_finishing()).
4933 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4934 for (i = 0; i < map->num_stripes; i++) {
4935 device = map->stripes[i].dev;
4936 dev_offset = map->stripes[i].physical;
4938 ret = btrfs_update_device(trans, device);
4941 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4942 dev_offset, stripe_size);
4947 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4951 stripe = &chunk->stripe;
4952 for (i = 0; i < map->num_stripes; i++) {
4953 device = map->stripes[i].dev;
4954 dev_offset = map->stripes[i].physical;
4956 btrfs_set_stack_stripe_devid(stripe, device->devid);
4957 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4958 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4961 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4963 btrfs_set_stack_chunk_length(chunk, chunk_size);
4964 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4965 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4966 btrfs_set_stack_chunk_type(chunk, map->type);
4967 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4968 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4969 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4970 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4971 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4973 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4974 key.type = BTRFS_CHUNK_ITEM_KEY;
4975 key.offset = chunk_offset;
4977 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4978 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4980 * TODO: Cleanup of inserted chunk root in case of
4983 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4988 free_extent_map(em);
4993 * Chunk allocation falls into two parts. The first part does works
4994 * that make the new allocated chunk useable, but not do any operation
4995 * that modifies the chunk tree. The second part does the works that
4996 * require modifying the chunk tree. This division is important for the
4997 * bootstrap process of adding storage to a seed btrfs.
4999 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5003 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5004 chunk_offset = find_next_chunk(trans->fs_info);
5005 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5008 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5009 struct btrfs_fs_info *fs_info)
5012 u64 sys_chunk_offset;
5016 chunk_offset = find_next_chunk(fs_info);
5017 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5018 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5022 sys_chunk_offset = find_next_chunk(fs_info);
5023 alloc_profile = btrfs_system_alloc_profile(fs_info);
5024 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5028 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5032 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5033 BTRFS_BLOCK_GROUP_RAID10 |
5034 BTRFS_BLOCK_GROUP_RAID5 |
5035 BTRFS_BLOCK_GROUP_DUP)) {
5037 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5046 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5048 struct extent_map *em;
5049 struct map_lookup *map;
5054 em = get_chunk_map(fs_info, chunk_offset, 1);
5058 map = em->map_lookup;
5059 for (i = 0; i < map->num_stripes; i++) {
5060 if (test_bit(BTRFS_DEV_STATE_MISSING,
5061 &map->stripes[i].dev->dev_state)) {
5065 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5066 &map->stripes[i].dev->dev_state)) {
5073 * If the number of missing devices is larger than max errors,
5074 * we can not write the data into that chunk successfully, so
5077 if (miss_ndevs > btrfs_chunk_max_errors(map))
5080 free_extent_map(em);
5084 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5086 extent_map_tree_init(&tree->map_tree);
5089 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5091 struct extent_map *em;
5094 write_lock(&tree->map_tree.lock);
5095 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5097 remove_extent_mapping(&tree->map_tree, em);
5098 write_unlock(&tree->map_tree.lock);
5102 free_extent_map(em);
5103 /* once for the tree */
5104 free_extent_map(em);
5108 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5110 struct extent_map *em;
5111 struct map_lookup *map;
5114 em = get_chunk_map(fs_info, logical, len);
5117 * We could return errors for these cases, but that could get
5118 * ugly and we'd probably do the same thing which is just not do
5119 * anything else and exit, so return 1 so the callers don't try
5120 * to use other copies.
5124 map = em->map_lookup;
5125 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5126 ret = map->num_stripes;
5127 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5128 ret = map->sub_stripes;
5129 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5131 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5133 * There could be two corrupted data stripes, we need
5134 * to loop retry in order to rebuild the correct data.
5136 * Fail a stripe at a time on every retry except the
5137 * stripe under reconstruction.
5139 ret = map->num_stripes;
5142 free_extent_map(em);
5144 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5145 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5146 fs_info->dev_replace.tgtdev)
5148 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5153 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5156 struct extent_map *em;
5157 struct map_lookup *map;
5158 unsigned long len = fs_info->sectorsize;
5160 em = get_chunk_map(fs_info, logical, len);
5162 if (!WARN_ON(IS_ERR(em))) {
5163 map = em->map_lookup;
5164 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5165 len = map->stripe_len * nr_data_stripes(map);
5166 free_extent_map(em);
5171 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5173 struct extent_map *em;
5174 struct map_lookup *map;
5177 em = get_chunk_map(fs_info, logical, len);
5179 if(!WARN_ON(IS_ERR(em))) {
5180 map = em->map_lookup;
5181 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5183 free_extent_map(em);
5188 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5189 struct map_lookup *map, int first,
5190 int dev_replace_is_ongoing)
5194 int preferred_mirror;
5196 struct btrfs_device *srcdev;
5199 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5201 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5202 num_stripes = map->sub_stripes;
5204 num_stripes = map->num_stripes;
5206 preferred_mirror = first + current->pid % num_stripes;
5208 if (dev_replace_is_ongoing &&
5209 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5210 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5211 srcdev = fs_info->dev_replace.srcdev;
5216 * try to avoid the drive that is the source drive for a
5217 * dev-replace procedure, only choose it if no other non-missing
5218 * mirror is available
5220 for (tolerance = 0; tolerance < 2; tolerance++) {
5221 if (map->stripes[preferred_mirror].dev->bdev &&
5222 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5223 return preferred_mirror;
5224 for (i = first; i < first + num_stripes; i++) {
5225 if (map->stripes[i].dev->bdev &&
5226 (tolerance || map->stripes[i].dev != srcdev))
5231 /* we couldn't find one that doesn't fail. Just return something
5232 * and the io error handling code will clean up eventually
5234 return preferred_mirror;
5237 static inline int parity_smaller(u64 a, u64 b)
5242 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5243 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5245 struct btrfs_bio_stripe s;
5252 for (i = 0; i < num_stripes - 1; i++) {
5253 if (parity_smaller(bbio->raid_map[i],
5254 bbio->raid_map[i+1])) {
5255 s = bbio->stripes[i];
5256 l = bbio->raid_map[i];
5257 bbio->stripes[i] = bbio->stripes[i+1];
5258 bbio->raid_map[i] = bbio->raid_map[i+1];
5259 bbio->stripes[i+1] = s;
5260 bbio->raid_map[i+1] = l;
5268 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5270 struct btrfs_bio *bbio = kzalloc(
5271 /* the size of the btrfs_bio */
5272 sizeof(struct btrfs_bio) +
5273 /* plus the variable array for the stripes */
5274 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5275 /* plus the variable array for the tgt dev */
5276 sizeof(int) * (real_stripes) +
5278 * plus the raid_map, which includes both the tgt dev
5281 sizeof(u64) * (total_stripes),
5282 GFP_NOFS|__GFP_NOFAIL);
5284 atomic_set(&bbio->error, 0);
5285 refcount_set(&bbio->refs, 1);
5290 void btrfs_get_bbio(struct btrfs_bio *bbio)
5292 WARN_ON(!refcount_read(&bbio->refs));
5293 refcount_inc(&bbio->refs);
5296 void btrfs_put_bbio(struct btrfs_bio *bbio)
5300 if (refcount_dec_and_test(&bbio->refs))
5304 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5306 * Please note that, discard won't be sent to target device of device
5309 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5310 u64 logical, u64 length,
5311 struct btrfs_bio **bbio_ret)
5313 struct extent_map *em;
5314 struct map_lookup *map;
5315 struct btrfs_bio *bbio;
5319 u64 stripe_end_offset;
5326 u32 sub_stripes = 0;
5327 u64 stripes_per_dev = 0;
5328 u32 remaining_stripes = 0;
5329 u32 last_stripe = 0;
5333 /* discard always return a bbio */
5336 em = get_chunk_map(fs_info, logical, length);
5340 map = em->map_lookup;
5341 /* we don't discard raid56 yet */
5342 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5347 offset = logical - em->start;
5348 length = min_t(u64, em->len - offset, length);
5350 stripe_len = map->stripe_len;
5352 * stripe_nr counts the total number of stripes we have to stride
5353 * to get to this block
5355 stripe_nr = div64_u64(offset, stripe_len);
5357 /* stripe_offset is the offset of this block in its stripe */
5358 stripe_offset = offset - stripe_nr * stripe_len;
5360 stripe_nr_end = round_up(offset + length, map->stripe_len);
5361 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5362 stripe_cnt = stripe_nr_end - stripe_nr;
5363 stripe_end_offset = stripe_nr_end * map->stripe_len -
5366 * after this, stripe_nr is the number of stripes on this
5367 * device we have to walk to find the data, and stripe_index is
5368 * the number of our device in the stripe array
5372 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5373 BTRFS_BLOCK_GROUP_RAID10)) {
5374 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5377 sub_stripes = map->sub_stripes;
5379 factor = map->num_stripes / sub_stripes;
5380 num_stripes = min_t(u64, map->num_stripes,
5381 sub_stripes * stripe_cnt);
5382 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5383 stripe_index *= sub_stripes;
5384 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5385 &remaining_stripes);
5386 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5387 last_stripe *= sub_stripes;
5388 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5389 BTRFS_BLOCK_GROUP_DUP)) {
5390 num_stripes = map->num_stripes;
5392 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5396 bbio = alloc_btrfs_bio(num_stripes, 0);
5402 for (i = 0; i < num_stripes; i++) {
5403 bbio->stripes[i].physical =
5404 map->stripes[stripe_index].physical +
5405 stripe_offset + stripe_nr * map->stripe_len;
5406 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5408 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5409 BTRFS_BLOCK_GROUP_RAID10)) {
5410 bbio->stripes[i].length = stripes_per_dev *
5413 if (i / sub_stripes < remaining_stripes)
5414 bbio->stripes[i].length +=
5418 * Special for the first stripe and
5421 * |-------|...|-------|
5425 if (i < sub_stripes)
5426 bbio->stripes[i].length -=
5429 if (stripe_index >= last_stripe &&
5430 stripe_index <= (last_stripe +
5432 bbio->stripes[i].length -=
5435 if (i == sub_stripes - 1)
5438 bbio->stripes[i].length = length;
5442 if (stripe_index == map->num_stripes) {
5449 bbio->map_type = map->type;
5450 bbio->num_stripes = num_stripes;
5452 free_extent_map(em);
5457 * In dev-replace case, for repair case (that's the only case where the mirror
5458 * is selected explicitly when calling btrfs_map_block), blocks left of the
5459 * left cursor can also be read from the target drive.
5461 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5463 * For READ, it also needs to be supported using the same mirror number.
5465 * If the requested block is not left of the left cursor, EIO is returned. This
5466 * can happen because btrfs_num_copies() returns one more in the dev-replace
5469 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5470 u64 logical, u64 length,
5471 u64 srcdev_devid, int *mirror_num,
5474 struct btrfs_bio *bbio = NULL;
5476 int index_srcdev = 0;
5478 u64 physical_of_found = 0;
5482 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5483 logical, &length, &bbio, 0, 0);
5485 ASSERT(bbio == NULL);
5489 num_stripes = bbio->num_stripes;
5490 if (*mirror_num > num_stripes) {
5492 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5493 * that means that the requested area is not left of the left
5496 btrfs_put_bbio(bbio);
5501 * process the rest of the function using the mirror_num of the source
5502 * drive. Therefore look it up first. At the end, patch the device
5503 * pointer to the one of the target drive.
5505 for (i = 0; i < num_stripes; i++) {
5506 if (bbio->stripes[i].dev->devid != srcdev_devid)
5510 * In case of DUP, in order to keep it simple, only add the
5511 * mirror with the lowest physical address
5514 physical_of_found <= bbio->stripes[i].physical)
5519 physical_of_found = bbio->stripes[i].physical;
5522 btrfs_put_bbio(bbio);
5528 *mirror_num = index_srcdev + 1;
5529 *physical = physical_of_found;
5533 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5534 struct btrfs_bio **bbio_ret,
5535 struct btrfs_dev_replace *dev_replace,
5536 int *num_stripes_ret, int *max_errors_ret)
5538 struct btrfs_bio *bbio = *bbio_ret;
5539 u64 srcdev_devid = dev_replace->srcdev->devid;
5540 int tgtdev_indexes = 0;
5541 int num_stripes = *num_stripes_ret;
5542 int max_errors = *max_errors_ret;
5545 if (op == BTRFS_MAP_WRITE) {
5546 int index_where_to_add;
5549 * duplicate the write operations while the dev replace
5550 * procedure is running. Since the copying of the old disk to
5551 * the new disk takes place at run time while the filesystem is
5552 * mounted writable, the regular write operations to the old
5553 * disk have to be duplicated to go to the new disk as well.
5555 * Note that device->missing is handled by the caller, and that
5556 * the write to the old disk is already set up in the stripes
5559 index_where_to_add = num_stripes;
5560 for (i = 0; i < num_stripes; i++) {
5561 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5562 /* write to new disk, too */
5563 struct btrfs_bio_stripe *new =
5564 bbio->stripes + index_where_to_add;
5565 struct btrfs_bio_stripe *old =
5568 new->physical = old->physical;
5569 new->length = old->length;
5570 new->dev = dev_replace->tgtdev;
5571 bbio->tgtdev_map[i] = index_where_to_add;
5572 index_where_to_add++;
5577 num_stripes = index_where_to_add;
5578 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5579 int index_srcdev = 0;
5581 u64 physical_of_found = 0;
5584 * During the dev-replace procedure, the target drive can also
5585 * be used to read data in case it is needed to repair a corrupt
5586 * block elsewhere. This is possible if the requested area is
5587 * left of the left cursor. In this area, the target drive is a
5588 * full copy of the source drive.
5590 for (i = 0; i < num_stripes; i++) {
5591 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5593 * In case of DUP, in order to keep it simple,
5594 * only add the mirror with the lowest physical
5598 physical_of_found <=
5599 bbio->stripes[i].physical)
5603 physical_of_found = bbio->stripes[i].physical;
5607 struct btrfs_bio_stripe *tgtdev_stripe =
5608 bbio->stripes + num_stripes;
5610 tgtdev_stripe->physical = physical_of_found;
5611 tgtdev_stripe->length =
5612 bbio->stripes[index_srcdev].length;
5613 tgtdev_stripe->dev = dev_replace->tgtdev;
5614 bbio->tgtdev_map[index_srcdev] = num_stripes;
5621 *num_stripes_ret = num_stripes;
5622 *max_errors_ret = max_errors;
5623 bbio->num_tgtdevs = tgtdev_indexes;
5627 static bool need_full_stripe(enum btrfs_map_op op)
5629 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5632 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5633 enum btrfs_map_op op,
5634 u64 logical, u64 *length,
5635 struct btrfs_bio **bbio_ret,
5636 int mirror_num, int need_raid_map)
5638 struct extent_map *em;
5639 struct map_lookup *map;
5649 int tgtdev_indexes = 0;
5650 struct btrfs_bio *bbio = NULL;
5651 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5652 int dev_replace_is_ongoing = 0;
5653 int num_alloc_stripes;
5654 int patch_the_first_stripe_for_dev_replace = 0;
5655 u64 physical_to_patch_in_first_stripe = 0;
5656 u64 raid56_full_stripe_start = (u64)-1;
5658 if (op == BTRFS_MAP_DISCARD)
5659 return __btrfs_map_block_for_discard(fs_info, logical,
5662 em = get_chunk_map(fs_info, logical, *length);
5666 map = em->map_lookup;
5667 offset = logical - em->start;
5669 stripe_len = map->stripe_len;
5672 * stripe_nr counts the total number of stripes we have to stride
5673 * to get to this block
5675 stripe_nr = div64_u64(stripe_nr, stripe_len);
5677 stripe_offset = stripe_nr * stripe_len;
5678 if (offset < stripe_offset) {
5680 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5681 stripe_offset, offset, em->start, logical,
5683 free_extent_map(em);
5687 /* stripe_offset is the offset of this block in its stripe*/
5688 stripe_offset = offset - stripe_offset;
5690 /* if we're here for raid56, we need to know the stripe aligned start */
5691 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5692 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5693 raid56_full_stripe_start = offset;
5695 /* allow a write of a full stripe, but make sure we don't
5696 * allow straddling of stripes
5698 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5700 raid56_full_stripe_start *= full_stripe_len;
5703 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5705 /* For writes to RAID[56], allow a full stripeset across all disks.
5706 For other RAID types and for RAID[56] reads, just allow a single
5707 stripe (on a single disk). */
5708 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5709 (op == BTRFS_MAP_WRITE)) {
5710 max_len = stripe_len * nr_data_stripes(map) -
5711 (offset - raid56_full_stripe_start);
5713 /* we limit the length of each bio to what fits in a stripe */
5714 max_len = stripe_len - stripe_offset;
5716 *length = min_t(u64, em->len - offset, max_len);
5718 *length = em->len - offset;
5721 /* This is for when we're called from btrfs_merge_bio_hook() and all
5722 it cares about is the length */
5726 btrfs_dev_replace_read_lock(dev_replace);
5727 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5728 if (!dev_replace_is_ongoing)
5729 btrfs_dev_replace_read_unlock(dev_replace);
5731 btrfs_dev_replace_set_lock_blocking(dev_replace);
5733 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5734 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5735 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5736 dev_replace->srcdev->devid,
5738 &physical_to_patch_in_first_stripe);
5742 patch_the_first_stripe_for_dev_replace = 1;
5743 } else if (mirror_num > map->num_stripes) {
5749 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5750 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5752 if (!need_full_stripe(op))
5754 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5755 if (need_full_stripe(op))
5756 num_stripes = map->num_stripes;
5757 else if (mirror_num)
5758 stripe_index = mirror_num - 1;
5760 stripe_index = find_live_mirror(fs_info, map, 0,
5761 dev_replace_is_ongoing);
5762 mirror_num = stripe_index + 1;
5765 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5766 if (need_full_stripe(op)) {
5767 num_stripes = map->num_stripes;
5768 } else if (mirror_num) {
5769 stripe_index = mirror_num - 1;
5774 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5775 u32 factor = map->num_stripes / map->sub_stripes;
5777 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5778 stripe_index *= map->sub_stripes;
5780 if (need_full_stripe(op))
5781 num_stripes = map->sub_stripes;
5782 else if (mirror_num)
5783 stripe_index += mirror_num - 1;
5785 int old_stripe_index = stripe_index;
5786 stripe_index = find_live_mirror(fs_info, map,
5788 dev_replace_is_ongoing);
5789 mirror_num = stripe_index - old_stripe_index + 1;
5792 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5793 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5794 /* push stripe_nr back to the start of the full stripe */
5795 stripe_nr = div64_u64(raid56_full_stripe_start,
5796 stripe_len * nr_data_stripes(map));
5798 /* RAID[56] write or recovery. Return all stripes */
5799 num_stripes = map->num_stripes;
5800 max_errors = nr_parity_stripes(map);
5802 *length = map->stripe_len;
5807 * Mirror #0 or #1 means the original data block.
5808 * Mirror #2 is RAID5 parity block.
5809 * Mirror #3 is RAID6 Q block.
5811 stripe_nr = div_u64_rem(stripe_nr,
5812 nr_data_stripes(map), &stripe_index);
5814 stripe_index = nr_data_stripes(map) +
5817 /* We distribute the parity blocks across stripes */
5818 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5820 if (!need_full_stripe(op) && mirror_num <= 1)
5825 * after this, stripe_nr is the number of stripes on this
5826 * device we have to walk to find the data, and stripe_index is
5827 * the number of our device in the stripe array
5829 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5831 mirror_num = stripe_index + 1;
5833 if (stripe_index >= map->num_stripes) {
5835 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5836 stripe_index, map->num_stripes);
5841 num_alloc_stripes = num_stripes;
5842 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5843 if (op == BTRFS_MAP_WRITE)
5844 num_alloc_stripes <<= 1;
5845 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5846 num_alloc_stripes++;
5847 tgtdev_indexes = num_stripes;
5850 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5855 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5856 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5858 /* build raid_map */
5859 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5860 (need_full_stripe(op) || mirror_num > 1)) {
5864 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5865 sizeof(struct btrfs_bio_stripe) *
5867 sizeof(int) * tgtdev_indexes);
5869 /* Work out the disk rotation on this stripe-set */
5870 div_u64_rem(stripe_nr, num_stripes, &rot);
5872 /* Fill in the logical address of each stripe */
5873 tmp = stripe_nr * nr_data_stripes(map);
5874 for (i = 0; i < nr_data_stripes(map); i++)
5875 bbio->raid_map[(i+rot) % num_stripes] =
5876 em->start + (tmp + i) * map->stripe_len;
5878 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5879 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5880 bbio->raid_map[(i+rot+1) % num_stripes] =
5885 for (i = 0; i < num_stripes; i++) {
5886 bbio->stripes[i].physical =
5887 map->stripes[stripe_index].physical +
5889 stripe_nr * map->stripe_len;
5890 bbio->stripes[i].dev =
5891 map->stripes[stripe_index].dev;
5895 if (need_full_stripe(op))
5896 max_errors = btrfs_chunk_max_errors(map);
5899 sort_parity_stripes(bbio, num_stripes);
5901 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5902 need_full_stripe(op)) {
5903 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5908 bbio->map_type = map->type;
5909 bbio->num_stripes = num_stripes;
5910 bbio->max_errors = max_errors;
5911 bbio->mirror_num = mirror_num;
5914 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5915 * mirror_num == num_stripes + 1 && dev_replace target drive is
5916 * available as a mirror
5918 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5919 WARN_ON(num_stripes > 1);
5920 bbio->stripes[0].dev = dev_replace->tgtdev;
5921 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5922 bbio->mirror_num = map->num_stripes + 1;
5925 if (dev_replace_is_ongoing) {
5926 ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
5927 btrfs_dev_replace_read_lock(dev_replace);
5928 /* Barrier implied by atomic_dec_and_test */
5929 if (atomic_dec_and_test(&dev_replace->blocking_readers))
5930 cond_wake_up_nomb(&dev_replace->read_lock_wq);
5931 btrfs_dev_replace_read_unlock(dev_replace);
5933 free_extent_map(em);
5937 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5938 u64 logical, u64 *length,
5939 struct btrfs_bio **bbio_ret, int mirror_num)
5941 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5945 /* For Scrub/replace */
5946 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5947 u64 logical, u64 *length,
5948 struct btrfs_bio **bbio_ret)
5950 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5953 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
5954 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
5956 struct extent_map *em;
5957 struct map_lookup *map;
5965 em = get_chunk_map(fs_info, chunk_start, 1);
5969 map = em->map_lookup;
5971 rmap_len = map->stripe_len;
5973 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5974 length = div_u64(length, map->num_stripes / map->sub_stripes);
5975 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5976 length = div_u64(length, map->num_stripes);
5977 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5978 length = div_u64(length, nr_data_stripes(map));
5979 rmap_len = map->stripe_len * nr_data_stripes(map);
5982 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5983 BUG_ON(!buf); /* -ENOMEM */
5985 for (i = 0; i < map->num_stripes; i++) {
5986 if (map->stripes[i].physical > physical ||
5987 map->stripes[i].physical + length <= physical)
5990 stripe_nr = physical - map->stripes[i].physical;
5991 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5993 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5994 stripe_nr = stripe_nr * map->num_stripes + i;
5995 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5996 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5997 stripe_nr = stripe_nr * map->num_stripes + i;
5998 } /* else if RAID[56], multiply by nr_data_stripes().
5999 * Alternatively, just use rmap_len below instead of
6000 * map->stripe_len */
6002 bytenr = chunk_start + stripe_nr * rmap_len;
6003 WARN_ON(nr >= map->num_stripes);
6004 for (j = 0; j < nr; j++) {
6005 if (buf[j] == bytenr)
6009 WARN_ON(nr >= map->num_stripes);
6016 *stripe_len = rmap_len;
6018 free_extent_map(em);
6022 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6024 bio->bi_private = bbio->private;
6025 bio->bi_end_io = bbio->end_io;
6028 btrfs_put_bbio(bbio);
6031 static void btrfs_end_bio(struct bio *bio)
6033 struct btrfs_bio *bbio = bio->bi_private;
6034 int is_orig_bio = 0;
6036 if (bio->bi_status) {
6037 atomic_inc(&bbio->error);
6038 if (bio->bi_status == BLK_STS_IOERR ||
6039 bio->bi_status == BLK_STS_TARGET) {
6040 unsigned int stripe_index =
6041 btrfs_io_bio(bio)->stripe_index;
6042 struct btrfs_device *dev;
6044 BUG_ON(stripe_index >= bbio->num_stripes);
6045 dev = bbio->stripes[stripe_index].dev;
6047 if (bio_op(bio) == REQ_OP_WRITE)
6048 btrfs_dev_stat_inc_and_print(dev,
6049 BTRFS_DEV_STAT_WRITE_ERRS);
6051 btrfs_dev_stat_inc_and_print(dev,
6052 BTRFS_DEV_STAT_READ_ERRS);
6053 if (bio->bi_opf & REQ_PREFLUSH)
6054 btrfs_dev_stat_inc_and_print(dev,
6055 BTRFS_DEV_STAT_FLUSH_ERRS);
6060 if (bio == bbio->orig_bio)
6063 btrfs_bio_counter_dec(bbio->fs_info);
6065 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6068 bio = bbio->orig_bio;
6071 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6072 /* only send an error to the higher layers if it is
6073 * beyond the tolerance of the btrfs bio
6075 if (atomic_read(&bbio->error) > bbio->max_errors) {
6076 bio->bi_status = BLK_STS_IOERR;
6079 * this bio is actually up to date, we didn't
6080 * go over the max number of errors
6082 bio->bi_status = BLK_STS_OK;
6085 btrfs_end_bbio(bbio, bio);
6086 } else if (!is_orig_bio) {
6092 * see run_scheduled_bios for a description of why bios are collected for
6095 * This will add one bio to the pending list for a device and make sure
6096 * the work struct is scheduled.
6098 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6101 struct btrfs_fs_info *fs_info = device->fs_info;
6102 int should_queue = 1;
6103 struct btrfs_pending_bios *pending_bios;
6105 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6111 /* don't bother with additional async steps for reads, right now */
6112 if (bio_op(bio) == REQ_OP_READ) {
6113 btrfsic_submit_bio(bio);
6117 WARN_ON(bio->bi_next);
6118 bio->bi_next = NULL;
6120 spin_lock(&device->io_lock);
6121 if (op_is_sync(bio->bi_opf))
6122 pending_bios = &device->pending_sync_bios;
6124 pending_bios = &device->pending_bios;
6126 if (pending_bios->tail)
6127 pending_bios->tail->bi_next = bio;
6129 pending_bios->tail = bio;
6130 if (!pending_bios->head)
6131 pending_bios->head = bio;
6132 if (device->running_pending)
6135 spin_unlock(&device->io_lock);
6138 btrfs_queue_work(fs_info->submit_workers, &device->work);
6141 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6142 u64 physical, int dev_nr, int async)
6144 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6145 struct btrfs_fs_info *fs_info = bbio->fs_info;
6147 bio->bi_private = bbio;
6148 btrfs_io_bio(bio)->stripe_index = dev_nr;
6149 bio->bi_end_io = btrfs_end_bio;
6150 bio->bi_iter.bi_sector = physical >> 9;
6151 btrfs_debug_in_rcu(fs_info,
6152 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6153 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6154 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6155 bio->bi_iter.bi_size);
6156 bio_set_dev(bio, dev->bdev);
6158 btrfs_bio_counter_inc_noblocked(fs_info);
6161 btrfs_schedule_bio(dev, bio);
6163 btrfsic_submit_bio(bio);
6166 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6168 atomic_inc(&bbio->error);
6169 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6170 /* Should be the original bio. */
6171 WARN_ON(bio != bbio->orig_bio);
6173 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6174 bio->bi_iter.bi_sector = logical >> 9;
6175 if (atomic_read(&bbio->error) > bbio->max_errors)
6176 bio->bi_status = BLK_STS_IOERR;
6178 bio->bi_status = BLK_STS_OK;
6179 btrfs_end_bbio(bbio, bio);
6183 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6184 int mirror_num, int async_submit)
6186 struct btrfs_device *dev;
6187 struct bio *first_bio = bio;
6188 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6194 struct btrfs_bio *bbio = NULL;
6196 length = bio->bi_iter.bi_size;
6197 map_length = length;
6199 btrfs_bio_counter_inc_blocked(fs_info);
6200 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6201 &map_length, &bbio, mirror_num, 1);
6203 btrfs_bio_counter_dec(fs_info);
6204 return errno_to_blk_status(ret);
6207 total_devs = bbio->num_stripes;
6208 bbio->orig_bio = first_bio;
6209 bbio->private = first_bio->bi_private;
6210 bbio->end_io = first_bio->bi_end_io;
6211 bbio->fs_info = fs_info;
6212 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6214 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6215 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6216 /* In this case, map_length has been set to the length of
6217 a single stripe; not the whole write */
6218 if (bio_op(bio) == REQ_OP_WRITE) {
6219 ret = raid56_parity_write(fs_info, bio, bbio,
6222 ret = raid56_parity_recover(fs_info, bio, bbio,
6223 map_length, mirror_num, 1);
6226 btrfs_bio_counter_dec(fs_info);
6227 return errno_to_blk_status(ret);
6230 if (map_length < length) {
6232 "mapping failed logical %llu bio len %llu len %llu",
6233 logical, length, map_length);
6237 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6238 dev = bbio->stripes[dev_nr].dev;
6239 if (!dev || !dev->bdev ||
6240 (bio_op(first_bio) == REQ_OP_WRITE &&
6241 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6242 bbio_error(bbio, first_bio, logical);
6246 if (dev_nr < total_devs - 1)
6247 bio = btrfs_bio_clone(first_bio);
6251 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6252 dev_nr, async_submit);
6254 btrfs_bio_counter_dec(fs_info);
6258 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6261 struct btrfs_device *device;
6262 struct btrfs_fs_devices *cur_devices;
6264 cur_devices = fs_info->fs_devices;
6265 while (cur_devices) {
6267 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6268 device = find_device(cur_devices, devid, uuid);
6272 cur_devices = cur_devices->seed;
6277 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6278 u64 devid, u8 *dev_uuid)
6280 struct btrfs_device *device;
6282 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6286 list_add(&device->dev_list, &fs_devices->devices);
6287 device->fs_devices = fs_devices;
6288 fs_devices->num_devices++;
6290 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6291 fs_devices->missing_devices++;
6297 * btrfs_alloc_device - allocate struct btrfs_device
6298 * @fs_info: used only for generating a new devid, can be NULL if
6299 * devid is provided (i.e. @devid != NULL).
6300 * @devid: a pointer to devid for this device. If NULL a new devid
6302 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6305 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6306 * on error. Returned struct is not linked onto any lists and must be
6307 * destroyed with btrfs_free_device.
6309 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6313 struct btrfs_device *dev;
6316 if (WARN_ON(!devid && !fs_info))
6317 return ERR_PTR(-EINVAL);
6319 dev = __alloc_device();
6328 ret = find_next_devid(fs_info, &tmp);
6330 btrfs_free_device(dev);
6331 return ERR_PTR(ret);
6337 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6339 generate_random_uuid(dev->uuid);
6341 btrfs_init_work(&dev->work, btrfs_submit_helper,
6342 pending_bios_fn, NULL, NULL);
6347 /* Return -EIO if any error, otherwise return 0. */
6348 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6349 struct extent_buffer *leaf,
6350 struct btrfs_chunk *chunk, u64 logical)
6360 length = btrfs_chunk_length(leaf, chunk);
6361 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6362 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6363 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6364 type = btrfs_chunk_type(leaf, chunk);
6367 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6371 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6372 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6375 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6376 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6377 btrfs_chunk_sector_size(leaf, chunk));
6380 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6381 btrfs_err(fs_info, "invalid chunk length %llu", length);
6384 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6385 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6389 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6391 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6392 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6393 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6394 btrfs_chunk_type(leaf, chunk));
6398 if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
6399 btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type);
6403 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
6404 (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
6406 "system chunk with data or metadata type: 0x%llx", type);
6410 features = btrfs_super_incompat_flags(fs_info->super_copy);
6411 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
6415 if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
6416 (type & BTRFS_BLOCK_GROUP_DATA)) {
6418 "mixed chunk type in non-mixed mode: 0x%llx", type);
6423 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6424 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6425 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6426 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6427 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6428 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6429 num_stripes != 1)) {
6431 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6432 num_stripes, sub_stripes,
6433 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6440 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6441 u64 devid, u8 *uuid, bool error)
6444 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6447 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6451 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6452 struct extent_buffer *leaf,
6453 struct btrfs_chunk *chunk)
6455 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6456 struct map_lookup *map;
6457 struct extent_map *em;
6461 u8 uuid[BTRFS_UUID_SIZE];
6466 logical = key->offset;
6467 length = btrfs_chunk_length(leaf, chunk);
6468 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6470 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6474 read_lock(&map_tree->map_tree.lock);
6475 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6476 read_unlock(&map_tree->map_tree.lock);
6478 /* already mapped? */
6479 if (em && em->start <= logical && em->start + em->len > logical) {
6480 free_extent_map(em);
6483 free_extent_map(em);
6486 em = alloc_extent_map();
6489 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6491 free_extent_map(em);
6495 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6496 em->map_lookup = map;
6497 em->start = logical;
6500 em->block_start = 0;
6501 em->block_len = em->len;
6503 map->num_stripes = num_stripes;
6504 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6505 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6506 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6507 map->type = btrfs_chunk_type(leaf, chunk);
6508 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6509 map->verified_stripes = 0;
6510 for (i = 0; i < num_stripes; i++) {
6511 map->stripes[i].physical =
6512 btrfs_stripe_offset_nr(leaf, chunk, i);
6513 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6514 read_extent_buffer(leaf, uuid, (unsigned long)
6515 btrfs_stripe_dev_uuid_nr(chunk, i),
6517 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6519 if (!map->stripes[i].dev &&
6520 !btrfs_test_opt(fs_info, DEGRADED)) {
6521 free_extent_map(em);
6522 btrfs_report_missing_device(fs_info, devid, uuid, true);
6525 if (!map->stripes[i].dev) {
6526 map->stripes[i].dev =
6527 add_missing_dev(fs_info->fs_devices, devid,
6529 if (IS_ERR(map->stripes[i].dev)) {
6530 free_extent_map(em);
6532 "failed to init missing dev %llu: %ld",
6533 devid, PTR_ERR(map->stripes[i].dev));
6534 return PTR_ERR(map->stripes[i].dev);
6536 btrfs_report_missing_device(fs_info, devid, uuid, false);
6538 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6539 &(map->stripes[i].dev->dev_state));
6543 write_lock(&map_tree->map_tree.lock);
6544 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6545 write_unlock(&map_tree->map_tree.lock);
6548 "failed to add chunk map, start=%llu len=%llu: %d",
6549 em->start, em->len, ret);
6551 free_extent_map(em);
6556 static void fill_device_from_item(struct extent_buffer *leaf,
6557 struct btrfs_dev_item *dev_item,
6558 struct btrfs_device *device)
6562 device->devid = btrfs_device_id(leaf, dev_item);
6563 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6564 device->total_bytes = device->disk_total_bytes;
6565 device->commit_total_bytes = device->disk_total_bytes;
6566 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6567 device->commit_bytes_used = device->bytes_used;
6568 device->type = btrfs_device_type(leaf, dev_item);
6569 device->io_align = btrfs_device_io_align(leaf, dev_item);
6570 device->io_width = btrfs_device_io_width(leaf, dev_item);
6571 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6572 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6573 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6575 ptr = btrfs_device_uuid(dev_item);
6576 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6579 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6582 struct btrfs_fs_devices *fs_devices;
6585 lockdep_assert_held(&uuid_mutex);
6588 fs_devices = fs_info->fs_devices->seed;
6589 while (fs_devices) {
6590 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6593 fs_devices = fs_devices->seed;
6596 fs_devices = find_fsid(fsid);
6598 if (!btrfs_test_opt(fs_info, DEGRADED))
6599 return ERR_PTR(-ENOENT);
6601 fs_devices = alloc_fs_devices(fsid);
6602 if (IS_ERR(fs_devices))
6605 fs_devices->seeding = 1;
6606 fs_devices->opened = 1;
6610 fs_devices = clone_fs_devices(fs_devices);
6611 if (IS_ERR(fs_devices))
6614 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6616 free_fs_devices(fs_devices);
6617 fs_devices = ERR_PTR(ret);
6621 if (!fs_devices->seeding) {
6622 close_fs_devices(fs_devices);
6623 free_fs_devices(fs_devices);
6624 fs_devices = ERR_PTR(-EINVAL);
6628 fs_devices->seed = fs_info->fs_devices->seed;
6629 fs_info->fs_devices->seed = fs_devices;
6634 static int read_one_dev(struct btrfs_fs_info *fs_info,
6635 struct extent_buffer *leaf,
6636 struct btrfs_dev_item *dev_item)
6638 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6639 struct btrfs_device *device;
6642 u8 fs_uuid[BTRFS_FSID_SIZE];
6643 u8 dev_uuid[BTRFS_UUID_SIZE];
6645 devid = btrfs_device_id(leaf, dev_item);
6646 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6648 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6651 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6652 fs_devices = open_seed_devices(fs_info, fs_uuid);
6653 if (IS_ERR(fs_devices))
6654 return PTR_ERR(fs_devices);
6657 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6659 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6660 btrfs_report_missing_device(fs_info, devid,
6665 device = add_missing_dev(fs_devices, devid, dev_uuid);
6666 if (IS_ERR(device)) {
6668 "failed to add missing dev %llu: %ld",
6669 devid, PTR_ERR(device));
6670 return PTR_ERR(device);
6672 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6674 if (!device->bdev) {
6675 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6676 btrfs_report_missing_device(fs_info,
6677 devid, dev_uuid, true);
6680 btrfs_report_missing_device(fs_info, devid,
6684 if (!device->bdev &&
6685 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6687 * this happens when a device that was properly setup
6688 * in the device info lists suddenly goes bad.
6689 * device->bdev is NULL, and so we have to set
6690 * device->missing to one here
6692 device->fs_devices->missing_devices++;
6693 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6696 /* Move the device to its own fs_devices */
6697 if (device->fs_devices != fs_devices) {
6698 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6699 &device->dev_state));
6701 list_move(&device->dev_list, &fs_devices->devices);
6702 device->fs_devices->num_devices--;
6703 fs_devices->num_devices++;
6705 device->fs_devices->missing_devices--;
6706 fs_devices->missing_devices++;
6708 device->fs_devices = fs_devices;
6712 if (device->fs_devices != fs_info->fs_devices) {
6713 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6714 if (device->generation !=
6715 btrfs_device_generation(leaf, dev_item))
6719 fill_device_from_item(leaf, dev_item, device);
6720 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6721 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6722 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6723 device->fs_devices->total_rw_bytes += device->total_bytes;
6724 atomic64_add(device->total_bytes - device->bytes_used,
6725 &fs_info->free_chunk_space);
6731 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6733 struct btrfs_root *root = fs_info->tree_root;
6734 struct btrfs_super_block *super_copy = fs_info->super_copy;
6735 struct extent_buffer *sb;
6736 struct btrfs_disk_key *disk_key;
6737 struct btrfs_chunk *chunk;
6739 unsigned long sb_array_offset;
6746 struct btrfs_key key;
6748 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6750 * This will create extent buffer of nodesize, superblock size is
6751 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6752 * overallocate but we can keep it as-is, only the first page is used.
6754 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6757 set_extent_buffer_uptodate(sb);
6758 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6760 * The sb extent buffer is artificial and just used to read the system array.
6761 * set_extent_buffer_uptodate() call does not properly mark all it's
6762 * pages up-to-date when the page is larger: extent does not cover the
6763 * whole page and consequently check_page_uptodate does not find all
6764 * the page's extents up-to-date (the hole beyond sb),
6765 * write_extent_buffer then triggers a WARN_ON.
6767 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6768 * but sb spans only this function. Add an explicit SetPageUptodate call
6769 * to silence the warning eg. on PowerPC 64.
6771 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6772 SetPageUptodate(sb->pages[0]);
6774 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6775 array_size = btrfs_super_sys_array_size(super_copy);
6777 array_ptr = super_copy->sys_chunk_array;
6778 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6781 while (cur_offset < array_size) {
6782 disk_key = (struct btrfs_disk_key *)array_ptr;
6783 len = sizeof(*disk_key);
6784 if (cur_offset + len > array_size)
6785 goto out_short_read;
6787 btrfs_disk_key_to_cpu(&key, disk_key);
6790 sb_array_offset += len;
6793 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6794 chunk = (struct btrfs_chunk *)sb_array_offset;
6796 * At least one btrfs_chunk with one stripe must be
6797 * present, exact stripe count check comes afterwards
6799 len = btrfs_chunk_item_size(1);
6800 if (cur_offset + len > array_size)
6801 goto out_short_read;
6803 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6806 "invalid number of stripes %u in sys_array at offset %u",
6807 num_stripes, cur_offset);
6812 type = btrfs_chunk_type(sb, chunk);
6813 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6815 "invalid chunk type %llu in sys_array at offset %u",
6821 len = btrfs_chunk_item_size(num_stripes);
6822 if (cur_offset + len > array_size)
6823 goto out_short_read;
6825 ret = read_one_chunk(fs_info, &key, sb, chunk);
6830 "unexpected item type %u in sys_array at offset %u",
6831 (u32)key.type, cur_offset);
6836 sb_array_offset += len;
6839 clear_extent_buffer_uptodate(sb);
6840 free_extent_buffer_stale(sb);
6844 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6846 clear_extent_buffer_uptodate(sb);
6847 free_extent_buffer_stale(sb);
6852 * Check if all chunks in the fs are OK for read-write degraded mount
6854 * If the @failing_dev is specified, it's accounted as missing.
6856 * Return true if all chunks meet the minimal RW mount requirements.
6857 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6859 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6860 struct btrfs_device *failing_dev)
6862 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6863 struct extent_map *em;
6867 read_lock(&map_tree->map_tree.lock);
6868 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6869 read_unlock(&map_tree->map_tree.lock);
6870 /* No chunk at all? Return false anyway */
6876 struct map_lookup *map;
6881 map = em->map_lookup;
6883 btrfs_get_num_tolerated_disk_barrier_failures(
6885 for (i = 0; i < map->num_stripes; i++) {
6886 struct btrfs_device *dev = map->stripes[i].dev;
6888 if (!dev || !dev->bdev ||
6889 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6890 dev->last_flush_error)
6892 else if (failing_dev && failing_dev == dev)
6895 if (missing > max_tolerated) {
6898 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6899 em->start, missing, max_tolerated);
6900 free_extent_map(em);
6904 next_start = extent_map_end(em);
6905 free_extent_map(em);
6907 read_lock(&map_tree->map_tree.lock);
6908 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6909 (u64)(-1) - next_start);
6910 read_unlock(&map_tree->map_tree.lock);
6916 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6918 struct btrfs_root *root = fs_info->chunk_root;
6919 struct btrfs_path *path;
6920 struct extent_buffer *leaf;
6921 struct btrfs_key key;
6922 struct btrfs_key found_key;
6927 path = btrfs_alloc_path();
6932 * uuid_mutex is needed only if we are mounting a sprout FS
6933 * otherwise we don't need it.
6935 mutex_lock(&uuid_mutex);
6936 mutex_lock(&fs_info->chunk_mutex);
6939 * Read all device items, and then all the chunk items. All
6940 * device items are found before any chunk item (their object id
6941 * is smaller than the lowest possible object id for a chunk
6942 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6944 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6947 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6951 leaf = path->nodes[0];
6952 slot = path->slots[0];
6953 if (slot >= btrfs_header_nritems(leaf)) {
6954 ret = btrfs_next_leaf(root, path);
6961 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6962 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6963 struct btrfs_dev_item *dev_item;
6964 dev_item = btrfs_item_ptr(leaf, slot,
6965 struct btrfs_dev_item);
6966 ret = read_one_dev(fs_info, leaf, dev_item);
6970 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6971 struct btrfs_chunk *chunk;
6972 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6973 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6981 * After loading chunk tree, we've got all device information,
6982 * do another round of validation checks.
6984 if (total_dev != fs_info->fs_devices->total_devices) {
6986 "super_num_devices %llu mismatch with num_devices %llu found here",
6987 btrfs_super_num_devices(fs_info->super_copy),
6992 if (btrfs_super_total_bytes(fs_info->super_copy) <
6993 fs_info->fs_devices->total_rw_bytes) {
6995 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6996 btrfs_super_total_bytes(fs_info->super_copy),
6997 fs_info->fs_devices->total_rw_bytes);
7003 mutex_unlock(&fs_info->chunk_mutex);
7004 mutex_unlock(&uuid_mutex);
7006 btrfs_free_path(path);
7010 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7012 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7013 struct btrfs_device *device;
7015 while (fs_devices) {
7016 mutex_lock(&fs_devices->device_list_mutex);
7017 list_for_each_entry(device, &fs_devices->devices, dev_list)
7018 device->fs_info = fs_info;
7019 mutex_unlock(&fs_devices->device_list_mutex);
7021 fs_devices = fs_devices->seed;
7025 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7029 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7030 btrfs_dev_stat_reset(dev, i);
7033 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7035 struct btrfs_key key;
7036 struct btrfs_key found_key;
7037 struct btrfs_root *dev_root = fs_info->dev_root;
7038 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7039 struct extent_buffer *eb;
7042 struct btrfs_device *device;
7043 struct btrfs_path *path = NULL;
7046 path = btrfs_alloc_path();
7052 mutex_lock(&fs_devices->device_list_mutex);
7053 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7055 struct btrfs_dev_stats_item *ptr;
7057 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7058 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7059 key.offset = device->devid;
7060 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7062 __btrfs_reset_dev_stats(device);
7063 device->dev_stats_valid = 1;
7064 btrfs_release_path(path);
7067 slot = path->slots[0];
7068 eb = path->nodes[0];
7069 btrfs_item_key_to_cpu(eb, &found_key, slot);
7070 item_size = btrfs_item_size_nr(eb, slot);
7072 ptr = btrfs_item_ptr(eb, slot,
7073 struct btrfs_dev_stats_item);
7075 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7076 if (item_size >= (1 + i) * sizeof(__le64))
7077 btrfs_dev_stat_set(device, i,
7078 btrfs_dev_stats_value(eb, ptr, i));
7080 btrfs_dev_stat_reset(device, i);
7083 device->dev_stats_valid = 1;
7084 btrfs_dev_stat_print_on_load(device);
7085 btrfs_release_path(path);
7087 mutex_unlock(&fs_devices->device_list_mutex);
7090 btrfs_free_path(path);
7091 return ret < 0 ? ret : 0;
7094 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7095 struct btrfs_device *device)
7097 struct btrfs_fs_info *fs_info = trans->fs_info;
7098 struct btrfs_root *dev_root = fs_info->dev_root;
7099 struct btrfs_path *path;
7100 struct btrfs_key key;
7101 struct extent_buffer *eb;
7102 struct btrfs_dev_stats_item *ptr;
7106 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7107 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7108 key.offset = device->devid;
7110 path = btrfs_alloc_path();
7113 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7115 btrfs_warn_in_rcu(fs_info,
7116 "error %d while searching for dev_stats item for device %s",
7117 ret, rcu_str_deref(device->name));
7122 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7123 /* need to delete old one and insert a new one */
7124 ret = btrfs_del_item(trans, dev_root, path);
7126 btrfs_warn_in_rcu(fs_info,
7127 "delete too small dev_stats item for device %s failed %d",
7128 rcu_str_deref(device->name), ret);
7135 /* need to insert a new item */
7136 btrfs_release_path(path);
7137 ret = btrfs_insert_empty_item(trans, dev_root, path,
7138 &key, sizeof(*ptr));
7140 btrfs_warn_in_rcu(fs_info,
7141 "insert dev_stats item for device %s failed %d",
7142 rcu_str_deref(device->name), ret);
7147 eb = path->nodes[0];
7148 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7149 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7150 btrfs_set_dev_stats_value(eb, ptr, i,
7151 btrfs_dev_stat_read(device, i));
7152 btrfs_mark_buffer_dirty(eb);
7155 btrfs_free_path(path);
7160 * called from commit_transaction. Writes all changed device stats to disk.
7162 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7163 struct btrfs_fs_info *fs_info)
7165 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7166 struct btrfs_device *device;
7170 mutex_lock(&fs_devices->device_list_mutex);
7171 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7172 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7173 if (!device->dev_stats_valid || stats_cnt == 0)
7178 * There is a LOAD-LOAD control dependency between the value of
7179 * dev_stats_ccnt and updating the on-disk values which requires
7180 * reading the in-memory counters. Such control dependencies
7181 * require explicit read memory barriers.
7183 * This memory barriers pairs with smp_mb__before_atomic in
7184 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7185 * barrier implied by atomic_xchg in
7186 * btrfs_dev_stats_read_and_reset
7190 ret = update_dev_stat_item(trans, device);
7192 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7194 mutex_unlock(&fs_devices->device_list_mutex);
7199 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7201 btrfs_dev_stat_inc(dev, index);
7202 btrfs_dev_stat_print_on_error(dev);
7205 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7207 if (!dev->dev_stats_valid)
7209 btrfs_err_rl_in_rcu(dev->fs_info,
7210 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7211 rcu_str_deref(dev->name),
7212 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7213 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7214 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7215 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7216 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7219 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7223 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7224 if (btrfs_dev_stat_read(dev, i) != 0)
7226 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7227 return; /* all values == 0, suppress message */
7229 btrfs_info_in_rcu(dev->fs_info,
7230 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7231 rcu_str_deref(dev->name),
7232 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7233 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7234 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7235 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7236 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7239 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7240 struct btrfs_ioctl_get_dev_stats *stats)
7242 struct btrfs_device *dev;
7243 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7246 mutex_lock(&fs_devices->device_list_mutex);
7247 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7248 mutex_unlock(&fs_devices->device_list_mutex);
7251 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7253 } else if (!dev->dev_stats_valid) {
7254 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7256 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7257 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7258 if (stats->nr_items > i)
7260 btrfs_dev_stat_read_and_reset(dev, i);
7262 btrfs_dev_stat_reset(dev, i);
7265 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7266 if (stats->nr_items > i)
7267 stats->values[i] = btrfs_dev_stat_read(dev, i);
7269 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7270 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7274 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7276 struct buffer_head *bh;
7277 struct btrfs_super_block *disk_super;
7283 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7286 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7289 disk_super = (struct btrfs_super_block *)bh->b_data;
7291 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7292 set_buffer_dirty(bh);
7293 sync_dirty_buffer(bh);
7297 /* Notify udev that device has changed */
7298 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7300 /* Update ctime/mtime for device path for libblkid */
7301 update_dev_time(device_path);
7305 * Update the size of all devices, which is used for writing out the
7308 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7310 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7311 struct btrfs_device *curr, *next;
7313 if (list_empty(&fs_devices->resized_devices))
7316 mutex_lock(&fs_devices->device_list_mutex);
7317 mutex_lock(&fs_info->chunk_mutex);
7318 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7320 list_del_init(&curr->resized_list);
7321 curr->commit_total_bytes = curr->disk_total_bytes;
7323 mutex_unlock(&fs_info->chunk_mutex);
7324 mutex_unlock(&fs_devices->device_list_mutex);
7327 /* Must be invoked during the transaction commit */
7328 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7330 struct btrfs_fs_info *fs_info = trans->fs_info;
7331 struct extent_map *em;
7332 struct map_lookup *map;
7333 struct btrfs_device *dev;
7336 if (list_empty(&trans->pending_chunks))
7339 /* In order to kick the device replace finish process */
7340 mutex_lock(&fs_info->chunk_mutex);
7341 list_for_each_entry(em, &trans->pending_chunks, list) {
7342 map = em->map_lookup;
7344 for (i = 0; i < map->num_stripes; i++) {
7345 dev = map->stripes[i].dev;
7346 dev->commit_bytes_used = dev->bytes_used;
7349 mutex_unlock(&fs_info->chunk_mutex);
7352 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7354 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7355 while (fs_devices) {
7356 fs_devices->fs_info = fs_info;
7357 fs_devices = fs_devices->seed;
7361 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7363 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7364 while (fs_devices) {
7365 fs_devices->fs_info = NULL;
7366 fs_devices = fs_devices->seed;
7371 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7373 int btrfs_bg_type_to_factor(u64 flags)
7375 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
7376 BTRFS_BLOCK_GROUP_RAID10))
7382 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7384 int index = btrfs_bg_flags_to_raid_index(type);
7385 int ncopies = btrfs_raid_array[index].ncopies;
7388 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
7389 case BTRFS_BLOCK_GROUP_RAID5:
7390 data_stripes = num_stripes - 1;
7392 case BTRFS_BLOCK_GROUP_RAID6:
7393 data_stripes = num_stripes - 2;
7396 data_stripes = num_stripes / ncopies;
7399 return div_u64(chunk_len, data_stripes);
7402 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7403 u64 chunk_offset, u64 devid,
7404 u64 physical_offset, u64 physical_len)
7406 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7407 struct extent_map *em;
7408 struct map_lookup *map;
7409 struct btrfs_device *dev;
7415 read_lock(&em_tree->lock);
7416 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7417 read_unlock(&em_tree->lock);
7421 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7422 physical_offset, devid);
7427 map = em->map_lookup;
7428 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7429 if (physical_len != stripe_len) {
7431 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7432 physical_offset, devid, em->start, physical_len,
7438 for (i = 0; i < map->num_stripes; i++) {
7439 if (map->stripes[i].dev->devid == devid &&
7440 map->stripes[i].physical == physical_offset) {
7442 if (map->verified_stripes >= map->num_stripes) {
7444 "too many dev extents for chunk %llu found",
7449 map->verified_stripes++;
7455 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7456 physical_offset, devid);
7460 /* Make sure no dev extent is beyond device bondary */
7461 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
7463 btrfs_err(fs_info, "failed to find devid %llu", devid);
7467 if (physical_offset + physical_len > dev->disk_total_bytes) {
7469 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7470 devid, physical_offset, physical_len,
7471 dev->disk_total_bytes);
7476 free_extent_map(em);
7480 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7482 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7483 struct extent_map *em;
7484 struct rb_node *node;
7487 read_lock(&em_tree->lock);
7488 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7489 em = rb_entry(node, struct extent_map, rb_node);
7490 if (em->map_lookup->num_stripes !=
7491 em->map_lookup->verified_stripes) {
7493 "chunk %llu has missing dev extent, have %d expect %d",
7494 em->start, em->map_lookup->verified_stripes,
7495 em->map_lookup->num_stripes);
7501 read_unlock(&em_tree->lock);
7506 * Ensure that all dev extents are mapped to correct chunk, otherwise
7507 * later chunk allocation/free would cause unexpected behavior.
7509 * NOTE: This will iterate through the whole device tree, which should be of
7510 * the same size level as the chunk tree. This slightly increases mount time.
7512 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7514 struct btrfs_path *path;
7515 struct btrfs_root *root = fs_info->dev_root;
7516 struct btrfs_key key;
7518 u64 prev_dev_ext_end = 0;
7522 key.type = BTRFS_DEV_EXTENT_KEY;
7525 path = btrfs_alloc_path();
7529 path->reada = READA_FORWARD;
7530 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7534 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7535 ret = btrfs_next_item(root, path);
7538 /* No dev extents at all? Not good */
7545 struct extent_buffer *leaf = path->nodes[0];
7546 struct btrfs_dev_extent *dext;
7547 int slot = path->slots[0];
7549 u64 physical_offset;
7553 btrfs_item_key_to_cpu(leaf, &key, slot);
7554 if (key.type != BTRFS_DEV_EXTENT_KEY)
7556 devid = key.objectid;
7557 physical_offset = key.offset;
7559 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7560 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7561 physical_len = btrfs_dev_extent_length(leaf, dext);
7563 /* Check if this dev extent overlaps with the previous one */
7564 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7566 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7567 devid, physical_offset, prev_dev_ext_end);
7572 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7573 physical_offset, physical_len);
7577 prev_dev_ext_end = physical_offset + physical_len;
7579 ret = btrfs_next_item(root, path);
7588 /* Ensure all chunks have corresponding dev extents */
7589 ret = verify_chunk_dev_extent_mapping(fs_info);
7591 btrfs_free_path(path);
7596 * Check whether the given block group or device is pinned by any inode being
7597 * used as a swapfile.
7599 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7601 struct btrfs_swapfile_pin *sp;
7602 struct rb_node *node;
7604 spin_lock(&fs_info->swapfile_pins_lock);
7605 node = fs_info->swapfile_pins.rb_node;
7607 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7609 node = node->rb_left;
7610 else if (ptr > sp->ptr)
7611 node = node->rb_right;
7615 spin_unlock(&fs_info->swapfile_pins_lock);
7616 return node != NULL;