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>
19 #include "extent_map.h"
21 #include "transaction.h"
22 #include "print-tree.h"
25 #include "async-thread.h"
26 #include "check-integrity.h"
27 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
32 #include "block-group.h"
34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
35 [BTRFS_RAID_RAID10] = {
38 .devs_max = 0, /* 0 == as many as possible */
40 .tolerated_failures = 1,
44 .raid_name = "raid10",
45 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
46 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
48 [BTRFS_RAID_RAID1] = {
53 .tolerated_failures = 1,
58 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
59 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
66 .tolerated_failures = 0,
71 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
74 [BTRFS_RAID_RAID0] = {
79 .tolerated_failures = 0,
84 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
87 [BTRFS_RAID_SINGLE] = {
92 .tolerated_failures = 0,
96 .raid_name = "single",
100 [BTRFS_RAID_RAID5] = {
105 .tolerated_failures = 1,
109 .raid_name = "raid5",
110 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
111 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
113 [BTRFS_RAID_RAID6] = {
118 .tolerated_failures = 2,
122 .raid_name = "raid6",
123 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
124 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
128 const char *btrfs_bg_type_to_raid_name(u64 flags)
130 const int index = btrfs_bg_flags_to_raid_index(flags);
132 if (index >= BTRFS_NR_RAID_TYPES)
135 return btrfs_raid_array[index].raid_name;
139 * Fill @buf with textual description of @bg_flags, no more than @size_buf
140 * bytes including terminating null byte.
142 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
147 u64 flags = bg_flags;
148 u32 size_bp = size_buf;
155 #define DESCRIBE_FLAG(flag, desc) \
157 if (flags & (flag)) { \
158 ret = snprintf(bp, size_bp, "%s|", (desc)); \
159 if (ret < 0 || ret >= size_bp) \
167 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
168 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
169 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
171 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
172 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
173 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
174 btrfs_raid_array[i].raid_name);
178 ret = snprintf(bp, size_bp, "0x%llx|", flags);
182 if (size_bp < size_buf)
183 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
186 * The text is trimmed, it's up to the caller to provide sufficiently
192 static int init_first_rw_device(struct btrfs_trans_handle *trans);
193 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
194 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
195 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
196 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
197 enum btrfs_map_op op,
198 u64 logical, u64 *length,
199 struct btrfs_bio **bbio_ret,
200 int mirror_num, int need_raid_map);
206 * There are several mutexes that protect manipulation of devices and low-level
207 * structures like chunks but not block groups, extents or files
209 * uuid_mutex (global lock)
210 * ------------------------
211 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
212 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
213 * device) or requested by the device= mount option
215 * the mutex can be very coarse and can cover long-running operations
217 * protects: updates to fs_devices counters like missing devices, rw devices,
218 * seeding, structure cloning, opening/closing devices at mount/umount time
220 * global::fs_devs - add, remove, updates to the global list
222 * does not protect: manipulation of the fs_devices::devices list!
224 * btrfs_device::name - renames (write side), read is RCU
226 * fs_devices::device_list_mutex (per-fs, with RCU)
227 * ------------------------------------------------
228 * protects updates to fs_devices::devices, ie. adding and deleting
230 * simple list traversal with read-only actions can be done with RCU protection
232 * may be used to exclude some operations from running concurrently without any
233 * modifications to the list (see write_all_supers)
237 * protects balance structures (status, state) and context accessed from
238 * several places (internally, ioctl)
242 * protects chunks, adding or removing during allocation, trim or when a new
243 * device is added/removed. Additionally it also protects post_commit_list of
244 * individual devices, since they can be added to the transaction's
245 * post_commit_list only with chunk_mutex held.
249 * a big lock that is held by the cleaner thread and prevents running subvolume
250 * cleaning together with relocation or delayed iputs
263 * Exclusive operations, BTRFS_FS_EXCL_OP
264 * ======================================
266 * Maintains the exclusivity of the following operations that apply to the
267 * whole filesystem and cannot run in parallel.
272 * - Device replace (*)
275 * The device operations (as above) can be in one of the following states:
281 * Only device operations marked with (*) can go into the Paused state for the
284 * - ioctl (only Balance can be Paused through ioctl)
285 * - filesystem remounted as read-only
286 * - filesystem unmounted and mounted as read-only
287 * - system power-cycle and filesystem mounted as read-only
288 * - filesystem or device errors leading to forced read-only
290 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
291 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
292 * A device operation in Paused or Running state can be canceled or resumed
293 * either by ioctl (Balance only) or when remounted as read-write.
294 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
298 DEFINE_MUTEX(uuid_mutex);
299 static LIST_HEAD(fs_uuids);
300 struct list_head *btrfs_get_fs_uuids(void)
306 * alloc_fs_devices - allocate struct btrfs_fs_devices
307 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
308 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
310 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
311 * The returned struct is not linked onto any lists and can be destroyed with
312 * kfree() right away.
314 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
315 const u8 *metadata_fsid)
317 struct btrfs_fs_devices *fs_devs;
319 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
321 return ERR_PTR(-ENOMEM);
323 mutex_init(&fs_devs->device_list_mutex);
325 INIT_LIST_HEAD(&fs_devs->devices);
326 INIT_LIST_HEAD(&fs_devs->alloc_list);
327 INIT_LIST_HEAD(&fs_devs->fs_list);
329 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
332 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
334 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
339 void btrfs_free_device(struct btrfs_device *device)
341 WARN_ON(!list_empty(&device->post_commit_list));
342 rcu_string_free(device->name);
343 extent_io_tree_release(&device->alloc_state);
344 bio_put(device->flush_bio);
348 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
350 struct btrfs_device *device;
351 WARN_ON(fs_devices->opened);
352 while (!list_empty(&fs_devices->devices)) {
353 device = list_entry(fs_devices->devices.next,
354 struct btrfs_device, dev_list);
355 list_del(&device->dev_list);
356 btrfs_free_device(device);
361 void __exit btrfs_cleanup_fs_uuids(void)
363 struct btrfs_fs_devices *fs_devices;
365 while (!list_empty(&fs_uuids)) {
366 fs_devices = list_entry(fs_uuids.next,
367 struct btrfs_fs_devices, fs_list);
368 list_del(&fs_devices->fs_list);
369 free_fs_devices(fs_devices);
374 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
375 * Returned struct is not linked onto any lists and must be destroyed using
378 static struct btrfs_device *__alloc_device(void)
380 struct btrfs_device *dev;
382 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
384 return ERR_PTR(-ENOMEM);
387 * Preallocate a bio that's always going to be used for flushing device
388 * barriers and matches the device lifespan
390 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
391 if (!dev->flush_bio) {
393 return ERR_PTR(-ENOMEM);
396 INIT_LIST_HEAD(&dev->dev_list);
397 INIT_LIST_HEAD(&dev->dev_alloc_list);
398 INIT_LIST_HEAD(&dev->post_commit_list);
400 spin_lock_init(&dev->io_lock);
402 atomic_set(&dev->reada_in_flight, 0);
403 atomic_set(&dev->dev_stats_ccnt, 0);
404 btrfs_device_data_ordered_init(dev);
405 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
406 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
407 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
412 static noinline struct btrfs_fs_devices *find_fsid(
413 const u8 *fsid, const u8 *metadata_fsid)
415 struct btrfs_fs_devices *fs_devices;
421 * Handle scanned device having completed its fsid change but
422 * belonging to a fs_devices that was created by first scanning
423 * a device which didn't have its fsid/metadata_uuid changed
424 * at all and the CHANGING_FSID_V2 flag set.
426 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
427 if (fs_devices->fsid_change &&
428 memcmp(metadata_fsid, fs_devices->fsid,
429 BTRFS_FSID_SIZE) == 0 &&
430 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
431 BTRFS_FSID_SIZE) == 0) {
436 * Handle scanned device having completed its fsid change but
437 * belonging to a fs_devices that was created by a device that
438 * has an outdated pair of fsid/metadata_uuid and
439 * CHANGING_FSID_V2 flag set.
441 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
442 if (fs_devices->fsid_change &&
443 memcmp(fs_devices->metadata_uuid,
444 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
445 memcmp(metadata_fsid, fs_devices->metadata_uuid,
446 BTRFS_FSID_SIZE) == 0) {
452 /* Handle non-split brain cases */
453 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
455 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
456 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
457 BTRFS_FSID_SIZE) == 0)
460 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
468 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
469 int flush, struct block_device **bdev,
470 struct buffer_head **bh)
474 *bdev = blkdev_get_by_path(device_path, flags, holder);
477 ret = PTR_ERR(*bdev);
482 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
483 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
485 blkdev_put(*bdev, flags);
488 invalidate_bdev(*bdev);
489 *bh = btrfs_read_dev_super(*bdev);
492 blkdev_put(*bdev, flags);
504 static void requeue_list(struct btrfs_pending_bios *pending_bios,
505 struct bio *head, struct bio *tail)
508 struct bio *old_head;
510 old_head = pending_bios->head;
511 pending_bios->head = head;
512 if (pending_bios->tail)
513 tail->bi_next = old_head;
515 pending_bios->tail = tail;
519 * we try to collect pending bios for a device so we don't get a large
520 * number of procs sending bios down to the same device. This greatly
521 * improves the schedulers ability to collect and merge the bios.
523 * But, it also turns into a long list of bios to process and that is sure
524 * to eventually make the worker thread block. The solution here is to
525 * make some progress and then put this work struct back at the end of
526 * the list if the block device is congested. This way, multiple devices
527 * can make progress from a single worker thread.
529 static noinline void run_scheduled_bios(struct btrfs_device *device)
531 struct btrfs_fs_info *fs_info = device->fs_info;
533 struct backing_dev_info *bdi;
534 struct btrfs_pending_bios *pending_bios;
538 unsigned long num_run;
539 unsigned long batch_run = 0;
540 unsigned long last_waited = 0;
542 int sync_pending = 0;
543 struct blk_plug plug;
546 * this function runs all the bios we've collected for
547 * a particular device. We don't want to wander off to
548 * another device without first sending all of these down.
549 * So, setup a plug here and finish it off before we return
551 blk_start_plug(&plug);
553 bdi = device->bdev->bd_bdi;
556 spin_lock(&device->io_lock);
561 /* take all the bios off the list at once and process them
562 * later on (without the lock held). But, remember the
563 * tail and other pointers so the bios can be properly reinserted
564 * into the list if we hit congestion
566 if (!force_reg && device->pending_sync_bios.head) {
567 pending_bios = &device->pending_sync_bios;
570 pending_bios = &device->pending_bios;
574 pending = pending_bios->head;
575 tail = pending_bios->tail;
576 WARN_ON(pending && !tail);
579 * if pending was null this time around, no bios need processing
580 * at all and we can stop. Otherwise it'll loop back up again
581 * and do an additional check so no bios are missed.
583 * device->running_pending is used to synchronize with the
586 if (device->pending_sync_bios.head == NULL &&
587 device->pending_bios.head == NULL) {
589 device->running_pending = 0;
592 device->running_pending = 1;
595 pending_bios->head = NULL;
596 pending_bios->tail = NULL;
598 spin_unlock(&device->io_lock);
603 /* we want to work on both lists, but do more bios on the
604 * sync list than the regular list
607 pending_bios != &device->pending_sync_bios &&
608 device->pending_sync_bios.head) ||
609 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
610 device->pending_bios.head)) {
611 spin_lock(&device->io_lock);
612 requeue_list(pending_bios, pending, tail);
617 pending = pending->bi_next;
620 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
623 * if we're doing the sync list, record that our
624 * plug has some sync requests on it
626 * If we're doing the regular list and there are
627 * sync requests sitting around, unplug before
630 if (pending_bios == &device->pending_sync_bios) {
632 } else if (sync_pending) {
633 blk_finish_plug(&plug);
634 blk_start_plug(&plug);
638 btrfsic_submit_bio(cur);
645 * we made progress, there is more work to do and the bdi
646 * is now congested. Back off and let other work structs
649 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
650 fs_info->fs_devices->open_devices > 1) {
651 struct io_context *ioc;
653 ioc = current->io_context;
656 * the main goal here is that we don't want to
657 * block if we're going to be able to submit
658 * more requests without blocking.
660 * This code does two great things, it pokes into
661 * the elevator code from a filesystem _and_
662 * it makes assumptions about how batching works.
664 if (ioc && ioc->nr_batch_requests > 0 &&
665 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
667 ioc->last_waited == last_waited)) {
669 * we want to go through our batch of
670 * requests and stop. So, we copy out
671 * the ioc->last_waited time and test
672 * against it before looping
674 last_waited = ioc->last_waited;
678 spin_lock(&device->io_lock);
679 requeue_list(pending_bios, pending, tail);
680 device->running_pending = 1;
682 spin_unlock(&device->io_lock);
683 btrfs_queue_work(fs_info->submit_workers,
693 spin_lock(&device->io_lock);
694 if (device->pending_bios.head || device->pending_sync_bios.head)
696 spin_unlock(&device->io_lock);
699 blk_finish_plug(&plug);
702 static void pending_bios_fn(struct btrfs_work *work)
704 struct btrfs_device *device;
706 device = container_of(work, struct btrfs_device, work);
707 run_scheduled_bios(device);
710 static bool device_path_matched(const char *path, struct btrfs_device *device)
715 found = strcmp(rcu_str_deref(device->name), path);
722 * Search and remove all stale (devices which are not mounted) devices.
723 * When both inputs are NULL, it will search and release all stale devices.
724 * path: Optional. When provided will it release all unmounted devices
725 * matching this path only.
726 * skip_dev: Optional. Will skip this device when searching for the stale
728 * Return: 0 for success or if @path is NULL.
729 * -EBUSY if @path is a mounted device.
730 * -ENOENT if @path does not match any device in the list.
732 static int btrfs_free_stale_devices(const char *path,
733 struct btrfs_device *skip_device)
735 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
736 struct btrfs_device *device, *tmp_device;
742 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
744 mutex_lock(&fs_devices->device_list_mutex);
745 list_for_each_entry_safe(device, tmp_device,
746 &fs_devices->devices, dev_list) {
747 if (skip_device && skip_device == device)
749 if (path && !device->name)
751 if (path && !device_path_matched(path, device))
753 if (fs_devices->opened) {
754 /* for an already deleted device return 0 */
755 if (path && ret != 0)
760 /* delete the stale device */
761 fs_devices->num_devices--;
762 list_del(&device->dev_list);
763 btrfs_free_device(device);
766 if (fs_devices->num_devices == 0)
769 mutex_unlock(&fs_devices->device_list_mutex);
771 if (fs_devices->num_devices == 0) {
772 btrfs_sysfs_remove_fsid(fs_devices);
773 list_del(&fs_devices->fs_list);
774 free_fs_devices(fs_devices);
781 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
782 struct btrfs_device *device, fmode_t flags,
785 struct request_queue *q;
786 struct block_device *bdev;
787 struct buffer_head *bh;
788 struct btrfs_super_block *disk_super;
797 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
802 disk_super = (struct btrfs_super_block *)bh->b_data;
803 devid = btrfs_stack_device_id(&disk_super->dev_item);
804 if (devid != device->devid)
807 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
810 device->generation = btrfs_super_generation(disk_super);
812 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
813 if (btrfs_super_incompat_flags(disk_super) &
814 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
816 "BTRFS: Invalid seeding and uuid-changed device detected\n");
820 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
821 fs_devices->seeding = 1;
823 if (bdev_read_only(bdev))
824 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
826 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
829 q = bdev_get_queue(bdev);
830 if (!blk_queue_nonrot(q))
831 fs_devices->rotating = 1;
834 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
835 device->mode = flags;
837 fs_devices->open_devices++;
838 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
839 device->devid != BTRFS_DEV_REPLACE_DEVID) {
840 fs_devices->rw_devices++;
841 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
849 blkdev_put(bdev, flags);
855 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
856 * being created with a disk that has already completed its fsid change.
858 static struct btrfs_fs_devices *find_fsid_inprogress(
859 struct btrfs_super_block *disk_super)
861 struct btrfs_fs_devices *fs_devices;
863 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
864 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
865 BTRFS_FSID_SIZE) != 0 &&
866 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
867 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
876 static struct btrfs_fs_devices *find_fsid_changed(
877 struct btrfs_super_block *disk_super)
879 struct btrfs_fs_devices *fs_devices;
882 * Handles the case where scanned device is part of an fs that had
883 * multiple successful changes of FSID but curently device didn't
884 * observe it. Meaning our fsid will be different than theirs.
886 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
887 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
888 BTRFS_FSID_SIZE) != 0 &&
889 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
890 BTRFS_FSID_SIZE) == 0 &&
891 memcmp(fs_devices->fsid, disk_super->fsid,
892 BTRFS_FSID_SIZE) != 0) {
900 * Add new device to list of registered devices
903 * device pointer which was just added or updated when successful
904 * error pointer when failed
906 static noinline struct btrfs_device *device_list_add(const char *path,
907 struct btrfs_super_block *disk_super,
908 bool *new_device_added)
910 struct btrfs_device *device;
911 struct btrfs_fs_devices *fs_devices = NULL;
912 struct rcu_string *name;
913 u64 found_transid = btrfs_super_generation(disk_super);
914 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
915 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
916 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
917 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
918 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
920 if (fsid_change_in_progress) {
921 if (!has_metadata_uuid) {
923 * When we have an image which has CHANGING_FSID_V2 set
924 * it might belong to either a filesystem which has
925 * disks with completed fsid change or it might belong
926 * to fs with no UUID changes in effect, handle both.
928 fs_devices = find_fsid_inprogress(disk_super);
930 fs_devices = find_fsid(disk_super->fsid, NULL);
932 fs_devices = find_fsid_changed(disk_super);
934 } else if (has_metadata_uuid) {
935 fs_devices = find_fsid(disk_super->fsid,
936 disk_super->metadata_uuid);
938 fs_devices = find_fsid(disk_super->fsid, NULL);
943 if (has_metadata_uuid)
944 fs_devices = alloc_fs_devices(disk_super->fsid,
945 disk_super->metadata_uuid);
947 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
949 if (IS_ERR(fs_devices))
950 return ERR_CAST(fs_devices);
952 fs_devices->fsid_change = fsid_change_in_progress;
954 mutex_lock(&fs_devices->device_list_mutex);
955 list_add(&fs_devices->fs_list, &fs_uuids);
959 mutex_lock(&fs_devices->device_list_mutex);
960 device = btrfs_find_device(fs_devices, devid,
961 disk_super->dev_item.uuid, NULL, false);
964 * If this disk has been pulled into an fs devices created by
965 * a device which had the CHANGING_FSID_V2 flag then replace the
966 * metadata_uuid/fsid values of the fs_devices.
968 if (has_metadata_uuid && fs_devices->fsid_change &&
969 found_transid > fs_devices->latest_generation) {
970 memcpy(fs_devices->fsid, disk_super->fsid,
972 memcpy(fs_devices->metadata_uuid,
973 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
975 fs_devices->fsid_change = false;
980 if (fs_devices->opened) {
981 mutex_unlock(&fs_devices->device_list_mutex);
982 return ERR_PTR(-EBUSY);
985 device = btrfs_alloc_device(NULL, &devid,
986 disk_super->dev_item.uuid);
987 if (IS_ERR(device)) {
988 mutex_unlock(&fs_devices->device_list_mutex);
989 /* we can safely leave the fs_devices entry around */
993 name = rcu_string_strdup(path, GFP_NOFS);
995 btrfs_free_device(device);
996 mutex_unlock(&fs_devices->device_list_mutex);
997 return ERR_PTR(-ENOMEM);
999 rcu_assign_pointer(device->name, name);
1001 list_add_rcu(&device->dev_list, &fs_devices->devices);
1002 fs_devices->num_devices++;
1004 device->fs_devices = fs_devices;
1005 *new_device_added = true;
1007 if (disk_super->label[0])
1008 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1009 disk_super->label, devid, found_transid, path);
1011 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1012 disk_super->fsid, devid, found_transid, path);
1014 } else if (!device->name || strcmp(device->name->str, path)) {
1016 * When FS is already mounted.
1017 * 1. If you are here and if the device->name is NULL that
1018 * means this device was missing at time of FS mount.
1019 * 2. If you are here and if the device->name is different
1020 * from 'path' that means either
1021 * a. The same device disappeared and reappeared with
1022 * different name. or
1023 * b. The missing-disk-which-was-replaced, has
1026 * We must allow 1 and 2a above. But 2b would be a spurious
1027 * and unintentional.
1029 * Further in case of 1 and 2a above, the disk at 'path'
1030 * would have missed some transaction when it was away and
1031 * in case of 2a the stale bdev has to be updated as well.
1032 * 2b must not be allowed at all time.
1036 * For now, we do allow update to btrfs_fs_device through the
1037 * btrfs dev scan cli after FS has been mounted. We're still
1038 * tracking a problem where systems fail mount by subvolume id
1039 * when we reject replacement on a mounted FS.
1041 if (!fs_devices->opened && found_transid < device->generation) {
1043 * That is if the FS is _not_ mounted and if you
1044 * are here, that means there is more than one
1045 * disk with same uuid and devid.We keep the one
1046 * with larger generation number or the last-in if
1047 * generation are equal.
1049 mutex_unlock(&fs_devices->device_list_mutex);
1050 return ERR_PTR(-EEXIST);
1054 * We are going to replace the device path for a given devid,
1055 * make sure it's the same device if the device is mounted
1058 struct block_device *path_bdev;
1060 path_bdev = lookup_bdev(path);
1061 if (IS_ERR(path_bdev)) {
1062 mutex_unlock(&fs_devices->device_list_mutex);
1063 return ERR_CAST(path_bdev);
1066 if (device->bdev != path_bdev) {
1068 mutex_unlock(&fs_devices->device_list_mutex);
1069 btrfs_warn_in_rcu(device->fs_info,
1070 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1071 disk_super->fsid, devid,
1072 rcu_str_deref(device->name), path);
1073 return ERR_PTR(-EEXIST);
1076 btrfs_info_in_rcu(device->fs_info,
1077 "device fsid %pU devid %llu moved old:%s new:%s",
1078 disk_super->fsid, devid,
1079 rcu_str_deref(device->name), path);
1082 name = rcu_string_strdup(path, GFP_NOFS);
1084 mutex_unlock(&fs_devices->device_list_mutex);
1085 return ERR_PTR(-ENOMEM);
1087 rcu_string_free(device->name);
1088 rcu_assign_pointer(device->name, name);
1089 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1090 fs_devices->missing_devices--;
1091 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1096 * Unmount does not free the btrfs_device struct but would zero
1097 * generation along with most of the other members. So just update
1098 * it back. We need it to pick the disk with largest generation
1101 if (!fs_devices->opened) {
1102 device->generation = found_transid;
1103 fs_devices->latest_generation = max_t(u64, found_transid,
1104 fs_devices->latest_generation);
1107 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1109 mutex_unlock(&fs_devices->device_list_mutex);
1113 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1115 struct btrfs_fs_devices *fs_devices;
1116 struct btrfs_device *device;
1117 struct btrfs_device *orig_dev;
1120 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1121 if (IS_ERR(fs_devices))
1124 mutex_lock(&orig->device_list_mutex);
1125 fs_devices->total_devices = orig->total_devices;
1127 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1128 struct rcu_string *name;
1130 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1132 if (IS_ERR(device)) {
1133 ret = PTR_ERR(device);
1138 * This is ok to do without rcu read locked because we hold the
1139 * uuid mutex so nothing we touch in here is going to disappear.
1141 if (orig_dev->name) {
1142 name = rcu_string_strdup(orig_dev->name->str,
1145 btrfs_free_device(device);
1149 rcu_assign_pointer(device->name, name);
1152 list_add(&device->dev_list, &fs_devices->devices);
1153 device->fs_devices = fs_devices;
1154 fs_devices->num_devices++;
1156 mutex_unlock(&orig->device_list_mutex);
1159 mutex_unlock(&orig->device_list_mutex);
1160 free_fs_devices(fs_devices);
1161 return ERR_PTR(ret);
1165 * After we have read the system tree and know devids belonging to
1166 * this filesystem, remove the device which does not belong there.
1168 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1170 struct btrfs_device *device, *next;
1171 struct btrfs_device *latest_dev = NULL;
1173 mutex_lock(&uuid_mutex);
1175 /* This is the initialized path, it is safe to release the devices. */
1176 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1177 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1178 &device->dev_state)) {
1179 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1180 &device->dev_state) &&
1182 device->generation > latest_dev->generation)) {
1183 latest_dev = device;
1188 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1190 * In the first step, keep the device which has
1191 * the correct fsid and the devid that is used
1192 * for the dev_replace procedure.
1193 * In the second step, the dev_replace state is
1194 * read from the device tree and it is known
1195 * whether the procedure is really active or
1196 * not, which means whether this device is
1197 * used or whether it should be removed.
1199 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1200 &device->dev_state)) {
1205 blkdev_put(device->bdev, device->mode);
1206 device->bdev = NULL;
1207 fs_devices->open_devices--;
1209 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1210 list_del_init(&device->dev_alloc_list);
1211 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1212 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1213 &device->dev_state))
1214 fs_devices->rw_devices--;
1216 list_del_init(&device->dev_list);
1217 fs_devices->num_devices--;
1218 btrfs_free_device(device);
1221 if (fs_devices->seed) {
1222 fs_devices = fs_devices->seed;
1226 fs_devices->latest_bdev = latest_dev->bdev;
1228 mutex_unlock(&uuid_mutex);
1231 static void btrfs_close_bdev(struct btrfs_device *device)
1236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1237 sync_blockdev(device->bdev);
1238 invalidate_bdev(device->bdev);
1241 blkdev_put(device->bdev, device->mode);
1244 static void btrfs_close_one_device(struct btrfs_device *device)
1246 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1247 struct btrfs_device *new_device;
1248 struct rcu_string *name;
1251 fs_devices->open_devices--;
1253 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1254 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1255 list_del_init(&device->dev_alloc_list);
1256 fs_devices->rw_devices--;
1259 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1260 fs_devices->missing_devices--;
1262 btrfs_close_bdev(device);
1264 new_device = btrfs_alloc_device(NULL, &device->devid,
1266 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1268 /* Safe because we are under uuid_mutex */
1270 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1271 BUG_ON(!name); /* -ENOMEM */
1272 rcu_assign_pointer(new_device->name, name);
1275 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1276 new_device->fs_devices = device->fs_devices;
1279 btrfs_free_device(device);
1282 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1284 struct btrfs_device *device, *tmp;
1286 if (--fs_devices->opened > 0)
1289 mutex_lock(&fs_devices->device_list_mutex);
1290 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1291 btrfs_close_one_device(device);
1293 mutex_unlock(&fs_devices->device_list_mutex);
1295 WARN_ON(fs_devices->open_devices);
1296 WARN_ON(fs_devices->rw_devices);
1297 fs_devices->opened = 0;
1298 fs_devices->seeding = 0;
1303 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1305 struct btrfs_fs_devices *seed_devices = NULL;
1308 mutex_lock(&uuid_mutex);
1309 ret = close_fs_devices(fs_devices);
1310 if (!fs_devices->opened) {
1311 seed_devices = fs_devices->seed;
1312 fs_devices->seed = NULL;
1314 mutex_unlock(&uuid_mutex);
1316 while (seed_devices) {
1317 fs_devices = seed_devices;
1318 seed_devices = fs_devices->seed;
1319 close_fs_devices(fs_devices);
1320 free_fs_devices(fs_devices);
1325 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1326 fmode_t flags, void *holder)
1328 struct btrfs_device *device;
1329 struct btrfs_device *latest_dev = NULL;
1332 flags |= FMODE_EXCL;
1334 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1335 /* Just open everything we can; ignore failures here */
1336 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1340 device->generation > latest_dev->generation)
1341 latest_dev = device;
1343 if (fs_devices->open_devices == 0) {
1347 fs_devices->opened = 1;
1348 fs_devices->latest_bdev = latest_dev->bdev;
1349 fs_devices->total_rw_bytes = 0;
1354 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1356 struct btrfs_device *dev1, *dev2;
1358 dev1 = list_entry(a, struct btrfs_device, dev_list);
1359 dev2 = list_entry(b, struct btrfs_device, dev_list);
1361 if (dev1->devid < dev2->devid)
1363 else if (dev1->devid > dev2->devid)
1368 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1369 fmode_t flags, void *holder)
1373 lockdep_assert_held(&uuid_mutex);
1375 mutex_lock(&fs_devices->device_list_mutex);
1376 if (fs_devices->opened) {
1377 fs_devices->opened++;
1380 list_sort(NULL, &fs_devices->devices, devid_cmp);
1381 ret = open_fs_devices(fs_devices, flags, holder);
1383 mutex_unlock(&fs_devices->device_list_mutex);
1388 static void btrfs_release_disk_super(struct page *page)
1394 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1396 struct btrfs_super_block **disk_super)
1401 /* make sure our super fits in the device */
1402 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1405 /* make sure our super fits in the page */
1406 if (sizeof(**disk_super) > PAGE_SIZE)
1409 /* make sure our super doesn't straddle pages on disk */
1410 index = bytenr >> PAGE_SHIFT;
1411 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1414 /* pull in the page with our super */
1415 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1418 if (IS_ERR_OR_NULL(*page))
1423 /* align our pointer to the offset of the super block */
1424 *disk_super = p + offset_in_page(bytenr);
1426 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1427 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1428 btrfs_release_disk_super(*page);
1432 if ((*disk_super)->label[0] &&
1433 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1434 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1439 int btrfs_forget_devices(const char *path)
1443 mutex_lock(&uuid_mutex);
1444 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1445 mutex_unlock(&uuid_mutex);
1451 * Look for a btrfs signature on a device. This may be called out of the mount path
1452 * and we are not allowed to call set_blocksize during the scan. The superblock
1453 * is read via pagecache
1455 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1458 struct btrfs_super_block *disk_super;
1459 bool new_device_added = false;
1460 struct btrfs_device *device = NULL;
1461 struct block_device *bdev;
1465 lockdep_assert_held(&uuid_mutex);
1468 * we would like to check all the supers, but that would make
1469 * a btrfs mount succeed after a mkfs from a different FS.
1470 * So, we need to add a special mount option to scan for
1471 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1473 bytenr = btrfs_sb_offset(0);
1474 flags |= FMODE_EXCL;
1476 bdev = blkdev_get_by_path(path, flags, holder);
1478 return ERR_CAST(bdev);
1480 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1481 device = ERR_PTR(-EINVAL);
1482 goto error_bdev_put;
1485 device = device_list_add(path, disk_super, &new_device_added);
1486 if (!IS_ERR(device)) {
1487 if (new_device_added)
1488 btrfs_free_stale_devices(path, device);
1491 btrfs_release_disk_super(page);
1494 blkdev_put(bdev, flags);
1500 * Try to find a chunk that intersects [start, start + len] range and when one
1501 * such is found, record the end of it in *start
1503 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1506 u64 physical_start, physical_end;
1508 lockdep_assert_held(&device->fs_info->chunk_mutex);
1510 if (!find_first_extent_bit(&device->alloc_state, *start,
1511 &physical_start, &physical_end,
1512 CHUNK_ALLOCATED, NULL)) {
1514 if (in_range(physical_start, *start, len) ||
1515 in_range(*start, physical_start,
1516 physical_end - physical_start)) {
1517 *start = physical_end + 1;
1526 * find_free_dev_extent_start - find free space in the specified device
1527 * @device: the device which we search the free space in
1528 * @num_bytes: the size of the free space that we need
1529 * @search_start: the position from which to begin the search
1530 * @start: store the start of the free space.
1531 * @len: the size of the free space. that we find, or the size
1532 * of the max free space if we don't find suitable free space
1534 * this uses a pretty simple search, the expectation is that it is
1535 * called very infrequently and that a given device has a small number
1538 * @start is used to store the start of the free space if we find. But if we
1539 * don't find suitable free space, it will be used to store the start position
1540 * of the max free space.
1542 * @len is used to store the size of the free space that we find.
1543 * But if we don't find suitable free space, it is used to store the size of
1544 * the max free space.
1546 * NOTE: This function will search *commit* root of device tree, and does extra
1547 * check to ensure dev extents are not double allocated.
1548 * This makes the function safe to allocate dev extents but may not report
1549 * correct usable device space, as device extent freed in current transaction
1550 * is not reported as avaiable.
1552 static int find_free_dev_extent_start(struct btrfs_device *device,
1553 u64 num_bytes, u64 search_start, u64 *start,
1556 struct btrfs_fs_info *fs_info = device->fs_info;
1557 struct btrfs_root *root = fs_info->dev_root;
1558 struct btrfs_key key;
1559 struct btrfs_dev_extent *dev_extent;
1560 struct btrfs_path *path;
1565 u64 search_end = device->total_bytes;
1568 struct extent_buffer *l;
1571 * We don't want to overwrite the superblock on the drive nor any area
1572 * used by the boot loader (grub for example), so we make sure to start
1573 * at an offset of at least 1MB.
1575 search_start = max_t(u64, search_start, SZ_1M);
1577 path = btrfs_alloc_path();
1581 max_hole_start = search_start;
1585 if (search_start >= search_end ||
1586 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1591 path->reada = READA_FORWARD;
1592 path->search_commit_root = 1;
1593 path->skip_locking = 1;
1595 key.objectid = device->devid;
1596 key.offset = search_start;
1597 key.type = BTRFS_DEV_EXTENT_KEY;
1599 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1603 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1610 slot = path->slots[0];
1611 if (slot >= btrfs_header_nritems(l)) {
1612 ret = btrfs_next_leaf(root, path);
1620 btrfs_item_key_to_cpu(l, &key, slot);
1622 if (key.objectid < device->devid)
1625 if (key.objectid > device->devid)
1628 if (key.type != BTRFS_DEV_EXTENT_KEY)
1631 if (key.offset > search_start) {
1632 hole_size = key.offset - search_start;
1635 * Have to check before we set max_hole_start, otherwise
1636 * we could end up sending back this offset anyway.
1638 if (contains_pending_extent(device, &search_start,
1640 if (key.offset >= search_start)
1641 hole_size = key.offset - search_start;
1646 if (hole_size > max_hole_size) {
1647 max_hole_start = search_start;
1648 max_hole_size = hole_size;
1652 * If this free space is greater than which we need,
1653 * it must be the max free space that we have found
1654 * until now, so max_hole_start must point to the start
1655 * of this free space and the length of this free space
1656 * is stored in max_hole_size. Thus, we return
1657 * max_hole_start and max_hole_size and go back to the
1660 if (hole_size >= num_bytes) {
1666 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1667 extent_end = key.offset + btrfs_dev_extent_length(l,
1669 if (extent_end > search_start)
1670 search_start = extent_end;
1677 * At this point, search_start should be the end of
1678 * allocated dev extents, and when shrinking the device,
1679 * search_end may be smaller than search_start.
1681 if (search_end > search_start) {
1682 hole_size = search_end - search_start;
1684 if (contains_pending_extent(device, &search_start, hole_size)) {
1685 btrfs_release_path(path);
1689 if (hole_size > max_hole_size) {
1690 max_hole_start = search_start;
1691 max_hole_size = hole_size;
1696 if (max_hole_size < num_bytes)
1702 btrfs_free_path(path);
1703 *start = max_hole_start;
1705 *len = max_hole_size;
1709 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1710 u64 *start, u64 *len)
1712 /* FIXME use last free of some kind */
1713 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1716 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1717 struct btrfs_device *device,
1718 u64 start, u64 *dev_extent_len)
1720 struct btrfs_fs_info *fs_info = device->fs_info;
1721 struct btrfs_root *root = fs_info->dev_root;
1723 struct btrfs_path *path;
1724 struct btrfs_key key;
1725 struct btrfs_key found_key;
1726 struct extent_buffer *leaf = NULL;
1727 struct btrfs_dev_extent *extent = NULL;
1729 path = btrfs_alloc_path();
1733 key.objectid = device->devid;
1735 key.type = BTRFS_DEV_EXTENT_KEY;
1737 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1739 ret = btrfs_previous_item(root, path, key.objectid,
1740 BTRFS_DEV_EXTENT_KEY);
1743 leaf = path->nodes[0];
1744 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1745 extent = btrfs_item_ptr(leaf, path->slots[0],
1746 struct btrfs_dev_extent);
1747 BUG_ON(found_key.offset > start || found_key.offset +
1748 btrfs_dev_extent_length(leaf, extent) < start);
1750 btrfs_release_path(path);
1752 } else if (ret == 0) {
1753 leaf = path->nodes[0];
1754 extent = btrfs_item_ptr(leaf, path->slots[0],
1755 struct btrfs_dev_extent);
1757 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1761 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1763 ret = btrfs_del_item(trans, root, path);
1765 btrfs_handle_fs_error(fs_info, ret,
1766 "Failed to remove dev extent item");
1768 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1771 btrfs_free_path(path);
1775 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1776 struct btrfs_device *device,
1777 u64 chunk_offset, u64 start, u64 num_bytes)
1780 struct btrfs_path *path;
1781 struct btrfs_fs_info *fs_info = device->fs_info;
1782 struct btrfs_root *root = fs_info->dev_root;
1783 struct btrfs_dev_extent *extent;
1784 struct extent_buffer *leaf;
1785 struct btrfs_key key;
1787 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1788 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1789 path = btrfs_alloc_path();
1793 key.objectid = device->devid;
1795 key.type = BTRFS_DEV_EXTENT_KEY;
1796 ret = btrfs_insert_empty_item(trans, root, path, &key,
1801 leaf = path->nodes[0];
1802 extent = btrfs_item_ptr(leaf, path->slots[0],
1803 struct btrfs_dev_extent);
1804 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1805 BTRFS_CHUNK_TREE_OBJECTID);
1806 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1807 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1808 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1810 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1811 btrfs_mark_buffer_dirty(leaf);
1813 btrfs_free_path(path);
1817 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1819 struct extent_map_tree *em_tree;
1820 struct extent_map *em;
1824 em_tree = &fs_info->mapping_tree;
1825 read_lock(&em_tree->lock);
1826 n = rb_last(&em_tree->map.rb_root);
1828 em = rb_entry(n, struct extent_map, rb_node);
1829 ret = em->start + em->len;
1831 read_unlock(&em_tree->lock);
1836 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1840 struct btrfs_key key;
1841 struct btrfs_key found_key;
1842 struct btrfs_path *path;
1844 path = btrfs_alloc_path();
1848 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1849 key.type = BTRFS_DEV_ITEM_KEY;
1850 key.offset = (u64)-1;
1852 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1858 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1863 ret = btrfs_previous_item(fs_info->chunk_root, path,
1864 BTRFS_DEV_ITEMS_OBJECTID,
1865 BTRFS_DEV_ITEM_KEY);
1869 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1871 *devid_ret = found_key.offset + 1;
1875 btrfs_free_path(path);
1880 * the device information is stored in the chunk root
1881 * the btrfs_device struct should be fully filled in
1883 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1884 struct btrfs_device *device)
1887 struct btrfs_path *path;
1888 struct btrfs_dev_item *dev_item;
1889 struct extent_buffer *leaf;
1890 struct btrfs_key key;
1893 path = btrfs_alloc_path();
1897 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1898 key.type = BTRFS_DEV_ITEM_KEY;
1899 key.offset = device->devid;
1901 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1902 &key, sizeof(*dev_item));
1906 leaf = path->nodes[0];
1907 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1909 btrfs_set_device_id(leaf, dev_item, device->devid);
1910 btrfs_set_device_generation(leaf, dev_item, 0);
1911 btrfs_set_device_type(leaf, dev_item, device->type);
1912 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1913 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1914 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1915 btrfs_set_device_total_bytes(leaf, dev_item,
1916 btrfs_device_get_disk_total_bytes(device));
1917 btrfs_set_device_bytes_used(leaf, dev_item,
1918 btrfs_device_get_bytes_used(device));
1919 btrfs_set_device_group(leaf, dev_item, 0);
1920 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1921 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1922 btrfs_set_device_start_offset(leaf, dev_item, 0);
1924 ptr = btrfs_device_uuid(dev_item);
1925 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1926 ptr = btrfs_device_fsid(dev_item);
1927 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1928 ptr, BTRFS_FSID_SIZE);
1929 btrfs_mark_buffer_dirty(leaf);
1933 btrfs_free_path(path);
1938 * Function to update ctime/mtime for a given device path.
1939 * Mainly used for ctime/mtime based probe like libblkid.
1941 static void update_dev_time(const char *path_name)
1945 filp = filp_open(path_name, O_RDWR, 0);
1948 file_update_time(filp);
1949 filp_close(filp, NULL);
1952 static int btrfs_rm_dev_item(struct btrfs_device *device)
1954 struct btrfs_root *root = device->fs_info->chunk_root;
1956 struct btrfs_path *path;
1957 struct btrfs_key key;
1958 struct btrfs_trans_handle *trans;
1960 path = btrfs_alloc_path();
1964 trans = btrfs_start_transaction(root, 0);
1965 if (IS_ERR(trans)) {
1966 btrfs_free_path(path);
1967 return PTR_ERR(trans);
1969 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1970 key.type = BTRFS_DEV_ITEM_KEY;
1971 key.offset = device->devid;
1973 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1977 btrfs_abort_transaction(trans, ret);
1978 btrfs_end_transaction(trans);
1982 ret = btrfs_del_item(trans, root, path);
1984 btrfs_abort_transaction(trans, ret);
1985 btrfs_end_transaction(trans);
1989 btrfs_free_path(path);
1991 ret = btrfs_commit_transaction(trans);
1996 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1997 * filesystem. It's up to the caller to adjust that number regarding eg. device
2000 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2008 seq = read_seqbegin(&fs_info->profiles_lock);
2010 all_avail = fs_info->avail_data_alloc_bits |
2011 fs_info->avail_system_alloc_bits |
2012 fs_info->avail_metadata_alloc_bits;
2013 } while (read_seqretry(&fs_info->profiles_lock, seq));
2015 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2016 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2019 if (num_devices < btrfs_raid_array[i].devs_min) {
2020 int ret = btrfs_raid_array[i].mindev_error;
2030 static struct btrfs_device * btrfs_find_next_active_device(
2031 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2033 struct btrfs_device *next_device;
2035 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2036 if (next_device != device &&
2037 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2038 && next_device->bdev)
2046 * Helper function to check if the given device is part of s_bdev / latest_bdev
2047 * and replace it with the provided or the next active device, in the context
2048 * where this function called, there should be always be another device (or
2049 * this_dev) which is active.
2051 void btrfs_assign_next_active_device(struct btrfs_device *device,
2052 struct btrfs_device *this_dev)
2054 struct btrfs_fs_info *fs_info = device->fs_info;
2055 struct btrfs_device *next_device;
2058 next_device = this_dev;
2060 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2062 ASSERT(next_device);
2064 if (fs_info->sb->s_bdev &&
2065 (fs_info->sb->s_bdev == device->bdev))
2066 fs_info->sb->s_bdev = next_device->bdev;
2068 if (fs_info->fs_devices->latest_bdev == device->bdev)
2069 fs_info->fs_devices->latest_bdev = next_device->bdev;
2073 * Return btrfs_fs_devices::num_devices excluding the device that's being
2074 * currently replaced.
2076 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2078 u64 num_devices = fs_info->fs_devices->num_devices;
2080 down_read(&fs_info->dev_replace.rwsem);
2081 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2082 ASSERT(num_devices > 1);
2085 up_read(&fs_info->dev_replace.rwsem);
2090 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2093 struct btrfs_device *device;
2094 struct btrfs_fs_devices *cur_devices;
2095 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2099 mutex_lock(&uuid_mutex);
2101 num_devices = btrfs_num_devices(fs_info);
2103 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2107 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2109 if (IS_ERR(device)) {
2110 if (PTR_ERR(device) == -ENOENT &&
2111 strcmp(device_path, "missing") == 0)
2112 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2114 ret = PTR_ERR(device);
2118 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2119 btrfs_warn_in_rcu(fs_info,
2120 "cannot remove device %s (devid %llu) due to active swapfile",
2121 rcu_str_deref(device->name), device->devid);
2126 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2127 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2131 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2132 fs_info->fs_devices->rw_devices == 1) {
2133 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2137 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2138 mutex_lock(&fs_info->chunk_mutex);
2139 list_del_init(&device->dev_alloc_list);
2140 device->fs_devices->rw_devices--;
2141 mutex_unlock(&fs_info->chunk_mutex);
2144 mutex_unlock(&uuid_mutex);
2145 ret = btrfs_shrink_device(device, 0);
2146 mutex_lock(&uuid_mutex);
2151 * TODO: the superblock still includes this device in its num_devices
2152 * counter although write_all_supers() is not locked out. This
2153 * could give a filesystem state which requires a degraded mount.
2155 ret = btrfs_rm_dev_item(device);
2159 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2160 btrfs_scrub_cancel_dev(device);
2163 * the device list mutex makes sure that we don't change
2164 * the device list while someone else is writing out all
2165 * the device supers. Whoever is writing all supers, should
2166 * lock the device list mutex before getting the number of
2167 * devices in the super block (super_copy). Conversely,
2168 * whoever updates the number of devices in the super block
2169 * (super_copy) should hold the device list mutex.
2173 * In normal cases the cur_devices == fs_devices. But in case
2174 * of deleting a seed device, the cur_devices should point to
2175 * its own fs_devices listed under the fs_devices->seed.
2177 cur_devices = device->fs_devices;
2178 mutex_lock(&fs_devices->device_list_mutex);
2179 list_del_rcu(&device->dev_list);
2181 cur_devices->num_devices--;
2182 cur_devices->total_devices--;
2183 /* Update total_devices of the parent fs_devices if it's seed */
2184 if (cur_devices != fs_devices)
2185 fs_devices->total_devices--;
2187 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2188 cur_devices->missing_devices--;
2190 btrfs_assign_next_active_device(device, NULL);
2193 cur_devices->open_devices--;
2194 /* remove sysfs entry */
2195 btrfs_sysfs_rm_device_link(fs_devices, device);
2198 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2199 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2200 mutex_unlock(&fs_devices->device_list_mutex);
2203 * at this point, the device is zero sized and detached from
2204 * the devices list. All that's left is to zero out the old
2205 * supers and free the device.
2207 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2208 btrfs_scratch_superblocks(device->bdev, device->name->str);
2210 btrfs_close_bdev(device);
2212 btrfs_free_device(device);
2214 if (cur_devices->open_devices == 0) {
2215 while (fs_devices) {
2216 if (fs_devices->seed == cur_devices) {
2217 fs_devices->seed = cur_devices->seed;
2220 fs_devices = fs_devices->seed;
2222 cur_devices->seed = NULL;
2223 close_fs_devices(cur_devices);
2224 free_fs_devices(cur_devices);
2228 mutex_unlock(&uuid_mutex);
2232 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2233 mutex_lock(&fs_info->chunk_mutex);
2234 list_add(&device->dev_alloc_list,
2235 &fs_devices->alloc_list);
2236 device->fs_devices->rw_devices++;
2237 mutex_unlock(&fs_info->chunk_mutex);
2242 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2244 struct btrfs_fs_devices *fs_devices;
2246 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2249 * in case of fs with no seed, srcdev->fs_devices will point
2250 * to fs_devices of fs_info. However when the dev being replaced is
2251 * a seed dev it will point to the seed's local fs_devices. In short
2252 * srcdev will have its correct fs_devices in both the cases.
2254 fs_devices = srcdev->fs_devices;
2256 list_del_rcu(&srcdev->dev_list);
2257 list_del(&srcdev->dev_alloc_list);
2258 fs_devices->num_devices--;
2259 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2260 fs_devices->missing_devices--;
2262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2263 fs_devices->rw_devices--;
2266 fs_devices->open_devices--;
2269 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2271 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2272 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2274 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2275 /* zero out the old super if it is writable */
2276 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2279 btrfs_close_bdev(srcdev);
2281 btrfs_free_device(srcdev);
2283 /* if this is no devs we rather delete the fs_devices */
2284 if (!fs_devices->num_devices) {
2285 struct btrfs_fs_devices *tmp_fs_devices;
2288 * On a mounted FS, num_devices can't be zero unless it's a
2289 * seed. In case of a seed device being replaced, the replace
2290 * target added to the sprout FS, so there will be no more
2291 * device left under the seed FS.
2293 ASSERT(fs_devices->seeding);
2295 tmp_fs_devices = fs_info->fs_devices;
2296 while (tmp_fs_devices) {
2297 if (tmp_fs_devices->seed == fs_devices) {
2298 tmp_fs_devices->seed = fs_devices->seed;
2301 tmp_fs_devices = tmp_fs_devices->seed;
2303 fs_devices->seed = NULL;
2304 close_fs_devices(fs_devices);
2305 free_fs_devices(fs_devices);
2309 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2311 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2314 mutex_lock(&fs_devices->device_list_mutex);
2316 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2319 fs_devices->open_devices--;
2321 fs_devices->num_devices--;
2323 btrfs_assign_next_active_device(tgtdev, NULL);
2325 list_del_rcu(&tgtdev->dev_list);
2327 mutex_unlock(&fs_devices->device_list_mutex);
2330 * The update_dev_time() with in btrfs_scratch_superblocks()
2331 * may lead to a call to btrfs_show_devname() which will try
2332 * to hold device_list_mutex. And here this device
2333 * is already out of device list, so we don't have to hold
2334 * the device_list_mutex lock.
2336 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2338 btrfs_close_bdev(tgtdev);
2340 btrfs_free_device(tgtdev);
2343 static struct btrfs_device *btrfs_find_device_by_path(
2344 struct btrfs_fs_info *fs_info, const char *device_path)
2347 struct btrfs_super_block *disk_super;
2350 struct block_device *bdev;
2351 struct buffer_head *bh;
2352 struct btrfs_device *device;
2354 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2355 fs_info->bdev_holder, 0, &bdev, &bh);
2357 return ERR_PTR(ret);
2358 disk_super = (struct btrfs_super_block *)bh->b_data;
2359 devid = btrfs_stack_device_id(&disk_super->dev_item);
2360 dev_uuid = disk_super->dev_item.uuid;
2361 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2362 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2363 disk_super->metadata_uuid, true);
2365 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2366 disk_super->fsid, true);
2370 device = ERR_PTR(-ENOENT);
2371 blkdev_put(bdev, FMODE_READ);
2376 * Lookup a device given by device id, or the path if the id is 0.
2378 struct btrfs_device *btrfs_find_device_by_devspec(
2379 struct btrfs_fs_info *fs_info, u64 devid,
2380 const char *device_path)
2382 struct btrfs_device *device;
2385 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2388 return ERR_PTR(-ENOENT);
2392 if (!device_path || !device_path[0])
2393 return ERR_PTR(-EINVAL);
2395 if (strcmp(device_path, "missing") == 0) {
2396 /* Find first missing device */
2397 list_for_each_entry(device, &fs_info->fs_devices->devices,
2399 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2400 &device->dev_state) && !device->bdev)
2403 return ERR_PTR(-ENOENT);
2406 return btrfs_find_device_by_path(fs_info, device_path);
2410 * does all the dirty work required for changing file system's UUID.
2412 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2414 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2415 struct btrfs_fs_devices *old_devices;
2416 struct btrfs_fs_devices *seed_devices;
2417 struct btrfs_super_block *disk_super = fs_info->super_copy;
2418 struct btrfs_device *device;
2421 lockdep_assert_held(&uuid_mutex);
2422 if (!fs_devices->seeding)
2425 seed_devices = alloc_fs_devices(NULL, NULL);
2426 if (IS_ERR(seed_devices))
2427 return PTR_ERR(seed_devices);
2429 old_devices = clone_fs_devices(fs_devices);
2430 if (IS_ERR(old_devices)) {
2431 kfree(seed_devices);
2432 return PTR_ERR(old_devices);
2435 list_add(&old_devices->fs_list, &fs_uuids);
2437 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2438 seed_devices->opened = 1;
2439 INIT_LIST_HEAD(&seed_devices->devices);
2440 INIT_LIST_HEAD(&seed_devices->alloc_list);
2441 mutex_init(&seed_devices->device_list_mutex);
2443 mutex_lock(&fs_devices->device_list_mutex);
2444 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2446 list_for_each_entry(device, &seed_devices->devices, dev_list)
2447 device->fs_devices = seed_devices;
2449 mutex_lock(&fs_info->chunk_mutex);
2450 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2451 mutex_unlock(&fs_info->chunk_mutex);
2453 fs_devices->seeding = 0;
2454 fs_devices->num_devices = 0;
2455 fs_devices->open_devices = 0;
2456 fs_devices->missing_devices = 0;
2457 fs_devices->rotating = 0;
2458 fs_devices->seed = seed_devices;
2460 generate_random_uuid(fs_devices->fsid);
2461 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2462 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2463 mutex_unlock(&fs_devices->device_list_mutex);
2465 super_flags = btrfs_super_flags(disk_super) &
2466 ~BTRFS_SUPER_FLAG_SEEDING;
2467 btrfs_set_super_flags(disk_super, super_flags);
2473 * Store the expected generation for seed devices in device items.
2475 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2477 struct btrfs_fs_info *fs_info = trans->fs_info;
2478 struct btrfs_root *root = fs_info->chunk_root;
2479 struct btrfs_path *path;
2480 struct extent_buffer *leaf;
2481 struct btrfs_dev_item *dev_item;
2482 struct btrfs_device *device;
2483 struct btrfs_key key;
2484 u8 fs_uuid[BTRFS_FSID_SIZE];
2485 u8 dev_uuid[BTRFS_UUID_SIZE];
2489 path = btrfs_alloc_path();
2493 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2495 key.type = BTRFS_DEV_ITEM_KEY;
2498 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2502 leaf = path->nodes[0];
2504 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2505 ret = btrfs_next_leaf(root, path);
2510 leaf = path->nodes[0];
2511 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2512 btrfs_release_path(path);
2516 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2517 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2518 key.type != BTRFS_DEV_ITEM_KEY)
2521 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2522 struct btrfs_dev_item);
2523 devid = btrfs_device_id(leaf, dev_item);
2524 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2526 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2528 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2530 BUG_ON(!device); /* Logic error */
2532 if (device->fs_devices->seeding) {
2533 btrfs_set_device_generation(leaf, dev_item,
2534 device->generation);
2535 btrfs_mark_buffer_dirty(leaf);
2543 btrfs_free_path(path);
2547 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2549 struct btrfs_root *root = fs_info->dev_root;
2550 struct request_queue *q;
2551 struct btrfs_trans_handle *trans;
2552 struct btrfs_device *device;
2553 struct block_device *bdev;
2554 struct super_block *sb = fs_info->sb;
2555 struct rcu_string *name;
2556 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2557 u64 orig_super_total_bytes;
2558 u64 orig_super_num_devices;
2559 int seeding_dev = 0;
2561 bool unlocked = false;
2563 if (sb_rdonly(sb) && !fs_devices->seeding)
2566 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2567 fs_info->bdev_holder);
2569 return PTR_ERR(bdev);
2571 if (fs_devices->seeding) {
2573 down_write(&sb->s_umount);
2574 mutex_lock(&uuid_mutex);
2577 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2579 mutex_lock(&fs_devices->device_list_mutex);
2580 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2581 if (device->bdev == bdev) {
2584 &fs_devices->device_list_mutex);
2588 mutex_unlock(&fs_devices->device_list_mutex);
2590 device = btrfs_alloc_device(fs_info, NULL, NULL);
2591 if (IS_ERR(device)) {
2592 /* we can safely leave the fs_devices entry around */
2593 ret = PTR_ERR(device);
2597 name = rcu_string_strdup(device_path, GFP_KERNEL);
2600 goto error_free_device;
2602 rcu_assign_pointer(device->name, name);
2604 trans = btrfs_start_transaction(root, 0);
2605 if (IS_ERR(trans)) {
2606 ret = PTR_ERR(trans);
2607 goto error_free_device;
2610 q = bdev_get_queue(bdev);
2611 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2612 device->generation = trans->transid;
2613 device->io_width = fs_info->sectorsize;
2614 device->io_align = fs_info->sectorsize;
2615 device->sector_size = fs_info->sectorsize;
2616 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2617 fs_info->sectorsize);
2618 device->disk_total_bytes = device->total_bytes;
2619 device->commit_total_bytes = device->total_bytes;
2620 device->fs_info = fs_info;
2621 device->bdev = bdev;
2622 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2623 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2624 device->mode = FMODE_EXCL;
2625 device->dev_stats_valid = 1;
2626 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2629 sb->s_flags &= ~SB_RDONLY;
2630 ret = btrfs_prepare_sprout(fs_info);
2632 btrfs_abort_transaction(trans, ret);
2637 device->fs_devices = fs_devices;
2639 mutex_lock(&fs_devices->device_list_mutex);
2640 mutex_lock(&fs_info->chunk_mutex);
2641 list_add_rcu(&device->dev_list, &fs_devices->devices);
2642 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2643 fs_devices->num_devices++;
2644 fs_devices->open_devices++;
2645 fs_devices->rw_devices++;
2646 fs_devices->total_devices++;
2647 fs_devices->total_rw_bytes += device->total_bytes;
2649 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2651 if (!blk_queue_nonrot(q))
2652 fs_devices->rotating = 1;
2654 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2655 btrfs_set_super_total_bytes(fs_info->super_copy,
2656 round_down(orig_super_total_bytes + device->total_bytes,
2657 fs_info->sectorsize));
2659 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2660 btrfs_set_super_num_devices(fs_info->super_copy,
2661 orig_super_num_devices + 1);
2663 /* add sysfs device entry */
2664 btrfs_sysfs_add_device_link(fs_devices, device);
2667 * we've got more storage, clear any full flags on the space
2670 btrfs_clear_space_info_full(fs_info);
2672 mutex_unlock(&fs_info->chunk_mutex);
2673 mutex_unlock(&fs_devices->device_list_mutex);
2676 mutex_lock(&fs_info->chunk_mutex);
2677 ret = init_first_rw_device(trans);
2678 mutex_unlock(&fs_info->chunk_mutex);
2680 btrfs_abort_transaction(trans, ret);
2685 ret = btrfs_add_dev_item(trans, device);
2687 btrfs_abort_transaction(trans, ret);
2692 ret = btrfs_finish_sprout(trans);
2694 btrfs_abort_transaction(trans, ret);
2698 btrfs_sysfs_update_sprout_fsid(fs_devices,
2699 fs_info->fs_devices->fsid);
2702 ret = btrfs_commit_transaction(trans);
2705 mutex_unlock(&uuid_mutex);
2706 up_write(&sb->s_umount);
2709 if (ret) /* transaction commit */
2712 ret = btrfs_relocate_sys_chunks(fs_info);
2714 btrfs_handle_fs_error(fs_info, ret,
2715 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2716 trans = btrfs_attach_transaction(root);
2717 if (IS_ERR(trans)) {
2718 if (PTR_ERR(trans) == -ENOENT)
2720 ret = PTR_ERR(trans);
2724 ret = btrfs_commit_transaction(trans);
2727 /* Update ctime/mtime for libblkid */
2728 update_dev_time(device_path);
2732 btrfs_sysfs_rm_device_link(fs_devices, device);
2733 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2734 mutex_lock(&fs_info->chunk_mutex);
2735 list_del_rcu(&device->dev_list);
2736 list_del(&device->dev_alloc_list);
2737 fs_info->fs_devices->num_devices--;
2738 fs_info->fs_devices->open_devices--;
2739 fs_info->fs_devices->rw_devices--;
2740 fs_info->fs_devices->total_devices--;
2741 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2742 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2743 btrfs_set_super_total_bytes(fs_info->super_copy,
2744 orig_super_total_bytes);
2745 btrfs_set_super_num_devices(fs_info->super_copy,
2746 orig_super_num_devices);
2747 mutex_unlock(&fs_info->chunk_mutex);
2748 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2751 sb->s_flags |= SB_RDONLY;
2753 btrfs_end_transaction(trans);
2755 btrfs_free_device(device);
2757 blkdev_put(bdev, FMODE_EXCL);
2758 if (seeding_dev && !unlocked) {
2759 mutex_unlock(&uuid_mutex);
2760 up_write(&sb->s_umount);
2765 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2766 struct btrfs_device *device)
2769 struct btrfs_path *path;
2770 struct btrfs_root *root = device->fs_info->chunk_root;
2771 struct btrfs_dev_item *dev_item;
2772 struct extent_buffer *leaf;
2773 struct btrfs_key key;
2775 path = btrfs_alloc_path();
2779 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2780 key.type = BTRFS_DEV_ITEM_KEY;
2781 key.offset = device->devid;
2783 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2792 leaf = path->nodes[0];
2793 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2795 btrfs_set_device_id(leaf, dev_item, device->devid);
2796 btrfs_set_device_type(leaf, dev_item, device->type);
2797 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2798 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2799 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2800 btrfs_set_device_total_bytes(leaf, dev_item,
2801 btrfs_device_get_disk_total_bytes(device));
2802 btrfs_set_device_bytes_used(leaf, dev_item,
2803 btrfs_device_get_bytes_used(device));
2804 btrfs_mark_buffer_dirty(leaf);
2807 btrfs_free_path(path);
2811 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2812 struct btrfs_device *device, u64 new_size)
2814 struct btrfs_fs_info *fs_info = device->fs_info;
2815 struct btrfs_super_block *super_copy = fs_info->super_copy;
2819 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2822 new_size = round_down(new_size, fs_info->sectorsize);
2824 mutex_lock(&fs_info->chunk_mutex);
2825 old_total = btrfs_super_total_bytes(super_copy);
2826 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2828 if (new_size <= device->total_bytes ||
2829 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2830 mutex_unlock(&fs_info->chunk_mutex);
2834 btrfs_set_super_total_bytes(super_copy,
2835 round_down(old_total + diff, fs_info->sectorsize));
2836 device->fs_devices->total_rw_bytes += diff;
2838 btrfs_device_set_total_bytes(device, new_size);
2839 btrfs_device_set_disk_total_bytes(device, new_size);
2840 btrfs_clear_space_info_full(device->fs_info);
2841 if (list_empty(&device->post_commit_list))
2842 list_add_tail(&device->post_commit_list,
2843 &trans->transaction->dev_update_list);
2844 mutex_unlock(&fs_info->chunk_mutex);
2846 return btrfs_update_device(trans, device);
2849 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2851 struct btrfs_fs_info *fs_info = trans->fs_info;
2852 struct btrfs_root *root = fs_info->chunk_root;
2854 struct btrfs_path *path;
2855 struct btrfs_key key;
2857 path = btrfs_alloc_path();
2861 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2862 key.offset = chunk_offset;
2863 key.type = BTRFS_CHUNK_ITEM_KEY;
2865 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2868 else if (ret > 0) { /* Logic error or corruption */
2869 btrfs_handle_fs_error(fs_info, -ENOENT,
2870 "Failed lookup while freeing chunk.");
2875 ret = btrfs_del_item(trans, root, path);
2877 btrfs_handle_fs_error(fs_info, ret,
2878 "Failed to delete chunk item.");
2880 btrfs_free_path(path);
2884 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2886 struct btrfs_super_block *super_copy = fs_info->super_copy;
2887 struct btrfs_disk_key *disk_key;
2888 struct btrfs_chunk *chunk;
2895 struct btrfs_key key;
2897 mutex_lock(&fs_info->chunk_mutex);
2898 array_size = btrfs_super_sys_array_size(super_copy);
2900 ptr = super_copy->sys_chunk_array;
2903 while (cur < array_size) {
2904 disk_key = (struct btrfs_disk_key *)ptr;
2905 btrfs_disk_key_to_cpu(&key, disk_key);
2907 len = sizeof(*disk_key);
2909 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2910 chunk = (struct btrfs_chunk *)(ptr + len);
2911 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2912 len += btrfs_chunk_item_size(num_stripes);
2917 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2918 key.offset == chunk_offset) {
2919 memmove(ptr, ptr + len, array_size - (cur + len));
2921 btrfs_set_super_sys_array_size(super_copy, array_size);
2927 mutex_unlock(&fs_info->chunk_mutex);
2932 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2933 * @logical: Logical block offset in bytes.
2934 * @length: Length of extent in bytes.
2936 * Return: Chunk mapping or ERR_PTR.
2938 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2939 u64 logical, u64 length)
2941 struct extent_map_tree *em_tree;
2942 struct extent_map *em;
2944 em_tree = &fs_info->mapping_tree;
2945 read_lock(&em_tree->lock);
2946 em = lookup_extent_mapping(em_tree, logical, length);
2947 read_unlock(&em_tree->lock);
2950 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2952 return ERR_PTR(-EINVAL);
2955 if (em->start > logical || em->start + em->len < logical) {
2957 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2958 logical, length, em->start, em->start + em->len);
2959 free_extent_map(em);
2960 return ERR_PTR(-EINVAL);
2963 /* callers are responsible for dropping em's ref. */
2967 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2969 struct btrfs_fs_info *fs_info = trans->fs_info;
2970 struct extent_map *em;
2971 struct map_lookup *map;
2972 u64 dev_extent_len = 0;
2974 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2976 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2979 * This is a logic error, but we don't want to just rely on the
2980 * user having built with ASSERT enabled, so if ASSERT doesn't
2981 * do anything we still error out.
2986 map = em->map_lookup;
2987 mutex_lock(&fs_info->chunk_mutex);
2988 check_system_chunk(trans, map->type);
2989 mutex_unlock(&fs_info->chunk_mutex);
2992 * Take the device list mutex to prevent races with the final phase of
2993 * a device replace operation that replaces the device object associated
2994 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2996 mutex_lock(&fs_devices->device_list_mutex);
2997 for (i = 0; i < map->num_stripes; i++) {
2998 struct btrfs_device *device = map->stripes[i].dev;
2999 ret = btrfs_free_dev_extent(trans, device,
3000 map->stripes[i].physical,
3003 mutex_unlock(&fs_devices->device_list_mutex);
3004 btrfs_abort_transaction(trans, ret);
3008 if (device->bytes_used > 0) {
3009 mutex_lock(&fs_info->chunk_mutex);
3010 btrfs_device_set_bytes_used(device,
3011 device->bytes_used - dev_extent_len);
3012 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3013 btrfs_clear_space_info_full(fs_info);
3014 mutex_unlock(&fs_info->chunk_mutex);
3017 ret = btrfs_update_device(trans, device);
3019 mutex_unlock(&fs_devices->device_list_mutex);
3020 btrfs_abort_transaction(trans, ret);
3024 mutex_unlock(&fs_devices->device_list_mutex);
3026 ret = btrfs_free_chunk(trans, chunk_offset);
3028 btrfs_abort_transaction(trans, ret);
3032 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3034 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3035 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3037 btrfs_abort_transaction(trans, ret);
3042 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3044 btrfs_abort_transaction(trans, ret);
3050 free_extent_map(em);
3054 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3056 struct btrfs_root *root = fs_info->chunk_root;
3057 struct btrfs_trans_handle *trans;
3061 * Prevent races with automatic removal of unused block groups.
3062 * After we relocate and before we remove the chunk with offset
3063 * chunk_offset, automatic removal of the block group can kick in,
3064 * resulting in a failure when calling btrfs_remove_chunk() below.
3066 * Make sure to acquire this mutex before doing a tree search (dev
3067 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3068 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3069 * we release the path used to search the chunk/dev tree and before
3070 * the current task acquires this mutex and calls us.
3072 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3074 /* step one, relocate all the extents inside this chunk */
3075 btrfs_scrub_pause(fs_info);
3076 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3077 btrfs_scrub_continue(fs_info);
3081 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3083 if (IS_ERR(trans)) {
3084 ret = PTR_ERR(trans);
3085 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3090 * step two, delete the device extents and the
3091 * chunk tree entries
3093 ret = btrfs_remove_chunk(trans, chunk_offset);
3094 btrfs_end_transaction(trans);
3098 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3100 struct btrfs_root *chunk_root = fs_info->chunk_root;
3101 struct btrfs_path *path;
3102 struct extent_buffer *leaf;
3103 struct btrfs_chunk *chunk;
3104 struct btrfs_key key;
3105 struct btrfs_key found_key;
3107 bool retried = false;
3111 path = btrfs_alloc_path();
3116 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3117 key.offset = (u64)-1;
3118 key.type = BTRFS_CHUNK_ITEM_KEY;
3121 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3122 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3124 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3127 BUG_ON(ret == 0); /* Corruption */
3129 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3132 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3138 leaf = path->nodes[0];
3139 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3141 chunk = btrfs_item_ptr(leaf, path->slots[0],
3142 struct btrfs_chunk);
3143 chunk_type = btrfs_chunk_type(leaf, chunk);
3144 btrfs_release_path(path);
3146 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3147 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3153 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3155 if (found_key.offset == 0)
3157 key.offset = found_key.offset - 1;
3160 if (failed && !retried) {
3164 } else if (WARN_ON(failed && retried)) {
3168 btrfs_free_path(path);
3173 * return 1 : allocate a data chunk successfully,
3174 * return <0: errors during allocating a data chunk,
3175 * return 0 : no need to allocate a data chunk.
3177 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3180 struct btrfs_block_group_cache *cache;
3184 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3186 chunk_type = cache->flags;
3187 btrfs_put_block_group(cache);
3189 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3190 spin_lock(&fs_info->data_sinfo->lock);
3191 bytes_used = fs_info->data_sinfo->bytes_used;
3192 spin_unlock(&fs_info->data_sinfo->lock);
3195 struct btrfs_trans_handle *trans;
3198 trans = btrfs_join_transaction(fs_info->tree_root);
3200 return PTR_ERR(trans);
3202 ret = btrfs_force_chunk_alloc(trans,
3203 BTRFS_BLOCK_GROUP_DATA);
3204 btrfs_end_transaction(trans);
3213 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3214 struct btrfs_balance_control *bctl)
3216 struct btrfs_root *root = fs_info->tree_root;
3217 struct btrfs_trans_handle *trans;
3218 struct btrfs_balance_item *item;
3219 struct btrfs_disk_balance_args disk_bargs;
3220 struct btrfs_path *path;
3221 struct extent_buffer *leaf;
3222 struct btrfs_key key;
3225 path = btrfs_alloc_path();
3229 trans = btrfs_start_transaction(root, 0);
3230 if (IS_ERR(trans)) {
3231 btrfs_free_path(path);
3232 return PTR_ERR(trans);
3235 key.objectid = BTRFS_BALANCE_OBJECTID;
3236 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3239 ret = btrfs_insert_empty_item(trans, root, path, &key,
3244 leaf = path->nodes[0];
3245 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3247 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3249 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3250 btrfs_set_balance_data(leaf, item, &disk_bargs);
3251 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3252 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3253 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3254 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3256 btrfs_set_balance_flags(leaf, item, bctl->flags);
3258 btrfs_mark_buffer_dirty(leaf);
3260 btrfs_free_path(path);
3261 err = btrfs_commit_transaction(trans);
3267 static int del_balance_item(struct btrfs_fs_info *fs_info)
3269 struct btrfs_root *root = fs_info->tree_root;
3270 struct btrfs_trans_handle *trans;
3271 struct btrfs_path *path;
3272 struct btrfs_key key;
3275 path = btrfs_alloc_path();
3279 trans = btrfs_start_transaction(root, 0);
3280 if (IS_ERR(trans)) {
3281 btrfs_free_path(path);
3282 return PTR_ERR(trans);
3285 key.objectid = BTRFS_BALANCE_OBJECTID;
3286 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3289 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3297 ret = btrfs_del_item(trans, root, path);
3299 btrfs_free_path(path);
3300 err = btrfs_commit_transaction(trans);
3307 * This is a heuristic used to reduce the number of chunks balanced on
3308 * resume after balance was interrupted.
3310 static void update_balance_args(struct btrfs_balance_control *bctl)
3313 * Turn on soft mode for chunk types that were being converted.
3315 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3316 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3317 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3318 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3319 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3320 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3323 * Turn on usage filter if is not already used. The idea is
3324 * that chunks that we have already balanced should be
3325 * reasonably full. Don't do it for chunks that are being
3326 * converted - that will keep us from relocating unconverted
3327 * (albeit full) chunks.
3329 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3330 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3331 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3332 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3333 bctl->data.usage = 90;
3335 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3336 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3337 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3338 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3339 bctl->sys.usage = 90;
3341 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3342 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3343 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3344 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3345 bctl->meta.usage = 90;
3350 * Clear the balance status in fs_info and delete the balance item from disk.
3352 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3354 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3357 BUG_ON(!fs_info->balance_ctl);
3359 spin_lock(&fs_info->balance_lock);
3360 fs_info->balance_ctl = NULL;
3361 spin_unlock(&fs_info->balance_lock);
3364 ret = del_balance_item(fs_info);
3366 btrfs_handle_fs_error(fs_info, ret, NULL);
3370 * Balance filters. Return 1 if chunk should be filtered out
3371 * (should not be balanced).
3373 static int chunk_profiles_filter(u64 chunk_type,
3374 struct btrfs_balance_args *bargs)
3376 chunk_type = chunk_to_extended(chunk_type) &
3377 BTRFS_EXTENDED_PROFILE_MASK;
3379 if (bargs->profiles & chunk_type)
3385 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3386 struct btrfs_balance_args *bargs)
3388 struct btrfs_block_group_cache *cache;
3390 u64 user_thresh_min;
3391 u64 user_thresh_max;
3394 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3395 chunk_used = btrfs_block_group_used(&cache->item);
3397 if (bargs->usage_min == 0)
3398 user_thresh_min = 0;
3400 user_thresh_min = div_factor_fine(cache->key.offset,
3403 if (bargs->usage_max == 0)
3404 user_thresh_max = 1;
3405 else if (bargs->usage_max > 100)
3406 user_thresh_max = cache->key.offset;
3408 user_thresh_max = div_factor_fine(cache->key.offset,
3411 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3414 btrfs_put_block_group(cache);
3418 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3419 u64 chunk_offset, struct btrfs_balance_args *bargs)
3421 struct btrfs_block_group_cache *cache;
3422 u64 chunk_used, user_thresh;
3425 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3426 chunk_used = btrfs_block_group_used(&cache->item);
3428 if (bargs->usage_min == 0)
3430 else if (bargs->usage > 100)
3431 user_thresh = cache->key.offset;
3433 user_thresh = div_factor_fine(cache->key.offset,
3436 if (chunk_used < user_thresh)
3439 btrfs_put_block_group(cache);
3443 static int chunk_devid_filter(struct extent_buffer *leaf,
3444 struct btrfs_chunk *chunk,
3445 struct btrfs_balance_args *bargs)
3447 struct btrfs_stripe *stripe;
3448 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3451 for (i = 0; i < num_stripes; i++) {
3452 stripe = btrfs_stripe_nr(chunk, i);
3453 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3460 static u64 calc_data_stripes(u64 type, int num_stripes)
3462 const int index = btrfs_bg_flags_to_raid_index(type);
3463 const int ncopies = btrfs_raid_array[index].ncopies;
3464 const int nparity = btrfs_raid_array[index].nparity;
3467 return num_stripes - nparity;
3469 return num_stripes / ncopies;
3472 /* [pstart, pend) */
3473 static int chunk_drange_filter(struct extent_buffer *leaf,
3474 struct btrfs_chunk *chunk,
3475 struct btrfs_balance_args *bargs)
3477 struct btrfs_stripe *stripe;
3478 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3485 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3488 type = btrfs_chunk_type(leaf, chunk);
3489 factor = calc_data_stripes(type, num_stripes);
3491 for (i = 0; i < num_stripes; i++) {
3492 stripe = btrfs_stripe_nr(chunk, i);
3493 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3496 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3497 stripe_length = btrfs_chunk_length(leaf, chunk);
3498 stripe_length = div_u64(stripe_length, factor);
3500 if (stripe_offset < bargs->pend &&
3501 stripe_offset + stripe_length > bargs->pstart)
3508 /* [vstart, vend) */
3509 static int chunk_vrange_filter(struct extent_buffer *leaf,
3510 struct btrfs_chunk *chunk,
3512 struct btrfs_balance_args *bargs)
3514 if (chunk_offset < bargs->vend &&
3515 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3516 /* at least part of the chunk is inside this vrange */
3522 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3523 struct btrfs_chunk *chunk,
3524 struct btrfs_balance_args *bargs)
3526 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3528 if (bargs->stripes_min <= num_stripes
3529 && num_stripes <= bargs->stripes_max)
3535 static int chunk_soft_convert_filter(u64 chunk_type,
3536 struct btrfs_balance_args *bargs)
3538 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3541 chunk_type = chunk_to_extended(chunk_type) &
3542 BTRFS_EXTENDED_PROFILE_MASK;
3544 if (bargs->target == chunk_type)
3550 static int should_balance_chunk(struct extent_buffer *leaf,
3551 struct btrfs_chunk *chunk, u64 chunk_offset)
3553 struct btrfs_fs_info *fs_info = leaf->fs_info;
3554 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3555 struct btrfs_balance_args *bargs = NULL;
3556 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3559 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3560 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3564 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3565 bargs = &bctl->data;
3566 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3568 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3569 bargs = &bctl->meta;
3571 /* profiles filter */
3572 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3573 chunk_profiles_filter(chunk_type, bargs)) {
3578 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3579 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3581 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3582 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3587 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3588 chunk_devid_filter(leaf, chunk, bargs)) {
3592 /* drange filter, makes sense only with devid filter */
3593 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3594 chunk_drange_filter(leaf, chunk, bargs)) {
3599 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3600 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3604 /* stripes filter */
3605 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3606 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3610 /* soft profile changing mode */
3611 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3612 chunk_soft_convert_filter(chunk_type, bargs)) {
3617 * limited by count, must be the last filter
3619 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3620 if (bargs->limit == 0)
3624 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3626 * Same logic as the 'limit' filter; the minimum cannot be
3627 * determined here because we do not have the global information
3628 * about the count of all chunks that satisfy the filters.
3630 if (bargs->limit_max == 0)
3639 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3641 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3642 struct btrfs_root *chunk_root = fs_info->chunk_root;
3644 struct btrfs_chunk *chunk;
3645 struct btrfs_path *path = NULL;
3646 struct btrfs_key key;
3647 struct btrfs_key found_key;
3648 struct extent_buffer *leaf;
3651 int enospc_errors = 0;
3652 bool counting = true;
3653 /* The single value limit and min/max limits use the same bytes in the */
3654 u64 limit_data = bctl->data.limit;
3655 u64 limit_meta = bctl->meta.limit;
3656 u64 limit_sys = bctl->sys.limit;
3660 int chunk_reserved = 0;
3662 path = btrfs_alloc_path();
3668 /* zero out stat counters */
3669 spin_lock(&fs_info->balance_lock);
3670 memset(&bctl->stat, 0, sizeof(bctl->stat));
3671 spin_unlock(&fs_info->balance_lock);
3675 * The single value limit and min/max limits use the same bytes
3678 bctl->data.limit = limit_data;
3679 bctl->meta.limit = limit_meta;
3680 bctl->sys.limit = limit_sys;
3682 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3683 key.offset = (u64)-1;
3684 key.type = BTRFS_CHUNK_ITEM_KEY;
3687 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3688 atomic_read(&fs_info->balance_cancel_req)) {
3693 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3694 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3696 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3701 * this shouldn't happen, it means the last relocate
3705 BUG(); /* FIXME break ? */
3707 ret = btrfs_previous_item(chunk_root, path, 0,
3708 BTRFS_CHUNK_ITEM_KEY);
3710 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3715 leaf = path->nodes[0];
3716 slot = path->slots[0];
3717 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3719 if (found_key.objectid != key.objectid) {
3720 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3724 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3725 chunk_type = btrfs_chunk_type(leaf, chunk);
3728 spin_lock(&fs_info->balance_lock);
3729 bctl->stat.considered++;
3730 spin_unlock(&fs_info->balance_lock);
3733 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3735 btrfs_release_path(path);
3737 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3742 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3743 spin_lock(&fs_info->balance_lock);
3744 bctl->stat.expected++;
3745 spin_unlock(&fs_info->balance_lock);
3747 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3749 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3751 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3758 * Apply limit_min filter, no need to check if the LIMITS
3759 * filter is used, limit_min is 0 by default
3761 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3762 count_data < bctl->data.limit_min)
3763 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3764 count_meta < bctl->meta.limit_min)
3765 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3766 count_sys < bctl->sys.limit_min)) {
3767 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3771 if (!chunk_reserved) {
3773 * We may be relocating the only data chunk we have,
3774 * which could potentially end up with losing data's
3775 * raid profile, so lets allocate an empty one in
3778 ret = btrfs_may_alloc_data_chunk(fs_info,
3781 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3783 } else if (ret == 1) {
3788 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3789 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3790 if (ret == -ENOSPC) {
3792 } else if (ret == -ETXTBSY) {
3794 "skipping relocation of block group %llu due to active swapfile",
3800 spin_lock(&fs_info->balance_lock);
3801 bctl->stat.completed++;
3802 spin_unlock(&fs_info->balance_lock);
3805 if (found_key.offset == 0)
3807 key.offset = found_key.offset - 1;
3811 btrfs_release_path(path);
3816 btrfs_free_path(path);
3817 if (enospc_errors) {
3818 btrfs_info(fs_info, "%d enospc errors during balance",
3828 * alloc_profile_is_valid - see if a given profile is valid and reduced
3829 * @flags: profile to validate
3830 * @extended: if true @flags is treated as an extended profile
3832 static int alloc_profile_is_valid(u64 flags, int extended)
3834 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3835 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3837 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3839 /* 1) check that all other bits are zeroed */
3843 /* 2) see if profile is reduced */
3845 return !extended; /* "0" is valid for usual profiles */
3847 /* true if exactly one bit set */
3848 return is_power_of_2(flags);
3851 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3853 /* cancel requested || normal exit path */
3854 return atomic_read(&fs_info->balance_cancel_req) ||
3855 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3856 atomic_read(&fs_info->balance_cancel_req) == 0);
3859 /* Non-zero return value signifies invalidity */
3860 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3863 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3864 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3865 (bctl_arg->target & ~allowed)));
3869 * Fill @buf with textual description of balance filter flags @bargs, up to
3870 * @size_buf including the terminating null. The output may be trimmed if it
3871 * does not fit into the provided buffer.
3873 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3877 u32 size_bp = size_buf;
3879 u64 flags = bargs->flags;
3880 char tmp_buf[128] = {'\0'};
3885 #define CHECK_APPEND_NOARG(a) \
3887 ret = snprintf(bp, size_bp, (a)); \
3888 if (ret < 0 || ret >= size_bp) \
3889 goto out_overflow; \
3894 #define CHECK_APPEND_1ARG(a, v1) \
3896 ret = snprintf(bp, size_bp, (a), (v1)); \
3897 if (ret < 0 || ret >= size_bp) \
3898 goto out_overflow; \
3903 #define CHECK_APPEND_2ARG(a, v1, v2) \
3905 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3906 if (ret < 0 || ret >= size_bp) \
3907 goto out_overflow; \
3912 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3913 CHECK_APPEND_1ARG("convert=%s,",
3914 btrfs_bg_type_to_raid_name(bargs->target));
3916 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3917 CHECK_APPEND_NOARG("soft,");
3919 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3920 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3922 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3925 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3926 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3928 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3929 CHECK_APPEND_2ARG("usage=%u..%u,",
3930 bargs->usage_min, bargs->usage_max);
3932 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3933 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3935 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3936 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3937 bargs->pstart, bargs->pend);
3939 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3940 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3941 bargs->vstart, bargs->vend);
3943 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3944 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3946 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3947 CHECK_APPEND_2ARG("limit=%u..%u,",
3948 bargs->limit_min, bargs->limit_max);
3950 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3951 CHECK_APPEND_2ARG("stripes=%u..%u,",
3952 bargs->stripes_min, bargs->stripes_max);
3954 #undef CHECK_APPEND_2ARG
3955 #undef CHECK_APPEND_1ARG
3956 #undef CHECK_APPEND_NOARG
3960 if (size_bp < size_buf)
3961 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3966 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3968 u32 size_buf = 1024;
3969 char tmp_buf[192] = {'\0'};
3972 u32 size_bp = size_buf;
3974 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3976 buf = kzalloc(size_buf, GFP_KERNEL);
3982 #define CHECK_APPEND_1ARG(a, v1) \
3984 ret = snprintf(bp, size_bp, (a), (v1)); \
3985 if (ret < 0 || ret >= size_bp) \
3986 goto out_overflow; \
3991 if (bctl->flags & BTRFS_BALANCE_FORCE)
3992 CHECK_APPEND_1ARG("%s", "-f ");
3994 if (bctl->flags & BTRFS_BALANCE_DATA) {
3995 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3996 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3999 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4000 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4001 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4004 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4005 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4006 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4009 #undef CHECK_APPEND_1ARG
4013 if (size_bp < size_buf)
4014 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4015 btrfs_info(fs_info, "balance: %s %s",
4016 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4017 "resume" : "start", buf);
4023 * Should be called with balance mutexe held
4025 int btrfs_balance(struct btrfs_fs_info *fs_info,
4026 struct btrfs_balance_control *bctl,
4027 struct btrfs_ioctl_balance_args *bargs)
4029 u64 meta_target, data_target;
4035 bool reducing_integrity;
4038 if (btrfs_fs_closing(fs_info) ||
4039 atomic_read(&fs_info->balance_pause_req) ||
4040 atomic_read(&fs_info->balance_cancel_req)) {
4045 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4046 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4050 * In case of mixed groups both data and meta should be picked,
4051 * and identical options should be given for both of them.
4053 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4054 if (mixed && (bctl->flags & allowed)) {
4055 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4056 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4057 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4059 "balance: mixed groups data and metadata options must be the same");
4065 num_devices = btrfs_num_devices(fs_info);
4067 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4068 if (num_devices >= btrfs_raid_array[i].devs_min)
4069 allowed |= btrfs_raid_array[i].bg_flag;
4071 if (validate_convert_profile(&bctl->data, allowed)) {
4073 "balance: invalid convert data profile %s",
4074 btrfs_bg_type_to_raid_name(bctl->data.target));
4078 if (validate_convert_profile(&bctl->meta, allowed)) {
4080 "balance: invalid convert metadata profile %s",
4081 btrfs_bg_type_to_raid_name(bctl->meta.target));
4085 if (validate_convert_profile(&bctl->sys, allowed)) {
4087 "balance: invalid convert system profile %s",
4088 btrfs_bg_type_to_raid_name(bctl->sys.target));
4094 * Allow to reduce metadata or system integrity only if force set for
4095 * profiles with redundancy (copies, parity)
4098 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4099 if (btrfs_raid_array[i].ncopies >= 2 ||
4100 btrfs_raid_array[i].tolerated_failures >= 1)
4101 allowed |= btrfs_raid_array[i].bg_flag;
4104 seq = read_seqbegin(&fs_info->profiles_lock);
4106 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4107 (fs_info->avail_system_alloc_bits & allowed) &&
4108 !(bctl->sys.target & allowed)) ||
4109 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4110 (fs_info->avail_metadata_alloc_bits & allowed) &&
4111 !(bctl->meta.target & allowed)))
4112 reducing_integrity = true;
4114 reducing_integrity = false;
4116 /* if we're not converting, the target field is uninitialized */
4117 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4118 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4119 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4120 bctl->data.target : fs_info->avail_data_alloc_bits;
4121 } while (read_seqretry(&fs_info->profiles_lock, seq));
4123 if (reducing_integrity) {
4124 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4126 "balance: force reducing metadata integrity");
4129 "balance: reduces metadata integrity, use --force if you want this");
4135 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4136 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4138 "balance: metadata profile %s has lower redundancy than data profile %s",
4139 btrfs_bg_type_to_raid_name(meta_target),
4140 btrfs_bg_type_to_raid_name(data_target));
4143 if (fs_info->send_in_progress) {
4144 btrfs_warn_rl(fs_info,
4145 "cannot run balance while send operations are in progress (%d in progress)",
4146 fs_info->send_in_progress);
4151 ret = insert_balance_item(fs_info, bctl);
4152 if (ret && ret != -EEXIST)
4155 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4156 BUG_ON(ret == -EEXIST);
4157 BUG_ON(fs_info->balance_ctl);
4158 spin_lock(&fs_info->balance_lock);
4159 fs_info->balance_ctl = bctl;
4160 spin_unlock(&fs_info->balance_lock);
4162 BUG_ON(ret != -EEXIST);
4163 spin_lock(&fs_info->balance_lock);
4164 update_balance_args(bctl);
4165 spin_unlock(&fs_info->balance_lock);
4168 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4169 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4170 describe_balance_start_or_resume(fs_info);
4171 mutex_unlock(&fs_info->balance_mutex);
4173 ret = __btrfs_balance(fs_info);
4175 mutex_lock(&fs_info->balance_mutex);
4176 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4177 btrfs_info(fs_info, "balance: paused");
4178 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4179 btrfs_info(fs_info, "balance: canceled");
4181 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4183 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4186 memset(bargs, 0, sizeof(*bargs));
4187 btrfs_update_ioctl_balance_args(fs_info, bargs);
4190 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4191 balance_need_close(fs_info)) {
4192 reset_balance_state(fs_info);
4193 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4196 wake_up(&fs_info->balance_wait_q);
4200 if (bctl->flags & BTRFS_BALANCE_RESUME)
4201 reset_balance_state(fs_info);
4204 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4209 static int balance_kthread(void *data)
4211 struct btrfs_fs_info *fs_info = data;
4214 mutex_lock(&fs_info->balance_mutex);
4215 if (fs_info->balance_ctl)
4216 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4217 mutex_unlock(&fs_info->balance_mutex);
4222 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4224 struct task_struct *tsk;
4226 mutex_lock(&fs_info->balance_mutex);
4227 if (!fs_info->balance_ctl) {
4228 mutex_unlock(&fs_info->balance_mutex);
4231 mutex_unlock(&fs_info->balance_mutex);
4233 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4234 btrfs_info(fs_info, "balance: resume skipped");
4239 * A ro->rw remount sequence should continue with the paused balance
4240 * regardless of who pauses it, system or the user as of now, so set
4243 spin_lock(&fs_info->balance_lock);
4244 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4245 spin_unlock(&fs_info->balance_lock);
4247 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4248 return PTR_ERR_OR_ZERO(tsk);
4251 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4253 struct btrfs_balance_control *bctl;
4254 struct btrfs_balance_item *item;
4255 struct btrfs_disk_balance_args disk_bargs;
4256 struct btrfs_path *path;
4257 struct extent_buffer *leaf;
4258 struct btrfs_key key;
4261 path = btrfs_alloc_path();
4265 key.objectid = BTRFS_BALANCE_OBJECTID;
4266 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4269 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4272 if (ret > 0) { /* ret = -ENOENT; */
4277 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4283 leaf = path->nodes[0];
4284 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4286 bctl->flags = btrfs_balance_flags(leaf, item);
4287 bctl->flags |= BTRFS_BALANCE_RESUME;
4289 btrfs_balance_data(leaf, item, &disk_bargs);
4290 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4291 btrfs_balance_meta(leaf, item, &disk_bargs);
4292 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4293 btrfs_balance_sys(leaf, item, &disk_bargs);
4294 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4297 * This should never happen, as the paused balance state is recovered
4298 * during mount without any chance of other exclusive ops to collide.
4300 * This gives the exclusive op status to balance and keeps in paused
4301 * state until user intervention (cancel or umount). If the ownership
4302 * cannot be assigned, show a message but do not fail. The balance
4303 * is in a paused state and must have fs_info::balance_ctl properly
4306 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4308 "balance: cannot set exclusive op status, resume manually");
4310 mutex_lock(&fs_info->balance_mutex);
4311 BUG_ON(fs_info->balance_ctl);
4312 spin_lock(&fs_info->balance_lock);
4313 fs_info->balance_ctl = bctl;
4314 spin_unlock(&fs_info->balance_lock);
4315 mutex_unlock(&fs_info->balance_mutex);
4317 btrfs_free_path(path);
4321 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4325 mutex_lock(&fs_info->balance_mutex);
4326 if (!fs_info->balance_ctl) {
4327 mutex_unlock(&fs_info->balance_mutex);
4331 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4332 atomic_inc(&fs_info->balance_pause_req);
4333 mutex_unlock(&fs_info->balance_mutex);
4335 wait_event(fs_info->balance_wait_q,
4336 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4338 mutex_lock(&fs_info->balance_mutex);
4339 /* we are good with balance_ctl ripped off from under us */
4340 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4341 atomic_dec(&fs_info->balance_pause_req);
4346 mutex_unlock(&fs_info->balance_mutex);
4350 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4352 mutex_lock(&fs_info->balance_mutex);
4353 if (!fs_info->balance_ctl) {
4354 mutex_unlock(&fs_info->balance_mutex);
4359 * A paused balance with the item stored on disk can be resumed at
4360 * mount time if the mount is read-write. Otherwise it's still paused
4361 * and we must not allow cancelling as it deletes the item.
4363 if (sb_rdonly(fs_info->sb)) {
4364 mutex_unlock(&fs_info->balance_mutex);
4368 atomic_inc(&fs_info->balance_cancel_req);
4370 * if we are running just wait and return, balance item is
4371 * deleted in btrfs_balance in this case
4373 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4374 mutex_unlock(&fs_info->balance_mutex);
4375 wait_event(fs_info->balance_wait_q,
4376 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4377 mutex_lock(&fs_info->balance_mutex);
4379 mutex_unlock(&fs_info->balance_mutex);
4381 * Lock released to allow other waiters to continue, we'll
4382 * reexamine the status again.
4384 mutex_lock(&fs_info->balance_mutex);
4386 if (fs_info->balance_ctl) {
4387 reset_balance_state(fs_info);
4388 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4389 btrfs_info(fs_info, "balance: canceled");
4393 BUG_ON(fs_info->balance_ctl ||
4394 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4395 atomic_dec(&fs_info->balance_cancel_req);
4396 mutex_unlock(&fs_info->balance_mutex);
4400 static int btrfs_uuid_scan_kthread(void *data)
4402 struct btrfs_fs_info *fs_info = data;
4403 struct btrfs_root *root = fs_info->tree_root;
4404 struct btrfs_key key;
4405 struct btrfs_path *path = NULL;
4407 struct extent_buffer *eb;
4409 struct btrfs_root_item root_item;
4411 struct btrfs_trans_handle *trans = NULL;
4413 path = btrfs_alloc_path();
4420 key.type = BTRFS_ROOT_ITEM_KEY;
4424 ret = btrfs_search_forward(root, &key, path,
4425 BTRFS_OLDEST_GENERATION);
4432 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4433 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4434 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4435 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4438 eb = path->nodes[0];
4439 slot = path->slots[0];
4440 item_size = btrfs_item_size_nr(eb, slot);
4441 if (item_size < sizeof(root_item))
4444 read_extent_buffer(eb, &root_item,
4445 btrfs_item_ptr_offset(eb, slot),
4446 (int)sizeof(root_item));
4447 if (btrfs_root_refs(&root_item) == 0)
4450 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4451 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4455 btrfs_release_path(path);
4457 * 1 - subvol uuid item
4458 * 1 - received_subvol uuid item
4460 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4461 if (IS_ERR(trans)) {
4462 ret = PTR_ERR(trans);
4470 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4471 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4472 BTRFS_UUID_KEY_SUBVOL,
4475 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4481 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4482 ret = btrfs_uuid_tree_add(trans,
4483 root_item.received_uuid,
4484 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4487 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4495 ret = btrfs_end_transaction(trans);
4501 btrfs_release_path(path);
4502 if (key.offset < (u64)-1) {
4504 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4506 key.type = BTRFS_ROOT_ITEM_KEY;
4507 } else if (key.objectid < (u64)-1) {
4509 key.type = BTRFS_ROOT_ITEM_KEY;
4518 btrfs_free_path(path);
4519 if (trans && !IS_ERR(trans))
4520 btrfs_end_transaction(trans);
4522 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4524 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4525 up(&fs_info->uuid_tree_rescan_sem);
4530 * Callback for btrfs_uuid_tree_iterate().
4532 * 0 check succeeded, the entry is not outdated.
4533 * < 0 if an error occurred.
4534 * > 0 if the check failed, which means the caller shall remove the entry.
4536 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4537 u8 *uuid, u8 type, u64 subid)
4539 struct btrfs_key key;
4541 struct btrfs_root *subvol_root;
4543 if (type != BTRFS_UUID_KEY_SUBVOL &&
4544 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4547 key.objectid = subid;
4548 key.type = BTRFS_ROOT_ITEM_KEY;
4549 key.offset = (u64)-1;
4550 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4551 if (IS_ERR(subvol_root)) {
4552 ret = PTR_ERR(subvol_root);
4559 case BTRFS_UUID_KEY_SUBVOL:
4560 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4563 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4564 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4574 static int btrfs_uuid_rescan_kthread(void *data)
4576 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4580 * 1st step is to iterate through the existing UUID tree and
4581 * to delete all entries that contain outdated data.
4582 * 2nd step is to add all missing entries to the UUID tree.
4584 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4586 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4587 up(&fs_info->uuid_tree_rescan_sem);
4590 return btrfs_uuid_scan_kthread(data);
4593 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4595 struct btrfs_trans_handle *trans;
4596 struct btrfs_root *tree_root = fs_info->tree_root;
4597 struct btrfs_root *uuid_root;
4598 struct task_struct *task;
4605 trans = btrfs_start_transaction(tree_root, 2);
4607 return PTR_ERR(trans);
4609 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4610 if (IS_ERR(uuid_root)) {
4611 ret = PTR_ERR(uuid_root);
4612 btrfs_abort_transaction(trans, ret);
4613 btrfs_end_transaction(trans);
4617 fs_info->uuid_root = uuid_root;
4619 ret = btrfs_commit_transaction(trans);
4623 down(&fs_info->uuid_tree_rescan_sem);
4624 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4626 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4627 btrfs_warn(fs_info, "failed to start uuid_scan task");
4628 up(&fs_info->uuid_tree_rescan_sem);
4629 return PTR_ERR(task);
4635 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4637 struct task_struct *task;
4639 down(&fs_info->uuid_tree_rescan_sem);
4640 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4642 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4643 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4644 up(&fs_info->uuid_tree_rescan_sem);
4645 return PTR_ERR(task);
4652 * shrinking a device means finding all of the device extents past
4653 * the new size, and then following the back refs to the chunks.
4654 * The chunk relocation code actually frees the device extent
4656 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4658 struct btrfs_fs_info *fs_info = device->fs_info;
4659 struct btrfs_root *root = fs_info->dev_root;
4660 struct btrfs_trans_handle *trans;
4661 struct btrfs_dev_extent *dev_extent = NULL;
4662 struct btrfs_path *path;
4668 bool retried = false;
4669 struct extent_buffer *l;
4670 struct btrfs_key key;
4671 struct btrfs_super_block *super_copy = fs_info->super_copy;
4672 u64 old_total = btrfs_super_total_bytes(super_copy);
4673 u64 old_size = btrfs_device_get_total_bytes(device);
4677 new_size = round_down(new_size, fs_info->sectorsize);
4679 diff = round_down(old_size - new_size, fs_info->sectorsize);
4681 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4684 path = btrfs_alloc_path();
4688 path->reada = READA_BACK;
4690 trans = btrfs_start_transaction(root, 0);
4691 if (IS_ERR(trans)) {
4692 btrfs_free_path(path);
4693 return PTR_ERR(trans);
4696 mutex_lock(&fs_info->chunk_mutex);
4698 btrfs_device_set_total_bytes(device, new_size);
4699 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4700 device->fs_devices->total_rw_bytes -= diff;
4701 atomic64_sub(diff, &fs_info->free_chunk_space);
4705 * Once the device's size has been set to the new size, ensure all
4706 * in-memory chunks are synced to disk so that the loop below sees them
4707 * and relocates them accordingly.
4709 if (contains_pending_extent(device, &start, diff)) {
4710 mutex_unlock(&fs_info->chunk_mutex);
4711 ret = btrfs_commit_transaction(trans);
4715 mutex_unlock(&fs_info->chunk_mutex);
4716 btrfs_end_transaction(trans);
4720 key.objectid = device->devid;
4721 key.offset = (u64)-1;
4722 key.type = BTRFS_DEV_EXTENT_KEY;
4725 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4726 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4728 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4732 ret = btrfs_previous_item(root, path, 0, key.type);
4734 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4739 btrfs_release_path(path);
4744 slot = path->slots[0];
4745 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4747 if (key.objectid != device->devid) {
4748 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4749 btrfs_release_path(path);
4753 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4754 length = btrfs_dev_extent_length(l, dev_extent);
4756 if (key.offset + length <= new_size) {
4757 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4758 btrfs_release_path(path);
4762 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4763 btrfs_release_path(path);
4766 * We may be relocating the only data chunk we have,
4767 * which could potentially end up with losing data's
4768 * raid profile, so lets allocate an empty one in
4771 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4773 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4777 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4778 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4779 if (ret == -ENOSPC) {
4782 if (ret == -ETXTBSY) {
4784 "could not shrink block group %llu due to active swapfile",
4789 } while (key.offset-- > 0);
4791 if (failed && !retried) {
4795 } else if (failed && retried) {
4800 /* Shrinking succeeded, else we would be at "done". */
4801 trans = btrfs_start_transaction(root, 0);
4802 if (IS_ERR(trans)) {
4803 ret = PTR_ERR(trans);
4807 mutex_lock(&fs_info->chunk_mutex);
4808 btrfs_device_set_disk_total_bytes(device, new_size);
4809 if (list_empty(&device->post_commit_list))
4810 list_add_tail(&device->post_commit_list,
4811 &trans->transaction->dev_update_list);
4813 WARN_ON(diff > old_total);
4814 btrfs_set_super_total_bytes(super_copy,
4815 round_down(old_total - diff, fs_info->sectorsize));
4816 mutex_unlock(&fs_info->chunk_mutex);
4818 /* Now btrfs_update_device() will change the on-disk size. */
4819 ret = btrfs_update_device(trans, device);
4821 btrfs_abort_transaction(trans, ret);
4822 btrfs_end_transaction(trans);
4824 ret = btrfs_commit_transaction(trans);
4827 btrfs_free_path(path);
4829 mutex_lock(&fs_info->chunk_mutex);
4830 btrfs_device_set_total_bytes(device, old_size);
4831 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4832 device->fs_devices->total_rw_bytes += diff;
4833 atomic64_add(diff, &fs_info->free_chunk_space);
4834 mutex_unlock(&fs_info->chunk_mutex);
4839 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4840 struct btrfs_key *key,
4841 struct btrfs_chunk *chunk, int item_size)
4843 struct btrfs_super_block *super_copy = fs_info->super_copy;
4844 struct btrfs_disk_key disk_key;
4848 mutex_lock(&fs_info->chunk_mutex);
4849 array_size = btrfs_super_sys_array_size(super_copy);
4850 if (array_size + item_size + sizeof(disk_key)
4851 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4852 mutex_unlock(&fs_info->chunk_mutex);
4856 ptr = super_copy->sys_chunk_array + array_size;
4857 btrfs_cpu_key_to_disk(&disk_key, key);
4858 memcpy(ptr, &disk_key, sizeof(disk_key));
4859 ptr += sizeof(disk_key);
4860 memcpy(ptr, chunk, item_size);
4861 item_size += sizeof(disk_key);
4862 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4863 mutex_unlock(&fs_info->chunk_mutex);
4869 * sort the devices in descending order by max_avail, total_avail
4871 static int btrfs_cmp_device_info(const void *a, const void *b)
4873 const struct btrfs_device_info *di_a = a;
4874 const struct btrfs_device_info *di_b = b;
4876 if (di_a->max_avail > di_b->max_avail)
4878 if (di_a->max_avail < di_b->max_avail)
4880 if (di_a->total_avail > di_b->total_avail)
4882 if (di_a->total_avail < di_b->total_avail)
4887 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4889 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4892 btrfs_set_fs_incompat(info, RAID56);
4895 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4896 u64 start, u64 type)
4898 struct btrfs_fs_info *info = trans->fs_info;
4899 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4900 struct btrfs_device *device;
4901 struct map_lookup *map = NULL;
4902 struct extent_map_tree *em_tree;
4903 struct extent_map *em;
4904 struct btrfs_device_info *devices_info = NULL;
4906 int num_stripes; /* total number of stripes to allocate */
4907 int data_stripes; /* number of stripes that count for
4909 int sub_stripes; /* sub_stripes info for map */
4910 int dev_stripes; /* stripes per dev */
4911 int devs_max; /* max devs to use */
4912 int devs_min; /* min devs needed */
4913 int devs_increment; /* ndevs has to be a multiple of this */
4914 int ncopies; /* how many copies to data has */
4915 int nparity; /* number of stripes worth of bytes to
4916 store parity information */
4918 u64 max_stripe_size;
4927 BUG_ON(!alloc_profile_is_valid(type, 0));
4929 if (list_empty(&fs_devices->alloc_list)) {
4930 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4931 btrfs_debug(info, "%s: no writable device", __func__);
4935 index = btrfs_bg_flags_to_raid_index(type);
4937 sub_stripes = btrfs_raid_array[index].sub_stripes;
4938 dev_stripes = btrfs_raid_array[index].dev_stripes;
4939 devs_max = btrfs_raid_array[index].devs_max;
4941 devs_max = BTRFS_MAX_DEVS(info);
4942 devs_min = btrfs_raid_array[index].devs_min;
4943 devs_increment = btrfs_raid_array[index].devs_increment;
4944 ncopies = btrfs_raid_array[index].ncopies;
4945 nparity = btrfs_raid_array[index].nparity;
4947 if (type & BTRFS_BLOCK_GROUP_DATA) {
4948 max_stripe_size = SZ_1G;
4949 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4950 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4951 /* for larger filesystems, use larger metadata chunks */
4952 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4953 max_stripe_size = SZ_1G;
4955 max_stripe_size = SZ_256M;
4956 max_chunk_size = max_stripe_size;
4957 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4958 max_stripe_size = SZ_32M;
4959 max_chunk_size = 2 * max_stripe_size;
4961 btrfs_err(info, "invalid chunk type 0x%llx requested",
4966 /* We don't want a chunk larger than 10% of writable space */
4967 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4970 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4976 * in the first pass through the devices list, we gather information
4977 * about the available holes on each device.
4980 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4984 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4986 "BTRFS: read-only device in alloc_list\n");
4990 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4991 &device->dev_state) ||
4992 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4995 if (device->total_bytes > device->bytes_used)
4996 total_avail = device->total_bytes - device->bytes_used;
5000 /* If there is no space on this device, skip it. */
5001 if (total_avail == 0)
5004 ret = find_free_dev_extent(device,
5005 max_stripe_size * dev_stripes,
5006 &dev_offset, &max_avail);
5007 if (ret && ret != -ENOSPC)
5011 max_avail = max_stripe_size * dev_stripes;
5013 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5014 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5016 "%s: devid %llu has no free space, have=%llu want=%u",
5017 __func__, device->devid, max_avail,
5018 BTRFS_STRIPE_LEN * dev_stripes);
5022 if (ndevs == fs_devices->rw_devices) {
5023 WARN(1, "%s: found more than %llu devices\n",
5024 __func__, fs_devices->rw_devices);
5027 devices_info[ndevs].dev_offset = dev_offset;
5028 devices_info[ndevs].max_avail = max_avail;
5029 devices_info[ndevs].total_avail = total_avail;
5030 devices_info[ndevs].dev = device;
5035 * now sort the devices by hole size / available space
5037 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5038 btrfs_cmp_device_info, NULL);
5040 /* round down to number of usable stripes */
5041 ndevs = round_down(ndevs, devs_increment);
5043 if (ndevs < devs_min) {
5045 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5047 "%s: not enough devices with free space: have=%d minimum required=%d",
5048 __func__, ndevs, devs_min);
5053 ndevs = min(ndevs, devs_max);
5056 * The primary goal is to maximize the number of stripes, so use as
5057 * many devices as possible, even if the stripes are not maximum sized.
5059 * The DUP profile stores more than one stripe per device, the
5060 * max_avail is the total size so we have to adjust.
5062 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5063 num_stripes = ndevs * dev_stripes;
5066 * this will have to be fixed for RAID1 and RAID10 over
5069 data_stripes = (num_stripes - nparity) / ncopies;
5072 * Use the number of data stripes to figure out how big this chunk
5073 * is really going to be in terms of logical address space,
5074 * and compare that answer with the max chunk size. If it's higher,
5075 * we try to reduce stripe_size.
5077 if (stripe_size * data_stripes > max_chunk_size) {
5079 * Reduce stripe_size, round it up to a 16MB boundary again and
5080 * then use it, unless it ends up being even bigger than the
5081 * previous value we had already.
5083 stripe_size = min(round_up(div_u64(max_chunk_size,
5084 data_stripes), SZ_16M),
5088 /* align to BTRFS_STRIPE_LEN */
5089 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5091 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5096 map->num_stripes = num_stripes;
5098 for (i = 0; i < ndevs; ++i) {
5099 for (j = 0; j < dev_stripes; ++j) {
5100 int s = i * dev_stripes + j;
5101 map->stripes[s].dev = devices_info[i].dev;
5102 map->stripes[s].physical = devices_info[i].dev_offset +
5106 map->stripe_len = BTRFS_STRIPE_LEN;
5107 map->io_align = BTRFS_STRIPE_LEN;
5108 map->io_width = BTRFS_STRIPE_LEN;
5110 map->sub_stripes = sub_stripes;
5112 chunk_size = stripe_size * data_stripes;
5114 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5116 em = alloc_extent_map();
5122 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5123 em->map_lookup = map;
5125 em->len = chunk_size;
5126 em->block_start = 0;
5127 em->block_len = em->len;
5128 em->orig_block_len = stripe_size;
5130 em_tree = &info->mapping_tree;
5131 write_lock(&em_tree->lock);
5132 ret = add_extent_mapping(em_tree, em, 0);
5134 write_unlock(&em_tree->lock);
5135 free_extent_map(em);
5138 write_unlock(&em_tree->lock);
5140 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5142 goto error_del_extent;
5144 for (i = 0; i < map->num_stripes; i++) {
5145 struct btrfs_device *dev = map->stripes[i].dev;
5147 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5148 if (list_empty(&dev->post_commit_list))
5149 list_add_tail(&dev->post_commit_list,
5150 &trans->transaction->dev_update_list);
5153 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5155 free_extent_map(em);
5156 check_raid56_incompat_flag(info, type);
5158 kfree(devices_info);
5162 write_lock(&em_tree->lock);
5163 remove_extent_mapping(em_tree, em);
5164 write_unlock(&em_tree->lock);
5166 /* One for our allocation */
5167 free_extent_map(em);
5168 /* One for the tree reference */
5169 free_extent_map(em);
5171 kfree(devices_info);
5175 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5176 u64 chunk_offset, u64 chunk_size)
5178 struct btrfs_fs_info *fs_info = trans->fs_info;
5179 struct btrfs_root *extent_root = fs_info->extent_root;
5180 struct btrfs_root *chunk_root = fs_info->chunk_root;
5181 struct btrfs_key key;
5182 struct btrfs_device *device;
5183 struct btrfs_chunk *chunk;
5184 struct btrfs_stripe *stripe;
5185 struct extent_map *em;
5186 struct map_lookup *map;
5193 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5197 map = em->map_lookup;
5198 item_size = btrfs_chunk_item_size(map->num_stripes);
5199 stripe_size = em->orig_block_len;
5201 chunk = kzalloc(item_size, GFP_NOFS);
5208 * Take the device list mutex to prevent races with the final phase of
5209 * a device replace operation that replaces the device object associated
5210 * with the map's stripes, because the device object's id can change
5211 * at any time during that final phase of the device replace operation
5212 * (dev-replace.c:btrfs_dev_replace_finishing()).
5214 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5215 for (i = 0; i < map->num_stripes; i++) {
5216 device = map->stripes[i].dev;
5217 dev_offset = map->stripes[i].physical;
5219 ret = btrfs_update_device(trans, device);
5222 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5223 dev_offset, stripe_size);
5228 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5232 stripe = &chunk->stripe;
5233 for (i = 0; i < map->num_stripes; i++) {
5234 device = map->stripes[i].dev;
5235 dev_offset = map->stripes[i].physical;
5237 btrfs_set_stack_stripe_devid(stripe, device->devid);
5238 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5239 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5242 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5244 btrfs_set_stack_chunk_length(chunk, chunk_size);
5245 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5246 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5247 btrfs_set_stack_chunk_type(chunk, map->type);
5248 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5249 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5250 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5251 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5252 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5254 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5255 key.type = BTRFS_CHUNK_ITEM_KEY;
5256 key.offset = chunk_offset;
5258 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5259 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5261 * TODO: Cleanup of inserted chunk root in case of
5264 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5269 free_extent_map(em);
5274 * Chunk allocation falls into two parts. The first part does work
5275 * that makes the new allocated chunk usable, but does not do any operation
5276 * that modifies the chunk tree. The second part does the work that
5277 * requires modifying the chunk tree. This division is important for the
5278 * bootstrap process of adding storage to a seed btrfs.
5280 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5284 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5285 chunk_offset = find_next_chunk(trans->fs_info);
5286 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5289 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5291 struct btrfs_fs_info *fs_info = trans->fs_info;
5293 u64 sys_chunk_offset;
5297 chunk_offset = find_next_chunk(fs_info);
5298 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5299 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5303 sys_chunk_offset = find_next_chunk(fs_info);
5304 alloc_profile = btrfs_system_alloc_profile(fs_info);
5305 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5309 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5311 const int index = btrfs_bg_flags_to_raid_index(map->type);
5313 return btrfs_raid_array[index].tolerated_failures;
5316 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5318 struct extent_map *em;
5319 struct map_lookup *map;
5324 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5328 map = em->map_lookup;
5329 for (i = 0; i < map->num_stripes; i++) {
5330 if (test_bit(BTRFS_DEV_STATE_MISSING,
5331 &map->stripes[i].dev->dev_state)) {
5335 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5336 &map->stripes[i].dev->dev_state)) {
5343 * If the number of missing devices is larger than max errors,
5344 * we can not write the data into that chunk successfully, so
5347 if (miss_ndevs > btrfs_chunk_max_errors(map))
5350 free_extent_map(em);
5354 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5356 struct extent_map *em;
5359 write_lock(&tree->lock);
5360 em = lookup_extent_mapping(tree, 0, (u64)-1);
5362 remove_extent_mapping(tree, em);
5363 write_unlock(&tree->lock);
5367 free_extent_map(em);
5368 /* once for the tree */
5369 free_extent_map(em);
5373 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5375 struct extent_map *em;
5376 struct map_lookup *map;
5379 em = btrfs_get_chunk_map(fs_info, logical, len);
5382 * We could return errors for these cases, but that could get
5383 * ugly and we'd probably do the same thing which is just not do
5384 * anything else and exit, so return 1 so the callers don't try
5385 * to use other copies.
5389 map = em->map_lookup;
5390 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5391 ret = map->num_stripes;
5392 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5393 ret = map->sub_stripes;
5394 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5396 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5398 * There could be two corrupted data stripes, we need
5399 * to loop retry in order to rebuild the correct data.
5401 * Fail a stripe at a time on every retry except the
5402 * stripe under reconstruction.
5404 ret = map->num_stripes;
5407 free_extent_map(em);
5409 down_read(&fs_info->dev_replace.rwsem);
5410 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5411 fs_info->dev_replace.tgtdev)
5413 up_read(&fs_info->dev_replace.rwsem);
5418 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5421 struct extent_map *em;
5422 struct map_lookup *map;
5423 unsigned long len = fs_info->sectorsize;
5425 em = btrfs_get_chunk_map(fs_info, logical, len);
5427 if (!WARN_ON(IS_ERR(em))) {
5428 map = em->map_lookup;
5429 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5430 len = map->stripe_len * nr_data_stripes(map);
5431 free_extent_map(em);
5436 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5438 struct extent_map *em;
5439 struct map_lookup *map;
5442 em = btrfs_get_chunk_map(fs_info, logical, len);
5444 if(!WARN_ON(IS_ERR(em))) {
5445 map = em->map_lookup;
5446 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5448 free_extent_map(em);
5453 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5454 struct map_lookup *map, int first,
5455 int dev_replace_is_ongoing)
5459 int preferred_mirror;
5461 struct btrfs_device *srcdev;
5464 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5466 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5467 num_stripes = map->sub_stripes;
5469 num_stripes = map->num_stripes;
5471 preferred_mirror = first + current->pid % num_stripes;
5473 if (dev_replace_is_ongoing &&
5474 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5475 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5476 srcdev = fs_info->dev_replace.srcdev;
5481 * try to avoid the drive that is the source drive for a
5482 * dev-replace procedure, only choose it if no other non-missing
5483 * mirror is available
5485 for (tolerance = 0; tolerance < 2; tolerance++) {
5486 if (map->stripes[preferred_mirror].dev->bdev &&
5487 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5488 return preferred_mirror;
5489 for (i = first; i < first + num_stripes; i++) {
5490 if (map->stripes[i].dev->bdev &&
5491 (tolerance || map->stripes[i].dev != srcdev))
5496 /* we couldn't find one that doesn't fail. Just return something
5497 * and the io error handling code will clean up eventually
5499 return preferred_mirror;
5502 static inline int parity_smaller(u64 a, u64 b)
5507 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5508 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5510 struct btrfs_bio_stripe s;
5517 for (i = 0; i < num_stripes - 1; i++) {
5518 if (parity_smaller(bbio->raid_map[i],
5519 bbio->raid_map[i+1])) {
5520 s = bbio->stripes[i];
5521 l = bbio->raid_map[i];
5522 bbio->stripes[i] = bbio->stripes[i+1];
5523 bbio->raid_map[i] = bbio->raid_map[i+1];
5524 bbio->stripes[i+1] = s;
5525 bbio->raid_map[i+1] = l;
5533 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5535 struct btrfs_bio *bbio = kzalloc(
5536 /* the size of the btrfs_bio */
5537 sizeof(struct btrfs_bio) +
5538 /* plus the variable array for the stripes */
5539 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5540 /* plus the variable array for the tgt dev */
5541 sizeof(int) * (real_stripes) +
5543 * plus the raid_map, which includes both the tgt dev
5546 sizeof(u64) * (total_stripes),
5547 GFP_NOFS|__GFP_NOFAIL);
5549 atomic_set(&bbio->error, 0);
5550 refcount_set(&bbio->refs, 1);
5555 void btrfs_get_bbio(struct btrfs_bio *bbio)
5557 WARN_ON(!refcount_read(&bbio->refs));
5558 refcount_inc(&bbio->refs);
5561 void btrfs_put_bbio(struct btrfs_bio *bbio)
5565 if (refcount_dec_and_test(&bbio->refs))
5569 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5571 * Please note that, discard won't be sent to target device of device
5574 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5575 u64 logical, u64 length,
5576 struct btrfs_bio **bbio_ret)
5578 struct extent_map *em;
5579 struct map_lookup *map;
5580 struct btrfs_bio *bbio;
5584 u64 stripe_end_offset;
5591 u32 sub_stripes = 0;
5592 u64 stripes_per_dev = 0;
5593 u32 remaining_stripes = 0;
5594 u32 last_stripe = 0;
5598 /* discard always return a bbio */
5601 em = btrfs_get_chunk_map(fs_info, logical, length);
5605 map = em->map_lookup;
5606 /* we don't discard raid56 yet */
5607 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5612 offset = logical - em->start;
5613 length = min_t(u64, em->len - offset, length);
5615 stripe_len = map->stripe_len;
5617 * stripe_nr counts the total number of stripes we have to stride
5618 * to get to this block
5620 stripe_nr = div64_u64(offset, stripe_len);
5622 /* stripe_offset is the offset of this block in its stripe */
5623 stripe_offset = offset - stripe_nr * stripe_len;
5625 stripe_nr_end = round_up(offset + length, map->stripe_len);
5626 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5627 stripe_cnt = stripe_nr_end - stripe_nr;
5628 stripe_end_offset = stripe_nr_end * map->stripe_len -
5631 * after this, stripe_nr is the number of stripes on this
5632 * device we have to walk to find the data, and stripe_index is
5633 * the number of our device in the stripe array
5637 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5638 BTRFS_BLOCK_GROUP_RAID10)) {
5639 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5642 sub_stripes = map->sub_stripes;
5644 factor = map->num_stripes / sub_stripes;
5645 num_stripes = min_t(u64, map->num_stripes,
5646 sub_stripes * stripe_cnt);
5647 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5648 stripe_index *= sub_stripes;
5649 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5650 &remaining_stripes);
5651 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5652 last_stripe *= sub_stripes;
5653 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5654 BTRFS_BLOCK_GROUP_DUP)) {
5655 num_stripes = map->num_stripes;
5657 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5661 bbio = alloc_btrfs_bio(num_stripes, 0);
5667 for (i = 0; i < num_stripes; i++) {
5668 bbio->stripes[i].physical =
5669 map->stripes[stripe_index].physical +
5670 stripe_offset + stripe_nr * map->stripe_len;
5671 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5673 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5674 BTRFS_BLOCK_GROUP_RAID10)) {
5675 bbio->stripes[i].length = stripes_per_dev *
5678 if (i / sub_stripes < remaining_stripes)
5679 bbio->stripes[i].length +=
5683 * Special for the first stripe and
5686 * |-------|...|-------|
5690 if (i < sub_stripes)
5691 bbio->stripes[i].length -=
5694 if (stripe_index >= last_stripe &&
5695 stripe_index <= (last_stripe +
5697 bbio->stripes[i].length -=
5700 if (i == sub_stripes - 1)
5703 bbio->stripes[i].length = length;
5707 if (stripe_index == map->num_stripes) {
5714 bbio->map_type = map->type;
5715 bbio->num_stripes = num_stripes;
5717 free_extent_map(em);
5722 * In dev-replace case, for repair case (that's the only case where the mirror
5723 * is selected explicitly when calling btrfs_map_block), blocks left of the
5724 * left cursor can also be read from the target drive.
5726 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5728 * For READ, it also needs to be supported using the same mirror number.
5730 * If the requested block is not left of the left cursor, EIO is returned. This
5731 * can happen because btrfs_num_copies() returns one more in the dev-replace
5734 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5735 u64 logical, u64 length,
5736 u64 srcdev_devid, int *mirror_num,
5739 struct btrfs_bio *bbio = NULL;
5741 int index_srcdev = 0;
5743 u64 physical_of_found = 0;
5747 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5748 logical, &length, &bbio, 0, 0);
5750 ASSERT(bbio == NULL);
5754 num_stripes = bbio->num_stripes;
5755 if (*mirror_num > num_stripes) {
5757 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5758 * that means that the requested area is not left of the left
5761 btrfs_put_bbio(bbio);
5766 * process the rest of the function using the mirror_num of the source
5767 * drive. Therefore look it up first. At the end, patch the device
5768 * pointer to the one of the target drive.
5770 for (i = 0; i < num_stripes; i++) {
5771 if (bbio->stripes[i].dev->devid != srcdev_devid)
5775 * In case of DUP, in order to keep it simple, only add the
5776 * mirror with the lowest physical address
5779 physical_of_found <= bbio->stripes[i].physical)
5784 physical_of_found = bbio->stripes[i].physical;
5787 btrfs_put_bbio(bbio);
5793 *mirror_num = index_srcdev + 1;
5794 *physical = physical_of_found;
5798 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5799 struct btrfs_bio **bbio_ret,
5800 struct btrfs_dev_replace *dev_replace,
5801 int *num_stripes_ret, int *max_errors_ret)
5803 struct btrfs_bio *bbio = *bbio_ret;
5804 u64 srcdev_devid = dev_replace->srcdev->devid;
5805 int tgtdev_indexes = 0;
5806 int num_stripes = *num_stripes_ret;
5807 int max_errors = *max_errors_ret;
5810 if (op == BTRFS_MAP_WRITE) {
5811 int index_where_to_add;
5814 * duplicate the write operations while the dev replace
5815 * procedure is running. Since the copying of the old disk to
5816 * the new disk takes place at run time while the filesystem is
5817 * mounted writable, the regular write operations to the old
5818 * disk have to be duplicated to go to the new disk as well.
5820 * Note that device->missing is handled by the caller, and that
5821 * the write to the old disk is already set up in the stripes
5824 index_where_to_add = num_stripes;
5825 for (i = 0; i < num_stripes; i++) {
5826 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5827 /* write to new disk, too */
5828 struct btrfs_bio_stripe *new =
5829 bbio->stripes + index_where_to_add;
5830 struct btrfs_bio_stripe *old =
5833 new->physical = old->physical;
5834 new->length = old->length;
5835 new->dev = dev_replace->tgtdev;
5836 bbio->tgtdev_map[i] = index_where_to_add;
5837 index_where_to_add++;
5842 num_stripes = index_where_to_add;
5843 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5844 int index_srcdev = 0;
5846 u64 physical_of_found = 0;
5849 * During the dev-replace procedure, the target drive can also
5850 * be used to read data in case it is needed to repair a corrupt
5851 * block elsewhere. This is possible if the requested area is
5852 * left of the left cursor. In this area, the target drive is a
5853 * full copy of the source drive.
5855 for (i = 0; i < num_stripes; i++) {
5856 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5858 * In case of DUP, in order to keep it simple,
5859 * only add the mirror with the lowest physical
5863 physical_of_found <=
5864 bbio->stripes[i].physical)
5868 physical_of_found = bbio->stripes[i].physical;
5872 struct btrfs_bio_stripe *tgtdev_stripe =
5873 bbio->stripes + num_stripes;
5875 tgtdev_stripe->physical = physical_of_found;
5876 tgtdev_stripe->length =
5877 bbio->stripes[index_srcdev].length;
5878 tgtdev_stripe->dev = dev_replace->tgtdev;
5879 bbio->tgtdev_map[index_srcdev] = num_stripes;
5886 *num_stripes_ret = num_stripes;
5887 *max_errors_ret = max_errors;
5888 bbio->num_tgtdevs = tgtdev_indexes;
5892 static bool need_full_stripe(enum btrfs_map_op op)
5894 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5898 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5899 * tuple. This information is used to calculate how big a
5900 * particular bio can get before it straddles a stripe.
5902 * @fs_info - the filesystem
5903 * @logical - address that we want to figure out the geometry of
5904 * @len - the length of IO we are going to perform, starting at @logical
5905 * @op - type of operation - write or read
5906 * @io_geom - pointer used to return values
5908 * Returns < 0 in case a chunk for the given logical address cannot be found,
5909 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5911 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5912 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5914 struct extent_map *em;
5915 struct map_lookup *map;
5920 u64 raid56_full_stripe_start = (u64)-1;
5924 ASSERT(op != BTRFS_MAP_DISCARD);
5926 em = btrfs_get_chunk_map(fs_info, logical, len);
5930 map = em->map_lookup;
5931 /* Offset of this logical address in the chunk */
5932 offset = logical - em->start;
5933 /* Len of a stripe in a chunk */
5934 stripe_len = map->stripe_len;
5935 /* Stripe wher this block falls in */
5936 stripe_nr = div64_u64(offset, stripe_len);
5937 /* Offset of stripe in the chunk */
5938 stripe_offset = stripe_nr * stripe_len;
5939 if (offset < stripe_offset) {
5941 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5942 stripe_offset, offset, em->start, logical, stripe_len);
5947 /* stripe_offset is the offset of this block in its stripe */
5948 stripe_offset = offset - stripe_offset;
5949 data_stripes = nr_data_stripes(map);
5951 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5952 u64 max_len = stripe_len - stripe_offset;
5955 * In case of raid56, we need to know the stripe aligned start
5957 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5958 unsigned long full_stripe_len = stripe_len * data_stripes;
5959 raid56_full_stripe_start = offset;
5962 * Allow a write of a full stripe, but make sure we
5963 * don't allow straddling of stripes
5965 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5967 raid56_full_stripe_start *= full_stripe_len;
5970 * For writes to RAID[56], allow a full stripeset across
5971 * all disks. For other RAID types and for RAID[56]
5972 * reads, just allow a single stripe (on a single disk).
5974 if (op == BTRFS_MAP_WRITE) {
5975 max_len = stripe_len * data_stripes -
5976 (offset - raid56_full_stripe_start);
5979 len = min_t(u64, em->len - offset, max_len);
5981 len = em->len - offset;
5985 io_geom->offset = offset;
5986 io_geom->stripe_len = stripe_len;
5987 io_geom->stripe_nr = stripe_nr;
5988 io_geom->stripe_offset = stripe_offset;
5989 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
5993 free_extent_map(em);
5997 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5998 enum btrfs_map_op op,
5999 u64 logical, u64 *length,
6000 struct btrfs_bio **bbio_ret,
6001 int mirror_num, int need_raid_map)
6003 struct extent_map *em;
6004 struct map_lookup *map;
6014 int tgtdev_indexes = 0;
6015 struct btrfs_bio *bbio = NULL;
6016 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6017 int dev_replace_is_ongoing = 0;
6018 int num_alloc_stripes;
6019 int patch_the_first_stripe_for_dev_replace = 0;
6020 u64 physical_to_patch_in_first_stripe = 0;
6021 u64 raid56_full_stripe_start = (u64)-1;
6022 struct btrfs_io_geometry geom;
6026 if (op == BTRFS_MAP_DISCARD)
6027 return __btrfs_map_block_for_discard(fs_info, logical,
6030 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6034 em = btrfs_get_chunk_map(fs_info, logical, *length);
6035 ASSERT(!IS_ERR(em));
6036 map = em->map_lookup;
6039 stripe_len = geom.stripe_len;
6040 stripe_nr = geom.stripe_nr;
6041 stripe_offset = geom.stripe_offset;
6042 raid56_full_stripe_start = geom.raid56_stripe_offset;
6043 data_stripes = nr_data_stripes(map);
6045 down_read(&dev_replace->rwsem);
6046 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6048 * Hold the semaphore for read during the whole operation, write is
6049 * requested at commit time but must wait.
6051 if (!dev_replace_is_ongoing)
6052 up_read(&dev_replace->rwsem);
6054 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6055 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6056 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6057 dev_replace->srcdev->devid,
6059 &physical_to_patch_in_first_stripe);
6063 patch_the_first_stripe_for_dev_replace = 1;
6064 } else if (mirror_num > map->num_stripes) {
6070 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6071 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6073 if (!need_full_stripe(op))
6075 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6076 if (need_full_stripe(op))
6077 num_stripes = map->num_stripes;
6078 else if (mirror_num)
6079 stripe_index = mirror_num - 1;
6081 stripe_index = find_live_mirror(fs_info, map, 0,
6082 dev_replace_is_ongoing);
6083 mirror_num = stripe_index + 1;
6086 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6087 if (need_full_stripe(op)) {
6088 num_stripes = map->num_stripes;
6089 } else if (mirror_num) {
6090 stripe_index = mirror_num - 1;
6095 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6096 u32 factor = map->num_stripes / map->sub_stripes;
6098 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6099 stripe_index *= map->sub_stripes;
6101 if (need_full_stripe(op))
6102 num_stripes = map->sub_stripes;
6103 else if (mirror_num)
6104 stripe_index += mirror_num - 1;
6106 int old_stripe_index = stripe_index;
6107 stripe_index = find_live_mirror(fs_info, map,
6109 dev_replace_is_ongoing);
6110 mirror_num = stripe_index - old_stripe_index + 1;
6113 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6114 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6115 /* push stripe_nr back to the start of the full stripe */
6116 stripe_nr = div64_u64(raid56_full_stripe_start,
6117 stripe_len * data_stripes);
6119 /* RAID[56] write or recovery. Return all stripes */
6120 num_stripes = map->num_stripes;
6121 max_errors = nr_parity_stripes(map);
6123 *length = map->stripe_len;
6128 * Mirror #0 or #1 means the original data block.
6129 * Mirror #2 is RAID5 parity block.
6130 * Mirror #3 is RAID6 Q block.
6132 stripe_nr = div_u64_rem(stripe_nr,
6133 data_stripes, &stripe_index);
6135 stripe_index = data_stripes + mirror_num - 2;
6137 /* We distribute the parity blocks across stripes */
6138 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6140 if (!need_full_stripe(op) && mirror_num <= 1)
6145 * after this, stripe_nr is the number of stripes on this
6146 * device we have to walk to find the data, and stripe_index is
6147 * the number of our device in the stripe array
6149 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6151 mirror_num = stripe_index + 1;
6153 if (stripe_index >= map->num_stripes) {
6155 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6156 stripe_index, map->num_stripes);
6161 num_alloc_stripes = num_stripes;
6162 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6163 if (op == BTRFS_MAP_WRITE)
6164 num_alloc_stripes <<= 1;
6165 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6166 num_alloc_stripes++;
6167 tgtdev_indexes = num_stripes;
6170 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6175 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6176 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6178 /* build raid_map */
6179 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6180 (need_full_stripe(op) || mirror_num > 1)) {
6184 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6185 sizeof(struct btrfs_bio_stripe) *
6187 sizeof(int) * tgtdev_indexes);
6189 /* Work out the disk rotation on this stripe-set */
6190 div_u64_rem(stripe_nr, num_stripes, &rot);
6192 /* Fill in the logical address of each stripe */
6193 tmp = stripe_nr * data_stripes;
6194 for (i = 0; i < data_stripes; i++)
6195 bbio->raid_map[(i+rot) % num_stripes] =
6196 em->start + (tmp + i) * map->stripe_len;
6198 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6199 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6200 bbio->raid_map[(i+rot+1) % num_stripes] =
6205 for (i = 0; i < num_stripes; i++) {
6206 bbio->stripes[i].physical =
6207 map->stripes[stripe_index].physical +
6209 stripe_nr * map->stripe_len;
6210 bbio->stripes[i].dev =
6211 map->stripes[stripe_index].dev;
6215 if (need_full_stripe(op))
6216 max_errors = btrfs_chunk_max_errors(map);
6219 sort_parity_stripes(bbio, num_stripes);
6221 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6222 need_full_stripe(op)) {
6223 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6228 bbio->map_type = map->type;
6229 bbio->num_stripes = num_stripes;
6230 bbio->max_errors = max_errors;
6231 bbio->mirror_num = mirror_num;
6234 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6235 * mirror_num == num_stripes + 1 && dev_replace target drive is
6236 * available as a mirror
6238 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6239 WARN_ON(num_stripes > 1);
6240 bbio->stripes[0].dev = dev_replace->tgtdev;
6241 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6242 bbio->mirror_num = map->num_stripes + 1;
6245 if (dev_replace_is_ongoing) {
6246 lockdep_assert_held(&dev_replace->rwsem);
6247 /* Unlock and let waiting writers proceed */
6248 up_read(&dev_replace->rwsem);
6250 free_extent_map(em);
6254 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6255 u64 logical, u64 *length,
6256 struct btrfs_bio **bbio_ret, int mirror_num)
6258 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6262 /* For Scrub/replace */
6263 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6264 u64 logical, u64 *length,
6265 struct btrfs_bio **bbio_ret)
6267 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6270 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6271 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6273 struct extent_map *em;
6274 struct map_lookup *map;
6282 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6286 map = em->map_lookup;
6288 rmap_len = map->stripe_len;
6290 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6291 length = div_u64(length, map->num_stripes / map->sub_stripes);
6292 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6293 length = div_u64(length, map->num_stripes);
6294 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6295 length = div_u64(length, nr_data_stripes(map));
6296 rmap_len = map->stripe_len * nr_data_stripes(map);
6299 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6300 BUG_ON(!buf); /* -ENOMEM */
6302 for (i = 0; i < map->num_stripes; i++) {
6303 if (map->stripes[i].physical > physical ||
6304 map->stripes[i].physical + length <= physical)
6307 stripe_nr = physical - map->stripes[i].physical;
6308 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6310 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6311 stripe_nr = stripe_nr * map->num_stripes + i;
6312 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6313 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6314 stripe_nr = stripe_nr * map->num_stripes + i;
6315 } /* else if RAID[56], multiply by nr_data_stripes().
6316 * Alternatively, just use rmap_len below instead of
6317 * map->stripe_len */
6319 bytenr = chunk_start + stripe_nr * rmap_len;
6320 WARN_ON(nr >= map->num_stripes);
6321 for (j = 0; j < nr; j++) {
6322 if (buf[j] == bytenr)
6326 WARN_ON(nr >= map->num_stripes);
6333 *stripe_len = rmap_len;
6335 free_extent_map(em);
6339 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6341 bio->bi_private = bbio->private;
6342 bio->bi_end_io = bbio->end_io;
6345 btrfs_put_bbio(bbio);
6348 static void btrfs_end_bio(struct bio *bio)
6350 struct btrfs_bio *bbio = bio->bi_private;
6351 int is_orig_bio = 0;
6353 if (bio->bi_status) {
6354 atomic_inc(&bbio->error);
6355 if (bio->bi_status == BLK_STS_IOERR ||
6356 bio->bi_status == BLK_STS_TARGET) {
6357 unsigned int stripe_index =
6358 btrfs_io_bio(bio)->stripe_index;
6359 struct btrfs_device *dev;
6361 BUG_ON(stripe_index >= bbio->num_stripes);
6362 dev = bbio->stripes[stripe_index].dev;
6364 if (bio_op(bio) == REQ_OP_WRITE)
6365 btrfs_dev_stat_inc_and_print(dev,
6366 BTRFS_DEV_STAT_WRITE_ERRS);
6367 else if (!(bio->bi_opf & REQ_RAHEAD))
6368 btrfs_dev_stat_inc_and_print(dev,
6369 BTRFS_DEV_STAT_READ_ERRS);
6370 if (bio->bi_opf & REQ_PREFLUSH)
6371 btrfs_dev_stat_inc_and_print(dev,
6372 BTRFS_DEV_STAT_FLUSH_ERRS);
6377 if (bio == bbio->orig_bio)
6380 btrfs_bio_counter_dec(bbio->fs_info);
6382 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6385 bio = bbio->orig_bio;
6388 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6389 /* only send an error to the higher layers if it is
6390 * beyond the tolerance of the btrfs bio
6392 if (atomic_read(&bbio->error) > bbio->max_errors) {
6393 bio->bi_status = BLK_STS_IOERR;
6396 * this bio is actually up to date, we didn't
6397 * go over the max number of errors
6399 bio->bi_status = BLK_STS_OK;
6402 btrfs_end_bbio(bbio, bio);
6403 } else if (!is_orig_bio) {
6409 * see run_scheduled_bios for a description of why bios are collected for
6412 * This will add one bio to the pending list for a device and make sure
6413 * the work struct is scheduled.
6415 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6418 struct btrfs_fs_info *fs_info = device->fs_info;
6419 int should_queue = 1;
6420 struct btrfs_pending_bios *pending_bios;
6422 /* don't bother with additional async steps for reads, right now */
6423 if (bio_op(bio) == REQ_OP_READ) {
6424 btrfsic_submit_bio(bio);
6428 WARN_ON(bio->bi_next);
6429 bio->bi_next = NULL;
6431 spin_lock(&device->io_lock);
6432 if (op_is_sync(bio->bi_opf))
6433 pending_bios = &device->pending_sync_bios;
6435 pending_bios = &device->pending_bios;
6437 if (pending_bios->tail)
6438 pending_bios->tail->bi_next = bio;
6440 pending_bios->tail = bio;
6441 if (!pending_bios->head)
6442 pending_bios->head = bio;
6443 if (device->running_pending)
6446 spin_unlock(&device->io_lock);
6449 btrfs_queue_work(fs_info->submit_workers, &device->work);
6452 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6453 u64 physical, int dev_nr, int async)
6455 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6456 struct btrfs_fs_info *fs_info = bbio->fs_info;
6458 bio->bi_private = bbio;
6459 btrfs_io_bio(bio)->stripe_index = dev_nr;
6460 bio->bi_end_io = btrfs_end_bio;
6461 bio->bi_iter.bi_sector = physical >> 9;
6462 btrfs_debug_in_rcu(fs_info,
6463 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6464 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6465 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6466 bio->bi_iter.bi_size);
6467 bio_set_dev(bio, dev->bdev);
6469 btrfs_bio_counter_inc_noblocked(fs_info);
6472 btrfs_schedule_bio(dev, bio);
6474 btrfsic_submit_bio(bio);
6477 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6479 atomic_inc(&bbio->error);
6480 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6481 /* Should be the original bio. */
6482 WARN_ON(bio != bbio->orig_bio);
6484 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6485 bio->bi_iter.bi_sector = logical >> 9;
6486 if (atomic_read(&bbio->error) > bbio->max_errors)
6487 bio->bi_status = BLK_STS_IOERR;
6489 bio->bi_status = BLK_STS_OK;
6490 btrfs_end_bbio(bbio, bio);
6494 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6495 int mirror_num, int async_submit)
6497 struct btrfs_device *dev;
6498 struct bio *first_bio = bio;
6499 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6505 struct btrfs_bio *bbio = NULL;
6507 length = bio->bi_iter.bi_size;
6508 map_length = length;
6510 btrfs_bio_counter_inc_blocked(fs_info);
6511 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6512 &map_length, &bbio, mirror_num, 1);
6514 btrfs_bio_counter_dec(fs_info);
6515 return errno_to_blk_status(ret);
6518 total_devs = bbio->num_stripes;
6519 bbio->orig_bio = first_bio;
6520 bbio->private = first_bio->bi_private;
6521 bbio->end_io = first_bio->bi_end_io;
6522 bbio->fs_info = fs_info;
6523 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6525 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6526 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6527 /* In this case, map_length has been set to the length of
6528 a single stripe; not the whole write */
6529 if (bio_op(bio) == REQ_OP_WRITE) {
6530 ret = raid56_parity_write(fs_info, bio, bbio,
6533 ret = raid56_parity_recover(fs_info, bio, bbio,
6534 map_length, mirror_num, 1);
6537 btrfs_bio_counter_dec(fs_info);
6538 return errno_to_blk_status(ret);
6541 if (map_length < length) {
6543 "mapping failed logical %llu bio len %llu len %llu",
6544 logical, length, map_length);
6548 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6549 dev = bbio->stripes[dev_nr].dev;
6550 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6552 (bio_op(first_bio) == REQ_OP_WRITE &&
6553 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6554 bbio_error(bbio, first_bio, logical);
6558 if (dev_nr < total_devs - 1)
6559 bio = btrfs_bio_clone(first_bio);
6563 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6564 dev_nr, async_submit);
6566 btrfs_bio_counter_dec(fs_info);
6571 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6574 * If devid and uuid are both specified, the match must be exact, otherwise
6575 * only devid is used.
6577 * If @seed is true, traverse through the seed devices.
6579 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6580 u64 devid, u8 *uuid, u8 *fsid,
6583 struct btrfs_device *device;
6585 while (fs_devices) {
6587 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6588 list_for_each_entry(device, &fs_devices->devices,
6590 if (device->devid == devid &&
6591 (!uuid || memcmp(device->uuid, uuid,
6592 BTRFS_UUID_SIZE) == 0))
6597 fs_devices = fs_devices->seed;
6604 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6605 u64 devid, u8 *dev_uuid)
6607 struct btrfs_device *device;
6609 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6613 list_add(&device->dev_list, &fs_devices->devices);
6614 device->fs_devices = fs_devices;
6615 fs_devices->num_devices++;
6617 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6618 fs_devices->missing_devices++;
6624 * btrfs_alloc_device - allocate struct btrfs_device
6625 * @fs_info: used only for generating a new devid, can be NULL if
6626 * devid is provided (i.e. @devid != NULL).
6627 * @devid: a pointer to devid for this device. If NULL a new devid
6629 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6632 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6633 * on error. Returned struct is not linked onto any lists and must be
6634 * destroyed with btrfs_free_device.
6636 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6640 struct btrfs_device *dev;
6643 if (WARN_ON(!devid && !fs_info))
6644 return ERR_PTR(-EINVAL);
6646 dev = __alloc_device();
6655 ret = find_next_devid(fs_info, &tmp);
6657 btrfs_free_device(dev);
6658 return ERR_PTR(ret);
6664 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6666 generate_random_uuid(dev->uuid);
6668 btrfs_init_work(&dev->work, btrfs_submit_helper,
6669 pending_bios_fn, NULL, NULL);
6674 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6675 u64 devid, u8 *uuid, bool error)
6678 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6681 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6685 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6687 int index = btrfs_bg_flags_to_raid_index(type);
6688 int ncopies = btrfs_raid_array[index].ncopies;
6691 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6692 case BTRFS_BLOCK_GROUP_RAID5:
6693 data_stripes = num_stripes - 1;
6695 case BTRFS_BLOCK_GROUP_RAID6:
6696 data_stripes = num_stripes - 2;
6699 data_stripes = num_stripes / ncopies;
6702 return div_u64(chunk_len, data_stripes);
6705 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6706 struct btrfs_chunk *chunk)
6708 struct btrfs_fs_info *fs_info = leaf->fs_info;
6709 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6710 struct map_lookup *map;
6711 struct extent_map *em;
6715 u8 uuid[BTRFS_UUID_SIZE];
6720 logical = key->offset;
6721 length = btrfs_chunk_length(leaf, chunk);
6722 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6725 * Only need to verify chunk item if we're reading from sys chunk array,
6726 * as chunk item in tree block is already verified by tree-checker.
6728 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6729 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6734 read_lock(&map_tree->lock);
6735 em = lookup_extent_mapping(map_tree, logical, 1);
6736 read_unlock(&map_tree->lock);
6738 /* already mapped? */
6739 if (em && em->start <= logical && em->start + em->len > logical) {
6740 free_extent_map(em);
6743 free_extent_map(em);
6746 em = alloc_extent_map();
6749 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6751 free_extent_map(em);
6755 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6756 em->map_lookup = map;
6757 em->start = logical;
6760 em->block_start = 0;
6761 em->block_len = em->len;
6763 map->num_stripes = num_stripes;
6764 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6765 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6766 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6767 map->type = btrfs_chunk_type(leaf, chunk);
6768 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6769 map->verified_stripes = 0;
6770 em->orig_block_len = calc_stripe_length(map->type, em->len,
6772 for (i = 0; i < num_stripes; i++) {
6773 map->stripes[i].physical =
6774 btrfs_stripe_offset_nr(leaf, chunk, i);
6775 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6776 read_extent_buffer(leaf, uuid, (unsigned long)
6777 btrfs_stripe_dev_uuid_nr(chunk, i),
6779 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6780 devid, uuid, NULL, true);
6781 if (!map->stripes[i].dev &&
6782 !btrfs_test_opt(fs_info, DEGRADED)) {
6783 free_extent_map(em);
6784 btrfs_report_missing_device(fs_info, devid, uuid, true);
6787 if (!map->stripes[i].dev) {
6788 map->stripes[i].dev =
6789 add_missing_dev(fs_info->fs_devices, devid,
6791 if (IS_ERR(map->stripes[i].dev)) {
6792 free_extent_map(em);
6794 "failed to init missing dev %llu: %ld",
6795 devid, PTR_ERR(map->stripes[i].dev));
6796 return PTR_ERR(map->stripes[i].dev);
6798 btrfs_report_missing_device(fs_info, devid, uuid, false);
6800 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6801 &(map->stripes[i].dev->dev_state));
6805 write_lock(&map_tree->lock);
6806 ret = add_extent_mapping(map_tree, em, 0);
6807 write_unlock(&map_tree->lock);
6810 "failed to add chunk map, start=%llu len=%llu: %d",
6811 em->start, em->len, ret);
6813 free_extent_map(em);
6818 static void fill_device_from_item(struct extent_buffer *leaf,
6819 struct btrfs_dev_item *dev_item,
6820 struct btrfs_device *device)
6824 device->devid = btrfs_device_id(leaf, dev_item);
6825 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6826 device->total_bytes = device->disk_total_bytes;
6827 device->commit_total_bytes = device->disk_total_bytes;
6828 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6829 device->commit_bytes_used = device->bytes_used;
6830 device->type = btrfs_device_type(leaf, dev_item);
6831 device->io_align = btrfs_device_io_align(leaf, dev_item);
6832 device->io_width = btrfs_device_io_width(leaf, dev_item);
6833 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6834 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6835 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6837 ptr = btrfs_device_uuid(dev_item);
6838 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6841 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6844 struct btrfs_fs_devices *fs_devices;
6847 lockdep_assert_held(&uuid_mutex);
6850 fs_devices = fs_info->fs_devices->seed;
6851 while (fs_devices) {
6852 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6855 fs_devices = fs_devices->seed;
6858 fs_devices = find_fsid(fsid, NULL);
6860 if (!btrfs_test_opt(fs_info, DEGRADED))
6861 return ERR_PTR(-ENOENT);
6863 fs_devices = alloc_fs_devices(fsid, NULL);
6864 if (IS_ERR(fs_devices))
6867 fs_devices->seeding = 1;
6868 fs_devices->opened = 1;
6872 fs_devices = clone_fs_devices(fs_devices);
6873 if (IS_ERR(fs_devices))
6876 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6878 free_fs_devices(fs_devices);
6879 fs_devices = ERR_PTR(ret);
6883 if (!fs_devices->seeding) {
6884 close_fs_devices(fs_devices);
6885 free_fs_devices(fs_devices);
6886 fs_devices = ERR_PTR(-EINVAL);
6890 fs_devices->seed = fs_info->fs_devices->seed;
6891 fs_info->fs_devices->seed = fs_devices;
6896 static int read_one_dev(struct extent_buffer *leaf,
6897 struct btrfs_dev_item *dev_item)
6899 struct btrfs_fs_info *fs_info = leaf->fs_info;
6900 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6901 struct btrfs_device *device;
6904 u8 fs_uuid[BTRFS_FSID_SIZE];
6905 u8 dev_uuid[BTRFS_UUID_SIZE];
6907 devid = btrfs_device_id(leaf, dev_item);
6908 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6910 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6913 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6914 fs_devices = open_seed_devices(fs_info, fs_uuid);
6915 if (IS_ERR(fs_devices))
6916 return PTR_ERR(fs_devices);
6919 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6922 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6923 btrfs_report_missing_device(fs_info, devid,
6928 device = add_missing_dev(fs_devices, devid, dev_uuid);
6929 if (IS_ERR(device)) {
6931 "failed to add missing dev %llu: %ld",
6932 devid, PTR_ERR(device));
6933 return PTR_ERR(device);
6935 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6937 if (!device->bdev) {
6938 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6939 btrfs_report_missing_device(fs_info,
6940 devid, dev_uuid, true);
6943 btrfs_report_missing_device(fs_info, devid,
6947 if (!device->bdev &&
6948 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6950 * this happens when a device that was properly setup
6951 * in the device info lists suddenly goes bad.
6952 * device->bdev is NULL, and so we have to set
6953 * device->missing to one here
6955 device->fs_devices->missing_devices++;
6956 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6959 /* Move the device to its own fs_devices */
6960 if (device->fs_devices != fs_devices) {
6961 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6962 &device->dev_state));
6964 list_move(&device->dev_list, &fs_devices->devices);
6965 device->fs_devices->num_devices--;
6966 fs_devices->num_devices++;
6968 device->fs_devices->missing_devices--;
6969 fs_devices->missing_devices++;
6971 device->fs_devices = fs_devices;
6975 if (device->fs_devices != fs_info->fs_devices) {
6976 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6977 if (device->generation !=
6978 btrfs_device_generation(leaf, dev_item))
6982 fill_device_from_item(leaf, dev_item, device);
6983 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6984 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6985 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6986 device->fs_devices->total_rw_bytes += device->total_bytes;
6987 atomic64_add(device->total_bytes - device->bytes_used,
6988 &fs_info->free_chunk_space);
6994 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6996 struct btrfs_root *root = fs_info->tree_root;
6997 struct btrfs_super_block *super_copy = fs_info->super_copy;
6998 struct extent_buffer *sb;
6999 struct btrfs_disk_key *disk_key;
7000 struct btrfs_chunk *chunk;
7002 unsigned long sb_array_offset;
7009 struct btrfs_key key;
7011 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7013 * This will create extent buffer of nodesize, superblock size is
7014 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7015 * overallocate but we can keep it as-is, only the first page is used.
7017 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7020 set_extent_buffer_uptodate(sb);
7021 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7023 * The sb extent buffer is artificial and just used to read the system array.
7024 * set_extent_buffer_uptodate() call does not properly mark all it's
7025 * pages up-to-date when the page is larger: extent does not cover the
7026 * whole page and consequently check_page_uptodate does not find all
7027 * the page's extents up-to-date (the hole beyond sb),
7028 * write_extent_buffer then triggers a WARN_ON.
7030 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7031 * but sb spans only this function. Add an explicit SetPageUptodate call
7032 * to silence the warning eg. on PowerPC 64.
7034 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7035 SetPageUptodate(sb->pages[0]);
7037 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7038 array_size = btrfs_super_sys_array_size(super_copy);
7040 array_ptr = super_copy->sys_chunk_array;
7041 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7044 while (cur_offset < array_size) {
7045 disk_key = (struct btrfs_disk_key *)array_ptr;
7046 len = sizeof(*disk_key);
7047 if (cur_offset + len > array_size)
7048 goto out_short_read;
7050 btrfs_disk_key_to_cpu(&key, disk_key);
7053 sb_array_offset += len;
7056 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7057 chunk = (struct btrfs_chunk *)sb_array_offset;
7059 * At least one btrfs_chunk with one stripe must be
7060 * present, exact stripe count check comes afterwards
7062 len = btrfs_chunk_item_size(1);
7063 if (cur_offset + len > array_size)
7064 goto out_short_read;
7066 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7069 "invalid number of stripes %u in sys_array at offset %u",
7070 num_stripes, cur_offset);
7075 type = btrfs_chunk_type(sb, chunk);
7076 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7078 "invalid chunk type %llu in sys_array at offset %u",
7084 len = btrfs_chunk_item_size(num_stripes);
7085 if (cur_offset + len > array_size)
7086 goto out_short_read;
7088 ret = read_one_chunk(&key, sb, chunk);
7093 "unexpected item type %u in sys_array at offset %u",
7094 (u32)key.type, cur_offset);
7099 sb_array_offset += len;
7102 clear_extent_buffer_uptodate(sb);
7103 free_extent_buffer_stale(sb);
7107 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7109 clear_extent_buffer_uptodate(sb);
7110 free_extent_buffer_stale(sb);
7115 * Check if all chunks in the fs are OK for read-write degraded mount
7117 * If the @failing_dev is specified, it's accounted as missing.
7119 * Return true if all chunks meet the minimal RW mount requirements.
7120 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7122 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7123 struct btrfs_device *failing_dev)
7125 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7126 struct extent_map *em;
7130 read_lock(&map_tree->lock);
7131 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7132 read_unlock(&map_tree->lock);
7133 /* No chunk at all? Return false anyway */
7139 struct map_lookup *map;
7144 map = em->map_lookup;
7146 btrfs_get_num_tolerated_disk_barrier_failures(
7148 for (i = 0; i < map->num_stripes; i++) {
7149 struct btrfs_device *dev = map->stripes[i].dev;
7151 if (!dev || !dev->bdev ||
7152 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7153 dev->last_flush_error)
7155 else if (failing_dev && failing_dev == dev)
7158 if (missing > max_tolerated) {
7161 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7162 em->start, missing, max_tolerated);
7163 free_extent_map(em);
7167 next_start = extent_map_end(em);
7168 free_extent_map(em);
7170 read_lock(&map_tree->lock);
7171 em = lookup_extent_mapping(map_tree, next_start,
7172 (u64)(-1) - next_start);
7173 read_unlock(&map_tree->lock);
7179 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7181 struct btrfs_root *root = fs_info->chunk_root;
7182 struct btrfs_path *path;
7183 struct extent_buffer *leaf;
7184 struct btrfs_key key;
7185 struct btrfs_key found_key;
7190 path = btrfs_alloc_path();
7195 * uuid_mutex is needed only if we are mounting a sprout FS
7196 * otherwise we don't need it.
7198 mutex_lock(&uuid_mutex);
7199 mutex_lock(&fs_info->chunk_mutex);
7202 * Read all device items, and then all the chunk items. All
7203 * device items are found before any chunk item (their object id
7204 * is smaller than the lowest possible object id for a chunk
7205 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7207 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7210 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7214 leaf = path->nodes[0];
7215 slot = path->slots[0];
7216 if (slot >= btrfs_header_nritems(leaf)) {
7217 ret = btrfs_next_leaf(root, path);
7224 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7225 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7226 struct btrfs_dev_item *dev_item;
7227 dev_item = btrfs_item_ptr(leaf, slot,
7228 struct btrfs_dev_item);
7229 ret = read_one_dev(leaf, dev_item);
7233 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7234 struct btrfs_chunk *chunk;
7235 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7236 ret = read_one_chunk(&found_key, leaf, chunk);
7244 * After loading chunk tree, we've got all device information,
7245 * do another round of validation checks.
7247 if (total_dev != fs_info->fs_devices->total_devices) {
7249 "super_num_devices %llu mismatch with num_devices %llu found here",
7250 btrfs_super_num_devices(fs_info->super_copy),
7255 if (btrfs_super_total_bytes(fs_info->super_copy) <
7256 fs_info->fs_devices->total_rw_bytes) {
7258 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7259 btrfs_super_total_bytes(fs_info->super_copy),
7260 fs_info->fs_devices->total_rw_bytes);
7266 mutex_unlock(&fs_info->chunk_mutex);
7267 mutex_unlock(&uuid_mutex);
7269 btrfs_free_path(path);
7273 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7275 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7276 struct btrfs_device *device;
7278 while (fs_devices) {
7279 mutex_lock(&fs_devices->device_list_mutex);
7280 list_for_each_entry(device, &fs_devices->devices, dev_list)
7281 device->fs_info = fs_info;
7282 mutex_unlock(&fs_devices->device_list_mutex);
7284 fs_devices = fs_devices->seed;
7288 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7289 const struct btrfs_dev_stats_item *ptr,
7294 read_extent_buffer(eb, &val,
7295 offsetof(struct btrfs_dev_stats_item, values) +
7296 ((unsigned long)ptr) + (index * sizeof(u64)),
7301 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7302 struct btrfs_dev_stats_item *ptr,
7305 write_extent_buffer(eb, &val,
7306 offsetof(struct btrfs_dev_stats_item, values) +
7307 ((unsigned long)ptr) + (index * sizeof(u64)),
7311 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7313 struct btrfs_key key;
7314 struct btrfs_root *dev_root = fs_info->dev_root;
7315 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7316 struct extent_buffer *eb;
7319 struct btrfs_device *device;
7320 struct btrfs_path *path = NULL;
7323 path = btrfs_alloc_path();
7327 mutex_lock(&fs_devices->device_list_mutex);
7328 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7330 struct btrfs_dev_stats_item *ptr;
7332 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7333 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7334 key.offset = device->devid;
7335 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7337 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7338 btrfs_dev_stat_set(device, i, 0);
7339 device->dev_stats_valid = 1;
7340 btrfs_release_path(path);
7343 slot = path->slots[0];
7344 eb = path->nodes[0];
7345 item_size = btrfs_item_size_nr(eb, slot);
7347 ptr = btrfs_item_ptr(eb, slot,
7348 struct btrfs_dev_stats_item);
7350 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7351 if (item_size >= (1 + i) * sizeof(__le64))
7352 btrfs_dev_stat_set(device, i,
7353 btrfs_dev_stats_value(eb, ptr, i));
7355 btrfs_dev_stat_set(device, i, 0);
7358 device->dev_stats_valid = 1;
7359 btrfs_dev_stat_print_on_load(device);
7360 btrfs_release_path(path);
7362 mutex_unlock(&fs_devices->device_list_mutex);
7364 btrfs_free_path(path);
7365 return ret < 0 ? ret : 0;
7368 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7369 struct btrfs_device *device)
7371 struct btrfs_fs_info *fs_info = trans->fs_info;
7372 struct btrfs_root *dev_root = fs_info->dev_root;
7373 struct btrfs_path *path;
7374 struct btrfs_key key;
7375 struct extent_buffer *eb;
7376 struct btrfs_dev_stats_item *ptr;
7380 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7381 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7382 key.offset = device->devid;
7384 path = btrfs_alloc_path();
7387 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7389 btrfs_warn_in_rcu(fs_info,
7390 "error %d while searching for dev_stats item for device %s",
7391 ret, rcu_str_deref(device->name));
7396 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7397 /* need to delete old one and insert a new one */
7398 ret = btrfs_del_item(trans, dev_root, path);
7400 btrfs_warn_in_rcu(fs_info,
7401 "delete too small dev_stats item for device %s failed %d",
7402 rcu_str_deref(device->name), ret);
7409 /* need to insert a new item */
7410 btrfs_release_path(path);
7411 ret = btrfs_insert_empty_item(trans, dev_root, path,
7412 &key, sizeof(*ptr));
7414 btrfs_warn_in_rcu(fs_info,
7415 "insert dev_stats item for device %s failed %d",
7416 rcu_str_deref(device->name), ret);
7421 eb = path->nodes[0];
7422 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7423 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7424 btrfs_set_dev_stats_value(eb, ptr, i,
7425 btrfs_dev_stat_read(device, i));
7426 btrfs_mark_buffer_dirty(eb);
7429 btrfs_free_path(path);
7434 * called from commit_transaction. Writes all changed device stats to disk.
7436 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7438 struct btrfs_fs_info *fs_info = trans->fs_info;
7439 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7440 struct btrfs_device *device;
7444 mutex_lock(&fs_devices->device_list_mutex);
7445 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7446 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7447 if (!device->dev_stats_valid || stats_cnt == 0)
7452 * There is a LOAD-LOAD control dependency between the value of
7453 * dev_stats_ccnt and updating the on-disk values which requires
7454 * reading the in-memory counters. Such control dependencies
7455 * require explicit read memory barriers.
7457 * This memory barriers pairs with smp_mb__before_atomic in
7458 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7459 * barrier implied by atomic_xchg in
7460 * btrfs_dev_stats_read_and_reset
7464 ret = update_dev_stat_item(trans, device);
7466 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7468 mutex_unlock(&fs_devices->device_list_mutex);
7473 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7475 btrfs_dev_stat_inc(dev, index);
7476 btrfs_dev_stat_print_on_error(dev);
7479 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7481 if (!dev->dev_stats_valid)
7483 btrfs_err_rl_in_rcu(dev->fs_info,
7484 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7485 rcu_str_deref(dev->name),
7486 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7487 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7488 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7489 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7490 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7493 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7497 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7498 if (btrfs_dev_stat_read(dev, i) != 0)
7500 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7501 return; /* all values == 0, suppress message */
7503 btrfs_info_in_rcu(dev->fs_info,
7504 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7505 rcu_str_deref(dev->name),
7506 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7507 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7508 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7509 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7510 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7513 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7514 struct btrfs_ioctl_get_dev_stats *stats)
7516 struct btrfs_device *dev;
7517 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7520 mutex_lock(&fs_devices->device_list_mutex);
7521 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7523 mutex_unlock(&fs_devices->device_list_mutex);
7526 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7528 } else if (!dev->dev_stats_valid) {
7529 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7531 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7532 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7533 if (stats->nr_items > i)
7535 btrfs_dev_stat_read_and_reset(dev, i);
7537 btrfs_dev_stat_set(dev, i, 0);
7540 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7541 if (stats->nr_items > i)
7542 stats->values[i] = btrfs_dev_stat_read(dev, i);
7544 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7545 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7549 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7551 struct buffer_head *bh;
7552 struct btrfs_super_block *disk_super;
7558 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7561 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7564 disk_super = (struct btrfs_super_block *)bh->b_data;
7566 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7567 set_buffer_dirty(bh);
7568 sync_dirty_buffer(bh);
7572 /* Notify udev that device has changed */
7573 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7575 /* Update ctime/mtime for device path for libblkid */
7576 update_dev_time(device_path);
7580 * Update the size and bytes used for each device where it changed. This is
7581 * delayed since we would otherwise get errors while writing out the
7584 * Must be invoked during transaction commit.
7586 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7588 struct btrfs_device *curr, *next;
7590 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7592 if (list_empty(&trans->dev_update_list))
7596 * We don't need the device_list_mutex here. This list is owned by the
7597 * transaction and the transaction must complete before the device is
7600 mutex_lock(&trans->fs_info->chunk_mutex);
7601 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7603 list_del_init(&curr->post_commit_list);
7604 curr->commit_total_bytes = curr->disk_total_bytes;
7605 curr->commit_bytes_used = curr->bytes_used;
7607 mutex_unlock(&trans->fs_info->chunk_mutex);
7610 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7612 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7613 while (fs_devices) {
7614 fs_devices->fs_info = fs_info;
7615 fs_devices = fs_devices->seed;
7619 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7621 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7622 while (fs_devices) {
7623 fs_devices->fs_info = NULL;
7624 fs_devices = fs_devices->seed;
7629 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7631 int btrfs_bg_type_to_factor(u64 flags)
7633 const int index = btrfs_bg_flags_to_raid_index(flags);
7635 return btrfs_raid_array[index].ncopies;
7640 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7641 u64 chunk_offset, u64 devid,
7642 u64 physical_offset, u64 physical_len)
7644 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7645 struct extent_map *em;
7646 struct map_lookup *map;
7647 struct btrfs_device *dev;
7653 read_lock(&em_tree->lock);
7654 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7655 read_unlock(&em_tree->lock);
7659 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7660 physical_offset, devid);
7665 map = em->map_lookup;
7666 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7667 if (physical_len != stripe_len) {
7669 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7670 physical_offset, devid, em->start, physical_len,
7676 for (i = 0; i < map->num_stripes; i++) {
7677 if (map->stripes[i].dev->devid == devid &&
7678 map->stripes[i].physical == physical_offset) {
7680 if (map->verified_stripes >= map->num_stripes) {
7682 "too many dev extents for chunk %llu found",
7687 map->verified_stripes++;
7693 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7694 physical_offset, devid);
7698 /* Make sure no dev extent is beyond device bondary */
7699 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7701 btrfs_err(fs_info, "failed to find devid %llu", devid);
7706 /* It's possible this device is a dummy for seed device */
7707 if (dev->disk_total_bytes == 0) {
7708 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7711 btrfs_err(fs_info, "failed to find seed devid %llu",
7718 if (physical_offset + physical_len > dev->disk_total_bytes) {
7720 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7721 devid, physical_offset, physical_len,
7722 dev->disk_total_bytes);
7727 free_extent_map(em);
7731 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7733 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7734 struct extent_map *em;
7735 struct rb_node *node;
7738 read_lock(&em_tree->lock);
7739 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7740 em = rb_entry(node, struct extent_map, rb_node);
7741 if (em->map_lookup->num_stripes !=
7742 em->map_lookup->verified_stripes) {
7744 "chunk %llu has missing dev extent, have %d expect %d",
7745 em->start, em->map_lookup->verified_stripes,
7746 em->map_lookup->num_stripes);
7752 read_unlock(&em_tree->lock);
7757 * Ensure that all dev extents are mapped to correct chunk, otherwise
7758 * later chunk allocation/free would cause unexpected behavior.
7760 * NOTE: This will iterate through the whole device tree, which should be of
7761 * the same size level as the chunk tree. This slightly increases mount time.
7763 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7765 struct btrfs_path *path;
7766 struct btrfs_root *root = fs_info->dev_root;
7767 struct btrfs_key key;
7769 u64 prev_dev_ext_end = 0;
7773 key.type = BTRFS_DEV_EXTENT_KEY;
7776 path = btrfs_alloc_path();
7780 path->reada = READA_FORWARD;
7781 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7785 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7786 ret = btrfs_next_item(root, path);
7789 /* No dev extents at all? Not good */
7796 struct extent_buffer *leaf = path->nodes[0];
7797 struct btrfs_dev_extent *dext;
7798 int slot = path->slots[0];
7800 u64 physical_offset;
7804 btrfs_item_key_to_cpu(leaf, &key, slot);
7805 if (key.type != BTRFS_DEV_EXTENT_KEY)
7807 devid = key.objectid;
7808 physical_offset = key.offset;
7810 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7811 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7812 physical_len = btrfs_dev_extent_length(leaf, dext);
7814 /* Check if this dev extent overlaps with the previous one */
7815 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7817 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7818 devid, physical_offset, prev_dev_ext_end);
7823 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7824 physical_offset, physical_len);
7828 prev_dev_ext_end = physical_offset + physical_len;
7830 ret = btrfs_next_item(root, path);
7839 /* Ensure all chunks have corresponding dev extents */
7840 ret = verify_chunk_dev_extent_mapping(fs_info);
7842 btrfs_free_path(path);
7847 * Check whether the given block group or device is pinned by any inode being
7848 * used as a swapfile.
7850 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7852 struct btrfs_swapfile_pin *sp;
7853 struct rb_node *node;
7855 spin_lock(&fs_info->swapfile_pins_lock);
7856 node = fs_info->swapfile_pins.rb_node;
7858 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7860 node = node->rb_left;
7861 else if (ptr > sp->ptr)
7862 node = node->rb_right;
7866 spin_unlock(&fs_info->swapfile_pins_lock);
7867 return node != NULL;