1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
33 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
34 [BTRFS_RAID_RAID10] = {
37 .devs_max = 0, /* 0 == as many as possible */
39 .tolerated_failures = 1,
43 .raid_name = "raid10",
44 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
45 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
47 [BTRFS_RAID_RAID1] = {
52 .tolerated_failures = 1,
57 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
58 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
65 .tolerated_failures = 0,
70 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
73 [BTRFS_RAID_RAID0] = {
78 .tolerated_failures = 0,
83 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
86 [BTRFS_RAID_SINGLE] = {
91 .tolerated_failures = 0,
95 .raid_name = "single",
99 [BTRFS_RAID_RAID5] = {
104 .tolerated_failures = 1,
108 .raid_name = "raid5",
109 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
110 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
112 [BTRFS_RAID_RAID6] = {
117 .tolerated_failures = 2,
121 .raid_name = "raid6",
122 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
123 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
127 const char *btrfs_bg_type_to_raid_name(u64 flags)
129 const int index = btrfs_bg_flags_to_raid_index(flags);
131 if (index >= BTRFS_NR_RAID_TYPES)
134 return btrfs_raid_array[index].raid_name;
138 * Fill @buf with textual description of @bg_flags, no more than @size_buf
139 * bytes including terminating null byte.
141 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
146 u64 flags = bg_flags;
147 u32 size_bp = size_buf;
154 #define DESCRIBE_FLAG(flag, desc) \
156 if (flags & (flag)) { \
157 ret = snprintf(bp, size_bp, "%s|", (desc)); \
158 if (ret < 0 || ret >= size_bp) \
166 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
167 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
168 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
170 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
171 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
172 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
173 btrfs_raid_array[i].raid_name);
177 ret = snprintf(bp, size_bp, "0x%llx|", flags);
181 if (size_bp < size_buf)
182 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
185 * The text is trimmed, it's up to the caller to provide sufficiently
191 static int init_first_rw_device(struct btrfs_trans_handle *trans);
192 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
193 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
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 static void btrfs_kobject_uevent(struct block_device *bdev,
362 enum kobject_action action)
366 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
368 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
370 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
371 &disk_to_dev(bdev->bd_disk)->kobj);
374 void __exit btrfs_cleanup_fs_uuids(void)
376 struct btrfs_fs_devices *fs_devices;
378 while (!list_empty(&fs_uuids)) {
379 fs_devices = list_entry(fs_uuids.next,
380 struct btrfs_fs_devices, fs_list);
381 list_del(&fs_devices->fs_list);
382 free_fs_devices(fs_devices);
387 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
388 * Returned struct is not linked onto any lists and must be destroyed using
391 static struct btrfs_device *__alloc_device(void)
393 struct btrfs_device *dev;
395 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
397 return ERR_PTR(-ENOMEM);
400 * Preallocate a bio that's always going to be used for flushing device
401 * barriers and matches the device lifespan
403 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
404 if (!dev->flush_bio) {
406 return ERR_PTR(-ENOMEM);
409 INIT_LIST_HEAD(&dev->dev_list);
410 INIT_LIST_HEAD(&dev->dev_alloc_list);
411 INIT_LIST_HEAD(&dev->post_commit_list);
413 spin_lock_init(&dev->io_lock);
415 atomic_set(&dev->reada_in_flight, 0);
416 atomic_set(&dev->dev_stats_ccnt, 0);
417 btrfs_device_data_ordered_init(dev);
418 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
419 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
420 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
425 static noinline struct btrfs_fs_devices *find_fsid(
426 const u8 *fsid, const u8 *metadata_fsid)
428 struct btrfs_fs_devices *fs_devices;
434 * Handle scanned device having completed its fsid change but
435 * belonging to a fs_devices that was created by first scanning
436 * a device which didn't have its fsid/metadata_uuid changed
437 * at all and the CHANGING_FSID_V2 flag set.
439 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
440 if (fs_devices->fsid_change &&
441 memcmp(metadata_fsid, fs_devices->fsid,
442 BTRFS_FSID_SIZE) == 0 &&
443 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
444 BTRFS_FSID_SIZE) == 0) {
449 * Handle scanned device having completed its fsid change but
450 * belonging to a fs_devices that was created by a device that
451 * has an outdated pair of fsid/metadata_uuid and
452 * CHANGING_FSID_V2 flag set.
454 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
455 if (fs_devices->fsid_change &&
456 memcmp(fs_devices->metadata_uuid,
457 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
458 memcmp(metadata_fsid, fs_devices->metadata_uuid,
459 BTRFS_FSID_SIZE) == 0) {
465 /* Handle non-split brain cases */
466 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
468 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
469 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
470 BTRFS_FSID_SIZE) == 0)
473 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
481 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
482 int flush, struct block_device **bdev,
483 struct buffer_head **bh)
487 *bdev = blkdev_get_by_path(device_path, flags, holder);
490 ret = PTR_ERR(*bdev);
495 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
496 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
498 blkdev_put(*bdev, flags);
501 invalidate_bdev(*bdev);
502 *bh = btrfs_read_dev_super(*bdev);
505 blkdev_put(*bdev, flags);
517 static void requeue_list(struct btrfs_pending_bios *pending_bios,
518 struct bio *head, struct bio *tail)
521 struct bio *old_head;
523 old_head = pending_bios->head;
524 pending_bios->head = head;
525 if (pending_bios->tail)
526 tail->bi_next = old_head;
528 pending_bios->tail = tail;
532 * we try to collect pending bios for a device so we don't get a large
533 * number of procs sending bios down to the same device. This greatly
534 * improves the schedulers ability to collect and merge the bios.
536 * But, it also turns into a long list of bios to process and that is sure
537 * to eventually make the worker thread block. The solution here is to
538 * make some progress and then put this work struct back at the end of
539 * the list if the block device is congested. This way, multiple devices
540 * can make progress from a single worker thread.
542 static noinline void run_scheduled_bios(struct btrfs_device *device)
544 struct btrfs_fs_info *fs_info = device->fs_info;
546 struct backing_dev_info *bdi;
547 struct btrfs_pending_bios *pending_bios;
551 unsigned long num_run;
552 unsigned long batch_run = 0;
553 unsigned long last_waited = 0;
555 int sync_pending = 0;
556 struct blk_plug plug;
559 * this function runs all the bios we've collected for
560 * a particular device. We don't want to wander off to
561 * another device without first sending all of these down.
562 * So, setup a plug here and finish it off before we return
564 blk_start_plug(&plug);
566 bdi = device->bdev->bd_bdi;
569 spin_lock(&device->io_lock);
574 /* take all the bios off the list at once and process them
575 * later on (without the lock held). But, remember the
576 * tail and other pointers so the bios can be properly reinserted
577 * into the list if we hit congestion
579 if (!force_reg && device->pending_sync_bios.head) {
580 pending_bios = &device->pending_sync_bios;
583 pending_bios = &device->pending_bios;
587 pending = pending_bios->head;
588 tail = pending_bios->tail;
589 WARN_ON(pending && !tail);
592 * if pending was null this time around, no bios need processing
593 * at all and we can stop. Otherwise it'll loop back up again
594 * and do an additional check so no bios are missed.
596 * device->running_pending is used to synchronize with the
599 if (device->pending_sync_bios.head == NULL &&
600 device->pending_bios.head == NULL) {
602 device->running_pending = 0;
605 device->running_pending = 1;
608 pending_bios->head = NULL;
609 pending_bios->tail = NULL;
611 spin_unlock(&device->io_lock);
616 /* we want to work on both lists, but do more bios on the
617 * sync list than the regular list
620 pending_bios != &device->pending_sync_bios &&
621 device->pending_sync_bios.head) ||
622 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
623 device->pending_bios.head)) {
624 spin_lock(&device->io_lock);
625 requeue_list(pending_bios, pending, tail);
630 pending = pending->bi_next;
633 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
636 * if we're doing the sync list, record that our
637 * plug has some sync requests on it
639 * If we're doing the regular list and there are
640 * sync requests sitting around, unplug before
643 if (pending_bios == &device->pending_sync_bios) {
645 } else if (sync_pending) {
646 blk_finish_plug(&plug);
647 blk_start_plug(&plug);
651 btrfsic_submit_bio(cur);
658 * we made progress, there is more work to do and the bdi
659 * is now congested. Back off and let other work structs
662 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
663 fs_info->fs_devices->open_devices > 1) {
664 struct io_context *ioc;
666 ioc = current->io_context;
669 * the main goal here is that we don't want to
670 * block if we're going to be able to submit
671 * more requests without blocking.
673 * This code does two great things, it pokes into
674 * the elevator code from a filesystem _and_
675 * it makes assumptions about how batching works.
677 if (ioc && ioc->nr_batch_requests > 0 &&
678 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
680 ioc->last_waited == last_waited)) {
682 * we want to go through our batch of
683 * requests and stop. So, we copy out
684 * the ioc->last_waited time and test
685 * against it before looping
687 last_waited = ioc->last_waited;
691 spin_lock(&device->io_lock);
692 requeue_list(pending_bios, pending, tail);
693 device->running_pending = 1;
695 spin_unlock(&device->io_lock);
696 btrfs_queue_work(fs_info->submit_workers,
706 spin_lock(&device->io_lock);
707 if (device->pending_bios.head || device->pending_sync_bios.head)
709 spin_unlock(&device->io_lock);
712 blk_finish_plug(&plug);
715 static void pending_bios_fn(struct btrfs_work *work)
717 struct btrfs_device *device;
719 device = container_of(work, struct btrfs_device, work);
720 run_scheduled_bios(device);
723 static bool device_path_matched(const char *path, struct btrfs_device *device)
728 found = strcmp(rcu_str_deref(device->name), path);
735 * Search and remove all stale (devices which are not mounted) devices.
736 * When both inputs are NULL, it will search and release all stale devices.
737 * path: Optional. When provided will it release all unmounted devices
738 * matching this path only.
739 * skip_dev: Optional. Will skip this device when searching for the stale
741 * Return: 0 for success or if @path is NULL.
742 * -EBUSY if @path is a mounted device.
743 * -ENOENT if @path does not match any device in the list.
745 static int btrfs_free_stale_devices(const char *path,
746 struct btrfs_device *skip_device)
748 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
749 struct btrfs_device *device, *tmp_device;
755 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
757 mutex_lock(&fs_devices->device_list_mutex);
758 list_for_each_entry_safe(device, tmp_device,
759 &fs_devices->devices, dev_list) {
760 if (skip_device && skip_device == device)
762 if (path && !device->name)
764 if (path && !device_path_matched(path, device))
766 if (fs_devices->opened) {
767 /* for an already deleted device return 0 */
768 if (path && ret != 0)
773 /* delete the stale device */
774 fs_devices->num_devices--;
775 list_del(&device->dev_list);
776 btrfs_free_device(device);
779 if (fs_devices->num_devices == 0)
782 mutex_unlock(&fs_devices->device_list_mutex);
784 if (fs_devices->num_devices == 0) {
785 btrfs_sysfs_remove_fsid(fs_devices);
786 list_del(&fs_devices->fs_list);
787 free_fs_devices(fs_devices);
794 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
795 struct btrfs_device *device, fmode_t flags,
798 struct request_queue *q;
799 struct block_device *bdev;
800 struct buffer_head *bh;
801 struct btrfs_super_block *disk_super;
810 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
815 disk_super = (struct btrfs_super_block *)bh->b_data;
816 devid = btrfs_stack_device_id(&disk_super->dev_item);
817 if (devid != device->devid)
820 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
823 device->generation = btrfs_super_generation(disk_super);
825 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
826 if (btrfs_super_incompat_flags(disk_super) &
827 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
829 "BTRFS: Invalid seeding and uuid-changed device detected\n");
833 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
834 fs_devices->seeding = 1;
836 if (bdev_read_only(bdev))
837 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
839 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
842 q = bdev_get_queue(bdev);
843 if (!blk_queue_nonrot(q))
844 fs_devices->rotating = 1;
847 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
848 device->mode = flags;
850 fs_devices->open_devices++;
851 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
852 device->devid != BTRFS_DEV_REPLACE_DEVID) {
853 fs_devices->rw_devices++;
854 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
862 blkdev_put(bdev, flags);
868 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
869 * being created with a disk that has already completed its fsid change.
871 static struct btrfs_fs_devices *find_fsid_inprogress(
872 struct btrfs_super_block *disk_super)
874 struct btrfs_fs_devices *fs_devices;
876 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
877 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
878 BTRFS_FSID_SIZE) != 0 &&
879 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
880 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
889 static struct btrfs_fs_devices *find_fsid_changed(
890 struct btrfs_super_block *disk_super)
892 struct btrfs_fs_devices *fs_devices;
895 * Handles the case where scanned device is part of an fs that had
896 * multiple successful changes of FSID but curently device didn't
897 * observe it. Meaning our fsid will be different than theirs.
899 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
900 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
901 BTRFS_FSID_SIZE) != 0 &&
902 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
903 BTRFS_FSID_SIZE) == 0 &&
904 memcmp(fs_devices->fsid, disk_super->fsid,
905 BTRFS_FSID_SIZE) != 0) {
913 * Add new device to list of registered devices
916 * device pointer which was just added or updated when successful
917 * error pointer when failed
919 static noinline struct btrfs_device *device_list_add(const char *path,
920 struct btrfs_super_block *disk_super,
921 bool *new_device_added)
923 struct btrfs_device *device;
924 struct btrfs_fs_devices *fs_devices = NULL;
925 struct rcu_string *name;
926 u64 found_transid = btrfs_super_generation(disk_super);
927 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
928 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
929 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
930 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
931 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
933 if (fsid_change_in_progress) {
934 if (!has_metadata_uuid) {
936 * When we have an image which has CHANGING_FSID_V2 set
937 * it might belong to either a filesystem which has
938 * disks with completed fsid change or it might belong
939 * to fs with no UUID changes in effect, handle both.
941 fs_devices = find_fsid_inprogress(disk_super);
943 fs_devices = find_fsid(disk_super->fsid, NULL);
945 fs_devices = find_fsid_changed(disk_super);
947 } else if (has_metadata_uuid) {
948 fs_devices = find_fsid(disk_super->fsid,
949 disk_super->metadata_uuid);
951 fs_devices = find_fsid(disk_super->fsid, NULL);
956 if (has_metadata_uuid)
957 fs_devices = alloc_fs_devices(disk_super->fsid,
958 disk_super->metadata_uuid);
960 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
962 if (IS_ERR(fs_devices))
963 return ERR_CAST(fs_devices);
965 fs_devices->fsid_change = fsid_change_in_progress;
967 mutex_lock(&fs_devices->device_list_mutex);
968 list_add(&fs_devices->fs_list, &fs_uuids);
972 mutex_lock(&fs_devices->device_list_mutex);
973 device = btrfs_find_device(fs_devices, devid,
974 disk_super->dev_item.uuid, NULL, false);
977 * If this disk has been pulled into an fs devices created by
978 * a device which had the CHANGING_FSID_V2 flag then replace the
979 * metadata_uuid/fsid values of the fs_devices.
981 if (has_metadata_uuid && fs_devices->fsid_change &&
982 found_transid > fs_devices->latest_generation) {
983 memcpy(fs_devices->fsid, disk_super->fsid,
985 memcpy(fs_devices->metadata_uuid,
986 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
988 fs_devices->fsid_change = false;
993 if (fs_devices->opened) {
994 mutex_unlock(&fs_devices->device_list_mutex);
995 return ERR_PTR(-EBUSY);
998 device = btrfs_alloc_device(NULL, &devid,
999 disk_super->dev_item.uuid);
1000 if (IS_ERR(device)) {
1001 mutex_unlock(&fs_devices->device_list_mutex);
1002 /* we can safely leave the fs_devices entry around */
1006 name = rcu_string_strdup(path, GFP_NOFS);
1008 btrfs_free_device(device);
1009 mutex_unlock(&fs_devices->device_list_mutex);
1010 return ERR_PTR(-ENOMEM);
1012 rcu_assign_pointer(device->name, name);
1014 list_add_rcu(&device->dev_list, &fs_devices->devices);
1015 fs_devices->num_devices++;
1017 device->fs_devices = fs_devices;
1018 *new_device_added = true;
1020 if (disk_super->label[0])
1021 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1022 disk_super->label, devid, found_transid, path);
1024 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1025 disk_super->fsid, devid, found_transid, path);
1027 } else if (!device->name || strcmp(device->name->str, path)) {
1029 * When FS is already mounted.
1030 * 1. If you are here and if the device->name is NULL that
1031 * means this device was missing at time of FS mount.
1032 * 2. If you are here and if the device->name is different
1033 * from 'path' that means either
1034 * a. The same device disappeared and reappeared with
1035 * different name. or
1036 * b. The missing-disk-which-was-replaced, has
1039 * We must allow 1 and 2a above. But 2b would be a spurious
1040 * and unintentional.
1042 * Further in case of 1 and 2a above, the disk at 'path'
1043 * would have missed some transaction when it was away and
1044 * in case of 2a the stale bdev has to be updated as well.
1045 * 2b must not be allowed at all time.
1049 * For now, we do allow update to btrfs_fs_device through the
1050 * btrfs dev scan cli after FS has been mounted. We're still
1051 * tracking a problem where systems fail mount by subvolume id
1052 * when we reject replacement on a mounted FS.
1054 if (!fs_devices->opened && found_transid < device->generation) {
1056 * That is if the FS is _not_ mounted and if you
1057 * are here, that means there is more than one
1058 * disk with same uuid and devid.We keep the one
1059 * with larger generation number or the last-in if
1060 * generation are equal.
1062 mutex_unlock(&fs_devices->device_list_mutex);
1063 return ERR_PTR(-EEXIST);
1067 * We are going to replace the device path for a given devid,
1068 * make sure it's the same device if the device is mounted
1071 struct block_device *path_bdev;
1073 path_bdev = lookup_bdev(path);
1074 if (IS_ERR(path_bdev)) {
1075 mutex_unlock(&fs_devices->device_list_mutex);
1076 return ERR_CAST(path_bdev);
1079 if (device->bdev != path_bdev) {
1081 mutex_unlock(&fs_devices->device_list_mutex);
1082 btrfs_warn_in_rcu(device->fs_info,
1083 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1084 disk_super->fsid, devid,
1085 rcu_str_deref(device->name), path);
1086 return ERR_PTR(-EEXIST);
1089 btrfs_info_in_rcu(device->fs_info,
1090 "device fsid %pU devid %llu moved old:%s new:%s",
1091 disk_super->fsid, devid,
1092 rcu_str_deref(device->name), path);
1095 name = rcu_string_strdup(path, GFP_NOFS);
1097 mutex_unlock(&fs_devices->device_list_mutex);
1098 return ERR_PTR(-ENOMEM);
1100 rcu_string_free(device->name);
1101 rcu_assign_pointer(device->name, name);
1102 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1103 fs_devices->missing_devices--;
1104 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1109 * Unmount does not free the btrfs_device struct but would zero
1110 * generation along with most of the other members. So just update
1111 * it back. We need it to pick the disk with largest generation
1114 if (!fs_devices->opened) {
1115 device->generation = found_transid;
1116 fs_devices->latest_generation = max_t(u64, found_transid,
1117 fs_devices->latest_generation);
1120 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1122 mutex_unlock(&fs_devices->device_list_mutex);
1126 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1128 struct btrfs_fs_devices *fs_devices;
1129 struct btrfs_device *device;
1130 struct btrfs_device *orig_dev;
1132 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1133 if (IS_ERR(fs_devices))
1136 mutex_lock(&orig->device_list_mutex);
1137 fs_devices->total_devices = orig->total_devices;
1139 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1140 struct rcu_string *name;
1142 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1148 * This is ok to do without rcu read locked because we hold the
1149 * uuid mutex so nothing we touch in here is going to disappear.
1151 if (orig_dev->name) {
1152 name = rcu_string_strdup(orig_dev->name->str,
1155 btrfs_free_device(device);
1158 rcu_assign_pointer(device->name, name);
1161 list_add(&device->dev_list, &fs_devices->devices);
1162 device->fs_devices = fs_devices;
1163 fs_devices->num_devices++;
1165 mutex_unlock(&orig->device_list_mutex);
1168 mutex_unlock(&orig->device_list_mutex);
1169 free_fs_devices(fs_devices);
1170 return ERR_PTR(-ENOMEM);
1174 * After we have read the system tree and know devids belonging to
1175 * this filesystem, remove the device which does not belong there.
1177 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1179 struct btrfs_device *device, *next;
1180 struct btrfs_device *latest_dev = NULL;
1182 mutex_lock(&uuid_mutex);
1184 /* This is the initialized path, it is safe to release the devices. */
1185 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1186 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1187 &device->dev_state)) {
1188 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1189 &device->dev_state) &&
1191 device->generation > latest_dev->generation)) {
1192 latest_dev = device;
1197 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1199 * In the first step, keep the device which has
1200 * the correct fsid and the devid that is used
1201 * for the dev_replace procedure.
1202 * In the second step, the dev_replace state is
1203 * read from the device tree and it is known
1204 * whether the procedure is really active or
1205 * not, which means whether this device is
1206 * used or whether it should be removed.
1208 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1209 &device->dev_state)) {
1214 blkdev_put(device->bdev, device->mode);
1215 device->bdev = NULL;
1216 fs_devices->open_devices--;
1218 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1219 list_del_init(&device->dev_alloc_list);
1220 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1221 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1222 &device->dev_state))
1223 fs_devices->rw_devices--;
1225 list_del_init(&device->dev_list);
1226 fs_devices->num_devices--;
1227 btrfs_free_device(device);
1230 if (fs_devices->seed) {
1231 fs_devices = fs_devices->seed;
1235 fs_devices->latest_bdev = latest_dev->bdev;
1237 mutex_unlock(&uuid_mutex);
1240 static void btrfs_close_bdev(struct btrfs_device *device)
1245 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1246 sync_blockdev(device->bdev);
1247 invalidate_bdev(device->bdev);
1250 blkdev_put(device->bdev, device->mode);
1253 static void btrfs_close_one_device(struct btrfs_device *device)
1255 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1256 struct btrfs_device *new_device;
1257 struct rcu_string *name;
1260 fs_devices->open_devices--;
1262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1263 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1264 list_del_init(&device->dev_alloc_list);
1265 fs_devices->rw_devices--;
1268 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1269 fs_devices->missing_devices--;
1271 btrfs_close_bdev(device);
1273 new_device = btrfs_alloc_device(NULL, &device->devid,
1275 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1277 /* Safe because we are under uuid_mutex */
1279 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1280 BUG_ON(!name); /* -ENOMEM */
1281 rcu_assign_pointer(new_device->name, name);
1284 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1285 new_device->fs_devices = device->fs_devices;
1288 btrfs_free_device(device);
1291 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1293 struct btrfs_device *device, *tmp;
1295 if (--fs_devices->opened > 0)
1298 mutex_lock(&fs_devices->device_list_mutex);
1299 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1300 btrfs_close_one_device(device);
1302 mutex_unlock(&fs_devices->device_list_mutex);
1304 WARN_ON(fs_devices->open_devices);
1305 WARN_ON(fs_devices->rw_devices);
1306 fs_devices->opened = 0;
1307 fs_devices->seeding = 0;
1312 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1314 struct btrfs_fs_devices *seed_devices = NULL;
1317 mutex_lock(&uuid_mutex);
1318 ret = close_fs_devices(fs_devices);
1319 if (!fs_devices->opened) {
1320 seed_devices = fs_devices->seed;
1321 fs_devices->seed = NULL;
1323 mutex_unlock(&uuid_mutex);
1325 while (seed_devices) {
1326 fs_devices = seed_devices;
1327 seed_devices = fs_devices->seed;
1328 close_fs_devices(fs_devices);
1329 free_fs_devices(fs_devices);
1334 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1335 fmode_t flags, void *holder)
1337 struct btrfs_device *device;
1338 struct btrfs_device *latest_dev = NULL;
1341 flags |= FMODE_EXCL;
1343 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1344 /* Just open everything we can; ignore failures here */
1345 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1349 device->generation > latest_dev->generation)
1350 latest_dev = device;
1352 if (fs_devices->open_devices == 0) {
1356 fs_devices->opened = 1;
1357 fs_devices->latest_bdev = latest_dev->bdev;
1358 fs_devices->total_rw_bytes = 0;
1363 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1365 struct btrfs_device *dev1, *dev2;
1367 dev1 = list_entry(a, struct btrfs_device, dev_list);
1368 dev2 = list_entry(b, struct btrfs_device, dev_list);
1370 if (dev1->devid < dev2->devid)
1372 else if (dev1->devid > dev2->devid)
1377 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1378 fmode_t flags, void *holder)
1382 lockdep_assert_held(&uuid_mutex);
1384 mutex_lock(&fs_devices->device_list_mutex);
1385 if (fs_devices->opened) {
1386 fs_devices->opened++;
1389 list_sort(NULL, &fs_devices->devices, devid_cmp);
1390 ret = open_fs_devices(fs_devices, flags, holder);
1392 mutex_unlock(&fs_devices->device_list_mutex);
1397 static void btrfs_release_disk_super(struct page *page)
1403 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1405 struct btrfs_super_block **disk_super)
1410 /* make sure our super fits in the device */
1411 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1414 /* make sure our super fits in the page */
1415 if (sizeof(**disk_super) > PAGE_SIZE)
1418 /* make sure our super doesn't straddle pages on disk */
1419 index = bytenr >> PAGE_SHIFT;
1420 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1423 /* pull in the page with our super */
1424 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1427 if (IS_ERR_OR_NULL(*page))
1432 /* align our pointer to the offset of the super block */
1433 *disk_super = p + offset_in_page(bytenr);
1435 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1436 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1437 btrfs_release_disk_super(*page);
1441 if ((*disk_super)->label[0] &&
1442 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1443 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1448 int btrfs_forget_devices(const char *path)
1452 mutex_lock(&uuid_mutex);
1453 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1454 mutex_unlock(&uuid_mutex);
1460 * Look for a btrfs signature on a device. This may be called out of the mount path
1461 * and we are not allowed to call set_blocksize during the scan. The superblock
1462 * is read via pagecache
1464 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1467 struct btrfs_super_block *disk_super;
1468 bool new_device_added = false;
1469 struct btrfs_device *device = NULL;
1470 struct block_device *bdev;
1474 lockdep_assert_held(&uuid_mutex);
1477 * we would like to check all the supers, but that would make
1478 * a btrfs mount succeed after a mkfs from a different FS.
1479 * So, we need to add a special mount option to scan for
1480 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1482 bytenr = btrfs_sb_offset(0);
1483 flags |= FMODE_EXCL;
1485 bdev = blkdev_get_by_path(path, flags, holder);
1487 return ERR_CAST(bdev);
1489 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1490 device = ERR_PTR(-EINVAL);
1491 goto error_bdev_put;
1494 device = device_list_add(path, disk_super, &new_device_added);
1495 if (!IS_ERR(device)) {
1496 if (new_device_added)
1497 btrfs_free_stale_devices(path, device);
1500 btrfs_release_disk_super(page);
1503 blkdev_put(bdev, flags);
1509 * Try to find a chunk that intersects [start, start + len] range and when one
1510 * such is found, record the end of it in *start
1512 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1515 u64 physical_start, physical_end;
1517 lockdep_assert_held(&device->fs_info->chunk_mutex);
1519 if (!find_first_extent_bit(&device->alloc_state, *start,
1520 &physical_start, &physical_end,
1521 CHUNK_ALLOCATED, NULL)) {
1523 if (in_range(physical_start, *start, len) ||
1524 in_range(*start, physical_start,
1525 physical_end - physical_start)) {
1526 *start = physical_end + 1;
1535 * find_free_dev_extent_start - find free space in the specified device
1536 * @device: the device which we search the free space in
1537 * @num_bytes: the size of the free space that we need
1538 * @search_start: the position from which to begin the search
1539 * @start: store the start of the free space.
1540 * @len: the size of the free space. that we find, or the size
1541 * of the max free space if we don't find suitable free space
1543 * this uses a pretty simple search, the expectation is that it is
1544 * called very infrequently and that a given device has a small number
1547 * @start is used to store the start of the free space if we find. But if we
1548 * don't find suitable free space, it will be used to store the start position
1549 * of the max free space.
1551 * @len is used to store the size of the free space that we find.
1552 * But if we don't find suitable free space, it is used to store the size of
1553 * the max free space.
1555 * NOTE: This function will search *commit* root of device tree, and does extra
1556 * check to ensure dev extents are not double allocated.
1557 * This makes the function safe to allocate dev extents but may not report
1558 * correct usable device space, as device extent freed in current transaction
1559 * is not reported as avaiable.
1561 static int find_free_dev_extent_start(struct btrfs_device *device,
1562 u64 num_bytes, u64 search_start, u64 *start,
1565 struct btrfs_fs_info *fs_info = device->fs_info;
1566 struct btrfs_root *root = fs_info->dev_root;
1567 struct btrfs_key key;
1568 struct btrfs_dev_extent *dev_extent;
1569 struct btrfs_path *path;
1574 u64 search_end = device->total_bytes;
1577 struct extent_buffer *l;
1580 * We don't want to overwrite the superblock on the drive nor any area
1581 * used by the boot loader (grub for example), so we make sure to start
1582 * at an offset of at least 1MB.
1584 search_start = max_t(u64, search_start, SZ_1M);
1586 path = btrfs_alloc_path();
1590 max_hole_start = search_start;
1594 if (search_start >= search_end ||
1595 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1600 path->reada = READA_FORWARD;
1601 path->search_commit_root = 1;
1602 path->skip_locking = 1;
1604 key.objectid = device->devid;
1605 key.offset = search_start;
1606 key.type = BTRFS_DEV_EXTENT_KEY;
1608 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1612 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1619 slot = path->slots[0];
1620 if (slot >= btrfs_header_nritems(l)) {
1621 ret = btrfs_next_leaf(root, path);
1629 btrfs_item_key_to_cpu(l, &key, slot);
1631 if (key.objectid < device->devid)
1634 if (key.objectid > device->devid)
1637 if (key.type != BTRFS_DEV_EXTENT_KEY)
1640 if (key.offset > search_start) {
1641 hole_size = key.offset - search_start;
1644 * Have to check before we set max_hole_start, otherwise
1645 * we could end up sending back this offset anyway.
1647 if (contains_pending_extent(device, &search_start,
1649 if (key.offset >= search_start)
1650 hole_size = key.offset - search_start;
1655 if (hole_size > max_hole_size) {
1656 max_hole_start = search_start;
1657 max_hole_size = hole_size;
1661 * If this free space is greater than which we need,
1662 * it must be the max free space that we have found
1663 * until now, so max_hole_start must point to the start
1664 * of this free space and the length of this free space
1665 * is stored in max_hole_size. Thus, we return
1666 * max_hole_start and max_hole_size and go back to the
1669 if (hole_size >= num_bytes) {
1675 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1676 extent_end = key.offset + btrfs_dev_extent_length(l,
1678 if (extent_end > search_start)
1679 search_start = extent_end;
1686 * At this point, search_start should be the end of
1687 * allocated dev extents, and when shrinking the device,
1688 * search_end may be smaller than search_start.
1690 if (search_end > search_start) {
1691 hole_size = search_end - search_start;
1693 if (contains_pending_extent(device, &search_start, hole_size)) {
1694 btrfs_release_path(path);
1698 if (hole_size > max_hole_size) {
1699 max_hole_start = search_start;
1700 max_hole_size = hole_size;
1705 if (max_hole_size < num_bytes)
1711 btrfs_free_path(path);
1712 *start = max_hole_start;
1714 *len = max_hole_size;
1718 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1719 u64 *start, u64 *len)
1721 /* FIXME use last free of some kind */
1722 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1725 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1726 struct btrfs_device *device,
1727 u64 start, u64 *dev_extent_len)
1729 struct btrfs_fs_info *fs_info = device->fs_info;
1730 struct btrfs_root *root = fs_info->dev_root;
1732 struct btrfs_path *path;
1733 struct btrfs_key key;
1734 struct btrfs_key found_key;
1735 struct extent_buffer *leaf = NULL;
1736 struct btrfs_dev_extent *extent = NULL;
1738 path = btrfs_alloc_path();
1742 key.objectid = device->devid;
1744 key.type = BTRFS_DEV_EXTENT_KEY;
1746 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1748 ret = btrfs_previous_item(root, path, key.objectid,
1749 BTRFS_DEV_EXTENT_KEY);
1752 leaf = path->nodes[0];
1753 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1754 extent = btrfs_item_ptr(leaf, path->slots[0],
1755 struct btrfs_dev_extent);
1756 BUG_ON(found_key.offset > start || found_key.offset +
1757 btrfs_dev_extent_length(leaf, extent) < start);
1759 btrfs_release_path(path);
1761 } else if (ret == 0) {
1762 leaf = path->nodes[0];
1763 extent = btrfs_item_ptr(leaf, path->slots[0],
1764 struct btrfs_dev_extent);
1766 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1770 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1772 ret = btrfs_del_item(trans, root, path);
1774 btrfs_handle_fs_error(fs_info, ret,
1775 "Failed to remove dev extent item");
1777 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1780 btrfs_free_path(path);
1784 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1785 struct btrfs_device *device,
1786 u64 chunk_offset, u64 start, u64 num_bytes)
1789 struct btrfs_path *path;
1790 struct btrfs_fs_info *fs_info = device->fs_info;
1791 struct btrfs_root *root = fs_info->dev_root;
1792 struct btrfs_dev_extent *extent;
1793 struct extent_buffer *leaf;
1794 struct btrfs_key key;
1796 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1797 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1798 path = btrfs_alloc_path();
1802 key.objectid = device->devid;
1804 key.type = BTRFS_DEV_EXTENT_KEY;
1805 ret = btrfs_insert_empty_item(trans, root, path, &key,
1810 leaf = path->nodes[0];
1811 extent = btrfs_item_ptr(leaf, path->slots[0],
1812 struct btrfs_dev_extent);
1813 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1814 BTRFS_CHUNK_TREE_OBJECTID);
1815 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1816 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1817 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1819 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1820 btrfs_mark_buffer_dirty(leaf);
1822 btrfs_free_path(path);
1826 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1828 struct extent_map_tree *em_tree;
1829 struct extent_map *em;
1833 em_tree = &fs_info->mapping_tree;
1834 read_lock(&em_tree->lock);
1835 n = rb_last(&em_tree->map.rb_root);
1837 em = rb_entry(n, struct extent_map, rb_node);
1838 ret = em->start + em->len;
1840 read_unlock(&em_tree->lock);
1845 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1849 struct btrfs_key key;
1850 struct btrfs_key found_key;
1851 struct btrfs_path *path;
1853 path = btrfs_alloc_path();
1857 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1858 key.type = BTRFS_DEV_ITEM_KEY;
1859 key.offset = (u64)-1;
1861 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1865 BUG_ON(ret == 0); /* Corruption */
1867 ret = btrfs_previous_item(fs_info->chunk_root, path,
1868 BTRFS_DEV_ITEMS_OBJECTID,
1869 BTRFS_DEV_ITEM_KEY);
1873 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1875 *devid_ret = found_key.offset + 1;
1879 btrfs_free_path(path);
1884 * the device information is stored in the chunk root
1885 * the btrfs_device struct should be fully filled in
1887 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1888 struct btrfs_device *device)
1891 struct btrfs_path *path;
1892 struct btrfs_dev_item *dev_item;
1893 struct extent_buffer *leaf;
1894 struct btrfs_key key;
1897 path = btrfs_alloc_path();
1901 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1902 key.type = BTRFS_DEV_ITEM_KEY;
1903 key.offset = device->devid;
1905 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1906 &key, sizeof(*dev_item));
1910 leaf = path->nodes[0];
1911 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1913 btrfs_set_device_id(leaf, dev_item, device->devid);
1914 btrfs_set_device_generation(leaf, dev_item, 0);
1915 btrfs_set_device_type(leaf, dev_item, device->type);
1916 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1917 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1918 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1919 btrfs_set_device_total_bytes(leaf, dev_item,
1920 btrfs_device_get_disk_total_bytes(device));
1921 btrfs_set_device_bytes_used(leaf, dev_item,
1922 btrfs_device_get_bytes_used(device));
1923 btrfs_set_device_group(leaf, dev_item, 0);
1924 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1925 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1926 btrfs_set_device_start_offset(leaf, dev_item, 0);
1928 ptr = btrfs_device_uuid(dev_item);
1929 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1930 ptr = btrfs_device_fsid(dev_item);
1931 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1932 ptr, BTRFS_FSID_SIZE);
1933 btrfs_mark_buffer_dirty(leaf);
1937 btrfs_free_path(path);
1942 * Function to update ctime/mtime for a given device path.
1943 * Mainly used for ctime/mtime based probe like libblkid.
1945 static void update_dev_time(const char *path_name)
1949 filp = filp_open(path_name, O_RDWR, 0);
1952 file_update_time(filp);
1953 filp_close(filp, NULL);
1956 static int btrfs_rm_dev_item(struct btrfs_device *device)
1958 struct btrfs_root *root = device->fs_info->chunk_root;
1960 struct btrfs_path *path;
1961 struct btrfs_key key;
1962 struct btrfs_trans_handle *trans;
1964 path = btrfs_alloc_path();
1968 trans = btrfs_start_transaction(root, 0);
1969 if (IS_ERR(trans)) {
1970 btrfs_free_path(path);
1971 return PTR_ERR(trans);
1973 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1974 key.type = BTRFS_DEV_ITEM_KEY;
1975 key.offset = device->devid;
1977 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1981 btrfs_abort_transaction(trans, ret);
1982 btrfs_end_transaction(trans);
1986 ret = btrfs_del_item(trans, root, path);
1988 btrfs_abort_transaction(trans, ret);
1989 btrfs_end_transaction(trans);
1993 btrfs_free_path(path);
1995 ret = btrfs_commit_transaction(trans);
2000 * Verify that @num_devices satisfies the RAID profile constraints in the whole
2001 * filesystem. It's up to the caller to adjust that number regarding eg. device
2004 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2012 seq = read_seqbegin(&fs_info->profiles_lock);
2014 all_avail = fs_info->avail_data_alloc_bits |
2015 fs_info->avail_system_alloc_bits |
2016 fs_info->avail_metadata_alloc_bits;
2017 } while (read_seqretry(&fs_info->profiles_lock, seq));
2019 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2020 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2023 if (num_devices < btrfs_raid_array[i].devs_min) {
2024 int ret = btrfs_raid_array[i].mindev_error;
2034 static struct btrfs_device * btrfs_find_next_active_device(
2035 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2037 struct btrfs_device *next_device;
2039 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2040 if (next_device != device &&
2041 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2042 && next_device->bdev)
2050 * Helper function to check if the given device is part of s_bdev / latest_bdev
2051 * and replace it with the provided or the next active device, in the context
2052 * where this function called, there should be always be another device (or
2053 * this_dev) which is active.
2055 void btrfs_assign_next_active_device(struct btrfs_device *device,
2056 struct btrfs_device *this_dev)
2058 struct btrfs_fs_info *fs_info = device->fs_info;
2059 struct btrfs_device *next_device;
2062 next_device = this_dev;
2064 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2066 ASSERT(next_device);
2068 if (fs_info->sb->s_bdev &&
2069 (fs_info->sb->s_bdev == device->bdev))
2070 fs_info->sb->s_bdev = next_device->bdev;
2072 if (fs_info->fs_devices->latest_bdev == device->bdev)
2073 fs_info->fs_devices->latest_bdev = next_device->bdev;
2077 * Return btrfs_fs_devices::num_devices excluding the device that's being
2078 * currently replaced.
2080 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2082 u64 num_devices = fs_info->fs_devices->num_devices;
2084 down_read(&fs_info->dev_replace.rwsem);
2085 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2086 ASSERT(num_devices > 1);
2089 up_read(&fs_info->dev_replace.rwsem);
2094 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2097 struct btrfs_device *device;
2098 struct btrfs_fs_devices *cur_devices;
2099 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2103 mutex_lock(&uuid_mutex);
2105 num_devices = btrfs_num_devices(fs_info);
2107 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2111 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2113 if (IS_ERR(device)) {
2114 if (PTR_ERR(device) == -ENOENT &&
2115 strcmp(device_path, "missing") == 0)
2116 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2118 ret = PTR_ERR(device);
2122 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2123 btrfs_warn_in_rcu(fs_info,
2124 "cannot remove device %s (devid %llu) due to active swapfile",
2125 rcu_str_deref(device->name), device->devid);
2130 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2131 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2135 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2136 fs_info->fs_devices->rw_devices == 1) {
2137 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2141 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2142 mutex_lock(&fs_info->chunk_mutex);
2143 list_del_init(&device->dev_alloc_list);
2144 device->fs_devices->rw_devices--;
2145 mutex_unlock(&fs_info->chunk_mutex);
2148 mutex_unlock(&uuid_mutex);
2149 ret = btrfs_shrink_device(device, 0);
2150 mutex_lock(&uuid_mutex);
2155 * TODO: the superblock still includes this device in its num_devices
2156 * counter although write_all_supers() is not locked out. This
2157 * could give a filesystem state which requires a degraded mount.
2159 ret = btrfs_rm_dev_item(device);
2163 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2164 btrfs_scrub_cancel_dev(device);
2167 * the device list mutex makes sure that we don't change
2168 * the device list while someone else is writing out all
2169 * the device supers. Whoever is writing all supers, should
2170 * lock the device list mutex before getting the number of
2171 * devices in the super block (super_copy). Conversely,
2172 * whoever updates the number of devices in the super block
2173 * (super_copy) should hold the device list mutex.
2177 * In normal cases the cur_devices == fs_devices. But in case
2178 * of deleting a seed device, the cur_devices should point to
2179 * its own fs_devices listed under the fs_devices->seed.
2181 cur_devices = device->fs_devices;
2182 mutex_lock(&fs_devices->device_list_mutex);
2183 list_del_rcu(&device->dev_list);
2185 cur_devices->num_devices--;
2186 cur_devices->total_devices--;
2187 /* Update total_devices of the parent fs_devices if it's seed */
2188 if (cur_devices != fs_devices)
2189 fs_devices->total_devices--;
2191 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2192 cur_devices->missing_devices--;
2194 btrfs_assign_next_active_device(device, NULL);
2197 cur_devices->open_devices--;
2198 /* remove sysfs entry */
2199 btrfs_sysfs_rm_device_link(fs_devices, device);
2202 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2203 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2204 mutex_unlock(&fs_devices->device_list_mutex);
2207 * at this point, the device is zero sized and detached from
2208 * the devices list. All that's left is to zero out the old
2209 * supers and free the device.
2211 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2212 btrfs_scratch_superblocks(device->bdev, device->name->str);
2214 btrfs_close_bdev(device);
2216 btrfs_free_device(device);
2218 if (cur_devices->open_devices == 0) {
2219 while (fs_devices) {
2220 if (fs_devices->seed == cur_devices) {
2221 fs_devices->seed = cur_devices->seed;
2224 fs_devices = fs_devices->seed;
2226 cur_devices->seed = NULL;
2227 close_fs_devices(cur_devices);
2228 free_fs_devices(cur_devices);
2232 mutex_unlock(&uuid_mutex);
2236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2237 mutex_lock(&fs_info->chunk_mutex);
2238 list_add(&device->dev_alloc_list,
2239 &fs_devices->alloc_list);
2240 device->fs_devices->rw_devices++;
2241 mutex_unlock(&fs_info->chunk_mutex);
2246 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2248 struct btrfs_fs_devices *fs_devices;
2250 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2253 * in case of fs with no seed, srcdev->fs_devices will point
2254 * to fs_devices of fs_info. However when the dev being replaced is
2255 * a seed dev it will point to the seed's local fs_devices. In short
2256 * srcdev will have its correct fs_devices in both the cases.
2258 fs_devices = srcdev->fs_devices;
2260 list_del_rcu(&srcdev->dev_list);
2261 list_del(&srcdev->dev_alloc_list);
2262 fs_devices->num_devices--;
2263 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2264 fs_devices->missing_devices--;
2266 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2267 fs_devices->rw_devices--;
2270 fs_devices->open_devices--;
2273 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2275 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2276 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2278 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2279 /* zero out the old super if it is writable */
2280 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2283 btrfs_close_bdev(srcdev);
2285 btrfs_free_device(srcdev);
2287 /* if this is no devs we rather delete the fs_devices */
2288 if (!fs_devices->num_devices) {
2289 struct btrfs_fs_devices *tmp_fs_devices;
2292 * On a mounted FS, num_devices can't be zero unless it's a
2293 * seed. In case of a seed device being replaced, the replace
2294 * target added to the sprout FS, so there will be no more
2295 * device left under the seed FS.
2297 ASSERT(fs_devices->seeding);
2299 tmp_fs_devices = fs_info->fs_devices;
2300 while (tmp_fs_devices) {
2301 if (tmp_fs_devices->seed == fs_devices) {
2302 tmp_fs_devices->seed = fs_devices->seed;
2305 tmp_fs_devices = tmp_fs_devices->seed;
2307 fs_devices->seed = NULL;
2308 close_fs_devices(fs_devices);
2309 free_fs_devices(fs_devices);
2313 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2315 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2318 mutex_lock(&fs_devices->device_list_mutex);
2320 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2323 fs_devices->open_devices--;
2325 fs_devices->num_devices--;
2327 btrfs_assign_next_active_device(tgtdev, NULL);
2329 list_del_rcu(&tgtdev->dev_list);
2331 mutex_unlock(&fs_devices->device_list_mutex);
2334 * The update_dev_time() with in btrfs_scratch_superblocks()
2335 * may lead to a call to btrfs_show_devname() which will try
2336 * to hold device_list_mutex. And here this device
2337 * is already out of device list, so we don't have to hold
2338 * the device_list_mutex lock.
2340 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2342 btrfs_close_bdev(tgtdev);
2344 btrfs_free_device(tgtdev);
2347 static struct btrfs_device *btrfs_find_device_by_path(
2348 struct btrfs_fs_info *fs_info, const char *device_path)
2351 struct btrfs_super_block *disk_super;
2354 struct block_device *bdev;
2355 struct buffer_head *bh;
2356 struct btrfs_device *device;
2358 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2359 fs_info->bdev_holder, 0, &bdev, &bh);
2361 return ERR_PTR(ret);
2362 disk_super = (struct btrfs_super_block *)bh->b_data;
2363 devid = btrfs_stack_device_id(&disk_super->dev_item);
2364 dev_uuid = disk_super->dev_item.uuid;
2365 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2366 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2367 disk_super->metadata_uuid, true);
2369 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2370 disk_super->fsid, true);
2374 device = ERR_PTR(-ENOENT);
2375 blkdev_put(bdev, FMODE_READ);
2380 * Lookup a device given by device id, or the path if the id is 0.
2382 struct btrfs_device *btrfs_find_device_by_devspec(
2383 struct btrfs_fs_info *fs_info, u64 devid,
2384 const char *device_path)
2386 struct btrfs_device *device;
2389 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2392 return ERR_PTR(-ENOENT);
2396 if (!device_path || !device_path[0])
2397 return ERR_PTR(-EINVAL);
2399 if (strcmp(device_path, "missing") == 0) {
2400 /* Find first missing device */
2401 list_for_each_entry(device, &fs_info->fs_devices->devices,
2403 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2404 &device->dev_state) && !device->bdev)
2407 return ERR_PTR(-ENOENT);
2410 return btrfs_find_device_by_path(fs_info, device_path);
2414 * does all the dirty work required for changing file system's UUID.
2416 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2418 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2419 struct btrfs_fs_devices *old_devices;
2420 struct btrfs_fs_devices *seed_devices;
2421 struct btrfs_super_block *disk_super = fs_info->super_copy;
2422 struct btrfs_device *device;
2425 lockdep_assert_held(&uuid_mutex);
2426 if (!fs_devices->seeding)
2429 seed_devices = alloc_fs_devices(NULL, NULL);
2430 if (IS_ERR(seed_devices))
2431 return PTR_ERR(seed_devices);
2433 old_devices = clone_fs_devices(fs_devices);
2434 if (IS_ERR(old_devices)) {
2435 kfree(seed_devices);
2436 return PTR_ERR(old_devices);
2439 list_add(&old_devices->fs_list, &fs_uuids);
2441 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2442 seed_devices->opened = 1;
2443 INIT_LIST_HEAD(&seed_devices->devices);
2444 INIT_LIST_HEAD(&seed_devices->alloc_list);
2445 mutex_init(&seed_devices->device_list_mutex);
2447 mutex_lock(&fs_devices->device_list_mutex);
2448 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2450 list_for_each_entry(device, &seed_devices->devices, dev_list)
2451 device->fs_devices = seed_devices;
2453 mutex_lock(&fs_info->chunk_mutex);
2454 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2455 mutex_unlock(&fs_info->chunk_mutex);
2457 fs_devices->seeding = 0;
2458 fs_devices->num_devices = 0;
2459 fs_devices->open_devices = 0;
2460 fs_devices->missing_devices = 0;
2461 fs_devices->rotating = 0;
2462 fs_devices->seed = seed_devices;
2464 generate_random_uuid(fs_devices->fsid);
2465 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2466 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2467 mutex_unlock(&fs_devices->device_list_mutex);
2469 super_flags = btrfs_super_flags(disk_super) &
2470 ~BTRFS_SUPER_FLAG_SEEDING;
2471 btrfs_set_super_flags(disk_super, super_flags);
2477 * Store the expected generation for seed devices in device items.
2479 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2481 struct btrfs_fs_info *fs_info = trans->fs_info;
2482 struct btrfs_root *root = fs_info->chunk_root;
2483 struct btrfs_path *path;
2484 struct extent_buffer *leaf;
2485 struct btrfs_dev_item *dev_item;
2486 struct btrfs_device *device;
2487 struct btrfs_key key;
2488 u8 fs_uuid[BTRFS_FSID_SIZE];
2489 u8 dev_uuid[BTRFS_UUID_SIZE];
2493 path = btrfs_alloc_path();
2497 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2499 key.type = BTRFS_DEV_ITEM_KEY;
2502 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2506 leaf = path->nodes[0];
2508 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2509 ret = btrfs_next_leaf(root, path);
2514 leaf = path->nodes[0];
2515 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2516 btrfs_release_path(path);
2520 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2521 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2522 key.type != BTRFS_DEV_ITEM_KEY)
2525 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2526 struct btrfs_dev_item);
2527 devid = btrfs_device_id(leaf, dev_item);
2528 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2530 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2532 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2534 BUG_ON(!device); /* Logic error */
2536 if (device->fs_devices->seeding) {
2537 btrfs_set_device_generation(leaf, dev_item,
2538 device->generation);
2539 btrfs_mark_buffer_dirty(leaf);
2547 btrfs_free_path(path);
2551 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2553 struct btrfs_root *root = fs_info->dev_root;
2554 struct request_queue *q;
2555 struct btrfs_trans_handle *trans;
2556 struct btrfs_device *device;
2557 struct block_device *bdev;
2558 struct super_block *sb = fs_info->sb;
2559 struct rcu_string *name;
2560 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2561 u64 orig_super_total_bytes;
2562 u64 orig_super_num_devices;
2563 int seeding_dev = 0;
2565 bool unlocked = false;
2567 if (sb_rdonly(sb) && !fs_devices->seeding)
2570 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2571 fs_info->bdev_holder);
2573 return PTR_ERR(bdev);
2575 if (fs_devices->seeding) {
2577 down_write(&sb->s_umount);
2578 mutex_lock(&uuid_mutex);
2581 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2583 mutex_lock(&fs_devices->device_list_mutex);
2584 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2585 if (device->bdev == bdev) {
2588 &fs_devices->device_list_mutex);
2592 mutex_unlock(&fs_devices->device_list_mutex);
2594 device = btrfs_alloc_device(fs_info, NULL, NULL);
2595 if (IS_ERR(device)) {
2596 /* we can safely leave the fs_devices entry around */
2597 ret = PTR_ERR(device);
2601 name = rcu_string_strdup(device_path, GFP_KERNEL);
2604 goto error_free_device;
2606 rcu_assign_pointer(device->name, name);
2608 trans = btrfs_start_transaction(root, 0);
2609 if (IS_ERR(trans)) {
2610 ret = PTR_ERR(trans);
2611 goto error_free_device;
2614 q = bdev_get_queue(bdev);
2615 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2616 device->generation = trans->transid;
2617 device->io_width = fs_info->sectorsize;
2618 device->io_align = fs_info->sectorsize;
2619 device->sector_size = fs_info->sectorsize;
2620 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2621 fs_info->sectorsize);
2622 device->disk_total_bytes = device->total_bytes;
2623 device->commit_total_bytes = device->total_bytes;
2624 device->fs_info = fs_info;
2625 device->bdev = bdev;
2626 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2627 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2628 device->mode = FMODE_EXCL;
2629 device->dev_stats_valid = 1;
2630 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2633 sb->s_flags &= ~SB_RDONLY;
2634 ret = btrfs_prepare_sprout(fs_info);
2636 btrfs_abort_transaction(trans, ret);
2641 device->fs_devices = fs_devices;
2643 mutex_lock(&fs_devices->device_list_mutex);
2644 mutex_lock(&fs_info->chunk_mutex);
2645 list_add_rcu(&device->dev_list, &fs_devices->devices);
2646 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2647 fs_devices->num_devices++;
2648 fs_devices->open_devices++;
2649 fs_devices->rw_devices++;
2650 fs_devices->total_devices++;
2651 fs_devices->total_rw_bytes += device->total_bytes;
2653 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2655 if (!blk_queue_nonrot(q))
2656 fs_devices->rotating = 1;
2658 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2659 btrfs_set_super_total_bytes(fs_info->super_copy,
2660 round_down(orig_super_total_bytes + device->total_bytes,
2661 fs_info->sectorsize));
2663 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2664 btrfs_set_super_num_devices(fs_info->super_copy,
2665 orig_super_num_devices + 1);
2667 /* add sysfs device entry */
2668 btrfs_sysfs_add_device_link(fs_devices, device);
2671 * we've got more storage, clear any full flags on the space
2674 btrfs_clear_space_info_full(fs_info);
2676 mutex_unlock(&fs_info->chunk_mutex);
2677 mutex_unlock(&fs_devices->device_list_mutex);
2680 mutex_lock(&fs_info->chunk_mutex);
2681 ret = init_first_rw_device(trans);
2682 mutex_unlock(&fs_info->chunk_mutex);
2684 btrfs_abort_transaction(trans, ret);
2689 ret = btrfs_add_dev_item(trans, device);
2691 btrfs_abort_transaction(trans, ret);
2696 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2698 ret = btrfs_finish_sprout(trans);
2700 btrfs_abort_transaction(trans, ret);
2704 /* Sprouting would change fsid of the mounted root,
2705 * so rename the fsid on the sysfs
2707 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2708 fs_info->fs_devices->fsid);
2709 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2711 "sysfs: failed to create fsid for sprout");
2714 ret = btrfs_commit_transaction(trans);
2717 mutex_unlock(&uuid_mutex);
2718 up_write(&sb->s_umount);
2721 if (ret) /* transaction commit */
2724 ret = btrfs_relocate_sys_chunks(fs_info);
2726 btrfs_handle_fs_error(fs_info, ret,
2727 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2728 trans = btrfs_attach_transaction(root);
2729 if (IS_ERR(trans)) {
2730 if (PTR_ERR(trans) == -ENOENT)
2732 ret = PTR_ERR(trans);
2736 ret = btrfs_commit_transaction(trans);
2739 /* Update ctime/mtime for libblkid */
2740 update_dev_time(device_path);
2744 btrfs_sysfs_rm_device_link(fs_devices, device);
2745 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2746 mutex_lock(&fs_info->chunk_mutex);
2747 list_del_rcu(&device->dev_list);
2748 list_del(&device->dev_alloc_list);
2749 fs_info->fs_devices->num_devices--;
2750 fs_info->fs_devices->open_devices--;
2751 fs_info->fs_devices->rw_devices--;
2752 fs_info->fs_devices->total_devices--;
2753 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2754 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2755 btrfs_set_super_total_bytes(fs_info->super_copy,
2756 orig_super_total_bytes);
2757 btrfs_set_super_num_devices(fs_info->super_copy,
2758 orig_super_num_devices);
2759 mutex_unlock(&fs_info->chunk_mutex);
2760 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2763 sb->s_flags |= SB_RDONLY;
2765 btrfs_end_transaction(trans);
2767 btrfs_free_device(device);
2769 blkdev_put(bdev, FMODE_EXCL);
2770 if (seeding_dev && !unlocked) {
2771 mutex_unlock(&uuid_mutex);
2772 up_write(&sb->s_umount);
2777 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2778 struct btrfs_device *device)
2781 struct btrfs_path *path;
2782 struct btrfs_root *root = device->fs_info->chunk_root;
2783 struct btrfs_dev_item *dev_item;
2784 struct extent_buffer *leaf;
2785 struct btrfs_key key;
2787 path = btrfs_alloc_path();
2791 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2792 key.type = BTRFS_DEV_ITEM_KEY;
2793 key.offset = device->devid;
2795 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2804 leaf = path->nodes[0];
2805 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2807 btrfs_set_device_id(leaf, dev_item, device->devid);
2808 btrfs_set_device_type(leaf, dev_item, device->type);
2809 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2810 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2811 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2812 btrfs_set_device_total_bytes(leaf, dev_item,
2813 btrfs_device_get_disk_total_bytes(device));
2814 btrfs_set_device_bytes_used(leaf, dev_item,
2815 btrfs_device_get_bytes_used(device));
2816 btrfs_mark_buffer_dirty(leaf);
2819 btrfs_free_path(path);
2823 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2824 struct btrfs_device *device, u64 new_size)
2826 struct btrfs_fs_info *fs_info = device->fs_info;
2827 struct btrfs_super_block *super_copy = fs_info->super_copy;
2831 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2834 new_size = round_down(new_size, fs_info->sectorsize);
2836 mutex_lock(&fs_info->chunk_mutex);
2837 old_total = btrfs_super_total_bytes(super_copy);
2838 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2840 if (new_size <= device->total_bytes ||
2841 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2842 mutex_unlock(&fs_info->chunk_mutex);
2846 btrfs_set_super_total_bytes(super_copy,
2847 round_down(old_total + diff, fs_info->sectorsize));
2848 device->fs_devices->total_rw_bytes += diff;
2850 btrfs_device_set_total_bytes(device, new_size);
2851 btrfs_device_set_disk_total_bytes(device, new_size);
2852 btrfs_clear_space_info_full(device->fs_info);
2853 if (list_empty(&device->post_commit_list))
2854 list_add_tail(&device->post_commit_list,
2855 &trans->transaction->dev_update_list);
2856 mutex_unlock(&fs_info->chunk_mutex);
2858 return btrfs_update_device(trans, device);
2861 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2863 struct btrfs_fs_info *fs_info = trans->fs_info;
2864 struct btrfs_root *root = fs_info->chunk_root;
2866 struct btrfs_path *path;
2867 struct btrfs_key key;
2869 path = btrfs_alloc_path();
2873 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2874 key.offset = chunk_offset;
2875 key.type = BTRFS_CHUNK_ITEM_KEY;
2877 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2880 else if (ret > 0) { /* Logic error or corruption */
2881 btrfs_handle_fs_error(fs_info, -ENOENT,
2882 "Failed lookup while freeing chunk.");
2887 ret = btrfs_del_item(trans, root, path);
2889 btrfs_handle_fs_error(fs_info, ret,
2890 "Failed to delete chunk item.");
2892 btrfs_free_path(path);
2896 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2898 struct btrfs_super_block *super_copy = fs_info->super_copy;
2899 struct btrfs_disk_key *disk_key;
2900 struct btrfs_chunk *chunk;
2907 struct btrfs_key key;
2909 mutex_lock(&fs_info->chunk_mutex);
2910 array_size = btrfs_super_sys_array_size(super_copy);
2912 ptr = super_copy->sys_chunk_array;
2915 while (cur < array_size) {
2916 disk_key = (struct btrfs_disk_key *)ptr;
2917 btrfs_disk_key_to_cpu(&key, disk_key);
2919 len = sizeof(*disk_key);
2921 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2922 chunk = (struct btrfs_chunk *)(ptr + len);
2923 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2924 len += btrfs_chunk_item_size(num_stripes);
2929 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2930 key.offset == chunk_offset) {
2931 memmove(ptr, ptr + len, array_size - (cur + len));
2933 btrfs_set_super_sys_array_size(super_copy, array_size);
2939 mutex_unlock(&fs_info->chunk_mutex);
2944 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2945 * @logical: Logical block offset in bytes.
2946 * @length: Length of extent in bytes.
2948 * Return: Chunk mapping or ERR_PTR.
2950 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2951 u64 logical, u64 length)
2953 struct extent_map_tree *em_tree;
2954 struct extent_map *em;
2956 em_tree = &fs_info->mapping_tree;
2957 read_lock(&em_tree->lock);
2958 em = lookup_extent_mapping(em_tree, logical, length);
2959 read_unlock(&em_tree->lock);
2962 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2964 return ERR_PTR(-EINVAL);
2967 if (em->start > logical || em->start + em->len < logical) {
2969 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2970 logical, length, em->start, em->start + em->len);
2971 free_extent_map(em);
2972 return ERR_PTR(-EINVAL);
2975 /* callers are responsible for dropping em's ref. */
2979 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2981 struct btrfs_fs_info *fs_info = trans->fs_info;
2982 struct extent_map *em;
2983 struct map_lookup *map;
2984 u64 dev_extent_len = 0;
2986 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2988 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2991 * This is a logic error, but we don't want to just rely on the
2992 * user having built with ASSERT enabled, so if ASSERT doesn't
2993 * do anything we still error out.
2998 map = em->map_lookup;
2999 mutex_lock(&fs_info->chunk_mutex);
3000 check_system_chunk(trans, map->type);
3001 mutex_unlock(&fs_info->chunk_mutex);
3004 * Take the device list mutex to prevent races with the final phase of
3005 * a device replace operation that replaces the device object associated
3006 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
3008 mutex_lock(&fs_devices->device_list_mutex);
3009 for (i = 0; i < map->num_stripes; i++) {
3010 struct btrfs_device *device = map->stripes[i].dev;
3011 ret = btrfs_free_dev_extent(trans, device,
3012 map->stripes[i].physical,
3015 mutex_unlock(&fs_devices->device_list_mutex);
3016 btrfs_abort_transaction(trans, ret);
3020 if (device->bytes_used > 0) {
3021 mutex_lock(&fs_info->chunk_mutex);
3022 btrfs_device_set_bytes_used(device,
3023 device->bytes_used - dev_extent_len);
3024 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3025 btrfs_clear_space_info_full(fs_info);
3026 mutex_unlock(&fs_info->chunk_mutex);
3029 ret = btrfs_update_device(trans, device);
3031 mutex_unlock(&fs_devices->device_list_mutex);
3032 btrfs_abort_transaction(trans, ret);
3036 mutex_unlock(&fs_devices->device_list_mutex);
3038 ret = btrfs_free_chunk(trans, chunk_offset);
3040 btrfs_abort_transaction(trans, ret);
3044 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3046 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3047 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3049 btrfs_abort_transaction(trans, ret);
3054 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3056 btrfs_abort_transaction(trans, ret);
3062 free_extent_map(em);
3066 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3068 struct btrfs_root *root = fs_info->chunk_root;
3069 struct btrfs_trans_handle *trans;
3073 * Prevent races with automatic removal of unused block groups.
3074 * After we relocate and before we remove the chunk with offset
3075 * chunk_offset, automatic removal of the block group can kick in,
3076 * resulting in a failure when calling btrfs_remove_chunk() below.
3078 * Make sure to acquire this mutex before doing a tree search (dev
3079 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3080 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3081 * we release the path used to search the chunk/dev tree and before
3082 * the current task acquires this mutex and calls us.
3084 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3086 ret = btrfs_can_relocate(fs_info, chunk_offset);
3090 /* step one, relocate all the extents inside this chunk */
3091 btrfs_scrub_pause(fs_info);
3092 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3093 btrfs_scrub_continue(fs_info);
3097 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3099 if (IS_ERR(trans)) {
3100 ret = PTR_ERR(trans);
3101 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3106 * step two, delete the device extents and the
3107 * chunk tree entries
3109 ret = btrfs_remove_chunk(trans, chunk_offset);
3110 btrfs_end_transaction(trans);
3114 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3116 struct btrfs_root *chunk_root = fs_info->chunk_root;
3117 struct btrfs_path *path;
3118 struct extent_buffer *leaf;
3119 struct btrfs_chunk *chunk;
3120 struct btrfs_key key;
3121 struct btrfs_key found_key;
3123 bool retried = false;
3127 path = btrfs_alloc_path();
3132 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3133 key.offset = (u64)-1;
3134 key.type = BTRFS_CHUNK_ITEM_KEY;
3137 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3138 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3140 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3143 BUG_ON(ret == 0); /* Corruption */
3145 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3148 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3154 leaf = path->nodes[0];
3155 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3157 chunk = btrfs_item_ptr(leaf, path->slots[0],
3158 struct btrfs_chunk);
3159 chunk_type = btrfs_chunk_type(leaf, chunk);
3160 btrfs_release_path(path);
3162 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3163 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3169 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3171 if (found_key.offset == 0)
3173 key.offset = found_key.offset - 1;
3176 if (failed && !retried) {
3180 } else if (WARN_ON(failed && retried)) {
3184 btrfs_free_path(path);
3189 * return 1 : allocate a data chunk successfully,
3190 * return <0: errors during allocating a data chunk,
3191 * return 0 : no need to allocate a data chunk.
3193 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3196 struct btrfs_block_group_cache *cache;
3200 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3202 chunk_type = cache->flags;
3203 btrfs_put_block_group(cache);
3205 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3206 spin_lock(&fs_info->data_sinfo->lock);
3207 bytes_used = fs_info->data_sinfo->bytes_used;
3208 spin_unlock(&fs_info->data_sinfo->lock);
3211 struct btrfs_trans_handle *trans;
3214 trans = btrfs_join_transaction(fs_info->tree_root);
3216 return PTR_ERR(trans);
3218 ret = btrfs_force_chunk_alloc(trans,
3219 BTRFS_BLOCK_GROUP_DATA);
3220 btrfs_end_transaction(trans);
3229 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3230 struct btrfs_balance_control *bctl)
3232 struct btrfs_root *root = fs_info->tree_root;
3233 struct btrfs_trans_handle *trans;
3234 struct btrfs_balance_item *item;
3235 struct btrfs_disk_balance_args disk_bargs;
3236 struct btrfs_path *path;
3237 struct extent_buffer *leaf;
3238 struct btrfs_key key;
3241 path = btrfs_alloc_path();
3245 trans = btrfs_start_transaction(root, 0);
3246 if (IS_ERR(trans)) {
3247 btrfs_free_path(path);
3248 return PTR_ERR(trans);
3251 key.objectid = BTRFS_BALANCE_OBJECTID;
3252 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3255 ret = btrfs_insert_empty_item(trans, root, path, &key,
3260 leaf = path->nodes[0];
3261 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3263 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3265 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3266 btrfs_set_balance_data(leaf, item, &disk_bargs);
3267 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3268 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3269 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3270 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3272 btrfs_set_balance_flags(leaf, item, bctl->flags);
3274 btrfs_mark_buffer_dirty(leaf);
3276 btrfs_free_path(path);
3277 err = btrfs_commit_transaction(trans);
3283 static int del_balance_item(struct btrfs_fs_info *fs_info)
3285 struct btrfs_root *root = fs_info->tree_root;
3286 struct btrfs_trans_handle *trans;
3287 struct btrfs_path *path;
3288 struct btrfs_key key;
3291 path = btrfs_alloc_path();
3295 trans = btrfs_start_transaction(root, 0);
3296 if (IS_ERR(trans)) {
3297 btrfs_free_path(path);
3298 return PTR_ERR(trans);
3301 key.objectid = BTRFS_BALANCE_OBJECTID;
3302 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3305 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3313 ret = btrfs_del_item(trans, root, path);
3315 btrfs_free_path(path);
3316 err = btrfs_commit_transaction(trans);
3323 * This is a heuristic used to reduce the number of chunks balanced on
3324 * resume after balance was interrupted.
3326 static void update_balance_args(struct btrfs_balance_control *bctl)
3329 * Turn on soft mode for chunk types that were being converted.
3331 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3332 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3333 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3334 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3335 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3336 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3339 * Turn on usage filter if is not already used. The idea is
3340 * that chunks that we have already balanced should be
3341 * reasonably full. Don't do it for chunks that are being
3342 * converted - that will keep us from relocating unconverted
3343 * (albeit full) chunks.
3345 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3346 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3347 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3348 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3349 bctl->data.usage = 90;
3351 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3352 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3353 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3354 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3355 bctl->sys.usage = 90;
3357 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3358 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3359 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3360 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3361 bctl->meta.usage = 90;
3366 * Clear the balance status in fs_info and delete the balance item from disk.
3368 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3370 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3373 BUG_ON(!fs_info->balance_ctl);
3375 spin_lock(&fs_info->balance_lock);
3376 fs_info->balance_ctl = NULL;
3377 spin_unlock(&fs_info->balance_lock);
3380 ret = del_balance_item(fs_info);
3382 btrfs_handle_fs_error(fs_info, ret, NULL);
3386 * Balance filters. Return 1 if chunk should be filtered out
3387 * (should not be balanced).
3389 static int chunk_profiles_filter(u64 chunk_type,
3390 struct btrfs_balance_args *bargs)
3392 chunk_type = chunk_to_extended(chunk_type) &
3393 BTRFS_EXTENDED_PROFILE_MASK;
3395 if (bargs->profiles & chunk_type)
3401 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3402 struct btrfs_balance_args *bargs)
3404 struct btrfs_block_group_cache *cache;
3406 u64 user_thresh_min;
3407 u64 user_thresh_max;
3410 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3411 chunk_used = btrfs_block_group_used(&cache->item);
3413 if (bargs->usage_min == 0)
3414 user_thresh_min = 0;
3416 user_thresh_min = div_factor_fine(cache->key.offset,
3419 if (bargs->usage_max == 0)
3420 user_thresh_max = 1;
3421 else if (bargs->usage_max > 100)
3422 user_thresh_max = cache->key.offset;
3424 user_thresh_max = div_factor_fine(cache->key.offset,
3427 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3430 btrfs_put_block_group(cache);
3434 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3435 u64 chunk_offset, struct btrfs_balance_args *bargs)
3437 struct btrfs_block_group_cache *cache;
3438 u64 chunk_used, user_thresh;
3441 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3442 chunk_used = btrfs_block_group_used(&cache->item);
3444 if (bargs->usage_min == 0)
3446 else if (bargs->usage > 100)
3447 user_thresh = cache->key.offset;
3449 user_thresh = div_factor_fine(cache->key.offset,
3452 if (chunk_used < user_thresh)
3455 btrfs_put_block_group(cache);
3459 static int chunk_devid_filter(struct extent_buffer *leaf,
3460 struct btrfs_chunk *chunk,
3461 struct btrfs_balance_args *bargs)
3463 struct btrfs_stripe *stripe;
3464 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3467 for (i = 0; i < num_stripes; i++) {
3468 stripe = btrfs_stripe_nr(chunk, i);
3469 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3476 static u64 calc_data_stripes(u64 type, int num_stripes)
3478 const int index = btrfs_bg_flags_to_raid_index(type);
3479 const int ncopies = btrfs_raid_array[index].ncopies;
3480 const int nparity = btrfs_raid_array[index].nparity;
3483 return num_stripes - nparity;
3485 return num_stripes / ncopies;
3488 /* [pstart, pend) */
3489 static int chunk_drange_filter(struct extent_buffer *leaf,
3490 struct btrfs_chunk *chunk,
3491 struct btrfs_balance_args *bargs)
3493 struct btrfs_stripe *stripe;
3494 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3501 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3504 type = btrfs_chunk_type(leaf, chunk);
3505 factor = calc_data_stripes(type, num_stripes);
3507 for (i = 0; i < num_stripes; i++) {
3508 stripe = btrfs_stripe_nr(chunk, i);
3509 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3512 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3513 stripe_length = btrfs_chunk_length(leaf, chunk);
3514 stripe_length = div_u64(stripe_length, factor);
3516 if (stripe_offset < bargs->pend &&
3517 stripe_offset + stripe_length > bargs->pstart)
3524 /* [vstart, vend) */
3525 static int chunk_vrange_filter(struct extent_buffer *leaf,
3526 struct btrfs_chunk *chunk,
3528 struct btrfs_balance_args *bargs)
3530 if (chunk_offset < bargs->vend &&
3531 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3532 /* at least part of the chunk is inside this vrange */
3538 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3539 struct btrfs_chunk *chunk,
3540 struct btrfs_balance_args *bargs)
3542 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3544 if (bargs->stripes_min <= num_stripes
3545 && num_stripes <= bargs->stripes_max)
3551 static int chunk_soft_convert_filter(u64 chunk_type,
3552 struct btrfs_balance_args *bargs)
3554 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3557 chunk_type = chunk_to_extended(chunk_type) &
3558 BTRFS_EXTENDED_PROFILE_MASK;
3560 if (bargs->target == chunk_type)
3566 static int should_balance_chunk(struct extent_buffer *leaf,
3567 struct btrfs_chunk *chunk, u64 chunk_offset)
3569 struct btrfs_fs_info *fs_info = leaf->fs_info;
3570 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3571 struct btrfs_balance_args *bargs = NULL;
3572 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3575 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3576 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3580 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3581 bargs = &bctl->data;
3582 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3584 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3585 bargs = &bctl->meta;
3587 /* profiles filter */
3588 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3589 chunk_profiles_filter(chunk_type, bargs)) {
3594 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3595 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3597 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3598 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3603 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3604 chunk_devid_filter(leaf, chunk, bargs)) {
3608 /* drange filter, makes sense only with devid filter */
3609 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3610 chunk_drange_filter(leaf, chunk, bargs)) {
3615 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3616 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3620 /* stripes filter */
3621 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3622 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3626 /* soft profile changing mode */
3627 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3628 chunk_soft_convert_filter(chunk_type, bargs)) {
3633 * limited by count, must be the last filter
3635 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3636 if (bargs->limit == 0)
3640 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3642 * Same logic as the 'limit' filter; the minimum cannot be
3643 * determined here because we do not have the global information
3644 * about the count of all chunks that satisfy the filters.
3646 if (bargs->limit_max == 0)
3655 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3657 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3658 struct btrfs_root *chunk_root = fs_info->chunk_root;
3660 struct btrfs_chunk *chunk;
3661 struct btrfs_path *path = NULL;
3662 struct btrfs_key key;
3663 struct btrfs_key found_key;
3664 struct extent_buffer *leaf;
3667 int enospc_errors = 0;
3668 bool counting = true;
3669 /* The single value limit and min/max limits use the same bytes in the */
3670 u64 limit_data = bctl->data.limit;
3671 u64 limit_meta = bctl->meta.limit;
3672 u64 limit_sys = bctl->sys.limit;
3676 int chunk_reserved = 0;
3678 path = btrfs_alloc_path();
3684 /* zero out stat counters */
3685 spin_lock(&fs_info->balance_lock);
3686 memset(&bctl->stat, 0, sizeof(bctl->stat));
3687 spin_unlock(&fs_info->balance_lock);
3691 * The single value limit and min/max limits use the same bytes
3694 bctl->data.limit = limit_data;
3695 bctl->meta.limit = limit_meta;
3696 bctl->sys.limit = limit_sys;
3698 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3699 key.offset = (u64)-1;
3700 key.type = BTRFS_CHUNK_ITEM_KEY;
3703 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3704 atomic_read(&fs_info->balance_cancel_req)) {
3709 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3710 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3712 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3717 * this shouldn't happen, it means the last relocate
3721 BUG(); /* FIXME break ? */
3723 ret = btrfs_previous_item(chunk_root, path, 0,
3724 BTRFS_CHUNK_ITEM_KEY);
3726 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3731 leaf = path->nodes[0];
3732 slot = path->slots[0];
3733 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3735 if (found_key.objectid != key.objectid) {
3736 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3740 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3741 chunk_type = btrfs_chunk_type(leaf, chunk);
3744 spin_lock(&fs_info->balance_lock);
3745 bctl->stat.considered++;
3746 spin_unlock(&fs_info->balance_lock);
3749 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3751 btrfs_release_path(path);
3753 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3758 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3759 spin_lock(&fs_info->balance_lock);
3760 bctl->stat.expected++;
3761 spin_unlock(&fs_info->balance_lock);
3763 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3765 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3767 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3774 * Apply limit_min filter, no need to check if the LIMITS
3775 * filter is used, limit_min is 0 by default
3777 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3778 count_data < bctl->data.limit_min)
3779 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3780 count_meta < bctl->meta.limit_min)
3781 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3782 count_sys < bctl->sys.limit_min)) {
3783 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3787 if (!chunk_reserved) {
3789 * We may be relocating the only data chunk we have,
3790 * which could potentially end up with losing data's
3791 * raid profile, so lets allocate an empty one in
3794 ret = btrfs_may_alloc_data_chunk(fs_info,
3797 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3799 } else if (ret == 1) {
3804 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3805 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3806 if (ret == -ENOSPC) {
3808 } else if (ret == -ETXTBSY) {
3810 "skipping relocation of block group %llu due to active swapfile",
3816 spin_lock(&fs_info->balance_lock);
3817 bctl->stat.completed++;
3818 spin_unlock(&fs_info->balance_lock);
3821 if (found_key.offset == 0)
3823 key.offset = found_key.offset - 1;
3827 btrfs_release_path(path);
3832 btrfs_free_path(path);
3833 if (enospc_errors) {
3834 btrfs_info(fs_info, "%d enospc errors during balance",
3844 * alloc_profile_is_valid - see if a given profile is valid and reduced
3845 * @flags: profile to validate
3846 * @extended: if true @flags is treated as an extended profile
3848 static int alloc_profile_is_valid(u64 flags, int extended)
3850 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3851 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3853 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3855 /* 1) check that all other bits are zeroed */
3859 /* 2) see if profile is reduced */
3861 return !extended; /* "0" is valid for usual profiles */
3863 /* true if exactly one bit set */
3864 return is_power_of_2(flags);
3867 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3869 /* cancel requested || normal exit path */
3870 return atomic_read(&fs_info->balance_cancel_req) ||
3871 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3872 atomic_read(&fs_info->balance_cancel_req) == 0);
3875 /* Non-zero return value signifies invalidity */
3876 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3879 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3880 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3881 (bctl_arg->target & ~allowed)));
3885 * Fill @buf with textual description of balance filter flags @bargs, up to
3886 * @size_buf including the terminating null. The output may be trimmed if it
3887 * does not fit into the provided buffer.
3889 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3893 u32 size_bp = size_buf;
3895 u64 flags = bargs->flags;
3896 char tmp_buf[128] = {'\0'};
3901 #define CHECK_APPEND_NOARG(a) \
3903 ret = snprintf(bp, size_bp, (a)); \
3904 if (ret < 0 || ret >= size_bp) \
3905 goto out_overflow; \
3910 #define CHECK_APPEND_1ARG(a, v1) \
3912 ret = snprintf(bp, size_bp, (a), (v1)); \
3913 if (ret < 0 || ret >= size_bp) \
3914 goto out_overflow; \
3919 #define CHECK_APPEND_2ARG(a, v1, v2) \
3921 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3922 if (ret < 0 || ret >= size_bp) \
3923 goto out_overflow; \
3928 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3929 CHECK_APPEND_1ARG("convert=%s,",
3930 btrfs_bg_type_to_raid_name(bargs->target));
3932 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3933 CHECK_APPEND_NOARG("soft,");
3935 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3936 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3938 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3941 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3942 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3944 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3945 CHECK_APPEND_2ARG("usage=%u..%u,",
3946 bargs->usage_min, bargs->usage_max);
3948 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3949 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3951 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3952 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3953 bargs->pstart, bargs->pend);
3955 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3956 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3957 bargs->vstart, bargs->vend);
3959 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3960 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3962 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3963 CHECK_APPEND_2ARG("limit=%u..%u,",
3964 bargs->limit_min, bargs->limit_max);
3966 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3967 CHECK_APPEND_2ARG("stripes=%u..%u,",
3968 bargs->stripes_min, bargs->stripes_max);
3970 #undef CHECK_APPEND_2ARG
3971 #undef CHECK_APPEND_1ARG
3972 #undef CHECK_APPEND_NOARG
3976 if (size_bp < size_buf)
3977 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3982 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3984 u32 size_buf = 1024;
3985 char tmp_buf[192] = {'\0'};
3988 u32 size_bp = size_buf;
3990 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3992 buf = kzalloc(size_buf, GFP_KERNEL);
3998 #define CHECK_APPEND_1ARG(a, v1) \
4000 ret = snprintf(bp, size_bp, (a), (v1)); \
4001 if (ret < 0 || ret >= size_bp) \
4002 goto out_overflow; \
4007 if (bctl->flags & BTRFS_BALANCE_FORCE)
4008 CHECK_APPEND_1ARG("%s", "-f ");
4010 if (bctl->flags & BTRFS_BALANCE_DATA) {
4011 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4012 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4015 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4016 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4017 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4020 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4021 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4022 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4025 #undef CHECK_APPEND_1ARG
4029 if (size_bp < size_buf)
4030 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4031 btrfs_info(fs_info, "balance: %s %s",
4032 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4033 "resume" : "start", buf);
4039 * Should be called with balance mutexe held
4041 int btrfs_balance(struct btrfs_fs_info *fs_info,
4042 struct btrfs_balance_control *bctl,
4043 struct btrfs_ioctl_balance_args *bargs)
4045 u64 meta_target, data_target;
4051 bool reducing_integrity;
4054 if (btrfs_fs_closing(fs_info) ||
4055 atomic_read(&fs_info->balance_pause_req) ||
4056 atomic_read(&fs_info->balance_cancel_req)) {
4061 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4062 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4066 * In case of mixed groups both data and meta should be picked,
4067 * and identical options should be given for both of them.
4069 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4070 if (mixed && (bctl->flags & allowed)) {
4071 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4072 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4073 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4075 "balance: mixed groups data and metadata options must be the same");
4081 num_devices = btrfs_num_devices(fs_info);
4083 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4084 if (num_devices >= btrfs_raid_array[i].devs_min)
4085 allowed |= btrfs_raid_array[i].bg_flag;
4087 if (validate_convert_profile(&bctl->data, allowed)) {
4089 "balance: invalid convert data profile %s",
4090 btrfs_bg_type_to_raid_name(bctl->data.target));
4094 if (validate_convert_profile(&bctl->meta, allowed)) {
4096 "balance: invalid convert metadata profile %s",
4097 btrfs_bg_type_to_raid_name(bctl->meta.target));
4101 if (validate_convert_profile(&bctl->sys, allowed)) {
4103 "balance: invalid convert system profile %s",
4104 btrfs_bg_type_to_raid_name(bctl->sys.target));
4110 * Allow to reduce metadata or system integrity only if force set for
4111 * profiles with redundancy (copies, parity)
4114 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4115 if (btrfs_raid_array[i].ncopies >= 2 ||
4116 btrfs_raid_array[i].tolerated_failures >= 1)
4117 allowed |= btrfs_raid_array[i].bg_flag;
4120 seq = read_seqbegin(&fs_info->profiles_lock);
4122 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4123 (fs_info->avail_system_alloc_bits & allowed) &&
4124 !(bctl->sys.target & allowed)) ||
4125 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4126 (fs_info->avail_metadata_alloc_bits & allowed) &&
4127 !(bctl->meta.target & allowed)))
4128 reducing_integrity = true;
4130 reducing_integrity = false;
4132 /* if we're not converting, the target field is uninitialized */
4133 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4134 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4135 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4136 bctl->data.target : fs_info->avail_data_alloc_bits;
4137 } while (read_seqretry(&fs_info->profiles_lock, seq));
4139 if (reducing_integrity) {
4140 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4142 "balance: force reducing metadata integrity");
4145 "balance: reduces metadata integrity, use --force if you want this");
4151 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4152 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4154 "balance: metadata profile %s has lower redundancy than data profile %s",
4155 btrfs_bg_type_to_raid_name(meta_target),
4156 btrfs_bg_type_to_raid_name(data_target));
4159 if (fs_info->send_in_progress) {
4160 btrfs_warn_rl(fs_info,
4161 "cannot run balance while send operations are in progress (%d in progress)",
4162 fs_info->send_in_progress);
4167 ret = insert_balance_item(fs_info, bctl);
4168 if (ret && ret != -EEXIST)
4171 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4172 BUG_ON(ret == -EEXIST);
4173 BUG_ON(fs_info->balance_ctl);
4174 spin_lock(&fs_info->balance_lock);
4175 fs_info->balance_ctl = bctl;
4176 spin_unlock(&fs_info->balance_lock);
4178 BUG_ON(ret != -EEXIST);
4179 spin_lock(&fs_info->balance_lock);
4180 update_balance_args(bctl);
4181 spin_unlock(&fs_info->balance_lock);
4184 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4185 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4186 describe_balance_start_or_resume(fs_info);
4187 mutex_unlock(&fs_info->balance_mutex);
4189 ret = __btrfs_balance(fs_info);
4191 mutex_lock(&fs_info->balance_mutex);
4192 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4193 btrfs_info(fs_info, "balance: paused");
4194 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4195 btrfs_info(fs_info, "balance: canceled");
4197 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4199 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4202 memset(bargs, 0, sizeof(*bargs));
4203 btrfs_update_ioctl_balance_args(fs_info, bargs);
4206 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4207 balance_need_close(fs_info)) {
4208 reset_balance_state(fs_info);
4209 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4212 wake_up(&fs_info->balance_wait_q);
4216 if (bctl->flags & BTRFS_BALANCE_RESUME)
4217 reset_balance_state(fs_info);
4220 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4225 static int balance_kthread(void *data)
4227 struct btrfs_fs_info *fs_info = data;
4230 mutex_lock(&fs_info->balance_mutex);
4231 if (fs_info->balance_ctl)
4232 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4233 mutex_unlock(&fs_info->balance_mutex);
4238 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4240 struct task_struct *tsk;
4242 mutex_lock(&fs_info->balance_mutex);
4243 if (!fs_info->balance_ctl) {
4244 mutex_unlock(&fs_info->balance_mutex);
4247 mutex_unlock(&fs_info->balance_mutex);
4249 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4250 btrfs_info(fs_info, "balance: resume skipped");
4255 * A ro->rw remount sequence should continue with the paused balance
4256 * regardless of who pauses it, system or the user as of now, so set
4259 spin_lock(&fs_info->balance_lock);
4260 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4261 spin_unlock(&fs_info->balance_lock);
4263 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4264 return PTR_ERR_OR_ZERO(tsk);
4267 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4269 struct btrfs_balance_control *bctl;
4270 struct btrfs_balance_item *item;
4271 struct btrfs_disk_balance_args disk_bargs;
4272 struct btrfs_path *path;
4273 struct extent_buffer *leaf;
4274 struct btrfs_key key;
4277 path = btrfs_alloc_path();
4281 key.objectid = BTRFS_BALANCE_OBJECTID;
4282 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4285 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4288 if (ret > 0) { /* ret = -ENOENT; */
4293 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4299 leaf = path->nodes[0];
4300 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4302 bctl->flags = btrfs_balance_flags(leaf, item);
4303 bctl->flags |= BTRFS_BALANCE_RESUME;
4305 btrfs_balance_data(leaf, item, &disk_bargs);
4306 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4307 btrfs_balance_meta(leaf, item, &disk_bargs);
4308 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4309 btrfs_balance_sys(leaf, item, &disk_bargs);
4310 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4313 * This should never happen, as the paused balance state is recovered
4314 * during mount without any chance of other exclusive ops to collide.
4316 * This gives the exclusive op status to balance and keeps in paused
4317 * state until user intervention (cancel or umount). If the ownership
4318 * cannot be assigned, show a message but do not fail. The balance
4319 * is in a paused state and must have fs_info::balance_ctl properly
4322 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4324 "balance: cannot set exclusive op status, resume manually");
4326 mutex_lock(&fs_info->balance_mutex);
4327 BUG_ON(fs_info->balance_ctl);
4328 spin_lock(&fs_info->balance_lock);
4329 fs_info->balance_ctl = bctl;
4330 spin_unlock(&fs_info->balance_lock);
4331 mutex_unlock(&fs_info->balance_mutex);
4333 btrfs_free_path(path);
4337 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4341 mutex_lock(&fs_info->balance_mutex);
4342 if (!fs_info->balance_ctl) {
4343 mutex_unlock(&fs_info->balance_mutex);
4347 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4348 atomic_inc(&fs_info->balance_pause_req);
4349 mutex_unlock(&fs_info->balance_mutex);
4351 wait_event(fs_info->balance_wait_q,
4352 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4354 mutex_lock(&fs_info->balance_mutex);
4355 /* we are good with balance_ctl ripped off from under us */
4356 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4357 atomic_dec(&fs_info->balance_pause_req);
4362 mutex_unlock(&fs_info->balance_mutex);
4366 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4368 mutex_lock(&fs_info->balance_mutex);
4369 if (!fs_info->balance_ctl) {
4370 mutex_unlock(&fs_info->balance_mutex);
4375 * A paused balance with the item stored on disk can be resumed at
4376 * mount time if the mount is read-write. Otherwise it's still paused
4377 * and we must not allow cancelling as it deletes the item.
4379 if (sb_rdonly(fs_info->sb)) {
4380 mutex_unlock(&fs_info->balance_mutex);
4384 atomic_inc(&fs_info->balance_cancel_req);
4386 * if we are running just wait and return, balance item is
4387 * deleted in btrfs_balance in this case
4389 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4390 mutex_unlock(&fs_info->balance_mutex);
4391 wait_event(fs_info->balance_wait_q,
4392 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4393 mutex_lock(&fs_info->balance_mutex);
4395 mutex_unlock(&fs_info->balance_mutex);
4397 * Lock released to allow other waiters to continue, we'll
4398 * reexamine the status again.
4400 mutex_lock(&fs_info->balance_mutex);
4402 if (fs_info->balance_ctl) {
4403 reset_balance_state(fs_info);
4404 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4405 btrfs_info(fs_info, "balance: canceled");
4409 BUG_ON(fs_info->balance_ctl ||
4410 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4411 atomic_dec(&fs_info->balance_cancel_req);
4412 mutex_unlock(&fs_info->balance_mutex);
4416 static int btrfs_uuid_scan_kthread(void *data)
4418 struct btrfs_fs_info *fs_info = data;
4419 struct btrfs_root *root = fs_info->tree_root;
4420 struct btrfs_key key;
4421 struct btrfs_path *path = NULL;
4423 struct extent_buffer *eb;
4425 struct btrfs_root_item root_item;
4427 struct btrfs_trans_handle *trans = NULL;
4429 path = btrfs_alloc_path();
4436 key.type = BTRFS_ROOT_ITEM_KEY;
4440 ret = btrfs_search_forward(root, &key, path,
4441 BTRFS_OLDEST_GENERATION);
4448 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4449 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4450 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4451 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4454 eb = path->nodes[0];
4455 slot = path->slots[0];
4456 item_size = btrfs_item_size_nr(eb, slot);
4457 if (item_size < sizeof(root_item))
4460 read_extent_buffer(eb, &root_item,
4461 btrfs_item_ptr_offset(eb, slot),
4462 (int)sizeof(root_item));
4463 if (btrfs_root_refs(&root_item) == 0)
4466 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4467 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4471 btrfs_release_path(path);
4473 * 1 - subvol uuid item
4474 * 1 - received_subvol uuid item
4476 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4477 if (IS_ERR(trans)) {
4478 ret = PTR_ERR(trans);
4486 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4487 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4488 BTRFS_UUID_KEY_SUBVOL,
4491 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4497 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4498 ret = btrfs_uuid_tree_add(trans,
4499 root_item.received_uuid,
4500 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4503 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4511 ret = btrfs_end_transaction(trans);
4517 btrfs_release_path(path);
4518 if (key.offset < (u64)-1) {
4520 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4522 key.type = BTRFS_ROOT_ITEM_KEY;
4523 } else if (key.objectid < (u64)-1) {
4525 key.type = BTRFS_ROOT_ITEM_KEY;
4534 btrfs_free_path(path);
4535 if (trans && !IS_ERR(trans))
4536 btrfs_end_transaction(trans);
4538 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4540 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4541 up(&fs_info->uuid_tree_rescan_sem);
4546 * Callback for btrfs_uuid_tree_iterate().
4548 * 0 check succeeded, the entry is not outdated.
4549 * < 0 if an error occurred.
4550 * > 0 if the check failed, which means the caller shall remove the entry.
4552 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4553 u8 *uuid, u8 type, u64 subid)
4555 struct btrfs_key key;
4557 struct btrfs_root *subvol_root;
4559 if (type != BTRFS_UUID_KEY_SUBVOL &&
4560 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4563 key.objectid = subid;
4564 key.type = BTRFS_ROOT_ITEM_KEY;
4565 key.offset = (u64)-1;
4566 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4567 if (IS_ERR(subvol_root)) {
4568 ret = PTR_ERR(subvol_root);
4575 case BTRFS_UUID_KEY_SUBVOL:
4576 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4579 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4580 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4590 static int btrfs_uuid_rescan_kthread(void *data)
4592 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4596 * 1st step is to iterate through the existing UUID tree and
4597 * to delete all entries that contain outdated data.
4598 * 2nd step is to add all missing entries to the UUID tree.
4600 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4602 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4603 up(&fs_info->uuid_tree_rescan_sem);
4606 return btrfs_uuid_scan_kthread(data);
4609 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4611 struct btrfs_trans_handle *trans;
4612 struct btrfs_root *tree_root = fs_info->tree_root;
4613 struct btrfs_root *uuid_root;
4614 struct task_struct *task;
4621 trans = btrfs_start_transaction(tree_root, 2);
4623 return PTR_ERR(trans);
4625 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4626 if (IS_ERR(uuid_root)) {
4627 ret = PTR_ERR(uuid_root);
4628 btrfs_abort_transaction(trans, ret);
4629 btrfs_end_transaction(trans);
4633 fs_info->uuid_root = uuid_root;
4635 ret = btrfs_commit_transaction(trans);
4639 down(&fs_info->uuid_tree_rescan_sem);
4640 task = kthread_run(btrfs_uuid_scan_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_scan task");
4644 up(&fs_info->uuid_tree_rescan_sem);
4645 return PTR_ERR(task);
4651 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4653 struct task_struct *task;
4655 down(&fs_info->uuid_tree_rescan_sem);
4656 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4658 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4659 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4660 up(&fs_info->uuid_tree_rescan_sem);
4661 return PTR_ERR(task);
4668 * shrinking a device means finding all of the device extents past
4669 * the new size, and then following the back refs to the chunks.
4670 * The chunk relocation code actually frees the device extent
4672 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4674 struct btrfs_fs_info *fs_info = device->fs_info;
4675 struct btrfs_root *root = fs_info->dev_root;
4676 struct btrfs_trans_handle *trans;
4677 struct btrfs_dev_extent *dev_extent = NULL;
4678 struct btrfs_path *path;
4684 bool retried = false;
4685 struct extent_buffer *l;
4686 struct btrfs_key key;
4687 struct btrfs_super_block *super_copy = fs_info->super_copy;
4688 u64 old_total = btrfs_super_total_bytes(super_copy);
4689 u64 old_size = btrfs_device_get_total_bytes(device);
4693 new_size = round_down(new_size, fs_info->sectorsize);
4695 diff = round_down(old_size - new_size, fs_info->sectorsize);
4697 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4700 path = btrfs_alloc_path();
4704 path->reada = READA_BACK;
4706 trans = btrfs_start_transaction(root, 0);
4707 if (IS_ERR(trans)) {
4708 btrfs_free_path(path);
4709 return PTR_ERR(trans);
4712 mutex_lock(&fs_info->chunk_mutex);
4714 btrfs_device_set_total_bytes(device, new_size);
4715 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4716 device->fs_devices->total_rw_bytes -= diff;
4717 atomic64_sub(diff, &fs_info->free_chunk_space);
4721 * Once the device's size has been set to the new size, ensure all
4722 * in-memory chunks are synced to disk so that the loop below sees them
4723 * and relocates them accordingly.
4725 if (contains_pending_extent(device, &start, diff)) {
4726 mutex_unlock(&fs_info->chunk_mutex);
4727 ret = btrfs_commit_transaction(trans);
4731 mutex_unlock(&fs_info->chunk_mutex);
4732 btrfs_end_transaction(trans);
4736 key.objectid = device->devid;
4737 key.offset = (u64)-1;
4738 key.type = BTRFS_DEV_EXTENT_KEY;
4741 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4742 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4744 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4748 ret = btrfs_previous_item(root, path, 0, key.type);
4750 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4755 btrfs_release_path(path);
4760 slot = path->slots[0];
4761 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4763 if (key.objectid != device->devid) {
4764 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4765 btrfs_release_path(path);
4769 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4770 length = btrfs_dev_extent_length(l, dev_extent);
4772 if (key.offset + length <= new_size) {
4773 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4774 btrfs_release_path(path);
4778 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4779 btrfs_release_path(path);
4782 * We may be relocating the only data chunk we have,
4783 * which could potentially end up with losing data's
4784 * raid profile, so lets allocate an empty one in
4787 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4789 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4793 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4794 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4795 if (ret == -ENOSPC) {
4798 if (ret == -ETXTBSY) {
4800 "could not shrink block group %llu due to active swapfile",
4805 } while (key.offset-- > 0);
4807 if (failed && !retried) {
4811 } else if (failed && retried) {
4816 /* Shrinking succeeded, else we would be at "done". */
4817 trans = btrfs_start_transaction(root, 0);
4818 if (IS_ERR(trans)) {
4819 ret = PTR_ERR(trans);
4823 mutex_lock(&fs_info->chunk_mutex);
4824 btrfs_device_set_disk_total_bytes(device, new_size);
4825 if (list_empty(&device->post_commit_list))
4826 list_add_tail(&device->post_commit_list,
4827 &trans->transaction->dev_update_list);
4829 WARN_ON(diff > old_total);
4830 btrfs_set_super_total_bytes(super_copy,
4831 round_down(old_total - diff, fs_info->sectorsize));
4832 mutex_unlock(&fs_info->chunk_mutex);
4834 /* Now btrfs_update_device() will change the on-disk size. */
4835 ret = btrfs_update_device(trans, device);
4837 btrfs_abort_transaction(trans, ret);
4838 btrfs_end_transaction(trans);
4840 ret = btrfs_commit_transaction(trans);
4843 btrfs_free_path(path);
4845 mutex_lock(&fs_info->chunk_mutex);
4846 btrfs_device_set_total_bytes(device, old_size);
4847 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4848 device->fs_devices->total_rw_bytes += diff;
4849 atomic64_add(diff, &fs_info->free_chunk_space);
4850 mutex_unlock(&fs_info->chunk_mutex);
4855 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4856 struct btrfs_key *key,
4857 struct btrfs_chunk *chunk, int item_size)
4859 struct btrfs_super_block *super_copy = fs_info->super_copy;
4860 struct btrfs_disk_key disk_key;
4864 mutex_lock(&fs_info->chunk_mutex);
4865 array_size = btrfs_super_sys_array_size(super_copy);
4866 if (array_size + item_size + sizeof(disk_key)
4867 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4868 mutex_unlock(&fs_info->chunk_mutex);
4872 ptr = super_copy->sys_chunk_array + array_size;
4873 btrfs_cpu_key_to_disk(&disk_key, key);
4874 memcpy(ptr, &disk_key, sizeof(disk_key));
4875 ptr += sizeof(disk_key);
4876 memcpy(ptr, chunk, item_size);
4877 item_size += sizeof(disk_key);
4878 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4879 mutex_unlock(&fs_info->chunk_mutex);
4885 * sort the devices in descending order by max_avail, total_avail
4887 static int btrfs_cmp_device_info(const void *a, const void *b)
4889 const struct btrfs_device_info *di_a = a;
4890 const struct btrfs_device_info *di_b = b;
4892 if (di_a->max_avail > di_b->max_avail)
4894 if (di_a->max_avail < di_b->max_avail)
4896 if (di_a->total_avail > di_b->total_avail)
4898 if (di_a->total_avail < di_b->total_avail)
4903 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4905 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4908 btrfs_set_fs_incompat(info, RAID56);
4911 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4912 u64 start, u64 type)
4914 struct btrfs_fs_info *info = trans->fs_info;
4915 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4916 struct btrfs_device *device;
4917 struct map_lookup *map = NULL;
4918 struct extent_map_tree *em_tree;
4919 struct extent_map *em;
4920 struct btrfs_device_info *devices_info = NULL;
4922 int num_stripes; /* total number of stripes to allocate */
4923 int data_stripes; /* number of stripes that count for
4925 int sub_stripes; /* sub_stripes info for map */
4926 int dev_stripes; /* stripes per dev */
4927 int devs_max; /* max devs to use */
4928 int devs_min; /* min devs needed */
4929 int devs_increment; /* ndevs has to be a multiple of this */
4930 int ncopies; /* how many copies to data has */
4931 int nparity; /* number of stripes worth of bytes to
4932 store parity information */
4934 u64 max_stripe_size;
4943 BUG_ON(!alloc_profile_is_valid(type, 0));
4945 if (list_empty(&fs_devices->alloc_list)) {
4946 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4947 btrfs_debug(info, "%s: no writable device", __func__);
4951 index = btrfs_bg_flags_to_raid_index(type);
4953 sub_stripes = btrfs_raid_array[index].sub_stripes;
4954 dev_stripes = btrfs_raid_array[index].dev_stripes;
4955 devs_max = btrfs_raid_array[index].devs_max;
4957 devs_max = BTRFS_MAX_DEVS(info);
4958 devs_min = btrfs_raid_array[index].devs_min;
4959 devs_increment = btrfs_raid_array[index].devs_increment;
4960 ncopies = btrfs_raid_array[index].ncopies;
4961 nparity = btrfs_raid_array[index].nparity;
4963 if (type & BTRFS_BLOCK_GROUP_DATA) {
4964 max_stripe_size = SZ_1G;
4965 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4966 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4967 /* for larger filesystems, use larger metadata chunks */
4968 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4969 max_stripe_size = SZ_1G;
4971 max_stripe_size = SZ_256M;
4972 max_chunk_size = max_stripe_size;
4973 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4974 max_stripe_size = SZ_32M;
4975 max_chunk_size = 2 * max_stripe_size;
4977 btrfs_err(info, "invalid chunk type 0x%llx requested",
4982 /* We don't want a chunk larger than 10% of writable space */
4983 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4986 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4992 * in the first pass through the devices list, we gather information
4993 * about the available holes on each device.
4996 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5000 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5002 "BTRFS: read-only device in alloc_list\n");
5006 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5007 &device->dev_state) ||
5008 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5011 if (device->total_bytes > device->bytes_used)
5012 total_avail = device->total_bytes - device->bytes_used;
5016 /* If there is no space on this device, skip it. */
5017 if (total_avail == 0)
5020 ret = find_free_dev_extent(device,
5021 max_stripe_size * dev_stripes,
5022 &dev_offset, &max_avail);
5023 if (ret && ret != -ENOSPC)
5027 max_avail = max_stripe_size * dev_stripes;
5029 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5030 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5032 "%s: devid %llu has no free space, have=%llu want=%u",
5033 __func__, device->devid, max_avail,
5034 BTRFS_STRIPE_LEN * dev_stripes);
5038 if (ndevs == fs_devices->rw_devices) {
5039 WARN(1, "%s: found more than %llu devices\n",
5040 __func__, fs_devices->rw_devices);
5043 devices_info[ndevs].dev_offset = dev_offset;
5044 devices_info[ndevs].max_avail = max_avail;
5045 devices_info[ndevs].total_avail = total_avail;
5046 devices_info[ndevs].dev = device;
5051 * now sort the devices by hole size / available space
5053 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5054 btrfs_cmp_device_info, NULL);
5056 /* round down to number of usable stripes */
5057 ndevs = round_down(ndevs, devs_increment);
5059 if (ndevs < devs_min) {
5061 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5063 "%s: not enough devices with free space: have=%d minimum required=%d",
5064 __func__, ndevs, devs_min);
5069 ndevs = min(ndevs, devs_max);
5072 * The primary goal is to maximize the number of stripes, so use as
5073 * many devices as possible, even if the stripes are not maximum sized.
5075 * The DUP profile stores more than one stripe per device, the
5076 * max_avail is the total size so we have to adjust.
5078 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5079 num_stripes = ndevs * dev_stripes;
5082 * this will have to be fixed for RAID1 and RAID10 over
5085 data_stripes = (num_stripes - nparity) / ncopies;
5088 * Use the number of data stripes to figure out how big this chunk
5089 * is really going to be in terms of logical address space,
5090 * and compare that answer with the max chunk size. If it's higher,
5091 * we try to reduce stripe_size.
5093 if (stripe_size * data_stripes > max_chunk_size) {
5095 * Reduce stripe_size, round it up to a 16MB boundary again and
5096 * then use it, unless it ends up being even bigger than the
5097 * previous value we had already.
5099 stripe_size = min(round_up(div_u64(max_chunk_size,
5100 data_stripes), SZ_16M),
5104 /* align to BTRFS_STRIPE_LEN */
5105 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5107 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5112 map->num_stripes = num_stripes;
5114 for (i = 0; i < ndevs; ++i) {
5115 for (j = 0; j < dev_stripes; ++j) {
5116 int s = i * dev_stripes + j;
5117 map->stripes[s].dev = devices_info[i].dev;
5118 map->stripes[s].physical = devices_info[i].dev_offset +
5122 map->stripe_len = BTRFS_STRIPE_LEN;
5123 map->io_align = BTRFS_STRIPE_LEN;
5124 map->io_width = BTRFS_STRIPE_LEN;
5126 map->sub_stripes = sub_stripes;
5128 chunk_size = stripe_size * data_stripes;
5130 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5132 em = alloc_extent_map();
5138 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5139 em->map_lookup = map;
5141 em->len = chunk_size;
5142 em->block_start = 0;
5143 em->block_len = em->len;
5144 em->orig_block_len = stripe_size;
5146 em_tree = &info->mapping_tree;
5147 write_lock(&em_tree->lock);
5148 ret = add_extent_mapping(em_tree, em, 0);
5150 write_unlock(&em_tree->lock);
5151 free_extent_map(em);
5154 write_unlock(&em_tree->lock);
5156 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5158 goto error_del_extent;
5160 for (i = 0; i < map->num_stripes; i++) {
5161 struct btrfs_device *dev = map->stripes[i].dev;
5163 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5164 if (list_empty(&dev->post_commit_list))
5165 list_add_tail(&dev->post_commit_list,
5166 &trans->transaction->dev_update_list);
5169 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5171 free_extent_map(em);
5172 check_raid56_incompat_flag(info, type);
5174 kfree(devices_info);
5178 write_lock(&em_tree->lock);
5179 remove_extent_mapping(em_tree, em);
5180 write_unlock(&em_tree->lock);
5182 /* One for our allocation */
5183 free_extent_map(em);
5184 /* One for the tree reference */
5185 free_extent_map(em);
5187 kfree(devices_info);
5191 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5192 u64 chunk_offset, u64 chunk_size)
5194 struct btrfs_fs_info *fs_info = trans->fs_info;
5195 struct btrfs_root *extent_root = fs_info->extent_root;
5196 struct btrfs_root *chunk_root = fs_info->chunk_root;
5197 struct btrfs_key key;
5198 struct btrfs_device *device;
5199 struct btrfs_chunk *chunk;
5200 struct btrfs_stripe *stripe;
5201 struct extent_map *em;
5202 struct map_lookup *map;
5209 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5213 map = em->map_lookup;
5214 item_size = btrfs_chunk_item_size(map->num_stripes);
5215 stripe_size = em->orig_block_len;
5217 chunk = kzalloc(item_size, GFP_NOFS);
5224 * Take the device list mutex to prevent races with the final phase of
5225 * a device replace operation that replaces the device object associated
5226 * with the map's stripes, because the device object's id can change
5227 * at any time during that final phase of the device replace operation
5228 * (dev-replace.c:btrfs_dev_replace_finishing()).
5230 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5231 for (i = 0; i < map->num_stripes; i++) {
5232 device = map->stripes[i].dev;
5233 dev_offset = map->stripes[i].physical;
5235 ret = btrfs_update_device(trans, device);
5238 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5239 dev_offset, stripe_size);
5244 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5248 stripe = &chunk->stripe;
5249 for (i = 0; i < map->num_stripes; i++) {
5250 device = map->stripes[i].dev;
5251 dev_offset = map->stripes[i].physical;
5253 btrfs_set_stack_stripe_devid(stripe, device->devid);
5254 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5255 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5258 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5260 btrfs_set_stack_chunk_length(chunk, chunk_size);
5261 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5262 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5263 btrfs_set_stack_chunk_type(chunk, map->type);
5264 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5265 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5266 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5267 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5268 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5270 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5271 key.type = BTRFS_CHUNK_ITEM_KEY;
5272 key.offset = chunk_offset;
5274 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5275 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5277 * TODO: Cleanup of inserted chunk root in case of
5280 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5285 free_extent_map(em);
5290 * Chunk allocation falls into two parts. The first part does work
5291 * that makes the new allocated chunk usable, but does not do any operation
5292 * that modifies the chunk tree. The second part does the work that
5293 * requires modifying the chunk tree. This division is important for the
5294 * bootstrap process of adding storage to a seed btrfs.
5296 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5300 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5301 chunk_offset = find_next_chunk(trans->fs_info);
5302 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5305 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5307 struct btrfs_fs_info *fs_info = trans->fs_info;
5309 u64 sys_chunk_offset;
5313 chunk_offset = find_next_chunk(fs_info);
5314 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5315 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5319 sys_chunk_offset = find_next_chunk(fs_info);
5320 alloc_profile = btrfs_system_alloc_profile(fs_info);
5321 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5325 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5327 const int index = btrfs_bg_flags_to_raid_index(map->type);
5329 return btrfs_raid_array[index].tolerated_failures;
5332 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5334 struct extent_map *em;
5335 struct map_lookup *map;
5340 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5344 map = em->map_lookup;
5345 for (i = 0; i < map->num_stripes; i++) {
5346 if (test_bit(BTRFS_DEV_STATE_MISSING,
5347 &map->stripes[i].dev->dev_state)) {
5351 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5352 &map->stripes[i].dev->dev_state)) {
5359 * If the number of missing devices is larger than max errors,
5360 * we can not write the data into that chunk successfully, so
5363 if (miss_ndevs > btrfs_chunk_max_errors(map))
5366 free_extent_map(em);
5370 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5372 struct extent_map *em;
5375 write_lock(&tree->lock);
5376 em = lookup_extent_mapping(tree, 0, (u64)-1);
5378 remove_extent_mapping(tree, em);
5379 write_unlock(&tree->lock);
5383 free_extent_map(em);
5384 /* once for the tree */
5385 free_extent_map(em);
5389 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5391 struct extent_map *em;
5392 struct map_lookup *map;
5395 em = btrfs_get_chunk_map(fs_info, logical, len);
5398 * We could return errors for these cases, but that could get
5399 * ugly and we'd probably do the same thing which is just not do
5400 * anything else and exit, so return 1 so the callers don't try
5401 * to use other copies.
5405 map = em->map_lookup;
5406 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5407 ret = map->num_stripes;
5408 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5409 ret = map->sub_stripes;
5410 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5412 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5414 * There could be two corrupted data stripes, we need
5415 * to loop retry in order to rebuild the correct data.
5417 * Fail a stripe at a time on every retry except the
5418 * stripe under reconstruction.
5420 ret = map->num_stripes;
5423 free_extent_map(em);
5425 down_read(&fs_info->dev_replace.rwsem);
5426 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5427 fs_info->dev_replace.tgtdev)
5429 up_read(&fs_info->dev_replace.rwsem);
5434 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5437 struct extent_map *em;
5438 struct map_lookup *map;
5439 unsigned long len = fs_info->sectorsize;
5441 em = btrfs_get_chunk_map(fs_info, logical, len);
5443 if (!WARN_ON(IS_ERR(em))) {
5444 map = em->map_lookup;
5445 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5446 len = map->stripe_len * nr_data_stripes(map);
5447 free_extent_map(em);
5452 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5454 struct extent_map *em;
5455 struct map_lookup *map;
5458 em = btrfs_get_chunk_map(fs_info, logical, len);
5460 if(!WARN_ON(IS_ERR(em))) {
5461 map = em->map_lookup;
5462 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5464 free_extent_map(em);
5469 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5470 struct map_lookup *map, int first,
5471 int dev_replace_is_ongoing)
5475 int preferred_mirror;
5477 struct btrfs_device *srcdev;
5480 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5482 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5483 num_stripes = map->sub_stripes;
5485 num_stripes = map->num_stripes;
5487 preferred_mirror = first + current->pid % num_stripes;
5489 if (dev_replace_is_ongoing &&
5490 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5491 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5492 srcdev = fs_info->dev_replace.srcdev;
5497 * try to avoid the drive that is the source drive for a
5498 * dev-replace procedure, only choose it if no other non-missing
5499 * mirror is available
5501 for (tolerance = 0; tolerance < 2; tolerance++) {
5502 if (map->stripes[preferred_mirror].dev->bdev &&
5503 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5504 return preferred_mirror;
5505 for (i = first; i < first + num_stripes; i++) {
5506 if (map->stripes[i].dev->bdev &&
5507 (tolerance || map->stripes[i].dev != srcdev))
5512 /* we couldn't find one that doesn't fail. Just return something
5513 * and the io error handling code will clean up eventually
5515 return preferred_mirror;
5518 static inline int parity_smaller(u64 a, u64 b)
5523 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5524 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5526 struct btrfs_bio_stripe s;
5533 for (i = 0; i < num_stripes - 1; i++) {
5534 if (parity_smaller(bbio->raid_map[i],
5535 bbio->raid_map[i+1])) {
5536 s = bbio->stripes[i];
5537 l = bbio->raid_map[i];
5538 bbio->stripes[i] = bbio->stripes[i+1];
5539 bbio->raid_map[i] = bbio->raid_map[i+1];
5540 bbio->stripes[i+1] = s;
5541 bbio->raid_map[i+1] = l;
5549 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5551 struct btrfs_bio *bbio = kzalloc(
5552 /* the size of the btrfs_bio */
5553 sizeof(struct btrfs_bio) +
5554 /* plus the variable array for the stripes */
5555 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5556 /* plus the variable array for the tgt dev */
5557 sizeof(int) * (real_stripes) +
5559 * plus the raid_map, which includes both the tgt dev
5562 sizeof(u64) * (total_stripes),
5563 GFP_NOFS|__GFP_NOFAIL);
5565 atomic_set(&bbio->error, 0);
5566 refcount_set(&bbio->refs, 1);
5571 void btrfs_get_bbio(struct btrfs_bio *bbio)
5573 WARN_ON(!refcount_read(&bbio->refs));
5574 refcount_inc(&bbio->refs);
5577 void btrfs_put_bbio(struct btrfs_bio *bbio)
5581 if (refcount_dec_and_test(&bbio->refs))
5585 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5587 * Please note that, discard won't be sent to target device of device
5590 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5591 u64 logical, u64 length,
5592 struct btrfs_bio **bbio_ret)
5594 struct extent_map *em;
5595 struct map_lookup *map;
5596 struct btrfs_bio *bbio;
5600 u64 stripe_end_offset;
5607 u32 sub_stripes = 0;
5608 u64 stripes_per_dev = 0;
5609 u32 remaining_stripes = 0;
5610 u32 last_stripe = 0;
5614 /* discard always return a bbio */
5617 em = btrfs_get_chunk_map(fs_info, logical, length);
5621 map = em->map_lookup;
5622 /* we don't discard raid56 yet */
5623 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5628 offset = logical - em->start;
5629 length = min_t(u64, em->len - offset, length);
5631 stripe_len = map->stripe_len;
5633 * stripe_nr counts the total number of stripes we have to stride
5634 * to get to this block
5636 stripe_nr = div64_u64(offset, stripe_len);
5638 /* stripe_offset is the offset of this block in its stripe */
5639 stripe_offset = offset - stripe_nr * stripe_len;
5641 stripe_nr_end = round_up(offset + length, map->stripe_len);
5642 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5643 stripe_cnt = stripe_nr_end - stripe_nr;
5644 stripe_end_offset = stripe_nr_end * map->stripe_len -
5647 * after this, stripe_nr is the number of stripes on this
5648 * device we have to walk to find the data, and stripe_index is
5649 * the number of our device in the stripe array
5653 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5654 BTRFS_BLOCK_GROUP_RAID10)) {
5655 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5658 sub_stripes = map->sub_stripes;
5660 factor = map->num_stripes / sub_stripes;
5661 num_stripes = min_t(u64, map->num_stripes,
5662 sub_stripes * stripe_cnt);
5663 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5664 stripe_index *= sub_stripes;
5665 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5666 &remaining_stripes);
5667 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5668 last_stripe *= sub_stripes;
5669 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5670 BTRFS_BLOCK_GROUP_DUP)) {
5671 num_stripes = map->num_stripes;
5673 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5677 bbio = alloc_btrfs_bio(num_stripes, 0);
5683 for (i = 0; i < num_stripes; i++) {
5684 bbio->stripes[i].physical =
5685 map->stripes[stripe_index].physical +
5686 stripe_offset + stripe_nr * map->stripe_len;
5687 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5689 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5690 BTRFS_BLOCK_GROUP_RAID10)) {
5691 bbio->stripes[i].length = stripes_per_dev *
5694 if (i / sub_stripes < remaining_stripes)
5695 bbio->stripes[i].length +=
5699 * Special for the first stripe and
5702 * |-------|...|-------|
5706 if (i < sub_stripes)
5707 bbio->stripes[i].length -=
5710 if (stripe_index >= last_stripe &&
5711 stripe_index <= (last_stripe +
5713 bbio->stripes[i].length -=
5716 if (i == sub_stripes - 1)
5719 bbio->stripes[i].length = length;
5723 if (stripe_index == map->num_stripes) {
5730 bbio->map_type = map->type;
5731 bbio->num_stripes = num_stripes;
5733 free_extent_map(em);
5738 * In dev-replace case, for repair case (that's the only case where the mirror
5739 * is selected explicitly when calling btrfs_map_block), blocks left of the
5740 * left cursor can also be read from the target drive.
5742 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5744 * For READ, it also needs to be supported using the same mirror number.
5746 * If the requested block is not left of the left cursor, EIO is returned. This
5747 * can happen because btrfs_num_copies() returns one more in the dev-replace
5750 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5751 u64 logical, u64 length,
5752 u64 srcdev_devid, int *mirror_num,
5755 struct btrfs_bio *bbio = NULL;
5757 int index_srcdev = 0;
5759 u64 physical_of_found = 0;
5763 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5764 logical, &length, &bbio, 0, 0);
5766 ASSERT(bbio == NULL);
5770 num_stripes = bbio->num_stripes;
5771 if (*mirror_num > num_stripes) {
5773 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5774 * that means that the requested area is not left of the left
5777 btrfs_put_bbio(bbio);
5782 * process the rest of the function using the mirror_num of the source
5783 * drive. Therefore look it up first. At the end, patch the device
5784 * pointer to the one of the target drive.
5786 for (i = 0; i < num_stripes; i++) {
5787 if (bbio->stripes[i].dev->devid != srcdev_devid)
5791 * In case of DUP, in order to keep it simple, only add the
5792 * mirror with the lowest physical address
5795 physical_of_found <= bbio->stripes[i].physical)
5800 physical_of_found = bbio->stripes[i].physical;
5803 btrfs_put_bbio(bbio);
5809 *mirror_num = index_srcdev + 1;
5810 *physical = physical_of_found;
5814 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5815 struct btrfs_bio **bbio_ret,
5816 struct btrfs_dev_replace *dev_replace,
5817 int *num_stripes_ret, int *max_errors_ret)
5819 struct btrfs_bio *bbio = *bbio_ret;
5820 u64 srcdev_devid = dev_replace->srcdev->devid;
5821 int tgtdev_indexes = 0;
5822 int num_stripes = *num_stripes_ret;
5823 int max_errors = *max_errors_ret;
5826 if (op == BTRFS_MAP_WRITE) {
5827 int index_where_to_add;
5830 * duplicate the write operations while the dev replace
5831 * procedure is running. Since the copying of the old disk to
5832 * the new disk takes place at run time while the filesystem is
5833 * mounted writable, the regular write operations to the old
5834 * disk have to be duplicated to go to the new disk as well.
5836 * Note that device->missing is handled by the caller, and that
5837 * the write to the old disk is already set up in the stripes
5840 index_where_to_add = num_stripes;
5841 for (i = 0; i < num_stripes; i++) {
5842 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5843 /* write to new disk, too */
5844 struct btrfs_bio_stripe *new =
5845 bbio->stripes + index_where_to_add;
5846 struct btrfs_bio_stripe *old =
5849 new->physical = old->physical;
5850 new->length = old->length;
5851 new->dev = dev_replace->tgtdev;
5852 bbio->tgtdev_map[i] = index_where_to_add;
5853 index_where_to_add++;
5858 num_stripes = index_where_to_add;
5859 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5860 int index_srcdev = 0;
5862 u64 physical_of_found = 0;
5865 * During the dev-replace procedure, the target drive can also
5866 * be used to read data in case it is needed to repair a corrupt
5867 * block elsewhere. This is possible if the requested area is
5868 * left of the left cursor. In this area, the target drive is a
5869 * full copy of the source drive.
5871 for (i = 0; i < num_stripes; i++) {
5872 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5874 * In case of DUP, in order to keep it simple,
5875 * only add the mirror with the lowest physical
5879 physical_of_found <=
5880 bbio->stripes[i].physical)
5884 physical_of_found = bbio->stripes[i].physical;
5888 struct btrfs_bio_stripe *tgtdev_stripe =
5889 bbio->stripes + num_stripes;
5891 tgtdev_stripe->physical = physical_of_found;
5892 tgtdev_stripe->length =
5893 bbio->stripes[index_srcdev].length;
5894 tgtdev_stripe->dev = dev_replace->tgtdev;
5895 bbio->tgtdev_map[index_srcdev] = num_stripes;
5902 *num_stripes_ret = num_stripes;
5903 *max_errors_ret = max_errors;
5904 bbio->num_tgtdevs = tgtdev_indexes;
5908 static bool need_full_stripe(enum btrfs_map_op op)
5910 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5914 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5915 * tuple. This information is used to calculate how big a
5916 * particular bio can get before it straddles a stripe.
5918 * @fs_info - the filesystem
5919 * @logical - address that we want to figure out the geometry of
5920 * @len - the length of IO we are going to perform, starting at @logical
5921 * @op - type of operation - write or read
5922 * @io_geom - pointer used to return values
5924 * Returns < 0 in case a chunk for the given logical address cannot be found,
5925 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5927 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5928 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5930 struct extent_map *em;
5931 struct map_lookup *map;
5936 u64 raid56_full_stripe_start = (u64)-1;
5940 ASSERT(op != BTRFS_MAP_DISCARD);
5942 em = btrfs_get_chunk_map(fs_info, logical, len);
5946 map = em->map_lookup;
5947 /* Offset of this logical address in the chunk */
5948 offset = logical - em->start;
5949 /* Len of a stripe in a chunk */
5950 stripe_len = map->stripe_len;
5951 /* Stripe wher this block falls in */
5952 stripe_nr = div64_u64(offset, stripe_len);
5953 /* Offset of stripe in the chunk */
5954 stripe_offset = stripe_nr * stripe_len;
5955 if (offset < stripe_offset) {
5957 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5958 stripe_offset, offset, em->start, logical, stripe_len);
5963 /* stripe_offset is the offset of this block in its stripe */
5964 stripe_offset = offset - stripe_offset;
5965 data_stripes = nr_data_stripes(map);
5967 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5968 u64 max_len = stripe_len - stripe_offset;
5971 * In case of raid56, we need to know the stripe aligned start
5973 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5974 unsigned long full_stripe_len = stripe_len * data_stripes;
5975 raid56_full_stripe_start = offset;
5978 * Allow a write of a full stripe, but make sure we
5979 * don't allow straddling of stripes
5981 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5983 raid56_full_stripe_start *= full_stripe_len;
5986 * For writes to RAID[56], allow a full stripeset across
5987 * all disks. For other RAID types and for RAID[56]
5988 * reads, just allow a single stripe (on a single disk).
5990 if (op == BTRFS_MAP_WRITE) {
5991 max_len = stripe_len * data_stripes -
5992 (offset - raid56_full_stripe_start);
5995 len = min_t(u64, em->len - offset, max_len);
5997 len = em->len - offset;
6001 io_geom->offset = offset;
6002 io_geom->stripe_len = stripe_len;
6003 io_geom->stripe_nr = stripe_nr;
6004 io_geom->stripe_offset = stripe_offset;
6005 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6009 free_extent_map(em);
6013 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6014 enum btrfs_map_op op,
6015 u64 logical, u64 *length,
6016 struct btrfs_bio **bbio_ret,
6017 int mirror_num, int need_raid_map)
6019 struct extent_map *em;
6020 struct map_lookup *map;
6030 int tgtdev_indexes = 0;
6031 struct btrfs_bio *bbio = NULL;
6032 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6033 int dev_replace_is_ongoing = 0;
6034 int num_alloc_stripes;
6035 int patch_the_first_stripe_for_dev_replace = 0;
6036 u64 physical_to_patch_in_first_stripe = 0;
6037 u64 raid56_full_stripe_start = (u64)-1;
6038 struct btrfs_io_geometry geom;
6042 if (op == BTRFS_MAP_DISCARD)
6043 return __btrfs_map_block_for_discard(fs_info, logical,
6046 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6050 em = btrfs_get_chunk_map(fs_info, logical, *length);
6052 map = em->map_lookup;
6055 stripe_len = geom.stripe_len;
6056 stripe_nr = geom.stripe_nr;
6057 stripe_offset = geom.stripe_offset;
6058 raid56_full_stripe_start = geom.raid56_stripe_offset;
6059 data_stripes = nr_data_stripes(map);
6061 down_read(&dev_replace->rwsem);
6062 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6064 * Hold the semaphore for read during the whole operation, write is
6065 * requested at commit time but must wait.
6067 if (!dev_replace_is_ongoing)
6068 up_read(&dev_replace->rwsem);
6070 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6071 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6072 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6073 dev_replace->srcdev->devid,
6075 &physical_to_patch_in_first_stripe);
6079 patch_the_first_stripe_for_dev_replace = 1;
6080 } else if (mirror_num > map->num_stripes) {
6086 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6087 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6089 if (!need_full_stripe(op))
6091 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6092 if (need_full_stripe(op))
6093 num_stripes = map->num_stripes;
6094 else if (mirror_num)
6095 stripe_index = mirror_num - 1;
6097 stripe_index = find_live_mirror(fs_info, map, 0,
6098 dev_replace_is_ongoing);
6099 mirror_num = stripe_index + 1;
6102 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6103 if (need_full_stripe(op)) {
6104 num_stripes = map->num_stripes;
6105 } else if (mirror_num) {
6106 stripe_index = mirror_num - 1;
6111 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6112 u32 factor = map->num_stripes / map->sub_stripes;
6114 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6115 stripe_index *= map->sub_stripes;
6117 if (need_full_stripe(op))
6118 num_stripes = map->sub_stripes;
6119 else if (mirror_num)
6120 stripe_index += mirror_num - 1;
6122 int old_stripe_index = stripe_index;
6123 stripe_index = find_live_mirror(fs_info, map,
6125 dev_replace_is_ongoing);
6126 mirror_num = stripe_index - old_stripe_index + 1;
6129 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6130 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6131 /* push stripe_nr back to the start of the full stripe */
6132 stripe_nr = div64_u64(raid56_full_stripe_start,
6133 stripe_len * data_stripes);
6135 /* RAID[56] write or recovery. Return all stripes */
6136 num_stripes = map->num_stripes;
6137 max_errors = nr_parity_stripes(map);
6139 *length = map->stripe_len;
6144 * Mirror #0 or #1 means the original data block.
6145 * Mirror #2 is RAID5 parity block.
6146 * Mirror #3 is RAID6 Q block.
6148 stripe_nr = div_u64_rem(stripe_nr,
6149 data_stripes, &stripe_index);
6151 stripe_index = data_stripes + mirror_num - 2;
6153 /* We distribute the parity blocks across stripes */
6154 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6156 if (!need_full_stripe(op) && mirror_num <= 1)
6161 * after this, stripe_nr is the number of stripes on this
6162 * device we have to walk to find the data, and stripe_index is
6163 * the number of our device in the stripe array
6165 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6167 mirror_num = stripe_index + 1;
6169 if (stripe_index >= map->num_stripes) {
6171 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6172 stripe_index, map->num_stripes);
6177 num_alloc_stripes = num_stripes;
6178 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6179 if (op == BTRFS_MAP_WRITE)
6180 num_alloc_stripes <<= 1;
6181 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6182 num_alloc_stripes++;
6183 tgtdev_indexes = num_stripes;
6186 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6191 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6192 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6194 /* build raid_map */
6195 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6196 (need_full_stripe(op) || mirror_num > 1)) {
6200 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6201 sizeof(struct btrfs_bio_stripe) *
6203 sizeof(int) * tgtdev_indexes);
6205 /* Work out the disk rotation on this stripe-set */
6206 div_u64_rem(stripe_nr, num_stripes, &rot);
6208 /* Fill in the logical address of each stripe */
6209 tmp = stripe_nr * data_stripes;
6210 for (i = 0; i < data_stripes; i++)
6211 bbio->raid_map[(i+rot) % num_stripes] =
6212 em->start + (tmp + i) * map->stripe_len;
6214 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6215 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6216 bbio->raid_map[(i+rot+1) % num_stripes] =
6221 for (i = 0; i < num_stripes; i++) {
6222 bbio->stripes[i].physical =
6223 map->stripes[stripe_index].physical +
6225 stripe_nr * map->stripe_len;
6226 bbio->stripes[i].dev =
6227 map->stripes[stripe_index].dev;
6231 if (need_full_stripe(op))
6232 max_errors = btrfs_chunk_max_errors(map);
6235 sort_parity_stripes(bbio, num_stripes);
6237 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6238 need_full_stripe(op)) {
6239 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6244 bbio->map_type = map->type;
6245 bbio->num_stripes = num_stripes;
6246 bbio->max_errors = max_errors;
6247 bbio->mirror_num = mirror_num;
6250 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6251 * mirror_num == num_stripes + 1 && dev_replace target drive is
6252 * available as a mirror
6254 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6255 WARN_ON(num_stripes > 1);
6256 bbio->stripes[0].dev = dev_replace->tgtdev;
6257 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6258 bbio->mirror_num = map->num_stripes + 1;
6261 if (dev_replace_is_ongoing) {
6262 lockdep_assert_held(&dev_replace->rwsem);
6263 /* Unlock and let waiting writers proceed */
6264 up_read(&dev_replace->rwsem);
6266 free_extent_map(em);
6270 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6271 u64 logical, u64 *length,
6272 struct btrfs_bio **bbio_ret, int mirror_num)
6274 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6278 /* For Scrub/replace */
6279 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6280 u64 logical, u64 *length,
6281 struct btrfs_bio **bbio_ret)
6283 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6286 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6287 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6289 struct extent_map *em;
6290 struct map_lookup *map;
6298 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6302 map = em->map_lookup;
6304 rmap_len = map->stripe_len;
6306 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6307 length = div_u64(length, map->num_stripes / map->sub_stripes);
6308 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6309 length = div_u64(length, map->num_stripes);
6310 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6311 length = div_u64(length, nr_data_stripes(map));
6312 rmap_len = map->stripe_len * nr_data_stripes(map);
6315 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6316 BUG_ON(!buf); /* -ENOMEM */
6318 for (i = 0; i < map->num_stripes; i++) {
6319 if (map->stripes[i].physical > physical ||
6320 map->stripes[i].physical + length <= physical)
6323 stripe_nr = physical - map->stripes[i].physical;
6324 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6326 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6327 stripe_nr = stripe_nr * map->num_stripes + i;
6328 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6329 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6330 stripe_nr = stripe_nr * map->num_stripes + i;
6331 } /* else if RAID[56], multiply by nr_data_stripes().
6332 * Alternatively, just use rmap_len below instead of
6333 * map->stripe_len */
6335 bytenr = chunk_start + stripe_nr * rmap_len;
6336 WARN_ON(nr >= map->num_stripes);
6337 for (j = 0; j < nr; j++) {
6338 if (buf[j] == bytenr)
6342 WARN_ON(nr >= map->num_stripes);
6349 *stripe_len = rmap_len;
6351 free_extent_map(em);
6355 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6357 bio->bi_private = bbio->private;
6358 bio->bi_end_io = bbio->end_io;
6361 btrfs_put_bbio(bbio);
6364 static void btrfs_end_bio(struct bio *bio)
6366 struct btrfs_bio *bbio = bio->bi_private;
6367 int is_orig_bio = 0;
6369 if (bio->bi_status) {
6370 atomic_inc(&bbio->error);
6371 if (bio->bi_status == BLK_STS_IOERR ||
6372 bio->bi_status == BLK_STS_TARGET) {
6373 unsigned int stripe_index =
6374 btrfs_io_bio(bio)->stripe_index;
6375 struct btrfs_device *dev;
6377 BUG_ON(stripe_index >= bbio->num_stripes);
6378 dev = bbio->stripes[stripe_index].dev;
6380 if (bio_op(bio) == REQ_OP_WRITE)
6381 btrfs_dev_stat_inc_and_print(dev,
6382 BTRFS_DEV_STAT_WRITE_ERRS);
6383 else if (!(bio->bi_opf & REQ_RAHEAD))
6384 btrfs_dev_stat_inc_and_print(dev,
6385 BTRFS_DEV_STAT_READ_ERRS);
6386 if (bio->bi_opf & REQ_PREFLUSH)
6387 btrfs_dev_stat_inc_and_print(dev,
6388 BTRFS_DEV_STAT_FLUSH_ERRS);
6393 if (bio == bbio->orig_bio)
6396 btrfs_bio_counter_dec(bbio->fs_info);
6398 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6401 bio = bbio->orig_bio;
6404 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6405 /* only send an error to the higher layers if it is
6406 * beyond the tolerance of the btrfs bio
6408 if (atomic_read(&bbio->error) > bbio->max_errors) {
6409 bio->bi_status = BLK_STS_IOERR;
6412 * this bio is actually up to date, we didn't
6413 * go over the max number of errors
6415 bio->bi_status = BLK_STS_OK;
6418 btrfs_end_bbio(bbio, bio);
6419 } else if (!is_orig_bio) {
6425 * see run_scheduled_bios for a description of why bios are collected for
6428 * This will add one bio to the pending list for a device and make sure
6429 * the work struct is scheduled.
6431 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6434 struct btrfs_fs_info *fs_info = device->fs_info;
6435 int should_queue = 1;
6436 struct btrfs_pending_bios *pending_bios;
6438 /* don't bother with additional async steps for reads, right now */
6439 if (bio_op(bio) == REQ_OP_READ) {
6440 btrfsic_submit_bio(bio);
6444 WARN_ON(bio->bi_next);
6445 bio->bi_next = NULL;
6447 spin_lock(&device->io_lock);
6448 if (op_is_sync(bio->bi_opf))
6449 pending_bios = &device->pending_sync_bios;
6451 pending_bios = &device->pending_bios;
6453 if (pending_bios->tail)
6454 pending_bios->tail->bi_next = bio;
6456 pending_bios->tail = bio;
6457 if (!pending_bios->head)
6458 pending_bios->head = bio;
6459 if (device->running_pending)
6462 spin_unlock(&device->io_lock);
6465 btrfs_queue_work(fs_info->submit_workers, &device->work);
6468 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6469 u64 physical, int dev_nr, int async)
6471 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6472 struct btrfs_fs_info *fs_info = bbio->fs_info;
6474 bio->bi_private = bbio;
6475 btrfs_io_bio(bio)->stripe_index = dev_nr;
6476 bio->bi_end_io = btrfs_end_bio;
6477 bio->bi_iter.bi_sector = physical >> 9;
6478 btrfs_debug_in_rcu(fs_info,
6479 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6480 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6481 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6482 bio->bi_iter.bi_size);
6483 bio_set_dev(bio, dev->bdev);
6485 btrfs_bio_counter_inc_noblocked(fs_info);
6488 btrfs_schedule_bio(dev, bio);
6490 btrfsic_submit_bio(bio);
6493 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6495 atomic_inc(&bbio->error);
6496 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6497 /* Should be the original bio. */
6498 WARN_ON(bio != bbio->orig_bio);
6500 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6501 bio->bi_iter.bi_sector = logical >> 9;
6502 if (atomic_read(&bbio->error) > bbio->max_errors)
6503 bio->bi_status = BLK_STS_IOERR;
6505 bio->bi_status = BLK_STS_OK;
6506 btrfs_end_bbio(bbio, bio);
6510 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6511 int mirror_num, int async_submit)
6513 struct btrfs_device *dev;
6514 struct bio *first_bio = bio;
6515 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6521 struct btrfs_bio *bbio = NULL;
6523 length = bio->bi_iter.bi_size;
6524 map_length = length;
6526 btrfs_bio_counter_inc_blocked(fs_info);
6527 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6528 &map_length, &bbio, mirror_num, 1);
6530 btrfs_bio_counter_dec(fs_info);
6531 return errno_to_blk_status(ret);
6534 total_devs = bbio->num_stripes;
6535 bbio->orig_bio = first_bio;
6536 bbio->private = first_bio->bi_private;
6537 bbio->end_io = first_bio->bi_end_io;
6538 bbio->fs_info = fs_info;
6539 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6541 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6542 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6543 /* In this case, map_length has been set to the length of
6544 a single stripe; not the whole write */
6545 if (bio_op(bio) == REQ_OP_WRITE) {
6546 ret = raid56_parity_write(fs_info, bio, bbio,
6549 ret = raid56_parity_recover(fs_info, bio, bbio,
6550 map_length, mirror_num, 1);
6553 btrfs_bio_counter_dec(fs_info);
6554 return errno_to_blk_status(ret);
6557 if (map_length < length) {
6559 "mapping failed logical %llu bio len %llu len %llu",
6560 logical, length, map_length);
6564 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6565 dev = bbio->stripes[dev_nr].dev;
6566 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6568 (bio_op(first_bio) == REQ_OP_WRITE &&
6569 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6570 bbio_error(bbio, first_bio, logical);
6574 if (dev_nr < total_devs - 1)
6575 bio = btrfs_bio_clone(first_bio);
6579 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6580 dev_nr, async_submit);
6582 btrfs_bio_counter_dec(fs_info);
6587 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6590 * If devid and uuid are both specified, the match must be exact, otherwise
6591 * only devid is used.
6593 * If @seed is true, traverse through the seed devices.
6595 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6596 u64 devid, u8 *uuid, u8 *fsid,
6599 struct btrfs_device *device;
6601 while (fs_devices) {
6603 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6604 list_for_each_entry(device, &fs_devices->devices,
6606 if (device->devid == devid &&
6607 (!uuid || memcmp(device->uuid, uuid,
6608 BTRFS_UUID_SIZE) == 0))
6613 fs_devices = fs_devices->seed;
6620 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6621 u64 devid, u8 *dev_uuid)
6623 struct btrfs_device *device;
6625 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6629 list_add(&device->dev_list, &fs_devices->devices);
6630 device->fs_devices = fs_devices;
6631 fs_devices->num_devices++;
6633 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6634 fs_devices->missing_devices++;
6640 * btrfs_alloc_device - allocate struct btrfs_device
6641 * @fs_info: used only for generating a new devid, can be NULL if
6642 * devid is provided (i.e. @devid != NULL).
6643 * @devid: a pointer to devid for this device. If NULL a new devid
6645 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6648 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6649 * on error. Returned struct is not linked onto any lists and must be
6650 * destroyed with btrfs_free_device.
6652 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6656 struct btrfs_device *dev;
6659 if (WARN_ON(!devid && !fs_info))
6660 return ERR_PTR(-EINVAL);
6662 dev = __alloc_device();
6671 ret = find_next_devid(fs_info, &tmp);
6673 btrfs_free_device(dev);
6674 return ERR_PTR(ret);
6680 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6682 generate_random_uuid(dev->uuid);
6684 btrfs_init_work(&dev->work, btrfs_submit_helper,
6685 pending_bios_fn, NULL, NULL);
6690 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6691 u64 devid, u8 *uuid, bool error)
6694 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6697 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6701 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6703 int index = btrfs_bg_flags_to_raid_index(type);
6704 int ncopies = btrfs_raid_array[index].ncopies;
6707 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6708 case BTRFS_BLOCK_GROUP_RAID5:
6709 data_stripes = num_stripes - 1;
6711 case BTRFS_BLOCK_GROUP_RAID6:
6712 data_stripes = num_stripes - 2;
6715 data_stripes = num_stripes / ncopies;
6718 return div_u64(chunk_len, data_stripes);
6721 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6722 struct btrfs_chunk *chunk)
6724 struct btrfs_fs_info *fs_info = leaf->fs_info;
6725 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6726 struct map_lookup *map;
6727 struct extent_map *em;
6731 u8 uuid[BTRFS_UUID_SIZE];
6736 logical = key->offset;
6737 length = btrfs_chunk_length(leaf, chunk);
6738 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6741 * Only need to verify chunk item if we're reading from sys chunk array,
6742 * as chunk item in tree block is already verified by tree-checker.
6744 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6745 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6750 read_lock(&map_tree->lock);
6751 em = lookup_extent_mapping(map_tree, logical, 1);
6752 read_unlock(&map_tree->lock);
6754 /* already mapped? */
6755 if (em && em->start <= logical && em->start + em->len > logical) {
6756 free_extent_map(em);
6759 free_extent_map(em);
6762 em = alloc_extent_map();
6765 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6767 free_extent_map(em);
6771 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6772 em->map_lookup = map;
6773 em->start = logical;
6776 em->block_start = 0;
6777 em->block_len = em->len;
6779 map->num_stripes = num_stripes;
6780 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6781 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6782 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6783 map->type = btrfs_chunk_type(leaf, chunk);
6784 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6785 map->verified_stripes = 0;
6786 em->orig_block_len = calc_stripe_length(map->type, em->len,
6788 for (i = 0; i < num_stripes; i++) {
6789 map->stripes[i].physical =
6790 btrfs_stripe_offset_nr(leaf, chunk, i);
6791 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6792 read_extent_buffer(leaf, uuid, (unsigned long)
6793 btrfs_stripe_dev_uuid_nr(chunk, i),
6795 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6796 devid, uuid, NULL, true);
6797 if (!map->stripes[i].dev &&
6798 !btrfs_test_opt(fs_info, DEGRADED)) {
6799 free_extent_map(em);
6800 btrfs_report_missing_device(fs_info, devid, uuid, true);
6803 if (!map->stripes[i].dev) {
6804 map->stripes[i].dev =
6805 add_missing_dev(fs_info->fs_devices, devid,
6807 if (IS_ERR(map->stripes[i].dev)) {
6808 free_extent_map(em);
6810 "failed to init missing dev %llu: %ld",
6811 devid, PTR_ERR(map->stripes[i].dev));
6812 return PTR_ERR(map->stripes[i].dev);
6814 btrfs_report_missing_device(fs_info, devid, uuid, false);
6816 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6817 &(map->stripes[i].dev->dev_state));
6821 write_lock(&map_tree->lock);
6822 ret = add_extent_mapping(map_tree, em, 0);
6823 write_unlock(&map_tree->lock);
6826 "failed to add chunk map, start=%llu len=%llu: %d",
6827 em->start, em->len, ret);
6829 free_extent_map(em);
6834 static void fill_device_from_item(struct extent_buffer *leaf,
6835 struct btrfs_dev_item *dev_item,
6836 struct btrfs_device *device)
6840 device->devid = btrfs_device_id(leaf, dev_item);
6841 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6842 device->total_bytes = device->disk_total_bytes;
6843 device->commit_total_bytes = device->disk_total_bytes;
6844 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6845 device->commit_bytes_used = device->bytes_used;
6846 device->type = btrfs_device_type(leaf, dev_item);
6847 device->io_align = btrfs_device_io_align(leaf, dev_item);
6848 device->io_width = btrfs_device_io_width(leaf, dev_item);
6849 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6850 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6851 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6853 ptr = btrfs_device_uuid(dev_item);
6854 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6857 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6860 struct btrfs_fs_devices *fs_devices;
6863 lockdep_assert_held(&uuid_mutex);
6866 fs_devices = fs_info->fs_devices->seed;
6867 while (fs_devices) {
6868 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6871 fs_devices = fs_devices->seed;
6874 fs_devices = find_fsid(fsid, NULL);
6876 if (!btrfs_test_opt(fs_info, DEGRADED))
6877 return ERR_PTR(-ENOENT);
6879 fs_devices = alloc_fs_devices(fsid, NULL);
6880 if (IS_ERR(fs_devices))
6883 fs_devices->seeding = 1;
6884 fs_devices->opened = 1;
6888 fs_devices = clone_fs_devices(fs_devices);
6889 if (IS_ERR(fs_devices))
6892 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6894 free_fs_devices(fs_devices);
6895 fs_devices = ERR_PTR(ret);
6899 if (!fs_devices->seeding) {
6900 close_fs_devices(fs_devices);
6901 free_fs_devices(fs_devices);
6902 fs_devices = ERR_PTR(-EINVAL);
6906 fs_devices->seed = fs_info->fs_devices->seed;
6907 fs_info->fs_devices->seed = fs_devices;
6912 static int read_one_dev(struct extent_buffer *leaf,
6913 struct btrfs_dev_item *dev_item)
6915 struct btrfs_fs_info *fs_info = leaf->fs_info;
6916 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6917 struct btrfs_device *device;
6920 u8 fs_uuid[BTRFS_FSID_SIZE];
6921 u8 dev_uuid[BTRFS_UUID_SIZE];
6923 devid = btrfs_device_id(leaf, dev_item);
6924 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6926 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6929 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6930 fs_devices = open_seed_devices(fs_info, fs_uuid);
6931 if (IS_ERR(fs_devices))
6932 return PTR_ERR(fs_devices);
6935 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6938 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6939 btrfs_report_missing_device(fs_info, devid,
6944 device = add_missing_dev(fs_devices, devid, dev_uuid);
6945 if (IS_ERR(device)) {
6947 "failed to add missing dev %llu: %ld",
6948 devid, PTR_ERR(device));
6949 return PTR_ERR(device);
6951 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6953 if (!device->bdev) {
6954 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6955 btrfs_report_missing_device(fs_info,
6956 devid, dev_uuid, true);
6959 btrfs_report_missing_device(fs_info, devid,
6963 if (!device->bdev &&
6964 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6966 * this happens when a device that was properly setup
6967 * in the device info lists suddenly goes bad.
6968 * device->bdev is NULL, and so we have to set
6969 * device->missing to one here
6971 device->fs_devices->missing_devices++;
6972 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6975 /* Move the device to its own fs_devices */
6976 if (device->fs_devices != fs_devices) {
6977 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6978 &device->dev_state));
6980 list_move(&device->dev_list, &fs_devices->devices);
6981 device->fs_devices->num_devices--;
6982 fs_devices->num_devices++;
6984 device->fs_devices->missing_devices--;
6985 fs_devices->missing_devices++;
6987 device->fs_devices = fs_devices;
6991 if (device->fs_devices != fs_info->fs_devices) {
6992 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6993 if (device->generation !=
6994 btrfs_device_generation(leaf, dev_item))
6998 fill_device_from_item(leaf, dev_item, device);
6999 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7000 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7001 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7002 device->fs_devices->total_rw_bytes += device->total_bytes;
7003 atomic64_add(device->total_bytes - device->bytes_used,
7004 &fs_info->free_chunk_space);
7010 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7012 struct btrfs_root *root = fs_info->tree_root;
7013 struct btrfs_super_block *super_copy = fs_info->super_copy;
7014 struct extent_buffer *sb;
7015 struct btrfs_disk_key *disk_key;
7016 struct btrfs_chunk *chunk;
7018 unsigned long sb_array_offset;
7025 struct btrfs_key key;
7027 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7029 * This will create extent buffer of nodesize, superblock size is
7030 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7031 * overallocate but we can keep it as-is, only the first page is used.
7033 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7036 set_extent_buffer_uptodate(sb);
7037 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7039 * The sb extent buffer is artificial and just used to read the system array.
7040 * set_extent_buffer_uptodate() call does not properly mark all it's
7041 * pages up-to-date when the page is larger: extent does not cover the
7042 * whole page and consequently check_page_uptodate does not find all
7043 * the page's extents up-to-date (the hole beyond sb),
7044 * write_extent_buffer then triggers a WARN_ON.
7046 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7047 * but sb spans only this function. Add an explicit SetPageUptodate call
7048 * to silence the warning eg. on PowerPC 64.
7050 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7051 SetPageUptodate(sb->pages[0]);
7053 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7054 array_size = btrfs_super_sys_array_size(super_copy);
7056 array_ptr = super_copy->sys_chunk_array;
7057 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7060 while (cur_offset < array_size) {
7061 disk_key = (struct btrfs_disk_key *)array_ptr;
7062 len = sizeof(*disk_key);
7063 if (cur_offset + len > array_size)
7064 goto out_short_read;
7066 btrfs_disk_key_to_cpu(&key, disk_key);
7069 sb_array_offset += len;
7072 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7073 chunk = (struct btrfs_chunk *)sb_array_offset;
7075 * At least one btrfs_chunk with one stripe must be
7076 * present, exact stripe count check comes afterwards
7078 len = btrfs_chunk_item_size(1);
7079 if (cur_offset + len > array_size)
7080 goto out_short_read;
7082 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7085 "invalid number of stripes %u in sys_array at offset %u",
7086 num_stripes, cur_offset);
7091 type = btrfs_chunk_type(sb, chunk);
7092 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7094 "invalid chunk type %llu in sys_array at offset %u",
7100 len = btrfs_chunk_item_size(num_stripes);
7101 if (cur_offset + len > array_size)
7102 goto out_short_read;
7104 ret = read_one_chunk(&key, sb, chunk);
7109 "unexpected item type %u in sys_array at offset %u",
7110 (u32)key.type, cur_offset);
7115 sb_array_offset += len;
7118 clear_extent_buffer_uptodate(sb);
7119 free_extent_buffer_stale(sb);
7123 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7125 clear_extent_buffer_uptodate(sb);
7126 free_extent_buffer_stale(sb);
7131 * Check if all chunks in the fs are OK for read-write degraded mount
7133 * If the @failing_dev is specified, it's accounted as missing.
7135 * Return true if all chunks meet the minimal RW mount requirements.
7136 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7138 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7139 struct btrfs_device *failing_dev)
7141 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7142 struct extent_map *em;
7146 read_lock(&map_tree->lock);
7147 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7148 read_unlock(&map_tree->lock);
7149 /* No chunk at all? Return false anyway */
7155 struct map_lookup *map;
7160 map = em->map_lookup;
7162 btrfs_get_num_tolerated_disk_barrier_failures(
7164 for (i = 0; i < map->num_stripes; i++) {
7165 struct btrfs_device *dev = map->stripes[i].dev;
7167 if (!dev || !dev->bdev ||
7168 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7169 dev->last_flush_error)
7171 else if (failing_dev && failing_dev == dev)
7174 if (missing > max_tolerated) {
7177 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7178 em->start, missing, max_tolerated);
7179 free_extent_map(em);
7183 next_start = extent_map_end(em);
7184 free_extent_map(em);
7186 read_lock(&map_tree->lock);
7187 em = lookup_extent_mapping(map_tree, next_start,
7188 (u64)(-1) - next_start);
7189 read_unlock(&map_tree->lock);
7195 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7197 struct btrfs_root *root = fs_info->chunk_root;
7198 struct btrfs_path *path;
7199 struct extent_buffer *leaf;
7200 struct btrfs_key key;
7201 struct btrfs_key found_key;
7206 path = btrfs_alloc_path();
7211 * uuid_mutex is needed only if we are mounting a sprout FS
7212 * otherwise we don't need it.
7214 mutex_lock(&uuid_mutex);
7215 mutex_lock(&fs_info->chunk_mutex);
7218 * Read all device items, and then all the chunk items. All
7219 * device items are found before any chunk item (their object id
7220 * is smaller than the lowest possible object id for a chunk
7221 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7223 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7226 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7230 leaf = path->nodes[0];
7231 slot = path->slots[0];
7232 if (slot >= btrfs_header_nritems(leaf)) {
7233 ret = btrfs_next_leaf(root, path);
7240 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7241 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7242 struct btrfs_dev_item *dev_item;
7243 dev_item = btrfs_item_ptr(leaf, slot,
7244 struct btrfs_dev_item);
7245 ret = read_one_dev(leaf, dev_item);
7249 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7250 struct btrfs_chunk *chunk;
7251 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7252 ret = read_one_chunk(&found_key, leaf, chunk);
7260 * After loading chunk tree, we've got all device information,
7261 * do another round of validation checks.
7263 if (total_dev != fs_info->fs_devices->total_devices) {
7265 "super_num_devices %llu mismatch with num_devices %llu found here",
7266 btrfs_super_num_devices(fs_info->super_copy),
7271 if (btrfs_super_total_bytes(fs_info->super_copy) <
7272 fs_info->fs_devices->total_rw_bytes) {
7274 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7275 btrfs_super_total_bytes(fs_info->super_copy),
7276 fs_info->fs_devices->total_rw_bytes);
7282 mutex_unlock(&fs_info->chunk_mutex);
7283 mutex_unlock(&uuid_mutex);
7285 btrfs_free_path(path);
7289 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7291 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7292 struct btrfs_device *device;
7294 while (fs_devices) {
7295 mutex_lock(&fs_devices->device_list_mutex);
7296 list_for_each_entry(device, &fs_devices->devices, dev_list)
7297 device->fs_info = fs_info;
7298 mutex_unlock(&fs_devices->device_list_mutex);
7300 fs_devices = fs_devices->seed;
7304 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7308 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7309 btrfs_dev_stat_reset(dev, i);
7312 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7314 struct btrfs_key key;
7315 struct btrfs_key found_key;
7316 struct btrfs_root *dev_root = fs_info->dev_root;
7317 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7318 struct extent_buffer *eb;
7321 struct btrfs_device *device;
7322 struct btrfs_path *path = NULL;
7325 path = btrfs_alloc_path();
7331 mutex_lock(&fs_devices->device_list_mutex);
7332 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7334 struct btrfs_dev_stats_item *ptr;
7336 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7337 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7338 key.offset = device->devid;
7339 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7341 __btrfs_reset_dev_stats(device);
7342 device->dev_stats_valid = 1;
7343 btrfs_release_path(path);
7346 slot = path->slots[0];
7347 eb = path->nodes[0];
7348 btrfs_item_key_to_cpu(eb, &found_key, slot);
7349 item_size = btrfs_item_size_nr(eb, slot);
7351 ptr = btrfs_item_ptr(eb, slot,
7352 struct btrfs_dev_stats_item);
7354 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7355 if (item_size >= (1 + i) * sizeof(__le64))
7356 btrfs_dev_stat_set(device, i,
7357 btrfs_dev_stats_value(eb, ptr, i));
7359 btrfs_dev_stat_reset(device, i);
7362 device->dev_stats_valid = 1;
7363 btrfs_dev_stat_print_on_load(device);
7364 btrfs_release_path(path);
7366 mutex_unlock(&fs_devices->device_list_mutex);
7369 btrfs_free_path(path);
7370 return ret < 0 ? ret : 0;
7373 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7374 struct btrfs_device *device)
7376 struct btrfs_fs_info *fs_info = trans->fs_info;
7377 struct btrfs_root *dev_root = fs_info->dev_root;
7378 struct btrfs_path *path;
7379 struct btrfs_key key;
7380 struct extent_buffer *eb;
7381 struct btrfs_dev_stats_item *ptr;
7385 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7386 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7387 key.offset = device->devid;
7389 path = btrfs_alloc_path();
7392 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7394 btrfs_warn_in_rcu(fs_info,
7395 "error %d while searching for dev_stats item for device %s",
7396 ret, rcu_str_deref(device->name));
7401 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7402 /* need to delete old one and insert a new one */
7403 ret = btrfs_del_item(trans, dev_root, path);
7405 btrfs_warn_in_rcu(fs_info,
7406 "delete too small dev_stats item for device %s failed %d",
7407 rcu_str_deref(device->name), ret);
7414 /* need to insert a new item */
7415 btrfs_release_path(path);
7416 ret = btrfs_insert_empty_item(trans, dev_root, path,
7417 &key, sizeof(*ptr));
7419 btrfs_warn_in_rcu(fs_info,
7420 "insert dev_stats item for device %s failed %d",
7421 rcu_str_deref(device->name), ret);
7426 eb = path->nodes[0];
7427 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7428 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7429 btrfs_set_dev_stats_value(eb, ptr, i,
7430 btrfs_dev_stat_read(device, i));
7431 btrfs_mark_buffer_dirty(eb);
7434 btrfs_free_path(path);
7439 * called from commit_transaction. Writes all changed device stats to disk.
7441 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7443 struct btrfs_fs_info *fs_info = trans->fs_info;
7444 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7445 struct btrfs_device *device;
7449 mutex_lock(&fs_devices->device_list_mutex);
7450 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7451 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7452 if (!device->dev_stats_valid || stats_cnt == 0)
7457 * There is a LOAD-LOAD control dependency between the value of
7458 * dev_stats_ccnt and updating the on-disk values which requires
7459 * reading the in-memory counters. Such control dependencies
7460 * require explicit read memory barriers.
7462 * This memory barriers pairs with smp_mb__before_atomic in
7463 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7464 * barrier implied by atomic_xchg in
7465 * btrfs_dev_stats_read_and_reset
7469 ret = update_dev_stat_item(trans, device);
7471 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7473 mutex_unlock(&fs_devices->device_list_mutex);
7478 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7480 btrfs_dev_stat_inc(dev, index);
7481 btrfs_dev_stat_print_on_error(dev);
7484 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7486 if (!dev->dev_stats_valid)
7488 btrfs_err_rl_in_rcu(dev->fs_info,
7489 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7490 rcu_str_deref(dev->name),
7491 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7492 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7493 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7494 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7495 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7498 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7502 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7503 if (btrfs_dev_stat_read(dev, i) != 0)
7505 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7506 return; /* all values == 0, suppress message */
7508 btrfs_info_in_rcu(dev->fs_info,
7509 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7510 rcu_str_deref(dev->name),
7511 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7512 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7513 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7514 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7515 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7518 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7519 struct btrfs_ioctl_get_dev_stats *stats)
7521 struct btrfs_device *dev;
7522 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7525 mutex_lock(&fs_devices->device_list_mutex);
7526 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7528 mutex_unlock(&fs_devices->device_list_mutex);
7531 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7533 } else if (!dev->dev_stats_valid) {
7534 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7536 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7537 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7538 if (stats->nr_items > i)
7540 btrfs_dev_stat_read_and_reset(dev, i);
7542 btrfs_dev_stat_reset(dev, i);
7545 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7546 if (stats->nr_items > i)
7547 stats->values[i] = btrfs_dev_stat_read(dev, i);
7549 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7550 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7554 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7556 struct buffer_head *bh;
7557 struct btrfs_super_block *disk_super;
7563 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7566 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7569 disk_super = (struct btrfs_super_block *)bh->b_data;
7571 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7572 set_buffer_dirty(bh);
7573 sync_dirty_buffer(bh);
7577 /* Notify udev that device has changed */
7578 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7580 /* Update ctime/mtime for device path for libblkid */
7581 update_dev_time(device_path);
7585 * Update the size and bytes used for each device where it changed. This is
7586 * delayed since we would otherwise get errors while writing out the
7589 * Must be invoked during transaction commit.
7591 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7593 struct btrfs_device *curr, *next;
7595 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7597 if (list_empty(&trans->dev_update_list))
7601 * We don't need the device_list_mutex here. This list is owned by the
7602 * transaction and the transaction must complete before the device is
7605 mutex_lock(&trans->fs_info->chunk_mutex);
7606 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7608 list_del_init(&curr->post_commit_list);
7609 curr->commit_total_bytes = curr->disk_total_bytes;
7610 curr->commit_bytes_used = curr->bytes_used;
7612 mutex_unlock(&trans->fs_info->chunk_mutex);
7615 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7617 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7618 while (fs_devices) {
7619 fs_devices->fs_info = fs_info;
7620 fs_devices = fs_devices->seed;
7624 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7626 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7627 while (fs_devices) {
7628 fs_devices->fs_info = NULL;
7629 fs_devices = fs_devices->seed;
7634 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7636 int btrfs_bg_type_to_factor(u64 flags)
7638 const int index = btrfs_bg_flags_to_raid_index(flags);
7640 return btrfs_raid_array[index].ncopies;
7645 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7646 u64 chunk_offset, u64 devid,
7647 u64 physical_offset, u64 physical_len)
7649 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7650 struct extent_map *em;
7651 struct map_lookup *map;
7652 struct btrfs_device *dev;
7658 read_lock(&em_tree->lock);
7659 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7660 read_unlock(&em_tree->lock);
7664 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7665 physical_offset, devid);
7670 map = em->map_lookup;
7671 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7672 if (physical_len != stripe_len) {
7674 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7675 physical_offset, devid, em->start, physical_len,
7681 for (i = 0; i < map->num_stripes; i++) {
7682 if (map->stripes[i].dev->devid == devid &&
7683 map->stripes[i].physical == physical_offset) {
7685 if (map->verified_stripes >= map->num_stripes) {
7687 "too many dev extents for chunk %llu found",
7692 map->verified_stripes++;
7698 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7699 physical_offset, devid);
7703 /* Make sure no dev extent is beyond device bondary */
7704 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7706 btrfs_err(fs_info, "failed to find devid %llu", devid);
7711 /* It's possible this device is a dummy for seed device */
7712 if (dev->disk_total_bytes == 0) {
7713 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7716 btrfs_err(fs_info, "failed to find seed devid %llu",
7723 if (physical_offset + physical_len > dev->disk_total_bytes) {
7725 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7726 devid, physical_offset, physical_len,
7727 dev->disk_total_bytes);
7732 free_extent_map(em);
7736 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7738 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7739 struct extent_map *em;
7740 struct rb_node *node;
7743 read_lock(&em_tree->lock);
7744 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7745 em = rb_entry(node, struct extent_map, rb_node);
7746 if (em->map_lookup->num_stripes !=
7747 em->map_lookup->verified_stripes) {
7749 "chunk %llu has missing dev extent, have %d expect %d",
7750 em->start, em->map_lookup->verified_stripes,
7751 em->map_lookup->num_stripes);
7757 read_unlock(&em_tree->lock);
7762 * Ensure that all dev extents are mapped to correct chunk, otherwise
7763 * later chunk allocation/free would cause unexpected behavior.
7765 * NOTE: This will iterate through the whole device tree, which should be of
7766 * the same size level as the chunk tree. This slightly increases mount time.
7768 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7770 struct btrfs_path *path;
7771 struct btrfs_root *root = fs_info->dev_root;
7772 struct btrfs_key key;
7774 u64 prev_dev_ext_end = 0;
7778 key.type = BTRFS_DEV_EXTENT_KEY;
7781 path = btrfs_alloc_path();
7785 path->reada = READA_FORWARD;
7786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7790 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7791 ret = btrfs_next_item(root, path);
7794 /* No dev extents at all? Not good */
7801 struct extent_buffer *leaf = path->nodes[0];
7802 struct btrfs_dev_extent *dext;
7803 int slot = path->slots[0];
7805 u64 physical_offset;
7809 btrfs_item_key_to_cpu(leaf, &key, slot);
7810 if (key.type != BTRFS_DEV_EXTENT_KEY)
7812 devid = key.objectid;
7813 physical_offset = key.offset;
7815 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7816 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7817 physical_len = btrfs_dev_extent_length(leaf, dext);
7819 /* Check if this dev extent overlaps with the previous one */
7820 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7822 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7823 devid, physical_offset, prev_dev_ext_end);
7828 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7829 physical_offset, physical_len);
7833 prev_dev_ext_end = physical_offset + physical_len;
7835 ret = btrfs_next_item(root, path);
7844 /* Ensure all chunks have corresponding dev extents */
7845 ret = verify_chunk_dev_extent_mapping(fs_info);
7847 btrfs_free_path(path);
7852 * Check whether the given block group or device is pinned by any inode being
7853 * used as a swapfile.
7855 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7857 struct btrfs_swapfile_pin *sp;
7858 struct rb_node *node;
7860 spin_lock(&fs_info->swapfile_pins_lock);
7861 node = fs_info->swapfile_pins.rb_node;
7863 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7865 node = node->rb_left;
7866 else if (ptr > sp->ptr)
7867 node = node->rb_right;
7871 spin_unlock(&fs_info->swapfile_pins_lock);
7872 return node != NULL;