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 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
1556 u64 search_start, u64 *start, u64 *len)
1558 struct btrfs_fs_info *fs_info = device->fs_info;
1559 struct btrfs_root *root = fs_info->dev_root;
1560 struct btrfs_key key;
1561 struct btrfs_dev_extent *dev_extent;
1562 struct btrfs_path *path;
1567 u64 search_end = device->total_bytes;
1570 struct extent_buffer *l;
1573 * We don't want to overwrite the superblock on the drive nor any area
1574 * used by the boot loader (grub for example), so we make sure to start
1575 * at an offset of at least 1MB.
1577 search_start = max_t(u64, search_start, SZ_1M);
1579 path = btrfs_alloc_path();
1583 max_hole_start = search_start;
1587 if (search_start >= search_end ||
1588 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1593 path->reada = READA_FORWARD;
1594 path->search_commit_root = 1;
1595 path->skip_locking = 1;
1597 key.objectid = device->devid;
1598 key.offset = search_start;
1599 key.type = BTRFS_DEV_EXTENT_KEY;
1601 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1605 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1612 slot = path->slots[0];
1613 if (slot >= btrfs_header_nritems(l)) {
1614 ret = btrfs_next_leaf(root, path);
1622 btrfs_item_key_to_cpu(l, &key, slot);
1624 if (key.objectid < device->devid)
1627 if (key.objectid > device->devid)
1630 if (key.type != BTRFS_DEV_EXTENT_KEY)
1633 if (key.offset > search_start) {
1634 hole_size = key.offset - search_start;
1637 * Have to check before we set max_hole_start, otherwise
1638 * we could end up sending back this offset anyway.
1640 if (contains_pending_extent(device, &search_start,
1642 if (key.offset >= search_start)
1643 hole_size = key.offset - search_start;
1648 if (hole_size > max_hole_size) {
1649 max_hole_start = search_start;
1650 max_hole_size = hole_size;
1654 * If this free space is greater than which we need,
1655 * it must be the max free space that we have found
1656 * until now, so max_hole_start must point to the start
1657 * of this free space and the length of this free space
1658 * is stored in max_hole_size. Thus, we return
1659 * max_hole_start and max_hole_size and go back to the
1662 if (hole_size >= num_bytes) {
1668 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1669 extent_end = key.offset + btrfs_dev_extent_length(l,
1671 if (extent_end > search_start)
1672 search_start = extent_end;
1679 * At this point, search_start should be the end of
1680 * allocated dev extents, and when shrinking the device,
1681 * search_end may be smaller than search_start.
1683 if (search_end > search_start) {
1684 hole_size = search_end - search_start;
1686 if (contains_pending_extent(device, &search_start, hole_size)) {
1687 btrfs_release_path(path);
1691 if (hole_size > max_hole_size) {
1692 max_hole_start = search_start;
1693 max_hole_size = hole_size;
1698 if (max_hole_size < num_bytes)
1704 btrfs_free_path(path);
1705 *start = max_hole_start;
1707 *len = max_hole_size;
1711 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1712 u64 *start, u64 *len)
1714 /* FIXME use last free of some kind */
1715 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1718 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1719 struct btrfs_device *device,
1720 u64 start, u64 *dev_extent_len)
1722 struct btrfs_fs_info *fs_info = device->fs_info;
1723 struct btrfs_root *root = fs_info->dev_root;
1725 struct btrfs_path *path;
1726 struct btrfs_key key;
1727 struct btrfs_key found_key;
1728 struct extent_buffer *leaf = NULL;
1729 struct btrfs_dev_extent *extent = NULL;
1731 path = btrfs_alloc_path();
1735 key.objectid = device->devid;
1737 key.type = BTRFS_DEV_EXTENT_KEY;
1739 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1741 ret = btrfs_previous_item(root, path, key.objectid,
1742 BTRFS_DEV_EXTENT_KEY);
1745 leaf = path->nodes[0];
1746 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1747 extent = btrfs_item_ptr(leaf, path->slots[0],
1748 struct btrfs_dev_extent);
1749 BUG_ON(found_key.offset > start || found_key.offset +
1750 btrfs_dev_extent_length(leaf, extent) < start);
1752 btrfs_release_path(path);
1754 } else if (ret == 0) {
1755 leaf = path->nodes[0];
1756 extent = btrfs_item_ptr(leaf, path->slots[0],
1757 struct btrfs_dev_extent);
1759 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1763 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1765 ret = btrfs_del_item(trans, root, path);
1767 btrfs_handle_fs_error(fs_info, ret,
1768 "Failed to remove dev extent item");
1770 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1773 btrfs_free_path(path);
1777 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1778 struct btrfs_device *device,
1779 u64 chunk_offset, u64 start, u64 num_bytes)
1782 struct btrfs_path *path;
1783 struct btrfs_fs_info *fs_info = device->fs_info;
1784 struct btrfs_root *root = fs_info->dev_root;
1785 struct btrfs_dev_extent *extent;
1786 struct extent_buffer *leaf;
1787 struct btrfs_key key;
1789 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1790 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1791 path = btrfs_alloc_path();
1795 key.objectid = device->devid;
1797 key.type = BTRFS_DEV_EXTENT_KEY;
1798 ret = btrfs_insert_empty_item(trans, root, path, &key,
1803 leaf = path->nodes[0];
1804 extent = btrfs_item_ptr(leaf, path->slots[0],
1805 struct btrfs_dev_extent);
1806 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1807 BTRFS_CHUNK_TREE_OBJECTID);
1808 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1809 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1810 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1812 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1813 btrfs_mark_buffer_dirty(leaf);
1815 btrfs_free_path(path);
1819 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1821 struct extent_map_tree *em_tree;
1822 struct extent_map *em;
1826 em_tree = &fs_info->mapping_tree;
1827 read_lock(&em_tree->lock);
1828 n = rb_last(&em_tree->map.rb_root);
1830 em = rb_entry(n, struct extent_map, rb_node);
1831 ret = em->start + em->len;
1833 read_unlock(&em_tree->lock);
1838 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1842 struct btrfs_key key;
1843 struct btrfs_key found_key;
1844 struct btrfs_path *path;
1846 path = btrfs_alloc_path();
1850 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1851 key.type = BTRFS_DEV_ITEM_KEY;
1852 key.offset = (u64)-1;
1854 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1858 BUG_ON(ret == 0); /* Corruption */
1860 ret = btrfs_previous_item(fs_info->chunk_root, path,
1861 BTRFS_DEV_ITEMS_OBJECTID,
1862 BTRFS_DEV_ITEM_KEY);
1866 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1868 *devid_ret = found_key.offset + 1;
1872 btrfs_free_path(path);
1877 * the device information is stored in the chunk root
1878 * the btrfs_device struct should be fully filled in
1880 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1881 struct btrfs_device *device)
1884 struct btrfs_path *path;
1885 struct btrfs_dev_item *dev_item;
1886 struct extent_buffer *leaf;
1887 struct btrfs_key key;
1890 path = btrfs_alloc_path();
1894 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1895 key.type = BTRFS_DEV_ITEM_KEY;
1896 key.offset = device->devid;
1898 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1899 &key, sizeof(*dev_item));
1903 leaf = path->nodes[0];
1904 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1906 btrfs_set_device_id(leaf, dev_item, device->devid);
1907 btrfs_set_device_generation(leaf, dev_item, 0);
1908 btrfs_set_device_type(leaf, dev_item, device->type);
1909 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1910 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1911 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1912 btrfs_set_device_total_bytes(leaf, dev_item,
1913 btrfs_device_get_disk_total_bytes(device));
1914 btrfs_set_device_bytes_used(leaf, dev_item,
1915 btrfs_device_get_bytes_used(device));
1916 btrfs_set_device_group(leaf, dev_item, 0);
1917 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1918 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1919 btrfs_set_device_start_offset(leaf, dev_item, 0);
1921 ptr = btrfs_device_uuid(dev_item);
1922 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1923 ptr = btrfs_device_fsid(dev_item);
1924 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1925 ptr, BTRFS_FSID_SIZE);
1926 btrfs_mark_buffer_dirty(leaf);
1930 btrfs_free_path(path);
1935 * Function to update ctime/mtime for a given device path.
1936 * Mainly used for ctime/mtime based probe like libblkid.
1938 static void update_dev_time(const char *path_name)
1942 filp = filp_open(path_name, O_RDWR, 0);
1945 file_update_time(filp);
1946 filp_close(filp, NULL);
1949 static int btrfs_rm_dev_item(struct btrfs_device *device)
1951 struct btrfs_root *root = device->fs_info->chunk_root;
1953 struct btrfs_path *path;
1954 struct btrfs_key key;
1955 struct btrfs_trans_handle *trans;
1957 path = btrfs_alloc_path();
1961 trans = btrfs_start_transaction(root, 0);
1962 if (IS_ERR(trans)) {
1963 btrfs_free_path(path);
1964 return PTR_ERR(trans);
1966 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1967 key.type = BTRFS_DEV_ITEM_KEY;
1968 key.offset = device->devid;
1970 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1974 btrfs_abort_transaction(trans, ret);
1975 btrfs_end_transaction(trans);
1979 ret = btrfs_del_item(trans, root, path);
1981 btrfs_abort_transaction(trans, ret);
1982 btrfs_end_transaction(trans);
1986 btrfs_free_path(path);
1988 ret = btrfs_commit_transaction(trans);
1993 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1994 * filesystem. It's up to the caller to adjust that number regarding eg. device
1997 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2005 seq = read_seqbegin(&fs_info->profiles_lock);
2007 all_avail = fs_info->avail_data_alloc_bits |
2008 fs_info->avail_system_alloc_bits |
2009 fs_info->avail_metadata_alloc_bits;
2010 } while (read_seqretry(&fs_info->profiles_lock, seq));
2012 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2013 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2016 if (num_devices < btrfs_raid_array[i].devs_min) {
2017 int ret = btrfs_raid_array[i].mindev_error;
2027 static struct btrfs_device * btrfs_find_next_active_device(
2028 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2030 struct btrfs_device *next_device;
2032 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2033 if (next_device != device &&
2034 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2035 && next_device->bdev)
2043 * Helper function to check if the given device is part of s_bdev / latest_bdev
2044 * and replace it with the provided or the next active device, in the context
2045 * where this function called, there should be always be another device (or
2046 * this_dev) which is active.
2048 void btrfs_assign_next_active_device(struct btrfs_device *device,
2049 struct btrfs_device *this_dev)
2051 struct btrfs_fs_info *fs_info = device->fs_info;
2052 struct btrfs_device *next_device;
2055 next_device = this_dev;
2057 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2059 ASSERT(next_device);
2061 if (fs_info->sb->s_bdev &&
2062 (fs_info->sb->s_bdev == device->bdev))
2063 fs_info->sb->s_bdev = next_device->bdev;
2065 if (fs_info->fs_devices->latest_bdev == device->bdev)
2066 fs_info->fs_devices->latest_bdev = next_device->bdev;
2070 * Return btrfs_fs_devices::num_devices excluding the device that's being
2071 * currently replaced.
2073 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2075 u64 num_devices = fs_info->fs_devices->num_devices;
2077 down_read(&fs_info->dev_replace.rwsem);
2078 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2079 ASSERT(num_devices > 1);
2082 up_read(&fs_info->dev_replace.rwsem);
2087 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2090 struct btrfs_device *device;
2091 struct btrfs_fs_devices *cur_devices;
2092 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2096 mutex_lock(&uuid_mutex);
2098 num_devices = btrfs_num_devices(fs_info);
2100 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2104 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2106 if (IS_ERR(device)) {
2107 if (PTR_ERR(device) == -ENOENT &&
2108 strcmp(device_path, "missing") == 0)
2109 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2111 ret = PTR_ERR(device);
2115 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2116 btrfs_warn_in_rcu(fs_info,
2117 "cannot remove device %s (devid %llu) due to active swapfile",
2118 rcu_str_deref(device->name), device->devid);
2123 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2124 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2128 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2129 fs_info->fs_devices->rw_devices == 1) {
2130 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2134 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2135 mutex_lock(&fs_info->chunk_mutex);
2136 list_del_init(&device->dev_alloc_list);
2137 device->fs_devices->rw_devices--;
2138 mutex_unlock(&fs_info->chunk_mutex);
2141 mutex_unlock(&uuid_mutex);
2142 ret = btrfs_shrink_device(device, 0);
2143 mutex_lock(&uuid_mutex);
2148 * TODO: the superblock still includes this device in its num_devices
2149 * counter although write_all_supers() is not locked out. This
2150 * could give a filesystem state which requires a degraded mount.
2152 ret = btrfs_rm_dev_item(device);
2156 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2157 btrfs_scrub_cancel_dev(device);
2160 * the device list mutex makes sure that we don't change
2161 * the device list while someone else is writing out all
2162 * the device supers. Whoever is writing all supers, should
2163 * lock the device list mutex before getting the number of
2164 * devices in the super block (super_copy). Conversely,
2165 * whoever updates the number of devices in the super block
2166 * (super_copy) should hold the device list mutex.
2170 * In normal cases the cur_devices == fs_devices. But in case
2171 * of deleting a seed device, the cur_devices should point to
2172 * its own fs_devices listed under the fs_devices->seed.
2174 cur_devices = device->fs_devices;
2175 mutex_lock(&fs_devices->device_list_mutex);
2176 list_del_rcu(&device->dev_list);
2178 cur_devices->num_devices--;
2179 cur_devices->total_devices--;
2180 /* Update total_devices of the parent fs_devices if it's seed */
2181 if (cur_devices != fs_devices)
2182 fs_devices->total_devices--;
2184 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2185 cur_devices->missing_devices--;
2187 btrfs_assign_next_active_device(device, NULL);
2190 cur_devices->open_devices--;
2191 /* remove sysfs entry */
2192 btrfs_sysfs_rm_device_link(fs_devices, device);
2195 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2196 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2197 mutex_unlock(&fs_devices->device_list_mutex);
2200 * at this point, the device is zero sized and detached from
2201 * the devices list. All that's left is to zero out the old
2202 * supers and free the device.
2204 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2205 btrfs_scratch_superblocks(device->bdev, device->name->str);
2207 btrfs_close_bdev(device);
2209 btrfs_free_device(device);
2211 if (cur_devices->open_devices == 0) {
2212 while (fs_devices) {
2213 if (fs_devices->seed == cur_devices) {
2214 fs_devices->seed = cur_devices->seed;
2217 fs_devices = fs_devices->seed;
2219 cur_devices->seed = NULL;
2220 close_fs_devices(cur_devices);
2221 free_fs_devices(cur_devices);
2225 mutex_unlock(&uuid_mutex);
2229 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2230 mutex_lock(&fs_info->chunk_mutex);
2231 list_add(&device->dev_alloc_list,
2232 &fs_devices->alloc_list);
2233 device->fs_devices->rw_devices++;
2234 mutex_unlock(&fs_info->chunk_mutex);
2239 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2241 struct btrfs_fs_devices *fs_devices;
2243 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2246 * in case of fs with no seed, srcdev->fs_devices will point
2247 * to fs_devices of fs_info. However when the dev being replaced is
2248 * a seed dev it will point to the seed's local fs_devices. In short
2249 * srcdev will have its correct fs_devices in both the cases.
2251 fs_devices = srcdev->fs_devices;
2253 list_del_rcu(&srcdev->dev_list);
2254 list_del(&srcdev->dev_alloc_list);
2255 fs_devices->num_devices--;
2256 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2257 fs_devices->missing_devices--;
2259 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2260 fs_devices->rw_devices--;
2263 fs_devices->open_devices--;
2266 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2268 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2269 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2271 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2272 /* zero out the old super if it is writable */
2273 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2276 btrfs_close_bdev(srcdev);
2278 btrfs_free_device(srcdev);
2280 /* if this is no devs we rather delete the fs_devices */
2281 if (!fs_devices->num_devices) {
2282 struct btrfs_fs_devices *tmp_fs_devices;
2285 * On a mounted FS, num_devices can't be zero unless it's a
2286 * seed. In case of a seed device being replaced, the replace
2287 * target added to the sprout FS, so there will be no more
2288 * device left under the seed FS.
2290 ASSERT(fs_devices->seeding);
2292 tmp_fs_devices = fs_info->fs_devices;
2293 while (tmp_fs_devices) {
2294 if (tmp_fs_devices->seed == fs_devices) {
2295 tmp_fs_devices->seed = fs_devices->seed;
2298 tmp_fs_devices = tmp_fs_devices->seed;
2300 fs_devices->seed = NULL;
2301 close_fs_devices(fs_devices);
2302 free_fs_devices(fs_devices);
2306 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2308 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2311 mutex_lock(&fs_devices->device_list_mutex);
2313 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2316 fs_devices->open_devices--;
2318 fs_devices->num_devices--;
2320 btrfs_assign_next_active_device(tgtdev, NULL);
2322 list_del_rcu(&tgtdev->dev_list);
2324 mutex_unlock(&fs_devices->device_list_mutex);
2327 * The update_dev_time() with in btrfs_scratch_superblocks()
2328 * may lead to a call to btrfs_show_devname() which will try
2329 * to hold device_list_mutex. And here this device
2330 * is already out of device list, so we don't have to hold
2331 * the device_list_mutex lock.
2333 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2335 btrfs_close_bdev(tgtdev);
2337 btrfs_free_device(tgtdev);
2340 static struct btrfs_device *btrfs_find_device_by_path(
2341 struct btrfs_fs_info *fs_info, const char *device_path)
2344 struct btrfs_super_block *disk_super;
2347 struct block_device *bdev;
2348 struct buffer_head *bh;
2349 struct btrfs_device *device;
2351 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2352 fs_info->bdev_holder, 0, &bdev, &bh);
2354 return ERR_PTR(ret);
2355 disk_super = (struct btrfs_super_block *)bh->b_data;
2356 devid = btrfs_stack_device_id(&disk_super->dev_item);
2357 dev_uuid = disk_super->dev_item.uuid;
2358 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2359 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2360 disk_super->metadata_uuid, true);
2362 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2363 disk_super->fsid, true);
2367 device = ERR_PTR(-ENOENT);
2368 blkdev_put(bdev, FMODE_READ);
2373 * Lookup a device given by device id, or the path if the id is 0.
2375 struct btrfs_device *btrfs_find_device_by_devspec(
2376 struct btrfs_fs_info *fs_info, u64 devid,
2377 const char *device_path)
2379 struct btrfs_device *device;
2382 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2385 return ERR_PTR(-ENOENT);
2389 if (!device_path || !device_path[0])
2390 return ERR_PTR(-EINVAL);
2392 if (strcmp(device_path, "missing") == 0) {
2393 /* Find first missing device */
2394 list_for_each_entry(device, &fs_info->fs_devices->devices,
2396 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2397 &device->dev_state) && !device->bdev)
2400 return ERR_PTR(-ENOENT);
2403 return btrfs_find_device_by_path(fs_info, device_path);
2407 * does all the dirty work required for changing file system's UUID.
2409 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2411 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2412 struct btrfs_fs_devices *old_devices;
2413 struct btrfs_fs_devices *seed_devices;
2414 struct btrfs_super_block *disk_super = fs_info->super_copy;
2415 struct btrfs_device *device;
2418 lockdep_assert_held(&uuid_mutex);
2419 if (!fs_devices->seeding)
2422 seed_devices = alloc_fs_devices(NULL, NULL);
2423 if (IS_ERR(seed_devices))
2424 return PTR_ERR(seed_devices);
2426 old_devices = clone_fs_devices(fs_devices);
2427 if (IS_ERR(old_devices)) {
2428 kfree(seed_devices);
2429 return PTR_ERR(old_devices);
2432 list_add(&old_devices->fs_list, &fs_uuids);
2434 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2435 seed_devices->opened = 1;
2436 INIT_LIST_HEAD(&seed_devices->devices);
2437 INIT_LIST_HEAD(&seed_devices->alloc_list);
2438 mutex_init(&seed_devices->device_list_mutex);
2440 mutex_lock(&fs_devices->device_list_mutex);
2441 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2443 list_for_each_entry(device, &seed_devices->devices, dev_list)
2444 device->fs_devices = seed_devices;
2446 mutex_lock(&fs_info->chunk_mutex);
2447 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2448 mutex_unlock(&fs_info->chunk_mutex);
2450 fs_devices->seeding = 0;
2451 fs_devices->num_devices = 0;
2452 fs_devices->open_devices = 0;
2453 fs_devices->missing_devices = 0;
2454 fs_devices->rotating = 0;
2455 fs_devices->seed = seed_devices;
2457 generate_random_uuid(fs_devices->fsid);
2458 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2459 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2460 mutex_unlock(&fs_devices->device_list_mutex);
2462 super_flags = btrfs_super_flags(disk_super) &
2463 ~BTRFS_SUPER_FLAG_SEEDING;
2464 btrfs_set_super_flags(disk_super, super_flags);
2470 * Store the expected generation for seed devices in device items.
2472 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2474 struct btrfs_fs_info *fs_info = trans->fs_info;
2475 struct btrfs_root *root = fs_info->chunk_root;
2476 struct btrfs_path *path;
2477 struct extent_buffer *leaf;
2478 struct btrfs_dev_item *dev_item;
2479 struct btrfs_device *device;
2480 struct btrfs_key key;
2481 u8 fs_uuid[BTRFS_FSID_SIZE];
2482 u8 dev_uuid[BTRFS_UUID_SIZE];
2486 path = btrfs_alloc_path();
2490 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2492 key.type = BTRFS_DEV_ITEM_KEY;
2495 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2499 leaf = path->nodes[0];
2501 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2502 ret = btrfs_next_leaf(root, path);
2507 leaf = path->nodes[0];
2508 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2509 btrfs_release_path(path);
2513 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2514 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2515 key.type != BTRFS_DEV_ITEM_KEY)
2518 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2519 struct btrfs_dev_item);
2520 devid = btrfs_device_id(leaf, dev_item);
2521 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2523 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2525 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2527 BUG_ON(!device); /* Logic error */
2529 if (device->fs_devices->seeding) {
2530 btrfs_set_device_generation(leaf, dev_item,
2531 device->generation);
2532 btrfs_mark_buffer_dirty(leaf);
2540 btrfs_free_path(path);
2544 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2546 struct btrfs_root *root = fs_info->dev_root;
2547 struct request_queue *q;
2548 struct btrfs_trans_handle *trans;
2549 struct btrfs_device *device;
2550 struct block_device *bdev;
2551 struct super_block *sb = fs_info->sb;
2552 struct rcu_string *name;
2553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2554 u64 orig_super_total_bytes;
2555 u64 orig_super_num_devices;
2556 int seeding_dev = 0;
2558 bool unlocked = false;
2560 if (sb_rdonly(sb) && !fs_devices->seeding)
2563 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2564 fs_info->bdev_holder);
2566 return PTR_ERR(bdev);
2568 if (fs_devices->seeding) {
2570 down_write(&sb->s_umount);
2571 mutex_lock(&uuid_mutex);
2574 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2576 mutex_lock(&fs_devices->device_list_mutex);
2577 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2578 if (device->bdev == bdev) {
2581 &fs_devices->device_list_mutex);
2585 mutex_unlock(&fs_devices->device_list_mutex);
2587 device = btrfs_alloc_device(fs_info, NULL, NULL);
2588 if (IS_ERR(device)) {
2589 /* we can safely leave the fs_devices entry around */
2590 ret = PTR_ERR(device);
2594 name = rcu_string_strdup(device_path, GFP_KERNEL);
2597 goto error_free_device;
2599 rcu_assign_pointer(device->name, name);
2601 trans = btrfs_start_transaction(root, 0);
2602 if (IS_ERR(trans)) {
2603 ret = PTR_ERR(trans);
2604 goto error_free_device;
2607 q = bdev_get_queue(bdev);
2608 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2609 device->generation = trans->transid;
2610 device->io_width = fs_info->sectorsize;
2611 device->io_align = fs_info->sectorsize;
2612 device->sector_size = fs_info->sectorsize;
2613 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2614 fs_info->sectorsize);
2615 device->disk_total_bytes = device->total_bytes;
2616 device->commit_total_bytes = device->total_bytes;
2617 device->fs_info = fs_info;
2618 device->bdev = bdev;
2619 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2620 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2621 device->mode = FMODE_EXCL;
2622 device->dev_stats_valid = 1;
2623 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2626 sb->s_flags &= ~SB_RDONLY;
2627 ret = btrfs_prepare_sprout(fs_info);
2629 btrfs_abort_transaction(trans, ret);
2634 device->fs_devices = fs_devices;
2636 mutex_lock(&fs_devices->device_list_mutex);
2637 mutex_lock(&fs_info->chunk_mutex);
2638 list_add_rcu(&device->dev_list, &fs_devices->devices);
2639 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2640 fs_devices->num_devices++;
2641 fs_devices->open_devices++;
2642 fs_devices->rw_devices++;
2643 fs_devices->total_devices++;
2644 fs_devices->total_rw_bytes += device->total_bytes;
2646 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2648 if (!blk_queue_nonrot(q))
2649 fs_devices->rotating = 1;
2651 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2652 btrfs_set_super_total_bytes(fs_info->super_copy,
2653 round_down(orig_super_total_bytes + device->total_bytes,
2654 fs_info->sectorsize));
2656 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2657 btrfs_set_super_num_devices(fs_info->super_copy,
2658 orig_super_num_devices + 1);
2660 /* add sysfs device entry */
2661 btrfs_sysfs_add_device_link(fs_devices, device);
2664 * we've got more storage, clear any full flags on the space
2667 btrfs_clear_space_info_full(fs_info);
2669 mutex_unlock(&fs_info->chunk_mutex);
2670 mutex_unlock(&fs_devices->device_list_mutex);
2673 mutex_lock(&fs_info->chunk_mutex);
2674 ret = init_first_rw_device(trans);
2675 mutex_unlock(&fs_info->chunk_mutex);
2677 btrfs_abort_transaction(trans, ret);
2682 ret = btrfs_add_dev_item(trans, device);
2684 btrfs_abort_transaction(trans, ret);
2689 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2691 ret = btrfs_finish_sprout(trans);
2693 btrfs_abort_transaction(trans, ret);
2697 /* Sprouting would change fsid of the mounted root,
2698 * so rename the fsid on the sysfs
2700 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2701 fs_info->fs_devices->fsid);
2702 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2704 "sysfs: failed to create fsid for sprout");
2707 ret = btrfs_commit_transaction(trans);
2710 mutex_unlock(&uuid_mutex);
2711 up_write(&sb->s_umount);
2714 if (ret) /* transaction commit */
2717 ret = btrfs_relocate_sys_chunks(fs_info);
2719 btrfs_handle_fs_error(fs_info, ret,
2720 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2721 trans = btrfs_attach_transaction(root);
2722 if (IS_ERR(trans)) {
2723 if (PTR_ERR(trans) == -ENOENT)
2725 ret = PTR_ERR(trans);
2729 ret = btrfs_commit_transaction(trans);
2732 /* Update ctime/mtime for libblkid */
2733 update_dev_time(device_path);
2737 btrfs_sysfs_rm_device_link(fs_devices, device);
2738 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2739 mutex_lock(&fs_info->chunk_mutex);
2740 list_del_rcu(&device->dev_list);
2741 list_del(&device->dev_alloc_list);
2742 fs_info->fs_devices->num_devices--;
2743 fs_info->fs_devices->open_devices--;
2744 fs_info->fs_devices->rw_devices--;
2745 fs_info->fs_devices->total_devices--;
2746 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2747 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2748 btrfs_set_super_total_bytes(fs_info->super_copy,
2749 orig_super_total_bytes);
2750 btrfs_set_super_num_devices(fs_info->super_copy,
2751 orig_super_num_devices);
2752 mutex_unlock(&fs_info->chunk_mutex);
2753 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2756 sb->s_flags |= SB_RDONLY;
2758 btrfs_end_transaction(trans);
2760 btrfs_free_device(device);
2762 blkdev_put(bdev, FMODE_EXCL);
2763 if (seeding_dev && !unlocked) {
2764 mutex_unlock(&uuid_mutex);
2765 up_write(&sb->s_umount);
2770 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2771 struct btrfs_device *device)
2774 struct btrfs_path *path;
2775 struct btrfs_root *root = device->fs_info->chunk_root;
2776 struct btrfs_dev_item *dev_item;
2777 struct extent_buffer *leaf;
2778 struct btrfs_key key;
2780 path = btrfs_alloc_path();
2784 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2785 key.type = BTRFS_DEV_ITEM_KEY;
2786 key.offset = device->devid;
2788 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2797 leaf = path->nodes[0];
2798 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2800 btrfs_set_device_id(leaf, dev_item, device->devid);
2801 btrfs_set_device_type(leaf, dev_item, device->type);
2802 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2803 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2804 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2805 btrfs_set_device_total_bytes(leaf, dev_item,
2806 btrfs_device_get_disk_total_bytes(device));
2807 btrfs_set_device_bytes_used(leaf, dev_item,
2808 btrfs_device_get_bytes_used(device));
2809 btrfs_mark_buffer_dirty(leaf);
2812 btrfs_free_path(path);
2816 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2817 struct btrfs_device *device, u64 new_size)
2819 struct btrfs_fs_info *fs_info = device->fs_info;
2820 struct btrfs_super_block *super_copy = fs_info->super_copy;
2824 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2827 new_size = round_down(new_size, fs_info->sectorsize);
2829 mutex_lock(&fs_info->chunk_mutex);
2830 old_total = btrfs_super_total_bytes(super_copy);
2831 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2833 if (new_size <= device->total_bytes ||
2834 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2835 mutex_unlock(&fs_info->chunk_mutex);
2839 btrfs_set_super_total_bytes(super_copy,
2840 round_down(old_total + diff, fs_info->sectorsize));
2841 device->fs_devices->total_rw_bytes += diff;
2843 btrfs_device_set_total_bytes(device, new_size);
2844 btrfs_device_set_disk_total_bytes(device, new_size);
2845 btrfs_clear_space_info_full(device->fs_info);
2846 if (list_empty(&device->post_commit_list))
2847 list_add_tail(&device->post_commit_list,
2848 &trans->transaction->dev_update_list);
2849 mutex_unlock(&fs_info->chunk_mutex);
2851 return btrfs_update_device(trans, device);
2854 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2856 struct btrfs_fs_info *fs_info = trans->fs_info;
2857 struct btrfs_root *root = fs_info->chunk_root;
2859 struct btrfs_path *path;
2860 struct btrfs_key key;
2862 path = btrfs_alloc_path();
2866 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2867 key.offset = chunk_offset;
2868 key.type = BTRFS_CHUNK_ITEM_KEY;
2870 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2873 else if (ret > 0) { /* Logic error or corruption */
2874 btrfs_handle_fs_error(fs_info, -ENOENT,
2875 "Failed lookup while freeing chunk.");
2880 ret = btrfs_del_item(trans, root, path);
2882 btrfs_handle_fs_error(fs_info, ret,
2883 "Failed to delete chunk item.");
2885 btrfs_free_path(path);
2889 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2891 struct btrfs_super_block *super_copy = fs_info->super_copy;
2892 struct btrfs_disk_key *disk_key;
2893 struct btrfs_chunk *chunk;
2900 struct btrfs_key key;
2902 mutex_lock(&fs_info->chunk_mutex);
2903 array_size = btrfs_super_sys_array_size(super_copy);
2905 ptr = super_copy->sys_chunk_array;
2908 while (cur < array_size) {
2909 disk_key = (struct btrfs_disk_key *)ptr;
2910 btrfs_disk_key_to_cpu(&key, disk_key);
2912 len = sizeof(*disk_key);
2914 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2915 chunk = (struct btrfs_chunk *)(ptr + len);
2916 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2917 len += btrfs_chunk_item_size(num_stripes);
2922 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2923 key.offset == chunk_offset) {
2924 memmove(ptr, ptr + len, array_size - (cur + len));
2926 btrfs_set_super_sys_array_size(super_copy, array_size);
2932 mutex_unlock(&fs_info->chunk_mutex);
2937 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2938 * @logical: Logical block offset in bytes.
2939 * @length: Length of extent in bytes.
2941 * Return: Chunk mapping or ERR_PTR.
2943 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2944 u64 logical, u64 length)
2946 struct extent_map_tree *em_tree;
2947 struct extent_map *em;
2949 em_tree = &fs_info->mapping_tree;
2950 read_lock(&em_tree->lock);
2951 em = lookup_extent_mapping(em_tree, logical, length);
2952 read_unlock(&em_tree->lock);
2955 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2957 return ERR_PTR(-EINVAL);
2960 if (em->start > logical || em->start + em->len < logical) {
2962 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2963 logical, length, em->start, em->start + em->len);
2964 free_extent_map(em);
2965 return ERR_PTR(-EINVAL);
2968 /* callers are responsible for dropping em's ref. */
2972 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2974 struct btrfs_fs_info *fs_info = trans->fs_info;
2975 struct extent_map *em;
2976 struct map_lookup *map;
2977 u64 dev_extent_len = 0;
2979 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2981 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2984 * This is a logic error, but we don't want to just rely on the
2985 * user having built with ASSERT enabled, so if ASSERT doesn't
2986 * do anything we still error out.
2991 map = em->map_lookup;
2992 mutex_lock(&fs_info->chunk_mutex);
2993 check_system_chunk(trans, map->type);
2994 mutex_unlock(&fs_info->chunk_mutex);
2997 * Take the device list mutex to prevent races with the final phase of
2998 * a device replace operation that replaces the device object associated
2999 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
3001 mutex_lock(&fs_devices->device_list_mutex);
3002 for (i = 0; i < map->num_stripes; i++) {
3003 struct btrfs_device *device = map->stripes[i].dev;
3004 ret = btrfs_free_dev_extent(trans, device,
3005 map->stripes[i].physical,
3008 mutex_unlock(&fs_devices->device_list_mutex);
3009 btrfs_abort_transaction(trans, ret);
3013 if (device->bytes_used > 0) {
3014 mutex_lock(&fs_info->chunk_mutex);
3015 btrfs_device_set_bytes_used(device,
3016 device->bytes_used - dev_extent_len);
3017 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3018 btrfs_clear_space_info_full(fs_info);
3019 mutex_unlock(&fs_info->chunk_mutex);
3022 ret = btrfs_update_device(trans, device);
3024 mutex_unlock(&fs_devices->device_list_mutex);
3025 btrfs_abort_transaction(trans, ret);
3029 mutex_unlock(&fs_devices->device_list_mutex);
3031 ret = btrfs_free_chunk(trans, chunk_offset);
3033 btrfs_abort_transaction(trans, ret);
3037 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3039 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3040 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3042 btrfs_abort_transaction(trans, ret);
3047 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3049 btrfs_abort_transaction(trans, ret);
3055 free_extent_map(em);
3059 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3061 struct btrfs_root *root = fs_info->chunk_root;
3062 struct btrfs_trans_handle *trans;
3066 * Prevent races with automatic removal of unused block groups.
3067 * After we relocate and before we remove the chunk with offset
3068 * chunk_offset, automatic removal of the block group can kick in,
3069 * resulting in a failure when calling btrfs_remove_chunk() below.
3071 * Make sure to acquire this mutex before doing a tree search (dev
3072 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3073 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3074 * we release the path used to search the chunk/dev tree and before
3075 * the current task acquires this mutex and calls us.
3077 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3079 ret = btrfs_can_relocate(fs_info, chunk_offset);
3083 /* step one, relocate all the extents inside this chunk */
3084 btrfs_scrub_pause(fs_info);
3085 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3086 btrfs_scrub_continue(fs_info);
3090 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3092 if (IS_ERR(trans)) {
3093 ret = PTR_ERR(trans);
3094 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3099 * step two, delete the device extents and the
3100 * chunk tree entries
3102 ret = btrfs_remove_chunk(trans, chunk_offset);
3103 btrfs_end_transaction(trans);
3107 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3109 struct btrfs_root *chunk_root = fs_info->chunk_root;
3110 struct btrfs_path *path;
3111 struct extent_buffer *leaf;
3112 struct btrfs_chunk *chunk;
3113 struct btrfs_key key;
3114 struct btrfs_key found_key;
3116 bool retried = false;
3120 path = btrfs_alloc_path();
3125 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3126 key.offset = (u64)-1;
3127 key.type = BTRFS_CHUNK_ITEM_KEY;
3130 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3131 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3133 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3136 BUG_ON(ret == 0); /* Corruption */
3138 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3141 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3147 leaf = path->nodes[0];
3148 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3150 chunk = btrfs_item_ptr(leaf, path->slots[0],
3151 struct btrfs_chunk);
3152 chunk_type = btrfs_chunk_type(leaf, chunk);
3153 btrfs_release_path(path);
3155 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3156 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3162 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3164 if (found_key.offset == 0)
3166 key.offset = found_key.offset - 1;
3169 if (failed && !retried) {
3173 } else if (WARN_ON(failed && retried)) {
3177 btrfs_free_path(path);
3182 * return 1 : allocate a data chunk successfully,
3183 * return <0: errors during allocating a data chunk,
3184 * return 0 : no need to allocate a data chunk.
3186 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3189 struct btrfs_block_group_cache *cache;
3193 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3195 chunk_type = cache->flags;
3196 btrfs_put_block_group(cache);
3198 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3199 spin_lock(&fs_info->data_sinfo->lock);
3200 bytes_used = fs_info->data_sinfo->bytes_used;
3201 spin_unlock(&fs_info->data_sinfo->lock);
3204 struct btrfs_trans_handle *trans;
3207 trans = btrfs_join_transaction(fs_info->tree_root);
3209 return PTR_ERR(trans);
3211 ret = btrfs_force_chunk_alloc(trans,
3212 BTRFS_BLOCK_GROUP_DATA);
3213 btrfs_end_transaction(trans);
3222 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3223 struct btrfs_balance_control *bctl)
3225 struct btrfs_root *root = fs_info->tree_root;
3226 struct btrfs_trans_handle *trans;
3227 struct btrfs_balance_item *item;
3228 struct btrfs_disk_balance_args disk_bargs;
3229 struct btrfs_path *path;
3230 struct extent_buffer *leaf;
3231 struct btrfs_key key;
3234 path = btrfs_alloc_path();
3238 trans = btrfs_start_transaction(root, 0);
3239 if (IS_ERR(trans)) {
3240 btrfs_free_path(path);
3241 return PTR_ERR(trans);
3244 key.objectid = BTRFS_BALANCE_OBJECTID;
3245 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3248 ret = btrfs_insert_empty_item(trans, root, path, &key,
3253 leaf = path->nodes[0];
3254 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3256 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3258 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3259 btrfs_set_balance_data(leaf, item, &disk_bargs);
3260 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3261 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3262 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3263 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3265 btrfs_set_balance_flags(leaf, item, bctl->flags);
3267 btrfs_mark_buffer_dirty(leaf);
3269 btrfs_free_path(path);
3270 err = btrfs_commit_transaction(trans);
3276 static int del_balance_item(struct btrfs_fs_info *fs_info)
3278 struct btrfs_root *root = fs_info->tree_root;
3279 struct btrfs_trans_handle *trans;
3280 struct btrfs_path *path;
3281 struct btrfs_key key;
3284 path = btrfs_alloc_path();
3288 trans = btrfs_start_transaction(root, 0);
3289 if (IS_ERR(trans)) {
3290 btrfs_free_path(path);
3291 return PTR_ERR(trans);
3294 key.objectid = BTRFS_BALANCE_OBJECTID;
3295 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3298 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3306 ret = btrfs_del_item(trans, root, path);
3308 btrfs_free_path(path);
3309 err = btrfs_commit_transaction(trans);
3316 * This is a heuristic used to reduce the number of chunks balanced on
3317 * resume after balance was interrupted.
3319 static void update_balance_args(struct btrfs_balance_control *bctl)
3322 * Turn on soft mode for chunk types that were being converted.
3324 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3325 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3326 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3327 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3328 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3329 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3332 * Turn on usage filter if is not already used. The idea is
3333 * that chunks that we have already balanced should be
3334 * reasonably full. Don't do it for chunks that are being
3335 * converted - that will keep us from relocating unconverted
3336 * (albeit full) chunks.
3338 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3339 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3340 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3341 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3342 bctl->data.usage = 90;
3344 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3345 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3346 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3347 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3348 bctl->sys.usage = 90;
3350 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3351 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3352 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3353 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3354 bctl->meta.usage = 90;
3359 * Clear the balance status in fs_info and delete the balance item from disk.
3361 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3363 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3366 BUG_ON(!fs_info->balance_ctl);
3368 spin_lock(&fs_info->balance_lock);
3369 fs_info->balance_ctl = NULL;
3370 spin_unlock(&fs_info->balance_lock);
3373 ret = del_balance_item(fs_info);
3375 btrfs_handle_fs_error(fs_info, ret, NULL);
3379 * Balance filters. Return 1 if chunk should be filtered out
3380 * (should not be balanced).
3382 static int chunk_profiles_filter(u64 chunk_type,
3383 struct btrfs_balance_args *bargs)
3385 chunk_type = chunk_to_extended(chunk_type) &
3386 BTRFS_EXTENDED_PROFILE_MASK;
3388 if (bargs->profiles & chunk_type)
3394 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3395 struct btrfs_balance_args *bargs)
3397 struct btrfs_block_group_cache *cache;
3399 u64 user_thresh_min;
3400 u64 user_thresh_max;
3403 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3404 chunk_used = btrfs_block_group_used(&cache->item);
3406 if (bargs->usage_min == 0)
3407 user_thresh_min = 0;
3409 user_thresh_min = div_factor_fine(cache->key.offset,
3412 if (bargs->usage_max == 0)
3413 user_thresh_max = 1;
3414 else if (bargs->usage_max > 100)
3415 user_thresh_max = cache->key.offset;
3417 user_thresh_max = div_factor_fine(cache->key.offset,
3420 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3423 btrfs_put_block_group(cache);
3427 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3428 u64 chunk_offset, struct btrfs_balance_args *bargs)
3430 struct btrfs_block_group_cache *cache;
3431 u64 chunk_used, user_thresh;
3434 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3435 chunk_used = btrfs_block_group_used(&cache->item);
3437 if (bargs->usage_min == 0)
3439 else if (bargs->usage > 100)
3440 user_thresh = cache->key.offset;
3442 user_thresh = div_factor_fine(cache->key.offset,
3445 if (chunk_used < user_thresh)
3448 btrfs_put_block_group(cache);
3452 static int chunk_devid_filter(struct extent_buffer *leaf,
3453 struct btrfs_chunk *chunk,
3454 struct btrfs_balance_args *bargs)
3456 struct btrfs_stripe *stripe;
3457 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3460 for (i = 0; i < num_stripes; i++) {
3461 stripe = btrfs_stripe_nr(chunk, i);
3462 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3469 static u64 calc_data_stripes(u64 type, int num_stripes)
3471 const int index = btrfs_bg_flags_to_raid_index(type);
3472 const int ncopies = btrfs_raid_array[index].ncopies;
3473 const int nparity = btrfs_raid_array[index].nparity;
3476 return num_stripes - nparity;
3478 return num_stripes / ncopies;
3481 /* [pstart, pend) */
3482 static int chunk_drange_filter(struct extent_buffer *leaf,
3483 struct btrfs_chunk *chunk,
3484 struct btrfs_balance_args *bargs)
3486 struct btrfs_stripe *stripe;
3487 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3494 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3497 type = btrfs_chunk_type(leaf, chunk);
3498 factor = calc_data_stripes(type, num_stripes);
3500 for (i = 0; i < num_stripes; i++) {
3501 stripe = btrfs_stripe_nr(chunk, i);
3502 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3505 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3506 stripe_length = btrfs_chunk_length(leaf, chunk);
3507 stripe_length = div_u64(stripe_length, factor);
3509 if (stripe_offset < bargs->pend &&
3510 stripe_offset + stripe_length > bargs->pstart)
3517 /* [vstart, vend) */
3518 static int chunk_vrange_filter(struct extent_buffer *leaf,
3519 struct btrfs_chunk *chunk,
3521 struct btrfs_balance_args *bargs)
3523 if (chunk_offset < bargs->vend &&
3524 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3525 /* at least part of the chunk is inside this vrange */
3531 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3532 struct btrfs_chunk *chunk,
3533 struct btrfs_balance_args *bargs)
3535 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3537 if (bargs->stripes_min <= num_stripes
3538 && num_stripes <= bargs->stripes_max)
3544 static int chunk_soft_convert_filter(u64 chunk_type,
3545 struct btrfs_balance_args *bargs)
3547 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3550 chunk_type = chunk_to_extended(chunk_type) &
3551 BTRFS_EXTENDED_PROFILE_MASK;
3553 if (bargs->target == chunk_type)
3559 static int should_balance_chunk(struct extent_buffer *leaf,
3560 struct btrfs_chunk *chunk, u64 chunk_offset)
3562 struct btrfs_fs_info *fs_info = leaf->fs_info;
3563 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3564 struct btrfs_balance_args *bargs = NULL;
3565 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3568 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3569 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3573 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3574 bargs = &bctl->data;
3575 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3577 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3578 bargs = &bctl->meta;
3580 /* profiles filter */
3581 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3582 chunk_profiles_filter(chunk_type, bargs)) {
3587 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3588 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3590 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3591 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3596 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3597 chunk_devid_filter(leaf, chunk, bargs)) {
3601 /* drange filter, makes sense only with devid filter */
3602 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3603 chunk_drange_filter(leaf, chunk, bargs)) {
3608 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3609 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3613 /* stripes filter */
3614 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3615 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3619 /* soft profile changing mode */
3620 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3621 chunk_soft_convert_filter(chunk_type, bargs)) {
3626 * limited by count, must be the last filter
3628 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3629 if (bargs->limit == 0)
3633 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3635 * Same logic as the 'limit' filter; the minimum cannot be
3636 * determined here because we do not have the global information
3637 * about the count of all chunks that satisfy the filters.
3639 if (bargs->limit_max == 0)
3648 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3650 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3651 struct btrfs_root *chunk_root = fs_info->chunk_root;
3653 struct btrfs_chunk *chunk;
3654 struct btrfs_path *path = NULL;
3655 struct btrfs_key key;
3656 struct btrfs_key found_key;
3657 struct extent_buffer *leaf;
3660 int enospc_errors = 0;
3661 bool counting = true;
3662 /* The single value limit and min/max limits use the same bytes in the */
3663 u64 limit_data = bctl->data.limit;
3664 u64 limit_meta = bctl->meta.limit;
3665 u64 limit_sys = bctl->sys.limit;
3669 int chunk_reserved = 0;
3671 path = btrfs_alloc_path();
3677 /* zero out stat counters */
3678 spin_lock(&fs_info->balance_lock);
3679 memset(&bctl->stat, 0, sizeof(bctl->stat));
3680 spin_unlock(&fs_info->balance_lock);
3684 * The single value limit and min/max limits use the same bytes
3687 bctl->data.limit = limit_data;
3688 bctl->meta.limit = limit_meta;
3689 bctl->sys.limit = limit_sys;
3691 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3692 key.offset = (u64)-1;
3693 key.type = BTRFS_CHUNK_ITEM_KEY;
3696 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3697 atomic_read(&fs_info->balance_cancel_req)) {
3702 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3703 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3705 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3710 * this shouldn't happen, it means the last relocate
3714 BUG(); /* FIXME break ? */
3716 ret = btrfs_previous_item(chunk_root, path, 0,
3717 BTRFS_CHUNK_ITEM_KEY);
3719 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3724 leaf = path->nodes[0];
3725 slot = path->slots[0];
3726 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3728 if (found_key.objectid != key.objectid) {
3729 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3733 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3734 chunk_type = btrfs_chunk_type(leaf, chunk);
3737 spin_lock(&fs_info->balance_lock);
3738 bctl->stat.considered++;
3739 spin_unlock(&fs_info->balance_lock);
3742 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3744 btrfs_release_path(path);
3746 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3751 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3752 spin_lock(&fs_info->balance_lock);
3753 bctl->stat.expected++;
3754 spin_unlock(&fs_info->balance_lock);
3756 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3758 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3760 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3767 * Apply limit_min filter, no need to check if the LIMITS
3768 * filter is used, limit_min is 0 by default
3770 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3771 count_data < bctl->data.limit_min)
3772 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3773 count_meta < bctl->meta.limit_min)
3774 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3775 count_sys < bctl->sys.limit_min)) {
3776 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3780 if (!chunk_reserved) {
3782 * We may be relocating the only data chunk we have,
3783 * which could potentially end up with losing data's
3784 * raid profile, so lets allocate an empty one in
3787 ret = btrfs_may_alloc_data_chunk(fs_info,
3790 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3792 } else if (ret == 1) {
3797 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3798 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3799 if (ret == -ENOSPC) {
3801 } else if (ret == -ETXTBSY) {
3803 "skipping relocation of block group %llu due to active swapfile",
3809 spin_lock(&fs_info->balance_lock);
3810 bctl->stat.completed++;
3811 spin_unlock(&fs_info->balance_lock);
3814 if (found_key.offset == 0)
3816 key.offset = found_key.offset - 1;
3820 btrfs_release_path(path);
3825 btrfs_free_path(path);
3826 if (enospc_errors) {
3827 btrfs_info(fs_info, "%d enospc errors during balance",
3837 * alloc_profile_is_valid - see if a given profile is valid and reduced
3838 * @flags: profile to validate
3839 * @extended: if true @flags is treated as an extended profile
3841 static int alloc_profile_is_valid(u64 flags, int extended)
3843 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3844 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3846 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3848 /* 1) check that all other bits are zeroed */
3852 /* 2) see if profile is reduced */
3854 return !extended; /* "0" is valid for usual profiles */
3856 /* true if exactly one bit set */
3857 return is_power_of_2(flags);
3860 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3862 /* cancel requested || normal exit path */
3863 return atomic_read(&fs_info->balance_cancel_req) ||
3864 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3865 atomic_read(&fs_info->balance_cancel_req) == 0);
3868 /* Non-zero return value signifies invalidity */
3869 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3872 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3873 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3874 (bctl_arg->target & ~allowed)));
3878 * Fill @buf with textual description of balance filter flags @bargs, up to
3879 * @size_buf including the terminating null. The output may be trimmed if it
3880 * does not fit into the provided buffer.
3882 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3886 u32 size_bp = size_buf;
3888 u64 flags = bargs->flags;
3889 char tmp_buf[128] = {'\0'};
3894 #define CHECK_APPEND_NOARG(a) \
3896 ret = snprintf(bp, size_bp, (a)); \
3897 if (ret < 0 || ret >= size_bp) \
3898 goto out_overflow; \
3903 #define CHECK_APPEND_1ARG(a, v1) \
3905 ret = snprintf(bp, size_bp, (a), (v1)); \
3906 if (ret < 0 || ret >= size_bp) \
3907 goto out_overflow; \
3912 #define CHECK_APPEND_2ARG(a, v1, v2) \
3914 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3915 if (ret < 0 || ret >= size_bp) \
3916 goto out_overflow; \
3921 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3922 CHECK_APPEND_1ARG("convert=%s,",
3923 btrfs_bg_type_to_raid_name(bargs->target));
3925 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3926 CHECK_APPEND_NOARG("soft,");
3928 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3929 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3931 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3934 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3935 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3937 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3938 CHECK_APPEND_2ARG("usage=%u..%u,",
3939 bargs->usage_min, bargs->usage_max);
3941 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3942 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3944 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3945 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3946 bargs->pstart, bargs->pend);
3948 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3949 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3950 bargs->vstart, bargs->vend);
3952 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3953 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3955 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3956 CHECK_APPEND_2ARG("limit=%u..%u,",
3957 bargs->limit_min, bargs->limit_max);
3959 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3960 CHECK_APPEND_2ARG("stripes=%u..%u,",
3961 bargs->stripes_min, bargs->stripes_max);
3963 #undef CHECK_APPEND_2ARG
3964 #undef CHECK_APPEND_1ARG
3965 #undef CHECK_APPEND_NOARG
3969 if (size_bp < size_buf)
3970 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3975 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3977 u32 size_buf = 1024;
3978 char tmp_buf[192] = {'\0'};
3981 u32 size_bp = size_buf;
3983 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3985 buf = kzalloc(size_buf, GFP_KERNEL);
3991 #define CHECK_APPEND_1ARG(a, v1) \
3993 ret = snprintf(bp, size_bp, (a), (v1)); \
3994 if (ret < 0 || ret >= size_bp) \
3995 goto out_overflow; \
4000 if (bctl->flags & BTRFS_BALANCE_FORCE)
4001 CHECK_APPEND_1ARG("%s", "-f ");
4003 if (bctl->flags & BTRFS_BALANCE_DATA) {
4004 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4005 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4008 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4009 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4010 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4013 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4014 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4015 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4018 #undef CHECK_APPEND_1ARG
4022 if (size_bp < size_buf)
4023 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4024 btrfs_info(fs_info, "balance: %s %s",
4025 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4026 "resume" : "start", buf);
4032 * Should be called with balance mutexe held
4034 int btrfs_balance(struct btrfs_fs_info *fs_info,
4035 struct btrfs_balance_control *bctl,
4036 struct btrfs_ioctl_balance_args *bargs)
4038 u64 meta_target, data_target;
4044 bool reducing_integrity;
4047 if (btrfs_fs_closing(fs_info) ||
4048 atomic_read(&fs_info->balance_pause_req) ||
4049 atomic_read(&fs_info->balance_cancel_req)) {
4054 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4055 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4059 * In case of mixed groups both data and meta should be picked,
4060 * and identical options should be given for both of them.
4062 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4063 if (mixed && (bctl->flags & allowed)) {
4064 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4065 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4066 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4068 "balance: mixed groups data and metadata options must be the same");
4074 num_devices = btrfs_num_devices(fs_info);
4076 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4077 if (num_devices >= btrfs_raid_array[i].devs_min)
4078 allowed |= btrfs_raid_array[i].bg_flag;
4080 if (validate_convert_profile(&bctl->data, allowed)) {
4082 "balance: invalid convert data profile %s",
4083 btrfs_bg_type_to_raid_name(bctl->data.target));
4087 if (validate_convert_profile(&bctl->meta, allowed)) {
4089 "balance: invalid convert metadata profile %s",
4090 btrfs_bg_type_to_raid_name(bctl->meta.target));
4094 if (validate_convert_profile(&bctl->sys, allowed)) {
4096 "balance: invalid convert system profile %s",
4097 btrfs_bg_type_to_raid_name(bctl->sys.target));
4103 * Allow to reduce metadata or system integrity only if force set for
4104 * profiles with redundancy (copies, parity)
4107 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4108 if (btrfs_raid_array[i].ncopies >= 2 ||
4109 btrfs_raid_array[i].tolerated_failures >= 1)
4110 allowed |= btrfs_raid_array[i].bg_flag;
4113 seq = read_seqbegin(&fs_info->profiles_lock);
4115 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4116 (fs_info->avail_system_alloc_bits & allowed) &&
4117 !(bctl->sys.target & allowed)) ||
4118 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4119 (fs_info->avail_metadata_alloc_bits & allowed) &&
4120 !(bctl->meta.target & allowed)))
4121 reducing_integrity = true;
4123 reducing_integrity = false;
4125 /* if we're not converting, the target field is uninitialized */
4126 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4127 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4128 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4129 bctl->data.target : fs_info->avail_data_alloc_bits;
4130 } while (read_seqretry(&fs_info->profiles_lock, seq));
4132 if (reducing_integrity) {
4133 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4135 "balance: force reducing metadata integrity");
4138 "balance: reduces metadata integrity, use --force if you want this");
4144 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4145 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4147 "balance: metadata profile %s has lower redundancy than data profile %s",
4148 btrfs_bg_type_to_raid_name(meta_target),
4149 btrfs_bg_type_to_raid_name(data_target));
4152 if (fs_info->send_in_progress) {
4153 btrfs_warn_rl(fs_info,
4154 "cannot run balance while send operations are in progress (%d in progress)",
4155 fs_info->send_in_progress);
4160 ret = insert_balance_item(fs_info, bctl);
4161 if (ret && ret != -EEXIST)
4164 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4165 BUG_ON(ret == -EEXIST);
4166 BUG_ON(fs_info->balance_ctl);
4167 spin_lock(&fs_info->balance_lock);
4168 fs_info->balance_ctl = bctl;
4169 spin_unlock(&fs_info->balance_lock);
4171 BUG_ON(ret != -EEXIST);
4172 spin_lock(&fs_info->balance_lock);
4173 update_balance_args(bctl);
4174 spin_unlock(&fs_info->balance_lock);
4177 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4178 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4179 describe_balance_start_or_resume(fs_info);
4180 mutex_unlock(&fs_info->balance_mutex);
4182 ret = __btrfs_balance(fs_info);
4184 mutex_lock(&fs_info->balance_mutex);
4185 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4186 btrfs_info(fs_info, "balance: paused");
4187 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4188 btrfs_info(fs_info, "balance: canceled");
4190 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4192 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4195 memset(bargs, 0, sizeof(*bargs));
4196 btrfs_update_ioctl_balance_args(fs_info, bargs);
4199 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4200 balance_need_close(fs_info)) {
4201 reset_balance_state(fs_info);
4202 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4205 wake_up(&fs_info->balance_wait_q);
4209 if (bctl->flags & BTRFS_BALANCE_RESUME)
4210 reset_balance_state(fs_info);
4213 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4218 static int balance_kthread(void *data)
4220 struct btrfs_fs_info *fs_info = data;
4223 mutex_lock(&fs_info->balance_mutex);
4224 if (fs_info->balance_ctl)
4225 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4226 mutex_unlock(&fs_info->balance_mutex);
4231 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4233 struct task_struct *tsk;
4235 mutex_lock(&fs_info->balance_mutex);
4236 if (!fs_info->balance_ctl) {
4237 mutex_unlock(&fs_info->balance_mutex);
4240 mutex_unlock(&fs_info->balance_mutex);
4242 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4243 btrfs_info(fs_info, "balance: resume skipped");
4248 * A ro->rw remount sequence should continue with the paused balance
4249 * regardless of who pauses it, system or the user as of now, so set
4252 spin_lock(&fs_info->balance_lock);
4253 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4254 spin_unlock(&fs_info->balance_lock);
4256 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4257 return PTR_ERR_OR_ZERO(tsk);
4260 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4262 struct btrfs_balance_control *bctl;
4263 struct btrfs_balance_item *item;
4264 struct btrfs_disk_balance_args disk_bargs;
4265 struct btrfs_path *path;
4266 struct extent_buffer *leaf;
4267 struct btrfs_key key;
4270 path = btrfs_alloc_path();
4274 key.objectid = BTRFS_BALANCE_OBJECTID;
4275 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4278 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4281 if (ret > 0) { /* ret = -ENOENT; */
4286 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4292 leaf = path->nodes[0];
4293 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4295 bctl->flags = btrfs_balance_flags(leaf, item);
4296 bctl->flags |= BTRFS_BALANCE_RESUME;
4298 btrfs_balance_data(leaf, item, &disk_bargs);
4299 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4300 btrfs_balance_meta(leaf, item, &disk_bargs);
4301 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4302 btrfs_balance_sys(leaf, item, &disk_bargs);
4303 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4306 * This should never happen, as the paused balance state is recovered
4307 * during mount without any chance of other exclusive ops to collide.
4309 * This gives the exclusive op status to balance and keeps in paused
4310 * state until user intervention (cancel or umount). If the ownership
4311 * cannot be assigned, show a message but do not fail. The balance
4312 * is in a paused state and must have fs_info::balance_ctl properly
4315 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4317 "balance: cannot set exclusive op status, resume manually");
4319 mutex_lock(&fs_info->balance_mutex);
4320 BUG_ON(fs_info->balance_ctl);
4321 spin_lock(&fs_info->balance_lock);
4322 fs_info->balance_ctl = bctl;
4323 spin_unlock(&fs_info->balance_lock);
4324 mutex_unlock(&fs_info->balance_mutex);
4326 btrfs_free_path(path);
4330 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4334 mutex_lock(&fs_info->balance_mutex);
4335 if (!fs_info->balance_ctl) {
4336 mutex_unlock(&fs_info->balance_mutex);
4340 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4341 atomic_inc(&fs_info->balance_pause_req);
4342 mutex_unlock(&fs_info->balance_mutex);
4344 wait_event(fs_info->balance_wait_q,
4345 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4347 mutex_lock(&fs_info->balance_mutex);
4348 /* we are good with balance_ctl ripped off from under us */
4349 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4350 atomic_dec(&fs_info->balance_pause_req);
4355 mutex_unlock(&fs_info->balance_mutex);
4359 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4361 mutex_lock(&fs_info->balance_mutex);
4362 if (!fs_info->balance_ctl) {
4363 mutex_unlock(&fs_info->balance_mutex);
4368 * A paused balance with the item stored on disk can be resumed at
4369 * mount time if the mount is read-write. Otherwise it's still paused
4370 * and we must not allow cancelling as it deletes the item.
4372 if (sb_rdonly(fs_info->sb)) {
4373 mutex_unlock(&fs_info->balance_mutex);
4377 atomic_inc(&fs_info->balance_cancel_req);
4379 * if we are running just wait and return, balance item is
4380 * deleted in btrfs_balance in this case
4382 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4383 mutex_unlock(&fs_info->balance_mutex);
4384 wait_event(fs_info->balance_wait_q,
4385 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4386 mutex_lock(&fs_info->balance_mutex);
4388 mutex_unlock(&fs_info->balance_mutex);
4390 * Lock released to allow other waiters to continue, we'll
4391 * reexamine the status again.
4393 mutex_lock(&fs_info->balance_mutex);
4395 if (fs_info->balance_ctl) {
4396 reset_balance_state(fs_info);
4397 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4398 btrfs_info(fs_info, "balance: canceled");
4402 BUG_ON(fs_info->balance_ctl ||
4403 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4404 atomic_dec(&fs_info->balance_cancel_req);
4405 mutex_unlock(&fs_info->balance_mutex);
4409 static int btrfs_uuid_scan_kthread(void *data)
4411 struct btrfs_fs_info *fs_info = data;
4412 struct btrfs_root *root = fs_info->tree_root;
4413 struct btrfs_key key;
4414 struct btrfs_path *path = NULL;
4416 struct extent_buffer *eb;
4418 struct btrfs_root_item root_item;
4420 struct btrfs_trans_handle *trans = NULL;
4422 path = btrfs_alloc_path();
4429 key.type = BTRFS_ROOT_ITEM_KEY;
4433 ret = btrfs_search_forward(root, &key, path,
4434 BTRFS_OLDEST_GENERATION);
4441 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4442 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4443 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4444 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4447 eb = path->nodes[0];
4448 slot = path->slots[0];
4449 item_size = btrfs_item_size_nr(eb, slot);
4450 if (item_size < sizeof(root_item))
4453 read_extent_buffer(eb, &root_item,
4454 btrfs_item_ptr_offset(eb, slot),
4455 (int)sizeof(root_item));
4456 if (btrfs_root_refs(&root_item) == 0)
4459 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4460 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4464 btrfs_release_path(path);
4466 * 1 - subvol uuid item
4467 * 1 - received_subvol uuid item
4469 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4470 if (IS_ERR(trans)) {
4471 ret = PTR_ERR(trans);
4479 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4480 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4481 BTRFS_UUID_KEY_SUBVOL,
4484 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4490 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4491 ret = btrfs_uuid_tree_add(trans,
4492 root_item.received_uuid,
4493 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4496 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4504 ret = btrfs_end_transaction(trans);
4510 btrfs_release_path(path);
4511 if (key.offset < (u64)-1) {
4513 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4515 key.type = BTRFS_ROOT_ITEM_KEY;
4516 } else if (key.objectid < (u64)-1) {
4518 key.type = BTRFS_ROOT_ITEM_KEY;
4527 btrfs_free_path(path);
4528 if (trans && !IS_ERR(trans))
4529 btrfs_end_transaction(trans);
4531 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4533 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4534 up(&fs_info->uuid_tree_rescan_sem);
4539 * Callback for btrfs_uuid_tree_iterate().
4541 * 0 check succeeded, the entry is not outdated.
4542 * < 0 if an error occurred.
4543 * > 0 if the check failed, which means the caller shall remove the entry.
4545 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4546 u8 *uuid, u8 type, u64 subid)
4548 struct btrfs_key key;
4550 struct btrfs_root *subvol_root;
4552 if (type != BTRFS_UUID_KEY_SUBVOL &&
4553 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4556 key.objectid = subid;
4557 key.type = BTRFS_ROOT_ITEM_KEY;
4558 key.offset = (u64)-1;
4559 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4560 if (IS_ERR(subvol_root)) {
4561 ret = PTR_ERR(subvol_root);
4568 case BTRFS_UUID_KEY_SUBVOL:
4569 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4572 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4573 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4583 static int btrfs_uuid_rescan_kthread(void *data)
4585 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4589 * 1st step is to iterate through the existing UUID tree and
4590 * to delete all entries that contain outdated data.
4591 * 2nd step is to add all missing entries to the UUID tree.
4593 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4595 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4596 up(&fs_info->uuid_tree_rescan_sem);
4599 return btrfs_uuid_scan_kthread(data);
4602 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4604 struct btrfs_trans_handle *trans;
4605 struct btrfs_root *tree_root = fs_info->tree_root;
4606 struct btrfs_root *uuid_root;
4607 struct task_struct *task;
4614 trans = btrfs_start_transaction(tree_root, 2);
4616 return PTR_ERR(trans);
4618 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4619 if (IS_ERR(uuid_root)) {
4620 ret = PTR_ERR(uuid_root);
4621 btrfs_abort_transaction(trans, ret);
4622 btrfs_end_transaction(trans);
4626 fs_info->uuid_root = uuid_root;
4628 ret = btrfs_commit_transaction(trans);
4632 down(&fs_info->uuid_tree_rescan_sem);
4633 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4635 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4636 btrfs_warn(fs_info, "failed to start uuid_scan task");
4637 up(&fs_info->uuid_tree_rescan_sem);
4638 return PTR_ERR(task);
4644 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4646 struct task_struct *task;
4648 down(&fs_info->uuid_tree_rescan_sem);
4649 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4651 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4652 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4653 up(&fs_info->uuid_tree_rescan_sem);
4654 return PTR_ERR(task);
4661 * shrinking a device means finding all of the device extents past
4662 * the new size, and then following the back refs to the chunks.
4663 * The chunk relocation code actually frees the device extent
4665 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4667 struct btrfs_fs_info *fs_info = device->fs_info;
4668 struct btrfs_root *root = fs_info->dev_root;
4669 struct btrfs_trans_handle *trans;
4670 struct btrfs_dev_extent *dev_extent = NULL;
4671 struct btrfs_path *path;
4677 bool retried = false;
4678 struct extent_buffer *l;
4679 struct btrfs_key key;
4680 struct btrfs_super_block *super_copy = fs_info->super_copy;
4681 u64 old_total = btrfs_super_total_bytes(super_copy);
4682 u64 old_size = btrfs_device_get_total_bytes(device);
4686 new_size = round_down(new_size, fs_info->sectorsize);
4688 diff = round_down(old_size - new_size, fs_info->sectorsize);
4690 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4693 path = btrfs_alloc_path();
4697 path->reada = READA_BACK;
4699 trans = btrfs_start_transaction(root, 0);
4700 if (IS_ERR(trans)) {
4701 btrfs_free_path(path);
4702 return PTR_ERR(trans);
4705 mutex_lock(&fs_info->chunk_mutex);
4707 btrfs_device_set_total_bytes(device, new_size);
4708 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4709 device->fs_devices->total_rw_bytes -= diff;
4710 atomic64_sub(diff, &fs_info->free_chunk_space);
4714 * Once the device's size has been set to the new size, ensure all
4715 * in-memory chunks are synced to disk so that the loop below sees them
4716 * and relocates them accordingly.
4718 if (contains_pending_extent(device, &start, diff)) {
4719 mutex_unlock(&fs_info->chunk_mutex);
4720 ret = btrfs_commit_transaction(trans);
4724 mutex_unlock(&fs_info->chunk_mutex);
4725 btrfs_end_transaction(trans);
4729 key.objectid = device->devid;
4730 key.offset = (u64)-1;
4731 key.type = BTRFS_DEV_EXTENT_KEY;
4734 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4735 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4737 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4741 ret = btrfs_previous_item(root, path, 0, key.type);
4743 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4748 btrfs_release_path(path);
4753 slot = path->slots[0];
4754 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4756 if (key.objectid != device->devid) {
4757 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4758 btrfs_release_path(path);
4762 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4763 length = btrfs_dev_extent_length(l, dev_extent);
4765 if (key.offset + length <= new_size) {
4766 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4767 btrfs_release_path(path);
4771 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4772 btrfs_release_path(path);
4775 * We may be relocating the only data chunk we have,
4776 * which could potentially end up with losing data's
4777 * raid profile, so lets allocate an empty one in
4780 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4782 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4786 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4787 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4788 if (ret == -ENOSPC) {
4791 if (ret == -ETXTBSY) {
4793 "could not shrink block group %llu due to active swapfile",
4798 } while (key.offset-- > 0);
4800 if (failed && !retried) {
4804 } else if (failed && retried) {
4809 /* Shrinking succeeded, else we would be at "done". */
4810 trans = btrfs_start_transaction(root, 0);
4811 if (IS_ERR(trans)) {
4812 ret = PTR_ERR(trans);
4816 mutex_lock(&fs_info->chunk_mutex);
4817 btrfs_device_set_disk_total_bytes(device, new_size);
4818 if (list_empty(&device->post_commit_list))
4819 list_add_tail(&device->post_commit_list,
4820 &trans->transaction->dev_update_list);
4822 WARN_ON(diff > old_total);
4823 btrfs_set_super_total_bytes(super_copy,
4824 round_down(old_total - diff, fs_info->sectorsize));
4825 mutex_unlock(&fs_info->chunk_mutex);
4827 /* Now btrfs_update_device() will change the on-disk size. */
4828 ret = btrfs_update_device(trans, device);
4830 btrfs_abort_transaction(trans, ret);
4831 btrfs_end_transaction(trans);
4833 ret = btrfs_commit_transaction(trans);
4836 btrfs_free_path(path);
4838 mutex_lock(&fs_info->chunk_mutex);
4839 btrfs_device_set_total_bytes(device, old_size);
4840 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4841 device->fs_devices->total_rw_bytes += diff;
4842 atomic64_add(diff, &fs_info->free_chunk_space);
4843 mutex_unlock(&fs_info->chunk_mutex);
4848 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4849 struct btrfs_key *key,
4850 struct btrfs_chunk *chunk, int item_size)
4852 struct btrfs_super_block *super_copy = fs_info->super_copy;
4853 struct btrfs_disk_key disk_key;
4857 mutex_lock(&fs_info->chunk_mutex);
4858 array_size = btrfs_super_sys_array_size(super_copy);
4859 if (array_size + item_size + sizeof(disk_key)
4860 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4861 mutex_unlock(&fs_info->chunk_mutex);
4865 ptr = super_copy->sys_chunk_array + array_size;
4866 btrfs_cpu_key_to_disk(&disk_key, key);
4867 memcpy(ptr, &disk_key, sizeof(disk_key));
4868 ptr += sizeof(disk_key);
4869 memcpy(ptr, chunk, item_size);
4870 item_size += sizeof(disk_key);
4871 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4872 mutex_unlock(&fs_info->chunk_mutex);
4878 * sort the devices in descending order by max_avail, total_avail
4880 static int btrfs_cmp_device_info(const void *a, const void *b)
4882 const struct btrfs_device_info *di_a = a;
4883 const struct btrfs_device_info *di_b = b;
4885 if (di_a->max_avail > di_b->max_avail)
4887 if (di_a->max_avail < di_b->max_avail)
4889 if (di_a->total_avail > di_b->total_avail)
4891 if (di_a->total_avail < di_b->total_avail)
4896 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4898 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4901 btrfs_set_fs_incompat(info, RAID56);
4904 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4905 u64 start, u64 type)
4907 struct btrfs_fs_info *info = trans->fs_info;
4908 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4909 struct btrfs_device *device;
4910 struct map_lookup *map = NULL;
4911 struct extent_map_tree *em_tree;
4912 struct extent_map *em;
4913 struct btrfs_device_info *devices_info = NULL;
4915 int num_stripes; /* total number of stripes to allocate */
4916 int data_stripes; /* number of stripes that count for
4918 int sub_stripes; /* sub_stripes info for map */
4919 int dev_stripes; /* stripes per dev */
4920 int devs_max; /* max devs to use */
4921 int devs_min; /* min devs needed */
4922 int devs_increment; /* ndevs has to be a multiple of this */
4923 int ncopies; /* how many copies to data has */
4924 int nparity; /* number of stripes worth of bytes to
4925 store parity information */
4927 u64 max_stripe_size;
4936 BUG_ON(!alloc_profile_is_valid(type, 0));
4938 if (list_empty(&fs_devices->alloc_list)) {
4939 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4940 btrfs_debug(info, "%s: no writable device", __func__);
4944 index = btrfs_bg_flags_to_raid_index(type);
4946 sub_stripes = btrfs_raid_array[index].sub_stripes;
4947 dev_stripes = btrfs_raid_array[index].dev_stripes;
4948 devs_max = btrfs_raid_array[index].devs_max;
4950 devs_max = BTRFS_MAX_DEVS(info);
4951 devs_min = btrfs_raid_array[index].devs_min;
4952 devs_increment = btrfs_raid_array[index].devs_increment;
4953 ncopies = btrfs_raid_array[index].ncopies;
4954 nparity = btrfs_raid_array[index].nparity;
4956 if (type & BTRFS_BLOCK_GROUP_DATA) {
4957 max_stripe_size = SZ_1G;
4958 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4959 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4960 /* for larger filesystems, use larger metadata chunks */
4961 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4962 max_stripe_size = SZ_1G;
4964 max_stripe_size = SZ_256M;
4965 max_chunk_size = max_stripe_size;
4966 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4967 max_stripe_size = SZ_32M;
4968 max_chunk_size = 2 * max_stripe_size;
4970 btrfs_err(info, "invalid chunk type 0x%llx requested",
4975 /* We don't want a chunk larger than 10% of writable space */
4976 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4979 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4985 * in the first pass through the devices list, we gather information
4986 * about the available holes on each device.
4989 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4993 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4995 "BTRFS: read-only device in alloc_list\n");
4999 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5000 &device->dev_state) ||
5001 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5004 if (device->total_bytes > device->bytes_used)
5005 total_avail = device->total_bytes - device->bytes_used;
5009 /* If there is no space on this device, skip it. */
5010 if (total_avail == 0)
5013 ret = find_free_dev_extent(device,
5014 max_stripe_size * dev_stripes,
5015 &dev_offset, &max_avail);
5016 if (ret && ret != -ENOSPC)
5020 max_avail = max_stripe_size * dev_stripes;
5022 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5023 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5025 "%s: devid %llu has no free space, have=%llu want=%u",
5026 __func__, device->devid, max_avail,
5027 BTRFS_STRIPE_LEN * dev_stripes);
5031 if (ndevs == fs_devices->rw_devices) {
5032 WARN(1, "%s: found more than %llu devices\n",
5033 __func__, fs_devices->rw_devices);
5036 devices_info[ndevs].dev_offset = dev_offset;
5037 devices_info[ndevs].max_avail = max_avail;
5038 devices_info[ndevs].total_avail = total_avail;
5039 devices_info[ndevs].dev = device;
5044 * now sort the devices by hole size / available space
5046 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5047 btrfs_cmp_device_info, NULL);
5049 /* round down to number of usable stripes */
5050 ndevs = round_down(ndevs, devs_increment);
5052 if (ndevs < devs_min) {
5054 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5056 "%s: not enough devices with free space: have=%d minimum required=%d",
5057 __func__, ndevs, devs_min);
5062 ndevs = min(ndevs, devs_max);
5065 * The primary goal is to maximize the number of stripes, so use as
5066 * many devices as possible, even if the stripes are not maximum sized.
5068 * The DUP profile stores more than one stripe per device, the
5069 * max_avail is the total size so we have to adjust.
5071 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5072 num_stripes = ndevs * dev_stripes;
5075 * this will have to be fixed for RAID1 and RAID10 over
5078 data_stripes = (num_stripes - nparity) / ncopies;
5081 * Use the number of data stripes to figure out how big this chunk
5082 * is really going to be in terms of logical address space,
5083 * and compare that answer with the max chunk size. If it's higher,
5084 * we try to reduce stripe_size.
5086 if (stripe_size * data_stripes > max_chunk_size) {
5088 * Reduce stripe_size, round it up to a 16MB boundary again and
5089 * then use it, unless it ends up being even bigger than the
5090 * previous value we had already.
5092 stripe_size = min(round_up(div_u64(max_chunk_size,
5093 data_stripes), SZ_16M),
5097 /* align to BTRFS_STRIPE_LEN */
5098 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5100 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5105 map->num_stripes = num_stripes;
5107 for (i = 0; i < ndevs; ++i) {
5108 for (j = 0; j < dev_stripes; ++j) {
5109 int s = i * dev_stripes + j;
5110 map->stripes[s].dev = devices_info[i].dev;
5111 map->stripes[s].physical = devices_info[i].dev_offset +
5115 map->stripe_len = BTRFS_STRIPE_LEN;
5116 map->io_align = BTRFS_STRIPE_LEN;
5117 map->io_width = BTRFS_STRIPE_LEN;
5119 map->sub_stripes = sub_stripes;
5121 chunk_size = stripe_size * data_stripes;
5123 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5125 em = alloc_extent_map();
5131 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5132 em->map_lookup = map;
5134 em->len = chunk_size;
5135 em->block_start = 0;
5136 em->block_len = em->len;
5137 em->orig_block_len = stripe_size;
5139 em_tree = &info->mapping_tree;
5140 write_lock(&em_tree->lock);
5141 ret = add_extent_mapping(em_tree, em, 0);
5143 write_unlock(&em_tree->lock);
5144 free_extent_map(em);
5147 write_unlock(&em_tree->lock);
5149 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5151 goto error_del_extent;
5153 for (i = 0; i < map->num_stripes; i++) {
5154 struct btrfs_device *dev = map->stripes[i].dev;
5156 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5157 if (list_empty(&dev->post_commit_list))
5158 list_add_tail(&dev->post_commit_list,
5159 &trans->transaction->dev_update_list);
5162 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5164 free_extent_map(em);
5165 check_raid56_incompat_flag(info, type);
5167 kfree(devices_info);
5171 write_lock(&em_tree->lock);
5172 remove_extent_mapping(em_tree, em);
5173 write_unlock(&em_tree->lock);
5175 /* One for our allocation */
5176 free_extent_map(em);
5177 /* One for the tree reference */
5178 free_extent_map(em);
5180 kfree(devices_info);
5184 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5185 u64 chunk_offset, u64 chunk_size)
5187 struct btrfs_fs_info *fs_info = trans->fs_info;
5188 struct btrfs_root *extent_root = fs_info->extent_root;
5189 struct btrfs_root *chunk_root = fs_info->chunk_root;
5190 struct btrfs_key key;
5191 struct btrfs_device *device;
5192 struct btrfs_chunk *chunk;
5193 struct btrfs_stripe *stripe;
5194 struct extent_map *em;
5195 struct map_lookup *map;
5202 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5206 map = em->map_lookup;
5207 item_size = btrfs_chunk_item_size(map->num_stripes);
5208 stripe_size = em->orig_block_len;
5210 chunk = kzalloc(item_size, GFP_NOFS);
5217 * Take the device list mutex to prevent races with the final phase of
5218 * a device replace operation that replaces the device object associated
5219 * with the map's stripes, because the device object's id can change
5220 * at any time during that final phase of the device replace operation
5221 * (dev-replace.c:btrfs_dev_replace_finishing()).
5223 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5224 for (i = 0; i < map->num_stripes; i++) {
5225 device = map->stripes[i].dev;
5226 dev_offset = map->stripes[i].physical;
5228 ret = btrfs_update_device(trans, device);
5231 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5232 dev_offset, stripe_size);
5237 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5241 stripe = &chunk->stripe;
5242 for (i = 0; i < map->num_stripes; i++) {
5243 device = map->stripes[i].dev;
5244 dev_offset = map->stripes[i].physical;
5246 btrfs_set_stack_stripe_devid(stripe, device->devid);
5247 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5248 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5251 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5253 btrfs_set_stack_chunk_length(chunk, chunk_size);
5254 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5255 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5256 btrfs_set_stack_chunk_type(chunk, map->type);
5257 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5258 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5259 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5260 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5261 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5263 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5264 key.type = BTRFS_CHUNK_ITEM_KEY;
5265 key.offset = chunk_offset;
5267 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5268 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5270 * TODO: Cleanup of inserted chunk root in case of
5273 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5278 free_extent_map(em);
5283 * Chunk allocation falls into two parts. The first part does work
5284 * that makes the new allocated chunk usable, but does not do any operation
5285 * that modifies the chunk tree. The second part does the work that
5286 * requires modifying the chunk tree. This division is important for the
5287 * bootstrap process of adding storage to a seed btrfs.
5289 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5293 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5294 chunk_offset = find_next_chunk(trans->fs_info);
5295 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5298 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5300 struct btrfs_fs_info *fs_info = trans->fs_info;
5302 u64 sys_chunk_offset;
5306 chunk_offset = find_next_chunk(fs_info);
5307 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5308 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5312 sys_chunk_offset = find_next_chunk(fs_info);
5313 alloc_profile = btrfs_system_alloc_profile(fs_info);
5314 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5318 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5320 const int index = btrfs_bg_flags_to_raid_index(map->type);
5322 return btrfs_raid_array[index].tolerated_failures;
5325 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5327 struct extent_map *em;
5328 struct map_lookup *map;
5333 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5337 map = em->map_lookup;
5338 for (i = 0; i < map->num_stripes; i++) {
5339 if (test_bit(BTRFS_DEV_STATE_MISSING,
5340 &map->stripes[i].dev->dev_state)) {
5344 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5345 &map->stripes[i].dev->dev_state)) {
5352 * If the number of missing devices is larger than max errors,
5353 * we can not write the data into that chunk successfully, so
5356 if (miss_ndevs > btrfs_chunk_max_errors(map))
5359 free_extent_map(em);
5363 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5365 struct extent_map *em;
5368 write_lock(&tree->lock);
5369 em = lookup_extent_mapping(tree, 0, (u64)-1);
5371 remove_extent_mapping(tree, em);
5372 write_unlock(&tree->lock);
5376 free_extent_map(em);
5377 /* once for the tree */
5378 free_extent_map(em);
5382 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5384 struct extent_map *em;
5385 struct map_lookup *map;
5388 em = btrfs_get_chunk_map(fs_info, logical, len);
5391 * We could return errors for these cases, but that could get
5392 * ugly and we'd probably do the same thing which is just not do
5393 * anything else and exit, so return 1 so the callers don't try
5394 * to use other copies.
5398 map = em->map_lookup;
5399 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5400 ret = map->num_stripes;
5401 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5402 ret = map->sub_stripes;
5403 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5405 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5407 * There could be two corrupted data stripes, we need
5408 * to loop retry in order to rebuild the correct data.
5410 * Fail a stripe at a time on every retry except the
5411 * stripe under reconstruction.
5413 ret = map->num_stripes;
5416 free_extent_map(em);
5418 down_read(&fs_info->dev_replace.rwsem);
5419 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5420 fs_info->dev_replace.tgtdev)
5422 up_read(&fs_info->dev_replace.rwsem);
5427 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5430 struct extent_map *em;
5431 struct map_lookup *map;
5432 unsigned long len = fs_info->sectorsize;
5434 em = btrfs_get_chunk_map(fs_info, logical, len);
5436 if (!WARN_ON(IS_ERR(em))) {
5437 map = em->map_lookup;
5438 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5439 len = map->stripe_len * nr_data_stripes(map);
5440 free_extent_map(em);
5445 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5447 struct extent_map *em;
5448 struct map_lookup *map;
5451 em = btrfs_get_chunk_map(fs_info, logical, len);
5453 if(!WARN_ON(IS_ERR(em))) {
5454 map = em->map_lookup;
5455 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5457 free_extent_map(em);
5462 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5463 struct map_lookup *map, int first,
5464 int dev_replace_is_ongoing)
5468 int preferred_mirror;
5470 struct btrfs_device *srcdev;
5473 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5475 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5476 num_stripes = map->sub_stripes;
5478 num_stripes = map->num_stripes;
5480 preferred_mirror = first + current->pid % num_stripes;
5482 if (dev_replace_is_ongoing &&
5483 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5484 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5485 srcdev = fs_info->dev_replace.srcdev;
5490 * try to avoid the drive that is the source drive for a
5491 * dev-replace procedure, only choose it if no other non-missing
5492 * mirror is available
5494 for (tolerance = 0; tolerance < 2; tolerance++) {
5495 if (map->stripes[preferred_mirror].dev->bdev &&
5496 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5497 return preferred_mirror;
5498 for (i = first; i < first + num_stripes; i++) {
5499 if (map->stripes[i].dev->bdev &&
5500 (tolerance || map->stripes[i].dev != srcdev))
5505 /* we couldn't find one that doesn't fail. Just return something
5506 * and the io error handling code will clean up eventually
5508 return preferred_mirror;
5511 static inline int parity_smaller(u64 a, u64 b)
5516 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5517 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5519 struct btrfs_bio_stripe s;
5526 for (i = 0; i < num_stripes - 1; i++) {
5527 if (parity_smaller(bbio->raid_map[i],
5528 bbio->raid_map[i+1])) {
5529 s = bbio->stripes[i];
5530 l = bbio->raid_map[i];
5531 bbio->stripes[i] = bbio->stripes[i+1];
5532 bbio->raid_map[i] = bbio->raid_map[i+1];
5533 bbio->stripes[i+1] = s;
5534 bbio->raid_map[i+1] = l;
5542 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5544 struct btrfs_bio *bbio = kzalloc(
5545 /* the size of the btrfs_bio */
5546 sizeof(struct btrfs_bio) +
5547 /* plus the variable array for the stripes */
5548 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5549 /* plus the variable array for the tgt dev */
5550 sizeof(int) * (real_stripes) +
5552 * plus the raid_map, which includes both the tgt dev
5555 sizeof(u64) * (total_stripes),
5556 GFP_NOFS|__GFP_NOFAIL);
5558 atomic_set(&bbio->error, 0);
5559 refcount_set(&bbio->refs, 1);
5564 void btrfs_get_bbio(struct btrfs_bio *bbio)
5566 WARN_ON(!refcount_read(&bbio->refs));
5567 refcount_inc(&bbio->refs);
5570 void btrfs_put_bbio(struct btrfs_bio *bbio)
5574 if (refcount_dec_and_test(&bbio->refs))
5578 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5580 * Please note that, discard won't be sent to target device of device
5583 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5584 u64 logical, u64 length,
5585 struct btrfs_bio **bbio_ret)
5587 struct extent_map *em;
5588 struct map_lookup *map;
5589 struct btrfs_bio *bbio;
5593 u64 stripe_end_offset;
5600 u32 sub_stripes = 0;
5601 u64 stripes_per_dev = 0;
5602 u32 remaining_stripes = 0;
5603 u32 last_stripe = 0;
5607 /* discard always return a bbio */
5610 em = btrfs_get_chunk_map(fs_info, logical, length);
5614 map = em->map_lookup;
5615 /* we don't discard raid56 yet */
5616 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5621 offset = logical - em->start;
5622 length = min_t(u64, em->len - offset, length);
5624 stripe_len = map->stripe_len;
5626 * stripe_nr counts the total number of stripes we have to stride
5627 * to get to this block
5629 stripe_nr = div64_u64(offset, stripe_len);
5631 /* stripe_offset is the offset of this block in its stripe */
5632 stripe_offset = offset - stripe_nr * stripe_len;
5634 stripe_nr_end = round_up(offset + length, map->stripe_len);
5635 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5636 stripe_cnt = stripe_nr_end - stripe_nr;
5637 stripe_end_offset = stripe_nr_end * map->stripe_len -
5640 * after this, stripe_nr is the number of stripes on this
5641 * device we have to walk to find the data, and stripe_index is
5642 * the number of our device in the stripe array
5646 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5647 BTRFS_BLOCK_GROUP_RAID10)) {
5648 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5651 sub_stripes = map->sub_stripes;
5653 factor = map->num_stripes / sub_stripes;
5654 num_stripes = min_t(u64, map->num_stripes,
5655 sub_stripes * stripe_cnt);
5656 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5657 stripe_index *= sub_stripes;
5658 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5659 &remaining_stripes);
5660 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5661 last_stripe *= sub_stripes;
5662 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5663 BTRFS_BLOCK_GROUP_DUP)) {
5664 num_stripes = map->num_stripes;
5666 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5670 bbio = alloc_btrfs_bio(num_stripes, 0);
5676 for (i = 0; i < num_stripes; i++) {
5677 bbio->stripes[i].physical =
5678 map->stripes[stripe_index].physical +
5679 stripe_offset + stripe_nr * map->stripe_len;
5680 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5682 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5683 BTRFS_BLOCK_GROUP_RAID10)) {
5684 bbio->stripes[i].length = stripes_per_dev *
5687 if (i / sub_stripes < remaining_stripes)
5688 bbio->stripes[i].length +=
5692 * Special for the first stripe and
5695 * |-------|...|-------|
5699 if (i < sub_stripes)
5700 bbio->stripes[i].length -=
5703 if (stripe_index >= last_stripe &&
5704 stripe_index <= (last_stripe +
5706 bbio->stripes[i].length -=
5709 if (i == sub_stripes - 1)
5712 bbio->stripes[i].length = length;
5716 if (stripe_index == map->num_stripes) {
5723 bbio->map_type = map->type;
5724 bbio->num_stripes = num_stripes;
5726 free_extent_map(em);
5731 * In dev-replace case, for repair case (that's the only case where the mirror
5732 * is selected explicitly when calling btrfs_map_block), blocks left of the
5733 * left cursor can also be read from the target drive.
5735 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5737 * For READ, it also needs to be supported using the same mirror number.
5739 * If the requested block is not left of the left cursor, EIO is returned. This
5740 * can happen because btrfs_num_copies() returns one more in the dev-replace
5743 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5744 u64 logical, u64 length,
5745 u64 srcdev_devid, int *mirror_num,
5748 struct btrfs_bio *bbio = NULL;
5750 int index_srcdev = 0;
5752 u64 physical_of_found = 0;
5756 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5757 logical, &length, &bbio, 0, 0);
5759 ASSERT(bbio == NULL);
5763 num_stripes = bbio->num_stripes;
5764 if (*mirror_num > num_stripes) {
5766 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5767 * that means that the requested area is not left of the left
5770 btrfs_put_bbio(bbio);
5775 * process the rest of the function using the mirror_num of the source
5776 * drive. Therefore look it up first. At the end, patch the device
5777 * pointer to the one of the target drive.
5779 for (i = 0; i < num_stripes; i++) {
5780 if (bbio->stripes[i].dev->devid != srcdev_devid)
5784 * In case of DUP, in order to keep it simple, only add the
5785 * mirror with the lowest physical address
5788 physical_of_found <= bbio->stripes[i].physical)
5793 physical_of_found = bbio->stripes[i].physical;
5796 btrfs_put_bbio(bbio);
5802 *mirror_num = index_srcdev + 1;
5803 *physical = physical_of_found;
5807 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5808 struct btrfs_bio **bbio_ret,
5809 struct btrfs_dev_replace *dev_replace,
5810 int *num_stripes_ret, int *max_errors_ret)
5812 struct btrfs_bio *bbio = *bbio_ret;
5813 u64 srcdev_devid = dev_replace->srcdev->devid;
5814 int tgtdev_indexes = 0;
5815 int num_stripes = *num_stripes_ret;
5816 int max_errors = *max_errors_ret;
5819 if (op == BTRFS_MAP_WRITE) {
5820 int index_where_to_add;
5823 * duplicate the write operations while the dev replace
5824 * procedure is running. Since the copying of the old disk to
5825 * the new disk takes place at run time while the filesystem is
5826 * mounted writable, the regular write operations to the old
5827 * disk have to be duplicated to go to the new disk as well.
5829 * Note that device->missing is handled by the caller, and that
5830 * the write to the old disk is already set up in the stripes
5833 index_where_to_add = num_stripes;
5834 for (i = 0; i < num_stripes; i++) {
5835 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5836 /* write to new disk, too */
5837 struct btrfs_bio_stripe *new =
5838 bbio->stripes + index_where_to_add;
5839 struct btrfs_bio_stripe *old =
5842 new->physical = old->physical;
5843 new->length = old->length;
5844 new->dev = dev_replace->tgtdev;
5845 bbio->tgtdev_map[i] = index_where_to_add;
5846 index_where_to_add++;
5851 num_stripes = index_where_to_add;
5852 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5853 int index_srcdev = 0;
5855 u64 physical_of_found = 0;
5858 * During the dev-replace procedure, the target drive can also
5859 * be used to read data in case it is needed to repair a corrupt
5860 * block elsewhere. This is possible if the requested area is
5861 * left of the left cursor. In this area, the target drive is a
5862 * full copy of the source drive.
5864 for (i = 0; i < num_stripes; i++) {
5865 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5867 * In case of DUP, in order to keep it simple,
5868 * only add the mirror with the lowest physical
5872 physical_of_found <=
5873 bbio->stripes[i].physical)
5877 physical_of_found = bbio->stripes[i].physical;
5881 struct btrfs_bio_stripe *tgtdev_stripe =
5882 bbio->stripes + num_stripes;
5884 tgtdev_stripe->physical = physical_of_found;
5885 tgtdev_stripe->length =
5886 bbio->stripes[index_srcdev].length;
5887 tgtdev_stripe->dev = dev_replace->tgtdev;
5888 bbio->tgtdev_map[index_srcdev] = num_stripes;
5895 *num_stripes_ret = num_stripes;
5896 *max_errors_ret = max_errors;
5897 bbio->num_tgtdevs = tgtdev_indexes;
5901 static bool need_full_stripe(enum btrfs_map_op op)
5903 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5907 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5908 * tuple. This information is used to calculate how big a
5909 * particular bio can get before it straddles a stripe.
5911 * @fs_info - the filesystem
5912 * @logical - address that we want to figure out the geometry of
5913 * @len - the length of IO we are going to perform, starting at @logical
5914 * @op - type of operation - write or read
5915 * @io_geom - pointer used to return values
5917 * Returns < 0 in case a chunk for the given logical address cannot be found,
5918 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5920 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5921 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5923 struct extent_map *em;
5924 struct map_lookup *map;
5929 u64 raid56_full_stripe_start = (u64)-1;
5933 ASSERT(op != BTRFS_MAP_DISCARD);
5935 em = btrfs_get_chunk_map(fs_info, logical, len);
5939 map = em->map_lookup;
5940 /* Offset of this logical address in the chunk */
5941 offset = logical - em->start;
5942 /* Len of a stripe in a chunk */
5943 stripe_len = map->stripe_len;
5944 /* Stripe wher this block falls in */
5945 stripe_nr = div64_u64(offset, stripe_len);
5946 /* Offset of stripe in the chunk */
5947 stripe_offset = stripe_nr * stripe_len;
5948 if (offset < stripe_offset) {
5950 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5951 stripe_offset, offset, em->start, logical, stripe_len);
5956 /* stripe_offset is the offset of this block in its stripe */
5957 stripe_offset = offset - stripe_offset;
5958 data_stripes = nr_data_stripes(map);
5960 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5961 u64 max_len = stripe_len - stripe_offset;
5964 * In case of raid56, we need to know the stripe aligned start
5966 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5967 unsigned long full_stripe_len = stripe_len * data_stripes;
5968 raid56_full_stripe_start = offset;
5971 * Allow a write of a full stripe, but make sure we
5972 * don't allow straddling of stripes
5974 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5976 raid56_full_stripe_start *= full_stripe_len;
5979 * For writes to RAID[56], allow a full stripeset across
5980 * all disks. For other RAID types and for RAID[56]
5981 * reads, just allow a single stripe (on a single disk).
5983 if (op == BTRFS_MAP_WRITE) {
5984 max_len = stripe_len * data_stripes -
5985 (offset - raid56_full_stripe_start);
5988 len = min_t(u64, em->len - offset, max_len);
5990 len = em->len - offset;
5994 io_geom->offset = offset;
5995 io_geom->stripe_len = stripe_len;
5996 io_geom->stripe_nr = stripe_nr;
5997 io_geom->stripe_offset = stripe_offset;
5998 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6002 free_extent_map(em);
6006 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6007 enum btrfs_map_op op,
6008 u64 logical, u64 *length,
6009 struct btrfs_bio **bbio_ret,
6010 int mirror_num, int need_raid_map)
6012 struct extent_map *em;
6013 struct map_lookup *map;
6024 int tgtdev_indexes = 0;
6025 struct btrfs_bio *bbio = NULL;
6026 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6027 int dev_replace_is_ongoing = 0;
6028 int num_alloc_stripes;
6029 int patch_the_first_stripe_for_dev_replace = 0;
6030 u64 physical_to_patch_in_first_stripe = 0;
6031 u64 raid56_full_stripe_start = (u64)-1;
6032 struct btrfs_io_geometry geom;
6036 if (op == BTRFS_MAP_DISCARD)
6037 return __btrfs_map_block_for_discard(fs_info, logical,
6040 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6044 em = btrfs_get_chunk_map(fs_info, logical, *length);
6046 map = em->map_lookup;
6049 offset = geom.offset;
6050 stripe_len = geom.stripe_len;
6051 stripe_nr = geom.stripe_nr;
6052 stripe_offset = geom.stripe_offset;
6053 raid56_full_stripe_start = geom.raid56_stripe_offset;
6054 data_stripes = nr_data_stripes(map);
6056 down_read(&dev_replace->rwsem);
6057 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6059 * Hold the semaphore for read during the whole operation, write is
6060 * requested at commit time but must wait.
6062 if (!dev_replace_is_ongoing)
6063 up_read(&dev_replace->rwsem);
6065 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6066 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6067 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6068 dev_replace->srcdev->devid,
6070 &physical_to_patch_in_first_stripe);
6074 patch_the_first_stripe_for_dev_replace = 1;
6075 } else if (mirror_num > map->num_stripes) {
6081 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6082 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6084 if (!need_full_stripe(op))
6086 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6087 if (need_full_stripe(op))
6088 num_stripes = map->num_stripes;
6089 else if (mirror_num)
6090 stripe_index = mirror_num - 1;
6092 stripe_index = find_live_mirror(fs_info, map, 0,
6093 dev_replace_is_ongoing);
6094 mirror_num = stripe_index + 1;
6097 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6098 if (need_full_stripe(op)) {
6099 num_stripes = map->num_stripes;
6100 } else if (mirror_num) {
6101 stripe_index = mirror_num - 1;
6106 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6107 u32 factor = map->num_stripes / map->sub_stripes;
6109 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6110 stripe_index *= map->sub_stripes;
6112 if (need_full_stripe(op))
6113 num_stripes = map->sub_stripes;
6114 else if (mirror_num)
6115 stripe_index += mirror_num - 1;
6117 int old_stripe_index = stripe_index;
6118 stripe_index = find_live_mirror(fs_info, map,
6120 dev_replace_is_ongoing);
6121 mirror_num = stripe_index - old_stripe_index + 1;
6124 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6125 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6126 /* push stripe_nr back to the start of the full stripe */
6127 stripe_nr = div64_u64(raid56_full_stripe_start,
6128 stripe_len * data_stripes);
6130 /* RAID[56] write or recovery. Return all stripes */
6131 num_stripes = map->num_stripes;
6132 max_errors = nr_parity_stripes(map);
6134 *length = map->stripe_len;
6139 * Mirror #0 or #1 means the original data block.
6140 * Mirror #2 is RAID5 parity block.
6141 * Mirror #3 is RAID6 Q block.
6143 stripe_nr = div_u64_rem(stripe_nr,
6144 data_stripes, &stripe_index);
6146 stripe_index = data_stripes + mirror_num - 2;
6148 /* We distribute the parity blocks across stripes */
6149 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6151 if (!need_full_stripe(op) && mirror_num <= 1)
6156 * after this, stripe_nr is the number of stripes on this
6157 * device we have to walk to find the data, and stripe_index is
6158 * the number of our device in the stripe array
6160 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6162 mirror_num = stripe_index + 1;
6164 if (stripe_index >= map->num_stripes) {
6166 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6167 stripe_index, map->num_stripes);
6172 num_alloc_stripes = num_stripes;
6173 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6174 if (op == BTRFS_MAP_WRITE)
6175 num_alloc_stripes <<= 1;
6176 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6177 num_alloc_stripes++;
6178 tgtdev_indexes = num_stripes;
6181 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6186 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6187 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6189 /* build raid_map */
6190 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6191 (need_full_stripe(op) || mirror_num > 1)) {
6195 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6196 sizeof(struct btrfs_bio_stripe) *
6198 sizeof(int) * tgtdev_indexes);
6200 /* Work out the disk rotation on this stripe-set */
6201 div_u64_rem(stripe_nr, num_stripes, &rot);
6203 /* Fill in the logical address of each stripe */
6204 tmp = stripe_nr * data_stripes;
6205 for (i = 0; i < data_stripes; i++)
6206 bbio->raid_map[(i+rot) % num_stripes] =
6207 em->start + (tmp + i) * map->stripe_len;
6209 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6210 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6211 bbio->raid_map[(i+rot+1) % num_stripes] =
6216 for (i = 0; i < num_stripes; i++) {
6217 bbio->stripes[i].physical =
6218 map->stripes[stripe_index].physical +
6220 stripe_nr * map->stripe_len;
6221 bbio->stripes[i].dev =
6222 map->stripes[stripe_index].dev;
6226 if (need_full_stripe(op))
6227 max_errors = btrfs_chunk_max_errors(map);
6230 sort_parity_stripes(bbio, num_stripes);
6232 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6233 need_full_stripe(op)) {
6234 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6239 bbio->map_type = map->type;
6240 bbio->num_stripes = num_stripes;
6241 bbio->max_errors = max_errors;
6242 bbio->mirror_num = mirror_num;
6245 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6246 * mirror_num == num_stripes + 1 && dev_replace target drive is
6247 * available as a mirror
6249 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6250 WARN_ON(num_stripes > 1);
6251 bbio->stripes[0].dev = dev_replace->tgtdev;
6252 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6253 bbio->mirror_num = map->num_stripes + 1;
6256 if (dev_replace_is_ongoing) {
6257 lockdep_assert_held(&dev_replace->rwsem);
6258 /* Unlock and let waiting writers proceed */
6259 up_read(&dev_replace->rwsem);
6261 free_extent_map(em);
6265 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6266 u64 logical, u64 *length,
6267 struct btrfs_bio **bbio_ret, int mirror_num)
6269 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6273 /* For Scrub/replace */
6274 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6275 u64 logical, u64 *length,
6276 struct btrfs_bio **bbio_ret)
6278 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6281 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6282 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6284 struct extent_map *em;
6285 struct map_lookup *map;
6293 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6297 map = em->map_lookup;
6299 rmap_len = map->stripe_len;
6301 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6302 length = div_u64(length, map->num_stripes / map->sub_stripes);
6303 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6304 length = div_u64(length, map->num_stripes);
6305 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6306 length = div_u64(length, nr_data_stripes(map));
6307 rmap_len = map->stripe_len * nr_data_stripes(map);
6310 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6311 BUG_ON(!buf); /* -ENOMEM */
6313 for (i = 0; i < map->num_stripes; i++) {
6314 if (map->stripes[i].physical > physical ||
6315 map->stripes[i].physical + length <= physical)
6318 stripe_nr = physical - map->stripes[i].physical;
6319 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6321 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6322 stripe_nr = stripe_nr * map->num_stripes + i;
6323 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6324 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6325 stripe_nr = stripe_nr * map->num_stripes + i;
6326 } /* else if RAID[56], multiply by nr_data_stripes().
6327 * Alternatively, just use rmap_len below instead of
6328 * map->stripe_len */
6330 bytenr = chunk_start + stripe_nr * rmap_len;
6331 WARN_ON(nr >= map->num_stripes);
6332 for (j = 0; j < nr; j++) {
6333 if (buf[j] == bytenr)
6337 WARN_ON(nr >= map->num_stripes);
6344 *stripe_len = rmap_len;
6346 free_extent_map(em);
6350 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6352 bio->bi_private = bbio->private;
6353 bio->bi_end_io = bbio->end_io;
6356 btrfs_put_bbio(bbio);
6359 static void btrfs_end_bio(struct bio *bio)
6361 struct btrfs_bio *bbio = bio->bi_private;
6362 int is_orig_bio = 0;
6364 if (bio->bi_status) {
6365 atomic_inc(&bbio->error);
6366 if (bio->bi_status == BLK_STS_IOERR ||
6367 bio->bi_status == BLK_STS_TARGET) {
6368 unsigned int stripe_index =
6369 btrfs_io_bio(bio)->stripe_index;
6370 struct btrfs_device *dev;
6372 BUG_ON(stripe_index >= bbio->num_stripes);
6373 dev = bbio->stripes[stripe_index].dev;
6375 if (bio_op(bio) == REQ_OP_WRITE)
6376 btrfs_dev_stat_inc_and_print(dev,
6377 BTRFS_DEV_STAT_WRITE_ERRS);
6378 else if (!(bio->bi_opf & REQ_RAHEAD))
6379 btrfs_dev_stat_inc_and_print(dev,
6380 BTRFS_DEV_STAT_READ_ERRS);
6381 if (bio->bi_opf & REQ_PREFLUSH)
6382 btrfs_dev_stat_inc_and_print(dev,
6383 BTRFS_DEV_STAT_FLUSH_ERRS);
6388 if (bio == bbio->orig_bio)
6391 btrfs_bio_counter_dec(bbio->fs_info);
6393 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6396 bio = bbio->orig_bio;
6399 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6400 /* only send an error to the higher layers if it is
6401 * beyond the tolerance of the btrfs bio
6403 if (atomic_read(&bbio->error) > bbio->max_errors) {
6404 bio->bi_status = BLK_STS_IOERR;
6407 * this bio is actually up to date, we didn't
6408 * go over the max number of errors
6410 bio->bi_status = BLK_STS_OK;
6413 btrfs_end_bbio(bbio, bio);
6414 } else if (!is_orig_bio) {
6420 * see run_scheduled_bios for a description of why bios are collected for
6423 * This will add one bio to the pending list for a device and make sure
6424 * the work struct is scheduled.
6426 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6429 struct btrfs_fs_info *fs_info = device->fs_info;
6430 int should_queue = 1;
6431 struct btrfs_pending_bios *pending_bios;
6433 /* don't bother with additional async steps for reads, right now */
6434 if (bio_op(bio) == REQ_OP_READ) {
6435 btrfsic_submit_bio(bio);
6439 WARN_ON(bio->bi_next);
6440 bio->bi_next = NULL;
6442 spin_lock(&device->io_lock);
6443 if (op_is_sync(bio->bi_opf))
6444 pending_bios = &device->pending_sync_bios;
6446 pending_bios = &device->pending_bios;
6448 if (pending_bios->tail)
6449 pending_bios->tail->bi_next = bio;
6451 pending_bios->tail = bio;
6452 if (!pending_bios->head)
6453 pending_bios->head = bio;
6454 if (device->running_pending)
6457 spin_unlock(&device->io_lock);
6460 btrfs_queue_work(fs_info->submit_workers, &device->work);
6463 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6464 u64 physical, int dev_nr, int async)
6466 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6467 struct btrfs_fs_info *fs_info = bbio->fs_info;
6469 bio->bi_private = bbio;
6470 btrfs_io_bio(bio)->stripe_index = dev_nr;
6471 bio->bi_end_io = btrfs_end_bio;
6472 bio->bi_iter.bi_sector = physical >> 9;
6473 btrfs_debug_in_rcu(fs_info,
6474 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6475 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6476 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6477 bio->bi_iter.bi_size);
6478 bio_set_dev(bio, dev->bdev);
6480 btrfs_bio_counter_inc_noblocked(fs_info);
6483 btrfs_schedule_bio(dev, bio);
6485 btrfsic_submit_bio(bio);
6488 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6490 atomic_inc(&bbio->error);
6491 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6492 /* Should be the original bio. */
6493 WARN_ON(bio != bbio->orig_bio);
6495 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6496 bio->bi_iter.bi_sector = logical >> 9;
6497 if (atomic_read(&bbio->error) > bbio->max_errors)
6498 bio->bi_status = BLK_STS_IOERR;
6500 bio->bi_status = BLK_STS_OK;
6501 btrfs_end_bbio(bbio, bio);
6505 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6506 int mirror_num, int async_submit)
6508 struct btrfs_device *dev;
6509 struct bio *first_bio = bio;
6510 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6516 struct btrfs_bio *bbio = NULL;
6518 length = bio->bi_iter.bi_size;
6519 map_length = length;
6521 btrfs_bio_counter_inc_blocked(fs_info);
6522 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6523 &map_length, &bbio, mirror_num, 1);
6525 btrfs_bio_counter_dec(fs_info);
6526 return errno_to_blk_status(ret);
6529 total_devs = bbio->num_stripes;
6530 bbio->orig_bio = first_bio;
6531 bbio->private = first_bio->bi_private;
6532 bbio->end_io = first_bio->bi_end_io;
6533 bbio->fs_info = fs_info;
6534 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6536 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6537 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6538 /* In this case, map_length has been set to the length of
6539 a single stripe; not the whole write */
6540 if (bio_op(bio) == REQ_OP_WRITE) {
6541 ret = raid56_parity_write(fs_info, bio, bbio,
6544 ret = raid56_parity_recover(fs_info, bio, bbio,
6545 map_length, mirror_num, 1);
6548 btrfs_bio_counter_dec(fs_info);
6549 return errno_to_blk_status(ret);
6552 if (map_length < length) {
6554 "mapping failed logical %llu bio len %llu len %llu",
6555 logical, length, map_length);
6559 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6560 dev = bbio->stripes[dev_nr].dev;
6561 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6563 (bio_op(first_bio) == REQ_OP_WRITE &&
6564 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6565 bbio_error(bbio, first_bio, logical);
6569 if (dev_nr < total_devs - 1)
6570 bio = btrfs_bio_clone(first_bio);
6574 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6575 dev_nr, async_submit);
6577 btrfs_bio_counter_dec(fs_info);
6582 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6585 * If devid and uuid are both specified, the match must be exact, otherwise
6586 * only devid is used.
6588 * If @seed is true, traverse through the seed devices.
6590 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6591 u64 devid, u8 *uuid, u8 *fsid,
6594 struct btrfs_device *device;
6596 while (fs_devices) {
6598 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6599 list_for_each_entry(device, &fs_devices->devices,
6601 if (device->devid == devid &&
6602 (!uuid || memcmp(device->uuid, uuid,
6603 BTRFS_UUID_SIZE) == 0))
6608 fs_devices = fs_devices->seed;
6615 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6616 u64 devid, u8 *dev_uuid)
6618 struct btrfs_device *device;
6620 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6624 list_add(&device->dev_list, &fs_devices->devices);
6625 device->fs_devices = fs_devices;
6626 fs_devices->num_devices++;
6628 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6629 fs_devices->missing_devices++;
6635 * btrfs_alloc_device - allocate struct btrfs_device
6636 * @fs_info: used only for generating a new devid, can be NULL if
6637 * devid is provided (i.e. @devid != NULL).
6638 * @devid: a pointer to devid for this device. If NULL a new devid
6640 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6643 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6644 * on error. Returned struct is not linked onto any lists and must be
6645 * destroyed with btrfs_free_device.
6647 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6651 struct btrfs_device *dev;
6654 if (WARN_ON(!devid && !fs_info))
6655 return ERR_PTR(-EINVAL);
6657 dev = __alloc_device();
6666 ret = find_next_devid(fs_info, &tmp);
6668 btrfs_free_device(dev);
6669 return ERR_PTR(ret);
6675 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6677 generate_random_uuid(dev->uuid);
6679 btrfs_init_work(&dev->work, btrfs_submit_helper,
6680 pending_bios_fn, NULL, NULL);
6685 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6686 u64 devid, u8 *uuid, bool error)
6689 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6692 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6696 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6698 int index = btrfs_bg_flags_to_raid_index(type);
6699 int ncopies = btrfs_raid_array[index].ncopies;
6702 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6703 case BTRFS_BLOCK_GROUP_RAID5:
6704 data_stripes = num_stripes - 1;
6706 case BTRFS_BLOCK_GROUP_RAID6:
6707 data_stripes = num_stripes - 2;
6710 data_stripes = num_stripes / ncopies;
6713 return div_u64(chunk_len, data_stripes);
6716 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6717 struct btrfs_chunk *chunk)
6719 struct btrfs_fs_info *fs_info = leaf->fs_info;
6720 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6721 struct map_lookup *map;
6722 struct extent_map *em;
6726 u8 uuid[BTRFS_UUID_SIZE];
6731 logical = key->offset;
6732 length = btrfs_chunk_length(leaf, chunk);
6733 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6736 * Only need to verify chunk item if we're reading from sys chunk array,
6737 * as chunk item in tree block is already verified by tree-checker.
6739 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6740 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6745 read_lock(&map_tree->lock);
6746 em = lookup_extent_mapping(map_tree, logical, 1);
6747 read_unlock(&map_tree->lock);
6749 /* already mapped? */
6750 if (em && em->start <= logical && em->start + em->len > logical) {
6751 free_extent_map(em);
6754 free_extent_map(em);
6757 em = alloc_extent_map();
6760 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6762 free_extent_map(em);
6766 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6767 em->map_lookup = map;
6768 em->start = logical;
6771 em->block_start = 0;
6772 em->block_len = em->len;
6774 map->num_stripes = num_stripes;
6775 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6776 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6777 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6778 map->type = btrfs_chunk_type(leaf, chunk);
6779 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6780 map->verified_stripes = 0;
6781 em->orig_block_len = calc_stripe_length(map->type, em->len,
6783 for (i = 0; i < num_stripes; i++) {
6784 map->stripes[i].physical =
6785 btrfs_stripe_offset_nr(leaf, chunk, i);
6786 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6787 read_extent_buffer(leaf, uuid, (unsigned long)
6788 btrfs_stripe_dev_uuid_nr(chunk, i),
6790 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6791 devid, uuid, NULL, true);
6792 if (!map->stripes[i].dev &&
6793 !btrfs_test_opt(fs_info, DEGRADED)) {
6794 free_extent_map(em);
6795 btrfs_report_missing_device(fs_info, devid, uuid, true);
6798 if (!map->stripes[i].dev) {
6799 map->stripes[i].dev =
6800 add_missing_dev(fs_info->fs_devices, devid,
6802 if (IS_ERR(map->stripes[i].dev)) {
6803 free_extent_map(em);
6805 "failed to init missing dev %llu: %ld",
6806 devid, PTR_ERR(map->stripes[i].dev));
6807 return PTR_ERR(map->stripes[i].dev);
6809 btrfs_report_missing_device(fs_info, devid, uuid, false);
6811 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6812 &(map->stripes[i].dev->dev_state));
6816 write_lock(&map_tree->lock);
6817 ret = add_extent_mapping(map_tree, em, 0);
6818 write_unlock(&map_tree->lock);
6821 "failed to add chunk map, start=%llu len=%llu: %d",
6822 em->start, em->len, ret);
6824 free_extent_map(em);
6829 static void fill_device_from_item(struct extent_buffer *leaf,
6830 struct btrfs_dev_item *dev_item,
6831 struct btrfs_device *device)
6835 device->devid = btrfs_device_id(leaf, dev_item);
6836 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6837 device->total_bytes = device->disk_total_bytes;
6838 device->commit_total_bytes = device->disk_total_bytes;
6839 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6840 device->commit_bytes_used = device->bytes_used;
6841 device->type = btrfs_device_type(leaf, dev_item);
6842 device->io_align = btrfs_device_io_align(leaf, dev_item);
6843 device->io_width = btrfs_device_io_width(leaf, dev_item);
6844 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6845 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6846 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6848 ptr = btrfs_device_uuid(dev_item);
6849 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6852 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6855 struct btrfs_fs_devices *fs_devices;
6858 lockdep_assert_held(&uuid_mutex);
6861 fs_devices = fs_info->fs_devices->seed;
6862 while (fs_devices) {
6863 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6866 fs_devices = fs_devices->seed;
6869 fs_devices = find_fsid(fsid, NULL);
6871 if (!btrfs_test_opt(fs_info, DEGRADED))
6872 return ERR_PTR(-ENOENT);
6874 fs_devices = alloc_fs_devices(fsid, NULL);
6875 if (IS_ERR(fs_devices))
6878 fs_devices->seeding = 1;
6879 fs_devices->opened = 1;
6883 fs_devices = clone_fs_devices(fs_devices);
6884 if (IS_ERR(fs_devices))
6887 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6889 free_fs_devices(fs_devices);
6890 fs_devices = ERR_PTR(ret);
6894 if (!fs_devices->seeding) {
6895 close_fs_devices(fs_devices);
6896 free_fs_devices(fs_devices);
6897 fs_devices = ERR_PTR(-EINVAL);
6901 fs_devices->seed = fs_info->fs_devices->seed;
6902 fs_info->fs_devices->seed = fs_devices;
6907 static int read_one_dev(struct extent_buffer *leaf,
6908 struct btrfs_dev_item *dev_item)
6910 struct btrfs_fs_info *fs_info = leaf->fs_info;
6911 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6912 struct btrfs_device *device;
6915 u8 fs_uuid[BTRFS_FSID_SIZE];
6916 u8 dev_uuid[BTRFS_UUID_SIZE];
6918 devid = btrfs_device_id(leaf, dev_item);
6919 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6921 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6924 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6925 fs_devices = open_seed_devices(fs_info, fs_uuid);
6926 if (IS_ERR(fs_devices))
6927 return PTR_ERR(fs_devices);
6930 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6933 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6934 btrfs_report_missing_device(fs_info, devid,
6939 device = add_missing_dev(fs_devices, devid, dev_uuid);
6940 if (IS_ERR(device)) {
6942 "failed to add missing dev %llu: %ld",
6943 devid, PTR_ERR(device));
6944 return PTR_ERR(device);
6946 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6948 if (!device->bdev) {
6949 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6950 btrfs_report_missing_device(fs_info,
6951 devid, dev_uuid, true);
6954 btrfs_report_missing_device(fs_info, devid,
6958 if (!device->bdev &&
6959 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6961 * this happens when a device that was properly setup
6962 * in the device info lists suddenly goes bad.
6963 * device->bdev is NULL, and so we have to set
6964 * device->missing to one here
6966 device->fs_devices->missing_devices++;
6967 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6970 /* Move the device to its own fs_devices */
6971 if (device->fs_devices != fs_devices) {
6972 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6973 &device->dev_state));
6975 list_move(&device->dev_list, &fs_devices->devices);
6976 device->fs_devices->num_devices--;
6977 fs_devices->num_devices++;
6979 device->fs_devices->missing_devices--;
6980 fs_devices->missing_devices++;
6982 device->fs_devices = fs_devices;
6986 if (device->fs_devices != fs_info->fs_devices) {
6987 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6988 if (device->generation !=
6989 btrfs_device_generation(leaf, dev_item))
6993 fill_device_from_item(leaf, dev_item, device);
6994 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6995 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6996 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6997 device->fs_devices->total_rw_bytes += device->total_bytes;
6998 atomic64_add(device->total_bytes - device->bytes_used,
6999 &fs_info->free_chunk_space);
7005 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7007 struct btrfs_root *root = fs_info->tree_root;
7008 struct btrfs_super_block *super_copy = fs_info->super_copy;
7009 struct extent_buffer *sb;
7010 struct btrfs_disk_key *disk_key;
7011 struct btrfs_chunk *chunk;
7013 unsigned long sb_array_offset;
7020 struct btrfs_key key;
7022 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7024 * This will create extent buffer of nodesize, superblock size is
7025 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7026 * overallocate but we can keep it as-is, only the first page is used.
7028 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7031 set_extent_buffer_uptodate(sb);
7032 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7034 * The sb extent buffer is artificial and just used to read the system array.
7035 * set_extent_buffer_uptodate() call does not properly mark all it's
7036 * pages up-to-date when the page is larger: extent does not cover the
7037 * whole page and consequently check_page_uptodate does not find all
7038 * the page's extents up-to-date (the hole beyond sb),
7039 * write_extent_buffer then triggers a WARN_ON.
7041 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7042 * but sb spans only this function. Add an explicit SetPageUptodate call
7043 * to silence the warning eg. on PowerPC 64.
7045 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7046 SetPageUptodate(sb->pages[0]);
7048 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7049 array_size = btrfs_super_sys_array_size(super_copy);
7051 array_ptr = super_copy->sys_chunk_array;
7052 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7055 while (cur_offset < array_size) {
7056 disk_key = (struct btrfs_disk_key *)array_ptr;
7057 len = sizeof(*disk_key);
7058 if (cur_offset + len > array_size)
7059 goto out_short_read;
7061 btrfs_disk_key_to_cpu(&key, disk_key);
7064 sb_array_offset += len;
7067 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7068 chunk = (struct btrfs_chunk *)sb_array_offset;
7070 * At least one btrfs_chunk with one stripe must be
7071 * present, exact stripe count check comes afterwards
7073 len = btrfs_chunk_item_size(1);
7074 if (cur_offset + len > array_size)
7075 goto out_short_read;
7077 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7080 "invalid number of stripes %u in sys_array at offset %u",
7081 num_stripes, cur_offset);
7086 type = btrfs_chunk_type(sb, chunk);
7087 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7089 "invalid chunk type %llu in sys_array at offset %u",
7095 len = btrfs_chunk_item_size(num_stripes);
7096 if (cur_offset + len > array_size)
7097 goto out_short_read;
7099 ret = read_one_chunk(&key, sb, chunk);
7104 "unexpected item type %u in sys_array at offset %u",
7105 (u32)key.type, cur_offset);
7110 sb_array_offset += len;
7113 clear_extent_buffer_uptodate(sb);
7114 free_extent_buffer_stale(sb);
7118 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7120 clear_extent_buffer_uptodate(sb);
7121 free_extent_buffer_stale(sb);
7126 * Check if all chunks in the fs are OK for read-write degraded mount
7128 * If the @failing_dev is specified, it's accounted as missing.
7130 * Return true if all chunks meet the minimal RW mount requirements.
7131 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7133 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7134 struct btrfs_device *failing_dev)
7136 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7137 struct extent_map *em;
7141 read_lock(&map_tree->lock);
7142 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7143 read_unlock(&map_tree->lock);
7144 /* No chunk at all? Return false anyway */
7150 struct map_lookup *map;
7155 map = em->map_lookup;
7157 btrfs_get_num_tolerated_disk_barrier_failures(
7159 for (i = 0; i < map->num_stripes; i++) {
7160 struct btrfs_device *dev = map->stripes[i].dev;
7162 if (!dev || !dev->bdev ||
7163 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7164 dev->last_flush_error)
7166 else if (failing_dev && failing_dev == dev)
7169 if (missing > max_tolerated) {
7172 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7173 em->start, missing, max_tolerated);
7174 free_extent_map(em);
7178 next_start = extent_map_end(em);
7179 free_extent_map(em);
7181 read_lock(&map_tree->lock);
7182 em = lookup_extent_mapping(map_tree, next_start,
7183 (u64)(-1) - next_start);
7184 read_unlock(&map_tree->lock);
7190 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7192 struct btrfs_root *root = fs_info->chunk_root;
7193 struct btrfs_path *path;
7194 struct extent_buffer *leaf;
7195 struct btrfs_key key;
7196 struct btrfs_key found_key;
7201 path = btrfs_alloc_path();
7206 * uuid_mutex is needed only if we are mounting a sprout FS
7207 * otherwise we don't need it.
7209 mutex_lock(&uuid_mutex);
7210 mutex_lock(&fs_info->chunk_mutex);
7213 * Read all device items, and then all the chunk items. All
7214 * device items are found before any chunk item (their object id
7215 * is smaller than the lowest possible object id for a chunk
7216 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7218 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7221 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7225 leaf = path->nodes[0];
7226 slot = path->slots[0];
7227 if (slot >= btrfs_header_nritems(leaf)) {
7228 ret = btrfs_next_leaf(root, path);
7235 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7236 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7237 struct btrfs_dev_item *dev_item;
7238 dev_item = btrfs_item_ptr(leaf, slot,
7239 struct btrfs_dev_item);
7240 ret = read_one_dev(leaf, dev_item);
7244 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7245 struct btrfs_chunk *chunk;
7246 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7247 ret = read_one_chunk(&found_key, leaf, chunk);
7255 * After loading chunk tree, we've got all device information,
7256 * do another round of validation checks.
7258 if (total_dev != fs_info->fs_devices->total_devices) {
7260 "super_num_devices %llu mismatch with num_devices %llu found here",
7261 btrfs_super_num_devices(fs_info->super_copy),
7266 if (btrfs_super_total_bytes(fs_info->super_copy) <
7267 fs_info->fs_devices->total_rw_bytes) {
7269 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7270 btrfs_super_total_bytes(fs_info->super_copy),
7271 fs_info->fs_devices->total_rw_bytes);
7277 mutex_unlock(&fs_info->chunk_mutex);
7278 mutex_unlock(&uuid_mutex);
7280 btrfs_free_path(path);
7284 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7286 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7287 struct btrfs_device *device;
7289 while (fs_devices) {
7290 mutex_lock(&fs_devices->device_list_mutex);
7291 list_for_each_entry(device, &fs_devices->devices, dev_list)
7292 device->fs_info = fs_info;
7293 mutex_unlock(&fs_devices->device_list_mutex);
7295 fs_devices = fs_devices->seed;
7299 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7303 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7304 btrfs_dev_stat_reset(dev, i);
7307 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7309 struct btrfs_key key;
7310 struct btrfs_key found_key;
7311 struct btrfs_root *dev_root = fs_info->dev_root;
7312 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7313 struct extent_buffer *eb;
7316 struct btrfs_device *device;
7317 struct btrfs_path *path = NULL;
7320 path = btrfs_alloc_path();
7326 mutex_lock(&fs_devices->device_list_mutex);
7327 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7329 struct btrfs_dev_stats_item *ptr;
7331 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7332 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7333 key.offset = device->devid;
7334 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7336 __btrfs_reset_dev_stats(device);
7337 device->dev_stats_valid = 1;
7338 btrfs_release_path(path);
7341 slot = path->slots[0];
7342 eb = path->nodes[0];
7343 btrfs_item_key_to_cpu(eb, &found_key, slot);
7344 item_size = btrfs_item_size_nr(eb, slot);
7346 ptr = btrfs_item_ptr(eb, slot,
7347 struct btrfs_dev_stats_item);
7349 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7350 if (item_size >= (1 + i) * sizeof(__le64))
7351 btrfs_dev_stat_set(device, i,
7352 btrfs_dev_stats_value(eb, ptr, i));
7354 btrfs_dev_stat_reset(device, i);
7357 device->dev_stats_valid = 1;
7358 btrfs_dev_stat_print_on_load(device);
7359 btrfs_release_path(path);
7361 mutex_unlock(&fs_devices->device_list_mutex);
7364 btrfs_free_path(path);
7365 return ret < 0 ? ret : 0;
7368 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7369 struct btrfs_device *device)
7371 struct btrfs_fs_info *fs_info = trans->fs_info;
7372 struct btrfs_root *dev_root = fs_info->dev_root;
7373 struct btrfs_path *path;
7374 struct btrfs_key key;
7375 struct extent_buffer *eb;
7376 struct btrfs_dev_stats_item *ptr;
7380 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7381 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7382 key.offset = device->devid;
7384 path = btrfs_alloc_path();
7387 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7389 btrfs_warn_in_rcu(fs_info,
7390 "error %d while searching for dev_stats item for device %s",
7391 ret, rcu_str_deref(device->name));
7396 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7397 /* need to delete old one and insert a new one */
7398 ret = btrfs_del_item(trans, dev_root, path);
7400 btrfs_warn_in_rcu(fs_info,
7401 "delete too small dev_stats item for device %s failed %d",
7402 rcu_str_deref(device->name), ret);
7409 /* need to insert a new item */
7410 btrfs_release_path(path);
7411 ret = btrfs_insert_empty_item(trans, dev_root, path,
7412 &key, sizeof(*ptr));
7414 btrfs_warn_in_rcu(fs_info,
7415 "insert dev_stats item for device %s failed %d",
7416 rcu_str_deref(device->name), ret);
7421 eb = path->nodes[0];
7422 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7423 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7424 btrfs_set_dev_stats_value(eb, ptr, i,
7425 btrfs_dev_stat_read(device, i));
7426 btrfs_mark_buffer_dirty(eb);
7429 btrfs_free_path(path);
7434 * called from commit_transaction. Writes all changed device stats to disk.
7436 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7438 struct btrfs_fs_info *fs_info = trans->fs_info;
7439 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7440 struct btrfs_device *device;
7444 mutex_lock(&fs_devices->device_list_mutex);
7445 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7446 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7447 if (!device->dev_stats_valid || stats_cnt == 0)
7452 * There is a LOAD-LOAD control dependency between the value of
7453 * dev_stats_ccnt and updating the on-disk values which requires
7454 * reading the in-memory counters. Such control dependencies
7455 * require explicit read memory barriers.
7457 * This memory barriers pairs with smp_mb__before_atomic in
7458 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7459 * barrier implied by atomic_xchg in
7460 * btrfs_dev_stats_read_and_reset
7464 ret = update_dev_stat_item(trans, device);
7466 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7468 mutex_unlock(&fs_devices->device_list_mutex);
7473 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7475 btrfs_dev_stat_inc(dev, index);
7476 btrfs_dev_stat_print_on_error(dev);
7479 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7481 if (!dev->dev_stats_valid)
7483 btrfs_err_rl_in_rcu(dev->fs_info,
7484 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7485 rcu_str_deref(dev->name),
7486 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7487 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7488 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7489 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7490 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7493 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7497 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7498 if (btrfs_dev_stat_read(dev, i) != 0)
7500 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7501 return; /* all values == 0, suppress message */
7503 btrfs_info_in_rcu(dev->fs_info,
7504 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7505 rcu_str_deref(dev->name),
7506 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7507 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7508 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7509 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7510 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7513 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7514 struct btrfs_ioctl_get_dev_stats *stats)
7516 struct btrfs_device *dev;
7517 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7520 mutex_lock(&fs_devices->device_list_mutex);
7521 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7523 mutex_unlock(&fs_devices->device_list_mutex);
7526 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7528 } else if (!dev->dev_stats_valid) {
7529 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7531 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7532 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7533 if (stats->nr_items > i)
7535 btrfs_dev_stat_read_and_reset(dev, i);
7537 btrfs_dev_stat_reset(dev, i);
7540 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7541 if (stats->nr_items > i)
7542 stats->values[i] = btrfs_dev_stat_read(dev, i);
7544 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7545 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7549 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7551 struct buffer_head *bh;
7552 struct btrfs_super_block *disk_super;
7558 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7561 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7564 disk_super = (struct btrfs_super_block *)bh->b_data;
7566 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7567 set_buffer_dirty(bh);
7568 sync_dirty_buffer(bh);
7572 /* Notify udev that device has changed */
7573 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7575 /* Update ctime/mtime for device path for libblkid */
7576 update_dev_time(device_path);
7580 * Update the size and bytes used for each device where it changed. This is
7581 * delayed since we would otherwise get errors while writing out the
7584 * Must be invoked during transaction commit.
7586 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7588 struct btrfs_device *curr, *next;
7590 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7592 if (list_empty(&trans->dev_update_list))
7596 * We don't need the device_list_mutex here. This list is owned by the
7597 * transaction and the transaction must complete before the device is
7600 mutex_lock(&trans->fs_info->chunk_mutex);
7601 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7603 list_del_init(&curr->post_commit_list);
7604 curr->commit_total_bytes = curr->disk_total_bytes;
7605 curr->commit_bytes_used = curr->bytes_used;
7607 mutex_unlock(&trans->fs_info->chunk_mutex);
7610 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7612 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7613 while (fs_devices) {
7614 fs_devices->fs_info = fs_info;
7615 fs_devices = fs_devices->seed;
7619 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7621 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7622 while (fs_devices) {
7623 fs_devices->fs_info = NULL;
7624 fs_devices = fs_devices->seed;
7629 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7631 int btrfs_bg_type_to_factor(u64 flags)
7633 const int index = btrfs_bg_flags_to_raid_index(flags);
7635 return btrfs_raid_array[index].ncopies;
7640 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7641 u64 chunk_offset, u64 devid,
7642 u64 physical_offset, u64 physical_len)
7644 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7645 struct extent_map *em;
7646 struct map_lookup *map;
7647 struct btrfs_device *dev;
7653 read_lock(&em_tree->lock);
7654 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7655 read_unlock(&em_tree->lock);
7659 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7660 physical_offset, devid);
7665 map = em->map_lookup;
7666 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7667 if (physical_len != stripe_len) {
7669 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7670 physical_offset, devid, em->start, physical_len,
7676 for (i = 0; i < map->num_stripes; i++) {
7677 if (map->stripes[i].dev->devid == devid &&
7678 map->stripes[i].physical == physical_offset) {
7680 if (map->verified_stripes >= map->num_stripes) {
7682 "too many dev extents for chunk %llu found",
7687 map->verified_stripes++;
7693 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7694 physical_offset, devid);
7698 /* Make sure no dev extent is beyond device bondary */
7699 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7701 btrfs_err(fs_info, "failed to find devid %llu", devid);
7706 /* It's possible this device is a dummy for seed device */
7707 if (dev->disk_total_bytes == 0) {
7708 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7711 btrfs_err(fs_info, "failed to find seed devid %llu",
7718 if (physical_offset + physical_len > dev->disk_total_bytes) {
7720 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7721 devid, physical_offset, physical_len,
7722 dev->disk_total_bytes);
7727 free_extent_map(em);
7731 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7733 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7734 struct extent_map *em;
7735 struct rb_node *node;
7738 read_lock(&em_tree->lock);
7739 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7740 em = rb_entry(node, struct extent_map, rb_node);
7741 if (em->map_lookup->num_stripes !=
7742 em->map_lookup->verified_stripes) {
7744 "chunk %llu has missing dev extent, have %d expect %d",
7745 em->start, em->map_lookup->verified_stripes,
7746 em->map_lookup->num_stripes);
7752 read_unlock(&em_tree->lock);
7757 * Ensure that all dev extents are mapped to correct chunk, otherwise
7758 * later chunk allocation/free would cause unexpected behavior.
7760 * NOTE: This will iterate through the whole device tree, which should be of
7761 * the same size level as the chunk tree. This slightly increases mount time.
7763 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7765 struct btrfs_path *path;
7766 struct btrfs_root *root = fs_info->dev_root;
7767 struct btrfs_key key;
7769 u64 prev_dev_ext_end = 0;
7773 key.type = BTRFS_DEV_EXTENT_KEY;
7776 path = btrfs_alloc_path();
7780 path->reada = READA_FORWARD;
7781 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7785 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7786 ret = btrfs_next_item(root, path);
7789 /* No dev extents at all? Not good */
7796 struct extent_buffer *leaf = path->nodes[0];
7797 struct btrfs_dev_extent *dext;
7798 int slot = path->slots[0];
7800 u64 physical_offset;
7804 btrfs_item_key_to_cpu(leaf, &key, slot);
7805 if (key.type != BTRFS_DEV_EXTENT_KEY)
7807 devid = key.objectid;
7808 physical_offset = key.offset;
7810 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7811 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7812 physical_len = btrfs_dev_extent_length(leaf, dext);
7814 /* Check if this dev extent overlaps with the previous one */
7815 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7817 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7818 devid, physical_offset, prev_dev_ext_end);
7823 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7824 physical_offset, physical_len);
7828 prev_dev_ext_end = physical_offset + physical_len;
7830 ret = btrfs_next_item(root, path);
7839 /* Ensure all chunks have corresponding dev extents */
7840 ret = verify_chunk_dev_extent_mapping(fs_info);
7842 btrfs_free_path(path);
7847 * Check whether the given block group or device is pinned by any inode being
7848 * used as a swapfile.
7850 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7852 struct btrfs_swapfile_pin *sp;
7853 struct rb_node *node;
7855 spin_lock(&fs_info->swapfile_pins_lock);
7856 node = fs_info->swapfile_pins.rb_node;
7858 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7860 node = node->rb_left;
7861 else if (ptr > sp->ptr)
7862 node = node->rb_right;
7866 spin_unlock(&fs_info->swapfile_pins_lock);
7867 return node != NULL;