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"
32 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
33 [BTRFS_RAID_RAID10] = {
36 .devs_max = 0, /* 0 == as many as possible */
38 .tolerated_failures = 1,
42 .raid_name = "raid10",
43 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
44 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
46 [BTRFS_RAID_RAID1] = {
51 .tolerated_failures = 1,
56 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
57 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
64 .tolerated_failures = 0,
69 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
72 [BTRFS_RAID_RAID0] = {
77 .tolerated_failures = 0,
82 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
85 [BTRFS_RAID_SINGLE] = {
90 .tolerated_failures = 0,
94 .raid_name = "single",
98 [BTRFS_RAID_RAID5] = {
103 .tolerated_failures = 1,
107 .raid_name = "raid5",
108 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
109 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
111 [BTRFS_RAID_RAID6] = {
116 .tolerated_failures = 2,
120 .raid_name = "raid6",
121 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
122 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
126 const char *get_raid_name(enum btrfs_raid_types type)
128 if (type >= BTRFS_NR_RAID_TYPES)
131 return btrfs_raid_array[type].raid_name;
135 * Fill @buf with textual description of @bg_flags, no more than @size_buf
136 * bytes including terminating null byte.
138 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
143 u64 flags = bg_flags;
144 u32 size_bp = size_buf;
151 #define DESCRIBE_FLAG(flag, desc) \
153 if (flags & (flag)) { \
154 ret = snprintf(bp, size_bp, "%s|", (desc)); \
155 if (ret < 0 || ret >= size_bp) \
163 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
164 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
165 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
167 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
168 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
169 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
170 btrfs_raid_array[i].raid_name);
174 ret = snprintf(bp, size_bp, "0x%llx|", flags);
178 if (size_bp < size_buf)
179 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
182 * The text is trimmed, it's up to the caller to provide sufficiently
188 static int init_first_rw_device(struct btrfs_trans_handle *trans,
189 struct btrfs_fs_info *fs_info);
190 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
191 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
192 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
193 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
194 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
195 enum btrfs_map_op op,
196 u64 logical, u64 *length,
197 struct btrfs_bio **bbio_ret,
198 int mirror_num, int need_raid_map);
204 * There are several mutexes that protect manipulation of devices and low-level
205 * structures like chunks but not block groups, extents or files
207 * uuid_mutex (global lock)
208 * ------------------------
209 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
210 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
211 * device) or requested by the device= mount option
213 * the mutex can be very coarse and can cover long-running operations
215 * protects: updates to fs_devices counters like missing devices, rw devices,
216 * seeding, structure cloning, opening/closing devices at mount/umount time
218 * global::fs_devs - add, remove, updates to the global list
220 * does not protect: manipulation of the fs_devices::devices list!
222 * btrfs_device::name - renames (write side), read is RCU
224 * fs_devices::device_list_mutex (per-fs, with RCU)
225 * ------------------------------------------------
226 * protects updates to fs_devices::devices, ie. adding and deleting
228 * simple list traversal with read-only actions can be done with RCU protection
230 * may be used to exclude some operations from running concurrently without any
231 * modifications to the list (see write_all_supers)
235 * protects balance structures (status, state) and context accessed from
236 * several places (internally, ioctl)
240 * protects chunks, adding or removing during allocation, trim or when a new
241 * device is added/removed
245 * a big lock that is held by the cleaner thread and prevents running subvolume
246 * cleaning together with relocation or delayed iputs
259 * Exclusive operations, BTRFS_FS_EXCL_OP
260 * ======================================
262 * Maintains the exclusivity of the following operations that apply to the
263 * whole filesystem and cannot run in parallel.
268 * - Device replace (*)
271 * The device operations (as above) can be in one of the following states:
277 * Only device operations marked with (*) can go into the Paused state for the
280 * - ioctl (only Balance can be Paused through ioctl)
281 * - filesystem remounted as read-only
282 * - filesystem unmounted and mounted as read-only
283 * - system power-cycle and filesystem mounted as read-only
284 * - filesystem or device errors leading to forced read-only
286 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
287 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
288 * A device operation in Paused or Running state can be canceled or resumed
289 * either by ioctl (Balance only) or when remounted as read-write.
290 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
294 DEFINE_MUTEX(uuid_mutex);
295 static LIST_HEAD(fs_uuids);
296 struct list_head *btrfs_get_fs_uuids(void)
302 * alloc_fs_devices - allocate struct btrfs_fs_devices
303 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
304 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
306 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
307 * The returned struct is not linked onto any lists and can be destroyed with
308 * kfree() right away.
310 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
311 const u8 *metadata_fsid)
313 struct btrfs_fs_devices *fs_devs;
315 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
317 return ERR_PTR(-ENOMEM);
319 mutex_init(&fs_devs->device_list_mutex);
321 INIT_LIST_HEAD(&fs_devs->devices);
322 INIT_LIST_HEAD(&fs_devs->alloc_list);
323 INIT_LIST_HEAD(&fs_devs->fs_list);
325 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
328 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
330 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
335 void btrfs_free_device(struct btrfs_device *device)
337 WARN_ON(!list_empty(&device->post_commit_list));
338 rcu_string_free(device->name);
339 extent_io_tree_release(&device->alloc_state);
340 bio_put(device->flush_bio);
344 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
346 struct btrfs_device *device;
347 WARN_ON(fs_devices->opened);
348 while (!list_empty(&fs_devices->devices)) {
349 device = list_entry(fs_devices->devices.next,
350 struct btrfs_device, dev_list);
351 list_del(&device->dev_list);
352 btrfs_free_device(device);
357 static void btrfs_kobject_uevent(struct block_device *bdev,
358 enum kobject_action action)
362 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
364 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
366 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
367 &disk_to_dev(bdev->bd_disk)->kobj);
370 void __exit btrfs_cleanup_fs_uuids(void)
372 struct btrfs_fs_devices *fs_devices;
374 while (!list_empty(&fs_uuids)) {
375 fs_devices = list_entry(fs_uuids.next,
376 struct btrfs_fs_devices, fs_list);
377 list_del(&fs_devices->fs_list);
378 free_fs_devices(fs_devices);
383 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
384 * Returned struct is not linked onto any lists and must be destroyed using
387 static struct btrfs_device *__alloc_device(void)
389 struct btrfs_device *dev;
391 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
393 return ERR_PTR(-ENOMEM);
396 * Preallocate a bio that's always going to be used for flushing device
397 * barriers and matches the device lifespan
399 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
400 if (!dev->flush_bio) {
402 return ERR_PTR(-ENOMEM);
405 INIT_LIST_HEAD(&dev->dev_list);
406 INIT_LIST_HEAD(&dev->dev_alloc_list);
407 INIT_LIST_HEAD(&dev->post_commit_list);
409 spin_lock_init(&dev->io_lock);
411 atomic_set(&dev->reada_in_flight, 0);
412 atomic_set(&dev->dev_stats_ccnt, 0);
413 btrfs_device_data_ordered_init(dev);
414 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
415 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
416 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
421 static noinline struct btrfs_fs_devices *find_fsid(
422 const u8 *fsid, const u8 *metadata_fsid)
424 struct btrfs_fs_devices *fs_devices;
430 * Handle scanned device having completed its fsid change but
431 * belonging to a fs_devices that was created by first scanning
432 * a device which didn't have its fsid/metadata_uuid changed
433 * at all and the CHANGING_FSID_V2 flag set.
435 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
436 if (fs_devices->fsid_change &&
437 memcmp(metadata_fsid, fs_devices->fsid,
438 BTRFS_FSID_SIZE) == 0 &&
439 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
440 BTRFS_FSID_SIZE) == 0) {
445 * Handle scanned device having completed its fsid change but
446 * belonging to a fs_devices that was created by a device that
447 * has an outdated pair of fsid/metadata_uuid and
448 * CHANGING_FSID_V2 flag set.
450 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
451 if (fs_devices->fsid_change &&
452 memcmp(fs_devices->metadata_uuid,
453 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
454 memcmp(metadata_fsid, fs_devices->metadata_uuid,
455 BTRFS_FSID_SIZE) == 0) {
461 /* Handle non-split brain cases */
462 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
464 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
465 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
466 BTRFS_FSID_SIZE) == 0)
469 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
477 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
478 int flush, struct block_device **bdev,
479 struct buffer_head **bh)
483 *bdev = blkdev_get_by_path(device_path, flags, holder);
486 ret = PTR_ERR(*bdev);
491 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
492 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
494 blkdev_put(*bdev, flags);
497 invalidate_bdev(*bdev);
498 *bh = btrfs_read_dev_super(*bdev);
501 blkdev_put(*bdev, flags);
513 static void requeue_list(struct btrfs_pending_bios *pending_bios,
514 struct bio *head, struct bio *tail)
517 struct bio *old_head;
519 old_head = pending_bios->head;
520 pending_bios->head = head;
521 if (pending_bios->tail)
522 tail->bi_next = old_head;
524 pending_bios->tail = tail;
528 * we try to collect pending bios for a device so we don't get a large
529 * number of procs sending bios down to the same device. This greatly
530 * improves the schedulers ability to collect and merge the bios.
532 * But, it also turns into a long list of bios to process and that is sure
533 * to eventually make the worker thread block. The solution here is to
534 * make some progress and then put this work struct back at the end of
535 * the list if the block device is congested. This way, multiple devices
536 * can make progress from a single worker thread.
538 static noinline void run_scheduled_bios(struct btrfs_device *device)
540 struct btrfs_fs_info *fs_info = device->fs_info;
542 struct backing_dev_info *bdi;
543 struct btrfs_pending_bios *pending_bios;
547 unsigned long num_run;
548 unsigned long batch_run = 0;
549 unsigned long last_waited = 0;
551 int sync_pending = 0;
552 struct blk_plug plug;
555 * this function runs all the bios we've collected for
556 * a particular device. We don't want to wander off to
557 * another device without first sending all of these down.
558 * So, setup a plug here and finish it off before we return
560 blk_start_plug(&plug);
562 bdi = device->bdev->bd_bdi;
565 spin_lock(&device->io_lock);
570 /* take all the bios off the list at once and process them
571 * later on (without the lock held). But, remember the
572 * tail and other pointers so the bios can be properly reinserted
573 * into the list if we hit congestion
575 if (!force_reg && device->pending_sync_bios.head) {
576 pending_bios = &device->pending_sync_bios;
579 pending_bios = &device->pending_bios;
583 pending = pending_bios->head;
584 tail = pending_bios->tail;
585 WARN_ON(pending && !tail);
588 * if pending was null this time around, no bios need processing
589 * at all and we can stop. Otherwise it'll loop back up again
590 * and do an additional check so no bios are missed.
592 * device->running_pending is used to synchronize with the
595 if (device->pending_sync_bios.head == NULL &&
596 device->pending_bios.head == NULL) {
598 device->running_pending = 0;
601 device->running_pending = 1;
604 pending_bios->head = NULL;
605 pending_bios->tail = NULL;
607 spin_unlock(&device->io_lock);
612 /* we want to work on both lists, but do more bios on the
613 * sync list than the regular list
616 pending_bios != &device->pending_sync_bios &&
617 device->pending_sync_bios.head) ||
618 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
619 device->pending_bios.head)) {
620 spin_lock(&device->io_lock);
621 requeue_list(pending_bios, pending, tail);
626 pending = pending->bi_next;
629 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
632 * if we're doing the sync list, record that our
633 * plug has some sync requests on it
635 * If we're doing the regular list and there are
636 * sync requests sitting around, unplug before
639 if (pending_bios == &device->pending_sync_bios) {
641 } else if (sync_pending) {
642 blk_finish_plug(&plug);
643 blk_start_plug(&plug);
647 btrfsic_submit_bio(cur);
654 * we made progress, there is more work to do and the bdi
655 * is now congested. Back off and let other work structs
658 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
659 fs_info->fs_devices->open_devices > 1) {
660 struct io_context *ioc;
662 ioc = current->io_context;
665 * the main goal here is that we don't want to
666 * block if we're going to be able to submit
667 * more requests without blocking.
669 * This code does two great things, it pokes into
670 * the elevator code from a filesystem _and_
671 * it makes assumptions about how batching works.
673 if (ioc && ioc->nr_batch_requests > 0 &&
674 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
676 ioc->last_waited == last_waited)) {
678 * we want to go through our batch of
679 * requests and stop. So, we copy out
680 * the ioc->last_waited time and test
681 * against it before looping
683 last_waited = ioc->last_waited;
687 spin_lock(&device->io_lock);
688 requeue_list(pending_bios, pending, tail);
689 device->running_pending = 1;
691 spin_unlock(&device->io_lock);
692 btrfs_queue_work(fs_info->submit_workers,
702 spin_lock(&device->io_lock);
703 if (device->pending_bios.head || device->pending_sync_bios.head)
705 spin_unlock(&device->io_lock);
708 blk_finish_plug(&plug);
711 static void pending_bios_fn(struct btrfs_work *work)
713 struct btrfs_device *device;
715 device = container_of(work, struct btrfs_device, work);
716 run_scheduled_bios(device);
719 static bool device_path_matched(const char *path, struct btrfs_device *device)
724 found = strcmp(rcu_str_deref(device->name), path);
731 * Search and remove all stale (devices which are not mounted) devices.
732 * When both inputs are NULL, it will search and release all stale devices.
733 * path: Optional. When provided will it release all unmounted devices
734 * matching this path only.
735 * skip_dev: Optional. Will skip this device when searching for the stale
737 * Return: 0 for success or if @path is NULL.
738 * -EBUSY if @path is a mounted device.
739 * -ENOENT if @path does not match any device in the list.
741 static int btrfs_free_stale_devices(const char *path,
742 struct btrfs_device *skip_device)
744 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
745 struct btrfs_device *device, *tmp_device;
751 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
753 mutex_lock(&fs_devices->device_list_mutex);
754 list_for_each_entry_safe(device, tmp_device,
755 &fs_devices->devices, dev_list) {
756 if (skip_device && skip_device == device)
758 if (path && !device->name)
760 if (path && !device_path_matched(path, device))
762 if (fs_devices->opened) {
763 /* for an already deleted device return 0 */
764 if (path && ret != 0)
769 /* delete the stale device */
770 fs_devices->num_devices--;
771 list_del(&device->dev_list);
772 btrfs_free_device(device);
775 if (fs_devices->num_devices == 0)
778 mutex_unlock(&fs_devices->device_list_mutex);
780 if (fs_devices->num_devices == 0) {
781 btrfs_sysfs_remove_fsid(fs_devices);
782 list_del(&fs_devices->fs_list);
783 free_fs_devices(fs_devices);
790 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
791 struct btrfs_device *device, fmode_t flags,
794 struct request_queue *q;
795 struct block_device *bdev;
796 struct buffer_head *bh;
797 struct btrfs_super_block *disk_super;
806 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
811 disk_super = (struct btrfs_super_block *)bh->b_data;
812 devid = btrfs_stack_device_id(&disk_super->dev_item);
813 if (devid != device->devid)
816 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
819 device->generation = btrfs_super_generation(disk_super);
821 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
822 if (btrfs_super_incompat_flags(disk_super) &
823 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
825 "BTRFS: Invalid seeding and uuid-changed device detected\n");
829 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
830 fs_devices->seeding = 1;
832 if (bdev_read_only(bdev))
833 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
835 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
838 q = bdev_get_queue(bdev);
839 if (!blk_queue_nonrot(q))
840 fs_devices->rotating = 1;
843 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
844 device->mode = flags;
846 fs_devices->open_devices++;
847 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
848 device->devid != BTRFS_DEV_REPLACE_DEVID) {
849 fs_devices->rw_devices++;
850 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
858 blkdev_put(bdev, flags);
864 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
865 * being created with a disk that has already completed its fsid change.
867 static struct btrfs_fs_devices *find_fsid_inprogress(
868 struct btrfs_super_block *disk_super)
870 struct btrfs_fs_devices *fs_devices;
872 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
873 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
874 BTRFS_FSID_SIZE) != 0 &&
875 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
876 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
885 static struct btrfs_fs_devices *find_fsid_changed(
886 struct btrfs_super_block *disk_super)
888 struct btrfs_fs_devices *fs_devices;
891 * Handles the case where scanned device is part of an fs that had
892 * multiple successful changes of FSID but curently device didn't
893 * observe it. Meaning our fsid will be different than theirs.
895 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
896 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
897 BTRFS_FSID_SIZE) != 0 &&
898 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
899 BTRFS_FSID_SIZE) == 0 &&
900 memcmp(fs_devices->fsid, disk_super->fsid,
901 BTRFS_FSID_SIZE) != 0) {
909 * Add new device to list of registered devices
912 * device pointer which was just added or updated when successful
913 * error pointer when failed
915 static noinline struct btrfs_device *device_list_add(const char *path,
916 struct btrfs_super_block *disk_super,
917 bool *new_device_added)
919 struct btrfs_device *device;
920 struct btrfs_fs_devices *fs_devices = NULL;
921 struct rcu_string *name;
922 u64 found_transid = btrfs_super_generation(disk_super);
923 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
924 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
925 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
926 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
927 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
929 if (fsid_change_in_progress) {
930 if (!has_metadata_uuid) {
932 * When we have an image which has CHANGING_FSID_V2 set
933 * it might belong to either a filesystem which has
934 * disks with completed fsid change or it might belong
935 * to fs with no UUID changes in effect, handle both.
937 fs_devices = find_fsid_inprogress(disk_super);
939 fs_devices = find_fsid(disk_super->fsid, NULL);
941 fs_devices = find_fsid_changed(disk_super);
943 } else if (has_metadata_uuid) {
944 fs_devices = find_fsid(disk_super->fsid,
945 disk_super->metadata_uuid);
947 fs_devices = find_fsid(disk_super->fsid, NULL);
952 if (has_metadata_uuid)
953 fs_devices = alloc_fs_devices(disk_super->fsid,
954 disk_super->metadata_uuid);
956 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
958 if (IS_ERR(fs_devices))
959 return ERR_CAST(fs_devices);
961 fs_devices->fsid_change = fsid_change_in_progress;
963 mutex_lock(&fs_devices->device_list_mutex);
964 list_add(&fs_devices->fs_list, &fs_uuids);
968 mutex_lock(&fs_devices->device_list_mutex);
969 device = btrfs_find_device(fs_devices, devid,
970 disk_super->dev_item.uuid, NULL, false);
973 * If this disk has been pulled into an fs devices created by
974 * a device which had the CHANGING_FSID_V2 flag then replace the
975 * metadata_uuid/fsid values of the fs_devices.
977 if (has_metadata_uuid && fs_devices->fsid_change &&
978 found_transid > fs_devices->latest_generation) {
979 memcpy(fs_devices->fsid, disk_super->fsid,
981 memcpy(fs_devices->metadata_uuid,
982 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
984 fs_devices->fsid_change = false;
989 if (fs_devices->opened) {
990 mutex_unlock(&fs_devices->device_list_mutex);
991 return ERR_PTR(-EBUSY);
994 device = btrfs_alloc_device(NULL, &devid,
995 disk_super->dev_item.uuid);
996 if (IS_ERR(device)) {
997 mutex_unlock(&fs_devices->device_list_mutex);
998 /* we can safely leave the fs_devices entry around */
1002 name = rcu_string_strdup(path, GFP_NOFS);
1004 btrfs_free_device(device);
1005 mutex_unlock(&fs_devices->device_list_mutex);
1006 return ERR_PTR(-ENOMEM);
1008 rcu_assign_pointer(device->name, name);
1010 list_add_rcu(&device->dev_list, &fs_devices->devices);
1011 fs_devices->num_devices++;
1013 device->fs_devices = fs_devices;
1014 *new_device_added = true;
1016 if (disk_super->label[0])
1017 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1018 disk_super->label, devid, found_transid, path);
1020 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1021 disk_super->fsid, devid, found_transid, path);
1023 } else if (!device->name || strcmp(device->name->str, path)) {
1025 * When FS is already mounted.
1026 * 1. If you are here and if the device->name is NULL that
1027 * means this device was missing at time of FS mount.
1028 * 2. If you are here and if the device->name is different
1029 * from 'path' that means either
1030 * a. The same device disappeared and reappeared with
1031 * different name. or
1032 * b. The missing-disk-which-was-replaced, has
1035 * We must allow 1 and 2a above. But 2b would be a spurious
1036 * and unintentional.
1038 * Further in case of 1 and 2a above, the disk at 'path'
1039 * would have missed some transaction when it was away and
1040 * in case of 2a the stale bdev has to be updated as well.
1041 * 2b must not be allowed at all time.
1045 * For now, we do allow update to btrfs_fs_device through the
1046 * btrfs dev scan cli after FS has been mounted. We're still
1047 * tracking a problem where systems fail mount by subvolume id
1048 * when we reject replacement on a mounted FS.
1050 if (!fs_devices->opened && found_transid < device->generation) {
1052 * That is if the FS is _not_ mounted and if you
1053 * are here, that means there is more than one
1054 * disk with same uuid and devid.We keep the one
1055 * with larger generation number or the last-in if
1056 * generation are equal.
1058 mutex_unlock(&fs_devices->device_list_mutex);
1059 return ERR_PTR(-EEXIST);
1063 * We are going to replace the device path for a given devid,
1064 * make sure it's the same device if the device is mounted
1067 struct block_device *path_bdev;
1069 path_bdev = lookup_bdev(path);
1070 if (IS_ERR(path_bdev)) {
1071 mutex_unlock(&fs_devices->device_list_mutex);
1072 return ERR_CAST(path_bdev);
1075 if (device->bdev != path_bdev) {
1077 mutex_unlock(&fs_devices->device_list_mutex);
1078 btrfs_warn_in_rcu(device->fs_info,
1079 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1080 disk_super->fsid, devid,
1081 rcu_str_deref(device->name), path);
1082 return ERR_PTR(-EEXIST);
1085 btrfs_info_in_rcu(device->fs_info,
1086 "device fsid %pU devid %llu moved old:%s new:%s",
1087 disk_super->fsid, devid,
1088 rcu_str_deref(device->name), path);
1091 name = rcu_string_strdup(path, GFP_NOFS);
1093 mutex_unlock(&fs_devices->device_list_mutex);
1094 return ERR_PTR(-ENOMEM);
1096 rcu_string_free(device->name);
1097 rcu_assign_pointer(device->name, name);
1098 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1099 fs_devices->missing_devices--;
1100 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1105 * Unmount does not free the btrfs_device struct but would zero
1106 * generation along with most of the other members. So just update
1107 * it back. We need it to pick the disk with largest generation
1110 if (!fs_devices->opened) {
1111 device->generation = found_transid;
1112 fs_devices->latest_generation = max_t(u64, found_transid,
1113 fs_devices->latest_generation);
1116 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1118 mutex_unlock(&fs_devices->device_list_mutex);
1122 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1124 struct btrfs_fs_devices *fs_devices;
1125 struct btrfs_device *device;
1126 struct btrfs_device *orig_dev;
1128 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1129 if (IS_ERR(fs_devices))
1132 mutex_lock(&orig->device_list_mutex);
1133 fs_devices->total_devices = orig->total_devices;
1135 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1136 struct rcu_string *name;
1138 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1144 * This is ok to do without rcu read locked because we hold the
1145 * uuid mutex so nothing we touch in here is going to disappear.
1147 if (orig_dev->name) {
1148 name = rcu_string_strdup(orig_dev->name->str,
1151 btrfs_free_device(device);
1154 rcu_assign_pointer(device->name, name);
1157 list_add(&device->dev_list, &fs_devices->devices);
1158 device->fs_devices = fs_devices;
1159 fs_devices->num_devices++;
1161 mutex_unlock(&orig->device_list_mutex);
1164 mutex_unlock(&orig->device_list_mutex);
1165 free_fs_devices(fs_devices);
1166 return ERR_PTR(-ENOMEM);
1170 * After we have read the system tree and know devids belonging to
1171 * this filesystem, remove the device which does not belong there.
1173 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1175 struct btrfs_device *device, *next;
1176 struct btrfs_device *latest_dev = NULL;
1178 mutex_lock(&uuid_mutex);
1180 /* This is the initialized path, it is safe to release the devices. */
1181 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1182 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1183 &device->dev_state)) {
1184 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1185 &device->dev_state) &&
1187 device->generation > latest_dev->generation)) {
1188 latest_dev = device;
1193 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1195 * In the first step, keep the device which has
1196 * the correct fsid and the devid that is used
1197 * for the dev_replace procedure.
1198 * In the second step, the dev_replace state is
1199 * read from the device tree and it is known
1200 * whether the procedure is really active or
1201 * not, which means whether this device is
1202 * used or whether it should be removed.
1204 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1205 &device->dev_state)) {
1210 blkdev_put(device->bdev, device->mode);
1211 device->bdev = NULL;
1212 fs_devices->open_devices--;
1214 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1215 list_del_init(&device->dev_alloc_list);
1216 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1217 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1218 &device->dev_state))
1219 fs_devices->rw_devices--;
1221 list_del_init(&device->dev_list);
1222 fs_devices->num_devices--;
1223 btrfs_free_device(device);
1226 if (fs_devices->seed) {
1227 fs_devices = fs_devices->seed;
1231 fs_devices->latest_bdev = latest_dev->bdev;
1233 mutex_unlock(&uuid_mutex);
1236 static void btrfs_close_bdev(struct btrfs_device *device)
1241 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1242 sync_blockdev(device->bdev);
1243 invalidate_bdev(device->bdev);
1246 blkdev_put(device->bdev, device->mode);
1249 static void btrfs_close_one_device(struct btrfs_device *device)
1251 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1252 struct btrfs_device *new_device;
1253 struct rcu_string *name;
1256 fs_devices->open_devices--;
1258 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1259 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1260 list_del_init(&device->dev_alloc_list);
1261 fs_devices->rw_devices--;
1264 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1265 fs_devices->missing_devices--;
1267 btrfs_close_bdev(device);
1269 new_device = btrfs_alloc_device(NULL, &device->devid,
1271 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1273 /* Safe because we are under uuid_mutex */
1275 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1276 BUG_ON(!name); /* -ENOMEM */
1277 rcu_assign_pointer(new_device->name, name);
1280 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1281 new_device->fs_devices = device->fs_devices;
1284 btrfs_free_device(device);
1287 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1289 struct btrfs_device *device, *tmp;
1291 if (--fs_devices->opened > 0)
1294 mutex_lock(&fs_devices->device_list_mutex);
1295 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1296 btrfs_close_one_device(device);
1298 mutex_unlock(&fs_devices->device_list_mutex);
1300 WARN_ON(fs_devices->open_devices);
1301 WARN_ON(fs_devices->rw_devices);
1302 fs_devices->opened = 0;
1303 fs_devices->seeding = 0;
1308 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1310 struct btrfs_fs_devices *seed_devices = NULL;
1313 mutex_lock(&uuid_mutex);
1314 ret = close_fs_devices(fs_devices);
1315 if (!fs_devices->opened) {
1316 seed_devices = fs_devices->seed;
1317 fs_devices->seed = NULL;
1319 mutex_unlock(&uuid_mutex);
1321 while (seed_devices) {
1322 fs_devices = seed_devices;
1323 seed_devices = fs_devices->seed;
1324 close_fs_devices(fs_devices);
1325 free_fs_devices(fs_devices);
1330 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1331 fmode_t flags, void *holder)
1333 struct btrfs_device *device;
1334 struct btrfs_device *latest_dev = NULL;
1337 flags |= FMODE_EXCL;
1339 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1340 /* Just open everything we can; ignore failures here */
1341 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1345 device->generation > latest_dev->generation)
1346 latest_dev = device;
1348 if (fs_devices->open_devices == 0) {
1352 fs_devices->opened = 1;
1353 fs_devices->latest_bdev = latest_dev->bdev;
1354 fs_devices->total_rw_bytes = 0;
1359 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1361 struct btrfs_device *dev1, *dev2;
1363 dev1 = list_entry(a, struct btrfs_device, dev_list);
1364 dev2 = list_entry(b, struct btrfs_device, dev_list);
1366 if (dev1->devid < dev2->devid)
1368 else if (dev1->devid > dev2->devid)
1373 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1374 fmode_t flags, void *holder)
1378 lockdep_assert_held(&uuid_mutex);
1380 mutex_lock(&fs_devices->device_list_mutex);
1381 if (fs_devices->opened) {
1382 fs_devices->opened++;
1385 list_sort(NULL, &fs_devices->devices, devid_cmp);
1386 ret = open_fs_devices(fs_devices, flags, holder);
1388 mutex_unlock(&fs_devices->device_list_mutex);
1393 static void btrfs_release_disk_super(struct page *page)
1399 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1401 struct btrfs_super_block **disk_super)
1406 /* make sure our super fits in the device */
1407 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1410 /* make sure our super fits in the page */
1411 if (sizeof(**disk_super) > PAGE_SIZE)
1414 /* make sure our super doesn't straddle pages on disk */
1415 index = bytenr >> PAGE_SHIFT;
1416 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1419 /* pull in the page with our super */
1420 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1423 if (IS_ERR_OR_NULL(*page))
1428 /* align our pointer to the offset of the super block */
1429 *disk_super = p + offset_in_page(bytenr);
1431 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1432 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1433 btrfs_release_disk_super(*page);
1437 if ((*disk_super)->label[0] &&
1438 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1439 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1444 int btrfs_forget_devices(const char *path)
1448 mutex_lock(&uuid_mutex);
1449 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1450 mutex_unlock(&uuid_mutex);
1456 * Look for a btrfs signature on a device. This may be called out of the mount path
1457 * and we are not allowed to call set_blocksize during the scan. The superblock
1458 * is read via pagecache
1460 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1463 struct btrfs_super_block *disk_super;
1464 bool new_device_added = false;
1465 struct btrfs_device *device = NULL;
1466 struct block_device *bdev;
1470 lockdep_assert_held(&uuid_mutex);
1473 * we would like to check all the supers, but that would make
1474 * a btrfs mount succeed after a mkfs from a different FS.
1475 * So, we need to add a special mount option to scan for
1476 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1478 bytenr = btrfs_sb_offset(0);
1479 flags |= FMODE_EXCL;
1481 bdev = blkdev_get_by_path(path, flags, holder);
1483 return ERR_CAST(bdev);
1485 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1486 device = ERR_PTR(-EINVAL);
1487 goto error_bdev_put;
1490 device = device_list_add(path, disk_super, &new_device_added);
1491 if (!IS_ERR(device)) {
1492 if (new_device_added)
1493 btrfs_free_stale_devices(path, device);
1496 btrfs_release_disk_super(page);
1499 blkdev_put(bdev, flags);
1505 * Try to find a chunk that intersects [start, start + len] range and when one
1506 * such is found, record the end of it in *start
1508 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1511 u64 physical_start, physical_end;
1513 lockdep_assert_held(&device->fs_info->chunk_mutex);
1515 if (!find_first_extent_bit(&device->alloc_state, *start,
1516 &physical_start, &physical_end,
1517 CHUNK_ALLOCATED, NULL)) {
1519 if (in_range(physical_start, *start, len) ||
1520 in_range(*start, physical_start,
1521 physical_end - physical_start)) {
1522 *start = physical_end + 1;
1531 * find_free_dev_extent_start - find free space in the specified device
1532 * @device: the device which we search the free space in
1533 * @num_bytes: the size of the free space that we need
1534 * @search_start: the position from which to begin the search
1535 * @start: store the start of the free space.
1536 * @len: the size of the free space. that we find, or the size
1537 * of the max free space if we don't find suitable free space
1539 * this uses a pretty simple search, the expectation is that it is
1540 * called very infrequently and that a given device has a small number
1543 * @start is used to store the start of the free space if we find. But if we
1544 * don't find suitable free space, it will be used to store the start position
1545 * of the max free space.
1547 * @len is used to store the size of the free space that we find.
1548 * But if we don't find suitable free space, it is used to store the size of
1549 * the max free space.
1551 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
1552 u64 search_start, u64 *start, u64 *len)
1554 struct btrfs_fs_info *fs_info = device->fs_info;
1555 struct btrfs_root *root = fs_info->dev_root;
1556 struct btrfs_key key;
1557 struct btrfs_dev_extent *dev_extent;
1558 struct btrfs_path *path;
1563 u64 search_end = device->total_bytes;
1566 struct extent_buffer *l;
1569 * We don't want to overwrite the superblock on the drive nor any area
1570 * used by the boot loader (grub for example), so we make sure to start
1571 * at an offset of at least 1MB.
1573 search_start = max_t(u64, search_start, SZ_1M);
1575 path = btrfs_alloc_path();
1579 max_hole_start = search_start;
1583 if (search_start >= search_end ||
1584 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1589 path->reada = READA_FORWARD;
1590 path->search_commit_root = 1;
1591 path->skip_locking = 1;
1593 key.objectid = device->devid;
1594 key.offset = search_start;
1595 key.type = BTRFS_DEV_EXTENT_KEY;
1597 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1601 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1608 slot = path->slots[0];
1609 if (slot >= btrfs_header_nritems(l)) {
1610 ret = btrfs_next_leaf(root, path);
1618 btrfs_item_key_to_cpu(l, &key, slot);
1620 if (key.objectid < device->devid)
1623 if (key.objectid > device->devid)
1626 if (key.type != BTRFS_DEV_EXTENT_KEY)
1629 if (key.offset > search_start) {
1630 hole_size = key.offset - search_start;
1633 * Have to check before we set max_hole_start, otherwise
1634 * we could end up sending back this offset anyway.
1636 if (contains_pending_extent(device, &search_start,
1638 if (key.offset >= search_start)
1639 hole_size = key.offset - search_start;
1644 if (hole_size > max_hole_size) {
1645 max_hole_start = search_start;
1646 max_hole_size = hole_size;
1650 * If this free space is greater than which we need,
1651 * it must be the max free space that we have found
1652 * until now, so max_hole_start must point to the start
1653 * of this free space and the length of this free space
1654 * is stored in max_hole_size. Thus, we return
1655 * max_hole_start and max_hole_size and go back to the
1658 if (hole_size >= num_bytes) {
1664 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1665 extent_end = key.offset + btrfs_dev_extent_length(l,
1667 if (extent_end > search_start)
1668 search_start = extent_end;
1675 * At this point, search_start should be the end of
1676 * allocated dev extents, and when shrinking the device,
1677 * search_end may be smaller than search_start.
1679 if (search_end > search_start) {
1680 hole_size = search_end - search_start;
1682 if (contains_pending_extent(device, &search_start, hole_size)) {
1683 btrfs_release_path(path);
1687 if (hole_size > max_hole_size) {
1688 max_hole_start = search_start;
1689 max_hole_size = hole_size;
1694 if (max_hole_size < num_bytes)
1700 btrfs_free_path(path);
1701 *start = max_hole_start;
1703 *len = max_hole_size;
1707 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1708 u64 *start, u64 *len)
1710 /* FIXME use last free of some kind */
1711 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1714 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1715 struct btrfs_device *device,
1716 u64 start, u64 *dev_extent_len)
1718 struct btrfs_fs_info *fs_info = device->fs_info;
1719 struct btrfs_root *root = fs_info->dev_root;
1721 struct btrfs_path *path;
1722 struct btrfs_key key;
1723 struct btrfs_key found_key;
1724 struct extent_buffer *leaf = NULL;
1725 struct btrfs_dev_extent *extent = NULL;
1727 path = btrfs_alloc_path();
1731 key.objectid = device->devid;
1733 key.type = BTRFS_DEV_EXTENT_KEY;
1735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1737 ret = btrfs_previous_item(root, path, key.objectid,
1738 BTRFS_DEV_EXTENT_KEY);
1741 leaf = path->nodes[0];
1742 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1743 extent = btrfs_item_ptr(leaf, path->slots[0],
1744 struct btrfs_dev_extent);
1745 BUG_ON(found_key.offset > start || found_key.offset +
1746 btrfs_dev_extent_length(leaf, extent) < start);
1748 btrfs_release_path(path);
1750 } else if (ret == 0) {
1751 leaf = path->nodes[0];
1752 extent = btrfs_item_ptr(leaf, path->slots[0],
1753 struct btrfs_dev_extent);
1755 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1759 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1761 ret = btrfs_del_item(trans, root, path);
1763 btrfs_handle_fs_error(fs_info, ret,
1764 "Failed to remove dev extent item");
1766 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1769 btrfs_free_path(path);
1773 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1774 struct btrfs_device *device,
1775 u64 chunk_offset, u64 start, u64 num_bytes)
1778 struct btrfs_path *path;
1779 struct btrfs_fs_info *fs_info = device->fs_info;
1780 struct btrfs_root *root = fs_info->dev_root;
1781 struct btrfs_dev_extent *extent;
1782 struct extent_buffer *leaf;
1783 struct btrfs_key key;
1785 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1786 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1787 path = btrfs_alloc_path();
1791 key.objectid = device->devid;
1793 key.type = BTRFS_DEV_EXTENT_KEY;
1794 ret = btrfs_insert_empty_item(trans, root, path, &key,
1799 leaf = path->nodes[0];
1800 extent = btrfs_item_ptr(leaf, path->slots[0],
1801 struct btrfs_dev_extent);
1802 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1803 BTRFS_CHUNK_TREE_OBJECTID);
1804 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1805 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1806 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1808 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1809 btrfs_mark_buffer_dirty(leaf);
1811 btrfs_free_path(path);
1815 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1817 struct extent_map_tree *em_tree;
1818 struct extent_map *em;
1822 em_tree = &fs_info->mapping_tree.map_tree;
1823 read_lock(&em_tree->lock);
1824 n = rb_last(&em_tree->map.rb_root);
1826 em = rb_entry(n, struct extent_map, rb_node);
1827 ret = em->start + em->len;
1829 read_unlock(&em_tree->lock);
1834 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1838 struct btrfs_key key;
1839 struct btrfs_key found_key;
1840 struct btrfs_path *path;
1842 path = btrfs_alloc_path();
1846 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1847 key.type = BTRFS_DEV_ITEM_KEY;
1848 key.offset = (u64)-1;
1850 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1854 BUG_ON(ret == 0); /* Corruption */
1856 ret = btrfs_previous_item(fs_info->chunk_root, path,
1857 BTRFS_DEV_ITEMS_OBJECTID,
1858 BTRFS_DEV_ITEM_KEY);
1862 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1864 *devid_ret = found_key.offset + 1;
1868 btrfs_free_path(path);
1873 * the device information is stored in the chunk root
1874 * the btrfs_device struct should be fully filled in
1876 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1877 struct btrfs_device *device)
1880 struct btrfs_path *path;
1881 struct btrfs_dev_item *dev_item;
1882 struct extent_buffer *leaf;
1883 struct btrfs_key key;
1886 path = btrfs_alloc_path();
1890 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1891 key.type = BTRFS_DEV_ITEM_KEY;
1892 key.offset = device->devid;
1894 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1895 &key, sizeof(*dev_item));
1899 leaf = path->nodes[0];
1900 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1902 btrfs_set_device_id(leaf, dev_item, device->devid);
1903 btrfs_set_device_generation(leaf, dev_item, 0);
1904 btrfs_set_device_type(leaf, dev_item, device->type);
1905 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1906 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1907 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1908 btrfs_set_device_total_bytes(leaf, dev_item,
1909 btrfs_device_get_disk_total_bytes(device));
1910 btrfs_set_device_bytes_used(leaf, dev_item,
1911 btrfs_device_get_bytes_used(device));
1912 btrfs_set_device_group(leaf, dev_item, 0);
1913 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1914 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1915 btrfs_set_device_start_offset(leaf, dev_item, 0);
1917 ptr = btrfs_device_uuid(dev_item);
1918 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1919 ptr = btrfs_device_fsid(dev_item);
1920 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1921 ptr, BTRFS_FSID_SIZE);
1922 btrfs_mark_buffer_dirty(leaf);
1926 btrfs_free_path(path);
1931 * Function to update ctime/mtime for a given device path.
1932 * Mainly used for ctime/mtime based probe like libblkid.
1934 static void update_dev_time(const char *path_name)
1938 filp = filp_open(path_name, O_RDWR, 0);
1941 file_update_time(filp);
1942 filp_close(filp, NULL);
1945 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1946 struct btrfs_device *device)
1948 struct btrfs_root *root = fs_info->chunk_root;
1950 struct btrfs_path *path;
1951 struct btrfs_key key;
1952 struct btrfs_trans_handle *trans;
1954 path = btrfs_alloc_path();
1958 trans = btrfs_start_transaction(root, 0);
1959 if (IS_ERR(trans)) {
1960 btrfs_free_path(path);
1961 return PTR_ERR(trans);
1963 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1964 key.type = BTRFS_DEV_ITEM_KEY;
1965 key.offset = device->devid;
1967 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1971 btrfs_abort_transaction(trans, ret);
1972 btrfs_end_transaction(trans);
1976 ret = btrfs_del_item(trans, root, path);
1978 btrfs_abort_transaction(trans, ret);
1979 btrfs_end_transaction(trans);
1983 btrfs_free_path(path);
1985 ret = btrfs_commit_transaction(trans);
1990 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1991 * filesystem. It's up to the caller to adjust that number regarding eg. device
1994 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2002 seq = read_seqbegin(&fs_info->profiles_lock);
2004 all_avail = fs_info->avail_data_alloc_bits |
2005 fs_info->avail_system_alloc_bits |
2006 fs_info->avail_metadata_alloc_bits;
2007 } while (read_seqretry(&fs_info->profiles_lock, seq));
2009 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2010 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2013 if (num_devices < btrfs_raid_array[i].devs_min) {
2014 int ret = btrfs_raid_array[i].mindev_error;
2024 static struct btrfs_device * btrfs_find_next_active_device(
2025 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2027 struct btrfs_device *next_device;
2029 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2030 if (next_device != device &&
2031 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2032 && next_device->bdev)
2040 * Helper function to check if the given device is part of s_bdev / latest_bdev
2041 * and replace it with the provided or the next active device, in the context
2042 * where this function called, there should be always be another device (or
2043 * this_dev) which is active.
2045 void btrfs_assign_next_active_device(struct btrfs_device *device,
2046 struct btrfs_device *this_dev)
2048 struct btrfs_fs_info *fs_info = device->fs_info;
2049 struct btrfs_device *next_device;
2052 next_device = this_dev;
2054 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2056 ASSERT(next_device);
2058 if (fs_info->sb->s_bdev &&
2059 (fs_info->sb->s_bdev == device->bdev))
2060 fs_info->sb->s_bdev = next_device->bdev;
2062 if (fs_info->fs_devices->latest_bdev == device->bdev)
2063 fs_info->fs_devices->latest_bdev = next_device->bdev;
2067 * Return btrfs_fs_devices::num_devices excluding the device that's being
2068 * currently replaced.
2070 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2072 u64 num_devices = fs_info->fs_devices->num_devices;
2074 down_read(&fs_info->dev_replace.rwsem);
2075 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2076 ASSERT(num_devices > 1);
2079 up_read(&fs_info->dev_replace.rwsem);
2084 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2087 struct btrfs_device *device;
2088 struct btrfs_fs_devices *cur_devices;
2089 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2093 mutex_lock(&uuid_mutex);
2095 num_devices = btrfs_num_devices(fs_info);
2097 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2101 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2103 if (IS_ERR(device)) {
2104 if (PTR_ERR(device) == -ENOENT &&
2105 strcmp(device_path, "missing") == 0)
2106 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2108 ret = PTR_ERR(device);
2112 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2113 btrfs_warn_in_rcu(fs_info,
2114 "cannot remove device %s (devid %llu) due to active swapfile",
2115 rcu_str_deref(device->name), device->devid);
2120 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2121 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2125 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2126 fs_info->fs_devices->rw_devices == 1) {
2127 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2131 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2132 mutex_lock(&fs_info->chunk_mutex);
2133 list_del_init(&device->dev_alloc_list);
2134 device->fs_devices->rw_devices--;
2135 mutex_unlock(&fs_info->chunk_mutex);
2138 mutex_unlock(&uuid_mutex);
2139 ret = btrfs_shrink_device(device, 0);
2140 mutex_lock(&uuid_mutex);
2145 * TODO: the superblock still includes this device in its num_devices
2146 * counter although write_all_supers() is not locked out. This
2147 * could give a filesystem state which requires a degraded mount.
2149 ret = btrfs_rm_dev_item(fs_info, device);
2153 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2154 btrfs_scrub_cancel_dev(fs_info, device);
2157 * the device list mutex makes sure that we don't change
2158 * the device list while someone else is writing out all
2159 * the device supers. Whoever is writing all supers, should
2160 * lock the device list mutex before getting the number of
2161 * devices in the super block (super_copy). Conversely,
2162 * whoever updates the number of devices in the super block
2163 * (super_copy) should hold the device list mutex.
2167 * In normal cases the cur_devices == fs_devices. But in case
2168 * of deleting a seed device, the cur_devices should point to
2169 * its own fs_devices listed under the fs_devices->seed.
2171 cur_devices = device->fs_devices;
2172 mutex_lock(&fs_devices->device_list_mutex);
2173 list_del_rcu(&device->dev_list);
2175 cur_devices->num_devices--;
2176 cur_devices->total_devices--;
2177 /* Update total_devices of the parent fs_devices if it's seed */
2178 if (cur_devices != fs_devices)
2179 fs_devices->total_devices--;
2181 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2182 cur_devices->missing_devices--;
2184 btrfs_assign_next_active_device(device, NULL);
2187 cur_devices->open_devices--;
2188 /* remove sysfs entry */
2189 btrfs_sysfs_rm_device_link(fs_devices, device);
2192 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2193 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2194 mutex_unlock(&fs_devices->device_list_mutex);
2197 * at this point, the device is zero sized and detached from
2198 * the devices list. All that's left is to zero out the old
2199 * supers and free the device.
2201 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2202 btrfs_scratch_superblocks(device->bdev, device->name->str);
2204 btrfs_close_bdev(device);
2206 btrfs_free_device(device);
2208 if (cur_devices->open_devices == 0) {
2209 while (fs_devices) {
2210 if (fs_devices->seed == cur_devices) {
2211 fs_devices->seed = cur_devices->seed;
2214 fs_devices = fs_devices->seed;
2216 cur_devices->seed = NULL;
2217 close_fs_devices(cur_devices);
2218 free_fs_devices(cur_devices);
2222 mutex_unlock(&uuid_mutex);
2226 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2227 mutex_lock(&fs_info->chunk_mutex);
2228 list_add(&device->dev_alloc_list,
2229 &fs_devices->alloc_list);
2230 device->fs_devices->rw_devices++;
2231 mutex_unlock(&fs_info->chunk_mutex);
2236 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2238 struct btrfs_fs_devices *fs_devices;
2240 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2243 * in case of fs with no seed, srcdev->fs_devices will point
2244 * to fs_devices of fs_info. However when the dev being replaced is
2245 * a seed dev it will point to the seed's local fs_devices. In short
2246 * srcdev will have its correct fs_devices in both the cases.
2248 fs_devices = srcdev->fs_devices;
2250 list_del_rcu(&srcdev->dev_list);
2251 list_del(&srcdev->dev_alloc_list);
2252 fs_devices->num_devices--;
2253 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2254 fs_devices->missing_devices--;
2256 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2257 fs_devices->rw_devices--;
2260 fs_devices->open_devices--;
2263 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2264 struct btrfs_device *srcdev)
2266 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2268 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2269 /* zero out the old super if it is writable */
2270 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2273 btrfs_close_bdev(srcdev);
2275 btrfs_free_device(srcdev);
2277 /* if this is no devs we rather delete the fs_devices */
2278 if (!fs_devices->num_devices) {
2279 struct btrfs_fs_devices *tmp_fs_devices;
2282 * On a mounted FS, num_devices can't be zero unless it's a
2283 * seed. In case of a seed device being replaced, the replace
2284 * target added to the sprout FS, so there will be no more
2285 * device left under the seed FS.
2287 ASSERT(fs_devices->seeding);
2289 tmp_fs_devices = fs_info->fs_devices;
2290 while (tmp_fs_devices) {
2291 if (tmp_fs_devices->seed == fs_devices) {
2292 tmp_fs_devices->seed = fs_devices->seed;
2295 tmp_fs_devices = tmp_fs_devices->seed;
2297 fs_devices->seed = NULL;
2298 close_fs_devices(fs_devices);
2299 free_fs_devices(fs_devices);
2303 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2305 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2308 mutex_lock(&fs_devices->device_list_mutex);
2310 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2313 fs_devices->open_devices--;
2315 fs_devices->num_devices--;
2317 btrfs_assign_next_active_device(tgtdev, NULL);
2319 list_del_rcu(&tgtdev->dev_list);
2321 mutex_unlock(&fs_devices->device_list_mutex);
2324 * The update_dev_time() with in btrfs_scratch_superblocks()
2325 * may lead to a call to btrfs_show_devname() which will try
2326 * to hold device_list_mutex. And here this device
2327 * is already out of device list, so we don't have to hold
2328 * the device_list_mutex lock.
2330 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2332 btrfs_close_bdev(tgtdev);
2334 btrfs_free_device(tgtdev);
2337 static struct btrfs_device *btrfs_find_device_by_path(
2338 struct btrfs_fs_info *fs_info, const char *device_path)
2341 struct btrfs_super_block *disk_super;
2344 struct block_device *bdev;
2345 struct buffer_head *bh;
2346 struct btrfs_device *device;
2348 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2349 fs_info->bdev_holder, 0, &bdev, &bh);
2351 return ERR_PTR(ret);
2352 disk_super = (struct btrfs_super_block *)bh->b_data;
2353 devid = btrfs_stack_device_id(&disk_super->dev_item);
2354 dev_uuid = disk_super->dev_item.uuid;
2355 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2356 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2357 disk_super->metadata_uuid, true);
2359 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2360 disk_super->fsid, true);
2364 device = ERR_PTR(-ENOENT);
2365 blkdev_put(bdev, FMODE_READ);
2370 * Lookup a device given by device id, or the path if the id is 0.
2372 struct btrfs_device *btrfs_find_device_by_devspec(
2373 struct btrfs_fs_info *fs_info, u64 devid,
2374 const char *device_path)
2376 struct btrfs_device *device;
2379 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2382 return ERR_PTR(-ENOENT);
2386 if (!device_path || !device_path[0])
2387 return ERR_PTR(-EINVAL);
2389 if (strcmp(device_path, "missing") == 0) {
2390 /* Find first missing device */
2391 list_for_each_entry(device, &fs_info->fs_devices->devices,
2393 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2394 &device->dev_state) && !device->bdev)
2397 return ERR_PTR(-ENOENT);
2400 return btrfs_find_device_by_path(fs_info, device_path);
2404 * does all the dirty work required for changing file system's UUID.
2406 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2408 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2409 struct btrfs_fs_devices *old_devices;
2410 struct btrfs_fs_devices *seed_devices;
2411 struct btrfs_super_block *disk_super = fs_info->super_copy;
2412 struct btrfs_device *device;
2415 lockdep_assert_held(&uuid_mutex);
2416 if (!fs_devices->seeding)
2419 seed_devices = alloc_fs_devices(NULL, NULL);
2420 if (IS_ERR(seed_devices))
2421 return PTR_ERR(seed_devices);
2423 old_devices = clone_fs_devices(fs_devices);
2424 if (IS_ERR(old_devices)) {
2425 kfree(seed_devices);
2426 return PTR_ERR(old_devices);
2429 list_add(&old_devices->fs_list, &fs_uuids);
2431 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2432 seed_devices->opened = 1;
2433 INIT_LIST_HEAD(&seed_devices->devices);
2434 INIT_LIST_HEAD(&seed_devices->alloc_list);
2435 mutex_init(&seed_devices->device_list_mutex);
2437 mutex_lock(&fs_devices->device_list_mutex);
2438 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2440 list_for_each_entry(device, &seed_devices->devices, dev_list)
2441 device->fs_devices = seed_devices;
2443 mutex_lock(&fs_info->chunk_mutex);
2444 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2445 mutex_unlock(&fs_info->chunk_mutex);
2447 fs_devices->seeding = 0;
2448 fs_devices->num_devices = 0;
2449 fs_devices->open_devices = 0;
2450 fs_devices->missing_devices = 0;
2451 fs_devices->rotating = 0;
2452 fs_devices->seed = seed_devices;
2454 generate_random_uuid(fs_devices->fsid);
2455 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2456 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2457 mutex_unlock(&fs_devices->device_list_mutex);
2459 super_flags = btrfs_super_flags(disk_super) &
2460 ~BTRFS_SUPER_FLAG_SEEDING;
2461 btrfs_set_super_flags(disk_super, super_flags);
2467 * Store the expected generation for seed devices in device items.
2469 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2470 struct btrfs_fs_info *fs_info)
2472 struct btrfs_root *root = fs_info->chunk_root;
2473 struct btrfs_path *path;
2474 struct extent_buffer *leaf;
2475 struct btrfs_dev_item *dev_item;
2476 struct btrfs_device *device;
2477 struct btrfs_key key;
2478 u8 fs_uuid[BTRFS_FSID_SIZE];
2479 u8 dev_uuid[BTRFS_UUID_SIZE];
2483 path = btrfs_alloc_path();
2487 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2489 key.type = BTRFS_DEV_ITEM_KEY;
2492 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2496 leaf = path->nodes[0];
2498 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2499 ret = btrfs_next_leaf(root, path);
2504 leaf = path->nodes[0];
2505 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2506 btrfs_release_path(path);
2510 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2511 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2512 key.type != BTRFS_DEV_ITEM_KEY)
2515 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2516 struct btrfs_dev_item);
2517 devid = btrfs_device_id(leaf, dev_item);
2518 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2520 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2522 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2524 BUG_ON(!device); /* Logic error */
2526 if (device->fs_devices->seeding) {
2527 btrfs_set_device_generation(leaf, dev_item,
2528 device->generation);
2529 btrfs_mark_buffer_dirty(leaf);
2537 btrfs_free_path(path);
2541 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2543 struct btrfs_root *root = fs_info->dev_root;
2544 struct request_queue *q;
2545 struct btrfs_trans_handle *trans;
2546 struct btrfs_device *device;
2547 struct block_device *bdev;
2548 struct super_block *sb = fs_info->sb;
2549 struct rcu_string *name;
2550 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2551 u64 orig_super_total_bytes;
2552 u64 orig_super_num_devices;
2553 int seeding_dev = 0;
2555 bool unlocked = false;
2557 if (sb_rdonly(sb) && !fs_devices->seeding)
2560 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2561 fs_info->bdev_holder);
2563 return PTR_ERR(bdev);
2565 if (fs_devices->seeding) {
2567 down_write(&sb->s_umount);
2568 mutex_lock(&uuid_mutex);
2571 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2573 mutex_lock(&fs_devices->device_list_mutex);
2574 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2575 if (device->bdev == bdev) {
2578 &fs_devices->device_list_mutex);
2582 mutex_unlock(&fs_devices->device_list_mutex);
2584 device = btrfs_alloc_device(fs_info, NULL, NULL);
2585 if (IS_ERR(device)) {
2586 /* we can safely leave the fs_devices entry around */
2587 ret = PTR_ERR(device);
2591 name = rcu_string_strdup(device_path, GFP_KERNEL);
2594 goto error_free_device;
2596 rcu_assign_pointer(device->name, name);
2598 trans = btrfs_start_transaction(root, 0);
2599 if (IS_ERR(trans)) {
2600 ret = PTR_ERR(trans);
2601 goto error_free_device;
2604 q = bdev_get_queue(bdev);
2605 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2606 device->generation = trans->transid;
2607 device->io_width = fs_info->sectorsize;
2608 device->io_align = fs_info->sectorsize;
2609 device->sector_size = fs_info->sectorsize;
2610 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2611 fs_info->sectorsize);
2612 device->disk_total_bytes = device->total_bytes;
2613 device->commit_total_bytes = device->total_bytes;
2614 device->fs_info = fs_info;
2615 device->bdev = bdev;
2616 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2617 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2618 device->mode = FMODE_EXCL;
2619 device->dev_stats_valid = 1;
2620 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2623 sb->s_flags &= ~SB_RDONLY;
2624 ret = btrfs_prepare_sprout(fs_info);
2626 btrfs_abort_transaction(trans, ret);
2631 device->fs_devices = fs_devices;
2633 mutex_lock(&fs_devices->device_list_mutex);
2634 mutex_lock(&fs_info->chunk_mutex);
2635 list_add_rcu(&device->dev_list, &fs_devices->devices);
2636 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2637 fs_devices->num_devices++;
2638 fs_devices->open_devices++;
2639 fs_devices->rw_devices++;
2640 fs_devices->total_devices++;
2641 fs_devices->total_rw_bytes += device->total_bytes;
2643 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2645 if (!blk_queue_nonrot(q))
2646 fs_devices->rotating = 1;
2648 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2649 btrfs_set_super_total_bytes(fs_info->super_copy,
2650 round_down(orig_super_total_bytes + device->total_bytes,
2651 fs_info->sectorsize));
2653 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2654 btrfs_set_super_num_devices(fs_info->super_copy,
2655 orig_super_num_devices + 1);
2657 /* add sysfs device entry */
2658 btrfs_sysfs_add_device_link(fs_devices, device);
2661 * we've got more storage, clear any full flags on the space
2664 btrfs_clear_space_info_full(fs_info);
2666 mutex_unlock(&fs_info->chunk_mutex);
2667 mutex_unlock(&fs_devices->device_list_mutex);
2670 mutex_lock(&fs_info->chunk_mutex);
2671 ret = init_first_rw_device(trans, fs_info);
2672 mutex_unlock(&fs_info->chunk_mutex);
2674 btrfs_abort_transaction(trans, ret);
2679 ret = btrfs_add_dev_item(trans, device);
2681 btrfs_abort_transaction(trans, ret);
2686 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2688 ret = btrfs_finish_sprout(trans, fs_info);
2690 btrfs_abort_transaction(trans, ret);
2694 /* Sprouting would change fsid of the mounted root,
2695 * so rename the fsid on the sysfs
2697 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2698 fs_info->fs_devices->fsid);
2699 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2701 "sysfs: failed to create fsid for sprout");
2704 ret = btrfs_commit_transaction(trans);
2707 mutex_unlock(&uuid_mutex);
2708 up_write(&sb->s_umount);
2711 if (ret) /* transaction commit */
2714 ret = btrfs_relocate_sys_chunks(fs_info);
2716 btrfs_handle_fs_error(fs_info, ret,
2717 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2718 trans = btrfs_attach_transaction(root);
2719 if (IS_ERR(trans)) {
2720 if (PTR_ERR(trans) == -ENOENT)
2722 ret = PTR_ERR(trans);
2726 ret = btrfs_commit_transaction(trans);
2729 /* Update ctime/mtime for libblkid */
2730 update_dev_time(device_path);
2734 btrfs_sysfs_rm_device_link(fs_devices, device);
2735 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2736 mutex_lock(&fs_info->chunk_mutex);
2737 list_del_rcu(&device->dev_list);
2738 list_del(&device->dev_alloc_list);
2739 fs_info->fs_devices->num_devices--;
2740 fs_info->fs_devices->open_devices--;
2741 fs_info->fs_devices->rw_devices--;
2742 fs_info->fs_devices->total_devices--;
2743 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2744 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2745 btrfs_set_super_total_bytes(fs_info->super_copy,
2746 orig_super_total_bytes);
2747 btrfs_set_super_num_devices(fs_info->super_copy,
2748 orig_super_num_devices);
2749 mutex_unlock(&fs_info->chunk_mutex);
2750 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2753 sb->s_flags |= SB_RDONLY;
2755 btrfs_end_transaction(trans);
2757 btrfs_free_device(device);
2759 blkdev_put(bdev, FMODE_EXCL);
2760 if (seeding_dev && !unlocked) {
2761 mutex_unlock(&uuid_mutex);
2762 up_write(&sb->s_umount);
2767 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2768 struct btrfs_device *device)
2771 struct btrfs_path *path;
2772 struct btrfs_root *root = device->fs_info->chunk_root;
2773 struct btrfs_dev_item *dev_item;
2774 struct extent_buffer *leaf;
2775 struct btrfs_key key;
2777 path = btrfs_alloc_path();
2781 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2782 key.type = BTRFS_DEV_ITEM_KEY;
2783 key.offset = device->devid;
2785 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2794 leaf = path->nodes[0];
2795 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2797 btrfs_set_device_id(leaf, dev_item, device->devid);
2798 btrfs_set_device_type(leaf, dev_item, device->type);
2799 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2800 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2801 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2802 btrfs_set_device_total_bytes(leaf, dev_item,
2803 btrfs_device_get_disk_total_bytes(device));
2804 btrfs_set_device_bytes_used(leaf, dev_item,
2805 btrfs_device_get_bytes_used(device));
2806 btrfs_mark_buffer_dirty(leaf);
2809 btrfs_free_path(path);
2813 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2814 struct btrfs_device *device, u64 new_size)
2816 struct btrfs_fs_info *fs_info = device->fs_info;
2817 struct btrfs_super_block *super_copy = fs_info->super_copy;
2821 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2824 new_size = round_down(new_size, fs_info->sectorsize);
2826 mutex_lock(&fs_info->chunk_mutex);
2827 old_total = btrfs_super_total_bytes(super_copy);
2828 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2830 if (new_size <= device->total_bytes ||
2831 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2832 mutex_unlock(&fs_info->chunk_mutex);
2836 btrfs_set_super_total_bytes(super_copy,
2837 round_down(old_total + diff, fs_info->sectorsize));
2838 device->fs_devices->total_rw_bytes += diff;
2840 btrfs_device_set_total_bytes(device, new_size);
2841 btrfs_device_set_disk_total_bytes(device, new_size);
2842 btrfs_clear_space_info_full(device->fs_info);
2843 if (list_empty(&device->post_commit_list))
2844 list_add_tail(&device->post_commit_list,
2845 &trans->transaction->dev_update_list);
2846 mutex_unlock(&fs_info->chunk_mutex);
2848 return btrfs_update_device(trans, device);
2851 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2853 struct btrfs_fs_info *fs_info = trans->fs_info;
2854 struct btrfs_root *root = fs_info->chunk_root;
2856 struct btrfs_path *path;
2857 struct btrfs_key key;
2859 path = btrfs_alloc_path();
2863 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2864 key.offset = chunk_offset;
2865 key.type = BTRFS_CHUNK_ITEM_KEY;
2867 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2870 else if (ret > 0) { /* Logic error or corruption */
2871 btrfs_handle_fs_error(fs_info, -ENOENT,
2872 "Failed lookup while freeing chunk.");
2877 ret = btrfs_del_item(trans, root, path);
2879 btrfs_handle_fs_error(fs_info, ret,
2880 "Failed to delete chunk item.");
2882 btrfs_free_path(path);
2886 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2888 struct btrfs_super_block *super_copy = fs_info->super_copy;
2889 struct btrfs_disk_key *disk_key;
2890 struct btrfs_chunk *chunk;
2897 struct btrfs_key key;
2899 mutex_lock(&fs_info->chunk_mutex);
2900 array_size = btrfs_super_sys_array_size(super_copy);
2902 ptr = super_copy->sys_chunk_array;
2905 while (cur < array_size) {
2906 disk_key = (struct btrfs_disk_key *)ptr;
2907 btrfs_disk_key_to_cpu(&key, disk_key);
2909 len = sizeof(*disk_key);
2911 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2912 chunk = (struct btrfs_chunk *)(ptr + len);
2913 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2914 len += btrfs_chunk_item_size(num_stripes);
2919 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2920 key.offset == chunk_offset) {
2921 memmove(ptr, ptr + len, array_size - (cur + len));
2923 btrfs_set_super_sys_array_size(super_copy, array_size);
2929 mutex_unlock(&fs_info->chunk_mutex);
2934 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2935 * @logical: Logical block offset in bytes.
2936 * @length: Length of extent in bytes.
2938 * Return: Chunk mapping or ERR_PTR.
2940 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2941 u64 logical, u64 length)
2943 struct extent_map_tree *em_tree;
2944 struct extent_map *em;
2946 em_tree = &fs_info->mapping_tree.map_tree;
2947 read_lock(&em_tree->lock);
2948 em = lookup_extent_mapping(em_tree, logical, length);
2949 read_unlock(&em_tree->lock);
2952 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2954 return ERR_PTR(-EINVAL);
2957 if (em->start > logical || em->start + em->len < logical) {
2959 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2960 logical, length, em->start, em->start + em->len);
2961 free_extent_map(em);
2962 return ERR_PTR(-EINVAL);
2965 /* callers are responsible for dropping em's ref. */
2969 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2971 struct btrfs_fs_info *fs_info = trans->fs_info;
2972 struct extent_map *em;
2973 struct map_lookup *map;
2974 u64 dev_extent_len = 0;
2976 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2978 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2981 * This is a logic error, but we don't want to just rely on the
2982 * user having built with ASSERT enabled, so if ASSERT doesn't
2983 * do anything we still error out.
2988 map = em->map_lookup;
2989 mutex_lock(&fs_info->chunk_mutex);
2990 check_system_chunk(trans, map->type);
2991 mutex_unlock(&fs_info->chunk_mutex);
2994 * Take the device list mutex to prevent races with the final phase of
2995 * a device replace operation that replaces the device object associated
2996 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2998 mutex_lock(&fs_devices->device_list_mutex);
2999 for (i = 0; i < map->num_stripes; i++) {
3000 struct btrfs_device *device = map->stripes[i].dev;
3001 ret = btrfs_free_dev_extent(trans, device,
3002 map->stripes[i].physical,
3005 mutex_unlock(&fs_devices->device_list_mutex);
3006 btrfs_abort_transaction(trans, ret);
3010 if (device->bytes_used > 0) {
3011 mutex_lock(&fs_info->chunk_mutex);
3012 btrfs_device_set_bytes_used(device,
3013 device->bytes_used - dev_extent_len);
3014 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3015 btrfs_clear_space_info_full(fs_info);
3016 mutex_unlock(&fs_info->chunk_mutex);
3019 ret = btrfs_update_device(trans, device);
3021 mutex_unlock(&fs_devices->device_list_mutex);
3022 btrfs_abort_transaction(trans, ret);
3026 mutex_unlock(&fs_devices->device_list_mutex);
3028 ret = btrfs_free_chunk(trans, chunk_offset);
3030 btrfs_abort_transaction(trans, ret);
3034 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3036 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3037 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3039 btrfs_abort_transaction(trans, ret);
3044 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3046 btrfs_abort_transaction(trans, ret);
3052 free_extent_map(em);
3056 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3058 struct btrfs_root *root = fs_info->chunk_root;
3059 struct btrfs_trans_handle *trans;
3063 * Prevent races with automatic removal of unused block groups.
3064 * After we relocate and before we remove the chunk with offset
3065 * chunk_offset, automatic removal of the block group can kick in,
3066 * resulting in a failure when calling btrfs_remove_chunk() below.
3068 * Make sure to acquire this mutex before doing a tree search (dev
3069 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3070 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3071 * we release the path used to search the chunk/dev tree and before
3072 * the current task acquires this mutex and calls us.
3074 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3076 ret = btrfs_can_relocate(fs_info, chunk_offset);
3080 /* step one, relocate all the extents inside this chunk */
3081 btrfs_scrub_pause(fs_info);
3082 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3083 btrfs_scrub_continue(fs_info);
3088 * We add the kobjects here (and after forcing data chunk creation)
3089 * since relocation is the only place we'll create chunks of a new
3090 * type at runtime. The only place where we'll remove the last
3091 * chunk of a type is the call immediately below this one. Even
3092 * so, we're protected against races with the cleaner thread since
3093 * we're covered by the delete_unused_bgs_mutex.
3095 btrfs_add_raid_kobjects(fs_info);
3097 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3099 if (IS_ERR(trans)) {
3100 ret = PTR_ERR(trans);
3101 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3106 * step two, delete the device extents and the
3107 * chunk tree entries
3109 ret = btrfs_remove_chunk(trans, chunk_offset);
3110 btrfs_end_transaction(trans);
3114 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3116 struct btrfs_root *chunk_root = fs_info->chunk_root;
3117 struct btrfs_path *path;
3118 struct extent_buffer *leaf;
3119 struct btrfs_chunk *chunk;
3120 struct btrfs_key key;
3121 struct btrfs_key found_key;
3123 bool retried = false;
3127 path = btrfs_alloc_path();
3132 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3133 key.offset = (u64)-1;
3134 key.type = BTRFS_CHUNK_ITEM_KEY;
3137 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3138 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3140 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3143 BUG_ON(ret == 0); /* Corruption */
3145 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3148 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3154 leaf = path->nodes[0];
3155 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3157 chunk = btrfs_item_ptr(leaf, path->slots[0],
3158 struct btrfs_chunk);
3159 chunk_type = btrfs_chunk_type(leaf, chunk);
3160 btrfs_release_path(path);
3162 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3163 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3169 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3171 if (found_key.offset == 0)
3173 key.offset = found_key.offset - 1;
3176 if (failed && !retried) {
3180 } else if (WARN_ON(failed && retried)) {
3184 btrfs_free_path(path);
3189 * return 1 : allocate a data chunk successfully,
3190 * return <0: errors during allocating a data chunk,
3191 * return 0 : no need to allocate a data chunk.
3193 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3196 struct btrfs_block_group_cache *cache;
3200 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3202 chunk_type = cache->flags;
3203 btrfs_put_block_group(cache);
3205 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3206 spin_lock(&fs_info->data_sinfo->lock);
3207 bytes_used = fs_info->data_sinfo->bytes_used;
3208 spin_unlock(&fs_info->data_sinfo->lock);
3211 struct btrfs_trans_handle *trans;
3214 trans = btrfs_join_transaction(fs_info->tree_root);
3216 return PTR_ERR(trans);
3218 ret = btrfs_force_chunk_alloc(trans,
3219 BTRFS_BLOCK_GROUP_DATA);
3220 btrfs_end_transaction(trans);
3224 btrfs_add_raid_kobjects(fs_info);
3232 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3233 struct btrfs_balance_control *bctl)
3235 struct btrfs_root *root = fs_info->tree_root;
3236 struct btrfs_trans_handle *trans;
3237 struct btrfs_balance_item *item;
3238 struct btrfs_disk_balance_args disk_bargs;
3239 struct btrfs_path *path;
3240 struct extent_buffer *leaf;
3241 struct btrfs_key key;
3244 path = btrfs_alloc_path();
3248 trans = btrfs_start_transaction(root, 0);
3249 if (IS_ERR(trans)) {
3250 btrfs_free_path(path);
3251 return PTR_ERR(trans);
3254 key.objectid = BTRFS_BALANCE_OBJECTID;
3255 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3258 ret = btrfs_insert_empty_item(trans, root, path, &key,
3263 leaf = path->nodes[0];
3264 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3266 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3268 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3269 btrfs_set_balance_data(leaf, item, &disk_bargs);
3270 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3271 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3272 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3273 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3275 btrfs_set_balance_flags(leaf, item, bctl->flags);
3277 btrfs_mark_buffer_dirty(leaf);
3279 btrfs_free_path(path);
3280 err = btrfs_commit_transaction(trans);
3286 static int del_balance_item(struct btrfs_fs_info *fs_info)
3288 struct btrfs_root *root = fs_info->tree_root;
3289 struct btrfs_trans_handle *trans;
3290 struct btrfs_path *path;
3291 struct btrfs_key key;
3294 path = btrfs_alloc_path();
3298 trans = btrfs_start_transaction(root, 0);
3299 if (IS_ERR(trans)) {
3300 btrfs_free_path(path);
3301 return PTR_ERR(trans);
3304 key.objectid = BTRFS_BALANCE_OBJECTID;
3305 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3308 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3316 ret = btrfs_del_item(trans, root, path);
3318 btrfs_free_path(path);
3319 err = btrfs_commit_transaction(trans);
3326 * This is a heuristic used to reduce the number of chunks balanced on
3327 * resume after balance was interrupted.
3329 static void update_balance_args(struct btrfs_balance_control *bctl)
3332 * Turn on soft mode for chunk types that were being converted.
3334 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3335 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3336 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3337 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3338 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3339 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3342 * Turn on usage filter if is not already used. The idea is
3343 * that chunks that we have already balanced should be
3344 * reasonably full. Don't do it for chunks that are being
3345 * converted - that will keep us from relocating unconverted
3346 * (albeit full) chunks.
3348 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3349 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3350 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3351 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3352 bctl->data.usage = 90;
3354 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3355 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3356 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3357 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3358 bctl->sys.usage = 90;
3360 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3361 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3362 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3363 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3364 bctl->meta.usage = 90;
3369 * Clear the balance status in fs_info and delete the balance item from disk.
3371 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3373 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3376 BUG_ON(!fs_info->balance_ctl);
3378 spin_lock(&fs_info->balance_lock);
3379 fs_info->balance_ctl = NULL;
3380 spin_unlock(&fs_info->balance_lock);
3383 ret = del_balance_item(fs_info);
3385 btrfs_handle_fs_error(fs_info, ret, NULL);
3389 * Balance filters. Return 1 if chunk should be filtered out
3390 * (should not be balanced).
3392 static int chunk_profiles_filter(u64 chunk_type,
3393 struct btrfs_balance_args *bargs)
3395 chunk_type = chunk_to_extended(chunk_type) &
3396 BTRFS_EXTENDED_PROFILE_MASK;
3398 if (bargs->profiles & chunk_type)
3404 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3405 struct btrfs_balance_args *bargs)
3407 struct btrfs_block_group_cache *cache;
3409 u64 user_thresh_min;
3410 u64 user_thresh_max;
3413 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3414 chunk_used = btrfs_block_group_used(&cache->item);
3416 if (bargs->usage_min == 0)
3417 user_thresh_min = 0;
3419 user_thresh_min = div_factor_fine(cache->key.offset,
3422 if (bargs->usage_max == 0)
3423 user_thresh_max = 1;
3424 else if (bargs->usage_max > 100)
3425 user_thresh_max = cache->key.offset;
3427 user_thresh_max = div_factor_fine(cache->key.offset,
3430 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3433 btrfs_put_block_group(cache);
3437 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3438 u64 chunk_offset, struct btrfs_balance_args *bargs)
3440 struct btrfs_block_group_cache *cache;
3441 u64 chunk_used, user_thresh;
3444 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3445 chunk_used = btrfs_block_group_used(&cache->item);
3447 if (bargs->usage_min == 0)
3449 else if (bargs->usage > 100)
3450 user_thresh = cache->key.offset;
3452 user_thresh = div_factor_fine(cache->key.offset,
3455 if (chunk_used < user_thresh)
3458 btrfs_put_block_group(cache);
3462 static int chunk_devid_filter(struct extent_buffer *leaf,
3463 struct btrfs_chunk *chunk,
3464 struct btrfs_balance_args *bargs)
3466 struct btrfs_stripe *stripe;
3467 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3470 for (i = 0; i < num_stripes; i++) {
3471 stripe = btrfs_stripe_nr(chunk, i);
3472 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3479 /* [pstart, pend) */
3480 static int chunk_drange_filter(struct extent_buffer *leaf,
3481 struct btrfs_chunk *chunk,
3482 struct btrfs_balance_args *bargs)
3484 struct btrfs_stripe *stripe;
3485 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3491 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3494 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3495 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3496 factor = num_stripes / 2;
3497 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3498 factor = num_stripes - 1;
3499 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3500 factor = num_stripes - 2;
3502 factor = num_stripes;
3505 for (i = 0; i < num_stripes; i++) {
3506 stripe = btrfs_stripe_nr(chunk, i);
3507 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3510 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3511 stripe_length = btrfs_chunk_length(leaf, chunk);
3512 stripe_length = div_u64(stripe_length, factor);
3514 if (stripe_offset < bargs->pend &&
3515 stripe_offset + stripe_length > bargs->pstart)
3522 /* [vstart, vend) */
3523 static int chunk_vrange_filter(struct extent_buffer *leaf,
3524 struct btrfs_chunk *chunk,
3526 struct btrfs_balance_args *bargs)
3528 if (chunk_offset < bargs->vend &&
3529 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3530 /* at least part of the chunk is inside this vrange */
3536 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3537 struct btrfs_chunk *chunk,
3538 struct btrfs_balance_args *bargs)
3540 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3542 if (bargs->stripes_min <= num_stripes
3543 && num_stripes <= bargs->stripes_max)
3549 static int chunk_soft_convert_filter(u64 chunk_type,
3550 struct btrfs_balance_args *bargs)
3552 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3555 chunk_type = chunk_to_extended(chunk_type) &
3556 BTRFS_EXTENDED_PROFILE_MASK;
3558 if (bargs->target == chunk_type)
3564 static int should_balance_chunk(struct extent_buffer *leaf,
3565 struct btrfs_chunk *chunk, u64 chunk_offset)
3567 struct btrfs_fs_info *fs_info = leaf->fs_info;
3568 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3569 struct btrfs_balance_args *bargs = NULL;
3570 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3573 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3574 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3578 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3579 bargs = &bctl->data;
3580 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3582 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3583 bargs = &bctl->meta;
3585 /* profiles filter */
3586 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3587 chunk_profiles_filter(chunk_type, bargs)) {
3592 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3593 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3595 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3596 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3601 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3602 chunk_devid_filter(leaf, chunk, bargs)) {
3606 /* drange filter, makes sense only with devid filter */
3607 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3608 chunk_drange_filter(leaf, chunk, bargs)) {
3613 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3614 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3618 /* stripes filter */
3619 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3620 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3624 /* soft profile changing mode */
3625 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3626 chunk_soft_convert_filter(chunk_type, bargs)) {
3631 * limited by count, must be the last filter
3633 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3634 if (bargs->limit == 0)
3638 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3640 * Same logic as the 'limit' filter; the minimum cannot be
3641 * determined here because we do not have the global information
3642 * about the count of all chunks that satisfy the filters.
3644 if (bargs->limit_max == 0)
3653 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3655 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3656 struct btrfs_root *chunk_root = fs_info->chunk_root;
3658 struct btrfs_chunk *chunk;
3659 struct btrfs_path *path = NULL;
3660 struct btrfs_key key;
3661 struct btrfs_key found_key;
3662 struct extent_buffer *leaf;
3665 int enospc_errors = 0;
3666 bool counting = true;
3667 /* The single value limit and min/max limits use the same bytes in the */
3668 u64 limit_data = bctl->data.limit;
3669 u64 limit_meta = bctl->meta.limit;
3670 u64 limit_sys = bctl->sys.limit;
3674 int chunk_reserved = 0;
3676 path = btrfs_alloc_path();
3682 /* zero out stat counters */
3683 spin_lock(&fs_info->balance_lock);
3684 memset(&bctl->stat, 0, sizeof(bctl->stat));
3685 spin_unlock(&fs_info->balance_lock);
3689 * The single value limit and min/max limits use the same bytes
3692 bctl->data.limit = limit_data;
3693 bctl->meta.limit = limit_meta;
3694 bctl->sys.limit = limit_sys;
3696 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3697 key.offset = (u64)-1;
3698 key.type = BTRFS_CHUNK_ITEM_KEY;
3701 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3702 atomic_read(&fs_info->balance_cancel_req)) {
3707 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3708 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3710 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3715 * this shouldn't happen, it means the last relocate
3719 BUG(); /* FIXME break ? */
3721 ret = btrfs_previous_item(chunk_root, path, 0,
3722 BTRFS_CHUNK_ITEM_KEY);
3724 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3729 leaf = path->nodes[0];
3730 slot = path->slots[0];
3731 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3733 if (found_key.objectid != key.objectid) {
3734 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3738 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3739 chunk_type = btrfs_chunk_type(leaf, chunk);
3742 spin_lock(&fs_info->balance_lock);
3743 bctl->stat.considered++;
3744 spin_unlock(&fs_info->balance_lock);
3747 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3749 btrfs_release_path(path);
3751 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3756 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3757 spin_lock(&fs_info->balance_lock);
3758 bctl->stat.expected++;
3759 spin_unlock(&fs_info->balance_lock);
3761 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3763 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3765 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3772 * Apply limit_min filter, no need to check if the LIMITS
3773 * filter is used, limit_min is 0 by default
3775 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3776 count_data < bctl->data.limit_min)
3777 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3778 count_meta < bctl->meta.limit_min)
3779 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3780 count_sys < bctl->sys.limit_min)) {
3781 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3785 if (!chunk_reserved) {
3787 * We may be relocating the only data chunk we have,
3788 * which could potentially end up with losing data's
3789 * raid profile, so lets allocate an empty one in
3792 ret = btrfs_may_alloc_data_chunk(fs_info,
3795 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3797 } else if (ret == 1) {
3802 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3803 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3804 if (ret == -ENOSPC) {
3806 } else if (ret == -ETXTBSY) {
3808 "skipping relocation of block group %llu due to active swapfile",
3814 spin_lock(&fs_info->balance_lock);
3815 bctl->stat.completed++;
3816 spin_unlock(&fs_info->balance_lock);
3819 if (found_key.offset == 0)
3821 key.offset = found_key.offset - 1;
3825 btrfs_release_path(path);
3830 btrfs_free_path(path);
3831 if (enospc_errors) {
3832 btrfs_info(fs_info, "%d enospc errors during balance",
3842 * alloc_profile_is_valid - see if a given profile is valid and reduced
3843 * @flags: profile to validate
3844 * @extended: if true @flags is treated as an extended profile
3846 static int alloc_profile_is_valid(u64 flags, int extended)
3848 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3849 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3851 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3853 /* 1) check that all other bits are zeroed */
3857 /* 2) see if profile is reduced */
3859 return !extended; /* "0" is valid for usual profiles */
3861 /* true if exactly one bit set */
3862 return is_power_of_2(flags);
3865 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3867 /* cancel requested || normal exit path */
3868 return atomic_read(&fs_info->balance_cancel_req) ||
3869 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3870 atomic_read(&fs_info->balance_cancel_req) == 0);
3873 /* Non-zero return value signifies invalidity */
3874 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3877 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3878 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3879 (bctl_arg->target & ~allowed)));
3883 * Fill @buf with textual description of balance filter flags @bargs, up to
3884 * @size_buf including the terminating null. The output may be trimmed if it
3885 * does not fit into the provided buffer.
3887 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3891 u32 size_bp = size_buf;
3893 u64 flags = bargs->flags;
3894 char tmp_buf[128] = {'\0'};
3899 #define CHECK_APPEND_NOARG(a) \
3901 ret = snprintf(bp, size_bp, (a)); \
3902 if (ret < 0 || ret >= size_bp) \
3903 goto out_overflow; \
3908 #define CHECK_APPEND_1ARG(a, v1) \
3910 ret = snprintf(bp, size_bp, (a), (v1)); \
3911 if (ret < 0 || ret >= size_bp) \
3912 goto out_overflow; \
3917 #define CHECK_APPEND_2ARG(a, v1, v2) \
3919 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3920 if (ret < 0 || ret >= size_bp) \
3921 goto out_overflow; \
3926 if (flags & BTRFS_BALANCE_ARGS_CONVERT) {
3927 int index = btrfs_bg_flags_to_raid_index(bargs->target);
3929 CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index));
3932 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3933 CHECK_APPEND_NOARG("soft,");
3935 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3936 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3938 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3941 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3942 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3944 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3945 CHECK_APPEND_2ARG("usage=%u..%u,",
3946 bargs->usage_min, bargs->usage_max);
3948 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3949 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3951 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3952 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3953 bargs->pstart, bargs->pend);
3955 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3956 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3957 bargs->vstart, bargs->vend);
3959 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3960 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3962 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3963 CHECK_APPEND_2ARG("limit=%u..%u,",
3964 bargs->limit_min, bargs->limit_max);
3966 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3967 CHECK_APPEND_2ARG("stripes=%u..%u,",
3968 bargs->stripes_min, bargs->stripes_max);
3970 #undef CHECK_APPEND_2ARG
3971 #undef CHECK_APPEND_1ARG
3972 #undef CHECK_APPEND_NOARG
3976 if (size_bp < size_buf)
3977 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3982 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3984 u32 size_buf = 1024;
3985 char tmp_buf[192] = {'\0'};
3988 u32 size_bp = size_buf;
3990 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3992 buf = kzalloc(size_buf, GFP_KERNEL);
3998 #define CHECK_APPEND_1ARG(a, v1) \
4000 ret = snprintf(bp, size_bp, (a), (v1)); \
4001 if (ret < 0 || ret >= size_bp) \
4002 goto out_overflow; \
4007 if (bctl->flags & BTRFS_BALANCE_FORCE)
4008 CHECK_APPEND_1ARG("%s", "-f ");
4010 if (bctl->flags & BTRFS_BALANCE_DATA) {
4011 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4012 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4015 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4016 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4017 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4020 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4021 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4022 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4025 #undef CHECK_APPEND_1ARG
4029 if (size_bp < size_buf)
4030 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4031 btrfs_info(fs_info, "balance: %s %s",
4032 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4033 "resume" : "start", buf);
4039 * Should be called with balance mutexe held
4041 int btrfs_balance(struct btrfs_fs_info *fs_info,
4042 struct btrfs_balance_control *bctl,
4043 struct btrfs_ioctl_balance_args *bargs)
4045 u64 meta_target, data_target;
4051 bool reducing_integrity;
4053 if (btrfs_fs_closing(fs_info) ||
4054 atomic_read(&fs_info->balance_pause_req) ||
4055 atomic_read(&fs_info->balance_cancel_req)) {
4060 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4061 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4065 * In case of mixed groups both data and meta should be picked,
4066 * and identical options should be given for both of them.
4068 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4069 if (mixed && (bctl->flags & allowed)) {
4070 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4071 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4072 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4074 "balance: mixed groups data and metadata options must be the same");
4080 num_devices = btrfs_num_devices(fs_info);
4082 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
4083 if (num_devices > 1)
4084 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
4085 if (num_devices > 2)
4086 allowed |= BTRFS_BLOCK_GROUP_RAID5;
4087 if (num_devices > 3)
4088 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
4089 BTRFS_BLOCK_GROUP_RAID6);
4090 if (validate_convert_profile(&bctl->data, allowed)) {
4091 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
4094 "balance: invalid convert data profile %s",
4095 get_raid_name(index));
4099 if (validate_convert_profile(&bctl->meta, allowed)) {
4100 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
4103 "balance: invalid convert metadata profile %s",
4104 get_raid_name(index));
4108 if (validate_convert_profile(&bctl->sys, allowed)) {
4109 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
4112 "balance: invalid convert system profile %s",
4113 get_raid_name(index));
4118 /* allow to reduce meta or sys integrity only if force set */
4119 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4120 BTRFS_BLOCK_GROUP_RAID10 |
4121 BTRFS_BLOCK_GROUP_RAID5 |
4122 BTRFS_BLOCK_GROUP_RAID6;
4124 seq = read_seqbegin(&fs_info->profiles_lock);
4126 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4127 (fs_info->avail_system_alloc_bits & allowed) &&
4128 !(bctl->sys.target & allowed)) ||
4129 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4130 (fs_info->avail_metadata_alloc_bits & allowed) &&
4131 !(bctl->meta.target & allowed)))
4132 reducing_integrity = true;
4134 reducing_integrity = false;
4136 /* if we're not converting, the target field is uninitialized */
4137 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4138 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4139 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4140 bctl->data.target : fs_info->avail_data_alloc_bits;
4141 } while (read_seqretry(&fs_info->profiles_lock, seq));
4143 if (reducing_integrity) {
4144 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4146 "balance: force reducing metadata integrity");
4149 "balance: reduces metadata integrity, use --force if you want this");
4155 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4156 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4157 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
4158 int data_index = btrfs_bg_flags_to_raid_index(data_target);
4161 "balance: metadata profile %s has lower redundancy than data profile %s",
4162 get_raid_name(meta_index), get_raid_name(data_index));
4165 ret = insert_balance_item(fs_info, bctl);
4166 if (ret && ret != -EEXIST)
4169 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4170 BUG_ON(ret == -EEXIST);
4171 BUG_ON(fs_info->balance_ctl);
4172 spin_lock(&fs_info->balance_lock);
4173 fs_info->balance_ctl = bctl;
4174 spin_unlock(&fs_info->balance_lock);
4176 BUG_ON(ret != -EEXIST);
4177 spin_lock(&fs_info->balance_lock);
4178 update_balance_args(bctl);
4179 spin_unlock(&fs_info->balance_lock);
4182 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4183 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4184 describe_balance_start_or_resume(fs_info);
4185 mutex_unlock(&fs_info->balance_mutex);
4187 ret = __btrfs_balance(fs_info);
4189 mutex_lock(&fs_info->balance_mutex);
4190 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4191 btrfs_info(fs_info, "balance: paused");
4192 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4193 btrfs_info(fs_info, "balance: canceled");
4195 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4197 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4200 memset(bargs, 0, sizeof(*bargs));
4201 btrfs_update_ioctl_balance_args(fs_info, bargs);
4204 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4205 balance_need_close(fs_info)) {
4206 reset_balance_state(fs_info);
4207 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4210 wake_up(&fs_info->balance_wait_q);
4214 if (bctl->flags & BTRFS_BALANCE_RESUME)
4215 reset_balance_state(fs_info);
4218 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4223 static int balance_kthread(void *data)
4225 struct btrfs_fs_info *fs_info = data;
4228 mutex_lock(&fs_info->balance_mutex);
4229 if (fs_info->balance_ctl)
4230 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4231 mutex_unlock(&fs_info->balance_mutex);
4236 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4238 struct task_struct *tsk;
4240 mutex_lock(&fs_info->balance_mutex);
4241 if (!fs_info->balance_ctl) {
4242 mutex_unlock(&fs_info->balance_mutex);
4245 mutex_unlock(&fs_info->balance_mutex);
4247 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4248 btrfs_info(fs_info, "balance: resume skipped");
4253 * A ro->rw remount sequence should continue with the paused balance
4254 * regardless of who pauses it, system or the user as of now, so set
4257 spin_lock(&fs_info->balance_lock);
4258 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4259 spin_unlock(&fs_info->balance_lock);
4261 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4262 return PTR_ERR_OR_ZERO(tsk);
4265 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4267 struct btrfs_balance_control *bctl;
4268 struct btrfs_balance_item *item;
4269 struct btrfs_disk_balance_args disk_bargs;
4270 struct btrfs_path *path;
4271 struct extent_buffer *leaf;
4272 struct btrfs_key key;
4275 path = btrfs_alloc_path();
4279 key.objectid = BTRFS_BALANCE_OBJECTID;
4280 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4283 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4286 if (ret > 0) { /* ret = -ENOENT; */
4291 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4297 leaf = path->nodes[0];
4298 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4300 bctl->flags = btrfs_balance_flags(leaf, item);
4301 bctl->flags |= BTRFS_BALANCE_RESUME;
4303 btrfs_balance_data(leaf, item, &disk_bargs);
4304 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4305 btrfs_balance_meta(leaf, item, &disk_bargs);
4306 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4307 btrfs_balance_sys(leaf, item, &disk_bargs);
4308 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4311 * This should never happen, as the paused balance state is recovered
4312 * during mount without any chance of other exclusive ops to collide.
4314 * This gives the exclusive op status to balance and keeps in paused
4315 * state until user intervention (cancel or umount). If the ownership
4316 * cannot be assigned, show a message but do not fail. The balance
4317 * is in a paused state and must have fs_info::balance_ctl properly
4320 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4322 "balance: cannot set exclusive op status, resume manually");
4324 mutex_lock(&fs_info->balance_mutex);
4325 BUG_ON(fs_info->balance_ctl);
4326 spin_lock(&fs_info->balance_lock);
4327 fs_info->balance_ctl = bctl;
4328 spin_unlock(&fs_info->balance_lock);
4329 mutex_unlock(&fs_info->balance_mutex);
4331 btrfs_free_path(path);
4335 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4339 mutex_lock(&fs_info->balance_mutex);
4340 if (!fs_info->balance_ctl) {
4341 mutex_unlock(&fs_info->balance_mutex);
4345 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4346 atomic_inc(&fs_info->balance_pause_req);
4347 mutex_unlock(&fs_info->balance_mutex);
4349 wait_event(fs_info->balance_wait_q,
4350 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4352 mutex_lock(&fs_info->balance_mutex);
4353 /* we are good with balance_ctl ripped off from under us */
4354 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4355 atomic_dec(&fs_info->balance_pause_req);
4360 mutex_unlock(&fs_info->balance_mutex);
4364 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4366 mutex_lock(&fs_info->balance_mutex);
4367 if (!fs_info->balance_ctl) {
4368 mutex_unlock(&fs_info->balance_mutex);
4373 * A paused balance with the item stored on disk can be resumed at
4374 * mount time if the mount is read-write. Otherwise it's still paused
4375 * and we must not allow cancelling as it deletes the item.
4377 if (sb_rdonly(fs_info->sb)) {
4378 mutex_unlock(&fs_info->balance_mutex);
4382 atomic_inc(&fs_info->balance_cancel_req);
4384 * if we are running just wait and return, balance item is
4385 * deleted in btrfs_balance in this case
4387 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4388 mutex_unlock(&fs_info->balance_mutex);
4389 wait_event(fs_info->balance_wait_q,
4390 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4391 mutex_lock(&fs_info->balance_mutex);
4393 mutex_unlock(&fs_info->balance_mutex);
4395 * Lock released to allow other waiters to continue, we'll
4396 * reexamine the status again.
4398 mutex_lock(&fs_info->balance_mutex);
4400 if (fs_info->balance_ctl) {
4401 reset_balance_state(fs_info);
4402 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4403 btrfs_info(fs_info, "balance: canceled");
4407 BUG_ON(fs_info->balance_ctl ||
4408 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4409 atomic_dec(&fs_info->balance_cancel_req);
4410 mutex_unlock(&fs_info->balance_mutex);
4414 static int btrfs_uuid_scan_kthread(void *data)
4416 struct btrfs_fs_info *fs_info = data;
4417 struct btrfs_root *root = fs_info->tree_root;
4418 struct btrfs_key key;
4419 struct btrfs_path *path = NULL;
4421 struct extent_buffer *eb;
4423 struct btrfs_root_item root_item;
4425 struct btrfs_trans_handle *trans = NULL;
4427 path = btrfs_alloc_path();
4434 key.type = BTRFS_ROOT_ITEM_KEY;
4438 ret = btrfs_search_forward(root, &key, path,
4439 BTRFS_OLDEST_GENERATION);
4446 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4447 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4448 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4449 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4452 eb = path->nodes[0];
4453 slot = path->slots[0];
4454 item_size = btrfs_item_size_nr(eb, slot);
4455 if (item_size < sizeof(root_item))
4458 read_extent_buffer(eb, &root_item,
4459 btrfs_item_ptr_offset(eb, slot),
4460 (int)sizeof(root_item));
4461 if (btrfs_root_refs(&root_item) == 0)
4464 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4465 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4469 btrfs_release_path(path);
4471 * 1 - subvol uuid item
4472 * 1 - received_subvol uuid item
4474 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4475 if (IS_ERR(trans)) {
4476 ret = PTR_ERR(trans);
4484 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4485 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4486 BTRFS_UUID_KEY_SUBVOL,
4489 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4495 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4496 ret = btrfs_uuid_tree_add(trans,
4497 root_item.received_uuid,
4498 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4501 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4509 ret = btrfs_end_transaction(trans);
4515 btrfs_release_path(path);
4516 if (key.offset < (u64)-1) {
4518 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4520 key.type = BTRFS_ROOT_ITEM_KEY;
4521 } else if (key.objectid < (u64)-1) {
4523 key.type = BTRFS_ROOT_ITEM_KEY;
4532 btrfs_free_path(path);
4533 if (trans && !IS_ERR(trans))
4534 btrfs_end_transaction(trans);
4536 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4538 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4539 up(&fs_info->uuid_tree_rescan_sem);
4544 * Callback for btrfs_uuid_tree_iterate().
4546 * 0 check succeeded, the entry is not outdated.
4547 * < 0 if an error occurred.
4548 * > 0 if the check failed, which means the caller shall remove the entry.
4550 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4551 u8 *uuid, u8 type, u64 subid)
4553 struct btrfs_key key;
4555 struct btrfs_root *subvol_root;
4557 if (type != BTRFS_UUID_KEY_SUBVOL &&
4558 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4561 key.objectid = subid;
4562 key.type = BTRFS_ROOT_ITEM_KEY;
4563 key.offset = (u64)-1;
4564 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4565 if (IS_ERR(subvol_root)) {
4566 ret = PTR_ERR(subvol_root);
4573 case BTRFS_UUID_KEY_SUBVOL:
4574 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4577 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4578 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4588 static int btrfs_uuid_rescan_kthread(void *data)
4590 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4594 * 1st step is to iterate through the existing UUID tree and
4595 * to delete all entries that contain outdated data.
4596 * 2nd step is to add all missing entries to the UUID tree.
4598 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4600 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4601 up(&fs_info->uuid_tree_rescan_sem);
4604 return btrfs_uuid_scan_kthread(data);
4607 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4609 struct btrfs_trans_handle *trans;
4610 struct btrfs_root *tree_root = fs_info->tree_root;
4611 struct btrfs_root *uuid_root;
4612 struct task_struct *task;
4619 trans = btrfs_start_transaction(tree_root, 2);
4621 return PTR_ERR(trans);
4623 uuid_root = btrfs_create_tree(trans, fs_info,
4624 BTRFS_UUID_TREE_OBJECTID);
4625 if (IS_ERR(uuid_root)) {
4626 ret = PTR_ERR(uuid_root);
4627 btrfs_abort_transaction(trans, ret);
4628 btrfs_end_transaction(trans);
4632 fs_info->uuid_root = uuid_root;
4634 ret = btrfs_commit_transaction(trans);
4638 down(&fs_info->uuid_tree_rescan_sem);
4639 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4641 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4642 btrfs_warn(fs_info, "failed to start uuid_scan task");
4643 up(&fs_info->uuid_tree_rescan_sem);
4644 return PTR_ERR(task);
4650 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4652 struct task_struct *task;
4654 down(&fs_info->uuid_tree_rescan_sem);
4655 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4657 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4658 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4659 up(&fs_info->uuid_tree_rescan_sem);
4660 return PTR_ERR(task);
4667 * shrinking a device means finding all of the device extents past
4668 * the new size, and then following the back refs to the chunks.
4669 * The chunk relocation code actually frees the device extent
4671 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4673 struct btrfs_fs_info *fs_info = device->fs_info;
4674 struct btrfs_root *root = fs_info->dev_root;
4675 struct btrfs_trans_handle *trans;
4676 struct btrfs_dev_extent *dev_extent = NULL;
4677 struct btrfs_path *path;
4683 bool retried = false;
4684 struct extent_buffer *l;
4685 struct btrfs_key key;
4686 struct btrfs_super_block *super_copy = fs_info->super_copy;
4687 u64 old_total = btrfs_super_total_bytes(super_copy);
4688 u64 old_size = btrfs_device_get_total_bytes(device);
4692 new_size = round_down(new_size, fs_info->sectorsize);
4694 diff = round_down(old_size - new_size, fs_info->sectorsize);
4696 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4699 path = btrfs_alloc_path();
4703 path->reada = READA_BACK;
4705 trans = btrfs_start_transaction(root, 0);
4706 if (IS_ERR(trans)) {
4707 btrfs_free_path(path);
4708 return PTR_ERR(trans);
4711 mutex_lock(&fs_info->chunk_mutex);
4713 btrfs_device_set_total_bytes(device, new_size);
4714 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4715 device->fs_devices->total_rw_bytes -= diff;
4716 atomic64_sub(diff, &fs_info->free_chunk_space);
4720 * Once the device's size has been set to the new size, ensure all
4721 * in-memory chunks are synced to disk so that the loop below sees them
4722 * and relocates them accordingly.
4724 if (contains_pending_extent(device, &start, diff)) {
4725 mutex_unlock(&fs_info->chunk_mutex);
4726 ret = btrfs_commit_transaction(trans);
4730 mutex_unlock(&fs_info->chunk_mutex);
4731 btrfs_end_transaction(trans);
4735 key.objectid = device->devid;
4736 key.offset = (u64)-1;
4737 key.type = BTRFS_DEV_EXTENT_KEY;
4740 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4741 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4743 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4747 ret = btrfs_previous_item(root, path, 0, key.type);
4749 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4754 btrfs_release_path(path);
4759 slot = path->slots[0];
4760 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4762 if (key.objectid != device->devid) {
4763 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4764 btrfs_release_path(path);
4768 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4769 length = btrfs_dev_extent_length(l, dev_extent);
4771 if (key.offset + length <= new_size) {
4772 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4773 btrfs_release_path(path);
4777 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4778 btrfs_release_path(path);
4781 * We may be relocating the only data chunk we have,
4782 * which could potentially end up with losing data's
4783 * raid profile, so lets allocate an empty one in
4786 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4788 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4792 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4793 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4794 if (ret == -ENOSPC) {
4797 if (ret == -ETXTBSY) {
4799 "could not shrink block group %llu due to active swapfile",
4804 } while (key.offset-- > 0);
4806 if (failed && !retried) {
4810 } else if (failed && retried) {
4815 /* Shrinking succeeded, else we would be at "done". */
4816 trans = btrfs_start_transaction(root, 0);
4817 if (IS_ERR(trans)) {
4818 ret = PTR_ERR(trans);
4822 mutex_lock(&fs_info->chunk_mutex);
4823 btrfs_device_set_disk_total_bytes(device, new_size);
4824 if (list_empty(&device->post_commit_list))
4825 list_add_tail(&device->post_commit_list,
4826 &trans->transaction->dev_update_list);
4828 WARN_ON(diff > old_total);
4829 btrfs_set_super_total_bytes(super_copy,
4830 round_down(old_total - diff, fs_info->sectorsize));
4831 mutex_unlock(&fs_info->chunk_mutex);
4833 /* Now btrfs_update_device() will change the on-disk size. */
4834 ret = btrfs_update_device(trans, device);
4836 btrfs_abort_transaction(trans, ret);
4837 btrfs_end_transaction(trans);
4839 ret = btrfs_commit_transaction(trans);
4842 btrfs_free_path(path);
4844 mutex_lock(&fs_info->chunk_mutex);
4845 btrfs_device_set_total_bytes(device, old_size);
4846 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4847 device->fs_devices->total_rw_bytes += diff;
4848 atomic64_add(diff, &fs_info->free_chunk_space);
4849 mutex_unlock(&fs_info->chunk_mutex);
4854 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4855 struct btrfs_key *key,
4856 struct btrfs_chunk *chunk, int item_size)
4858 struct btrfs_super_block *super_copy = fs_info->super_copy;
4859 struct btrfs_disk_key disk_key;
4863 mutex_lock(&fs_info->chunk_mutex);
4864 array_size = btrfs_super_sys_array_size(super_copy);
4865 if (array_size + item_size + sizeof(disk_key)
4866 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4867 mutex_unlock(&fs_info->chunk_mutex);
4871 ptr = super_copy->sys_chunk_array + array_size;
4872 btrfs_cpu_key_to_disk(&disk_key, key);
4873 memcpy(ptr, &disk_key, sizeof(disk_key));
4874 ptr += sizeof(disk_key);
4875 memcpy(ptr, chunk, item_size);
4876 item_size += sizeof(disk_key);
4877 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4878 mutex_unlock(&fs_info->chunk_mutex);
4884 * sort the devices in descending order by max_avail, total_avail
4886 static int btrfs_cmp_device_info(const void *a, const void *b)
4888 const struct btrfs_device_info *di_a = a;
4889 const struct btrfs_device_info *di_b = b;
4891 if (di_a->max_avail > di_b->max_avail)
4893 if (di_a->max_avail < di_b->max_avail)
4895 if (di_a->total_avail > di_b->total_avail)
4897 if (di_a->total_avail < di_b->total_avail)
4902 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4904 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4907 btrfs_set_fs_incompat(info, RAID56);
4910 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4911 u64 start, u64 type)
4913 struct btrfs_fs_info *info = trans->fs_info;
4914 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4915 struct btrfs_device *device;
4916 struct map_lookup *map = NULL;
4917 struct extent_map_tree *em_tree;
4918 struct extent_map *em;
4919 struct btrfs_device_info *devices_info = NULL;
4921 int num_stripes; /* total number of stripes to allocate */
4922 int data_stripes; /* number of stripes that count for
4924 int sub_stripes; /* sub_stripes info for map */
4925 int dev_stripes; /* stripes per dev */
4926 int devs_max; /* max devs to use */
4927 int devs_min; /* min devs needed */
4928 int devs_increment; /* ndevs has to be a multiple of this */
4929 int ncopies; /* how many copies to data has */
4930 int nparity; /* number of stripes worth of bytes to
4931 store parity information */
4933 u64 max_stripe_size;
4942 BUG_ON(!alloc_profile_is_valid(type, 0));
4944 if (list_empty(&fs_devices->alloc_list)) {
4945 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4946 btrfs_debug(info, "%s: no writable device", __func__);
4950 index = btrfs_bg_flags_to_raid_index(type);
4952 sub_stripes = btrfs_raid_array[index].sub_stripes;
4953 dev_stripes = btrfs_raid_array[index].dev_stripes;
4954 devs_max = btrfs_raid_array[index].devs_max;
4955 devs_min = btrfs_raid_array[index].devs_min;
4956 devs_increment = btrfs_raid_array[index].devs_increment;
4957 ncopies = btrfs_raid_array[index].ncopies;
4958 nparity = btrfs_raid_array[index].nparity;
4960 if (type & BTRFS_BLOCK_GROUP_DATA) {
4961 max_stripe_size = SZ_1G;
4962 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4964 devs_max = BTRFS_MAX_DEVS(info);
4965 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4966 /* for larger filesystems, use larger metadata chunks */
4967 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4968 max_stripe_size = SZ_1G;
4970 max_stripe_size = SZ_256M;
4971 max_chunk_size = max_stripe_size;
4973 devs_max = BTRFS_MAX_DEVS(info);
4974 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4975 max_stripe_size = SZ_32M;
4976 max_chunk_size = 2 * max_stripe_size;
4978 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4980 btrfs_err(info, "invalid chunk type 0x%llx requested",
4985 /* We don't want a chunk larger than 10% of writable space */
4986 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4989 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4995 * in the first pass through the devices list, we gather information
4996 * about the available holes on each device.
4999 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5003 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5005 "BTRFS: read-only device in alloc_list\n");
5009 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5010 &device->dev_state) ||
5011 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5014 if (device->total_bytes > device->bytes_used)
5015 total_avail = device->total_bytes - device->bytes_used;
5019 /* If there is no space on this device, skip it. */
5020 if (total_avail == 0)
5023 ret = find_free_dev_extent(device,
5024 max_stripe_size * dev_stripes,
5025 &dev_offset, &max_avail);
5026 if (ret && ret != -ENOSPC)
5030 max_avail = max_stripe_size * dev_stripes;
5032 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5033 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5035 "%s: devid %llu has no free space, have=%llu want=%u",
5036 __func__, device->devid, max_avail,
5037 BTRFS_STRIPE_LEN * dev_stripes);
5041 if (ndevs == fs_devices->rw_devices) {
5042 WARN(1, "%s: found more than %llu devices\n",
5043 __func__, fs_devices->rw_devices);
5046 devices_info[ndevs].dev_offset = dev_offset;
5047 devices_info[ndevs].max_avail = max_avail;
5048 devices_info[ndevs].total_avail = total_avail;
5049 devices_info[ndevs].dev = device;
5054 * now sort the devices by hole size / available space
5056 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5057 btrfs_cmp_device_info, NULL);
5059 /* round down to number of usable stripes */
5060 ndevs = round_down(ndevs, devs_increment);
5062 if (ndevs < devs_min) {
5064 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5066 "%s: not enough devices with free space: have=%d minimum required=%d",
5067 __func__, ndevs, devs_min);
5072 ndevs = min(ndevs, devs_max);
5075 * The primary goal is to maximize the number of stripes, so use as
5076 * many devices as possible, even if the stripes are not maximum sized.
5078 * The DUP profile stores more than one stripe per device, the
5079 * max_avail is the total size so we have to adjust.
5081 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5082 num_stripes = ndevs * dev_stripes;
5085 * this will have to be fixed for RAID1 and RAID10 over
5088 data_stripes = (num_stripes - nparity) / ncopies;
5091 * Use the number of data stripes to figure out how big this chunk
5092 * is really going to be in terms of logical address space,
5093 * and compare that answer with the max chunk size. If it's higher,
5094 * we try to reduce stripe_size.
5096 if (stripe_size * data_stripes > max_chunk_size) {
5098 * Reduce stripe_size, round it up to a 16MB boundary again and
5099 * then use it, unless it ends up being even bigger than the
5100 * previous value we had already.
5102 stripe_size = min(round_up(div_u64(max_chunk_size,
5103 data_stripes), SZ_16M),
5107 /* align to BTRFS_STRIPE_LEN */
5108 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5110 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5115 map->num_stripes = num_stripes;
5117 for (i = 0; i < ndevs; ++i) {
5118 for (j = 0; j < dev_stripes; ++j) {
5119 int s = i * dev_stripes + j;
5120 map->stripes[s].dev = devices_info[i].dev;
5121 map->stripes[s].physical = devices_info[i].dev_offset +
5125 map->stripe_len = BTRFS_STRIPE_LEN;
5126 map->io_align = BTRFS_STRIPE_LEN;
5127 map->io_width = BTRFS_STRIPE_LEN;
5129 map->sub_stripes = sub_stripes;
5131 chunk_size = stripe_size * data_stripes;
5133 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5135 em = alloc_extent_map();
5141 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5142 em->map_lookup = map;
5144 em->len = chunk_size;
5145 em->block_start = 0;
5146 em->block_len = em->len;
5147 em->orig_block_len = stripe_size;
5149 em_tree = &info->mapping_tree.map_tree;
5150 write_lock(&em_tree->lock);
5151 ret = add_extent_mapping(em_tree, em, 0);
5153 write_unlock(&em_tree->lock);
5154 free_extent_map(em);
5157 write_unlock(&em_tree->lock);
5159 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5161 goto error_del_extent;
5163 for (i = 0; i < map->num_stripes; i++) {
5164 struct btrfs_device *dev = map->stripes[i].dev;
5166 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5167 if (list_empty(&dev->post_commit_list))
5168 list_add_tail(&dev->post_commit_list,
5169 &trans->transaction->dev_update_list);
5172 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5174 free_extent_map(em);
5175 check_raid56_incompat_flag(info, type);
5177 kfree(devices_info);
5181 write_lock(&em_tree->lock);
5182 remove_extent_mapping(em_tree, em);
5183 write_unlock(&em_tree->lock);
5185 /* One for our allocation */
5186 free_extent_map(em);
5187 /* One for the tree reference */
5188 free_extent_map(em);
5190 kfree(devices_info);
5194 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5195 u64 chunk_offset, u64 chunk_size)
5197 struct btrfs_fs_info *fs_info = trans->fs_info;
5198 struct btrfs_root *extent_root = fs_info->extent_root;
5199 struct btrfs_root *chunk_root = fs_info->chunk_root;
5200 struct btrfs_key key;
5201 struct btrfs_device *device;
5202 struct btrfs_chunk *chunk;
5203 struct btrfs_stripe *stripe;
5204 struct extent_map *em;
5205 struct map_lookup *map;
5212 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5216 map = em->map_lookup;
5217 item_size = btrfs_chunk_item_size(map->num_stripes);
5218 stripe_size = em->orig_block_len;
5220 chunk = kzalloc(item_size, GFP_NOFS);
5227 * Take the device list mutex to prevent races with the final phase of
5228 * a device replace operation that replaces the device object associated
5229 * with the map's stripes, because the device object's id can change
5230 * at any time during that final phase of the device replace operation
5231 * (dev-replace.c:btrfs_dev_replace_finishing()).
5233 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5234 for (i = 0; i < map->num_stripes; i++) {
5235 device = map->stripes[i].dev;
5236 dev_offset = map->stripes[i].physical;
5238 ret = btrfs_update_device(trans, device);
5241 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5242 dev_offset, stripe_size);
5247 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5251 stripe = &chunk->stripe;
5252 for (i = 0; i < map->num_stripes; i++) {
5253 device = map->stripes[i].dev;
5254 dev_offset = map->stripes[i].physical;
5256 btrfs_set_stack_stripe_devid(stripe, device->devid);
5257 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5258 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5261 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5263 btrfs_set_stack_chunk_length(chunk, chunk_size);
5264 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5265 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5266 btrfs_set_stack_chunk_type(chunk, map->type);
5267 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5268 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5269 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5270 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5271 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5273 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5274 key.type = BTRFS_CHUNK_ITEM_KEY;
5275 key.offset = chunk_offset;
5277 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5278 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5280 * TODO: Cleanup of inserted chunk root in case of
5283 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5288 free_extent_map(em);
5293 * Chunk allocation falls into two parts. The first part does work
5294 * that makes the new allocated chunk usable, but does not do any operation
5295 * that modifies the chunk tree. The second part does the work that
5296 * requires modifying the chunk tree. This division is important for the
5297 * bootstrap process of adding storage to a seed btrfs.
5299 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5303 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5304 chunk_offset = find_next_chunk(trans->fs_info);
5305 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5308 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5309 struct btrfs_fs_info *fs_info)
5312 u64 sys_chunk_offset;
5316 chunk_offset = find_next_chunk(fs_info);
5317 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5318 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5322 sys_chunk_offset = find_next_chunk(fs_info);
5323 alloc_profile = btrfs_system_alloc_profile(fs_info);
5324 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5328 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5332 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5333 BTRFS_BLOCK_GROUP_RAID10 |
5334 BTRFS_BLOCK_GROUP_RAID5 |
5335 BTRFS_BLOCK_GROUP_DUP)) {
5337 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5346 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5348 struct extent_map *em;
5349 struct map_lookup *map;
5354 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5358 map = em->map_lookup;
5359 for (i = 0; i < map->num_stripes; i++) {
5360 if (test_bit(BTRFS_DEV_STATE_MISSING,
5361 &map->stripes[i].dev->dev_state)) {
5365 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5366 &map->stripes[i].dev->dev_state)) {
5373 * If the number of missing devices is larger than max errors,
5374 * we can not write the data into that chunk successfully, so
5377 if (miss_ndevs > btrfs_chunk_max_errors(map))
5380 free_extent_map(em);
5384 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5386 extent_map_tree_init(&tree->map_tree);
5389 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5391 struct extent_map *em;
5394 write_lock(&tree->map_tree.lock);
5395 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5397 remove_extent_mapping(&tree->map_tree, em);
5398 write_unlock(&tree->map_tree.lock);
5402 free_extent_map(em);
5403 /* once for the tree */
5404 free_extent_map(em);
5408 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5410 struct extent_map *em;
5411 struct map_lookup *map;
5414 em = btrfs_get_chunk_map(fs_info, logical, len);
5417 * We could return errors for these cases, but that could get
5418 * ugly and we'd probably do the same thing which is just not do
5419 * anything else and exit, so return 1 so the callers don't try
5420 * to use other copies.
5424 map = em->map_lookup;
5425 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5426 ret = map->num_stripes;
5427 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5428 ret = map->sub_stripes;
5429 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5431 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5433 * There could be two corrupted data stripes, we need
5434 * to loop retry in order to rebuild the correct data.
5436 * Fail a stripe at a time on every retry except the
5437 * stripe under reconstruction.
5439 ret = map->num_stripes;
5442 free_extent_map(em);
5444 down_read(&fs_info->dev_replace.rwsem);
5445 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5446 fs_info->dev_replace.tgtdev)
5448 up_read(&fs_info->dev_replace.rwsem);
5453 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5456 struct extent_map *em;
5457 struct map_lookup *map;
5458 unsigned long len = fs_info->sectorsize;
5460 em = btrfs_get_chunk_map(fs_info, logical, len);
5462 if (!WARN_ON(IS_ERR(em))) {
5463 map = em->map_lookup;
5464 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5465 len = map->stripe_len * nr_data_stripes(map);
5466 free_extent_map(em);
5471 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5473 struct extent_map *em;
5474 struct map_lookup *map;
5477 em = btrfs_get_chunk_map(fs_info, logical, len);
5479 if(!WARN_ON(IS_ERR(em))) {
5480 map = em->map_lookup;
5481 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5483 free_extent_map(em);
5488 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5489 struct map_lookup *map, int first,
5490 int dev_replace_is_ongoing)
5494 int preferred_mirror;
5496 struct btrfs_device *srcdev;
5499 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5501 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5502 num_stripes = map->sub_stripes;
5504 num_stripes = map->num_stripes;
5506 preferred_mirror = first + current->pid % num_stripes;
5508 if (dev_replace_is_ongoing &&
5509 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5510 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5511 srcdev = fs_info->dev_replace.srcdev;
5516 * try to avoid the drive that is the source drive for a
5517 * dev-replace procedure, only choose it if no other non-missing
5518 * mirror is available
5520 for (tolerance = 0; tolerance < 2; tolerance++) {
5521 if (map->stripes[preferred_mirror].dev->bdev &&
5522 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5523 return preferred_mirror;
5524 for (i = first; i < first + num_stripes; i++) {
5525 if (map->stripes[i].dev->bdev &&
5526 (tolerance || map->stripes[i].dev != srcdev))
5531 /* we couldn't find one that doesn't fail. Just return something
5532 * and the io error handling code will clean up eventually
5534 return preferred_mirror;
5537 static inline int parity_smaller(u64 a, u64 b)
5542 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5543 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5545 struct btrfs_bio_stripe s;
5552 for (i = 0; i < num_stripes - 1; i++) {
5553 if (parity_smaller(bbio->raid_map[i],
5554 bbio->raid_map[i+1])) {
5555 s = bbio->stripes[i];
5556 l = bbio->raid_map[i];
5557 bbio->stripes[i] = bbio->stripes[i+1];
5558 bbio->raid_map[i] = bbio->raid_map[i+1];
5559 bbio->stripes[i+1] = s;
5560 bbio->raid_map[i+1] = l;
5568 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5570 struct btrfs_bio *bbio = kzalloc(
5571 /* the size of the btrfs_bio */
5572 sizeof(struct btrfs_bio) +
5573 /* plus the variable array for the stripes */
5574 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5575 /* plus the variable array for the tgt dev */
5576 sizeof(int) * (real_stripes) +
5578 * plus the raid_map, which includes both the tgt dev
5581 sizeof(u64) * (total_stripes),
5582 GFP_NOFS|__GFP_NOFAIL);
5584 atomic_set(&bbio->error, 0);
5585 refcount_set(&bbio->refs, 1);
5590 void btrfs_get_bbio(struct btrfs_bio *bbio)
5592 WARN_ON(!refcount_read(&bbio->refs));
5593 refcount_inc(&bbio->refs);
5596 void btrfs_put_bbio(struct btrfs_bio *bbio)
5600 if (refcount_dec_and_test(&bbio->refs))
5604 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5606 * Please note that, discard won't be sent to target device of device
5609 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5610 u64 logical, u64 length,
5611 struct btrfs_bio **bbio_ret)
5613 struct extent_map *em;
5614 struct map_lookup *map;
5615 struct btrfs_bio *bbio;
5619 u64 stripe_end_offset;
5626 u32 sub_stripes = 0;
5627 u64 stripes_per_dev = 0;
5628 u32 remaining_stripes = 0;
5629 u32 last_stripe = 0;
5633 /* discard always return a bbio */
5636 em = btrfs_get_chunk_map(fs_info, logical, length);
5640 map = em->map_lookup;
5641 /* we don't discard raid56 yet */
5642 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5647 offset = logical - em->start;
5648 length = min_t(u64, em->len - offset, length);
5650 stripe_len = map->stripe_len;
5652 * stripe_nr counts the total number of stripes we have to stride
5653 * to get to this block
5655 stripe_nr = div64_u64(offset, stripe_len);
5657 /* stripe_offset is the offset of this block in its stripe */
5658 stripe_offset = offset - stripe_nr * stripe_len;
5660 stripe_nr_end = round_up(offset + length, map->stripe_len);
5661 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5662 stripe_cnt = stripe_nr_end - stripe_nr;
5663 stripe_end_offset = stripe_nr_end * map->stripe_len -
5666 * after this, stripe_nr is the number of stripes on this
5667 * device we have to walk to find the data, and stripe_index is
5668 * the number of our device in the stripe array
5672 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5673 BTRFS_BLOCK_GROUP_RAID10)) {
5674 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5677 sub_stripes = map->sub_stripes;
5679 factor = map->num_stripes / sub_stripes;
5680 num_stripes = min_t(u64, map->num_stripes,
5681 sub_stripes * stripe_cnt);
5682 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5683 stripe_index *= sub_stripes;
5684 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5685 &remaining_stripes);
5686 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5687 last_stripe *= sub_stripes;
5688 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5689 BTRFS_BLOCK_GROUP_DUP)) {
5690 num_stripes = map->num_stripes;
5692 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5696 bbio = alloc_btrfs_bio(num_stripes, 0);
5702 for (i = 0; i < num_stripes; i++) {
5703 bbio->stripes[i].physical =
5704 map->stripes[stripe_index].physical +
5705 stripe_offset + stripe_nr * map->stripe_len;
5706 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5708 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5709 BTRFS_BLOCK_GROUP_RAID10)) {
5710 bbio->stripes[i].length = stripes_per_dev *
5713 if (i / sub_stripes < remaining_stripes)
5714 bbio->stripes[i].length +=
5718 * Special for the first stripe and
5721 * |-------|...|-------|
5725 if (i < sub_stripes)
5726 bbio->stripes[i].length -=
5729 if (stripe_index >= last_stripe &&
5730 stripe_index <= (last_stripe +
5732 bbio->stripes[i].length -=
5735 if (i == sub_stripes - 1)
5738 bbio->stripes[i].length = length;
5742 if (stripe_index == map->num_stripes) {
5749 bbio->map_type = map->type;
5750 bbio->num_stripes = num_stripes;
5752 free_extent_map(em);
5757 * In dev-replace case, for repair case (that's the only case where the mirror
5758 * is selected explicitly when calling btrfs_map_block), blocks left of the
5759 * left cursor can also be read from the target drive.
5761 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5763 * For READ, it also needs to be supported using the same mirror number.
5765 * If the requested block is not left of the left cursor, EIO is returned. This
5766 * can happen because btrfs_num_copies() returns one more in the dev-replace
5769 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5770 u64 logical, u64 length,
5771 u64 srcdev_devid, int *mirror_num,
5774 struct btrfs_bio *bbio = NULL;
5776 int index_srcdev = 0;
5778 u64 physical_of_found = 0;
5782 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5783 logical, &length, &bbio, 0, 0);
5785 ASSERT(bbio == NULL);
5789 num_stripes = bbio->num_stripes;
5790 if (*mirror_num > num_stripes) {
5792 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5793 * that means that the requested area is not left of the left
5796 btrfs_put_bbio(bbio);
5801 * process the rest of the function using the mirror_num of the source
5802 * drive. Therefore look it up first. At the end, patch the device
5803 * pointer to the one of the target drive.
5805 for (i = 0; i < num_stripes; i++) {
5806 if (bbio->stripes[i].dev->devid != srcdev_devid)
5810 * In case of DUP, in order to keep it simple, only add the
5811 * mirror with the lowest physical address
5814 physical_of_found <= bbio->stripes[i].physical)
5819 physical_of_found = bbio->stripes[i].physical;
5822 btrfs_put_bbio(bbio);
5828 *mirror_num = index_srcdev + 1;
5829 *physical = physical_of_found;
5833 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5834 struct btrfs_bio **bbio_ret,
5835 struct btrfs_dev_replace *dev_replace,
5836 int *num_stripes_ret, int *max_errors_ret)
5838 struct btrfs_bio *bbio = *bbio_ret;
5839 u64 srcdev_devid = dev_replace->srcdev->devid;
5840 int tgtdev_indexes = 0;
5841 int num_stripes = *num_stripes_ret;
5842 int max_errors = *max_errors_ret;
5845 if (op == BTRFS_MAP_WRITE) {
5846 int index_where_to_add;
5849 * duplicate the write operations while the dev replace
5850 * procedure is running. Since the copying of the old disk to
5851 * the new disk takes place at run time while the filesystem is
5852 * mounted writable, the regular write operations to the old
5853 * disk have to be duplicated to go to the new disk as well.
5855 * Note that device->missing is handled by the caller, and that
5856 * the write to the old disk is already set up in the stripes
5859 index_where_to_add = num_stripes;
5860 for (i = 0; i < num_stripes; i++) {
5861 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5862 /* write to new disk, too */
5863 struct btrfs_bio_stripe *new =
5864 bbio->stripes + index_where_to_add;
5865 struct btrfs_bio_stripe *old =
5868 new->physical = old->physical;
5869 new->length = old->length;
5870 new->dev = dev_replace->tgtdev;
5871 bbio->tgtdev_map[i] = index_where_to_add;
5872 index_where_to_add++;
5877 num_stripes = index_where_to_add;
5878 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5879 int index_srcdev = 0;
5881 u64 physical_of_found = 0;
5884 * During the dev-replace procedure, the target drive can also
5885 * be used to read data in case it is needed to repair a corrupt
5886 * block elsewhere. This is possible if the requested area is
5887 * left of the left cursor. In this area, the target drive is a
5888 * full copy of the source drive.
5890 for (i = 0; i < num_stripes; i++) {
5891 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5893 * In case of DUP, in order to keep it simple,
5894 * only add the mirror with the lowest physical
5898 physical_of_found <=
5899 bbio->stripes[i].physical)
5903 physical_of_found = bbio->stripes[i].physical;
5907 struct btrfs_bio_stripe *tgtdev_stripe =
5908 bbio->stripes + num_stripes;
5910 tgtdev_stripe->physical = physical_of_found;
5911 tgtdev_stripe->length =
5912 bbio->stripes[index_srcdev].length;
5913 tgtdev_stripe->dev = dev_replace->tgtdev;
5914 bbio->tgtdev_map[index_srcdev] = num_stripes;
5921 *num_stripes_ret = num_stripes;
5922 *max_errors_ret = max_errors;
5923 bbio->num_tgtdevs = tgtdev_indexes;
5927 static bool need_full_stripe(enum btrfs_map_op op)
5929 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5932 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5933 enum btrfs_map_op op,
5934 u64 logical, u64 *length,
5935 struct btrfs_bio **bbio_ret,
5936 int mirror_num, int need_raid_map)
5938 struct extent_map *em;
5939 struct map_lookup *map;
5949 int tgtdev_indexes = 0;
5950 struct btrfs_bio *bbio = NULL;
5951 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5952 int dev_replace_is_ongoing = 0;
5953 int num_alloc_stripes;
5954 int patch_the_first_stripe_for_dev_replace = 0;
5955 u64 physical_to_patch_in_first_stripe = 0;
5956 u64 raid56_full_stripe_start = (u64)-1;
5958 if (op == BTRFS_MAP_DISCARD)
5959 return __btrfs_map_block_for_discard(fs_info, logical,
5962 em = btrfs_get_chunk_map(fs_info, logical, *length);
5966 map = em->map_lookup;
5967 offset = logical - em->start;
5969 stripe_len = map->stripe_len;
5972 * stripe_nr counts the total number of stripes we have to stride
5973 * to get to this block
5975 stripe_nr = div64_u64(stripe_nr, stripe_len);
5977 stripe_offset = stripe_nr * stripe_len;
5978 if (offset < stripe_offset) {
5980 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5981 stripe_offset, offset, em->start, logical,
5983 free_extent_map(em);
5987 /* stripe_offset is the offset of this block in its stripe*/
5988 stripe_offset = offset - stripe_offset;
5990 /* if we're here for raid56, we need to know the stripe aligned start */
5991 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5992 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5993 raid56_full_stripe_start = offset;
5995 /* allow a write of a full stripe, but make sure we don't
5996 * allow straddling of stripes
5998 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6000 raid56_full_stripe_start *= full_stripe_len;
6003 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6005 /* For writes to RAID[56], allow a full stripeset across all disks.
6006 For other RAID types and for RAID[56] reads, just allow a single
6007 stripe (on a single disk). */
6008 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6009 (op == BTRFS_MAP_WRITE)) {
6010 max_len = stripe_len * nr_data_stripes(map) -
6011 (offset - raid56_full_stripe_start);
6013 /* we limit the length of each bio to what fits in a stripe */
6014 max_len = stripe_len - stripe_offset;
6016 *length = min_t(u64, em->len - offset, max_len);
6018 *length = em->len - offset;
6022 * This is for when we're called from btrfs_bio_fits_in_stripe and all
6023 * it cares about is the length
6028 down_read(&dev_replace->rwsem);
6029 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6031 * Hold the semaphore for read during the whole operation, write is
6032 * requested at commit time but must wait.
6034 if (!dev_replace_is_ongoing)
6035 up_read(&dev_replace->rwsem);
6037 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6038 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6039 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6040 dev_replace->srcdev->devid,
6042 &physical_to_patch_in_first_stripe);
6046 patch_the_first_stripe_for_dev_replace = 1;
6047 } else if (mirror_num > map->num_stripes) {
6053 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6054 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6056 if (!need_full_stripe(op))
6058 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
6059 if (need_full_stripe(op))
6060 num_stripes = map->num_stripes;
6061 else if (mirror_num)
6062 stripe_index = mirror_num - 1;
6064 stripe_index = find_live_mirror(fs_info, map, 0,
6065 dev_replace_is_ongoing);
6066 mirror_num = stripe_index + 1;
6069 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6070 if (need_full_stripe(op)) {
6071 num_stripes = map->num_stripes;
6072 } else if (mirror_num) {
6073 stripe_index = mirror_num - 1;
6078 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6079 u32 factor = map->num_stripes / map->sub_stripes;
6081 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6082 stripe_index *= map->sub_stripes;
6084 if (need_full_stripe(op))
6085 num_stripes = map->sub_stripes;
6086 else if (mirror_num)
6087 stripe_index += mirror_num - 1;
6089 int old_stripe_index = stripe_index;
6090 stripe_index = find_live_mirror(fs_info, map,
6092 dev_replace_is_ongoing);
6093 mirror_num = stripe_index - old_stripe_index + 1;
6096 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6097 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6098 /* push stripe_nr back to the start of the full stripe */
6099 stripe_nr = div64_u64(raid56_full_stripe_start,
6100 stripe_len * nr_data_stripes(map));
6102 /* RAID[56] write or recovery. Return all stripes */
6103 num_stripes = map->num_stripes;
6104 max_errors = nr_parity_stripes(map);
6106 *length = map->stripe_len;
6111 * Mirror #0 or #1 means the original data block.
6112 * Mirror #2 is RAID5 parity block.
6113 * Mirror #3 is RAID6 Q block.
6115 stripe_nr = div_u64_rem(stripe_nr,
6116 nr_data_stripes(map), &stripe_index);
6118 stripe_index = nr_data_stripes(map) +
6121 /* We distribute the parity blocks across stripes */
6122 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6124 if (!need_full_stripe(op) && mirror_num <= 1)
6129 * after this, stripe_nr is the number of stripes on this
6130 * device we have to walk to find the data, and stripe_index is
6131 * the number of our device in the stripe array
6133 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6135 mirror_num = stripe_index + 1;
6137 if (stripe_index >= map->num_stripes) {
6139 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6140 stripe_index, map->num_stripes);
6145 num_alloc_stripes = num_stripes;
6146 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6147 if (op == BTRFS_MAP_WRITE)
6148 num_alloc_stripes <<= 1;
6149 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6150 num_alloc_stripes++;
6151 tgtdev_indexes = num_stripes;
6154 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6159 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6160 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6162 /* build raid_map */
6163 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6164 (need_full_stripe(op) || mirror_num > 1)) {
6168 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6169 sizeof(struct btrfs_bio_stripe) *
6171 sizeof(int) * tgtdev_indexes);
6173 /* Work out the disk rotation on this stripe-set */
6174 div_u64_rem(stripe_nr, num_stripes, &rot);
6176 /* Fill in the logical address of each stripe */
6177 tmp = stripe_nr * nr_data_stripes(map);
6178 for (i = 0; i < nr_data_stripes(map); i++)
6179 bbio->raid_map[(i+rot) % num_stripes] =
6180 em->start + (tmp + i) * map->stripe_len;
6182 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6183 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6184 bbio->raid_map[(i+rot+1) % num_stripes] =
6189 for (i = 0; i < num_stripes; i++) {
6190 bbio->stripes[i].physical =
6191 map->stripes[stripe_index].physical +
6193 stripe_nr * map->stripe_len;
6194 bbio->stripes[i].dev =
6195 map->stripes[stripe_index].dev;
6199 if (need_full_stripe(op))
6200 max_errors = btrfs_chunk_max_errors(map);
6203 sort_parity_stripes(bbio, num_stripes);
6205 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6206 need_full_stripe(op)) {
6207 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6212 bbio->map_type = map->type;
6213 bbio->num_stripes = num_stripes;
6214 bbio->max_errors = max_errors;
6215 bbio->mirror_num = mirror_num;
6218 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6219 * mirror_num == num_stripes + 1 && dev_replace target drive is
6220 * available as a mirror
6222 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6223 WARN_ON(num_stripes > 1);
6224 bbio->stripes[0].dev = dev_replace->tgtdev;
6225 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6226 bbio->mirror_num = map->num_stripes + 1;
6229 if (dev_replace_is_ongoing) {
6230 lockdep_assert_held(&dev_replace->rwsem);
6231 /* Unlock and let waiting writers proceed */
6232 up_read(&dev_replace->rwsem);
6234 free_extent_map(em);
6238 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6239 u64 logical, u64 *length,
6240 struct btrfs_bio **bbio_ret, int mirror_num)
6242 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6246 /* For Scrub/replace */
6247 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6248 u64 logical, u64 *length,
6249 struct btrfs_bio **bbio_ret)
6251 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6254 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6255 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6257 struct extent_map *em;
6258 struct map_lookup *map;
6266 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6270 map = em->map_lookup;
6272 rmap_len = map->stripe_len;
6274 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6275 length = div_u64(length, map->num_stripes / map->sub_stripes);
6276 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6277 length = div_u64(length, map->num_stripes);
6278 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6279 length = div_u64(length, nr_data_stripes(map));
6280 rmap_len = map->stripe_len * nr_data_stripes(map);
6283 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6284 BUG_ON(!buf); /* -ENOMEM */
6286 for (i = 0; i < map->num_stripes; i++) {
6287 if (map->stripes[i].physical > physical ||
6288 map->stripes[i].physical + length <= physical)
6291 stripe_nr = physical - map->stripes[i].physical;
6292 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6294 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6295 stripe_nr = stripe_nr * map->num_stripes + i;
6296 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6297 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6298 stripe_nr = stripe_nr * map->num_stripes + i;
6299 } /* else if RAID[56], multiply by nr_data_stripes().
6300 * Alternatively, just use rmap_len below instead of
6301 * map->stripe_len */
6303 bytenr = chunk_start + stripe_nr * rmap_len;
6304 WARN_ON(nr >= map->num_stripes);
6305 for (j = 0; j < nr; j++) {
6306 if (buf[j] == bytenr)
6310 WARN_ON(nr >= map->num_stripes);
6317 *stripe_len = rmap_len;
6319 free_extent_map(em);
6323 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6325 bio->bi_private = bbio->private;
6326 bio->bi_end_io = bbio->end_io;
6329 btrfs_put_bbio(bbio);
6332 static void btrfs_end_bio(struct bio *bio)
6334 struct btrfs_bio *bbio = bio->bi_private;
6335 int is_orig_bio = 0;
6337 if (bio->bi_status) {
6338 atomic_inc(&bbio->error);
6339 if (bio->bi_status == BLK_STS_IOERR ||
6340 bio->bi_status == BLK_STS_TARGET) {
6341 unsigned int stripe_index =
6342 btrfs_io_bio(bio)->stripe_index;
6343 struct btrfs_device *dev;
6345 BUG_ON(stripe_index >= bbio->num_stripes);
6346 dev = bbio->stripes[stripe_index].dev;
6348 if (bio_op(bio) == REQ_OP_WRITE)
6349 btrfs_dev_stat_inc_and_print(dev,
6350 BTRFS_DEV_STAT_WRITE_ERRS);
6351 else if (!(bio->bi_opf & REQ_RAHEAD))
6352 btrfs_dev_stat_inc_and_print(dev,
6353 BTRFS_DEV_STAT_READ_ERRS);
6354 if (bio->bi_opf & REQ_PREFLUSH)
6355 btrfs_dev_stat_inc_and_print(dev,
6356 BTRFS_DEV_STAT_FLUSH_ERRS);
6361 if (bio == bbio->orig_bio)
6364 btrfs_bio_counter_dec(bbio->fs_info);
6366 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6369 bio = bbio->orig_bio;
6372 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6373 /* only send an error to the higher layers if it is
6374 * beyond the tolerance of the btrfs bio
6376 if (atomic_read(&bbio->error) > bbio->max_errors) {
6377 bio->bi_status = BLK_STS_IOERR;
6380 * this bio is actually up to date, we didn't
6381 * go over the max number of errors
6383 bio->bi_status = BLK_STS_OK;
6386 btrfs_end_bbio(bbio, bio);
6387 } else if (!is_orig_bio) {
6393 * see run_scheduled_bios for a description of why bios are collected for
6396 * This will add one bio to the pending list for a device and make sure
6397 * the work struct is scheduled.
6399 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6402 struct btrfs_fs_info *fs_info = device->fs_info;
6403 int should_queue = 1;
6404 struct btrfs_pending_bios *pending_bios;
6406 /* don't bother with additional async steps for reads, right now */
6407 if (bio_op(bio) == REQ_OP_READ) {
6408 btrfsic_submit_bio(bio);
6412 WARN_ON(bio->bi_next);
6413 bio->bi_next = NULL;
6415 spin_lock(&device->io_lock);
6416 if (op_is_sync(bio->bi_opf))
6417 pending_bios = &device->pending_sync_bios;
6419 pending_bios = &device->pending_bios;
6421 if (pending_bios->tail)
6422 pending_bios->tail->bi_next = bio;
6424 pending_bios->tail = bio;
6425 if (!pending_bios->head)
6426 pending_bios->head = bio;
6427 if (device->running_pending)
6430 spin_unlock(&device->io_lock);
6433 btrfs_queue_work(fs_info->submit_workers, &device->work);
6436 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6437 u64 physical, int dev_nr, int async)
6439 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6440 struct btrfs_fs_info *fs_info = bbio->fs_info;
6442 bio->bi_private = bbio;
6443 btrfs_io_bio(bio)->stripe_index = dev_nr;
6444 bio->bi_end_io = btrfs_end_bio;
6445 bio->bi_iter.bi_sector = physical >> 9;
6446 btrfs_debug_in_rcu(fs_info,
6447 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6448 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6449 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6450 bio->bi_iter.bi_size);
6451 bio_set_dev(bio, dev->bdev);
6453 btrfs_bio_counter_inc_noblocked(fs_info);
6456 btrfs_schedule_bio(dev, bio);
6458 btrfsic_submit_bio(bio);
6461 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6463 atomic_inc(&bbio->error);
6464 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6465 /* Should be the original bio. */
6466 WARN_ON(bio != bbio->orig_bio);
6468 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6469 bio->bi_iter.bi_sector = logical >> 9;
6470 if (atomic_read(&bbio->error) > bbio->max_errors)
6471 bio->bi_status = BLK_STS_IOERR;
6473 bio->bi_status = BLK_STS_OK;
6474 btrfs_end_bbio(bbio, bio);
6478 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6479 int mirror_num, int async_submit)
6481 struct btrfs_device *dev;
6482 struct bio *first_bio = bio;
6483 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6489 struct btrfs_bio *bbio = NULL;
6491 length = bio->bi_iter.bi_size;
6492 map_length = length;
6494 btrfs_bio_counter_inc_blocked(fs_info);
6495 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6496 &map_length, &bbio, mirror_num, 1);
6498 btrfs_bio_counter_dec(fs_info);
6499 return errno_to_blk_status(ret);
6502 total_devs = bbio->num_stripes;
6503 bbio->orig_bio = first_bio;
6504 bbio->private = first_bio->bi_private;
6505 bbio->end_io = first_bio->bi_end_io;
6506 bbio->fs_info = fs_info;
6507 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6509 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6510 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6511 /* In this case, map_length has been set to the length of
6512 a single stripe; not the whole write */
6513 if (bio_op(bio) == REQ_OP_WRITE) {
6514 ret = raid56_parity_write(fs_info, bio, bbio,
6517 ret = raid56_parity_recover(fs_info, bio, bbio,
6518 map_length, mirror_num, 1);
6521 btrfs_bio_counter_dec(fs_info);
6522 return errno_to_blk_status(ret);
6525 if (map_length < length) {
6527 "mapping failed logical %llu bio len %llu len %llu",
6528 logical, length, map_length);
6532 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6533 dev = bbio->stripes[dev_nr].dev;
6534 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6536 (bio_op(first_bio) == REQ_OP_WRITE &&
6537 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6538 bbio_error(bbio, first_bio, logical);
6542 if (dev_nr < total_devs - 1)
6543 bio = btrfs_bio_clone(first_bio);
6547 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6548 dev_nr, async_submit);
6550 btrfs_bio_counter_dec(fs_info);
6555 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6558 * If devid and uuid are both specified, the match must be exact, otherwise
6559 * only devid is used.
6561 * If @seed is true, traverse through the seed devices.
6563 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6564 u64 devid, u8 *uuid, u8 *fsid,
6567 struct btrfs_device *device;
6569 while (fs_devices) {
6571 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6572 list_for_each_entry(device, &fs_devices->devices,
6574 if (device->devid == devid &&
6575 (!uuid || memcmp(device->uuid, uuid,
6576 BTRFS_UUID_SIZE) == 0))
6581 fs_devices = fs_devices->seed;
6588 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6589 u64 devid, u8 *dev_uuid)
6591 struct btrfs_device *device;
6593 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6597 list_add(&device->dev_list, &fs_devices->devices);
6598 device->fs_devices = fs_devices;
6599 fs_devices->num_devices++;
6601 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6602 fs_devices->missing_devices++;
6608 * btrfs_alloc_device - allocate struct btrfs_device
6609 * @fs_info: used only for generating a new devid, can be NULL if
6610 * devid is provided (i.e. @devid != NULL).
6611 * @devid: a pointer to devid for this device. If NULL a new devid
6613 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6616 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6617 * on error. Returned struct is not linked onto any lists and must be
6618 * destroyed with btrfs_free_device.
6620 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6624 struct btrfs_device *dev;
6627 if (WARN_ON(!devid && !fs_info))
6628 return ERR_PTR(-EINVAL);
6630 dev = __alloc_device();
6639 ret = find_next_devid(fs_info, &tmp);
6641 btrfs_free_device(dev);
6642 return ERR_PTR(ret);
6648 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6650 generate_random_uuid(dev->uuid);
6652 btrfs_init_work(&dev->work, btrfs_submit_helper,
6653 pending_bios_fn, NULL, NULL);
6658 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6659 u64 devid, u8 *uuid, bool error)
6662 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6665 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6669 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6671 int index = btrfs_bg_flags_to_raid_index(type);
6672 int ncopies = btrfs_raid_array[index].ncopies;
6675 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6676 case BTRFS_BLOCK_GROUP_RAID5:
6677 data_stripes = num_stripes - 1;
6679 case BTRFS_BLOCK_GROUP_RAID6:
6680 data_stripes = num_stripes - 2;
6683 data_stripes = num_stripes / ncopies;
6686 return div_u64(chunk_len, data_stripes);
6689 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6690 struct btrfs_chunk *chunk)
6692 struct btrfs_fs_info *fs_info = leaf->fs_info;
6693 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6694 struct map_lookup *map;
6695 struct extent_map *em;
6699 u8 uuid[BTRFS_UUID_SIZE];
6704 logical = key->offset;
6705 length = btrfs_chunk_length(leaf, chunk);
6706 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6709 * Only need to verify chunk item if we're reading from sys chunk array,
6710 * as chunk item in tree block is already verified by tree-checker.
6712 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6713 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6718 read_lock(&map_tree->map_tree.lock);
6719 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6720 read_unlock(&map_tree->map_tree.lock);
6722 /* already mapped? */
6723 if (em && em->start <= logical && em->start + em->len > logical) {
6724 free_extent_map(em);
6727 free_extent_map(em);
6730 em = alloc_extent_map();
6733 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6735 free_extent_map(em);
6739 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6740 em->map_lookup = map;
6741 em->start = logical;
6744 em->block_start = 0;
6745 em->block_len = em->len;
6747 map->num_stripes = num_stripes;
6748 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6749 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6750 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6751 map->type = btrfs_chunk_type(leaf, chunk);
6752 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6753 map->verified_stripes = 0;
6754 em->orig_block_len = calc_stripe_length(map->type, em->len,
6756 for (i = 0; i < num_stripes; i++) {
6757 map->stripes[i].physical =
6758 btrfs_stripe_offset_nr(leaf, chunk, i);
6759 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6760 read_extent_buffer(leaf, uuid, (unsigned long)
6761 btrfs_stripe_dev_uuid_nr(chunk, i),
6763 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6764 devid, uuid, NULL, true);
6765 if (!map->stripes[i].dev &&
6766 !btrfs_test_opt(fs_info, DEGRADED)) {
6767 free_extent_map(em);
6768 btrfs_report_missing_device(fs_info, devid, uuid, true);
6771 if (!map->stripes[i].dev) {
6772 map->stripes[i].dev =
6773 add_missing_dev(fs_info->fs_devices, devid,
6775 if (IS_ERR(map->stripes[i].dev)) {
6776 free_extent_map(em);
6778 "failed to init missing dev %llu: %ld",
6779 devid, PTR_ERR(map->stripes[i].dev));
6780 return PTR_ERR(map->stripes[i].dev);
6782 btrfs_report_missing_device(fs_info, devid, uuid, false);
6784 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6785 &(map->stripes[i].dev->dev_state));
6789 write_lock(&map_tree->map_tree.lock);
6790 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6791 write_unlock(&map_tree->map_tree.lock);
6794 "failed to add chunk map, start=%llu len=%llu: %d",
6795 em->start, em->len, ret);
6797 free_extent_map(em);
6802 static void fill_device_from_item(struct extent_buffer *leaf,
6803 struct btrfs_dev_item *dev_item,
6804 struct btrfs_device *device)
6808 device->devid = btrfs_device_id(leaf, dev_item);
6809 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6810 device->total_bytes = device->disk_total_bytes;
6811 device->commit_total_bytes = device->disk_total_bytes;
6812 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6813 device->commit_bytes_used = device->bytes_used;
6814 device->type = btrfs_device_type(leaf, dev_item);
6815 device->io_align = btrfs_device_io_align(leaf, dev_item);
6816 device->io_width = btrfs_device_io_width(leaf, dev_item);
6817 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6818 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6819 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6821 ptr = btrfs_device_uuid(dev_item);
6822 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6825 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6828 struct btrfs_fs_devices *fs_devices;
6831 lockdep_assert_held(&uuid_mutex);
6834 fs_devices = fs_info->fs_devices->seed;
6835 while (fs_devices) {
6836 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6839 fs_devices = fs_devices->seed;
6842 fs_devices = find_fsid(fsid, NULL);
6844 if (!btrfs_test_opt(fs_info, DEGRADED))
6845 return ERR_PTR(-ENOENT);
6847 fs_devices = alloc_fs_devices(fsid, NULL);
6848 if (IS_ERR(fs_devices))
6851 fs_devices->seeding = 1;
6852 fs_devices->opened = 1;
6856 fs_devices = clone_fs_devices(fs_devices);
6857 if (IS_ERR(fs_devices))
6860 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6862 free_fs_devices(fs_devices);
6863 fs_devices = ERR_PTR(ret);
6867 if (!fs_devices->seeding) {
6868 close_fs_devices(fs_devices);
6869 free_fs_devices(fs_devices);
6870 fs_devices = ERR_PTR(-EINVAL);
6874 fs_devices->seed = fs_info->fs_devices->seed;
6875 fs_info->fs_devices->seed = fs_devices;
6880 static int read_one_dev(struct btrfs_fs_info *fs_info,
6881 struct extent_buffer *leaf,
6882 struct btrfs_dev_item *dev_item)
6884 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6885 struct btrfs_device *device;
6888 u8 fs_uuid[BTRFS_FSID_SIZE];
6889 u8 dev_uuid[BTRFS_UUID_SIZE];
6891 devid = btrfs_device_id(leaf, dev_item);
6892 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6894 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6897 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6898 fs_devices = open_seed_devices(fs_info, fs_uuid);
6899 if (IS_ERR(fs_devices))
6900 return PTR_ERR(fs_devices);
6903 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6906 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6907 btrfs_report_missing_device(fs_info, devid,
6912 device = add_missing_dev(fs_devices, devid, dev_uuid);
6913 if (IS_ERR(device)) {
6915 "failed to add missing dev %llu: %ld",
6916 devid, PTR_ERR(device));
6917 return PTR_ERR(device);
6919 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6921 if (!device->bdev) {
6922 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6923 btrfs_report_missing_device(fs_info,
6924 devid, dev_uuid, true);
6927 btrfs_report_missing_device(fs_info, devid,
6931 if (!device->bdev &&
6932 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6934 * this happens when a device that was properly setup
6935 * in the device info lists suddenly goes bad.
6936 * device->bdev is NULL, and so we have to set
6937 * device->missing to one here
6939 device->fs_devices->missing_devices++;
6940 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6943 /* Move the device to its own fs_devices */
6944 if (device->fs_devices != fs_devices) {
6945 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6946 &device->dev_state));
6948 list_move(&device->dev_list, &fs_devices->devices);
6949 device->fs_devices->num_devices--;
6950 fs_devices->num_devices++;
6952 device->fs_devices->missing_devices--;
6953 fs_devices->missing_devices++;
6955 device->fs_devices = fs_devices;
6959 if (device->fs_devices != fs_info->fs_devices) {
6960 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6961 if (device->generation !=
6962 btrfs_device_generation(leaf, dev_item))
6966 fill_device_from_item(leaf, dev_item, device);
6967 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6968 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6969 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6970 device->fs_devices->total_rw_bytes += device->total_bytes;
6971 atomic64_add(device->total_bytes - device->bytes_used,
6972 &fs_info->free_chunk_space);
6978 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6980 struct btrfs_root *root = fs_info->tree_root;
6981 struct btrfs_super_block *super_copy = fs_info->super_copy;
6982 struct extent_buffer *sb;
6983 struct btrfs_disk_key *disk_key;
6984 struct btrfs_chunk *chunk;
6986 unsigned long sb_array_offset;
6993 struct btrfs_key key;
6995 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6997 * This will create extent buffer of nodesize, superblock size is
6998 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6999 * overallocate but we can keep it as-is, only the first page is used.
7001 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7004 set_extent_buffer_uptodate(sb);
7005 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7007 * The sb extent buffer is artificial and just used to read the system array.
7008 * set_extent_buffer_uptodate() call does not properly mark all it's
7009 * pages up-to-date when the page is larger: extent does not cover the
7010 * whole page and consequently check_page_uptodate does not find all
7011 * the page's extents up-to-date (the hole beyond sb),
7012 * write_extent_buffer then triggers a WARN_ON.
7014 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7015 * but sb spans only this function. Add an explicit SetPageUptodate call
7016 * to silence the warning eg. on PowerPC 64.
7018 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7019 SetPageUptodate(sb->pages[0]);
7021 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7022 array_size = btrfs_super_sys_array_size(super_copy);
7024 array_ptr = super_copy->sys_chunk_array;
7025 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7028 while (cur_offset < array_size) {
7029 disk_key = (struct btrfs_disk_key *)array_ptr;
7030 len = sizeof(*disk_key);
7031 if (cur_offset + len > array_size)
7032 goto out_short_read;
7034 btrfs_disk_key_to_cpu(&key, disk_key);
7037 sb_array_offset += len;
7040 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7041 chunk = (struct btrfs_chunk *)sb_array_offset;
7043 * At least one btrfs_chunk with one stripe must be
7044 * present, exact stripe count check comes afterwards
7046 len = btrfs_chunk_item_size(1);
7047 if (cur_offset + len > array_size)
7048 goto out_short_read;
7050 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7053 "invalid number of stripes %u in sys_array at offset %u",
7054 num_stripes, cur_offset);
7059 type = btrfs_chunk_type(sb, chunk);
7060 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7062 "invalid chunk type %llu in sys_array at offset %u",
7068 len = btrfs_chunk_item_size(num_stripes);
7069 if (cur_offset + len > array_size)
7070 goto out_short_read;
7072 ret = read_one_chunk(&key, sb, chunk);
7077 "unexpected item type %u in sys_array at offset %u",
7078 (u32)key.type, cur_offset);
7083 sb_array_offset += len;
7086 clear_extent_buffer_uptodate(sb);
7087 free_extent_buffer_stale(sb);
7091 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7093 clear_extent_buffer_uptodate(sb);
7094 free_extent_buffer_stale(sb);
7099 * Check if all chunks in the fs are OK for read-write degraded mount
7101 * If the @failing_dev is specified, it's accounted as missing.
7103 * Return true if all chunks meet the minimal RW mount requirements.
7104 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7106 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7107 struct btrfs_device *failing_dev)
7109 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
7110 struct extent_map *em;
7114 read_lock(&map_tree->map_tree.lock);
7115 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
7116 read_unlock(&map_tree->map_tree.lock);
7117 /* No chunk at all? Return false anyway */
7123 struct map_lookup *map;
7128 map = em->map_lookup;
7130 btrfs_get_num_tolerated_disk_barrier_failures(
7132 for (i = 0; i < map->num_stripes; i++) {
7133 struct btrfs_device *dev = map->stripes[i].dev;
7135 if (!dev || !dev->bdev ||
7136 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7137 dev->last_flush_error)
7139 else if (failing_dev && failing_dev == dev)
7142 if (missing > max_tolerated) {
7145 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7146 em->start, missing, max_tolerated);
7147 free_extent_map(em);
7151 next_start = extent_map_end(em);
7152 free_extent_map(em);
7154 read_lock(&map_tree->map_tree.lock);
7155 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
7156 (u64)(-1) - next_start);
7157 read_unlock(&map_tree->map_tree.lock);
7163 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7165 struct btrfs_root *root = fs_info->chunk_root;
7166 struct btrfs_path *path;
7167 struct extent_buffer *leaf;
7168 struct btrfs_key key;
7169 struct btrfs_key found_key;
7174 path = btrfs_alloc_path();
7179 * uuid_mutex is needed only if we are mounting a sprout FS
7180 * otherwise we don't need it.
7182 mutex_lock(&uuid_mutex);
7183 mutex_lock(&fs_info->chunk_mutex);
7186 * Read all device items, and then all the chunk items. All
7187 * device items are found before any chunk item (their object id
7188 * is smaller than the lowest possible object id for a chunk
7189 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7191 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7194 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7198 leaf = path->nodes[0];
7199 slot = path->slots[0];
7200 if (slot >= btrfs_header_nritems(leaf)) {
7201 ret = btrfs_next_leaf(root, path);
7208 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7209 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7210 struct btrfs_dev_item *dev_item;
7211 dev_item = btrfs_item_ptr(leaf, slot,
7212 struct btrfs_dev_item);
7213 ret = read_one_dev(fs_info, leaf, dev_item);
7217 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7218 struct btrfs_chunk *chunk;
7219 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7220 ret = read_one_chunk(&found_key, leaf, chunk);
7228 * After loading chunk tree, we've got all device information,
7229 * do another round of validation checks.
7231 if (total_dev != fs_info->fs_devices->total_devices) {
7233 "super_num_devices %llu mismatch with num_devices %llu found here",
7234 btrfs_super_num_devices(fs_info->super_copy),
7239 if (btrfs_super_total_bytes(fs_info->super_copy) <
7240 fs_info->fs_devices->total_rw_bytes) {
7242 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7243 btrfs_super_total_bytes(fs_info->super_copy),
7244 fs_info->fs_devices->total_rw_bytes);
7250 mutex_unlock(&fs_info->chunk_mutex);
7251 mutex_unlock(&uuid_mutex);
7253 btrfs_free_path(path);
7257 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7259 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7260 struct btrfs_device *device;
7262 while (fs_devices) {
7263 mutex_lock(&fs_devices->device_list_mutex);
7264 list_for_each_entry(device, &fs_devices->devices, dev_list)
7265 device->fs_info = fs_info;
7266 mutex_unlock(&fs_devices->device_list_mutex);
7268 fs_devices = fs_devices->seed;
7272 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7276 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7277 btrfs_dev_stat_reset(dev, i);
7280 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7282 struct btrfs_key key;
7283 struct btrfs_key found_key;
7284 struct btrfs_root *dev_root = fs_info->dev_root;
7285 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7286 struct extent_buffer *eb;
7289 struct btrfs_device *device;
7290 struct btrfs_path *path = NULL;
7293 path = btrfs_alloc_path();
7299 mutex_lock(&fs_devices->device_list_mutex);
7300 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7302 struct btrfs_dev_stats_item *ptr;
7304 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7305 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7306 key.offset = device->devid;
7307 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7309 __btrfs_reset_dev_stats(device);
7310 device->dev_stats_valid = 1;
7311 btrfs_release_path(path);
7314 slot = path->slots[0];
7315 eb = path->nodes[0];
7316 btrfs_item_key_to_cpu(eb, &found_key, slot);
7317 item_size = btrfs_item_size_nr(eb, slot);
7319 ptr = btrfs_item_ptr(eb, slot,
7320 struct btrfs_dev_stats_item);
7322 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7323 if (item_size >= (1 + i) * sizeof(__le64))
7324 btrfs_dev_stat_set(device, i,
7325 btrfs_dev_stats_value(eb, ptr, i));
7327 btrfs_dev_stat_reset(device, i);
7330 device->dev_stats_valid = 1;
7331 btrfs_dev_stat_print_on_load(device);
7332 btrfs_release_path(path);
7334 mutex_unlock(&fs_devices->device_list_mutex);
7337 btrfs_free_path(path);
7338 return ret < 0 ? ret : 0;
7341 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7342 struct btrfs_device *device)
7344 struct btrfs_fs_info *fs_info = trans->fs_info;
7345 struct btrfs_root *dev_root = fs_info->dev_root;
7346 struct btrfs_path *path;
7347 struct btrfs_key key;
7348 struct extent_buffer *eb;
7349 struct btrfs_dev_stats_item *ptr;
7353 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7354 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7355 key.offset = device->devid;
7357 path = btrfs_alloc_path();
7360 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7362 btrfs_warn_in_rcu(fs_info,
7363 "error %d while searching for dev_stats item for device %s",
7364 ret, rcu_str_deref(device->name));
7369 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7370 /* need to delete old one and insert a new one */
7371 ret = btrfs_del_item(trans, dev_root, path);
7373 btrfs_warn_in_rcu(fs_info,
7374 "delete too small dev_stats item for device %s failed %d",
7375 rcu_str_deref(device->name), ret);
7382 /* need to insert a new item */
7383 btrfs_release_path(path);
7384 ret = btrfs_insert_empty_item(trans, dev_root, path,
7385 &key, sizeof(*ptr));
7387 btrfs_warn_in_rcu(fs_info,
7388 "insert dev_stats item for device %s failed %d",
7389 rcu_str_deref(device->name), ret);
7394 eb = path->nodes[0];
7395 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7396 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7397 btrfs_set_dev_stats_value(eb, ptr, i,
7398 btrfs_dev_stat_read(device, i));
7399 btrfs_mark_buffer_dirty(eb);
7402 btrfs_free_path(path);
7407 * called from commit_transaction. Writes all changed device stats to disk.
7409 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7410 struct btrfs_fs_info *fs_info)
7412 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7413 struct btrfs_device *device;
7417 mutex_lock(&fs_devices->device_list_mutex);
7418 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7419 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7420 if (!device->dev_stats_valid || stats_cnt == 0)
7425 * There is a LOAD-LOAD control dependency between the value of
7426 * dev_stats_ccnt and updating the on-disk values which requires
7427 * reading the in-memory counters. Such control dependencies
7428 * require explicit read memory barriers.
7430 * This memory barriers pairs with smp_mb__before_atomic in
7431 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7432 * barrier implied by atomic_xchg in
7433 * btrfs_dev_stats_read_and_reset
7437 ret = update_dev_stat_item(trans, device);
7439 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7441 mutex_unlock(&fs_devices->device_list_mutex);
7446 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7448 btrfs_dev_stat_inc(dev, index);
7449 btrfs_dev_stat_print_on_error(dev);
7452 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7454 if (!dev->dev_stats_valid)
7456 btrfs_err_rl_in_rcu(dev->fs_info,
7457 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7458 rcu_str_deref(dev->name),
7459 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7460 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7461 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7462 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7463 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7466 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7470 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7471 if (btrfs_dev_stat_read(dev, i) != 0)
7473 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7474 return; /* all values == 0, suppress message */
7476 btrfs_info_in_rcu(dev->fs_info,
7477 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7478 rcu_str_deref(dev->name),
7479 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7480 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7481 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7482 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7483 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7486 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7487 struct btrfs_ioctl_get_dev_stats *stats)
7489 struct btrfs_device *dev;
7490 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7493 mutex_lock(&fs_devices->device_list_mutex);
7494 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7496 mutex_unlock(&fs_devices->device_list_mutex);
7499 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7501 } else if (!dev->dev_stats_valid) {
7502 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7504 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7505 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7506 if (stats->nr_items > i)
7508 btrfs_dev_stat_read_and_reset(dev, i);
7510 btrfs_dev_stat_reset(dev, i);
7513 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7514 if (stats->nr_items > i)
7515 stats->values[i] = btrfs_dev_stat_read(dev, i);
7517 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7518 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7522 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7524 struct buffer_head *bh;
7525 struct btrfs_super_block *disk_super;
7531 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7534 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7537 disk_super = (struct btrfs_super_block *)bh->b_data;
7539 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7540 set_buffer_dirty(bh);
7541 sync_dirty_buffer(bh);
7545 /* Notify udev that device has changed */
7546 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7548 /* Update ctime/mtime for device path for libblkid */
7549 update_dev_time(device_path);
7553 * Update the size and bytes used for each device where it changed. This is
7554 * delayed since we would otherwise get errors while writing out the
7557 * Must be invoked during transaction commit.
7559 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7561 struct btrfs_device *curr, *next;
7563 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7565 if (list_empty(&trans->dev_update_list))
7569 * We don't need the device_list_mutex here. This list is owned by the
7570 * transaction and the transaction must complete before the device is
7573 mutex_lock(&trans->fs_info->chunk_mutex);
7574 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7576 list_del_init(&curr->post_commit_list);
7577 curr->commit_total_bytes = curr->disk_total_bytes;
7578 curr->commit_bytes_used = curr->bytes_used;
7580 mutex_unlock(&trans->fs_info->chunk_mutex);
7583 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7585 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7586 while (fs_devices) {
7587 fs_devices->fs_info = fs_info;
7588 fs_devices = fs_devices->seed;
7592 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7594 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7595 while (fs_devices) {
7596 fs_devices->fs_info = NULL;
7597 fs_devices = fs_devices->seed;
7602 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7604 int btrfs_bg_type_to_factor(u64 flags)
7606 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
7607 BTRFS_BLOCK_GROUP_RAID10))
7614 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7615 u64 chunk_offset, u64 devid,
7616 u64 physical_offset, u64 physical_len)
7618 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7619 struct extent_map *em;
7620 struct map_lookup *map;
7621 struct btrfs_device *dev;
7627 read_lock(&em_tree->lock);
7628 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7629 read_unlock(&em_tree->lock);
7633 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7634 physical_offset, devid);
7639 map = em->map_lookup;
7640 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7641 if (physical_len != stripe_len) {
7643 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7644 physical_offset, devid, em->start, physical_len,
7650 for (i = 0; i < map->num_stripes; i++) {
7651 if (map->stripes[i].dev->devid == devid &&
7652 map->stripes[i].physical == physical_offset) {
7654 if (map->verified_stripes >= map->num_stripes) {
7656 "too many dev extents for chunk %llu found",
7661 map->verified_stripes++;
7667 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7668 physical_offset, devid);
7672 /* Make sure no dev extent is beyond device bondary */
7673 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7675 btrfs_err(fs_info, "failed to find devid %llu", devid);
7680 /* It's possible this device is a dummy for seed device */
7681 if (dev->disk_total_bytes == 0) {
7682 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7685 btrfs_err(fs_info, "failed to find seed devid %llu",
7692 if (physical_offset + physical_len > dev->disk_total_bytes) {
7694 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7695 devid, physical_offset, physical_len,
7696 dev->disk_total_bytes);
7701 free_extent_map(em);
7705 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7707 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7708 struct extent_map *em;
7709 struct rb_node *node;
7712 read_lock(&em_tree->lock);
7713 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7714 em = rb_entry(node, struct extent_map, rb_node);
7715 if (em->map_lookup->num_stripes !=
7716 em->map_lookup->verified_stripes) {
7718 "chunk %llu has missing dev extent, have %d expect %d",
7719 em->start, em->map_lookup->verified_stripes,
7720 em->map_lookup->num_stripes);
7726 read_unlock(&em_tree->lock);
7731 * Ensure that all dev extents are mapped to correct chunk, otherwise
7732 * later chunk allocation/free would cause unexpected behavior.
7734 * NOTE: This will iterate through the whole device tree, which should be of
7735 * the same size level as the chunk tree. This slightly increases mount time.
7737 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7739 struct btrfs_path *path;
7740 struct btrfs_root *root = fs_info->dev_root;
7741 struct btrfs_key key;
7743 u64 prev_dev_ext_end = 0;
7747 key.type = BTRFS_DEV_EXTENT_KEY;
7750 path = btrfs_alloc_path();
7754 path->reada = READA_FORWARD;
7755 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7759 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7760 ret = btrfs_next_item(root, path);
7763 /* No dev extents at all? Not good */
7770 struct extent_buffer *leaf = path->nodes[0];
7771 struct btrfs_dev_extent *dext;
7772 int slot = path->slots[0];
7774 u64 physical_offset;
7778 btrfs_item_key_to_cpu(leaf, &key, slot);
7779 if (key.type != BTRFS_DEV_EXTENT_KEY)
7781 devid = key.objectid;
7782 physical_offset = key.offset;
7784 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7785 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7786 physical_len = btrfs_dev_extent_length(leaf, dext);
7788 /* Check if this dev extent overlaps with the previous one */
7789 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7791 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7792 devid, physical_offset, prev_dev_ext_end);
7797 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7798 physical_offset, physical_len);
7802 prev_dev_ext_end = physical_offset + physical_len;
7804 ret = btrfs_next_item(root, path);
7813 /* Ensure all chunks have corresponding dev extents */
7814 ret = verify_chunk_dev_extent_mapping(fs_info);
7816 btrfs_free_path(path);
7821 * Check whether the given block group or device is pinned by any inode being
7822 * used as a swapfile.
7824 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7826 struct btrfs_swapfile_pin *sp;
7827 struct rb_node *node;
7829 spin_lock(&fs_info->swapfile_pins_lock);
7830 node = fs_info->swapfile_pins.rb_node;
7832 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7834 node = node->rb_left;
7835 else if (ptr > sp->ptr)
7836 node = node->rb_right;
7840 spin_unlock(&fs_info->swapfile_pins_lock);
7841 return node != NULL;