1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
31 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
32 [BTRFS_RAID_RAID10] = {
35 .devs_max = 0, /* 0 == as many as possible */
37 .tolerated_failures = 1,
41 .raid_name = "raid10",
42 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
43 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
45 [BTRFS_RAID_RAID1] = {
50 .tolerated_failures = 1,
55 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
56 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
63 .tolerated_failures = 0,
68 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
71 [BTRFS_RAID_RAID0] = {
76 .tolerated_failures = 0,
81 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
84 [BTRFS_RAID_SINGLE] = {
89 .tolerated_failures = 0,
93 .raid_name = "single",
97 [BTRFS_RAID_RAID5] = {
102 .tolerated_failures = 1,
106 .raid_name = "raid5",
107 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
108 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
110 [BTRFS_RAID_RAID6] = {
115 .tolerated_failures = 2,
119 .raid_name = "raid6",
120 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
121 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
125 const char *get_raid_name(enum btrfs_raid_types type)
127 if (type >= BTRFS_NR_RAID_TYPES)
130 return btrfs_raid_array[type].raid_name;
134 * Fill @buf with textual description of @bg_flags, no more than @size_buf
135 * bytes including terminating null byte.
137 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
142 u64 flags = bg_flags;
143 u32 size_bp = size_buf;
150 #define DESCRIBE_FLAG(flag, desc) \
152 if (flags & (flag)) { \
153 ret = snprintf(bp, size_bp, "%s|", (desc)); \
154 if (ret < 0 || ret >= size_bp) \
162 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
163 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
164 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
166 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
167 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
168 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
169 btrfs_raid_array[i].raid_name);
173 ret = snprintf(bp, size_bp, "0x%llx|", flags);
177 if (size_bp < size_buf)
178 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
181 * The text is trimmed, it's up to the caller to provide sufficiently
187 static int init_first_rw_device(struct btrfs_trans_handle *trans,
188 struct btrfs_fs_info *fs_info);
189 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
190 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
191 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
192 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
193 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
194 enum btrfs_map_op op,
195 u64 logical, u64 *length,
196 struct btrfs_bio **bbio_ret,
197 int mirror_num, int need_raid_map);
203 * There are several mutexes that protect manipulation of devices and low-level
204 * structures like chunks but not block groups, extents or files
206 * uuid_mutex (global lock)
207 * ------------------------
208 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
209 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
210 * device) or requested by the device= mount option
212 * the mutex can be very coarse and can cover long-running operations
214 * protects: updates to fs_devices counters like missing devices, rw devices,
215 * seeding, structure cloning, openning/closing devices at mount/umount time
217 * global::fs_devs - add, remove, updates to the global list
219 * does not protect: manipulation of the fs_devices::devices list!
221 * btrfs_device::name - renames (write side), read is RCU
223 * fs_devices::device_list_mutex (per-fs, with RCU)
224 * ------------------------------------------------
225 * protects updates to fs_devices::devices, ie. adding and deleting
227 * simple list traversal with read-only actions can be done with RCU protection
229 * may be used to exclude some operations from running concurrently without any
230 * modifications to the list (see write_all_supers)
234 * protects balance structures (status, state) and context accessed from
235 * several places (internally, ioctl)
239 * protects chunks, adding or removing during allocation, trim or when a new
240 * device is added/removed
244 * a big lock that is held by the cleaner thread and prevents running subvolume
245 * cleaning together with relocation or delayed iputs
258 * Exclusive operations, BTRFS_FS_EXCL_OP
259 * ======================================
261 * Maintains the exclusivity of the following operations that apply to the
262 * whole filesystem and cannot run in parallel.
267 * - Device replace (*)
270 * The device operations (as above) can be in one of the following states:
276 * Only device operations marked with (*) can go into the Paused state for the
279 * - ioctl (only Balance can be Paused through ioctl)
280 * - filesystem remounted as read-only
281 * - filesystem unmounted and mounted as read-only
282 * - system power-cycle and filesystem mounted as read-only
283 * - filesystem or device errors leading to forced read-only
285 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
286 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
287 * A device operation in Paused or Running state can be canceled or resumed
288 * either by ioctl (Balance only) or when remounted as read-write.
289 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
293 DEFINE_MUTEX(uuid_mutex);
294 static LIST_HEAD(fs_uuids);
295 struct list_head *btrfs_get_fs_uuids(void)
301 * alloc_fs_devices - allocate struct btrfs_fs_devices
302 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
303 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
305 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
306 * The returned struct is not linked onto any lists and can be destroyed with
307 * kfree() right away.
309 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
310 const u8 *metadata_fsid)
312 struct btrfs_fs_devices *fs_devs;
314 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
316 return ERR_PTR(-ENOMEM);
318 mutex_init(&fs_devs->device_list_mutex);
320 INIT_LIST_HEAD(&fs_devs->devices);
321 INIT_LIST_HEAD(&fs_devs->resized_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 rcu_string_free(device->name);
338 bio_put(device->flush_bio);
342 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
344 struct btrfs_device *device;
345 WARN_ON(fs_devices->opened);
346 while (!list_empty(&fs_devices->devices)) {
347 device = list_entry(fs_devices->devices.next,
348 struct btrfs_device, dev_list);
349 list_del(&device->dev_list);
350 btrfs_free_device(device);
355 static void btrfs_kobject_uevent(struct block_device *bdev,
356 enum kobject_action action)
360 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
362 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
364 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
365 &disk_to_dev(bdev->bd_disk)->kobj);
368 void __exit btrfs_cleanup_fs_uuids(void)
370 struct btrfs_fs_devices *fs_devices;
372 while (!list_empty(&fs_uuids)) {
373 fs_devices = list_entry(fs_uuids.next,
374 struct btrfs_fs_devices, fs_list);
375 list_del(&fs_devices->fs_list);
376 free_fs_devices(fs_devices);
381 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
382 * Returned struct is not linked onto any lists and must be destroyed using
385 static struct btrfs_device *__alloc_device(void)
387 struct btrfs_device *dev;
389 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
391 return ERR_PTR(-ENOMEM);
394 * Preallocate a bio that's always going to be used for flushing device
395 * barriers and matches the device lifespan
397 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
398 if (!dev->flush_bio) {
400 return ERR_PTR(-ENOMEM);
403 INIT_LIST_HEAD(&dev->dev_list);
404 INIT_LIST_HEAD(&dev->dev_alloc_list);
405 INIT_LIST_HEAD(&dev->resized_list);
407 spin_lock_init(&dev->io_lock);
409 atomic_set(&dev->reada_in_flight, 0);
410 atomic_set(&dev->dev_stats_ccnt, 0);
411 btrfs_device_data_ordered_init(dev);
412 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
413 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
419 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
422 * If devid and uuid are both specified, the match must be exact, otherwise
423 * only devid is used.
425 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
426 u64 devid, const u8 *uuid)
428 struct btrfs_device *dev;
430 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
431 if (dev->devid == devid &&
432 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
439 static noinline struct btrfs_fs_devices *find_fsid(
440 const u8 *fsid, const u8 *metadata_fsid)
442 struct btrfs_fs_devices *fs_devices;
448 * Handle scanned device having completed its fsid change but
449 * belonging to a fs_devices that was created by first scanning
450 * a device which didn't have its fsid/metadata_uuid changed
451 * at all and the CHANGING_FSID_V2 flag set.
453 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
454 if (fs_devices->fsid_change &&
455 memcmp(metadata_fsid, fs_devices->fsid,
456 BTRFS_FSID_SIZE) == 0 &&
457 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
458 BTRFS_FSID_SIZE) == 0) {
463 * Handle scanned device having completed its fsid change but
464 * belonging to a fs_devices that was created by a device that
465 * has an outdated pair of fsid/metadata_uuid and
466 * CHANGING_FSID_V2 flag set.
468 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
469 if (fs_devices->fsid_change &&
470 memcmp(fs_devices->metadata_uuid,
471 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
472 memcmp(metadata_fsid, fs_devices->metadata_uuid,
473 BTRFS_FSID_SIZE) == 0) {
479 /* Handle non-split brain cases */
480 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
482 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
483 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
484 BTRFS_FSID_SIZE) == 0)
487 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
495 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
496 int flush, struct block_device **bdev,
497 struct buffer_head **bh)
501 *bdev = blkdev_get_by_path(device_path, flags, holder);
504 ret = PTR_ERR(*bdev);
509 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
510 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
512 blkdev_put(*bdev, flags);
515 invalidate_bdev(*bdev);
516 *bh = btrfs_read_dev_super(*bdev);
519 blkdev_put(*bdev, flags);
531 static void requeue_list(struct btrfs_pending_bios *pending_bios,
532 struct bio *head, struct bio *tail)
535 struct bio *old_head;
537 old_head = pending_bios->head;
538 pending_bios->head = head;
539 if (pending_bios->tail)
540 tail->bi_next = old_head;
542 pending_bios->tail = tail;
546 * we try to collect pending bios for a device so we don't get a large
547 * number of procs sending bios down to the same device. This greatly
548 * improves the schedulers ability to collect and merge the bios.
550 * But, it also turns into a long list of bios to process and that is sure
551 * to eventually make the worker thread block. The solution here is to
552 * make some progress and then put this work struct back at the end of
553 * the list if the block device is congested. This way, multiple devices
554 * can make progress from a single worker thread.
556 static noinline void run_scheduled_bios(struct btrfs_device *device)
558 struct btrfs_fs_info *fs_info = device->fs_info;
560 struct backing_dev_info *bdi;
561 struct btrfs_pending_bios *pending_bios;
565 unsigned long num_run;
566 unsigned long batch_run = 0;
567 unsigned long last_waited = 0;
569 int sync_pending = 0;
570 struct blk_plug plug;
573 * this function runs all the bios we've collected for
574 * a particular device. We don't want to wander off to
575 * another device without first sending all of these down.
576 * So, setup a plug here and finish it off before we return
578 blk_start_plug(&plug);
580 bdi = device->bdev->bd_bdi;
583 spin_lock(&device->io_lock);
588 /* take all the bios off the list at once and process them
589 * later on (without the lock held). But, remember the
590 * tail and other pointers so the bios can be properly reinserted
591 * into the list if we hit congestion
593 if (!force_reg && device->pending_sync_bios.head) {
594 pending_bios = &device->pending_sync_bios;
597 pending_bios = &device->pending_bios;
601 pending = pending_bios->head;
602 tail = pending_bios->tail;
603 WARN_ON(pending && !tail);
606 * if pending was null this time around, no bios need processing
607 * at all and we can stop. Otherwise it'll loop back up again
608 * and do an additional check so no bios are missed.
610 * device->running_pending is used to synchronize with the
613 if (device->pending_sync_bios.head == NULL &&
614 device->pending_bios.head == NULL) {
616 device->running_pending = 0;
619 device->running_pending = 1;
622 pending_bios->head = NULL;
623 pending_bios->tail = NULL;
625 spin_unlock(&device->io_lock);
630 /* we want to work on both lists, but do more bios on the
631 * sync list than the regular list
634 pending_bios != &device->pending_sync_bios &&
635 device->pending_sync_bios.head) ||
636 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
637 device->pending_bios.head)) {
638 spin_lock(&device->io_lock);
639 requeue_list(pending_bios, pending, tail);
644 pending = pending->bi_next;
647 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
650 * if we're doing the sync list, record that our
651 * plug has some sync requests on it
653 * If we're doing the regular list and there are
654 * sync requests sitting around, unplug before
657 if (pending_bios == &device->pending_sync_bios) {
659 } else if (sync_pending) {
660 blk_finish_plug(&plug);
661 blk_start_plug(&plug);
665 btrfsic_submit_bio(cur);
672 * we made progress, there is more work to do and the bdi
673 * is now congested. Back off and let other work structs
676 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
677 fs_info->fs_devices->open_devices > 1) {
678 struct io_context *ioc;
680 ioc = current->io_context;
683 * the main goal here is that we don't want to
684 * block if we're going to be able to submit
685 * more requests without blocking.
687 * This code does two great things, it pokes into
688 * the elevator code from a filesystem _and_
689 * it makes assumptions about how batching works.
691 if (ioc && ioc->nr_batch_requests > 0 &&
692 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
694 ioc->last_waited == last_waited)) {
696 * we want to go through our batch of
697 * requests and stop. So, we copy out
698 * the ioc->last_waited time and test
699 * against it before looping
701 last_waited = ioc->last_waited;
705 spin_lock(&device->io_lock);
706 requeue_list(pending_bios, pending, tail);
707 device->running_pending = 1;
709 spin_unlock(&device->io_lock);
710 btrfs_queue_work(fs_info->submit_workers,
720 spin_lock(&device->io_lock);
721 if (device->pending_bios.head || device->pending_sync_bios.head)
723 spin_unlock(&device->io_lock);
726 blk_finish_plug(&plug);
729 static void pending_bios_fn(struct btrfs_work *work)
731 struct btrfs_device *device;
733 device = container_of(work, struct btrfs_device, work);
734 run_scheduled_bios(device);
738 * Search and remove all stale (devices which are not mounted) devices.
739 * When both inputs are NULL, it will search and release all stale devices.
740 * path: Optional. When provided will it release all unmounted devices
741 * matching this path only.
742 * skip_dev: Optional. Will skip this device when searching for the stale
745 static void btrfs_free_stale_devices(const char *path,
746 struct btrfs_device *skip_device)
748 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
749 struct btrfs_device *device, *tmp_device;
751 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
752 mutex_lock(&fs_devices->device_list_mutex);
753 if (fs_devices->opened) {
754 mutex_unlock(&fs_devices->device_list_mutex);
758 list_for_each_entry_safe(device, tmp_device,
759 &fs_devices->devices, dev_list) {
762 if (skip_device && skip_device == device)
764 if (path && !device->name)
769 not_found = strcmp(rcu_str_deref(device->name),
775 /* delete the stale device */
776 fs_devices->num_devices--;
777 list_del(&device->dev_list);
778 btrfs_free_device(device);
780 if (fs_devices->num_devices == 0)
783 mutex_unlock(&fs_devices->device_list_mutex);
784 if (fs_devices->num_devices == 0) {
785 btrfs_sysfs_remove_fsid(fs_devices);
786 list_del(&fs_devices->fs_list);
787 free_fs_devices(fs_devices);
792 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
793 struct btrfs_device *device, fmode_t flags,
796 struct request_queue *q;
797 struct block_device *bdev;
798 struct buffer_head *bh;
799 struct btrfs_super_block *disk_super;
808 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
813 disk_super = (struct btrfs_super_block *)bh->b_data;
814 devid = btrfs_stack_device_id(&disk_super->dev_item);
815 if (devid != device->devid)
818 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
821 device->generation = btrfs_super_generation(disk_super);
823 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
824 if (btrfs_super_incompat_flags(disk_super) &
825 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
827 "BTRFS: Invalid seeding and uuid-changed device detected\n");
831 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
832 fs_devices->seeding = 1;
834 if (bdev_read_only(bdev))
835 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
837 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
840 q = bdev_get_queue(bdev);
841 if (!blk_queue_nonrot(q))
842 fs_devices->rotating = 1;
845 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
846 device->mode = flags;
848 fs_devices->open_devices++;
849 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
850 device->devid != BTRFS_DEV_REPLACE_DEVID) {
851 fs_devices->rw_devices++;
852 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
860 blkdev_put(bdev, flags);
866 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
867 * being created with a disk that has already completed its fsid change.
869 static struct btrfs_fs_devices *find_fsid_inprogress(
870 struct btrfs_super_block *disk_super)
872 struct btrfs_fs_devices *fs_devices;
874 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
875 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
876 BTRFS_FSID_SIZE) != 0 &&
877 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
878 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
887 static struct btrfs_fs_devices *find_fsid_changed(
888 struct btrfs_super_block *disk_super)
890 struct btrfs_fs_devices *fs_devices;
893 * Handles the case where scanned device is part of an fs that had
894 * multiple successful changes of FSID but curently device didn't
895 * observe it. Meaning our fsid will be different than theirs.
897 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
898 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
899 BTRFS_FSID_SIZE) != 0 &&
900 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
901 BTRFS_FSID_SIZE) == 0 &&
902 memcmp(fs_devices->fsid, disk_super->fsid,
903 BTRFS_FSID_SIZE) != 0) {
911 * Add new device to list of registered devices
914 * device pointer which was just added or updated when successful
915 * error pointer when failed
917 static noinline struct btrfs_device *device_list_add(const char *path,
918 struct btrfs_super_block *disk_super,
919 bool *new_device_added)
921 struct btrfs_device *device;
922 struct btrfs_fs_devices *fs_devices = NULL;
923 struct rcu_string *name;
924 u64 found_transid = btrfs_super_generation(disk_super);
925 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
926 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
927 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
928 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
929 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
931 if (fsid_change_in_progress) {
932 if (!has_metadata_uuid) {
934 * When we have an image which has CHANGING_FSID_V2 set
935 * it might belong to either a filesystem which has
936 * disks with completed fsid change or it might belong
937 * to fs with no UUID changes in effect, handle both.
939 fs_devices = find_fsid_inprogress(disk_super);
941 fs_devices = find_fsid(disk_super->fsid, NULL);
943 fs_devices = find_fsid_changed(disk_super);
945 } else if (has_metadata_uuid) {
946 fs_devices = find_fsid(disk_super->fsid,
947 disk_super->metadata_uuid);
949 fs_devices = find_fsid(disk_super->fsid, NULL);
954 if (has_metadata_uuid)
955 fs_devices = alloc_fs_devices(disk_super->fsid,
956 disk_super->metadata_uuid);
958 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
960 fs_devices->fsid_change = fsid_change_in_progress;
962 if (IS_ERR(fs_devices))
963 return ERR_CAST(fs_devices);
965 mutex_lock(&fs_devices->device_list_mutex);
966 list_add(&fs_devices->fs_list, &fs_uuids);
970 mutex_lock(&fs_devices->device_list_mutex);
971 device = find_device(fs_devices, devid,
972 disk_super->dev_item.uuid);
975 * If this disk has been pulled into an fs devices created by
976 * a device which had the CHANGING_FSID_V2 flag then replace the
977 * metadata_uuid/fsid values of the fs_devices.
979 if (has_metadata_uuid && fs_devices->fsid_change &&
980 found_transid > fs_devices->latest_generation) {
981 memcpy(fs_devices->fsid, disk_super->fsid,
983 memcpy(fs_devices->metadata_uuid,
984 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
986 fs_devices->fsid_change = false;
991 if (fs_devices->opened) {
992 mutex_unlock(&fs_devices->device_list_mutex);
993 return ERR_PTR(-EBUSY);
996 device = btrfs_alloc_device(NULL, &devid,
997 disk_super->dev_item.uuid);
998 if (IS_ERR(device)) {
999 mutex_unlock(&fs_devices->device_list_mutex);
1000 /* we can safely leave the fs_devices entry around */
1004 name = rcu_string_strdup(path, GFP_NOFS);
1006 btrfs_free_device(device);
1007 mutex_unlock(&fs_devices->device_list_mutex);
1008 return ERR_PTR(-ENOMEM);
1010 rcu_assign_pointer(device->name, name);
1012 list_add_rcu(&device->dev_list, &fs_devices->devices);
1013 fs_devices->num_devices++;
1015 device->fs_devices = fs_devices;
1016 *new_device_added = true;
1018 if (disk_super->label[0])
1019 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1020 disk_super->label, devid, found_transid, path);
1022 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1023 disk_super->fsid, devid, found_transid, path);
1025 } else if (!device->name || strcmp(device->name->str, path)) {
1027 * When FS is already mounted.
1028 * 1. If you are here and if the device->name is NULL that
1029 * means this device was missing at time of FS mount.
1030 * 2. If you are here and if the device->name is different
1031 * from 'path' that means either
1032 * a. The same device disappeared and reappeared with
1033 * different name. or
1034 * b. The missing-disk-which-was-replaced, has
1037 * We must allow 1 and 2a above. But 2b would be a spurious
1038 * and unintentional.
1040 * Further in case of 1 and 2a above, the disk at 'path'
1041 * would have missed some transaction when it was away and
1042 * in case of 2a the stale bdev has to be updated as well.
1043 * 2b must not be allowed at all time.
1047 * For now, we do allow update to btrfs_fs_device through the
1048 * btrfs dev scan cli after FS has been mounted. We're still
1049 * tracking a problem where systems fail mount by subvolume id
1050 * when we reject replacement on a mounted FS.
1052 if (!fs_devices->opened && found_transid < device->generation) {
1054 * That is if the FS is _not_ mounted and if you
1055 * are here, that means there is more than one
1056 * disk with same uuid and devid.We keep the one
1057 * with larger generation number or the last-in if
1058 * generation are equal.
1060 mutex_unlock(&fs_devices->device_list_mutex);
1061 return ERR_PTR(-EEXIST);
1065 * We are going to replace the device path for a given devid,
1066 * make sure it's the same device if the device is mounted
1069 struct block_device *path_bdev;
1071 path_bdev = lookup_bdev(path);
1072 if (IS_ERR(path_bdev)) {
1073 mutex_unlock(&fs_devices->device_list_mutex);
1074 return ERR_CAST(path_bdev);
1077 if (device->bdev != path_bdev) {
1079 mutex_unlock(&fs_devices->device_list_mutex);
1080 btrfs_warn_in_rcu(device->fs_info,
1081 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1082 disk_super->fsid, devid,
1083 rcu_str_deref(device->name), path);
1084 return ERR_PTR(-EEXIST);
1087 btrfs_info_in_rcu(device->fs_info,
1088 "device fsid %pU devid %llu moved old:%s new:%s",
1089 disk_super->fsid, devid,
1090 rcu_str_deref(device->name), path);
1093 name = rcu_string_strdup(path, GFP_NOFS);
1095 mutex_unlock(&fs_devices->device_list_mutex);
1096 return ERR_PTR(-ENOMEM);
1098 rcu_string_free(device->name);
1099 rcu_assign_pointer(device->name, name);
1100 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1101 fs_devices->missing_devices--;
1102 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1107 * Unmount does not free the btrfs_device struct but would zero
1108 * generation along with most of the other members. So just update
1109 * it back. We need it to pick the disk with largest generation
1112 if (!fs_devices->opened) {
1113 device->generation = found_transid;
1114 fs_devices->latest_generation = max_t(u64, found_transid,
1115 fs_devices->latest_generation);
1118 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1120 mutex_unlock(&fs_devices->device_list_mutex);
1124 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1126 struct btrfs_fs_devices *fs_devices;
1127 struct btrfs_device *device;
1128 struct btrfs_device *orig_dev;
1130 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1131 if (IS_ERR(fs_devices))
1134 mutex_lock(&orig->device_list_mutex);
1135 fs_devices->total_devices = orig->total_devices;
1137 /* We have held the volume lock, it is safe to get the devices. */
1138 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1139 struct rcu_string *name;
1141 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1147 * This is ok to do without rcu read locked because we hold the
1148 * uuid mutex so nothing we touch in here is going to disappear.
1150 if (orig_dev->name) {
1151 name = rcu_string_strdup(orig_dev->name->str,
1154 btrfs_free_device(device);
1157 rcu_assign_pointer(device->name, name);
1160 list_add(&device->dev_list, &fs_devices->devices);
1161 device->fs_devices = fs_devices;
1162 fs_devices->num_devices++;
1164 mutex_unlock(&orig->device_list_mutex);
1167 mutex_unlock(&orig->device_list_mutex);
1168 free_fs_devices(fs_devices);
1169 return ERR_PTR(-ENOMEM);
1173 * After we have read the system tree and know devids belonging to
1174 * this filesystem, remove the device which does not belong there.
1176 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1178 struct btrfs_device *device, *next;
1179 struct btrfs_device *latest_dev = NULL;
1181 mutex_lock(&uuid_mutex);
1183 /* This is the initialized path, it is safe to release the devices. */
1184 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1185 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1186 &device->dev_state)) {
1187 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1188 &device->dev_state) &&
1190 device->generation > latest_dev->generation)) {
1191 latest_dev = device;
1196 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1198 * In the first step, keep the device which has
1199 * the correct fsid and the devid that is used
1200 * for the dev_replace procedure.
1201 * In the second step, the dev_replace state is
1202 * read from the device tree and it is known
1203 * whether the procedure is really active or
1204 * not, which means whether this device is
1205 * used or whether it should be removed.
1207 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1208 &device->dev_state)) {
1213 blkdev_put(device->bdev, device->mode);
1214 device->bdev = NULL;
1215 fs_devices->open_devices--;
1217 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1218 list_del_init(&device->dev_alloc_list);
1219 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1220 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1221 &device->dev_state))
1222 fs_devices->rw_devices--;
1224 list_del_init(&device->dev_list);
1225 fs_devices->num_devices--;
1226 btrfs_free_device(device);
1229 if (fs_devices->seed) {
1230 fs_devices = fs_devices->seed;
1234 fs_devices->latest_bdev = latest_dev->bdev;
1236 mutex_unlock(&uuid_mutex);
1239 static void free_device_rcu(struct rcu_head *head)
1241 struct btrfs_device *device;
1243 device = container_of(head, struct btrfs_device, rcu);
1244 btrfs_free_device(device);
1247 static void btrfs_close_bdev(struct btrfs_device *device)
1252 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1253 sync_blockdev(device->bdev);
1254 invalidate_bdev(device->bdev);
1257 blkdev_put(device->bdev, device->mode);
1260 static void btrfs_close_one_device(struct btrfs_device *device)
1262 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1263 struct btrfs_device *new_device;
1264 struct rcu_string *name;
1267 fs_devices->open_devices--;
1269 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1270 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1271 list_del_init(&device->dev_alloc_list);
1272 fs_devices->rw_devices--;
1275 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1276 fs_devices->missing_devices--;
1278 btrfs_close_bdev(device);
1280 new_device = btrfs_alloc_device(NULL, &device->devid,
1282 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1284 /* Safe because we are under uuid_mutex */
1286 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1287 BUG_ON(!name); /* -ENOMEM */
1288 rcu_assign_pointer(new_device->name, name);
1291 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1292 new_device->fs_devices = device->fs_devices;
1294 call_rcu(&device->rcu, free_device_rcu);
1297 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1299 struct btrfs_device *device, *tmp;
1301 if (--fs_devices->opened > 0)
1304 mutex_lock(&fs_devices->device_list_mutex);
1305 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1306 btrfs_close_one_device(device);
1308 mutex_unlock(&fs_devices->device_list_mutex);
1310 WARN_ON(fs_devices->open_devices);
1311 WARN_ON(fs_devices->rw_devices);
1312 fs_devices->opened = 0;
1313 fs_devices->seeding = 0;
1318 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1320 struct btrfs_fs_devices *seed_devices = NULL;
1323 mutex_lock(&uuid_mutex);
1324 ret = close_fs_devices(fs_devices);
1325 if (!fs_devices->opened) {
1326 seed_devices = fs_devices->seed;
1327 fs_devices->seed = NULL;
1329 mutex_unlock(&uuid_mutex);
1331 while (seed_devices) {
1332 fs_devices = seed_devices;
1333 seed_devices = fs_devices->seed;
1334 close_fs_devices(fs_devices);
1335 free_fs_devices(fs_devices);
1340 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1341 fmode_t flags, void *holder)
1343 struct btrfs_device *device;
1344 struct btrfs_device *latest_dev = NULL;
1347 flags |= FMODE_EXCL;
1349 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1350 /* Just open everything we can; ignore failures here */
1351 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1355 device->generation > latest_dev->generation)
1356 latest_dev = device;
1358 if (fs_devices->open_devices == 0) {
1362 fs_devices->opened = 1;
1363 fs_devices->latest_bdev = latest_dev->bdev;
1364 fs_devices->total_rw_bytes = 0;
1369 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1371 struct btrfs_device *dev1, *dev2;
1373 dev1 = list_entry(a, struct btrfs_device, dev_list);
1374 dev2 = list_entry(b, struct btrfs_device, dev_list);
1376 if (dev1->devid < dev2->devid)
1378 else if (dev1->devid > dev2->devid)
1383 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1384 fmode_t flags, void *holder)
1388 lockdep_assert_held(&uuid_mutex);
1390 mutex_lock(&fs_devices->device_list_mutex);
1391 if (fs_devices->opened) {
1392 fs_devices->opened++;
1395 list_sort(NULL, &fs_devices->devices, devid_cmp);
1396 ret = open_fs_devices(fs_devices, flags, holder);
1398 mutex_unlock(&fs_devices->device_list_mutex);
1403 static void btrfs_release_disk_super(struct page *page)
1409 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1411 struct btrfs_super_block **disk_super)
1416 /* make sure our super fits in the device */
1417 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1420 /* make sure our super fits in the page */
1421 if (sizeof(**disk_super) > PAGE_SIZE)
1424 /* make sure our super doesn't straddle pages on disk */
1425 index = bytenr >> PAGE_SHIFT;
1426 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1429 /* pull in the page with our super */
1430 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1433 if (IS_ERR_OR_NULL(*page))
1438 /* align our pointer to the offset of the super block */
1439 *disk_super = p + offset_in_page(bytenr);
1441 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1442 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1443 btrfs_release_disk_super(*page);
1447 if ((*disk_super)->label[0] &&
1448 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1449 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1455 * Look for a btrfs signature on a device. This may be called out of the mount path
1456 * and we are not allowed to call set_blocksize during the scan. The superblock
1457 * is read via pagecache
1459 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1462 struct btrfs_super_block *disk_super;
1463 bool new_device_added = false;
1464 struct btrfs_device *device = NULL;
1465 struct block_device *bdev;
1469 lockdep_assert_held(&uuid_mutex);
1472 * we would like to check all the supers, but that would make
1473 * a btrfs mount succeed after a mkfs from a different FS.
1474 * So, we need to add a special mount option to scan for
1475 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1477 bytenr = btrfs_sb_offset(0);
1478 flags |= FMODE_EXCL;
1480 bdev = blkdev_get_by_path(path, flags, holder);
1482 return ERR_CAST(bdev);
1484 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1485 device = ERR_PTR(-EINVAL);
1486 goto error_bdev_put;
1489 device = device_list_add(path, disk_super, &new_device_added);
1490 if (!IS_ERR(device)) {
1491 if (new_device_added)
1492 btrfs_free_stale_devices(path, device);
1495 btrfs_release_disk_super(page);
1498 blkdev_put(bdev, flags);
1503 static int contains_pending_extent(struct btrfs_transaction *transaction,
1504 struct btrfs_device *device,
1505 u64 *start, u64 len)
1507 struct btrfs_fs_info *fs_info = device->fs_info;
1508 struct extent_map *em;
1509 struct list_head *search_list = &fs_info->pinned_chunks;
1511 u64 physical_start = *start;
1514 search_list = &transaction->pending_chunks;
1516 list_for_each_entry(em, search_list, list) {
1517 struct map_lookup *map;
1520 map = em->map_lookup;
1521 for (i = 0; i < map->num_stripes; i++) {
1524 if (map->stripes[i].dev != device)
1526 if (map->stripes[i].physical >= physical_start + len ||
1527 map->stripes[i].physical + em->orig_block_len <=
1531 * Make sure that while processing the pinned list we do
1532 * not override our *start with a lower value, because
1533 * we can have pinned chunks that fall within this
1534 * device hole and that have lower physical addresses
1535 * than the pending chunks we processed before. If we
1536 * do not take this special care we can end up getting
1537 * 2 pending chunks that start at the same physical
1538 * device offsets because the end offset of a pinned
1539 * chunk can be equal to the start offset of some
1542 end = map->stripes[i].physical + em->orig_block_len;
1549 if (search_list != &fs_info->pinned_chunks) {
1550 search_list = &fs_info->pinned_chunks;
1559 * find_free_dev_extent_start - find free space in the specified device
1560 * @device: the device which we search the free space in
1561 * @num_bytes: the size of the free space that we need
1562 * @search_start: the position from which to begin the search
1563 * @start: store the start of the free space.
1564 * @len: the size of the free space. that we find, or the size
1565 * of the max free space if we don't find suitable free space
1567 * this uses a pretty simple search, the expectation is that it is
1568 * called very infrequently and that a given device has a small number
1571 * @start is used to store the start of the free space if we find. But if we
1572 * don't find suitable free space, it will be used to store the start position
1573 * of the max free space.
1575 * @len is used to store the size of the free space that we find.
1576 * But if we don't find suitable free space, it is used to store the size of
1577 * the max free space.
1579 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1580 struct btrfs_device *device, u64 num_bytes,
1581 u64 search_start, u64 *start, u64 *len)
1583 struct btrfs_fs_info *fs_info = device->fs_info;
1584 struct btrfs_root *root = fs_info->dev_root;
1585 struct btrfs_key key;
1586 struct btrfs_dev_extent *dev_extent;
1587 struct btrfs_path *path;
1592 u64 search_end = device->total_bytes;
1595 struct extent_buffer *l;
1598 * We don't want to overwrite the superblock on the drive nor any area
1599 * used by the boot loader (grub for example), so we make sure to start
1600 * at an offset of at least 1MB.
1602 search_start = max_t(u64, search_start, SZ_1M);
1604 path = btrfs_alloc_path();
1608 max_hole_start = search_start;
1612 if (search_start >= search_end ||
1613 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1618 path->reada = READA_FORWARD;
1619 path->search_commit_root = 1;
1620 path->skip_locking = 1;
1622 key.objectid = device->devid;
1623 key.offset = search_start;
1624 key.type = BTRFS_DEV_EXTENT_KEY;
1626 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1630 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1637 slot = path->slots[0];
1638 if (slot >= btrfs_header_nritems(l)) {
1639 ret = btrfs_next_leaf(root, path);
1647 btrfs_item_key_to_cpu(l, &key, slot);
1649 if (key.objectid < device->devid)
1652 if (key.objectid > device->devid)
1655 if (key.type != BTRFS_DEV_EXTENT_KEY)
1658 if (key.offset > search_start) {
1659 hole_size = key.offset - search_start;
1662 * Have to check before we set max_hole_start, otherwise
1663 * we could end up sending back this offset anyway.
1665 if (contains_pending_extent(transaction, device,
1668 if (key.offset >= search_start) {
1669 hole_size = key.offset - search_start;
1676 if (hole_size > max_hole_size) {
1677 max_hole_start = search_start;
1678 max_hole_size = hole_size;
1682 * If this free space is greater than which we need,
1683 * it must be the max free space that we have found
1684 * until now, so max_hole_start must point to the start
1685 * of this free space and the length of this free space
1686 * is stored in max_hole_size. Thus, we return
1687 * max_hole_start and max_hole_size and go back to the
1690 if (hole_size >= num_bytes) {
1696 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1697 extent_end = key.offset + btrfs_dev_extent_length(l,
1699 if (extent_end > search_start)
1700 search_start = extent_end;
1707 * At this point, search_start should be the end of
1708 * allocated dev extents, and when shrinking the device,
1709 * search_end may be smaller than search_start.
1711 if (search_end > search_start) {
1712 hole_size = search_end - search_start;
1714 if (contains_pending_extent(transaction, device, &search_start,
1716 btrfs_release_path(path);
1720 if (hole_size > max_hole_size) {
1721 max_hole_start = search_start;
1722 max_hole_size = hole_size;
1727 if (max_hole_size < num_bytes)
1733 btrfs_free_path(path);
1734 *start = max_hole_start;
1736 *len = max_hole_size;
1740 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1741 struct btrfs_device *device, u64 num_bytes,
1742 u64 *start, u64 *len)
1744 /* FIXME use last free of some kind */
1745 return find_free_dev_extent_start(trans->transaction, device,
1746 num_bytes, 0, start, len);
1749 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1750 struct btrfs_device *device,
1751 u64 start, u64 *dev_extent_len)
1753 struct btrfs_fs_info *fs_info = device->fs_info;
1754 struct btrfs_root *root = fs_info->dev_root;
1756 struct btrfs_path *path;
1757 struct btrfs_key key;
1758 struct btrfs_key found_key;
1759 struct extent_buffer *leaf = NULL;
1760 struct btrfs_dev_extent *extent = NULL;
1762 path = btrfs_alloc_path();
1766 key.objectid = device->devid;
1768 key.type = BTRFS_DEV_EXTENT_KEY;
1770 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1772 ret = btrfs_previous_item(root, path, key.objectid,
1773 BTRFS_DEV_EXTENT_KEY);
1776 leaf = path->nodes[0];
1777 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1778 extent = btrfs_item_ptr(leaf, path->slots[0],
1779 struct btrfs_dev_extent);
1780 BUG_ON(found_key.offset > start || found_key.offset +
1781 btrfs_dev_extent_length(leaf, extent) < start);
1783 btrfs_release_path(path);
1785 } else if (ret == 0) {
1786 leaf = path->nodes[0];
1787 extent = btrfs_item_ptr(leaf, path->slots[0],
1788 struct btrfs_dev_extent);
1790 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1794 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1796 ret = btrfs_del_item(trans, root, path);
1798 btrfs_handle_fs_error(fs_info, ret,
1799 "Failed to remove dev extent item");
1801 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1804 btrfs_free_path(path);
1808 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1809 struct btrfs_device *device,
1810 u64 chunk_offset, u64 start, u64 num_bytes)
1813 struct btrfs_path *path;
1814 struct btrfs_fs_info *fs_info = device->fs_info;
1815 struct btrfs_root *root = fs_info->dev_root;
1816 struct btrfs_dev_extent *extent;
1817 struct extent_buffer *leaf;
1818 struct btrfs_key key;
1820 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1821 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1822 path = btrfs_alloc_path();
1826 key.objectid = device->devid;
1828 key.type = BTRFS_DEV_EXTENT_KEY;
1829 ret = btrfs_insert_empty_item(trans, root, path, &key,
1834 leaf = path->nodes[0];
1835 extent = btrfs_item_ptr(leaf, path->slots[0],
1836 struct btrfs_dev_extent);
1837 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1838 BTRFS_CHUNK_TREE_OBJECTID);
1839 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1840 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1841 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1843 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1844 btrfs_mark_buffer_dirty(leaf);
1846 btrfs_free_path(path);
1850 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1852 struct extent_map_tree *em_tree;
1853 struct extent_map *em;
1857 em_tree = &fs_info->mapping_tree.map_tree;
1858 read_lock(&em_tree->lock);
1859 n = rb_last(&em_tree->map.rb_root);
1861 em = rb_entry(n, struct extent_map, rb_node);
1862 ret = em->start + em->len;
1864 read_unlock(&em_tree->lock);
1869 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1873 struct btrfs_key key;
1874 struct btrfs_key found_key;
1875 struct btrfs_path *path;
1877 path = btrfs_alloc_path();
1881 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1882 key.type = BTRFS_DEV_ITEM_KEY;
1883 key.offset = (u64)-1;
1885 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1889 BUG_ON(ret == 0); /* Corruption */
1891 ret = btrfs_previous_item(fs_info->chunk_root, path,
1892 BTRFS_DEV_ITEMS_OBJECTID,
1893 BTRFS_DEV_ITEM_KEY);
1897 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1899 *devid_ret = found_key.offset + 1;
1903 btrfs_free_path(path);
1908 * the device information is stored in the chunk root
1909 * the btrfs_device struct should be fully filled in
1911 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1912 struct btrfs_device *device)
1915 struct btrfs_path *path;
1916 struct btrfs_dev_item *dev_item;
1917 struct extent_buffer *leaf;
1918 struct btrfs_key key;
1921 path = btrfs_alloc_path();
1925 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1926 key.type = BTRFS_DEV_ITEM_KEY;
1927 key.offset = device->devid;
1929 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1930 &key, sizeof(*dev_item));
1934 leaf = path->nodes[0];
1935 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1937 btrfs_set_device_id(leaf, dev_item, device->devid);
1938 btrfs_set_device_generation(leaf, dev_item, 0);
1939 btrfs_set_device_type(leaf, dev_item, device->type);
1940 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1941 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1942 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1943 btrfs_set_device_total_bytes(leaf, dev_item,
1944 btrfs_device_get_disk_total_bytes(device));
1945 btrfs_set_device_bytes_used(leaf, dev_item,
1946 btrfs_device_get_bytes_used(device));
1947 btrfs_set_device_group(leaf, dev_item, 0);
1948 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1949 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1950 btrfs_set_device_start_offset(leaf, dev_item, 0);
1952 ptr = btrfs_device_uuid(dev_item);
1953 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1954 ptr = btrfs_device_fsid(dev_item);
1955 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1956 ptr, BTRFS_FSID_SIZE);
1957 btrfs_mark_buffer_dirty(leaf);
1961 btrfs_free_path(path);
1966 * Function to update ctime/mtime for a given device path.
1967 * Mainly used for ctime/mtime based probe like libblkid.
1969 static void update_dev_time(const char *path_name)
1973 filp = filp_open(path_name, O_RDWR, 0);
1976 file_update_time(filp);
1977 filp_close(filp, NULL);
1980 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1981 struct btrfs_device *device)
1983 struct btrfs_root *root = fs_info->chunk_root;
1985 struct btrfs_path *path;
1986 struct btrfs_key key;
1987 struct btrfs_trans_handle *trans;
1989 path = btrfs_alloc_path();
1993 trans = btrfs_start_transaction(root, 0);
1994 if (IS_ERR(trans)) {
1995 btrfs_free_path(path);
1996 return PTR_ERR(trans);
1998 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1999 key.type = BTRFS_DEV_ITEM_KEY;
2000 key.offset = device->devid;
2002 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2006 btrfs_abort_transaction(trans, ret);
2007 btrfs_end_transaction(trans);
2011 ret = btrfs_del_item(trans, root, path);
2013 btrfs_abort_transaction(trans, ret);
2014 btrfs_end_transaction(trans);
2018 btrfs_free_path(path);
2020 ret = btrfs_commit_transaction(trans);
2025 * Verify that @num_devices satisfies the RAID profile constraints in the whole
2026 * filesystem. It's up to the caller to adjust that number regarding eg. device
2029 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2037 seq = read_seqbegin(&fs_info->profiles_lock);
2039 all_avail = fs_info->avail_data_alloc_bits |
2040 fs_info->avail_system_alloc_bits |
2041 fs_info->avail_metadata_alloc_bits;
2042 } while (read_seqretry(&fs_info->profiles_lock, seq));
2044 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2045 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2048 if (num_devices < btrfs_raid_array[i].devs_min) {
2049 int ret = btrfs_raid_array[i].mindev_error;
2059 static struct btrfs_device * btrfs_find_next_active_device(
2060 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2062 struct btrfs_device *next_device;
2064 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2065 if (next_device != device &&
2066 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2067 && next_device->bdev)
2075 * Helper function to check if the given device is part of s_bdev / latest_bdev
2076 * and replace it with the provided or the next active device, in the context
2077 * where this function called, there should be always be another device (or
2078 * this_dev) which is active.
2080 void btrfs_assign_next_active_device(struct btrfs_device *device,
2081 struct btrfs_device *this_dev)
2083 struct btrfs_fs_info *fs_info = device->fs_info;
2084 struct btrfs_device *next_device;
2087 next_device = this_dev;
2089 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2091 ASSERT(next_device);
2093 if (fs_info->sb->s_bdev &&
2094 (fs_info->sb->s_bdev == device->bdev))
2095 fs_info->sb->s_bdev = next_device->bdev;
2097 if (fs_info->fs_devices->latest_bdev == device->bdev)
2098 fs_info->fs_devices->latest_bdev = next_device->bdev;
2102 * Return btrfs_fs_devices::num_devices excluding the device that's being
2103 * currently replaced.
2105 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2107 u64 num_devices = fs_info->fs_devices->num_devices;
2109 down_read(&fs_info->dev_replace.rwsem);
2110 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2111 ASSERT(num_devices > 1);
2114 up_read(&fs_info->dev_replace.rwsem);
2119 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2122 struct btrfs_device *device;
2123 struct btrfs_fs_devices *cur_devices;
2124 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2128 mutex_lock(&uuid_mutex);
2130 num_devices = btrfs_num_devices(fs_info);
2132 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2136 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2138 if (IS_ERR(device)) {
2139 if (PTR_ERR(device) == -ENOENT &&
2140 strcmp(device_path, "missing") == 0)
2141 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2143 ret = PTR_ERR(device);
2147 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2148 btrfs_warn_in_rcu(fs_info,
2149 "cannot remove device %s (devid %llu) due to active swapfile",
2150 rcu_str_deref(device->name), device->devid);
2155 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2156 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2160 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2161 fs_info->fs_devices->rw_devices == 1) {
2162 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2166 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2167 mutex_lock(&fs_info->chunk_mutex);
2168 list_del_init(&device->dev_alloc_list);
2169 device->fs_devices->rw_devices--;
2170 mutex_unlock(&fs_info->chunk_mutex);
2173 mutex_unlock(&uuid_mutex);
2174 ret = btrfs_shrink_device(device, 0);
2175 mutex_lock(&uuid_mutex);
2180 * TODO: the superblock still includes this device in its num_devices
2181 * counter although write_all_supers() is not locked out. This
2182 * could give a filesystem state which requires a degraded mount.
2184 ret = btrfs_rm_dev_item(fs_info, device);
2188 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2189 btrfs_scrub_cancel_dev(fs_info, device);
2192 * the device list mutex makes sure that we don't change
2193 * the device list while someone else is writing out all
2194 * the device supers. Whoever is writing all supers, should
2195 * lock the device list mutex before getting the number of
2196 * devices in the super block (super_copy). Conversely,
2197 * whoever updates the number of devices in the super block
2198 * (super_copy) should hold the device list mutex.
2202 * In normal cases the cur_devices == fs_devices. But in case
2203 * of deleting a seed device, the cur_devices should point to
2204 * its own fs_devices listed under the fs_devices->seed.
2206 cur_devices = device->fs_devices;
2207 mutex_lock(&fs_devices->device_list_mutex);
2208 list_del_rcu(&device->dev_list);
2210 cur_devices->num_devices--;
2211 cur_devices->total_devices--;
2212 /* Update total_devices of the parent fs_devices if it's seed */
2213 if (cur_devices != fs_devices)
2214 fs_devices->total_devices--;
2216 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2217 cur_devices->missing_devices--;
2219 btrfs_assign_next_active_device(device, NULL);
2222 cur_devices->open_devices--;
2223 /* remove sysfs entry */
2224 btrfs_sysfs_rm_device_link(fs_devices, device);
2227 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2228 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2229 mutex_unlock(&fs_devices->device_list_mutex);
2232 * at this point, the device is zero sized and detached from
2233 * the devices list. All that's left is to zero out the old
2234 * supers and free the device.
2236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2237 btrfs_scratch_superblocks(device->bdev, device->name->str);
2239 btrfs_close_bdev(device);
2240 call_rcu(&device->rcu, free_device_rcu);
2242 if (cur_devices->open_devices == 0) {
2243 while (fs_devices) {
2244 if (fs_devices->seed == cur_devices) {
2245 fs_devices->seed = cur_devices->seed;
2248 fs_devices = fs_devices->seed;
2250 cur_devices->seed = NULL;
2251 close_fs_devices(cur_devices);
2252 free_fs_devices(cur_devices);
2256 mutex_unlock(&uuid_mutex);
2260 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2261 mutex_lock(&fs_info->chunk_mutex);
2262 list_add(&device->dev_alloc_list,
2263 &fs_devices->alloc_list);
2264 device->fs_devices->rw_devices++;
2265 mutex_unlock(&fs_info->chunk_mutex);
2270 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2272 struct btrfs_fs_devices *fs_devices;
2274 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2277 * in case of fs with no seed, srcdev->fs_devices will point
2278 * to fs_devices of fs_info. However when the dev being replaced is
2279 * a seed dev it will point to the seed's local fs_devices. In short
2280 * srcdev will have its correct fs_devices in both the cases.
2282 fs_devices = srcdev->fs_devices;
2284 list_del_rcu(&srcdev->dev_list);
2285 list_del(&srcdev->dev_alloc_list);
2286 fs_devices->num_devices--;
2287 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2288 fs_devices->missing_devices--;
2290 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2291 fs_devices->rw_devices--;
2294 fs_devices->open_devices--;
2297 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2298 struct btrfs_device *srcdev)
2300 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2302 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2303 /* zero out the old super if it is writable */
2304 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2307 btrfs_close_bdev(srcdev);
2308 call_rcu(&srcdev->rcu, free_device_rcu);
2310 /* if this is no devs we rather delete the fs_devices */
2311 if (!fs_devices->num_devices) {
2312 struct btrfs_fs_devices *tmp_fs_devices;
2315 * On a mounted FS, num_devices can't be zero unless it's a
2316 * seed. In case of a seed device being replaced, the replace
2317 * target added to the sprout FS, so there will be no more
2318 * device left under the seed FS.
2320 ASSERT(fs_devices->seeding);
2322 tmp_fs_devices = fs_info->fs_devices;
2323 while (tmp_fs_devices) {
2324 if (tmp_fs_devices->seed == fs_devices) {
2325 tmp_fs_devices->seed = fs_devices->seed;
2328 tmp_fs_devices = tmp_fs_devices->seed;
2330 fs_devices->seed = NULL;
2331 close_fs_devices(fs_devices);
2332 free_fs_devices(fs_devices);
2336 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2338 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2341 mutex_lock(&fs_devices->device_list_mutex);
2343 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2346 fs_devices->open_devices--;
2348 fs_devices->num_devices--;
2350 btrfs_assign_next_active_device(tgtdev, NULL);
2352 list_del_rcu(&tgtdev->dev_list);
2354 mutex_unlock(&fs_devices->device_list_mutex);
2357 * The update_dev_time() with in btrfs_scratch_superblocks()
2358 * may lead to a call to btrfs_show_devname() which will try
2359 * to hold device_list_mutex. And here this device
2360 * is already out of device list, so we don't have to hold
2361 * the device_list_mutex lock.
2363 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2365 btrfs_close_bdev(tgtdev);
2366 call_rcu(&tgtdev->rcu, free_device_rcu);
2369 static struct btrfs_device *btrfs_find_device_by_path(
2370 struct btrfs_fs_info *fs_info, const char *device_path)
2373 struct btrfs_super_block *disk_super;
2376 struct block_device *bdev;
2377 struct buffer_head *bh;
2378 struct btrfs_device *device;
2380 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2381 fs_info->bdev_holder, 0, &bdev, &bh);
2383 return ERR_PTR(ret);
2384 disk_super = (struct btrfs_super_block *)bh->b_data;
2385 devid = btrfs_stack_device_id(&disk_super->dev_item);
2386 dev_uuid = disk_super->dev_item.uuid;
2387 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2388 device = btrfs_find_device(fs_info, devid, dev_uuid,
2389 disk_super->metadata_uuid);
2391 device = btrfs_find_device(fs_info, devid,
2392 dev_uuid, disk_super->fsid);
2396 device = ERR_PTR(-ENOENT);
2397 blkdev_put(bdev, FMODE_READ);
2401 static struct btrfs_device *btrfs_find_device_missing_or_by_path(
2402 struct btrfs_fs_info *fs_info, const char *device_path)
2404 struct btrfs_device *device = NULL;
2405 if (strcmp(device_path, "missing") == 0) {
2406 struct list_head *devices;
2407 struct btrfs_device *tmp;
2409 devices = &fs_info->fs_devices->devices;
2410 list_for_each_entry(tmp, devices, dev_list) {
2411 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2412 &tmp->dev_state) && !tmp->bdev) {
2419 return ERR_PTR(-ENOENT);
2421 device = btrfs_find_device_by_path(fs_info, device_path);
2428 * Lookup a device given by device id, or the path if the id is 0.
2430 struct btrfs_device *btrfs_find_device_by_devspec(
2431 struct btrfs_fs_info *fs_info, u64 devid, const char *devpath)
2433 struct btrfs_device *device;
2436 device = btrfs_find_device(fs_info, devid, NULL, NULL);
2438 return ERR_PTR(-ENOENT);
2440 if (!devpath || !devpath[0])
2441 return ERR_PTR(-EINVAL);
2442 device = btrfs_find_device_missing_or_by_path(fs_info, devpath);
2448 * does all the dirty work required for changing file system's UUID.
2450 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2452 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2453 struct btrfs_fs_devices *old_devices;
2454 struct btrfs_fs_devices *seed_devices;
2455 struct btrfs_super_block *disk_super = fs_info->super_copy;
2456 struct btrfs_device *device;
2459 lockdep_assert_held(&uuid_mutex);
2460 if (!fs_devices->seeding)
2463 seed_devices = alloc_fs_devices(NULL, NULL);
2464 if (IS_ERR(seed_devices))
2465 return PTR_ERR(seed_devices);
2467 old_devices = clone_fs_devices(fs_devices);
2468 if (IS_ERR(old_devices)) {
2469 kfree(seed_devices);
2470 return PTR_ERR(old_devices);
2473 list_add(&old_devices->fs_list, &fs_uuids);
2475 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2476 seed_devices->opened = 1;
2477 INIT_LIST_HEAD(&seed_devices->devices);
2478 INIT_LIST_HEAD(&seed_devices->alloc_list);
2479 mutex_init(&seed_devices->device_list_mutex);
2481 mutex_lock(&fs_devices->device_list_mutex);
2482 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2484 list_for_each_entry(device, &seed_devices->devices, dev_list)
2485 device->fs_devices = seed_devices;
2487 mutex_lock(&fs_info->chunk_mutex);
2488 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2489 mutex_unlock(&fs_info->chunk_mutex);
2491 fs_devices->seeding = 0;
2492 fs_devices->num_devices = 0;
2493 fs_devices->open_devices = 0;
2494 fs_devices->missing_devices = 0;
2495 fs_devices->rotating = 0;
2496 fs_devices->seed = seed_devices;
2498 generate_random_uuid(fs_devices->fsid);
2499 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2500 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2501 mutex_unlock(&fs_devices->device_list_mutex);
2503 super_flags = btrfs_super_flags(disk_super) &
2504 ~BTRFS_SUPER_FLAG_SEEDING;
2505 btrfs_set_super_flags(disk_super, super_flags);
2511 * Store the expected generation for seed devices in device items.
2513 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2514 struct btrfs_fs_info *fs_info)
2516 struct btrfs_root *root = fs_info->chunk_root;
2517 struct btrfs_path *path;
2518 struct extent_buffer *leaf;
2519 struct btrfs_dev_item *dev_item;
2520 struct btrfs_device *device;
2521 struct btrfs_key key;
2522 u8 fs_uuid[BTRFS_FSID_SIZE];
2523 u8 dev_uuid[BTRFS_UUID_SIZE];
2527 path = btrfs_alloc_path();
2531 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2533 key.type = BTRFS_DEV_ITEM_KEY;
2536 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2540 leaf = path->nodes[0];
2542 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2543 ret = btrfs_next_leaf(root, path);
2548 leaf = path->nodes[0];
2549 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2550 btrfs_release_path(path);
2554 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2555 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2556 key.type != BTRFS_DEV_ITEM_KEY)
2559 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2560 struct btrfs_dev_item);
2561 devid = btrfs_device_id(leaf, dev_item);
2562 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2564 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2566 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2567 BUG_ON(!device); /* Logic error */
2569 if (device->fs_devices->seeding) {
2570 btrfs_set_device_generation(leaf, dev_item,
2571 device->generation);
2572 btrfs_mark_buffer_dirty(leaf);
2580 btrfs_free_path(path);
2584 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2586 struct btrfs_root *root = fs_info->dev_root;
2587 struct request_queue *q;
2588 struct btrfs_trans_handle *trans;
2589 struct btrfs_device *device;
2590 struct block_device *bdev;
2591 struct super_block *sb = fs_info->sb;
2592 struct rcu_string *name;
2593 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2594 u64 orig_super_total_bytes;
2595 u64 orig_super_num_devices;
2596 int seeding_dev = 0;
2598 bool unlocked = false;
2600 if (sb_rdonly(sb) && !fs_devices->seeding)
2603 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2604 fs_info->bdev_holder);
2606 return PTR_ERR(bdev);
2608 if (fs_devices->seeding) {
2610 down_write(&sb->s_umount);
2611 mutex_lock(&uuid_mutex);
2614 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2616 mutex_lock(&fs_devices->device_list_mutex);
2617 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2618 if (device->bdev == bdev) {
2621 &fs_devices->device_list_mutex);
2625 mutex_unlock(&fs_devices->device_list_mutex);
2627 device = btrfs_alloc_device(fs_info, NULL, NULL);
2628 if (IS_ERR(device)) {
2629 /* we can safely leave the fs_devices entry around */
2630 ret = PTR_ERR(device);
2634 name = rcu_string_strdup(device_path, GFP_KERNEL);
2637 goto error_free_device;
2639 rcu_assign_pointer(device->name, name);
2641 trans = btrfs_start_transaction(root, 0);
2642 if (IS_ERR(trans)) {
2643 ret = PTR_ERR(trans);
2644 goto error_free_device;
2647 q = bdev_get_queue(bdev);
2648 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2649 device->generation = trans->transid;
2650 device->io_width = fs_info->sectorsize;
2651 device->io_align = fs_info->sectorsize;
2652 device->sector_size = fs_info->sectorsize;
2653 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2654 fs_info->sectorsize);
2655 device->disk_total_bytes = device->total_bytes;
2656 device->commit_total_bytes = device->total_bytes;
2657 device->fs_info = fs_info;
2658 device->bdev = bdev;
2659 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2660 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2661 device->mode = FMODE_EXCL;
2662 device->dev_stats_valid = 1;
2663 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2666 sb->s_flags &= ~SB_RDONLY;
2667 ret = btrfs_prepare_sprout(fs_info);
2669 btrfs_abort_transaction(trans, ret);
2674 device->fs_devices = fs_devices;
2676 mutex_lock(&fs_devices->device_list_mutex);
2677 mutex_lock(&fs_info->chunk_mutex);
2678 list_add_rcu(&device->dev_list, &fs_devices->devices);
2679 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2680 fs_devices->num_devices++;
2681 fs_devices->open_devices++;
2682 fs_devices->rw_devices++;
2683 fs_devices->total_devices++;
2684 fs_devices->total_rw_bytes += device->total_bytes;
2686 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2688 if (!blk_queue_nonrot(q))
2689 fs_devices->rotating = 1;
2691 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2692 btrfs_set_super_total_bytes(fs_info->super_copy,
2693 round_down(orig_super_total_bytes + device->total_bytes,
2694 fs_info->sectorsize));
2696 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2697 btrfs_set_super_num_devices(fs_info->super_copy,
2698 orig_super_num_devices + 1);
2700 /* add sysfs device entry */
2701 btrfs_sysfs_add_device_link(fs_devices, device);
2704 * we've got more storage, clear any full flags on the space
2707 btrfs_clear_space_info_full(fs_info);
2709 mutex_unlock(&fs_info->chunk_mutex);
2710 mutex_unlock(&fs_devices->device_list_mutex);
2713 mutex_lock(&fs_info->chunk_mutex);
2714 ret = init_first_rw_device(trans, fs_info);
2715 mutex_unlock(&fs_info->chunk_mutex);
2717 btrfs_abort_transaction(trans, ret);
2722 ret = btrfs_add_dev_item(trans, device);
2724 btrfs_abort_transaction(trans, ret);
2729 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2731 ret = btrfs_finish_sprout(trans, fs_info);
2733 btrfs_abort_transaction(trans, ret);
2737 /* Sprouting would change fsid of the mounted root,
2738 * so rename the fsid on the sysfs
2740 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2741 fs_info->fs_devices->fsid);
2742 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2744 "sysfs: failed to create fsid for sprout");
2747 ret = btrfs_commit_transaction(trans);
2750 mutex_unlock(&uuid_mutex);
2751 up_write(&sb->s_umount);
2754 if (ret) /* transaction commit */
2757 ret = btrfs_relocate_sys_chunks(fs_info);
2759 btrfs_handle_fs_error(fs_info, ret,
2760 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2761 trans = btrfs_attach_transaction(root);
2762 if (IS_ERR(trans)) {
2763 if (PTR_ERR(trans) == -ENOENT)
2765 ret = PTR_ERR(trans);
2769 ret = btrfs_commit_transaction(trans);
2772 /* Update ctime/mtime for libblkid */
2773 update_dev_time(device_path);
2777 btrfs_sysfs_rm_device_link(fs_devices, device);
2778 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2779 mutex_lock(&fs_info->chunk_mutex);
2780 list_del_rcu(&device->dev_list);
2781 list_del(&device->dev_alloc_list);
2782 fs_info->fs_devices->num_devices--;
2783 fs_info->fs_devices->open_devices--;
2784 fs_info->fs_devices->rw_devices--;
2785 fs_info->fs_devices->total_devices--;
2786 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2787 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2788 btrfs_set_super_total_bytes(fs_info->super_copy,
2789 orig_super_total_bytes);
2790 btrfs_set_super_num_devices(fs_info->super_copy,
2791 orig_super_num_devices);
2792 mutex_unlock(&fs_info->chunk_mutex);
2793 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2796 sb->s_flags |= SB_RDONLY;
2798 btrfs_end_transaction(trans);
2800 btrfs_free_device(device);
2802 blkdev_put(bdev, FMODE_EXCL);
2803 if (seeding_dev && !unlocked) {
2804 mutex_unlock(&uuid_mutex);
2805 up_write(&sb->s_umount);
2810 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2811 struct btrfs_device *device)
2814 struct btrfs_path *path;
2815 struct btrfs_root *root = device->fs_info->chunk_root;
2816 struct btrfs_dev_item *dev_item;
2817 struct extent_buffer *leaf;
2818 struct btrfs_key key;
2820 path = btrfs_alloc_path();
2824 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2825 key.type = BTRFS_DEV_ITEM_KEY;
2826 key.offset = device->devid;
2828 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2837 leaf = path->nodes[0];
2838 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2840 btrfs_set_device_id(leaf, dev_item, device->devid);
2841 btrfs_set_device_type(leaf, dev_item, device->type);
2842 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2843 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2844 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2845 btrfs_set_device_total_bytes(leaf, dev_item,
2846 btrfs_device_get_disk_total_bytes(device));
2847 btrfs_set_device_bytes_used(leaf, dev_item,
2848 btrfs_device_get_bytes_used(device));
2849 btrfs_mark_buffer_dirty(leaf);
2852 btrfs_free_path(path);
2856 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2857 struct btrfs_device *device, u64 new_size)
2859 struct btrfs_fs_info *fs_info = device->fs_info;
2860 struct btrfs_super_block *super_copy = fs_info->super_copy;
2861 struct btrfs_fs_devices *fs_devices;
2865 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2868 new_size = round_down(new_size, fs_info->sectorsize);
2870 mutex_lock(&fs_info->chunk_mutex);
2871 old_total = btrfs_super_total_bytes(super_copy);
2872 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2874 if (new_size <= device->total_bytes ||
2875 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2876 mutex_unlock(&fs_info->chunk_mutex);
2880 fs_devices = fs_info->fs_devices;
2882 btrfs_set_super_total_bytes(super_copy,
2883 round_down(old_total + diff, fs_info->sectorsize));
2884 device->fs_devices->total_rw_bytes += diff;
2886 btrfs_device_set_total_bytes(device, new_size);
2887 btrfs_device_set_disk_total_bytes(device, new_size);
2888 btrfs_clear_space_info_full(device->fs_info);
2889 if (list_empty(&device->resized_list))
2890 list_add_tail(&device->resized_list,
2891 &fs_devices->resized_devices);
2892 mutex_unlock(&fs_info->chunk_mutex);
2894 return btrfs_update_device(trans, device);
2897 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2899 struct btrfs_fs_info *fs_info = trans->fs_info;
2900 struct btrfs_root *root = fs_info->chunk_root;
2902 struct btrfs_path *path;
2903 struct btrfs_key key;
2905 path = btrfs_alloc_path();
2909 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2910 key.offset = chunk_offset;
2911 key.type = BTRFS_CHUNK_ITEM_KEY;
2913 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2916 else if (ret > 0) { /* Logic error or corruption */
2917 btrfs_handle_fs_error(fs_info, -ENOENT,
2918 "Failed lookup while freeing chunk.");
2923 ret = btrfs_del_item(trans, root, path);
2925 btrfs_handle_fs_error(fs_info, ret,
2926 "Failed to delete chunk item.");
2928 btrfs_free_path(path);
2932 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2934 struct btrfs_super_block *super_copy = fs_info->super_copy;
2935 struct btrfs_disk_key *disk_key;
2936 struct btrfs_chunk *chunk;
2943 struct btrfs_key key;
2945 mutex_lock(&fs_info->chunk_mutex);
2946 array_size = btrfs_super_sys_array_size(super_copy);
2948 ptr = super_copy->sys_chunk_array;
2951 while (cur < array_size) {
2952 disk_key = (struct btrfs_disk_key *)ptr;
2953 btrfs_disk_key_to_cpu(&key, disk_key);
2955 len = sizeof(*disk_key);
2957 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2958 chunk = (struct btrfs_chunk *)(ptr + len);
2959 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2960 len += btrfs_chunk_item_size(num_stripes);
2965 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2966 key.offset == chunk_offset) {
2967 memmove(ptr, ptr + len, array_size - (cur + len));
2969 btrfs_set_super_sys_array_size(super_copy, array_size);
2975 mutex_unlock(&fs_info->chunk_mutex);
2980 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2981 * @logical: Logical block offset in bytes.
2982 * @length: Length of extent in bytes.
2984 * Return: Chunk mapping or ERR_PTR.
2986 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2987 u64 logical, u64 length)
2989 struct extent_map_tree *em_tree;
2990 struct extent_map *em;
2992 em_tree = &fs_info->mapping_tree.map_tree;
2993 read_lock(&em_tree->lock);
2994 em = lookup_extent_mapping(em_tree, logical, length);
2995 read_unlock(&em_tree->lock);
2998 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3000 return ERR_PTR(-EINVAL);
3003 if (em->start > logical || em->start + em->len < logical) {
3005 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3006 logical, length, em->start, em->start + em->len);
3007 free_extent_map(em);
3008 return ERR_PTR(-EINVAL);
3011 /* callers are responsible for dropping em's ref. */
3015 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3017 struct btrfs_fs_info *fs_info = trans->fs_info;
3018 struct extent_map *em;
3019 struct map_lookup *map;
3020 u64 dev_extent_len = 0;
3022 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3024 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3027 * This is a logic error, but we don't want to just rely on the
3028 * user having built with ASSERT enabled, so if ASSERT doesn't
3029 * do anything we still error out.
3034 map = em->map_lookup;
3035 mutex_lock(&fs_info->chunk_mutex);
3036 check_system_chunk(trans, map->type);
3037 mutex_unlock(&fs_info->chunk_mutex);
3040 * Take the device list mutex to prevent races with the final phase of
3041 * a device replace operation that replaces the device object associated
3042 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
3044 mutex_lock(&fs_devices->device_list_mutex);
3045 for (i = 0; i < map->num_stripes; i++) {
3046 struct btrfs_device *device = map->stripes[i].dev;
3047 ret = btrfs_free_dev_extent(trans, device,
3048 map->stripes[i].physical,
3051 mutex_unlock(&fs_devices->device_list_mutex);
3052 btrfs_abort_transaction(trans, ret);
3056 if (device->bytes_used > 0) {
3057 mutex_lock(&fs_info->chunk_mutex);
3058 btrfs_device_set_bytes_used(device,
3059 device->bytes_used - dev_extent_len);
3060 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3061 btrfs_clear_space_info_full(fs_info);
3062 mutex_unlock(&fs_info->chunk_mutex);
3065 ret = btrfs_update_device(trans, device);
3067 mutex_unlock(&fs_devices->device_list_mutex);
3068 btrfs_abort_transaction(trans, ret);
3072 mutex_unlock(&fs_devices->device_list_mutex);
3074 ret = btrfs_free_chunk(trans, chunk_offset);
3076 btrfs_abort_transaction(trans, ret);
3080 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3082 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3083 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3085 btrfs_abort_transaction(trans, ret);
3090 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3092 btrfs_abort_transaction(trans, ret);
3098 free_extent_map(em);
3102 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3104 struct btrfs_root *root = fs_info->chunk_root;
3105 struct btrfs_trans_handle *trans;
3109 * Prevent races with automatic removal of unused block groups.
3110 * After we relocate and before we remove the chunk with offset
3111 * chunk_offset, automatic removal of the block group can kick in,
3112 * resulting in a failure when calling btrfs_remove_chunk() below.
3114 * Make sure to acquire this mutex before doing a tree search (dev
3115 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3116 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3117 * we release the path used to search the chunk/dev tree and before
3118 * the current task acquires this mutex and calls us.
3120 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3122 ret = btrfs_can_relocate(fs_info, chunk_offset);
3126 /* step one, relocate all the extents inside this chunk */
3127 btrfs_scrub_pause(fs_info);
3128 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3129 btrfs_scrub_continue(fs_info);
3134 * We add the kobjects here (and after forcing data chunk creation)
3135 * since relocation is the only place we'll create chunks of a new
3136 * type at runtime. The only place where we'll remove the last
3137 * chunk of a type is the call immediately below this one. Even
3138 * so, we're protected against races with the cleaner thread since
3139 * we're covered by the delete_unused_bgs_mutex.
3141 btrfs_add_raid_kobjects(fs_info);
3143 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3145 if (IS_ERR(trans)) {
3146 ret = PTR_ERR(trans);
3147 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3152 * step two, delete the device extents and the
3153 * chunk tree entries
3155 ret = btrfs_remove_chunk(trans, chunk_offset);
3156 btrfs_end_transaction(trans);
3160 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3162 struct btrfs_root *chunk_root = fs_info->chunk_root;
3163 struct btrfs_path *path;
3164 struct extent_buffer *leaf;
3165 struct btrfs_chunk *chunk;
3166 struct btrfs_key key;
3167 struct btrfs_key found_key;
3169 bool retried = false;
3173 path = btrfs_alloc_path();
3178 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3179 key.offset = (u64)-1;
3180 key.type = BTRFS_CHUNK_ITEM_KEY;
3183 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3184 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3186 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3189 BUG_ON(ret == 0); /* Corruption */
3191 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3194 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3200 leaf = path->nodes[0];
3201 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3203 chunk = btrfs_item_ptr(leaf, path->slots[0],
3204 struct btrfs_chunk);
3205 chunk_type = btrfs_chunk_type(leaf, chunk);
3206 btrfs_release_path(path);
3208 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3209 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3215 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3217 if (found_key.offset == 0)
3219 key.offset = found_key.offset - 1;
3222 if (failed && !retried) {
3226 } else if (WARN_ON(failed && retried)) {
3230 btrfs_free_path(path);
3235 * return 1 : allocate a data chunk successfully,
3236 * return <0: errors during allocating a data chunk,
3237 * return 0 : no need to allocate a data chunk.
3239 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3242 struct btrfs_block_group_cache *cache;
3246 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3248 chunk_type = cache->flags;
3249 btrfs_put_block_group(cache);
3251 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3252 spin_lock(&fs_info->data_sinfo->lock);
3253 bytes_used = fs_info->data_sinfo->bytes_used;
3254 spin_unlock(&fs_info->data_sinfo->lock);
3257 struct btrfs_trans_handle *trans;
3260 trans = btrfs_join_transaction(fs_info->tree_root);
3262 return PTR_ERR(trans);
3264 ret = btrfs_force_chunk_alloc(trans,
3265 BTRFS_BLOCK_GROUP_DATA);
3266 btrfs_end_transaction(trans);
3270 btrfs_add_raid_kobjects(fs_info);
3278 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3279 struct btrfs_balance_control *bctl)
3281 struct btrfs_root *root = fs_info->tree_root;
3282 struct btrfs_trans_handle *trans;
3283 struct btrfs_balance_item *item;
3284 struct btrfs_disk_balance_args disk_bargs;
3285 struct btrfs_path *path;
3286 struct extent_buffer *leaf;
3287 struct btrfs_key key;
3290 path = btrfs_alloc_path();
3294 trans = btrfs_start_transaction(root, 0);
3295 if (IS_ERR(trans)) {
3296 btrfs_free_path(path);
3297 return PTR_ERR(trans);
3300 key.objectid = BTRFS_BALANCE_OBJECTID;
3301 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3304 ret = btrfs_insert_empty_item(trans, root, path, &key,
3309 leaf = path->nodes[0];
3310 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3312 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3314 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3315 btrfs_set_balance_data(leaf, item, &disk_bargs);
3316 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3317 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3318 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3319 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3321 btrfs_set_balance_flags(leaf, item, bctl->flags);
3323 btrfs_mark_buffer_dirty(leaf);
3325 btrfs_free_path(path);
3326 err = btrfs_commit_transaction(trans);
3332 static int del_balance_item(struct btrfs_fs_info *fs_info)
3334 struct btrfs_root *root = fs_info->tree_root;
3335 struct btrfs_trans_handle *trans;
3336 struct btrfs_path *path;
3337 struct btrfs_key key;
3340 path = btrfs_alloc_path();
3344 trans = btrfs_start_transaction(root, 0);
3345 if (IS_ERR(trans)) {
3346 btrfs_free_path(path);
3347 return PTR_ERR(trans);
3350 key.objectid = BTRFS_BALANCE_OBJECTID;
3351 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3354 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3362 ret = btrfs_del_item(trans, root, path);
3364 btrfs_free_path(path);
3365 err = btrfs_commit_transaction(trans);
3372 * This is a heuristic used to reduce the number of chunks balanced on
3373 * resume after balance was interrupted.
3375 static void update_balance_args(struct btrfs_balance_control *bctl)
3378 * Turn on soft mode for chunk types that were being converted.
3380 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3381 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3382 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3383 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3384 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3385 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3388 * Turn on usage filter if is not already used. The idea is
3389 * that chunks that we have already balanced should be
3390 * reasonably full. Don't do it for chunks that are being
3391 * converted - that will keep us from relocating unconverted
3392 * (albeit full) chunks.
3394 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3395 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3396 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3397 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3398 bctl->data.usage = 90;
3400 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3401 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3402 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3403 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3404 bctl->sys.usage = 90;
3406 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3407 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3408 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3409 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3410 bctl->meta.usage = 90;
3415 * Clear the balance status in fs_info and delete the balance item from disk.
3417 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3419 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3422 BUG_ON(!fs_info->balance_ctl);
3424 spin_lock(&fs_info->balance_lock);
3425 fs_info->balance_ctl = NULL;
3426 spin_unlock(&fs_info->balance_lock);
3429 ret = del_balance_item(fs_info);
3431 btrfs_handle_fs_error(fs_info, ret, NULL);
3435 * Balance filters. Return 1 if chunk should be filtered out
3436 * (should not be balanced).
3438 static int chunk_profiles_filter(u64 chunk_type,
3439 struct btrfs_balance_args *bargs)
3441 chunk_type = chunk_to_extended(chunk_type) &
3442 BTRFS_EXTENDED_PROFILE_MASK;
3444 if (bargs->profiles & chunk_type)
3450 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3451 struct btrfs_balance_args *bargs)
3453 struct btrfs_block_group_cache *cache;
3455 u64 user_thresh_min;
3456 u64 user_thresh_max;
3459 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3460 chunk_used = btrfs_block_group_used(&cache->item);
3462 if (bargs->usage_min == 0)
3463 user_thresh_min = 0;
3465 user_thresh_min = div_factor_fine(cache->key.offset,
3468 if (bargs->usage_max == 0)
3469 user_thresh_max = 1;
3470 else if (bargs->usage_max > 100)
3471 user_thresh_max = cache->key.offset;
3473 user_thresh_max = div_factor_fine(cache->key.offset,
3476 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3479 btrfs_put_block_group(cache);
3483 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3484 u64 chunk_offset, struct btrfs_balance_args *bargs)
3486 struct btrfs_block_group_cache *cache;
3487 u64 chunk_used, user_thresh;
3490 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3491 chunk_used = btrfs_block_group_used(&cache->item);
3493 if (bargs->usage_min == 0)
3495 else if (bargs->usage > 100)
3496 user_thresh = cache->key.offset;
3498 user_thresh = div_factor_fine(cache->key.offset,
3501 if (chunk_used < user_thresh)
3504 btrfs_put_block_group(cache);
3508 static int chunk_devid_filter(struct extent_buffer *leaf,
3509 struct btrfs_chunk *chunk,
3510 struct btrfs_balance_args *bargs)
3512 struct btrfs_stripe *stripe;
3513 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3516 for (i = 0; i < num_stripes; i++) {
3517 stripe = btrfs_stripe_nr(chunk, i);
3518 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3525 /* [pstart, pend) */
3526 static int chunk_drange_filter(struct extent_buffer *leaf,
3527 struct btrfs_chunk *chunk,
3528 struct btrfs_balance_args *bargs)
3530 struct btrfs_stripe *stripe;
3531 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3537 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3540 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3541 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3542 factor = num_stripes / 2;
3543 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3544 factor = num_stripes - 1;
3545 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3546 factor = num_stripes - 2;
3548 factor = num_stripes;
3551 for (i = 0; i < num_stripes; i++) {
3552 stripe = btrfs_stripe_nr(chunk, i);
3553 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3556 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3557 stripe_length = btrfs_chunk_length(leaf, chunk);
3558 stripe_length = div_u64(stripe_length, factor);
3560 if (stripe_offset < bargs->pend &&
3561 stripe_offset + stripe_length > bargs->pstart)
3568 /* [vstart, vend) */
3569 static int chunk_vrange_filter(struct extent_buffer *leaf,
3570 struct btrfs_chunk *chunk,
3572 struct btrfs_balance_args *bargs)
3574 if (chunk_offset < bargs->vend &&
3575 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3576 /* at least part of the chunk is inside this vrange */
3582 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3583 struct btrfs_chunk *chunk,
3584 struct btrfs_balance_args *bargs)
3586 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3588 if (bargs->stripes_min <= num_stripes
3589 && num_stripes <= bargs->stripes_max)
3595 static int chunk_soft_convert_filter(u64 chunk_type,
3596 struct btrfs_balance_args *bargs)
3598 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3601 chunk_type = chunk_to_extended(chunk_type) &
3602 BTRFS_EXTENDED_PROFILE_MASK;
3604 if (bargs->target == chunk_type)
3610 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3611 struct extent_buffer *leaf,
3612 struct btrfs_chunk *chunk, u64 chunk_offset)
3614 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3615 struct btrfs_balance_args *bargs = NULL;
3616 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3619 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3620 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3624 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3625 bargs = &bctl->data;
3626 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3629 bargs = &bctl->meta;
3631 /* profiles filter */
3632 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3633 chunk_profiles_filter(chunk_type, bargs)) {
3638 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3639 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3641 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3642 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3647 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3648 chunk_devid_filter(leaf, chunk, bargs)) {
3652 /* drange filter, makes sense only with devid filter */
3653 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3654 chunk_drange_filter(leaf, chunk, bargs)) {
3659 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3660 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3664 /* stripes filter */
3665 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3666 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3670 /* soft profile changing mode */
3671 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3672 chunk_soft_convert_filter(chunk_type, bargs)) {
3677 * limited by count, must be the last filter
3679 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3680 if (bargs->limit == 0)
3684 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3686 * Same logic as the 'limit' filter; the minimum cannot be
3687 * determined here because we do not have the global information
3688 * about the count of all chunks that satisfy the filters.
3690 if (bargs->limit_max == 0)
3699 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3701 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3702 struct btrfs_root *chunk_root = fs_info->chunk_root;
3703 struct btrfs_root *dev_root = fs_info->dev_root;
3704 struct list_head *devices;
3705 struct btrfs_device *device;
3709 struct btrfs_chunk *chunk;
3710 struct btrfs_path *path = NULL;
3711 struct btrfs_key key;
3712 struct btrfs_key found_key;
3713 struct btrfs_trans_handle *trans;
3714 struct extent_buffer *leaf;
3717 int enospc_errors = 0;
3718 bool counting = true;
3719 /* The single value limit and min/max limits use the same bytes in the */
3720 u64 limit_data = bctl->data.limit;
3721 u64 limit_meta = bctl->meta.limit;
3722 u64 limit_sys = bctl->sys.limit;
3726 int chunk_reserved = 0;
3728 /* step one make some room on all the devices */
3729 devices = &fs_info->fs_devices->devices;
3730 list_for_each_entry(device, devices, dev_list) {
3731 old_size = btrfs_device_get_total_bytes(device);
3732 size_to_free = div_factor(old_size, 1);
3733 size_to_free = min_t(u64, size_to_free, SZ_1M);
3734 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3735 btrfs_device_get_total_bytes(device) -
3736 btrfs_device_get_bytes_used(device) > size_to_free ||
3737 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3740 ret = btrfs_shrink_device(device, old_size - size_to_free);
3744 /* btrfs_shrink_device never returns ret > 0 */
3749 trans = btrfs_start_transaction(dev_root, 0);
3750 if (IS_ERR(trans)) {
3751 ret = PTR_ERR(trans);
3752 btrfs_info_in_rcu(fs_info,
3753 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3754 rcu_str_deref(device->name), ret,
3755 old_size, old_size - size_to_free);
3759 ret = btrfs_grow_device(trans, device, old_size);
3761 btrfs_end_transaction(trans);
3762 /* btrfs_grow_device never returns ret > 0 */
3764 btrfs_info_in_rcu(fs_info,
3765 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3766 rcu_str_deref(device->name), ret,
3767 old_size, old_size - size_to_free);
3771 btrfs_end_transaction(trans);
3774 /* step two, relocate all the chunks */
3775 path = btrfs_alloc_path();
3781 /* zero out stat counters */
3782 spin_lock(&fs_info->balance_lock);
3783 memset(&bctl->stat, 0, sizeof(bctl->stat));
3784 spin_unlock(&fs_info->balance_lock);
3788 * The single value limit and min/max limits use the same bytes
3791 bctl->data.limit = limit_data;
3792 bctl->meta.limit = limit_meta;
3793 bctl->sys.limit = limit_sys;
3795 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3796 key.offset = (u64)-1;
3797 key.type = BTRFS_CHUNK_ITEM_KEY;
3800 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3801 atomic_read(&fs_info->balance_cancel_req)) {
3806 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3807 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3809 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3814 * this shouldn't happen, it means the last relocate
3818 BUG(); /* FIXME break ? */
3820 ret = btrfs_previous_item(chunk_root, path, 0,
3821 BTRFS_CHUNK_ITEM_KEY);
3823 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3828 leaf = path->nodes[0];
3829 slot = path->slots[0];
3830 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3832 if (found_key.objectid != key.objectid) {
3833 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3837 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3838 chunk_type = btrfs_chunk_type(leaf, chunk);
3841 spin_lock(&fs_info->balance_lock);
3842 bctl->stat.considered++;
3843 spin_unlock(&fs_info->balance_lock);
3846 ret = should_balance_chunk(fs_info, leaf, chunk,
3849 btrfs_release_path(path);
3851 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3856 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3857 spin_lock(&fs_info->balance_lock);
3858 bctl->stat.expected++;
3859 spin_unlock(&fs_info->balance_lock);
3861 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3863 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3865 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3872 * Apply limit_min filter, no need to check if the LIMITS
3873 * filter is used, limit_min is 0 by default
3875 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3876 count_data < bctl->data.limit_min)
3877 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3878 count_meta < bctl->meta.limit_min)
3879 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3880 count_sys < bctl->sys.limit_min)) {
3881 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3885 if (!chunk_reserved) {
3887 * We may be relocating the only data chunk we have,
3888 * which could potentially end up with losing data's
3889 * raid profile, so lets allocate an empty one in
3892 ret = btrfs_may_alloc_data_chunk(fs_info,
3895 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3897 } else if (ret == 1) {
3902 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3903 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3904 if (ret == -ENOSPC) {
3906 } else if (ret == -ETXTBSY) {
3908 "skipping relocation of block group %llu due to active swapfile",
3914 spin_lock(&fs_info->balance_lock);
3915 bctl->stat.completed++;
3916 spin_unlock(&fs_info->balance_lock);
3919 if (found_key.offset == 0)
3921 key.offset = found_key.offset - 1;
3925 btrfs_release_path(path);
3930 btrfs_free_path(path);
3931 if (enospc_errors) {
3932 btrfs_info(fs_info, "%d enospc errors during balance",
3942 * alloc_profile_is_valid - see if a given profile is valid and reduced
3943 * @flags: profile to validate
3944 * @extended: if true @flags is treated as an extended profile
3946 static int alloc_profile_is_valid(u64 flags, int extended)
3948 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3949 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3951 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3953 /* 1) check that all other bits are zeroed */
3957 /* 2) see if profile is reduced */
3959 return !extended; /* "0" is valid for usual profiles */
3961 /* true if exactly one bit set */
3962 return is_power_of_2(flags);
3965 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3967 /* cancel requested || normal exit path */
3968 return atomic_read(&fs_info->balance_cancel_req) ||
3969 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3970 atomic_read(&fs_info->balance_cancel_req) == 0);
3973 /* Non-zero return value signifies invalidity */
3974 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3977 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3978 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3979 (bctl_arg->target & ~allowed)));
3983 * Fill @buf with textual description of balance filter flags @bargs, up to
3984 * @size_buf including the terminating null. The output may be trimmed if it
3985 * does not fit into the provided buffer.
3987 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3991 u32 size_bp = size_buf;
3993 u64 flags = bargs->flags;
3994 char tmp_buf[128] = {'\0'};
3999 #define CHECK_APPEND_NOARG(a) \
4001 ret = snprintf(bp, size_bp, (a)); \
4002 if (ret < 0 || ret >= size_bp) \
4003 goto out_overflow; \
4008 #define CHECK_APPEND_1ARG(a, v1) \
4010 ret = snprintf(bp, size_bp, (a), (v1)); \
4011 if (ret < 0 || ret >= size_bp) \
4012 goto out_overflow; \
4017 #define CHECK_APPEND_2ARG(a, v1, v2) \
4019 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4020 if (ret < 0 || ret >= size_bp) \
4021 goto out_overflow; \
4026 if (flags & BTRFS_BALANCE_ARGS_CONVERT) {
4027 int index = btrfs_bg_flags_to_raid_index(bargs->target);
4029 CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index));
4032 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4033 CHECK_APPEND_NOARG("soft,");
4035 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4036 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4038 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4041 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4042 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4044 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4045 CHECK_APPEND_2ARG("usage=%u..%u,",
4046 bargs->usage_min, bargs->usage_max);
4048 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4049 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4051 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4052 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4053 bargs->pstart, bargs->pend);
4055 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4056 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4057 bargs->vstart, bargs->vend);
4059 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4060 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4062 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4063 CHECK_APPEND_2ARG("limit=%u..%u,",
4064 bargs->limit_min, bargs->limit_max);
4066 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4067 CHECK_APPEND_2ARG("stripes=%u..%u,",
4068 bargs->stripes_min, bargs->stripes_max);
4070 #undef CHECK_APPEND_2ARG
4071 #undef CHECK_APPEND_1ARG
4072 #undef CHECK_APPEND_NOARG
4076 if (size_bp < size_buf)
4077 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4082 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4084 u32 size_buf = 1024;
4085 char tmp_buf[192] = {'\0'};
4088 u32 size_bp = size_buf;
4090 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4092 buf = kzalloc(size_buf, GFP_KERNEL);
4098 #define CHECK_APPEND_1ARG(a, v1) \
4100 ret = snprintf(bp, size_bp, (a), (v1)); \
4101 if (ret < 0 || ret >= size_bp) \
4102 goto out_overflow; \
4107 if (bctl->flags & BTRFS_BALANCE_FORCE)
4108 CHECK_APPEND_1ARG("%s", "-f ");
4110 if (bctl->flags & BTRFS_BALANCE_DATA) {
4111 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4112 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4115 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4116 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4117 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4120 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4121 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4122 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4125 #undef CHECK_APPEND_1ARG
4129 if (size_bp < size_buf)
4130 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4131 btrfs_info(fs_info, "balance: %s %s",
4132 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4133 "resume" : "start", buf);
4139 * Should be called with balance mutexe held
4141 int btrfs_balance(struct btrfs_fs_info *fs_info,
4142 struct btrfs_balance_control *bctl,
4143 struct btrfs_ioctl_balance_args *bargs)
4145 u64 meta_target, data_target;
4151 bool reducing_integrity;
4153 if (btrfs_fs_closing(fs_info) ||
4154 atomic_read(&fs_info->balance_pause_req) ||
4155 atomic_read(&fs_info->balance_cancel_req)) {
4160 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4161 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4165 * In case of mixed groups both data and meta should be picked,
4166 * and identical options should be given for both of them.
4168 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4169 if (mixed && (bctl->flags & allowed)) {
4170 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4171 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4172 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4174 "balance: mixed groups data and metadata options must be the same");
4180 num_devices = btrfs_num_devices(fs_info);
4182 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
4183 if (num_devices > 1)
4184 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
4185 if (num_devices > 2)
4186 allowed |= BTRFS_BLOCK_GROUP_RAID5;
4187 if (num_devices > 3)
4188 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
4189 BTRFS_BLOCK_GROUP_RAID6);
4190 if (validate_convert_profile(&bctl->data, allowed)) {
4191 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
4194 "balance: invalid convert data profile %s",
4195 get_raid_name(index));
4199 if (validate_convert_profile(&bctl->meta, allowed)) {
4200 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
4203 "balance: invalid convert metadata profile %s",
4204 get_raid_name(index));
4208 if (validate_convert_profile(&bctl->sys, allowed)) {
4209 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
4212 "balance: invalid convert system profile %s",
4213 get_raid_name(index));
4218 /* allow to reduce meta or sys integrity only if force set */
4219 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4220 BTRFS_BLOCK_GROUP_RAID10 |
4221 BTRFS_BLOCK_GROUP_RAID5 |
4222 BTRFS_BLOCK_GROUP_RAID6;
4224 seq = read_seqbegin(&fs_info->profiles_lock);
4226 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4227 (fs_info->avail_system_alloc_bits & allowed) &&
4228 !(bctl->sys.target & allowed)) ||
4229 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4230 (fs_info->avail_metadata_alloc_bits & allowed) &&
4231 !(bctl->meta.target & allowed)))
4232 reducing_integrity = true;
4234 reducing_integrity = false;
4236 /* if we're not converting, the target field is uninitialized */
4237 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4238 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4239 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4240 bctl->data.target : fs_info->avail_data_alloc_bits;
4241 } while (read_seqretry(&fs_info->profiles_lock, seq));
4243 if (reducing_integrity) {
4244 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4246 "balance: force reducing metadata integrity");
4249 "balance: reduces metadata integrity, use --force if you want this");
4255 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4256 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4257 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
4258 int data_index = btrfs_bg_flags_to_raid_index(data_target);
4261 "balance: metadata profile %s has lower redundancy than data profile %s",
4262 get_raid_name(meta_index), get_raid_name(data_index));
4265 ret = insert_balance_item(fs_info, bctl);
4266 if (ret && ret != -EEXIST)
4269 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4270 BUG_ON(ret == -EEXIST);
4271 BUG_ON(fs_info->balance_ctl);
4272 spin_lock(&fs_info->balance_lock);
4273 fs_info->balance_ctl = bctl;
4274 spin_unlock(&fs_info->balance_lock);
4276 BUG_ON(ret != -EEXIST);
4277 spin_lock(&fs_info->balance_lock);
4278 update_balance_args(bctl);
4279 spin_unlock(&fs_info->balance_lock);
4282 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4283 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4284 describe_balance_start_or_resume(fs_info);
4285 mutex_unlock(&fs_info->balance_mutex);
4287 ret = __btrfs_balance(fs_info);
4289 mutex_lock(&fs_info->balance_mutex);
4290 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4291 btrfs_info(fs_info, "balance: paused");
4292 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4293 btrfs_info(fs_info, "balance: canceled");
4295 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4297 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4300 memset(bargs, 0, sizeof(*bargs));
4301 btrfs_update_ioctl_balance_args(fs_info, bargs);
4304 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4305 balance_need_close(fs_info)) {
4306 reset_balance_state(fs_info);
4307 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4310 wake_up(&fs_info->balance_wait_q);
4314 if (bctl->flags & BTRFS_BALANCE_RESUME)
4315 reset_balance_state(fs_info);
4318 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4323 static int balance_kthread(void *data)
4325 struct btrfs_fs_info *fs_info = data;
4328 mutex_lock(&fs_info->balance_mutex);
4329 if (fs_info->balance_ctl)
4330 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4331 mutex_unlock(&fs_info->balance_mutex);
4336 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4338 struct task_struct *tsk;
4340 mutex_lock(&fs_info->balance_mutex);
4341 if (!fs_info->balance_ctl) {
4342 mutex_unlock(&fs_info->balance_mutex);
4345 mutex_unlock(&fs_info->balance_mutex);
4347 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4348 btrfs_info(fs_info, "balance: resume skipped");
4353 * A ro->rw remount sequence should continue with the paused balance
4354 * regardless of who pauses it, system or the user as of now, so set
4357 spin_lock(&fs_info->balance_lock);
4358 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4359 spin_unlock(&fs_info->balance_lock);
4361 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4362 return PTR_ERR_OR_ZERO(tsk);
4365 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4367 struct btrfs_balance_control *bctl;
4368 struct btrfs_balance_item *item;
4369 struct btrfs_disk_balance_args disk_bargs;
4370 struct btrfs_path *path;
4371 struct extent_buffer *leaf;
4372 struct btrfs_key key;
4375 path = btrfs_alloc_path();
4379 key.objectid = BTRFS_BALANCE_OBJECTID;
4380 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4386 if (ret > 0) { /* ret = -ENOENT; */
4391 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4397 leaf = path->nodes[0];
4398 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4400 bctl->flags = btrfs_balance_flags(leaf, item);
4401 bctl->flags |= BTRFS_BALANCE_RESUME;
4403 btrfs_balance_data(leaf, item, &disk_bargs);
4404 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4405 btrfs_balance_meta(leaf, item, &disk_bargs);
4406 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4407 btrfs_balance_sys(leaf, item, &disk_bargs);
4408 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4411 * This should never happen, as the paused balance state is recovered
4412 * during mount without any chance of other exclusive ops to collide.
4414 * This gives the exclusive op status to balance and keeps in paused
4415 * state until user intervention (cancel or umount). If the ownership
4416 * cannot be assigned, show a message but do not fail. The balance
4417 * is in a paused state and must have fs_info::balance_ctl properly
4420 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4422 "balance: cannot set exclusive op status, resume manually");
4424 mutex_lock(&fs_info->balance_mutex);
4425 BUG_ON(fs_info->balance_ctl);
4426 spin_lock(&fs_info->balance_lock);
4427 fs_info->balance_ctl = bctl;
4428 spin_unlock(&fs_info->balance_lock);
4429 mutex_unlock(&fs_info->balance_mutex);
4431 btrfs_free_path(path);
4435 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4439 mutex_lock(&fs_info->balance_mutex);
4440 if (!fs_info->balance_ctl) {
4441 mutex_unlock(&fs_info->balance_mutex);
4445 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4446 atomic_inc(&fs_info->balance_pause_req);
4447 mutex_unlock(&fs_info->balance_mutex);
4449 wait_event(fs_info->balance_wait_q,
4450 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4452 mutex_lock(&fs_info->balance_mutex);
4453 /* we are good with balance_ctl ripped off from under us */
4454 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4455 atomic_dec(&fs_info->balance_pause_req);
4460 mutex_unlock(&fs_info->balance_mutex);
4464 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4466 mutex_lock(&fs_info->balance_mutex);
4467 if (!fs_info->balance_ctl) {
4468 mutex_unlock(&fs_info->balance_mutex);
4473 * A paused balance with the item stored on disk can be resumed at
4474 * mount time if the mount is read-write. Otherwise it's still paused
4475 * and we must not allow cancelling as it deletes the item.
4477 if (sb_rdonly(fs_info->sb)) {
4478 mutex_unlock(&fs_info->balance_mutex);
4482 atomic_inc(&fs_info->balance_cancel_req);
4484 * if we are running just wait and return, balance item is
4485 * deleted in btrfs_balance in this case
4487 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4488 mutex_unlock(&fs_info->balance_mutex);
4489 wait_event(fs_info->balance_wait_q,
4490 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4491 mutex_lock(&fs_info->balance_mutex);
4493 mutex_unlock(&fs_info->balance_mutex);
4495 * Lock released to allow other waiters to continue, we'll
4496 * reexamine the status again.
4498 mutex_lock(&fs_info->balance_mutex);
4500 if (fs_info->balance_ctl) {
4501 reset_balance_state(fs_info);
4502 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4503 btrfs_info(fs_info, "balance: canceled");
4507 BUG_ON(fs_info->balance_ctl ||
4508 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4509 atomic_dec(&fs_info->balance_cancel_req);
4510 mutex_unlock(&fs_info->balance_mutex);
4514 static int btrfs_uuid_scan_kthread(void *data)
4516 struct btrfs_fs_info *fs_info = data;
4517 struct btrfs_root *root = fs_info->tree_root;
4518 struct btrfs_key key;
4519 struct btrfs_path *path = NULL;
4521 struct extent_buffer *eb;
4523 struct btrfs_root_item root_item;
4525 struct btrfs_trans_handle *trans = NULL;
4527 path = btrfs_alloc_path();
4534 key.type = BTRFS_ROOT_ITEM_KEY;
4538 ret = btrfs_search_forward(root, &key, path,
4539 BTRFS_OLDEST_GENERATION);
4546 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4547 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4548 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4549 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4552 eb = path->nodes[0];
4553 slot = path->slots[0];
4554 item_size = btrfs_item_size_nr(eb, slot);
4555 if (item_size < sizeof(root_item))
4558 read_extent_buffer(eb, &root_item,
4559 btrfs_item_ptr_offset(eb, slot),
4560 (int)sizeof(root_item));
4561 if (btrfs_root_refs(&root_item) == 0)
4564 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4565 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4569 btrfs_release_path(path);
4571 * 1 - subvol uuid item
4572 * 1 - received_subvol uuid item
4574 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4575 if (IS_ERR(trans)) {
4576 ret = PTR_ERR(trans);
4584 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4585 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4586 BTRFS_UUID_KEY_SUBVOL,
4589 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4595 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4596 ret = btrfs_uuid_tree_add(trans,
4597 root_item.received_uuid,
4598 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4601 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4609 ret = btrfs_end_transaction(trans);
4615 btrfs_release_path(path);
4616 if (key.offset < (u64)-1) {
4618 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4620 key.type = BTRFS_ROOT_ITEM_KEY;
4621 } else if (key.objectid < (u64)-1) {
4623 key.type = BTRFS_ROOT_ITEM_KEY;
4632 btrfs_free_path(path);
4633 if (trans && !IS_ERR(trans))
4634 btrfs_end_transaction(trans);
4636 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4638 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4639 up(&fs_info->uuid_tree_rescan_sem);
4644 * Callback for btrfs_uuid_tree_iterate().
4646 * 0 check succeeded, the entry is not outdated.
4647 * < 0 if an error occurred.
4648 * > 0 if the check failed, which means the caller shall remove the entry.
4650 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4651 u8 *uuid, u8 type, u64 subid)
4653 struct btrfs_key key;
4655 struct btrfs_root *subvol_root;
4657 if (type != BTRFS_UUID_KEY_SUBVOL &&
4658 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4661 key.objectid = subid;
4662 key.type = BTRFS_ROOT_ITEM_KEY;
4663 key.offset = (u64)-1;
4664 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4665 if (IS_ERR(subvol_root)) {
4666 ret = PTR_ERR(subvol_root);
4673 case BTRFS_UUID_KEY_SUBVOL:
4674 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4677 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4678 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4688 static int btrfs_uuid_rescan_kthread(void *data)
4690 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4694 * 1st step is to iterate through the existing UUID tree and
4695 * to delete all entries that contain outdated data.
4696 * 2nd step is to add all missing entries to the UUID tree.
4698 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4700 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4701 up(&fs_info->uuid_tree_rescan_sem);
4704 return btrfs_uuid_scan_kthread(data);
4707 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4709 struct btrfs_trans_handle *trans;
4710 struct btrfs_root *tree_root = fs_info->tree_root;
4711 struct btrfs_root *uuid_root;
4712 struct task_struct *task;
4719 trans = btrfs_start_transaction(tree_root, 2);
4721 return PTR_ERR(trans);
4723 uuid_root = btrfs_create_tree(trans, fs_info,
4724 BTRFS_UUID_TREE_OBJECTID);
4725 if (IS_ERR(uuid_root)) {
4726 ret = PTR_ERR(uuid_root);
4727 btrfs_abort_transaction(trans, ret);
4728 btrfs_end_transaction(trans);
4732 fs_info->uuid_root = uuid_root;
4734 ret = btrfs_commit_transaction(trans);
4738 down(&fs_info->uuid_tree_rescan_sem);
4739 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4741 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4742 btrfs_warn(fs_info, "failed to start uuid_scan task");
4743 up(&fs_info->uuid_tree_rescan_sem);
4744 return PTR_ERR(task);
4750 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4752 struct task_struct *task;
4754 down(&fs_info->uuid_tree_rescan_sem);
4755 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4757 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4758 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4759 up(&fs_info->uuid_tree_rescan_sem);
4760 return PTR_ERR(task);
4767 * shrinking a device means finding all of the device extents past
4768 * the new size, and then following the back refs to the chunks.
4769 * The chunk relocation code actually frees the device extent
4771 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4773 struct btrfs_fs_info *fs_info = device->fs_info;
4774 struct btrfs_root *root = fs_info->dev_root;
4775 struct btrfs_trans_handle *trans;
4776 struct btrfs_dev_extent *dev_extent = NULL;
4777 struct btrfs_path *path;
4783 bool retried = false;
4784 bool checked_pending_chunks = false;
4785 struct extent_buffer *l;
4786 struct btrfs_key key;
4787 struct btrfs_super_block *super_copy = fs_info->super_copy;
4788 u64 old_total = btrfs_super_total_bytes(super_copy);
4789 u64 old_size = btrfs_device_get_total_bytes(device);
4792 new_size = round_down(new_size, fs_info->sectorsize);
4793 diff = round_down(old_size - new_size, fs_info->sectorsize);
4795 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4798 path = btrfs_alloc_path();
4802 path->reada = READA_BACK;
4804 mutex_lock(&fs_info->chunk_mutex);
4806 btrfs_device_set_total_bytes(device, new_size);
4807 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4808 device->fs_devices->total_rw_bytes -= diff;
4809 atomic64_sub(diff, &fs_info->free_chunk_space);
4811 mutex_unlock(&fs_info->chunk_mutex);
4814 key.objectid = device->devid;
4815 key.offset = (u64)-1;
4816 key.type = BTRFS_DEV_EXTENT_KEY;
4819 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4822 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4826 ret = btrfs_previous_item(root, path, 0, key.type);
4828 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4833 btrfs_release_path(path);
4838 slot = path->slots[0];
4839 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4841 if (key.objectid != device->devid) {
4842 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4843 btrfs_release_path(path);
4847 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4848 length = btrfs_dev_extent_length(l, dev_extent);
4850 if (key.offset + length <= new_size) {
4851 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4852 btrfs_release_path(path);
4856 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4857 btrfs_release_path(path);
4860 * We may be relocating the only data chunk we have,
4861 * which could potentially end up with losing data's
4862 * raid profile, so lets allocate an empty one in
4865 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4867 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4871 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4872 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4873 if (ret == -ENOSPC) {
4876 if (ret == -ETXTBSY) {
4878 "could not shrink block group %llu due to active swapfile",
4883 } while (key.offset-- > 0);
4885 if (failed && !retried) {
4889 } else if (failed && retried) {
4894 /* Shrinking succeeded, else we would be at "done". */
4895 trans = btrfs_start_transaction(root, 0);
4896 if (IS_ERR(trans)) {
4897 ret = PTR_ERR(trans);
4901 mutex_lock(&fs_info->chunk_mutex);
4904 * We checked in the above loop all device extents that were already in
4905 * the device tree. However before we have updated the device's
4906 * total_bytes to the new size, we might have had chunk allocations that
4907 * have not complete yet (new block groups attached to transaction
4908 * handles), and therefore their device extents were not yet in the
4909 * device tree and we missed them in the loop above. So if we have any
4910 * pending chunk using a device extent that overlaps the device range
4911 * that we can not use anymore, commit the current transaction and
4912 * repeat the search on the device tree - this way we guarantee we will
4913 * not have chunks using device extents that end beyond 'new_size'.
4915 if (!checked_pending_chunks) {
4916 u64 start = new_size;
4917 u64 len = old_size - new_size;
4919 if (contains_pending_extent(trans->transaction, device,
4921 mutex_unlock(&fs_info->chunk_mutex);
4922 checked_pending_chunks = true;
4925 ret = btrfs_commit_transaction(trans);
4932 btrfs_device_set_disk_total_bytes(device, new_size);
4933 if (list_empty(&device->resized_list))
4934 list_add_tail(&device->resized_list,
4935 &fs_info->fs_devices->resized_devices);
4937 WARN_ON(diff > old_total);
4938 btrfs_set_super_total_bytes(super_copy,
4939 round_down(old_total - diff, fs_info->sectorsize));
4940 mutex_unlock(&fs_info->chunk_mutex);
4942 /* Now btrfs_update_device() will change the on-disk size. */
4943 ret = btrfs_update_device(trans, device);
4945 btrfs_abort_transaction(trans, ret);
4946 btrfs_end_transaction(trans);
4948 ret = btrfs_commit_transaction(trans);
4951 btrfs_free_path(path);
4953 mutex_lock(&fs_info->chunk_mutex);
4954 btrfs_device_set_total_bytes(device, old_size);
4955 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4956 device->fs_devices->total_rw_bytes += diff;
4957 atomic64_add(diff, &fs_info->free_chunk_space);
4958 mutex_unlock(&fs_info->chunk_mutex);
4963 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4964 struct btrfs_key *key,
4965 struct btrfs_chunk *chunk, int item_size)
4967 struct btrfs_super_block *super_copy = fs_info->super_copy;
4968 struct btrfs_disk_key disk_key;
4972 mutex_lock(&fs_info->chunk_mutex);
4973 array_size = btrfs_super_sys_array_size(super_copy);
4974 if (array_size + item_size + sizeof(disk_key)
4975 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4976 mutex_unlock(&fs_info->chunk_mutex);
4980 ptr = super_copy->sys_chunk_array + array_size;
4981 btrfs_cpu_key_to_disk(&disk_key, key);
4982 memcpy(ptr, &disk_key, sizeof(disk_key));
4983 ptr += sizeof(disk_key);
4984 memcpy(ptr, chunk, item_size);
4985 item_size += sizeof(disk_key);
4986 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4987 mutex_unlock(&fs_info->chunk_mutex);
4993 * sort the devices in descending order by max_avail, total_avail
4995 static int btrfs_cmp_device_info(const void *a, const void *b)
4997 const struct btrfs_device_info *di_a = a;
4998 const struct btrfs_device_info *di_b = b;
5000 if (di_a->max_avail > di_b->max_avail)
5002 if (di_a->max_avail < di_b->max_avail)
5004 if (di_a->total_avail > di_b->total_avail)
5006 if (di_a->total_avail < di_b->total_avail)
5011 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5013 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5016 btrfs_set_fs_incompat(info, RAID56);
5019 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
5020 - sizeof(struct btrfs_chunk)) \
5021 / sizeof(struct btrfs_stripe) + 1)
5023 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
5024 - 2 * sizeof(struct btrfs_disk_key) \
5025 - 2 * sizeof(struct btrfs_chunk)) \
5026 / sizeof(struct btrfs_stripe) + 1)
5028 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5029 u64 start, u64 type)
5031 struct btrfs_fs_info *info = trans->fs_info;
5032 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5033 struct btrfs_device *device;
5034 struct map_lookup *map = NULL;
5035 struct extent_map_tree *em_tree;
5036 struct extent_map *em;
5037 struct btrfs_device_info *devices_info = NULL;
5039 int num_stripes; /* total number of stripes to allocate */
5040 int data_stripes; /* number of stripes that count for
5042 int sub_stripes; /* sub_stripes info for map */
5043 int dev_stripes; /* stripes per dev */
5044 int devs_max; /* max devs to use */
5045 int devs_min; /* min devs needed */
5046 int devs_increment; /* ndevs has to be a multiple of this */
5047 int ncopies; /* how many copies to data has */
5048 int nparity; /* number of stripes worth of bytes to
5049 store parity information */
5051 u64 max_stripe_size;
5060 BUG_ON(!alloc_profile_is_valid(type, 0));
5062 if (list_empty(&fs_devices->alloc_list)) {
5063 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5064 btrfs_debug(info, "%s: no writable device", __func__);
5068 index = btrfs_bg_flags_to_raid_index(type);
5070 sub_stripes = btrfs_raid_array[index].sub_stripes;
5071 dev_stripes = btrfs_raid_array[index].dev_stripes;
5072 devs_max = btrfs_raid_array[index].devs_max;
5073 devs_min = btrfs_raid_array[index].devs_min;
5074 devs_increment = btrfs_raid_array[index].devs_increment;
5075 ncopies = btrfs_raid_array[index].ncopies;
5076 nparity = btrfs_raid_array[index].nparity;
5078 if (type & BTRFS_BLOCK_GROUP_DATA) {
5079 max_stripe_size = SZ_1G;
5080 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5082 devs_max = BTRFS_MAX_DEVS(info);
5083 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5084 /* for larger filesystems, use larger metadata chunks */
5085 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5086 max_stripe_size = SZ_1G;
5088 max_stripe_size = SZ_256M;
5089 max_chunk_size = max_stripe_size;
5091 devs_max = BTRFS_MAX_DEVS(info);
5092 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5093 max_stripe_size = SZ_32M;
5094 max_chunk_size = 2 * max_stripe_size;
5096 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
5098 btrfs_err(info, "invalid chunk type 0x%llx requested",
5103 /* we don't want a chunk larger than 10% of writeable space */
5104 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5107 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5113 * in the first pass through the devices list, we gather information
5114 * about the available holes on each device.
5117 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5121 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5123 "BTRFS: read-only device in alloc_list\n");
5127 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5128 &device->dev_state) ||
5129 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5132 if (device->total_bytes > device->bytes_used)
5133 total_avail = device->total_bytes - device->bytes_used;
5137 /* If there is no space on this device, skip it. */
5138 if (total_avail == 0)
5141 ret = find_free_dev_extent(trans, device,
5142 max_stripe_size * dev_stripes,
5143 &dev_offset, &max_avail);
5144 if (ret && ret != -ENOSPC)
5148 max_avail = max_stripe_size * dev_stripes;
5150 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5151 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5153 "%s: devid %llu has no free space, have=%llu want=%u",
5154 __func__, device->devid, max_avail,
5155 BTRFS_STRIPE_LEN * dev_stripes);
5159 if (ndevs == fs_devices->rw_devices) {
5160 WARN(1, "%s: found more than %llu devices\n",
5161 __func__, fs_devices->rw_devices);
5164 devices_info[ndevs].dev_offset = dev_offset;
5165 devices_info[ndevs].max_avail = max_avail;
5166 devices_info[ndevs].total_avail = total_avail;
5167 devices_info[ndevs].dev = device;
5172 * now sort the devices by hole size / available space
5174 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5175 btrfs_cmp_device_info, NULL);
5177 /* round down to number of usable stripes */
5178 ndevs = round_down(ndevs, devs_increment);
5180 if (ndevs < devs_min) {
5182 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5184 "%s: not enough devices with free space: have=%d minimum required=%d",
5185 __func__, ndevs, devs_min);
5190 ndevs = min(ndevs, devs_max);
5193 * The primary goal is to maximize the number of stripes, so use as
5194 * many devices as possible, even if the stripes are not maximum sized.
5196 * The DUP profile stores more than one stripe per device, the
5197 * max_avail is the total size so we have to adjust.
5199 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5200 num_stripes = ndevs * dev_stripes;
5203 * this will have to be fixed for RAID1 and RAID10 over
5206 data_stripes = (num_stripes - nparity) / ncopies;
5209 * Use the number of data stripes to figure out how big this chunk
5210 * is really going to be in terms of logical address space,
5211 * and compare that answer with the max chunk size. If it's higher,
5212 * we try to reduce stripe_size.
5214 if (stripe_size * data_stripes > max_chunk_size) {
5216 * Reduce stripe_size, round it up to a 16MB boundary again and
5217 * then use it, unless it ends up being even bigger than the
5218 * previous value we had already.
5220 stripe_size = min(round_up(div_u64(max_chunk_size,
5221 data_stripes), SZ_16M),
5225 /* align to BTRFS_STRIPE_LEN */
5226 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5228 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5233 map->num_stripes = num_stripes;
5235 for (i = 0; i < ndevs; ++i) {
5236 for (j = 0; j < dev_stripes; ++j) {
5237 int s = i * dev_stripes + j;
5238 map->stripes[s].dev = devices_info[i].dev;
5239 map->stripes[s].physical = devices_info[i].dev_offset +
5243 map->stripe_len = BTRFS_STRIPE_LEN;
5244 map->io_align = BTRFS_STRIPE_LEN;
5245 map->io_width = BTRFS_STRIPE_LEN;
5247 map->sub_stripes = sub_stripes;
5249 chunk_size = stripe_size * data_stripes;
5251 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5253 em = alloc_extent_map();
5259 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5260 em->map_lookup = map;
5262 em->len = chunk_size;
5263 em->block_start = 0;
5264 em->block_len = em->len;
5265 em->orig_block_len = stripe_size;
5267 em_tree = &info->mapping_tree.map_tree;
5268 write_lock(&em_tree->lock);
5269 ret = add_extent_mapping(em_tree, em, 0);
5271 write_unlock(&em_tree->lock);
5272 free_extent_map(em);
5276 list_add_tail(&em->list, &trans->transaction->pending_chunks);
5277 refcount_inc(&em->refs);
5278 write_unlock(&em_tree->lock);
5280 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5282 goto error_del_extent;
5284 for (i = 0; i < map->num_stripes; i++)
5285 btrfs_device_set_bytes_used(map->stripes[i].dev,
5286 map->stripes[i].dev->bytes_used + stripe_size);
5288 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5290 free_extent_map(em);
5291 check_raid56_incompat_flag(info, type);
5293 kfree(devices_info);
5297 write_lock(&em_tree->lock);
5298 remove_extent_mapping(em_tree, em);
5299 write_unlock(&em_tree->lock);
5301 /* One for our allocation */
5302 free_extent_map(em);
5303 /* One for the tree reference */
5304 free_extent_map(em);
5305 /* One for the pending_chunks list reference */
5306 free_extent_map(em);
5308 kfree(devices_info);
5312 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5313 u64 chunk_offset, u64 chunk_size)
5315 struct btrfs_fs_info *fs_info = trans->fs_info;
5316 struct btrfs_root *extent_root = fs_info->extent_root;
5317 struct btrfs_root *chunk_root = fs_info->chunk_root;
5318 struct btrfs_key key;
5319 struct btrfs_device *device;
5320 struct btrfs_chunk *chunk;
5321 struct btrfs_stripe *stripe;
5322 struct extent_map *em;
5323 struct map_lookup *map;
5330 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5334 map = em->map_lookup;
5335 item_size = btrfs_chunk_item_size(map->num_stripes);
5336 stripe_size = em->orig_block_len;
5338 chunk = kzalloc(item_size, GFP_NOFS);
5345 * Take the device list mutex to prevent races with the final phase of
5346 * a device replace operation that replaces the device object associated
5347 * with the map's stripes, because the device object's id can change
5348 * at any time during that final phase of the device replace operation
5349 * (dev-replace.c:btrfs_dev_replace_finishing()).
5351 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5352 for (i = 0; i < map->num_stripes; i++) {
5353 device = map->stripes[i].dev;
5354 dev_offset = map->stripes[i].physical;
5356 ret = btrfs_update_device(trans, device);
5359 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5360 dev_offset, stripe_size);
5365 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5369 stripe = &chunk->stripe;
5370 for (i = 0; i < map->num_stripes; i++) {
5371 device = map->stripes[i].dev;
5372 dev_offset = map->stripes[i].physical;
5374 btrfs_set_stack_stripe_devid(stripe, device->devid);
5375 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5376 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5379 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5381 btrfs_set_stack_chunk_length(chunk, chunk_size);
5382 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5383 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5384 btrfs_set_stack_chunk_type(chunk, map->type);
5385 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5386 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5387 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5388 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5389 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5391 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5392 key.type = BTRFS_CHUNK_ITEM_KEY;
5393 key.offset = chunk_offset;
5395 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5396 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5398 * TODO: Cleanup of inserted chunk root in case of
5401 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5406 free_extent_map(em);
5411 * Chunk allocation falls into two parts. The first part does works
5412 * that make the new allocated chunk useable, but not do any operation
5413 * that modifies the chunk tree. The second part does the works that
5414 * require modifying the chunk tree. This division is important for the
5415 * bootstrap process of adding storage to a seed btrfs.
5417 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5421 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5422 chunk_offset = find_next_chunk(trans->fs_info);
5423 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5426 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5427 struct btrfs_fs_info *fs_info)
5430 u64 sys_chunk_offset;
5434 chunk_offset = find_next_chunk(fs_info);
5435 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5436 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5440 sys_chunk_offset = find_next_chunk(fs_info);
5441 alloc_profile = btrfs_system_alloc_profile(fs_info);
5442 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5446 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5450 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5451 BTRFS_BLOCK_GROUP_RAID10 |
5452 BTRFS_BLOCK_GROUP_RAID5 |
5453 BTRFS_BLOCK_GROUP_DUP)) {
5455 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5464 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5466 struct extent_map *em;
5467 struct map_lookup *map;
5472 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5476 map = em->map_lookup;
5477 for (i = 0; i < map->num_stripes; i++) {
5478 if (test_bit(BTRFS_DEV_STATE_MISSING,
5479 &map->stripes[i].dev->dev_state)) {
5483 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5484 &map->stripes[i].dev->dev_state)) {
5491 * If the number of missing devices is larger than max errors,
5492 * we can not write the data into that chunk successfully, so
5495 if (miss_ndevs > btrfs_chunk_max_errors(map))
5498 free_extent_map(em);
5502 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5504 extent_map_tree_init(&tree->map_tree);
5507 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5509 struct extent_map *em;
5512 write_lock(&tree->map_tree.lock);
5513 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5515 remove_extent_mapping(&tree->map_tree, em);
5516 write_unlock(&tree->map_tree.lock);
5520 free_extent_map(em);
5521 /* once for the tree */
5522 free_extent_map(em);
5526 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5528 struct extent_map *em;
5529 struct map_lookup *map;
5532 em = btrfs_get_chunk_map(fs_info, logical, len);
5535 * We could return errors for these cases, but that could get
5536 * ugly and we'd probably do the same thing which is just not do
5537 * anything else and exit, so return 1 so the callers don't try
5538 * to use other copies.
5542 map = em->map_lookup;
5543 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5544 ret = map->num_stripes;
5545 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5546 ret = map->sub_stripes;
5547 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5549 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5551 * There could be two corrupted data stripes, we need
5552 * to loop retry in order to rebuild the correct data.
5554 * Fail a stripe at a time on every retry except the
5555 * stripe under reconstruction.
5557 ret = map->num_stripes;
5560 free_extent_map(em);
5562 down_read(&fs_info->dev_replace.rwsem);
5563 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5564 fs_info->dev_replace.tgtdev)
5566 up_read(&fs_info->dev_replace.rwsem);
5571 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5574 struct extent_map *em;
5575 struct map_lookup *map;
5576 unsigned long len = fs_info->sectorsize;
5578 em = btrfs_get_chunk_map(fs_info, logical, len);
5580 if (!WARN_ON(IS_ERR(em))) {
5581 map = em->map_lookup;
5582 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5583 len = map->stripe_len * nr_data_stripes(map);
5584 free_extent_map(em);
5589 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5591 struct extent_map *em;
5592 struct map_lookup *map;
5595 em = btrfs_get_chunk_map(fs_info, logical, len);
5597 if(!WARN_ON(IS_ERR(em))) {
5598 map = em->map_lookup;
5599 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5601 free_extent_map(em);
5606 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5607 struct map_lookup *map, int first,
5608 int dev_replace_is_ongoing)
5612 int preferred_mirror;
5614 struct btrfs_device *srcdev;
5617 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5619 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5620 num_stripes = map->sub_stripes;
5622 num_stripes = map->num_stripes;
5624 preferred_mirror = first + current->pid % num_stripes;
5626 if (dev_replace_is_ongoing &&
5627 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5628 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5629 srcdev = fs_info->dev_replace.srcdev;
5634 * try to avoid the drive that is the source drive for a
5635 * dev-replace procedure, only choose it if no other non-missing
5636 * mirror is available
5638 for (tolerance = 0; tolerance < 2; tolerance++) {
5639 if (map->stripes[preferred_mirror].dev->bdev &&
5640 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5641 return preferred_mirror;
5642 for (i = first; i < first + num_stripes; i++) {
5643 if (map->stripes[i].dev->bdev &&
5644 (tolerance || map->stripes[i].dev != srcdev))
5649 /* we couldn't find one that doesn't fail. Just return something
5650 * and the io error handling code will clean up eventually
5652 return preferred_mirror;
5655 static inline int parity_smaller(u64 a, u64 b)
5660 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5661 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5663 struct btrfs_bio_stripe s;
5670 for (i = 0; i < num_stripes - 1; i++) {
5671 if (parity_smaller(bbio->raid_map[i],
5672 bbio->raid_map[i+1])) {
5673 s = bbio->stripes[i];
5674 l = bbio->raid_map[i];
5675 bbio->stripes[i] = bbio->stripes[i+1];
5676 bbio->raid_map[i] = bbio->raid_map[i+1];
5677 bbio->stripes[i+1] = s;
5678 bbio->raid_map[i+1] = l;
5686 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5688 struct btrfs_bio *bbio = kzalloc(
5689 /* the size of the btrfs_bio */
5690 sizeof(struct btrfs_bio) +
5691 /* plus the variable array for the stripes */
5692 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5693 /* plus the variable array for the tgt dev */
5694 sizeof(int) * (real_stripes) +
5696 * plus the raid_map, which includes both the tgt dev
5699 sizeof(u64) * (total_stripes),
5700 GFP_NOFS|__GFP_NOFAIL);
5702 atomic_set(&bbio->error, 0);
5703 refcount_set(&bbio->refs, 1);
5708 void btrfs_get_bbio(struct btrfs_bio *bbio)
5710 WARN_ON(!refcount_read(&bbio->refs));
5711 refcount_inc(&bbio->refs);
5714 void btrfs_put_bbio(struct btrfs_bio *bbio)
5718 if (refcount_dec_and_test(&bbio->refs))
5722 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5724 * Please note that, discard won't be sent to target device of device
5727 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5728 u64 logical, u64 length,
5729 struct btrfs_bio **bbio_ret)
5731 struct extent_map *em;
5732 struct map_lookup *map;
5733 struct btrfs_bio *bbio;
5737 u64 stripe_end_offset;
5744 u32 sub_stripes = 0;
5745 u64 stripes_per_dev = 0;
5746 u32 remaining_stripes = 0;
5747 u32 last_stripe = 0;
5751 /* discard always return a bbio */
5754 em = btrfs_get_chunk_map(fs_info, logical, length);
5758 map = em->map_lookup;
5759 /* we don't discard raid56 yet */
5760 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5765 offset = logical - em->start;
5766 length = min_t(u64, em->len - offset, length);
5768 stripe_len = map->stripe_len;
5770 * stripe_nr counts the total number of stripes we have to stride
5771 * to get to this block
5773 stripe_nr = div64_u64(offset, stripe_len);
5775 /* stripe_offset is the offset of this block in its stripe */
5776 stripe_offset = offset - stripe_nr * stripe_len;
5778 stripe_nr_end = round_up(offset + length, map->stripe_len);
5779 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5780 stripe_cnt = stripe_nr_end - stripe_nr;
5781 stripe_end_offset = stripe_nr_end * map->stripe_len -
5784 * after this, stripe_nr is the number of stripes on this
5785 * device we have to walk to find the data, and stripe_index is
5786 * the number of our device in the stripe array
5790 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5791 BTRFS_BLOCK_GROUP_RAID10)) {
5792 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5795 sub_stripes = map->sub_stripes;
5797 factor = map->num_stripes / sub_stripes;
5798 num_stripes = min_t(u64, map->num_stripes,
5799 sub_stripes * stripe_cnt);
5800 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5801 stripe_index *= sub_stripes;
5802 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5803 &remaining_stripes);
5804 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5805 last_stripe *= sub_stripes;
5806 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5807 BTRFS_BLOCK_GROUP_DUP)) {
5808 num_stripes = map->num_stripes;
5810 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5814 bbio = alloc_btrfs_bio(num_stripes, 0);
5820 for (i = 0; i < num_stripes; i++) {
5821 bbio->stripes[i].physical =
5822 map->stripes[stripe_index].physical +
5823 stripe_offset + stripe_nr * map->stripe_len;
5824 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5826 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5827 BTRFS_BLOCK_GROUP_RAID10)) {
5828 bbio->stripes[i].length = stripes_per_dev *
5831 if (i / sub_stripes < remaining_stripes)
5832 bbio->stripes[i].length +=
5836 * Special for the first stripe and
5839 * |-------|...|-------|
5843 if (i < sub_stripes)
5844 bbio->stripes[i].length -=
5847 if (stripe_index >= last_stripe &&
5848 stripe_index <= (last_stripe +
5850 bbio->stripes[i].length -=
5853 if (i == sub_stripes - 1)
5856 bbio->stripes[i].length = length;
5860 if (stripe_index == map->num_stripes) {
5867 bbio->map_type = map->type;
5868 bbio->num_stripes = num_stripes;
5870 free_extent_map(em);
5875 * In dev-replace case, for repair case (that's the only case where the mirror
5876 * is selected explicitly when calling btrfs_map_block), blocks left of the
5877 * left cursor can also be read from the target drive.
5879 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5881 * For READ, it also needs to be supported using the same mirror number.
5883 * If the requested block is not left of the left cursor, EIO is returned. This
5884 * can happen because btrfs_num_copies() returns one more in the dev-replace
5887 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5888 u64 logical, u64 length,
5889 u64 srcdev_devid, int *mirror_num,
5892 struct btrfs_bio *bbio = NULL;
5894 int index_srcdev = 0;
5896 u64 physical_of_found = 0;
5900 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5901 logical, &length, &bbio, 0, 0);
5903 ASSERT(bbio == NULL);
5907 num_stripes = bbio->num_stripes;
5908 if (*mirror_num > num_stripes) {
5910 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5911 * that means that the requested area is not left of the left
5914 btrfs_put_bbio(bbio);
5919 * process the rest of the function using the mirror_num of the source
5920 * drive. Therefore look it up first. At the end, patch the device
5921 * pointer to the one of the target drive.
5923 for (i = 0; i < num_stripes; i++) {
5924 if (bbio->stripes[i].dev->devid != srcdev_devid)
5928 * In case of DUP, in order to keep it simple, only add the
5929 * mirror with the lowest physical address
5932 physical_of_found <= bbio->stripes[i].physical)
5937 physical_of_found = bbio->stripes[i].physical;
5940 btrfs_put_bbio(bbio);
5946 *mirror_num = index_srcdev + 1;
5947 *physical = physical_of_found;
5951 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5952 struct btrfs_bio **bbio_ret,
5953 struct btrfs_dev_replace *dev_replace,
5954 int *num_stripes_ret, int *max_errors_ret)
5956 struct btrfs_bio *bbio = *bbio_ret;
5957 u64 srcdev_devid = dev_replace->srcdev->devid;
5958 int tgtdev_indexes = 0;
5959 int num_stripes = *num_stripes_ret;
5960 int max_errors = *max_errors_ret;
5963 if (op == BTRFS_MAP_WRITE) {
5964 int index_where_to_add;
5967 * duplicate the write operations while the dev replace
5968 * procedure is running. Since the copying of the old disk to
5969 * the new disk takes place at run time while the filesystem is
5970 * mounted writable, the regular write operations to the old
5971 * disk have to be duplicated to go to the new disk as well.
5973 * Note that device->missing is handled by the caller, and that
5974 * the write to the old disk is already set up in the stripes
5977 index_where_to_add = num_stripes;
5978 for (i = 0; i < num_stripes; i++) {
5979 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5980 /* write to new disk, too */
5981 struct btrfs_bio_stripe *new =
5982 bbio->stripes + index_where_to_add;
5983 struct btrfs_bio_stripe *old =
5986 new->physical = old->physical;
5987 new->length = old->length;
5988 new->dev = dev_replace->tgtdev;
5989 bbio->tgtdev_map[i] = index_where_to_add;
5990 index_where_to_add++;
5995 num_stripes = index_where_to_add;
5996 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5997 int index_srcdev = 0;
5999 u64 physical_of_found = 0;
6002 * During the dev-replace procedure, the target drive can also
6003 * be used to read data in case it is needed to repair a corrupt
6004 * block elsewhere. This is possible if the requested area is
6005 * left of the left cursor. In this area, the target drive is a
6006 * full copy of the source drive.
6008 for (i = 0; i < num_stripes; i++) {
6009 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6011 * In case of DUP, in order to keep it simple,
6012 * only add the mirror with the lowest physical
6016 physical_of_found <=
6017 bbio->stripes[i].physical)
6021 physical_of_found = bbio->stripes[i].physical;
6025 struct btrfs_bio_stripe *tgtdev_stripe =
6026 bbio->stripes + num_stripes;
6028 tgtdev_stripe->physical = physical_of_found;
6029 tgtdev_stripe->length =
6030 bbio->stripes[index_srcdev].length;
6031 tgtdev_stripe->dev = dev_replace->tgtdev;
6032 bbio->tgtdev_map[index_srcdev] = num_stripes;
6039 *num_stripes_ret = num_stripes;
6040 *max_errors_ret = max_errors;
6041 bbio->num_tgtdevs = tgtdev_indexes;
6045 static bool need_full_stripe(enum btrfs_map_op op)
6047 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6050 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6051 enum btrfs_map_op op,
6052 u64 logical, u64 *length,
6053 struct btrfs_bio **bbio_ret,
6054 int mirror_num, int need_raid_map)
6056 struct extent_map *em;
6057 struct map_lookup *map;
6067 int tgtdev_indexes = 0;
6068 struct btrfs_bio *bbio = NULL;
6069 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6070 int dev_replace_is_ongoing = 0;
6071 int num_alloc_stripes;
6072 int patch_the_first_stripe_for_dev_replace = 0;
6073 u64 physical_to_patch_in_first_stripe = 0;
6074 u64 raid56_full_stripe_start = (u64)-1;
6076 if (op == BTRFS_MAP_DISCARD)
6077 return __btrfs_map_block_for_discard(fs_info, logical,
6080 em = btrfs_get_chunk_map(fs_info, logical, *length);
6084 map = em->map_lookup;
6085 offset = logical - em->start;
6087 stripe_len = map->stripe_len;
6090 * stripe_nr counts the total number of stripes we have to stride
6091 * to get to this block
6093 stripe_nr = div64_u64(stripe_nr, stripe_len);
6095 stripe_offset = stripe_nr * stripe_len;
6096 if (offset < stripe_offset) {
6098 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
6099 stripe_offset, offset, em->start, logical,
6101 free_extent_map(em);
6105 /* stripe_offset is the offset of this block in its stripe*/
6106 stripe_offset = offset - stripe_offset;
6108 /* if we're here for raid56, we need to know the stripe aligned start */
6109 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6110 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
6111 raid56_full_stripe_start = offset;
6113 /* allow a write of a full stripe, but make sure we don't
6114 * allow straddling of stripes
6116 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6118 raid56_full_stripe_start *= full_stripe_len;
6121 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6123 /* For writes to RAID[56], allow a full stripeset across all disks.
6124 For other RAID types and for RAID[56] reads, just allow a single
6125 stripe (on a single disk). */
6126 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6127 (op == BTRFS_MAP_WRITE)) {
6128 max_len = stripe_len * nr_data_stripes(map) -
6129 (offset - raid56_full_stripe_start);
6131 /* we limit the length of each bio to what fits in a stripe */
6132 max_len = stripe_len - stripe_offset;
6134 *length = min_t(u64, em->len - offset, max_len);
6136 *length = em->len - offset;
6140 * This is for when we're called from btrfs_bio_fits_in_stripe and all
6141 * it cares about is the length
6146 down_read(&dev_replace->rwsem);
6147 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6149 * Hold the semaphore for read during the whole operation, write is
6150 * requested at commit time but must wait.
6152 if (!dev_replace_is_ongoing)
6153 up_read(&dev_replace->rwsem);
6155 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6156 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6157 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6158 dev_replace->srcdev->devid,
6160 &physical_to_patch_in_first_stripe);
6164 patch_the_first_stripe_for_dev_replace = 1;
6165 } else if (mirror_num > map->num_stripes) {
6171 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6172 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6174 if (!need_full_stripe(op))
6176 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
6177 if (need_full_stripe(op))
6178 num_stripes = map->num_stripes;
6179 else if (mirror_num)
6180 stripe_index = mirror_num - 1;
6182 stripe_index = find_live_mirror(fs_info, map, 0,
6183 dev_replace_is_ongoing);
6184 mirror_num = stripe_index + 1;
6187 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6188 if (need_full_stripe(op)) {
6189 num_stripes = map->num_stripes;
6190 } else if (mirror_num) {
6191 stripe_index = mirror_num - 1;
6196 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6197 u32 factor = map->num_stripes / map->sub_stripes;
6199 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6200 stripe_index *= map->sub_stripes;
6202 if (need_full_stripe(op))
6203 num_stripes = map->sub_stripes;
6204 else if (mirror_num)
6205 stripe_index += mirror_num - 1;
6207 int old_stripe_index = stripe_index;
6208 stripe_index = find_live_mirror(fs_info, map,
6210 dev_replace_is_ongoing);
6211 mirror_num = stripe_index - old_stripe_index + 1;
6214 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6215 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6216 /* push stripe_nr back to the start of the full stripe */
6217 stripe_nr = div64_u64(raid56_full_stripe_start,
6218 stripe_len * nr_data_stripes(map));
6220 /* RAID[56] write or recovery. Return all stripes */
6221 num_stripes = map->num_stripes;
6222 max_errors = nr_parity_stripes(map);
6224 *length = map->stripe_len;
6229 * Mirror #0 or #1 means the original data block.
6230 * Mirror #2 is RAID5 parity block.
6231 * Mirror #3 is RAID6 Q block.
6233 stripe_nr = div_u64_rem(stripe_nr,
6234 nr_data_stripes(map), &stripe_index);
6236 stripe_index = nr_data_stripes(map) +
6239 /* We distribute the parity blocks across stripes */
6240 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6242 if (!need_full_stripe(op) && mirror_num <= 1)
6247 * after this, stripe_nr is the number of stripes on this
6248 * device we have to walk to find the data, and stripe_index is
6249 * the number of our device in the stripe array
6251 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6253 mirror_num = stripe_index + 1;
6255 if (stripe_index >= map->num_stripes) {
6257 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6258 stripe_index, map->num_stripes);
6263 num_alloc_stripes = num_stripes;
6264 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6265 if (op == BTRFS_MAP_WRITE)
6266 num_alloc_stripes <<= 1;
6267 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6268 num_alloc_stripes++;
6269 tgtdev_indexes = num_stripes;
6272 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6277 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6278 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6280 /* build raid_map */
6281 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6282 (need_full_stripe(op) || mirror_num > 1)) {
6286 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6287 sizeof(struct btrfs_bio_stripe) *
6289 sizeof(int) * tgtdev_indexes);
6291 /* Work out the disk rotation on this stripe-set */
6292 div_u64_rem(stripe_nr, num_stripes, &rot);
6294 /* Fill in the logical address of each stripe */
6295 tmp = stripe_nr * nr_data_stripes(map);
6296 for (i = 0; i < nr_data_stripes(map); i++)
6297 bbio->raid_map[(i+rot) % num_stripes] =
6298 em->start + (tmp + i) * map->stripe_len;
6300 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6301 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6302 bbio->raid_map[(i+rot+1) % num_stripes] =
6307 for (i = 0; i < num_stripes; i++) {
6308 bbio->stripes[i].physical =
6309 map->stripes[stripe_index].physical +
6311 stripe_nr * map->stripe_len;
6312 bbio->stripes[i].dev =
6313 map->stripes[stripe_index].dev;
6317 if (need_full_stripe(op))
6318 max_errors = btrfs_chunk_max_errors(map);
6321 sort_parity_stripes(bbio, num_stripes);
6323 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6324 need_full_stripe(op)) {
6325 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6330 bbio->map_type = map->type;
6331 bbio->num_stripes = num_stripes;
6332 bbio->max_errors = max_errors;
6333 bbio->mirror_num = mirror_num;
6336 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6337 * mirror_num == num_stripes + 1 && dev_replace target drive is
6338 * available as a mirror
6340 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6341 WARN_ON(num_stripes > 1);
6342 bbio->stripes[0].dev = dev_replace->tgtdev;
6343 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6344 bbio->mirror_num = map->num_stripes + 1;
6347 if (dev_replace_is_ongoing) {
6348 lockdep_assert_held(&dev_replace->rwsem);
6349 /* Unlock and let waiting writers proceed */
6350 up_read(&dev_replace->rwsem);
6352 free_extent_map(em);
6356 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6357 u64 logical, u64 *length,
6358 struct btrfs_bio **bbio_ret, int mirror_num)
6360 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6364 /* For Scrub/replace */
6365 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6366 u64 logical, u64 *length,
6367 struct btrfs_bio **bbio_ret)
6369 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6372 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6373 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6375 struct extent_map *em;
6376 struct map_lookup *map;
6384 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6388 map = em->map_lookup;
6390 rmap_len = map->stripe_len;
6392 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6393 length = div_u64(length, map->num_stripes / map->sub_stripes);
6394 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6395 length = div_u64(length, map->num_stripes);
6396 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6397 length = div_u64(length, nr_data_stripes(map));
6398 rmap_len = map->stripe_len * nr_data_stripes(map);
6401 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6402 BUG_ON(!buf); /* -ENOMEM */
6404 for (i = 0; i < map->num_stripes; i++) {
6405 if (map->stripes[i].physical > physical ||
6406 map->stripes[i].physical + length <= physical)
6409 stripe_nr = physical - map->stripes[i].physical;
6410 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6412 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6413 stripe_nr = stripe_nr * map->num_stripes + i;
6414 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6415 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6416 stripe_nr = stripe_nr * map->num_stripes + i;
6417 } /* else if RAID[56], multiply by nr_data_stripes().
6418 * Alternatively, just use rmap_len below instead of
6419 * map->stripe_len */
6421 bytenr = chunk_start + stripe_nr * rmap_len;
6422 WARN_ON(nr >= map->num_stripes);
6423 for (j = 0; j < nr; j++) {
6424 if (buf[j] == bytenr)
6428 WARN_ON(nr >= map->num_stripes);
6435 *stripe_len = rmap_len;
6437 free_extent_map(em);
6441 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6443 bio->bi_private = bbio->private;
6444 bio->bi_end_io = bbio->end_io;
6447 btrfs_put_bbio(bbio);
6450 static void btrfs_end_bio(struct bio *bio)
6452 struct btrfs_bio *bbio = bio->bi_private;
6453 int is_orig_bio = 0;
6455 if (bio->bi_status) {
6456 atomic_inc(&bbio->error);
6457 if (bio->bi_status == BLK_STS_IOERR ||
6458 bio->bi_status == BLK_STS_TARGET) {
6459 unsigned int stripe_index =
6460 btrfs_io_bio(bio)->stripe_index;
6461 struct btrfs_device *dev;
6463 BUG_ON(stripe_index >= bbio->num_stripes);
6464 dev = bbio->stripes[stripe_index].dev;
6466 if (bio_op(bio) == REQ_OP_WRITE)
6467 btrfs_dev_stat_inc_and_print(dev,
6468 BTRFS_DEV_STAT_WRITE_ERRS);
6470 btrfs_dev_stat_inc_and_print(dev,
6471 BTRFS_DEV_STAT_READ_ERRS);
6472 if (bio->bi_opf & REQ_PREFLUSH)
6473 btrfs_dev_stat_inc_and_print(dev,
6474 BTRFS_DEV_STAT_FLUSH_ERRS);
6479 if (bio == bbio->orig_bio)
6482 btrfs_bio_counter_dec(bbio->fs_info);
6484 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6487 bio = bbio->orig_bio;
6490 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6491 /* only send an error to the higher layers if it is
6492 * beyond the tolerance of the btrfs bio
6494 if (atomic_read(&bbio->error) > bbio->max_errors) {
6495 bio->bi_status = BLK_STS_IOERR;
6498 * this bio is actually up to date, we didn't
6499 * go over the max number of errors
6501 bio->bi_status = BLK_STS_OK;
6504 btrfs_end_bbio(bbio, bio);
6505 } else if (!is_orig_bio) {
6511 * see run_scheduled_bios for a description of why bios are collected for
6514 * This will add one bio to the pending list for a device and make sure
6515 * the work struct is scheduled.
6517 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6520 struct btrfs_fs_info *fs_info = device->fs_info;
6521 int should_queue = 1;
6522 struct btrfs_pending_bios *pending_bios;
6524 /* don't bother with additional async steps for reads, right now */
6525 if (bio_op(bio) == REQ_OP_READ) {
6526 btrfsic_submit_bio(bio);
6530 WARN_ON(bio->bi_next);
6531 bio->bi_next = NULL;
6533 spin_lock(&device->io_lock);
6534 if (op_is_sync(bio->bi_opf))
6535 pending_bios = &device->pending_sync_bios;
6537 pending_bios = &device->pending_bios;
6539 if (pending_bios->tail)
6540 pending_bios->tail->bi_next = bio;
6542 pending_bios->tail = bio;
6543 if (!pending_bios->head)
6544 pending_bios->head = bio;
6545 if (device->running_pending)
6548 spin_unlock(&device->io_lock);
6551 btrfs_queue_work(fs_info->submit_workers, &device->work);
6554 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6555 u64 physical, int dev_nr, int async)
6557 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6558 struct btrfs_fs_info *fs_info = bbio->fs_info;
6560 bio->bi_private = bbio;
6561 btrfs_io_bio(bio)->stripe_index = dev_nr;
6562 bio->bi_end_io = btrfs_end_bio;
6563 bio->bi_iter.bi_sector = physical >> 9;
6564 btrfs_debug_in_rcu(fs_info,
6565 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6566 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6567 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6568 bio->bi_iter.bi_size);
6569 bio_set_dev(bio, dev->bdev);
6571 btrfs_bio_counter_inc_noblocked(fs_info);
6574 btrfs_schedule_bio(dev, bio);
6576 btrfsic_submit_bio(bio);
6579 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6581 atomic_inc(&bbio->error);
6582 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6583 /* Should be the original bio. */
6584 WARN_ON(bio != bbio->orig_bio);
6586 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6587 bio->bi_iter.bi_sector = logical >> 9;
6588 if (atomic_read(&bbio->error) > bbio->max_errors)
6589 bio->bi_status = BLK_STS_IOERR;
6591 bio->bi_status = BLK_STS_OK;
6592 btrfs_end_bbio(bbio, bio);
6596 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6597 int mirror_num, int async_submit)
6599 struct btrfs_device *dev;
6600 struct bio *first_bio = bio;
6601 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6607 struct btrfs_bio *bbio = NULL;
6609 length = bio->bi_iter.bi_size;
6610 map_length = length;
6612 btrfs_bio_counter_inc_blocked(fs_info);
6613 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6614 &map_length, &bbio, mirror_num, 1);
6616 btrfs_bio_counter_dec(fs_info);
6617 return errno_to_blk_status(ret);
6620 total_devs = bbio->num_stripes;
6621 bbio->orig_bio = first_bio;
6622 bbio->private = first_bio->bi_private;
6623 bbio->end_io = first_bio->bi_end_io;
6624 bbio->fs_info = fs_info;
6625 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6627 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6628 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6629 /* In this case, map_length has been set to the length of
6630 a single stripe; not the whole write */
6631 if (bio_op(bio) == REQ_OP_WRITE) {
6632 ret = raid56_parity_write(fs_info, bio, bbio,
6635 ret = raid56_parity_recover(fs_info, bio, bbio,
6636 map_length, mirror_num, 1);
6639 btrfs_bio_counter_dec(fs_info);
6640 return errno_to_blk_status(ret);
6643 if (map_length < length) {
6645 "mapping failed logical %llu bio len %llu len %llu",
6646 logical, length, map_length);
6650 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6651 dev = bbio->stripes[dev_nr].dev;
6652 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6654 (bio_op(first_bio) == REQ_OP_WRITE &&
6655 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6656 bbio_error(bbio, first_bio, logical);
6660 if (dev_nr < total_devs - 1)
6661 bio = btrfs_bio_clone(first_bio);
6665 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6666 dev_nr, async_submit);
6668 btrfs_bio_counter_dec(fs_info);
6672 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6675 struct btrfs_device *device;
6676 struct btrfs_fs_devices *cur_devices;
6678 cur_devices = fs_info->fs_devices;
6679 while (cur_devices) {
6681 !memcmp(cur_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6682 device = find_device(cur_devices, devid, uuid);
6686 cur_devices = cur_devices->seed;
6691 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6692 u64 devid, u8 *dev_uuid)
6694 struct btrfs_device *device;
6696 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6700 list_add(&device->dev_list, &fs_devices->devices);
6701 device->fs_devices = fs_devices;
6702 fs_devices->num_devices++;
6704 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6705 fs_devices->missing_devices++;
6711 * btrfs_alloc_device - allocate struct btrfs_device
6712 * @fs_info: used only for generating a new devid, can be NULL if
6713 * devid is provided (i.e. @devid != NULL).
6714 * @devid: a pointer to devid for this device. If NULL a new devid
6716 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6719 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6720 * on error. Returned struct is not linked onto any lists and must be
6721 * destroyed with btrfs_free_device.
6723 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6727 struct btrfs_device *dev;
6730 if (WARN_ON(!devid && !fs_info))
6731 return ERR_PTR(-EINVAL);
6733 dev = __alloc_device();
6742 ret = find_next_devid(fs_info, &tmp);
6744 btrfs_free_device(dev);
6745 return ERR_PTR(ret);
6751 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6753 generate_random_uuid(dev->uuid);
6755 btrfs_init_work(&dev->work, btrfs_submit_helper,
6756 pending_bios_fn, NULL, NULL);
6761 /* Return -EIO if any error, otherwise return 0. */
6762 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6763 struct extent_buffer *leaf,
6764 struct btrfs_chunk *chunk, u64 logical)
6774 length = btrfs_chunk_length(leaf, chunk);
6775 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6776 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6777 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6778 type = btrfs_chunk_type(leaf, chunk);
6781 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6785 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6786 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6789 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6790 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6791 btrfs_chunk_sector_size(leaf, chunk));
6794 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6795 btrfs_err(fs_info, "invalid chunk length %llu", length);
6798 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6799 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6803 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6805 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6806 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6807 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6808 btrfs_chunk_type(leaf, chunk));
6812 if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
6813 btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type);
6817 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
6818 (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
6820 "system chunk with data or metadata type: 0x%llx", type);
6824 features = btrfs_super_incompat_flags(fs_info->super_copy);
6825 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
6829 if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
6830 (type & BTRFS_BLOCK_GROUP_DATA)) {
6832 "mixed chunk type in non-mixed mode: 0x%llx", type);
6837 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6838 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6839 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6840 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6841 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6842 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6843 num_stripes != 1)) {
6845 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6846 num_stripes, sub_stripes,
6847 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6854 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6855 u64 devid, u8 *uuid, bool error)
6858 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6861 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6865 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6866 struct extent_buffer *leaf,
6867 struct btrfs_chunk *chunk)
6869 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6870 struct map_lookup *map;
6871 struct extent_map *em;
6875 u8 uuid[BTRFS_UUID_SIZE];
6880 logical = key->offset;
6881 length = btrfs_chunk_length(leaf, chunk);
6882 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6884 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6888 read_lock(&map_tree->map_tree.lock);
6889 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6890 read_unlock(&map_tree->map_tree.lock);
6892 /* already mapped? */
6893 if (em && em->start <= logical && em->start + em->len > logical) {
6894 free_extent_map(em);
6897 free_extent_map(em);
6900 em = alloc_extent_map();
6903 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6905 free_extent_map(em);
6909 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6910 em->map_lookup = map;
6911 em->start = logical;
6914 em->block_start = 0;
6915 em->block_len = em->len;
6917 map->num_stripes = num_stripes;
6918 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6919 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6920 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6921 map->type = btrfs_chunk_type(leaf, chunk);
6922 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6923 map->verified_stripes = 0;
6924 for (i = 0; i < num_stripes; i++) {
6925 map->stripes[i].physical =
6926 btrfs_stripe_offset_nr(leaf, chunk, i);
6927 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6928 read_extent_buffer(leaf, uuid, (unsigned long)
6929 btrfs_stripe_dev_uuid_nr(chunk, i),
6931 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6933 if (!map->stripes[i].dev &&
6934 !btrfs_test_opt(fs_info, DEGRADED)) {
6935 free_extent_map(em);
6936 btrfs_report_missing_device(fs_info, devid, uuid, true);
6939 if (!map->stripes[i].dev) {
6940 map->stripes[i].dev =
6941 add_missing_dev(fs_info->fs_devices, devid,
6943 if (IS_ERR(map->stripes[i].dev)) {
6944 free_extent_map(em);
6946 "failed to init missing dev %llu: %ld",
6947 devid, PTR_ERR(map->stripes[i].dev));
6948 return PTR_ERR(map->stripes[i].dev);
6950 btrfs_report_missing_device(fs_info, devid, uuid, false);
6952 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6953 &(map->stripes[i].dev->dev_state));
6957 write_lock(&map_tree->map_tree.lock);
6958 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6959 write_unlock(&map_tree->map_tree.lock);
6962 "failed to add chunk map, start=%llu len=%llu: %d",
6963 em->start, em->len, ret);
6965 free_extent_map(em);
6970 static void fill_device_from_item(struct extent_buffer *leaf,
6971 struct btrfs_dev_item *dev_item,
6972 struct btrfs_device *device)
6976 device->devid = btrfs_device_id(leaf, dev_item);
6977 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6978 device->total_bytes = device->disk_total_bytes;
6979 device->commit_total_bytes = device->disk_total_bytes;
6980 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6981 device->commit_bytes_used = device->bytes_used;
6982 device->type = btrfs_device_type(leaf, dev_item);
6983 device->io_align = btrfs_device_io_align(leaf, dev_item);
6984 device->io_width = btrfs_device_io_width(leaf, dev_item);
6985 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6986 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6987 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6989 ptr = btrfs_device_uuid(dev_item);
6990 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6993 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6996 struct btrfs_fs_devices *fs_devices;
6999 lockdep_assert_held(&uuid_mutex);
7002 fs_devices = fs_info->fs_devices->seed;
7003 while (fs_devices) {
7004 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7007 fs_devices = fs_devices->seed;
7010 fs_devices = find_fsid(fsid, NULL);
7012 if (!btrfs_test_opt(fs_info, DEGRADED))
7013 return ERR_PTR(-ENOENT);
7015 fs_devices = alloc_fs_devices(fsid, NULL);
7016 if (IS_ERR(fs_devices))
7019 fs_devices->seeding = 1;
7020 fs_devices->opened = 1;
7024 fs_devices = clone_fs_devices(fs_devices);
7025 if (IS_ERR(fs_devices))
7028 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7030 free_fs_devices(fs_devices);
7031 fs_devices = ERR_PTR(ret);
7035 if (!fs_devices->seeding) {
7036 close_fs_devices(fs_devices);
7037 free_fs_devices(fs_devices);
7038 fs_devices = ERR_PTR(-EINVAL);
7042 fs_devices->seed = fs_info->fs_devices->seed;
7043 fs_info->fs_devices->seed = fs_devices;
7048 static int read_one_dev(struct btrfs_fs_info *fs_info,
7049 struct extent_buffer *leaf,
7050 struct btrfs_dev_item *dev_item)
7052 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7053 struct btrfs_device *device;
7056 u8 fs_uuid[BTRFS_FSID_SIZE];
7057 u8 dev_uuid[BTRFS_UUID_SIZE];
7059 devid = btrfs_device_id(leaf, dev_item);
7060 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7062 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7065 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7066 fs_devices = open_seed_devices(fs_info, fs_uuid);
7067 if (IS_ERR(fs_devices))
7068 return PTR_ERR(fs_devices);
7071 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
7073 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7074 btrfs_report_missing_device(fs_info, devid,
7079 device = add_missing_dev(fs_devices, devid, dev_uuid);
7080 if (IS_ERR(device)) {
7082 "failed to add missing dev %llu: %ld",
7083 devid, PTR_ERR(device));
7084 return PTR_ERR(device);
7086 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7088 if (!device->bdev) {
7089 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7090 btrfs_report_missing_device(fs_info,
7091 devid, dev_uuid, true);
7094 btrfs_report_missing_device(fs_info, devid,
7098 if (!device->bdev &&
7099 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7101 * this happens when a device that was properly setup
7102 * in the device info lists suddenly goes bad.
7103 * device->bdev is NULL, and so we have to set
7104 * device->missing to one here
7106 device->fs_devices->missing_devices++;
7107 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7110 /* Move the device to its own fs_devices */
7111 if (device->fs_devices != fs_devices) {
7112 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7113 &device->dev_state));
7115 list_move(&device->dev_list, &fs_devices->devices);
7116 device->fs_devices->num_devices--;
7117 fs_devices->num_devices++;
7119 device->fs_devices->missing_devices--;
7120 fs_devices->missing_devices++;
7122 device->fs_devices = fs_devices;
7126 if (device->fs_devices != fs_info->fs_devices) {
7127 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7128 if (device->generation !=
7129 btrfs_device_generation(leaf, dev_item))
7133 fill_device_from_item(leaf, dev_item, device);
7134 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7135 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7136 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7137 device->fs_devices->total_rw_bytes += device->total_bytes;
7138 atomic64_add(device->total_bytes - device->bytes_used,
7139 &fs_info->free_chunk_space);
7145 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7147 struct btrfs_root *root = fs_info->tree_root;
7148 struct btrfs_super_block *super_copy = fs_info->super_copy;
7149 struct extent_buffer *sb;
7150 struct btrfs_disk_key *disk_key;
7151 struct btrfs_chunk *chunk;
7153 unsigned long sb_array_offset;
7160 struct btrfs_key key;
7162 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7164 * This will create extent buffer of nodesize, superblock size is
7165 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7166 * overallocate but we can keep it as-is, only the first page is used.
7168 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7171 set_extent_buffer_uptodate(sb);
7172 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7174 * The sb extent buffer is artificial and just used to read the system array.
7175 * set_extent_buffer_uptodate() call does not properly mark all it's
7176 * pages up-to-date when the page is larger: extent does not cover the
7177 * whole page and consequently check_page_uptodate does not find all
7178 * the page's extents up-to-date (the hole beyond sb),
7179 * write_extent_buffer then triggers a WARN_ON.
7181 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7182 * but sb spans only this function. Add an explicit SetPageUptodate call
7183 * to silence the warning eg. on PowerPC 64.
7185 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7186 SetPageUptodate(sb->pages[0]);
7188 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7189 array_size = btrfs_super_sys_array_size(super_copy);
7191 array_ptr = super_copy->sys_chunk_array;
7192 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7195 while (cur_offset < array_size) {
7196 disk_key = (struct btrfs_disk_key *)array_ptr;
7197 len = sizeof(*disk_key);
7198 if (cur_offset + len > array_size)
7199 goto out_short_read;
7201 btrfs_disk_key_to_cpu(&key, disk_key);
7204 sb_array_offset += len;
7207 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7208 chunk = (struct btrfs_chunk *)sb_array_offset;
7210 * At least one btrfs_chunk with one stripe must be
7211 * present, exact stripe count check comes afterwards
7213 len = btrfs_chunk_item_size(1);
7214 if (cur_offset + len > array_size)
7215 goto out_short_read;
7217 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7220 "invalid number of stripes %u in sys_array at offset %u",
7221 num_stripes, cur_offset);
7226 type = btrfs_chunk_type(sb, chunk);
7227 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7229 "invalid chunk type %llu in sys_array at offset %u",
7235 len = btrfs_chunk_item_size(num_stripes);
7236 if (cur_offset + len > array_size)
7237 goto out_short_read;
7239 ret = read_one_chunk(fs_info, &key, sb, chunk);
7244 "unexpected item type %u in sys_array at offset %u",
7245 (u32)key.type, cur_offset);
7250 sb_array_offset += len;
7253 clear_extent_buffer_uptodate(sb);
7254 free_extent_buffer_stale(sb);
7258 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7260 clear_extent_buffer_uptodate(sb);
7261 free_extent_buffer_stale(sb);
7266 * Check if all chunks in the fs are OK for read-write degraded mount
7268 * If the @failing_dev is specified, it's accounted as missing.
7270 * Return true if all chunks meet the minimal RW mount requirements.
7271 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7273 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7274 struct btrfs_device *failing_dev)
7276 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
7277 struct extent_map *em;
7281 read_lock(&map_tree->map_tree.lock);
7282 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
7283 read_unlock(&map_tree->map_tree.lock);
7284 /* No chunk at all? Return false anyway */
7290 struct map_lookup *map;
7295 map = em->map_lookup;
7297 btrfs_get_num_tolerated_disk_barrier_failures(
7299 for (i = 0; i < map->num_stripes; i++) {
7300 struct btrfs_device *dev = map->stripes[i].dev;
7302 if (!dev || !dev->bdev ||
7303 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7304 dev->last_flush_error)
7306 else if (failing_dev && failing_dev == dev)
7309 if (missing > max_tolerated) {
7312 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
7313 em->start, missing, max_tolerated);
7314 free_extent_map(em);
7318 next_start = extent_map_end(em);
7319 free_extent_map(em);
7321 read_lock(&map_tree->map_tree.lock);
7322 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
7323 (u64)(-1) - next_start);
7324 read_unlock(&map_tree->map_tree.lock);
7330 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7332 struct btrfs_root *root = fs_info->chunk_root;
7333 struct btrfs_path *path;
7334 struct extent_buffer *leaf;
7335 struct btrfs_key key;
7336 struct btrfs_key found_key;
7341 path = btrfs_alloc_path();
7346 * uuid_mutex is needed only if we are mounting a sprout FS
7347 * otherwise we don't need it.
7349 mutex_lock(&uuid_mutex);
7350 mutex_lock(&fs_info->chunk_mutex);
7353 * Read all device items, and then all the chunk items. All
7354 * device items are found before any chunk item (their object id
7355 * is smaller than the lowest possible object id for a chunk
7356 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7358 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7361 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7365 leaf = path->nodes[0];
7366 slot = path->slots[0];
7367 if (slot >= btrfs_header_nritems(leaf)) {
7368 ret = btrfs_next_leaf(root, path);
7375 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7376 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7377 struct btrfs_dev_item *dev_item;
7378 dev_item = btrfs_item_ptr(leaf, slot,
7379 struct btrfs_dev_item);
7380 ret = read_one_dev(fs_info, leaf, dev_item);
7384 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7385 struct btrfs_chunk *chunk;
7386 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7387 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
7395 * After loading chunk tree, we've got all device information,
7396 * do another round of validation checks.
7398 if (total_dev != fs_info->fs_devices->total_devices) {
7400 "super_num_devices %llu mismatch with num_devices %llu found here",
7401 btrfs_super_num_devices(fs_info->super_copy),
7406 if (btrfs_super_total_bytes(fs_info->super_copy) <
7407 fs_info->fs_devices->total_rw_bytes) {
7409 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7410 btrfs_super_total_bytes(fs_info->super_copy),
7411 fs_info->fs_devices->total_rw_bytes);
7417 mutex_unlock(&fs_info->chunk_mutex);
7418 mutex_unlock(&uuid_mutex);
7420 btrfs_free_path(path);
7424 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7426 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7427 struct btrfs_device *device;
7429 while (fs_devices) {
7430 mutex_lock(&fs_devices->device_list_mutex);
7431 list_for_each_entry(device, &fs_devices->devices, dev_list)
7432 device->fs_info = fs_info;
7433 mutex_unlock(&fs_devices->device_list_mutex);
7435 fs_devices = fs_devices->seed;
7439 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7443 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7444 btrfs_dev_stat_reset(dev, i);
7447 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7449 struct btrfs_key key;
7450 struct btrfs_key found_key;
7451 struct btrfs_root *dev_root = fs_info->dev_root;
7452 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7453 struct extent_buffer *eb;
7456 struct btrfs_device *device;
7457 struct btrfs_path *path = NULL;
7460 path = btrfs_alloc_path();
7466 mutex_lock(&fs_devices->device_list_mutex);
7467 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7469 struct btrfs_dev_stats_item *ptr;
7471 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7472 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7473 key.offset = device->devid;
7474 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7476 __btrfs_reset_dev_stats(device);
7477 device->dev_stats_valid = 1;
7478 btrfs_release_path(path);
7481 slot = path->slots[0];
7482 eb = path->nodes[0];
7483 btrfs_item_key_to_cpu(eb, &found_key, slot);
7484 item_size = btrfs_item_size_nr(eb, slot);
7486 ptr = btrfs_item_ptr(eb, slot,
7487 struct btrfs_dev_stats_item);
7489 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7490 if (item_size >= (1 + i) * sizeof(__le64))
7491 btrfs_dev_stat_set(device, i,
7492 btrfs_dev_stats_value(eb, ptr, i));
7494 btrfs_dev_stat_reset(device, i);
7497 device->dev_stats_valid = 1;
7498 btrfs_dev_stat_print_on_load(device);
7499 btrfs_release_path(path);
7501 mutex_unlock(&fs_devices->device_list_mutex);
7504 btrfs_free_path(path);
7505 return ret < 0 ? ret : 0;
7508 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7509 struct btrfs_device *device)
7511 struct btrfs_fs_info *fs_info = trans->fs_info;
7512 struct btrfs_root *dev_root = fs_info->dev_root;
7513 struct btrfs_path *path;
7514 struct btrfs_key key;
7515 struct extent_buffer *eb;
7516 struct btrfs_dev_stats_item *ptr;
7520 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7521 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7522 key.offset = device->devid;
7524 path = btrfs_alloc_path();
7527 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7529 btrfs_warn_in_rcu(fs_info,
7530 "error %d while searching for dev_stats item for device %s",
7531 ret, rcu_str_deref(device->name));
7536 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7537 /* need to delete old one and insert a new one */
7538 ret = btrfs_del_item(trans, dev_root, path);
7540 btrfs_warn_in_rcu(fs_info,
7541 "delete too small dev_stats item for device %s failed %d",
7542 rcu_str_deref(device->name), ret);
7549 /* need to insert a new item */
7550 btrfs_release_path(path);
7551 ret = btrfs_insert_empty_item(trans, dev_root, path,
7552 &key, sizeof(*ptr));
7554 btrfs_warn_in_rcu(fs_info,
7555 "insert dev_stats item for device %s failed %d",
7556 rcu_str_deref(device->name), ret);
7561 eb = path->nodes[0];
7562 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7563 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7564 btrfs_set_dev_stats_value(eb, ptr, i,
7565 btrfs_dev_stat_read(device, i));
7566 btrfs_mark_buffer_dirty(eb);
7569 btrfs_free_path(path);
7574 * called from commit_transaction. Writes all changed device stats to disk.
7576 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7577 struct btrfs_fs_info *fs_info)
7579 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7580 struct btrfs_device *device;
7584 mutex_lock(&fs_devices->device_list_mutex);
7585 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7586 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7587 if (!device->dev_stats_valid || stats_cnt == 0)
7592 * There is a LOAD-LOAD control dependency between the value of
7593 * dev_stats_ccnt and updating the on-disk values which requires
7594 * reading the in-memory counters. Such control dependencies
7595 * require explicit read memory barriers.
7597 * This memory barriers pairs with smp_mb__before_atomic in
7598 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7599 * barrier implied by atomic_xchg in
7600 * btrfs_dev_stats_read_and_reset
7604 ret = update_dev_stat_item(trans, device);
7606 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7608 mutex_unlock(&fs_devices->device_list_mutex);
7613 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7615 btrfs_dev_stat_inc(dev, index);
7616 btrfs_dev_stat_print_on_error(dev);
7619 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7621 if (!dev->dev_stats_valid)
7623 btrfs_err_rl_in_rcu(dev->fs_info,
7624 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7625 rcu_str_deref(dev->name),
7626 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7627 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7628 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7629 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7630 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7633 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7637 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7638 if (btrfs_dev_stat_read(dev, i) != 0)
7640 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7641 return; /* all values == 0, suppress message */
7643 btrfs_info_in_rcu(dev->fs_info,
7644 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7645 rcu_str_deref(dev->name),
7646 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7647 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7648 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7649 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7650 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7653 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7654 struct btrfs_ioctl_get_dev_stats *stats)
7656 struct btrfs_device *dev;
7657 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7660 mutex_lock(&fs_devices->device_list_mutex);
7661 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7662 mutex_unlock(&fs_devices->device_list_mutex);
7665 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7667 } else if (!dev->dev_stats_valid) {
7668 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7670 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7671 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7672 if (stats->nr_items > i)
7674 btrfs_dev_stat_read_and_reset(dev, i);
7676 btrfs_dev_stat_reset(dev, i);
7679 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7680 if (stats->nr_items > i)
7681 stats->values[i] = btrfs_dev_stat_read(dev, i);
7683 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7684 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7688 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7690 struct buffer_head *bh;
7691 struct btrfs_super_block *disk_super;
7697 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7700 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7703 disk_super = (struct btrfs_super_block *)bh->b_data;
7705 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7706 set_buffer_dirty(bh);
7707 sync_dirty_buffer(bh);
7711 /* Notify udev that device has changed */
7712 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7714 /* Update ctime/mtime for device path for libblkid */
7715 update_dev_time(device_path);
7719 * Update the size of all devices, which is used for writing out the
7722 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7724 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7725 struct btrfs_device *curr, *next;
7727 if (list_empty(&fs_devices->resized_devices))
7730 mutex_lock(&fs_devices->device_list_mutex);
7731 mutex_lock(&fs_info->chunk_mutex);
7732 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7734 list_del_init(&curr->resized_list);
7735 curr->commit_total_bytes = curr->disk_total_bytes;
7737 mutex_unlock(&fs_info->chunk_mutex);
7738 mutex_unlock(&fs_devices->device_list_mutex);
7741 /* Must be invoked during the transaction commit */
7742 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7744 struct btrfs_fs_info *fs_info = trans->fs_info;
7745 struct extent_map *em;
7746 struct map_lookup *map;
7747 struct btrfs_device *dev;
7750 if (list_empty(&trans->pending_chunks))
7753 /* In order to kick the device replace finish process */
7754 mutex_lock(&fs_info->chunk_mutex);
7755 list_for_each_entry(em, &trans->pending_chunks, list) {
7756 map = em->map_lookup;
7758 for (i = 0; i < map->num_stripes; i++) {
7759 dev = map->stripes[i].dev;
7760 dev->commit_bytes_used = dev->bytes_used;
7763 mutex_unlock(&fs_info->chunk_mutex);
7766 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7768 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7769 while (fs_devices) {
7770 fs_devices->fs_info = fs_info;
7771 fs_devices = fs_devices->seed;
7775 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7777 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7778 while (fs_devices) {
7779 fs_devices->fs_info = NULL;
7780 fs_devices = fs_devices->seed;
7785 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7787 int btrfs_bg_type_to_factor(u64 flags)
7789 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
7790 BTRFS_BLOCK_GROUP_RAID10))
7796 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7798 int index = btrfs_bg_flags_to_raid_index(type);
7799 int ncopies = btrfs_raid_array[index].ncopies;
7802 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
7803 case BTRFS_BLOCK_GROUP_RAID5:
7804 data_stripes = num_stripes - 1;
7806 case BTRFS_BLOCK_GROUP_RAID6:
7807 data_stripes = num_stripes - 2;
7810 data_stripes = num_stripes / ncopies;
7813 return div_u64(chunk_len, data_stripes);
7816 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7817 u64 chunk_offset, u64 devid,
7818 u64 physical_offset, u64 physical_len)
7820 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7821 struct extent_map *em;
7822 struct map_lookup *map;
7823 struct btrfs_device *dev;
7829 read_lock(&em_tree->lock);
7830 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7831 read_unlock(&em_tree->lock);
7835 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7836 physical_offset, devid);
7841 map = em->map_lookup;
7842 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7843 if (physical_len != stripe_len) {
7845 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7846 physical_offset, devid, em->start, physical_len,
7852 for (i = 0; i < map->num_stripes; i++) {
7853 if (map->stripes[i].dev->devid == devid &&
7854 map->stripes[i].physical == physical_offset) {
7856 if (map->verified_stripes >= map->num_stripes) {
7858 "too many dev extents for chunk %llu found",
7863 map->verified_stripes++;
7869 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7870 physical_offset, devid);
7874 /* Make sure no dev extent is beyond device bondary */
7875 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
7877 btrfs_err(fs_info, "failed to find devid %llu", devid);
7881 if (physical_offset + physical_len > dev->disk_total_bytes) {
7883 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7884 devid, physical_offset, physical_len,
7885 dev->disk_total_bytes);
7890 free_extent_map(em);
7894 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7896 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7897 struct extent_map *em;
7898 struct rb_node *node;
7901 read_lock(&em_tree->lock);
7902 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7903 em = rb_entry(node, struct extent_map, rb_node);
7904 if (em->map_lookup->num_stripes !=
7905 em->map_lookup->verified_stripes) {
7907 "chunk %llu has missing dev extent, have %d expect %d",
7908 em->start, em->map_lookup->verified_stripes,
7909 em->map_lookup->num_stripes);
7915 read_unlock(&em_tree->lock);
7920 * Ensure that all dev extents are mapped to correct chunk, otherwise
7921 * later chunk allocation/free would cause unexpected behavior.
7923 * NOTE: This will iterate through the whole device tree, which should be of
7924 * the same size level as the chunk tree. This slightly increases mount time.
7926 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7928 struct btrfs_path *path;
7929 struct btrfs_root *root = fs_info->dev_root;
7930 struct btrfs_key key;
7932 u64 prev_dev_ext_end = 0;
7936 key.type = BTRFS_DEV_EXTENT_KEY;
7939 path = btrfs_alloc_path();
7943 path->reada = READA_FORWARD;
7944 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7948 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7949 ret = btrfs_next_item(root, path);
7952 /* No dev extents at all? Not good */
7959 struct extent_buffer *leaf = path->nodes[0];
7960 struct btrfs_dev_extent *dext;
7961 int slot = path->slots[0];
7963 u64 physical_offset;
7967 btrfs_item_key_to_cpu(leaf, &key, slot);
7968 if (key.type != BTRFS_DEV_EXTENT_KEY)
7970 devid = key.objectid;
7971 physical_offset = key.offset;
7973 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7974 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7975 physical_len = btrfs_dev_extent_length(leaf, dext);
7977 /* Check if this dev extent overlaps with the previous one */
7978 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7980 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7981 devid, physical_offset, prev_dev_ext_end);
7986 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7987 physical_offset, physical_len);
7991 prev_dev_ext_end = physical_offset + physical_len;
7993 ret = btrfs_next_item(root, path);
8002 /* Ensure all chunks have corresponding dev extents */
8003 ret = verify_chunk_dev_extent_mapping(fs_info);
8005 btrfs_free_path(path);
8010 * Check whether the given block group or device is pinned by any inode being
8011 * used as a swapfile.
8013 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8015 struct btrfs_swapfile_pin *sp;
8016 struct rb_node *node;
8018 spin_lock(&fs_info->swapfile_pins_lock);
8019 node = fs_info->swapfile_pins.rb_node;
8021 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8023 node = node->rb_left;
8024 else if (ptr > sp->ptr)
8025 node = node->rb_right;
8029 spin_unlock(&fs_info->swapfile_pins_lock);
8030 return node != NULL;