1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
32 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
33 [BTRFS_RAID_RAID10] = {
36 .devs_max = 0, /* 0 == as many as possible */
38 .tolerated_failures = 1,
42 .raid_name = "raid10",
43 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
44 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
46 [BTRFS_RAID_RAID1] = {
51 .tolerated_failures = 1,
56 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
57 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
64 .tolerated_failures = 0,
69 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
72 [BTRFS_RAID_RAID0] = {
77 .tolerated_failures = 0,
82 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
85 [BTRFS_RAID_SINGLE] = {
90 .tolerated_failures = 0,
94 .raid_name = "single",
98 [BTRFS_RAID_RAID5] = {
103 .tolerated_failures = 1,
107 .raid_name = "raid5",
108 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
109 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
111 [BTRFS_RAID_RAID6] = {
116 .tolerated_failures = 2,
120 .raid_name = "raid6",
121 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
122 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
126 const char *get_raid_name(enum btrfs_raid_types type)
128 if (type >= BTRFS_NR_RAID_TYPES)
131 return btrfs_raid_array[type].raid_name;
135 * Fill @buf with textual description of @bg_flags, no more than @size_buf
136 * bytes including terminating null byte.
138 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
143 u64 flags = bg_flags;
144 u32 size_bp = size_buf;
151 #define DESCRIBE_FLAG(flag, desc) \
153 if (flags & (flag)) { \
154 ret = snprintf(bp, size_bp, "%s|", (desc)); \
155 if (ret < 0 || ret >= size_bp) \
163 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
164 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
165 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
167 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
168 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
169 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
170 btrfs_raid_array[i].raid_name);
174 ret = snprintf(bp, size_bp, "0x%llx|", flags);
178 if (size_bp < size_buf)
179 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
182 * The text is trimmed, it's up to the caller to provide sufficiently
188 static int init_first_rw_device(struct btrfs_trans_handle *trans);
189 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, opening/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->alloc_list);
322 INIT_LIST_HEAD(&fs_devs->fs_list);
324 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
327 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
329 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
334 void btrfs_free_device(struct btrfs_device *device)
336 WARN_ON(!list_empty(&device->post_commit_list));
337 rcu_string_free(device->name);
338 extent_io_tree_release(&device->alloc_state);
339 bio_put(device->flush_bio);
343 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
345 struct btrfs_device *device;
346 WARN_ON(fs_devices->opened);
347 while (!list_empty(&fs_devices->devices)) {
348 device = list_entry(fs_devices->devices.next,
349 struct btrfs_device, dev_list);
350 list_del(&device->dev_list);
351 btrfs_free_device(device);
356 static void btrfs_kobject_uevent(struct block_device *bdev,
357 enum kobject_action action)
361 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
363 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
365 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
366 &disk_to_dev(bdev->bd_disk)->kobj);
369 void __exit btrfs_cleanup_fs_uuids(void)
371 struct btrfs_fs_devices *fs_devices;
373 while (!list_empty(&fs_uuids)) {
374 fs_devices = list_entry(fs_uuids.next,
375 struct btrfs_fs_devices, fs_list);
376 list_del(&fs_devices->fs_list);
377 free_fs_devices(fs_devices);
382 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
383 * Returned struct is not linked onto any lists and must be destroyed using
386 static struct btrfs_device *__alloc_device(void)
388 struct btrfs_device *dev;
390 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
392 return ERR_PTR(-ENOMEM);
395 * Preallocate a bio that's always going to be used for flushing device
396 * barriers and matches the device lifespan
398 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
399 if (!dev->flush_bio) {
401 return ERR_PTR(-ENOMEM);
404 INIT_LIST_HEAD(&dev->dev_list);
405 INIT_LIST_HEAD(&dev->dev_alloc_list);
406 INIT_LIST_HEAD(&dev->post_commit_list);
408 spin_lock_init(&dev->io_lock);
410 atomic_set(&dev->reada_in_flight, 0);
411 atomic_set(&dev->dev_stats_ccnt, 0);
412 btrfs_device_data_ordered_init(dev);
413 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
414 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
415 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
420 static noinline struct btrfs_fs_devices *find_fsid(
421 const u8 *fsid, const u8 *metadata_fsid)
423 struct btrfs_fs_devices *fs_devices;
429 * Handle scanned device having completed its fsid change but
430 * belonging to a fs_devices that was created by first scanning
431 * a device which didn't have its fsid/metadata_uuid changed
432 * at all and the CHANGING_FSID_V2 flag set.
434 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
435 if (fs_devices->fsid_change &&
436 memcmp(metadata_fsid, fs_devices->fsid,
437 BTRFS_FSID_SIZE) == 0 &&
438 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
439 BTRFS_FSID_SIZE) == 0) {
444 * Handle scanned device having completed its fsid change but
445 * belonging to a fs_devices that was created by a device that
446 * has an outdated pair of fsid/metadata_uuid and
447 * CHANGING_FSID_V2 flag set.
449 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
450 if (fs_devices->fsid_change &&
451 memcmp(fs_devices->metadata_uuid,
452 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
453 memcmp(metadata_fsid, fs_devices->metadata_uuid,
454 BTRFS_FSID_SIZE) == 0) {
460 /* Handle non-split brain cases */
461 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
463 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
464 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
465 BTRFS_FSID_SIZE) == 0)
468 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
476 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
477 int flush, struct block_device **bdev,
478 struct buffer_head **bh)
482 *bdev = blkdev_get_by_path(device_path, flags, holder);
485 ret = PTR_ERR(*bdev);
490 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
491 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
493 blkdev_put(*bdev, flags);
496 invalidate_bdev(*bdev);
497 *bh = btrfs_read_dev_super(*bdev);
500 blkdev_put(*bdev, flags);
512 static void requeue_list(struct btrfs_pending_bios *pending_bios,
513 struct bio *head, struct bio *tail)
516 struct bio *old_head;
518 old_head = pending_bios->head;
519 pending_bios->head = head;
520 if (pending_bios->tail)
521 tail->bi_next = old_head;
523 pending_bios->tail = tail;
527 * we try to collect pending bios for a device so we don't get a large
528 * number of procs sending bios down to the same device. This greatly
529 * improves the schedulers ability to collect and merge the bios.
531 * But, it also turns into a long list of bios to process and that is sure
532 * to eventually make the worker thread block. The solution here is to
533 * make some progress and then put this work struct back at the end of
534 * the list if the block device is congested. This way, multiple devices
535 * can make progress from a single worker thread.
537 static noinline void run_scheduled_bios(struct btrfs_device *device)
539 struct btrfs_fs_info *fs_info = device->fs_info;
541 struct backing_dev_info *bdi;
542 struct btrfs_pending_bios *pending_bios;
546 unsigned long num_run;
547 unsigned long batch_run = 0;
548 unsigned long last_waited = 0;
550 int sync_pending = 0;
551 struct blk_plug plug;
554 * this function runs all the bios we've collected for
555 * a particular device. We don't want to wander off to
556 * another device without first sending all of these down.
557 * So, setup a plug here and finish it off before we return
559 blk_start_plug(&plug);
561 bdi = device->bdev->bd_bdi;
564 spin_lock(&device->io_lock);
569 /* take all the bios off the list at once and process them
570 * later on (without the lock held). But, remember the
571 * tail and other pointers so the bios can be properly reinserted
572 * into the list if we hit congestion
574 if (!force_reg && device->pending_sync_bios.head) {
575 pending_bios = &device->pending_sync_bios;
578 pending_bios = &device->pending_bios;
582 pending = pending_bios->head;
583 tail = pending_bios->tail;
584 WARN_ON(pending && !tail);
587 * if pending was null this time around, no bios need processing
588 * at all and we can stop. Otherwise it'll loop back up again
589 * and do an additional check so no bios are missed.
591 * device->running_pending is used to synchronize with the
594 if (device->pending_sync_bios.head == NULL &&
595 device->pending_bios.head == NULL) {
597 device->running_pending = 0;
600 device->running_pending = 1;
603 pending_bios->head = NULL;
604 pending_bios->tail = NULL;
606 spin_unlock(&device->io_lock);
611 /* we want to work on both lists, but do more bios on the
612 * sync list than the regular list
615 pending_bios != &device->pending_sync_bios &&
616 device->pending_sync_bios.head) ||
617 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
618 device->pending_bios.head)) {
619 spin_lock(&device->io_lock);
620 requeue_list(pending_bios, pending, tail);
625 pending = pending->bi_next;
628 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
631 * if we're doing the sync list, record that our
632 * plug has some sync requests on it
634 * If we're doing the regular list and there are
635 * sync requests sitting around, unplug before
638 if (pending_bios == &device->pending_sync_bios) {
640 } else if (sync_pending) {
641 blk_finish_plug(&plug);
642 blk_start_plug(&plug);
646 btrfsic_submit_bio(cur);
653 * we made progress, there is more work to do and the bdi
654 * is now congested. Back off and let other work structs
657 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
658 fs_info->fs_devices->open_devices > 1) {
659 struct io_context *ioc;
661 ioc = current->io_context;
664 * the main goal here is that we don't want to
665 * block if we're going to be able to submit
666 * more requests without blocking.
668 * This code does two great things, it pokes into
669 * the elevator code from a filesystem _and_
670 * it makes assumptions about how batching works.
672 if (ioc && ioc->nr_batch_requests > 0 &&
673 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
675 ioc->last_waited == last_waited)) {
677 * we want to go through our batch of
678 * requests and stop. So, we copy out
679 * the ioc->last_waited time and test
680 * against it before looping
682 last_waited = ioc->last_waited;
686 spin_lock(&device->io_lock);
687 requeue_list(pending_bios, pending, tail);
688 device->running_pending = 1;
690 spin_unlock(&device->io_lock);
691 btrfs_queue_work(fs_info->submit_workers,
701 spin_lock(&device->io_lock);
702 if (device->pending_bios.head || device->pending_sync_bios.head)
704 spin_unlock(&device->io_lock);
707 blk_finish_plug(&plug);
710 static void pending_bios_fn(struct btrfs_work *work)
712 struct btrfs_device *device;
714 device = container_of(work, struct btrfs_device, work);
715 run_scheduled_bios(device);
718 static bool device_path_matched(const char *path, struct btrfs_device *device)
723 found = strcmp(rcu_str_deref(device->name), path);
730 * Search and remove all stale (devices which are not mounted) devices.
731 * When both inputs are NULL, it will search and release all stale devices.
732 * path: Optional. When provided will it release all unmounted devices
733 * matching this path only.
734 * skip_dev: Optional. Will skip this device when searching for the stale
736 * Return: 0 for success or if @path is NULL.
737 * -EBUSY if @path is a mounted device.
738 * -ENOENT if @path does not match any device in the list.
740 static int btrfs_free_stale_devices(const char *path,
741 struct btrfs_device *skip_device)
743 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
744 struct btrfs_device *device, *tmp_device;
750 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
752 mutex_lock(&fs_devices->device_list_mutex);
753 list_for_each_entry_safe(device, tmp_device,
754 &fs_devices->devices, dev_list) {
755 if (skip_device && skip_device == device)
757 if (path && !device->name)
759 if (path && !device_path_matched(path, device))
761 if (fs_devices->opened) {
762 /* for an already deleted device return 0 */
763 if (path && ret != 0)
768 /* delete the stale device */
769 fs_devices->num_devices--;
770 list_del(&device->dev_list);
771 btrfs_free_device(device);
774 if (fs_devices->num_devices == 0)
777 mutex_unlock(&fs_devices->device_list_mutex);
779 if (fs_devices->num_devices == 0) {
780 btrfs_sysfs_remove_fsid(fs_devices);
781 list_del(&fs_devices->fs_list);
782 free_fs_devices(fs_devices);
789 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
790 struct btrfs_device *device, fmode_t flags,
793 struct request_queue *q;
794 struct block_device *bdev;
795 struct buffer_head *bh;
796 struct btrfs_super_block *disk_super;
805 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
810 disk_super = (struct btrfs_super_block *)bh->b_data;
811 devid = btrfs_stack_device_id(&disk_super->dev_item);
812 if (devid != device->devid)
815 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
818 device->generation = btrfs_super_generation(disk_super);
820 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
821 if (btrfs_super_incompat_flags(disk_super) &
822 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
824 "BTRFS: Invalid seeding and uuid-changed device detected\n");
828 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
829 fs_devices->seeding = 1;
831 if (bdev_read_only(bdev))
832 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
834 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
837 q = bdev_get_queue(bdev);
838 if (!blk_queue_nonrot(q))
839 fs_devices->rotating = 1;
842 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
843 device->mode = flags;
845 fs_devices->open_devices++;
846 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
847 device->devid != BTRFS_DEV_REPLACE_DEVID) {
848 fs_devices->rw_devices++;
849 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
857 blkdev_put(bdev, flags);
863 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
864 * being created with a disk that has already completed its fsid change.
866 static struct btrfs_fs_devices *find_fsid_inprogress(
867 struct btrfs_super_block *disk_super)
869 struct btrfs_fs_devices *fs_devices;
871 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
872 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
873 BTRFS_FSID_SIZE) != 0 &&
874 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
875 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
884 static struct btrfs_fs_devices *find_fsid_changed(
885 struct btrfs_super_block *disk_super)
887 struct btrfs_fs_devices *fs_devices;
890 * Handles the case where scanned device is part of an fs that had
891 * multiple successful changes of FSID but curently device didn't
892 * observe it. Meaning our fsid will be different than theirs.
894 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
895 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
896 BTRFS_FSID_SIZE) != 0 &&
897 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
898 BTRFS_FSID_SIZE) == 0 &&
899 memcmp(fs_devices->fsid, disk_super->fsid,
900 BTRFS_FSID_SIZE) != 0) {
908 * Add new device to list of registered devices
911 * device pointer which was just added or updated when successful
912 * error pointer when failed
914 static noinline struct btrfs_device *device_list_add(const char *path,
915 struct btrfs_super_block *disk_super,
916 bool *new_device_added)
918 struct btrfs_device *device;
919 struct btrfs_fs_devices *fs_devices = NULL;
920 struct rcu_string *name;
921 u64 found_transid = btrfs_super_generation(disk_super);
922 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
923 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
924 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
925 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
926 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
928 if (fsid_change_in_progress) {
929 if (!has_metadata_uuid) {
931 * When we have an image which has CHANGING_FSID_V2 set
932 * it might belong to either a filesystem which has
933 * disks with completed fsid change or it might belong
934 * to fs with no UUID changes in effect, handle both.
936 fs_devices = find_fsid_inprogress(disk_super);
938 fs_devices = find_fsid(disk_super->fsid, NULL);
940 fs_devices = find_fsid_changed(disk_super);
942 } else if (has_metadata_uuid) {
943 fs_devices = find_fsid(disk_super->fsid,
944 disk_super->metadata_uuid);
946 fs_devices = find_fsid(disk_super->fsid, NULL);
951 if (has_metadata_uuid)
952 fs_devices = alloc_fs_devices(disk_super->fsid,
953 disk_super->metadata_uuid);
955 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
957 if (IS_ERR(fs_devices))
958 return ERR_CAST(fs_devices);
960 fs_devices->fsid_change = fsid_change_in_progress;
962 mutex_lock(&fs_devices->device_list_mutex);
963 list_add(&fs_devices->fs_list, &fs_uuids);
967 mutex_lock(&fs_devices->device_list_mutex);
968 device = btrfs_find_device(fs_devices, devid,
969 disk_super->dev_item.uuid, NULL, false);
972 * If this disk has been pulled into an fs devices created by
973 * a device which had the CHANGING_FSID_V2 flag then replace the
974 * metadata_uuid/fsid values of the fs_devices.
976 if (has_metadata_uuid && fs_devices->fsid_change &&
977 found_transid > fs_devices->latest_generation) {
978 memcpy(fs_devices->fsid, disk_super->fsid,
980 memcpy(fs_devices->metadata_uuid,
981 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
983 fs_devices->fsid_change = false;
988 if (fs_devices->opened) {
989 mutex_unlock(&fs_devices->device_list_mutex);
990 return ERR_PTR(-EBUSY);
993 device = btrfs_alloc_device(NULL, &devid,
994 disk_super->dev_item.uuid);
995 if (IS_ERR(device)) {
996 mutex_unlock(&fs_devices->device_list_mutex);
997 /* we can safely leave the fs_devices entry around */
1001 name = rcu_string_strdup(path, GFP_NOFS);
1003 btrfs_free_device(device);
1004 mutex_unlock(&fs_devices->device_list_mutex);
1005 return ERR_PTR(-ENOMEM);
1007 rcu_assign_pointer(device->name, name);
1009 list_add_rcu(&device->dev_list, &fs_devices->devices);
1010 fs_devices->num_devices++;
1012 device->fs_devices = fs_devices;
1013 *new_device_added = true;
1015 if (disk_super->label[0])
1016 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1017 disk_super->label, devid, found_transid, path);
1019 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1020 disk_super->fsid, devid, found_transid, path);
1022 } else if (!device->name || strcmp(device->name->str, path)) {
1024 * When FS is already mounted.
1025 * 1. If you are here and if the device->name is NULL that
1026 * means this device was missing at time of FS mount.
1027 * 2. If you are here and if the device->name is different
1028 * from 'path' that means either
1029 * a. The same device disappeared and reappeared with
1030 * different name. or
1031 * b. The missing-disk-which-was-replaced, has
1034 * We must allow 1 and 2a above. But 2b would be a spurious
1035 * and unintentional.
1037 * Further in case of 1 and 2a above, the disk at 'path'
1038 * would have missed some transaction when it was away and
1039 * in case of 2a the stale bdev has to be updated as well.
1040 * 2b must not be allowed at all time.
1044 * For now, we do allow update to btrfs_fs_device through the
1045 * btrfs dev scan cli after FS has been mounted. We're still
1046 * tracking a problem where systems fail mount by subvolume id
1047 * when we reject replacement on a mounted FS.
1049 if (!fs_devices->opened && found_transid < device->generation) {
1051 * That is if the FS is _not_ mounted and if you
1052 * are here, that means there is more than one
1053 * disk with same uuid and devid.We keep the one
1054 * with larger generation number or the last-in if
1055 * generation are equal.
1057 mutex_unlock(&fs_devices->device_list_mutex);
1058 return ERR_PTR(-EEXIST);
1062 * We are going to replace the device path for a given devid,
1063 * make sure it's the same device if the device is mounted
1066 struct block_device *path_bdev;
1068 path_bdev = lookup_bdev(path);
1069 if (IS_ERR(path_bdev)) {
1070 mutex_unlock(&fs_devices->device_list_mutex);
1071 return ERR_CAST(path_bdev);
1074 if (device->bdev != path_bdev) {
1076 mutex_unlock(&fs_devices->device_list_mutex);
1077 btrfs_warn_in_rcu(device->fs_info,
1078 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1079 disk_super->fsid, devid,
1080 rcu_str_deref(device->name), path);
1081 return ERR_PTR(-EEXIST);
1084 btrfs_info_in_rcu(device->fs_info,
1085 "device fsid %pU devid %llu moved old:%s new:%s",
1086 disk_super->fsid, devid,
1087 rcu_str_deref(device->name), path);
1090 name = rcu_string_strdup(path, GFP_NOFS);
1092 mutex_unlock(&fs_devices->device_list_mutex);
1093 return ERR_PTR(-ENOMEM);
1095 rcu_string_free(device->name);
1096 rcu_assign_pointer(device->name, name);
1097 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1098 fs_devices->missing_devices--;
1099 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1104 * Unmount does not free the btrfs_device struct but would zero
1105 * generation along with most of the other members. So just update
1106 * it back. We need it to pick the disk with largest generation
1109 if (!fs_devices->opened) {
1110 device->generation = found_transid;
1111 fs_devices->latest_generation = max_t(u64, found_transid,
1112 fs_devices->latest_generation);
1115 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1117 mutex_unlock(&fs_devices->device_list_mutex);
1121 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1123 struct btrfs_fs_devices *fs_devices;
1124 struct btrfs_device *device;
1125 struct btrfs_device *orig_dev;
1127 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1128 if (IS_ERR(fs_devices))
1131 mutex_lock(&orig->device_list_mutex);
1132 fs_devices->total_devices = orig->total_devices;
1134 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1135 struct rcu_string *name;
1137 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1143 * This is ok to do without rcu read locked because we hold the
1144 * uuid mutex so nothing we touch in here is going to disappear.
1146 if (orig_dev->name) {
1147 name = rcu_string_strdup(orig_dev->name->str,
1150 btrfs_free_device(device);
1153 rcu_assign_pointer(device->name, name);
1156 list_add(&device->dev_list, &fs_devices->devices);
1157 device->fs_devices = fs_devices;
1158 fs_devices->num_devices++;
1160 mutex_unlock(&orig->device_list_mutex);
1163 mutex_unlock(&orig->device_list_mutex);
1164 free_fs_devices(fs_devices);
1165 return ERR_PTR(-ENOMEM);
1169 * After we have read the system tree and know devids belonging to
1170 * this filesystem, remove the device which does not belong there.
1172 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1174 struct btrfs_device *device, *next;
1175 struct btrfs_device *latest_dev = NULL;
1177 mutex_lock(&uuid_mutex);
1179 /* This is the initialized path, it is safe to release the devices. */
1180 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1181 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1182 &device->dev_state)) {
1183 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1184 &device->dev_state) &&
1186 device->generation > latest_dev->generation)) {
1187 latest_dev = device;
1192 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1194 * In the first step, keep the device which has
1195 * the correct fsid and the devid that is used
1196 * for the dev_replace procedure.
1197 * In the second step, the dev_replace state is
1198 * read from the device tree and it is known
1199 * whether the procedure is really active or
1200 * not, which means whether this device is
1201 * used or whether it should be removed.
1203 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1204 &device->dev_state)) {
1209 blkdev_put(device->bdev, device->mode);
1210 device->bdev = NULL;
1211 fs_devices->open_devices--;
1213 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1214 list_del_init(&device->dev_alloc_list);
1215 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1216 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1217 &device->dev_state))
1218 fs_devices->rw_devices--;
1220 list_del_init(&device->dev_list);
1221 fs_devices->num_devices--;
1222 btrfs_free_device(device);
1225 if (fs_devices->seed) {
1226 fs_devices = fs_devices->seed;
1230 fs_devices->latest_bdev = latest_dev->bdev;
1232 mutex_unlock(&uuid_mutex);
1235 static void btrfs_close_bdev(struct btrfs_device *device)
1240 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1241 sync_blockdev(device->bdev);
1242 invalidate_bdev(device->bdev);
1245 blkdev_put(device->bdev, device->mode);
1248 static void btrfs_close_one_device(struct btrfs_device *device)
1250 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1251 struct btrfs_device *new_device;
1252 struct rcu_string *name;
1255 fs_devices->open_devices--;
1257 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1258 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1259 list_del_init(&device->dev_alloc_list);
1260 fs_devices->rw_devices--;
1263 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1264 fs_devices->missing_devices--;
1266 btrfs_close_bdev(device);
1268 new_device = btrfs_alloc_device(NULL, &device->devid,
1270 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1272 /* Safe because we are under uuid_mutex */
1274 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1275 BUG_ON(!name); /* -ENOMEM */
1276 rcu_assign_pointer(new_device->name, name);
1279 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1280 new_device->fs_devices = device->fs_devices;
1283 btrfs_free_device(device);
1286 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1288 struct btrfs_device *device, *tmp;
1290 if (--fs_devices->opened > 0)
1293 mutex_lock(&fs_devices->device_list_mutex);
1294 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1295 btrfs_close_one_device(device);
1297 mutex_unlock(&fs_devices->device_list_mutex);
1299 WARN_ON(fs_devices->open_devices);
1300 WARN_ON(fs_devices->rw_devices);
1301 fs_devices->opened = 0;
1302 fs_devices->seeding = 0;
1307 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1309 struct btrfs_fs_devices *seed_devices = NULL;
1312 mutex_lock(&uuid_mutex);
1313 ret = close_fs_devices(fs_devices);
1314 if (!fs_devices->opened) {
1315 seed_devices = fs_devices->seed;
1316 fs_devices->seed = NULL;
1318 mutex_unlock(&uuid_mutex);
1320 while (seed_devices) {
1321 fs_devices = seed_devices;
1322 seed_devices = fs_devices->seed;
1323 close_fs_devices(fs_devices);
1324 free_fs_devices(fs_devices);
1329 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1330 fmode_t flags, void *holder)
1332 struct btrfs_device *device;
1333 struct btrfs_device *latest_dev = NULL;
1336 flags |= FMODE_EXCL;
1338 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1339 /* Just open everything we can; ignore failures here */
1340 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1344 device->generation > latest_dev->generation)
1345 latest_dev = device;
1347 if (fs_devices->open_devices == 0) {
1351 fs_devices->opened = 1;
1352 fs_devices->latest_bdev = latest_dev->bdev;
1353 fs_devices->total_rw_bytes = 0;
1358 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1360 struct btrfs_device *dev1, *dev2;
1362 dev1 = list_entry(a, struct btrfs_device, dev_list);
1363 dev2 = list_entry(b, struct btrfs_device, dev_list);
1365 if (dev1->devid < dev2->devid)
1367 else if (dev1->devid > dev2->devid)
1372 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1373 fmode_t flags, void *holder)
1377 lockdep_assert_held(&uuid_mutex);
1379 mutex_lock(&fs_devices->device_list_mutex);
1380 if (fs_devices->opened) {
1381 fs_devices->opened++;
1384 list_sort(NULL, &fs_devices->devices, devid_cmp);
1385 ret = open_fs_devices(fs_devices, flags, holder);
1387 mutex_unlock(&fs_devices->device_list_mutex);
1392 static void btrfs_release_disk_super(struct page *page)
1398 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1400 struct btrfs_super_block **disk_super)
1405 /* make sure our super fits in the device */
1406 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1409 /* make sure our super fits in the page */
1410 if (sizeof(**disk_super) > PAGE_SIZE)
1413 /* make sure our super doesn't straddle pages on disk */
1414 index = bytenr >> PAGE_SHIFT;
1415 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1418 /* pull in the page with our super */
1419 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1422 if (IS_ERR_OR_NULL(*page))
1427 /* align our pointer to the offset of the super block */
1428 *disk_super = p + offset_in_page(bytenr);
1430 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1431 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1432 btrfs_release_disk_super(*page);
1436 if ((*disk_super)->label[0] &&
1437 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1438 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1443 int btrfs_forget_devices(const char *path)
1447 mutex_lock(&uuid_mutex);
1448 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1449 mutex_unlock(&uuid_mutex);
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);
1504 * Try to find a chunk that intersects [start, start + len] range and when one
1505 * such is found, record the end of it in *start
1507 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1510 u64 physical_start, physical_end;
1512 lockdep_assert_held(&device->fs_info->chunk_mutex);
1514 if (!find_first_extent_bit(&device->alloc_state, *start,
1515 &physical_start, &physical_end,
1516 CHUNK_ALLOCATED, NULL)) {
1518 if (in_range(physical_start, *start, len) ||
1519 in_range(*start, physical_start,
1520 physical_end - physical_start)) {
1521 *start = physical_end + 1;
1530 * find_free_dev_extent_start - find free space in the specified device
1531 * @device: the device which we search the free space in
1532 * @num_bytes: the size of the free space that we need
1533 * @search_start: the position from which to begin the search
1534 * @start: store the start of the free space.
1535 * @len: the size of the free space. that we find, or the size
1536 * of the max free space if we don't find suitable free space
1538 * this uses a pretty simple search, the expectation is that it is
1539 * called very infrequently and that a given device has a small number
1542 * @start is used to store the start of the free space if we find. But if we
1543 * don't find suitable free space, it will be used to store the start position
1544 * of the max free space.
1546 * @len is used to store the size of the free space that we find.
1547 * But if we don't find suitable free space, it is used to store the size of
1548 * the max free space.
1550 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
1551 u64 search_start, u64 *start, u64 *len)
1553 struct btrfs_fs_info *fs_info = device->fs_info;
1554 struct btrfs_root *root = fs_info->dev_root;
1555 struct btrfs_key key;
1556 struct btrfs_dev_extent *dev_extent;
1557 struct btrfs_path *path;
1562 u64 search_end = device->total_bytes;
1565 struct extent_buffer *l;
1568 * We don't want to overwrite the superblock on the drive nor any area
1569 * used by the boot loader (grub for example), so we make sure to start
1570 * at an offset of at least 1MB.
1572 search_start = max_t(u64, search_start, SZ_1M);
1574 path = btrfs_alloc_path();
1578 max_hole_start = search_start;
1582 if (search_start >= search_end ||
1583 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1588 path->reada = READA_FORWARD;
1589 path->search_commit_root = 1;
1590 path->skip_locking = 1;
1592 key.objectid = device->devid;
1593 key.offset = search_start;
1594 key.type = BTRFS_DEV_EXTENT_KEY;
1596 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1600 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1607 slot = path->slots[0];
1608 if (slot >= btrfs_header_nritems(l)) {
1609 ret = btrfs_next_leaf(root, path);
1617 btrfs_item_key_to_cpu(l, &key, slot);
1619 if (key.objectid < device->devid)
1622 if (key.objectid > device->devid)
1625 if (key.type != BTRFS_DEV_EXTENT_KEY)
1628 if (key.offset > search_start) {
1629 hole_size = key.offset - search_start;
1632 * Have to check before we set max_hole_start, otherwise
1633 * we could end up sending back this offset anyway.
1635 if (contains_pending_extent(device, &search_start,
1637 if (key.offset >= search_start)
1638 hole_size = key.offset - search_start;
1643 if (hole_size > max_hole_size) {
1644 max_hole_start = search_start;
1645 max_hole_size = hole_size;
1649 * If this free space is greater than which we need,
1650 * it must be the max free space that we have found
1651 * until now, so max_hole_start must point to the start
1652 * of this free space and the length of this free space
1653 * is stored in max_hole_size. Thus, we return
1654 * max_hole_start and max_hole_size and go back to the
1657 if (hole_size >= num_bytes) {
1663 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1664 extent_end = key.offset + btrfs_dev_extent_length(l,
1666 if (extent_end > search_start)
1667 search_start = extent_end;
1674 * At this point, search_start should be the end of
1675 * allocated dev extents, and when shrinking the device,
1676 * search_end may be smaller than search_start.
1678 if (search_end > search_start) {
1679 hole_size = search_end - search_start;
1681 if (contains_pending_extent(device, &search_start, hole_size)) {
1682 btrfs_release_path(path);
1686 if (hole_size > max_hole_size) {
1687 max_hole_start = search_start;
1688 max_hole_size = hole_size;
1693 if (max_hole_size < num_bytes)
1699 btrfs_free_path(path);
1700 *start = max_hole_start;
1702 *len = max_hole_size;
1706 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1707 u64 *start, u64 *len)
1709 /* FIXME use last free of some kind */
1710 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1713 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1714 struct btrfs_device *device,
1715 u64 start, u64 *dev_extent_len)
1717 struct btrfs_fs_info *fs_info = device->fs_info;
1718 struct btrfs_root *root = fs_info->dev_root;
1720 struct btrfs_path *path;
1721 struct btrfs_key key;
1722 struct btrfs_key found_key;
1723 struct extent_buffer *leaf = NULL;
1724 struct btrfs_dev_extent *extent = NULL;
1726 path = btrfs_alloc_path();
1730 key.objectid = device->devid;
1732 key.type = BTRFS_DEV_EXTENT_KEY;
1734 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1736 ret = btrfs_previous_item(root, path, key.objectid,
1737 BTRFS_DEV_EXTENT_KEY);
1740 leaf = path->nodes[0];
1741 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1742 extent = btrfs_item_ptr(leaf, path->slots[0],
1743 struct btrfs_dev_extent);
1744 BUG_ON(found_key.offset > start || found_key.offset +
1745 btrfs_dev_extent_length(leaf, extent) < start);
1747 btrfs_release_path(path);
1749 } else if (ret == 0) {
1750 leaf = path->nodes[0];
1751 extent = btrfs_item_ptr(leaf, path->slots[0],
1752 struct btrfs_dev_extent);
1754 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1758 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1760 ret = btrfs_del_item(trans, root, path);
1762 btrfs_handle_fs_error(fs_info, ret,
1763 "Failed to remove dev extent item");
1765 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1768 btrfs_free_path(path);
1772 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1773 struct btrfs_device *device,
1774 u64 chunk_offset, u64 start, u64 num_bytes)
1777 struct btrfs_path *path;
1778 struct btrfs_fs_info *fs_info = device->fs_info;
1779 struct btrfs_root *root = fs_info->dev_root;
1780 struct btrfs_dev_extent *extent;
1781 struct extent_buffer *leaf;
1782 struct btrfs_key key;
1784 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1785 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1786 path = btrfs_alloc_path();
1790 key.objectid = device->devid;
1792 key.type = BTRFS_DEV_EXTENT_KEY;
1793 ret = btrfs_insert_empty_item(trans, root, path, &key,
1798 leaf = path->nodes[0];
1799 extent = btrfs_item_ptr(leaf, path->slots[0],
1800 struct btrfs_dev_extent);
1801 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1802 BTRFS_CHUNK_TREE_OBJECTID);
1803 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1804 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1805 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1807 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1808 btrfs_mark_buffer_dirty(leaf);
1810 btrfs_free_path(path);
1814 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1816 struct extent_map_tree *em_tree;
1817 struct extent_map *em;
1821 em_tree = &fs_info->mapping_tree.map_tree;
1822 read_lock(&em_tree->lock);
1823 n = rb_last(&em_tree->map.rb_root);
1825 em = rb_entry(n, struct extent_map, rb_node);
1826 ret = em->start + em->len;
1828 read_unlock(&em_tree->lock);
1833 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1837 struct btrfs_key key;
1838 struct btrfs_key found_key;
1839 struct btrfs_path *path;
1841 path = btrfs_alloc_path();
1845 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1846 key.type = BTRFS_DEV_ITEM_KEY;
1847 key.offset = (u64)-1;
1849 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1853 BUG_ON(ret == 0); /* Corruption */
1855 ret = btrfs_previous_item(fs_info->chunk_root, path,
1856 BTRFS_DEV_ITEMS_OBJECTID,
1857 BTRFS_DEV_ITEM_KEY);
1861 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1863 *devid_ret = found_key.offset + 1;
1867 btrfs_free_path(path);
1872 * the device information is stored in the chunk root
1873 * the btrfs_device struct should be fully filled in
1875 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1876 struct btrfs_device *device)
1879 struct btrfs_path *path;
1880 struct btrfs_dev_item *dev_item;
1881 struct extent_buffer *leaf;
1882 struct btrfs_key key;
1885 path = btrfs_alloc_path();
1889 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1890 key.type = BTRFS_DEV_ITEM_KEY;
1891 key.offset = device->devid;
1893 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1894 &key, sizeof(*dev_item));
1898 leaf = path->nodes[0];
1899 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1901 btrfs_set_device_id(leaf, dev_item, device->devid);
1902 btrfs_set_device_generation(leaf, dev_item, 0);
1903 btrfs_set_device_type(leaf, dev_item, device->type);
1904 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1905 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1906 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1907 btrfs_set_device_total_bytes(leaf, dev_item,
1908 btrfs_device_get_disk_total_bytes(device));
1909 btrfs_set_device_bytes_used(leaf, dev_item,
1910 btrfs_device_get_bytes_used(device));
1911 btrfs_set_device_group(leaf, dev_item, 0);
1912 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1913 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1914 btrfs_set_device_start_offset(leaf, dev_item, 0);
1916 ptr = btrfs_device_uuid(dev_item);
1917 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1918 ptr = btrfs_device_fsid(dev_item);
1919 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1920 ptr, BTRFS_FSID_SIZE);
1921 btrfs_mark_buffer_dirty(leaf);
1925 btrfs_free_path(path);
1930 * Function to update ctime/mtime for a given device path.
1931 * Mainly used for ctime/mtime based probe like libblkid.
1933 static void update_dev_time(const char *path_name)
1937 filp = filp_open(path_name, O_RDWR, 0);
1940 file_update_time(filp);
1941 filp_close(filp, NULL);
1944 static int btrfs_rm_dev_item(struct btrfs_device *device)
1946 struct btrfs_root *root = device->fs_info->chunk_root;
1948 struct btrfs_path *path;
1949 struct btrfs_key key;
1950 struct btrfs_trans_handle *trans;
1952 path = btrfs_alloc_path();
1956 trans = btrfs_start_transaction(root, 0);
1957 if (IS_ERR(trans)) {
1958 btrfs_free_path(path);
1959 return PTR_ERR(trans);
1961 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1962 key.type = BTRFS_DEV_ITEM_KEY;
1963 key.offset = device->devid;
1965 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1969 btrfs_abort_transaction(trans, ret);
1970 btrfs_end_transaction(trans);
1974 ret = btrfs_del_item(trans, root, path);
1976 btrfs_abort_transaction(trans, ret);
1977 btrfs_end_transaction(trans);
1981 btrfs_free_path(path);
1983 ret = btrfs_commit_transaction(trans);
1988 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1989 * filesystem. It's up to the caller to adjust that number regarding eg. device
1992 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
2000 seq = read_seqbegin(&fs_info->profiles_lock);
2002 all_avail = fs_info->avail_data_alloc_bits |
2003 fs_info->avail_system_alloc_bits |
2004 fs_info->avail_metadata_alloc_bits;
2005 } while (read_seqretry(&fs_info->profiles_lock, seq));
2007 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2008 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2011 if (num_devices < btrfs_raid_array[i].devs_min) {
2012 int ret = btrfs_raid_array[i].mindev_error;
2022 static struct btrfs_device * btrfs_find_next_active_device(
2023 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2025 struct btrfs_device *next_device;
2027 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2028 if (next_device != device &&
2029 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2030 && next_device->bdev)
2038 * Helper function to check if the given device is part of s_bdev / latest_bdev
2039 * and replace it with the provided or the next active device, in the context
2040 * where this function called, there should be always be another device (or
2041 * this_dev) which is active.
2043 void btrfs_assign_next_active_device(struct btrfs_device *device,
2044 struct btrfs_device *this_dev)
2046 struct btrfs_fs_info *fs_info = device->fs_info;
2047 struct btrfs_device *next_device;
2050 next_device = this_dev;
2052 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2054 ASSERT(next_device);
2056 if (fs_info->sb->s_bdev &&
2057 (fs_info->sb->s_bdev == device->bdev))
2058 fs_info->sb->s_bdev = next_device->bdev;
2060 if (fs_info->fs_devices->latest_bdev == device->bdev)
2061 fs_info->fs_devices->latest_bdev = next_device->bdev;
2065 * Return btrfs_fs_devices::num_devices excluding the device that's being
2066 * currently replaced.
2068 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2070 u64 num_devices = fs_info->fs_devices->num_devices;
2072 down_read(&fs_info->dev_replace.rwsem);
2073 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2074 ASSERT(num_devices > 1);
2077 up_read(&fs_info->dev_replace.rwsem);
2082 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2085 struct btrfs_device *device;
2086 struct btrfs_fs_devices *cur_devices;
2087 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2091 mutex_lock(&uuid_mutex);
2093 num_devices = btrfs_num_devices(fs_info);
2095 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2099 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2101 if (IS_ERR(device)) {
2102 if (PTR_ERR(device) == -ENOENT &&
2103 strcmp(device_path, "missing") == 0)
2104 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2106 ret = PTR_ERR(device);
2110 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2111 btrfs_warn_in_rcu(fs_info,
2112 "cannot remove device %s (devid %llu) due to active swapfile",
2113 rcu_str_deref(device->name), device->devid);
2118 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2119 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2124 fs_info->fs_devices->rw_devices == 1) {
2125 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2129 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2130 mutex_lock(&fs_info->chunk_mutex);
2131 list_del_init(&device->dev_alloc_list);
2132 device->fs_devices->rw_devices--;
2133 mutex_unlock(&fs_info->chunk_mutex);
2136 mutex_unlock(&uuid_mutex);
2137 ret = btrfs_shrink_device(device, 0);
2138 mutex_lock(&uuid_mutex);
2143 * TODO: the superblock still includes this device in its num_devices
2144 * counter although write_all_supers() is not locked out. This
2145 * could give a filesystem state which requires a degraded mount.
2147 ret = btrfs_rm_dev_item(device);
2151 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2152 btrfs_scrub_cancel_dev(device);
2155 * the device list mutex makes sure that we don't change
2156 * the device list while someone else is writing out all
2157 * the device supers. Whoever is writing all supers, should
2158 * lock the device list mutex before getting the number of
2159 * devices in the super block (super_copy). Conversely,
2160 * whoever updates the number of devices in the super block
2161 * (super_copy) should hold the device list mutex.
2165 * In normal cases the cur_devices == fs_devices. But in case
2166 * of deleting a seed device, the cur_devices should point to
2167 * its own fs_devices listed under the fs_devices->seed.
2169 cur_devices = device->fs_devices;
2170 mutex_lock(&fs_devices->device_list_mutex);
2171 list_del_rcu(&device->dev_list);
2173 cur_devices->num_devices--;
2174 cur_devices->total_devices--;
2175 /* Update total_devices of the parent fs_devices if it's seed */
2176 if (cur_devices != fs_devices)
2177 fs_devices->total_devices--;
2179 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2180 cur_devices->missing_devices--;
2182 btrfs_assign_next_active_device(device, NULL);
2185 cur_devices->open_devices--;
2186 /* remove sysfs entry */
2187 btrfs_sysfs_rm_device_link(fs_devices, device);
2190 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2191 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2192 mutex_unlock(&fs_devices->device_list_mutex);
2195 * at this point, the device is zero sized and detached from
2196 * the devices list. All that's left is to zero out the old
2197 * supers and free the device.
2199 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2200 btrfs_scratch_superblocks(device->bdev, device->name->str);
2202 btrfs_close_bdev(device);
2204 btrfs_free_device(device);
2206 if (cur_devices->open_devices == 0) {
2207 while (fs_devices) {
2208 if (fs_devices->seed == cur_devices) {
2209 fs_devices->seed = cur_devices->seed;
2212 fs_devices = fs_devices->seed;
2214 cur_devices->seed = NULL;
2215 close_fs_devices(cur_devices);
2216 free_fs_devices(cur_devices);
2220 mutex_unlock(&uuid_mutex);
2224 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2225 mutex_lock(&fs_info->chunk_mutex);
2226 list_add(&device->dev_alloc_list,
2227 &fs_devices->alloc_list);
2228 device->fs_devices->rw_devices++;
2229 mutex_unlock(&fs_info->chunk_mutex);
2234 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2236 struct btrfs_fs_devices *fs_devices;
2238 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2241 * in case of fs with no seed, srcdev->fs_devices will point
2242 * to fs_devices of fs_info. However when the dev being replaced is
2243 * a seed dev it will point to the seed's local fs_devices. In short
2244 * srcdev will have its correct fs_devices in both the cases.
2246 fs_devices = srcdev->fs_devices;
2248 list_del_rcu(&srcdev->dev_list);
2249 list_del(&srcdev->dev_alloc_list);
2250 fs_devices->num_devices--;
2251 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2252 fs_devices->missing_devices--;
2254 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2255 fs_devices->rw_devices--;
2258 fs_devices->open_devices--;
2261 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2263 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2264 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2266 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2267 /* zero out the old super if it is writable */
2268 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2271 btrfs_close_bdev(srcdev);
2273 btrfs_free_device(srcdev);
2275 /* if this is no devs we rather delete the fs_devices */
2276 if (!fs_devices->num_devices) {
2277 struct btrfs_fs_devices *tmp_fs_devices;
2280 * On a mounted FS, num_devices can't be zero unless it's a
2281 * seed. In case of a seed device being replaced, the replace
2282 * target added to the sprout FS, so there will be no more
2283 * device left under the seed FS.
2285 ASSERT(fs_devices->seeding);
2287 tmp_fs_devices = fs_info->fs_devices;
2288 while (tmp_fs_devices) {
2289 if (tmp_fs_devices->seed == fs_devices) {
2290 tmp_fs_devices->seed = fs_devices->seed;
2293 tmp_fs_devices = tmp_fs_devices->seed;
2295 fs_devices->seed = NULL;
2296 close_fs_devices(fs_devices);
2297 free_fs_devices(fs_devices);
2301 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2303 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2306 mutex_lock(&fs_devices->device_list_mutex);
2308 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2311 fs_devices->open_devices--;
2313 fs_devices->num_devices--;
2315 btrfs_assign_next_active_device(tgtdev, NULL);
2317 list_del_rcu(&tgtdev->dev_list);
2319 mutex_unlock(&fs_devices->device_list_mutex);
2322 * The update_dev_time() with in btrfs_scratch_superblocks()
2323 * may lead to a call to btrfs_show_devname() which will try
2324 * to hold device_list_mutex. And here this device
2325 * is already out of device list, so we don't have to hold
2326 * the device_list_mutex lock.
2328 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2330 btrfs_close_bdev(tgtdev);
2332 btrfs_free_device(tgtdev);
2335 static struct btrfs_device *btrfs_find_device_by_path(
2336 struct btrfs_fs_info *fs_info, const char *device_path)
2339 struct btrfs_super_block *disk_super;
2342 struct block_device *bdev;
2343 struct buffer_head *bh;
2344 struct btrfs_device *device;
2346 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2347 fs_info->bdev_holder, 0, &bdev, &bh);
2349 return ERR_PTR(ret);
2350 disk_super = (struct btrfs_super_block *)bh->b_data;
2351 devid = btrfs_stack_device_id(&disk_super->dev_item);
2352 dev_uuid = disk_super->dev_item.uuid;
2353 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2354 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2355 disk_super->metadata_uuid, true);
2357 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2358 disk_super->fsid, true);
2362 device = ERR_PTR(-ENOENT);
2363 blkdev_put(bdev, FMODE_READ);
2368 * Lookup a device given by device id, or the path if the id is 0.
2370 struct btrfs_device *btrfs_find_device_by_devspec(
2371 struct btrfs_fs_info *fs_info, u64 devid,
2372 const char *device_path)
2374 struct btrfs_device *device;
2377 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2380 return ERR_PTR(-ENOENT);
2384 if (!device_path || !device_path[0])
2385 return ERR_PTR(-EINVAL);
2387 if (strcmp(device_path, "missing") == 0) {
2388 /* Find first missing device */
2389 list_for_each_entry(device, &fs_info->fs_devices->devices,
2391 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2392 &device->dev_state) && !device->bdev)
2395 return ERR_PTR(-ENOENT);
2398 return btrfs_find_device_by_path(fs_info, device_path);
2402 * does all the dirty work required for changing file system's UUID.
2404 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2406 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2407 struct btrfs_fs_devices *old_devices;
2408 struct btrfs_fs_devices *seed_devices;
2409 struct btrfs_super_block *disk_super = fs_info->super_copy;
2410 struct btrfs_device *device;
2413 lockdep_assert_held(&uuid_mutex);
2414 if (!fs_devices->seeding)
2417 seed_devices = alloc_fs_devices(NULL, NULL);
2418 if (IS_ERR(seed_devices))
2419 return PTR_ERR(seed_devices);
2421 old_devices = clone_fs_devices(fs_devices);
2422 if (IS_ERR(old_devices)) {
2423 kfree(seed_devices);
2424 return PTR_ERR(old_devices);
2427 list_add(&old_devices->fs_list, &fs_uuids);
2429 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2430 seed_devices->opened = 1;
2431 INIT_LIST_HEAD(&seed_devices->devices);
2432 INIT_LIST_HEAD(&seed_devices->alloc_list);
2433 mutex_init(&seed_devices->device_list_mutex);
2435 mutex_lock(&fs_devices->device_list_mutex);
2436 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2438 list_for_each_entry(device, &seed_devices->devices, dev_list)
2439 device->fs_devices = seed_devices;
2441 mutex_lock(&fs_info->chunk_mutex);
2442 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2443 mutex_unlock(&fs_info->chunk_mutex);
2445 fs_devices->seeding = 0;
2446 fs_devices->num_devices = 0;
2447 fs_devices->open_devices = 0;
2448 fs_devices->missing_devices = 0;
2449 fs_devices->rotating = 0;
2450 fs_devices->seed = seed_devices;
2452 generate_random_uuid(fs_devices->fsid);
2453 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2454 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2455 mutex_unlock(&fs_devices->device_list_mutex);
2457 super_flags = btrfs_super_flags(disk_super) &
2458 ~BTRFS_SUPER_FLAG_SEEDING;
2459 btrfs_set_super_flags(disk_super, super_flags);
2465 * Store the expected generation for seed devices in device items.
2467 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2469 struct btrfs_fs_info *fs_info = trans->fs_info;
2470 struct btrfs_root *root = fs_info->chunk_root;
2471 struct btrfs_path *path;
2472 struct extent_buffer *leaf;
2473 struct btrfs_dev_item *dev_item;
2474 struct btrfs_device *device;
2475 struct btrfs_key key;
2476 u8 fs_uuid[BTRFS_FSID_SIZE];
2477 u8 dev_uuid[BTRFS_UUID_SIZE];
2481 path = btrfs_alloc_path();
2485 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2487 key.type = BTRFS_DEV_ITEM_KEY;
2490 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2494 leaf = path->nodes[0];
2496 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2497 ret = btrfs_next_leaf(root, path);
2502 leaf = path->nodes[0];
2503 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2504 btrfs_release_path(path);
2508 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2509 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2510 key.type != BTRFS_DEV_ITEM_KEY)
2513 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2514 struct btrfs_dev_item);
2515 devid = btrfs_device_id(leaf, dev_item);
2516 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2518 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2520 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2522 BUG_ON(!device); /* Logic error */
2524 if (device->fs_devices->seeding) {
2525 btrfs_set_device_generation(leaf, dev_item,
2526 device->generation);
2527 btrfs_mark_buffer_dirty(leaf);
2535 btrfs_free_path(path);
2539 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2541 struct btrfs_root *root = fs_info->dev_root;
2542 struct request_queue *q;
2543 struct btrfs_trans_handle *trans;
2544 struct btrfs_device *device;
2545 struct block_device *bdev;
2546 struct super_block *sb = fs_info->sb;
2547 struct rcu_string *name;
2548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2549 u64 orig_super_total_bytes;
2550 u64 orig_super_num_devices;
2551 int seeding_dev = 0;
2553 bool unlocked = false;
2555 if (sb_rdonly(sb) && !fs_devices->seeding)
2558 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2559 fs_info->bdev_holder);
2561 return PTR_ERR(bdev);
2563 if (fs_devices->seeding) {
2565 down_write(&sb->s_umount);
2566 mutex_lock(&uuid_mutex);
2569 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2571 mutex_lock(&fs_devices->device_list_mutex);
2572 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2573 if (device->bdev == bdev) {
2576 &fs_devices->device_list_mutex);
2580 mutex_unlock(&fs_devices->device_list_mutex);
2582 device = btrfs_alloc_device(fs_info, NULL, NULL);
2583 if (IS_ERR(device)) {
2584 /* we can safely leave the fs_devices entry around */
2585 ret = PTR_ERR(device);
2589 name = rcu_string_strdup(device_path, GFP_KERNEL);
2592 goto error_free_device;
2594 rcu_assign_pointer(device->name, name);
2596 trans = btrfs_start_transaction(root, 0);
2597 if (IS_ERR(trans)) {
2598 ret = PTR_ERR(trans);
2599 goto error_free_device;
2602 q = bdev_get_queue(bdev);
2603 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2604 device->generation = trans->transid;
2605 device->io_width = fs_info->sectorsize;
2606 device->io_align = fs_info->sectorsize;
2607 device->sector_size = fs_info->sectorsize;
2608 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2609 fs_info->sectorsize);
2610 device->disk_total_bytes = device->total_bytes;
2611 device->commit_total_bytes = device->total_bytes;
2612 device->fs_info = fs_info;
2613 device->bdev = bdev;
2614 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2615 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2616 device->mode = FMODE_EXCL;
2617 device->dev_stats_valid = 1;
2618 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2621 sb->s_flags &= ~SB_RDONLY;
2622 ret = btrfs_prepare_sprout(fs_info);
2624 btrfs_abort_transaction(trans, ret);
2629 device->fs_devices = fs_devices;
2631 mutex_lock(&fs_devices->device_list_mutex);
2632 mutex_lock(&fs_info->chunk_mutex);
2633 list_add_rcu(&device->dev_list, &fs_devices->devices);
2634 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2635 fs_devices->num_devices++;
2636 fs_devices->open_devices++;
2637 fs_devices->rw_devices++;
2638 fs_devices->total_devices++;
2639 fs_devices->total_rw_bytes += device->total_bytes;
2641 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2643 if (!blk_queue_nonrot(q))
2644 fs_devices->rotating = 1;
2646 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2647 btrfs_set_super_total_bytes(fs_info->super_copy,
2648 round_down(orig_super_total_bytes + device->total_bytes,
2649 fs_info->sectorsize));
2651 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2652 btrfs_set_super_num_devices(fs_info->super_copy,
2653 orig_super_num_devices + 1);
2655 /* add sysfs device entry */
2656 btrfs_sysfs_add_device_link(fs_devices, device);
2659 * we've got more storage, clear any full flags on the space
2662 btrfs_clear_space_info_full(fs_info);
2664 mutex_unlock(&fs_info->chunk_mutex);
2665 mutex_unlock(&fs_devices->device_list_mutex);
2668 mutex_lock(&fs_info->chunk_mutex);
2669 ret = init_first_rw_device(trans);
2670 mutex_unlock(&fs_info->chunk_mutex);
2672 btrfs_abort_transaction(trans, ret);
2677 ret = btrfs_add_dev_item(trans, device);
2679 btrfs_abort_transaction(trans, ret);
2684 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2686 ret = btrfs_finish_sprout(trans);
2688 btrfs_abort_transaction(trans, ret);
2692 /* Sprouting would change fsid of the mounted root,
2693 * so rename the fsid on the sysfs
2695 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2696 fs_info->fs_devices->fsid);
2697 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2699 "sysfs: failed to create fsid for sprout");
2702 ret = btrfs_commit_transaction(trans);
2705 mutex_unlock(&uuid_mutex);
2706 up_write(&sb->s_umount);
2709 if (ret) /* transaction commit */
2712 ret = btrfs_relocate_sys_chunks(fs_info);
2714 btrfs_handle_fs_error(fs_info, ret,
2715 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2716 trans = btrfs_attach_transaction(root);
2717 if (IS_ERR(trans)) {
2718 if (PTR_ERR(trans) == -ENOENT)
2720 ret = PTR_ERR(trans);
2724 ret = btrfs_commit_transaction(trans);
2727 /* Update ctime/mtime for libblkid */
2728 update_dev_time(device_path);
2732 btrfs_sysfs_rm_device_link(fs_devices, device);
2733 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2734 mutex_lock(&fs_info->chunk_mutex);
2735 list_del_rcu(&device->dev_list);
2736 list_del(&device->dev_alloc_list);
2737 fs_info->fs_devices->num_devices--;
2738 fs_info->fs_devices->open_devices--;
2739 fs_info->fs_devices->rw_devices--;
2740 fs_info->fs_devices->total_devices--;
2741 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2742 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2743 btrfs_set_super_total_bytes(fs_info->super_copy,
2744 orig_super_total_bytes);
2745 btrfs_set_super_num_devices(fs_info->super_copy,
2746 orig_super_num_devices);
2747 mutex_unlock(&fs_info->chunk_mutex);
2748 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2751 sb->s_flags |= SB_RDONLY;
2753 btrfs_end_transaction(trans);
2755 btrfs_free_device(device);
2757 blkdev_put(bdev, FMODE_EXCL);
2758 if (seeding_dev && !unlocked) {
2759 mutex_unlock(&uuid_mutex);
2760 up_write(&sb->s_umount);
2765 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2766 struct btrfs_device *device)
2769 struct btrfs_path *path;
2770 struct btrfs_root *root = device->fs_info->chunk_root;
2771 struct btrfs_dev_item *dev_item;
2772 struct extent_buffer *leaf;
2773 struct btrfs_key key;
2775 path = btrfs_alloc_path();
2779 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2780 key.type = BTRFS_DEV_ITEM_KEY;
2781 key.offset = device->devid;
2783 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2792 leaf = path->nodes[0];
2793 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2795 btrfs_set_device_id(leaf, dev_item, device->devid);
2796 btrfs_set_device_type(leaf, dev_item, device->type);
2797 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2798 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2799 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2800 btrfs_set_device_total_bytes(leaf, dev_item,
2801 btrfs_device_get_disk_total_bytes(device));
2802 btrfs_set_device_bytes_used(leaf, dev_item,
2803 btrfs_device_get_bytes_used(device));
2804 btrfs_mark_buffer_dirty(leaf);
2807 btrfs_free_path(path);
2811 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2812 struct btrfs_device *device, u64 new_size)
2814 struct btrfs_fs_info *fs_info = device->fs_info;
2815 struct btrfs_super_block *super_copy = fs_info->super_copy;
2819 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2822 new_size = round_down(new_size, fs_info->sectorsize);
2824 mutex_lock(&fs_info->chunk_mutex);
2825 old_total = btrfs_super_total_bytes(super_copy);
2826 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2828 if (new_size <= device->total_bytes ||
2829 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2830 mutex_unlock(&fs_info->chunk_mutex);
2834 btrfs_set_super_total_bytes(super_copy,
2835 round_down(old_total + diff, fs_info->sectorsize));
2836 device->fs_devices->total_rw_bytes += diff;
2838 btrfs_device_set_total_bytes(device, new_size);
2839 btrfs_device_set_disk_total_bytes(device, new_size);
2840 btrfs_clear_space_info_full(device->fs_info);
2841 if (list_empty(&device->post_commit_list))
2842 list_add_tail(&device->post_commit_list,
2843 &trans->transaction->dev_update_list);
2844 mutex_unlock(&fs_info->chunk_mutex);
2846 return btrfs_update_device(trans, device);
2849 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2851 struct btrfs_fs_info *fs_info = trans->fs_info;
2852 struct btrfs_root *root = fs_info->chunk_root;
2854 struct btrfs_path *path;
2855 struct btrfs_key key;
2857 path = btrfs_alloc_path();
2861 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2862 key.offset = chunk_offset;
2863 key.type = BTRFS_CHUNK_ITEM_KEY;
2865 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2868 else if (ret > 0) { /* Logic error or corruption */
2869 btrfs_handle_fs_error(fs_info, -ENOENT,
2870 "Failed lookup while freeing chunk.");
2875 ret = btrfs_del_item(trans, root, path);
2877 btrfs_handle_fs_error(fs_info, ret,
2878 "Failed to delete chunk item.");
2880 btrfs_free_path(path);
2884 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2886 struct btrfs_super_block *super_copy = fs_info->super_copy;
2887 struct btrfs_disk_key *disk_key;
2888 struct btrfs_chunk *chunk;
2895 struct btrfs_key key;
2897 mutex_lock(&fs_info->chunk_mutex);
2898 array_size = btrfs_super_sys_array_size(super_copy);
2900 ptr = super_copy->sys_chunk_array;
2903 while (cur < array_size) {
2904 disk_key = (struct btrfs_disk_key *)ptr;
2905 btrfs_disk_key_to_cpu(&key, disk_key);
2907 len = sizeof(*disk_key);
2909 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2910 chunk = (struct btrfs_chunk *)(ptr + len);
2911 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2912 len += btrfs_chunk_item_size(num_stripes);
2917 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2918 key.offset == chunk_offset) {
2919 memmove(ptr, ptr + len, array_size - (cur + len));
2921 btrfs_set_super_sys_array_size(super_copy, array_size);
2927 mutex_unlock(&fs_info->chunk_mutex);
2932 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2933 * @logical: Logical block offset in bytes.
2934 * @length: Length of extent in bytes.
2936 * Return: Chunk mapping or ERR_PTR.
2938 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2939 u64 logical, u64 length)
2941 struct extent_map_tree *em_tree;
2942 struct extent_map *em;
2944 em_tree = &fs_info->mapping_tree.map_tree;
2945 read_lock(&em_tree->lock);
2946 em = lookup_extent_mapping(em_tree, logical, length);
2947 read_unlock(&em_tree->lock);
2950 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2952 return ERR_PTR(-EINVAL);
2955 if (em->start > logical || em->start + em->len < logical) {
2957 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2958 logical, length, em->start, em->start + em->len);
2959 free_extent_map(em);
2960 return ERR_PTR(-EINVAL);
2963 /* callers are responsible for dropping em's ref. */
2967 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2969 struct btrfs_fs_info *fs_info = trans->fs_info;
2970 struct extent_map *em;
2971 struct map_lookup *map;
2972 u64 dev_extent_len = 0;
2974 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2976 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2979 * This is a logic error, but we don't want to just rely on the
2980 * user having built with ASSERT enabled, so if ASSERT doesn't
2981 * do anything we still error out.
2986 map = em->map_lookup;
2987 mutex_lock(&fs_info->chunk_mutex);
2988 check_system_chunk(trans, map->type);
2989 mutex_unlock(&fs_info->chunk_mutex);
2992 * Take the device list mutex to prevent races with the final phase of
2993 * a device replace operation that replaces the device object associated
2994 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2996 mutex_lock(&fs_devices->device_list_mutex);
2997 for (i = 0; i < map->num_stripes; i++) {
2998 struct btrfs_device *device = map->stripes[i].dev;
2999 ret = btrfs_free_dev_extent(trans, device,
3000 map->stripes[i].physical,
3003 mutex_unlock(&fs_devices->device_list_mutex);
3004 btrfs_abort_transaction(trans, ret);
3008 if (device->bytes_used > 0) {
3009 mutex_lock(&fs_info->chunk_mutex);
3010 btrfs_device_set_bytes_used(device,
3011 device->bytes_used - dev_extent_len);
3012 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3013 btrfs_clear_space_info_full(fs_info);
3014 mutex_unlock(&fs_info->chunk_mutex);
3017 ret = btrfs_update_device(trans, device);
3019 mutex_unlock(&fs_devices->device_list_mutex);
3020 btrfs_abort_transaction(trans, ret);
3024 mutex_unlock(&fs_devices->device_list_mutex);
3026 ret = btrfs_free_chunk(trans, chunk_offset);
3028 btrfs_abort_transaction(trans, ret);
3032 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3034 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3035 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3037 btrfs_abort_transaction(trans, ret);
3042 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3044 btrfs_abort_transaction(trans, ret);
3050 free_extent_map(em);
3054 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3056 struct btrfs_root *root = fs_info->chunk_root;
3057 struct btrfs_trans_handle *trans;
3061 * Prevent races with automatic removal of unused block groups.
3062 * After we relocate and before we remove the chunk with offset
3063 * chunk_offset, automatic removal of the block group can kick in,
3064 * resulting in a failure when calling btrfs_remove_chunk() below.
3066 * Make sure to acquire this mutex before doing a tree search (dev
3067 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3068 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3069 * we release the path used to search the chunk/dev tree and before
3070 * the current task acquires this mutex and calls us.
3072 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3074 ret = btrfs_can_relocate(fs_info, chunk_offset);
3078 /* step one, relocate all the extents inside this chunk */
3079 btrfs_scrub_pause(fs_info);
3080 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3081 btrfs_scrub_continue(fs_info);
3086 * We add the kobjects here (and after forcing data chunk creation)
3087 * since relocation is the only place we'll create chunks of a new
3088 * type at runtime. The only place where we'll remove the last
3089 * chunk of a type is the call immediately below this one. Even
3090 * so, we're protected against races with the cleaner thread since
3091 * we're covered by the delete_unused_bgs_mutex.
3093 btrfs_add_raid_kobjects(fs_info);
3095 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3097 if (IS_ERR(trans)) {
3098 ret = PTR_ERR(trans);
3099 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3104 * step two, delete the device extents and the
3105 * chunk tree entries
3107 ret = btrfs_remove_chunk(trans, chunk_offset);
3108 btrfs_end_transaction(trans);
3112 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3114 struct btrfs_root *chunk_root = fs_info->chunk_root;
3115 struct btrfs_path *path;
3116 struct extent_buffer *leaf;
3117 struct btrfs_chunk *chunk;
3118 struct btrfs_key key;
3119 struct btrfs_key found_key;
3121 bool retried = false;
3125 path = btrfs_alloc_path();
3130 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3131 key.offset = (u64)-1;
3132 key.type = BTRFS_CHUNK_ITEM_KEY;
3135 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3136 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3138 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3141 BUG_ON(ret == 0); /* Corruption */
3143 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3146 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3152 leaf = path->nodes[0];
3153 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3155 chunk = btrfs_item_ptr(leaf, path->slots[0],
3156 struct btrfs_chunk);
3157 chunk_type = btrfs_chunk_type(leaf, chunk);
3158 btrfs_release_path(path);
3160 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3161 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3167 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3169 if (found_key.offset == 0)
3171 key.offset = found_key.offset - 1;
3174 if (failed && !retried) {
3178 } else if (WARN_ON(failed && retried)) {
3182 btrfs_free_path(path);
3187 * return 1 : allocate a data chunk successfully,
3188 * return <0: errors during allocating a data chunk,
3189 * return 0 : no need to allocate a data chunk.
3191 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3194 struct btrfs_block_group_cache *cache;
3198 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3200 chunk_type = cache->flags;
3201 btrfs_put_block_group(cache);
3203 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3204 spin_lock(&fs_info->data_sinfo->lock);
3205 bytes_used = fs_info->data_sinfo->bytes_used;
3206 spin_unlock(&fs_info->data_sinfo->lock);
3209 struct btrfs_trans_handle *trans;
3212 trans = btrfs_join_transaction(fs_info->tree_root);
3214 return PTR_ERR(trans);
3216 ret = btrfs_force_chunk_alloc(trans,
3217 BTRFS_BLOCK_GROUP_DATA);
3218 btrfs_end_transaction(trans);
3222 btrfs_add_raid_kobjects(fs_info);
3230 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3231 struct btrfs_balance_control *bctl)
3233 struct btrfs_root *root = fs_info->tree_root;
3234 struct btrfs_trans_handle *trans;
3235 struct btrfs_balance_item *item;
3236 struct btrfs_disk_balance_args disk_bargs;
3237 struct btrfs_path *path;
3238 struct extent_buffer *leaf;
3239 struct btrfs_key key;
3242 path = btrfs_alloc_path();
3246 trans = btrfs_start_transaction(root, 0);
3247 if (IS_ERR(trans)) {
3248 btrfs_free_path(path);
3249 return PTR_ERR(trans);
3252 key.objectid = BTRFS_BALANCE_OBJECTID;
3253 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3256 ret = btrfs_insert_empty_item(trans, root, path, &key,
3261 leaf = path->nodes[0];
3262 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3264 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3266 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3267 btrfs_set_balance_data(leaf, item, &disk_bargs);
3268 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3269 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3270 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3271 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3273 btrfs_set_balance_flags(leaf, item, bctl->flags);
3275 btrfs_mark_buffer_dirty(leaf);
3277 btrfs_free_path(path);
3278 err = btrfs_commit_transaction(trans);
3284 static int del_balance_item(struct btrfs_fs_info *fs_info)
3286 struct btrfs_root *root = fs_info->tree_root;
3287 struct btrfs_trans_handle *trans;
3288 struct btrfs_path *path;
3289 struct btrfs_key key;
3292 path = btrfs_alloc_path();
3296 trans = btrfs_start_transaction(root, 0);
3297 if (IS_ERR(trans)) {
3298 btrfs_free_path(path);
3299 return PTR_ERR(trans);
3302 key.objectid = BTRFS_BALANCE_OBJECTID;
3303 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3306 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3314 ret = btrfs_del_item(trans, root, path);
3316 btrfs_free_path(path);
3317 err = btrfs_commit_transaction(trans);
3324 * This is a heuristic used to reduce the number of chunks balanced on
3325 * resume after balance was interrupted.
3327 static void update_balance_args(struct btrfs_balance_control *bctl)
3330 * Turn on soft mode for chunk types that were being converted.
3332 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3333 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3334 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3335 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3336 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3337 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3340 * Turn on usage filter if is not already used. The idea is
3341 * that chunks that we have already balanced should be
3342 * reasonably full. Don't do it for chunks that are being
3343 * converted - that will keep us from relocating unconverted
3344 * (albeit full) chunks.
3346 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3347 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3348 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3349 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3350 bctl->data.usage = 90;
3352 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3353 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3354 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3355 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3356 bctl->sys.usage = 90;
3358 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3359 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3360 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3361 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3362 bctl->meta.usage = 90;
3367 * Clear the balance status in fs_info and delete the balance item from disk.
3369 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3371 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3374 BUG_ON(!fs_info->balance_ctl);
3376 spin_lock(&fs_info->balance_lock);
3377 fs_info->balance_ctl = NULL;
3378 spin_unlock(&fs_info->balance_lock);
3381 ret = del_balance_item(fs_info);
3383 btrfs_handle_fs_error(fs_info, ret, NULL);
3387 * Balance filters. Return 1 if chunk should be filtered out
3388 * (should not be balanced).
3390 static int chunk_profiles_filter(u64 chunk_type,
3391 struct btrfs_balance_args *bargs)
3393 chunk_type = chunk_to_extended(chunk_type) &
3394 BTRFS_EXTENDED_PROFILE_MASK;
3396 if (bargs->profiles & chunk_type)
3402 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3403 struct btrfs_balance_args *bargs)
3405 struct btrfs_block_group_cache *cache;
3407 u64 user_thresh_min;
3408 u64 user_thresh_max;
3411 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3412 chunk_used = btrfs_block_group_used(&cache->item);
3414 if (bargs->usage_min == 0)
3415 user_thresh_min = 0;
3417 user_thresh_min = div_factor_fine(cache->key.offset,
3420 if (bargs->usage_max == 0)
3421 user_thresh_max = 1;
3422 else if (bargs->usage_max > 100)
3423 user_thresh_max = cache->key.offset;
3425 user_thresh_max = div_factor_fine(cache->key.offset,
3428 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3431 btrfs_put_block_group(cache);
3435 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3436 u64 chunk_offset, struct btrfs_balance_args *bargs)
3438 struct btrfs_block_group_cache *cache;
3439 u64 chunk_used, user_thresh;
3442 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3443 chunk_used = btrfs_block_group_used(&cache->item);
3445 if (bargs->usage_min == 0)
3447 else if (bargs->usage > 100)
3448 user_thresh = cache->key.offset;
3450 user_thresh = div_factor_fine(cache->key.offset,
3453 if (chunk_used < user_thresh)
3456 btrfs_put_block_group(cache);
3460 static int chunk_devid_filter(struct extent_buffer *leaf,
3461 struct btrfs_chunk *chunk,
3462 struct btrfs_balance_args *bargs)
3464 struct btrfs_stripe *stripe;
3465 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3468 for (i = 0; i < num_stripes; i++) {
3469 stripe = btrfs_stripe_nr(chunk, i);
3470 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3477 /* [pstart, pend) */
3478 static int chunk_drange_filter(struct extent_buffer *leaf,
3479 struct btrfs_chunk *chunk,
3480 struct btrfs_balance_args *bargs)
3482 struct btrfs_stripe *stripe;
3483 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3489 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3492 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3493 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3494 factor = num_stripes / 2;
3495 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3496 factor = num_stripes - 1;
3497 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3498 factor = num_stripes - 2;
3500 factor = num_stripes;
3503 for (i = 0; i < num_stripes; i++) {
3504 stripe = btrfs_stripe_nr(chunk, i);
3505 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3508 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3509 stripe_length = btrfs_chunk_length(leaf, chunk);
3510 stripe_length = div_u64(stripe_length, factor);
3512 if (stripe_offset < bargs->pend &&
3513 stripe_offset + stripe_length > bargs->pstart)
3520 /* [vstart, vend) */
3521 static int chunk_vrange_filter(struct extent_buffer *leaf,
3522 struct btrfs_chunk *chunk,
3524 struct btrfs_balance_args *bargs)
3526 if (chunk_offset < bargs->vend &&
3527 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3528 /* at least part of the chunk is inside this vrange */
3534 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3535 struct btrfs_chunk *chunk,
3536 struct btrfs_balance_args *bargs)
3538 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3540 if (bargs->stripes_min <= num_stripes
3541 && num_stripes <= bargs->stripes_max)
3547 static int chunk_soft_convert_filter(u64 chunk_type,
3548 struct btrfs_balance_args *bargs)
3550 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3553 chunk_type = chunk_to_extended(chunk_type) &
3554 BTRFS_EXTENDED_PROFILE_MASK;
3556 if (bargs->target == chunk_type)
3562 static int should_balance_chunk(struct extent_buffer *leaf,
3563 struct btrfs_chunk *chunk, u64 chunk_offset)
3565 struct btrfs_fs_info *fs_info = leaf->fs_info;
3566 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3567 struct btrfs_balance_args *bargs = NULL;
3568 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3571 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3572 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3576 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3577 bargs = &bctl->data;
3578 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3580 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3581 bargs = &bctl->meta;
3583 /* profiles filter */
3584 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3585 chunk_profiles_filter(chunk_type, bargs)) {
3590 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3591 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3593 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3594 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3599 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3600 chunk_devid_filter(leaf, chunk, bargs)) {
3604 /* drange filter, makes sense only with devid filter */
3605 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3606 chunk_drange_filter(leaf, chunk, bargs)) {
3611 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3612 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3616 /* stripes filter */
3617 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3618 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3622 /* soft profile changing mode */
3623 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3624 chunk_soft_convert_filter(chunk_type, bargs)) {
3629 * limited by count, must be the last filter
3631 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3632 if (bargs->limit == 0)
3636 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3638 * Same logic as the 'limit' filter; the minimum cannot be
3639 * determined here because we do not have the global information
3640 * about the count of all chunks that satisfy the filters.
3642 if (bargs->limit_max == 0)
3651 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3653 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3654 struct btrfs_root *chunk_root = fs_info->chunk_root;
3656 struct btrfs_chunk *chunk;
3657 struct btrfs_path *path = NULL;
3658 struct btrfs_key key;
3659 struct btrfs_key found_key;
3660 struct extent_buffer *leaf;
3663 int enospc_errors = 0;
3664 bool counting = true;
3665 /* The single value limit and min/max limits use the same bytes in the */
3666 u64 limit_data = bctl->data.limit;
3667 u64 limit_meta = bctl->meta.limit;
3668 u64 limit_sys = bctl->sys.limit;
3672 int chunk_reserved = 0;
3674 path = btrfs_alloc_path();
3680 /* zero out stat counters */
3681 spin_lock(&fs_info->balance_lock);
3682 memset(&bctl->stat, 0, sizeof(bctl->stat));
3683 spin_unlock(&fs_info->balance_lock);
3687 * The single value limit and min/max limits use the same bytes
3690 bctl->data.limit = limit_data;
3691 bctl->meta.limit = limit_meta;
3692 bctl->sys.limit = limit_sys;
3694 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3695 key.offset = (u64)-1;
3696 key.type = BTRFS_CHUNK_ITEM_KEY;
3699 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3700 atomic_read(&fs_info->balance_cancel_req)) {
3705 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3706 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3708 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3713 * this shouldn't happen, it means the last relocate
3717 BUG(); /* FIXME break ? */
3719 ret = btrfs_previous_item(chunk_root, path, 0,
3720 BTRFS_CHUNK_ITEM_KEY);
3722 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3727 leaf = path->nodes[0];
3728 slot = path->slots[0];
3729 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3731 if (found_key.objectid != key.objectid) {
3732 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3736 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3737 chunk_type = btrfs_chunk_type(leaf, chunk);
3740 spin_lock(&fs_info->balance_lock);
3741 bctl->stat.considered++;
3742 spin_unlock(&fs_info->balance_lock);
3745 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3747 btrfs_release_path(path);
3749 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3754 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3755 spin_lock(&fs_info->balance_lock);
3756 bctl->stat.expected++;
3757 spin_unlock(&fs_info->balance_lock);
3759 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3761 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3763 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3770 * Apply limit_min filter, no need to check if the LIMITS
3771 * filter is used, limit_min is 0 by default
3773 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3774 count_data < bctl->data.limit_min)
3775 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3776 count_meta < bctl->meta.limit_min)
3777 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3778 count_sys < bctl->sys.limit_min)) {
3779 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3783 if (!chunk_reserved) {
3785 * We may be relocating the only data chunk we have,
3786 * which could potentially end up with losing data's
3787 * raid profile, so lets allocate an empty one in
3790 ret = btrfs_may_alloc_data_chunk(fs_info,
3793 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3795 } else if (ret == 1) {
3800 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3801 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3802 if (ret == -ENOSPC) {
3804 } else if (ret == -ETXTBSY) {
3806 "skipping relocation of block group %llu due to active swapfile",
3812 spin_lock(&fs_info->balance_lock);
3813 bctl->stat.completed++;
3814 spin_unlock(&fs_info->balance_lock);
3817 if (found_key.offset == 0)
3819 key.offset = found_key.offset - 1;
3823 btrfs_release_path(path);
3828 btrfs_free_path(path);
3829 if (enospc_errors) {
3830 btrfs_info(fs_info, "%d enospc errors during balance",
3840 * alloc_profile_is_valid - see if a given profile is valid and reduced
3841 * @flags: profile to validate
3842 * @extended: if true @flags is treated as an extended profile
3844 static int alloc_profile_is_valid(u64 flags, int extended)
3846 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3847 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3849 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3851 /* 1) check that all other bits are zeroed */
3855 /* 2) see if profile is reduced */
3857 return !extended; /* "0" is valid for usual profiles */
3859 /* true if exactly one bit set */
3860 return is_power_of_2(flags);
3863 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3865 /* cancel requested || normal exit path */
3866 return atomic_read(&fs_info->balance_cancel_req) ||
3867 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3868 atomic_read(&fs_info->balance_cancel_req) == 0);
3871 /* Non-zero return value signifies invalidity */
3872 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3875 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3876 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3877 (bctl_arg->target & ~allowed)));
3881 * Fill @buf with textual description of balance filter flags @bargs, up to
3882 * @size_buf including the terminating null. The output may be trimmed if it
3883 * does not fit into the provided buffer.
3885 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3889 u32 size_bp = size_buf;
3891 u64 flags = bargs->flags;
3892 char tmp_buf[128] = {'\0'};
3897 #define CHECK_APPEND_NOARG(a) \
3899 ret = snprintf(bp, size_bp, (a)); \
3900 if (ret < 0 || ret >= size_bp) \
3901 goto out_overflow; \
3906 #define CHECK_APPEND_1ARG(a, v1) \
3908 ret = snprintf(bp, size_bp, (a), (v1)); \
3909 if (ret < 0 || ret >= size_bp) \
3910 goto out_overflow; \
3915 #define CHECK_APPEND_2ARG(a, v1, v2) \
3917 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3918 if (ret < 0 || ret >= size_bp) \
3919 goto out_overflow; \
3924 if (flags & BTRFS_BALANCE_ARGS_CONVERT) {
3925 int index = btrfs_bg_flags_to_raid_index(bargs->target);
3927 CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index));
3930 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3931 CHECK_APPEND_NOARG("soft,");
3933 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3934 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3936 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3939 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3940 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3942 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3943 CHECK_APPEND_2ARG("usage=%u..%u,",
3944 bargs->usage_min, bargs->usage_max);
3946 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3947 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3949 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3950 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3951 bargs->pstart, bargs->pend);
3953 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3954 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3955 bargs->vstart, bargs->vend);
3957 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3958 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3960 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3961 CHECK_APPEND_2ARG("limit=%u..%u,",
3962 bargs->limit_min, bargs->limit_max);
3964 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3965 CHECK_APPEND_2ARG("stripes=%u..%u,",
3966 bargs->stripes_min, bargs->stripes_max);
3968 #undef CHECK_APPEND_2ARG
3969 #undef CHECK_APPEND_1ARG
3970 #undef CHECK_APPEND_NOARG
3974 if (size_bp < size_buf)
3975 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3980 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3982 u32 size_buf = 1024;
3983 char tmp_buf[192] = {'\0'};
3986 u32 size_bp = size_buf;
3988 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3990 buf = kzalloc(size_buf, GFP_KERNEL);
3996 #define CHECK_APPEND_1ARG(a, v1) \
3998 ret = snprintf(bp, size_bp, (a), (v1)); \
3999 if (ret < 0 || ret >= size_bp) \
4000 goto out_overflow; \
4005 if (bctl->flags & BTRFS_BALANCE_FORCE)
4006 CHECK_APPEND_1ARG("%s", "-f ");
4008 if (bctl->flags & BTRFS_BALANCE_DATA) {
4009 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4010 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4013 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4014 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4015 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4018 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4019 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4020 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4023 #undef CHECK_APPEND_1ARG
4027 if (size_bp < size_buf)
4028 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4029 btrfs_info(fs_info, "balance: %s %s",
4030 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4031 "resume" : "start", buf);
4037 * Should be called with balance mutexe held
4039 int btrfs_balance(struct btrfs_fs_info *fs_info,
4040 struct btrfs_balance_control *bctl,
4041 struct btrfs_ioctl_balance_args *bargs)
4043 u64 meta_target, data_target;
4049 bool reducing_integrity;
4051 if (btrfs_fs_closing(fs_info) ||
4052 atomic_read(&fs_info->balance_pause_req) ||
4053 atomic_read(&fs_info->balance_cancel_req)) {
4058 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4059 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4063 * In case of mixed groups both data and meta should be picked,
4064 * and identical options should be given for both of them.
4066 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4067 if (mixed && (bctl->flags & allowed)) {
4068 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4069 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4070 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4072 "balance: mixed groups data and metadata options must be the same");
4078 num_devices = btrfs_num_devices(fs_info);
4080 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
4081 if (num_devices > 1)
4082 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
4083 if (num_devices > 2)
4084 allowed |= BTRFS_BLOCK_GROUP_RAID5;
4085 if (num_devices > 3)
4086 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
4087 BTRFS_BLOCK_GROUP_RAID6);
4088 if (validate_convert_profile(&bctl->data, allowed)) {
4089 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
4092 "balance: invalid convert data profile %s",
4093 get_raid_name(index));
4097 if (validate_convert_profile(&bctl->meta, allowed)) {
4098 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
4101 "balance: invalid convert metadata profile %s",
4102 get_raid_name(index));
4106 if (validate_convert_profile(&bctl->sys, allowed)) {
4107 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
4110 "balance: invalid convert system profile %s",
4111 get_raid_name(index));
4116 /* allow to reduce meta or sys integrity only if force set */
4117 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4118 BTRFS_BLOCK_GROUP_RAID10 |
4119 BTRFS_BLOCK_GROUP_RAID5 |
4120 BTRFS_BLOCK_GROUP_RAID6;
4122 seq = read_seqbegin(&fs_info->profiles_lock);
4124 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4125 (fs_info->avail_system_alloc_bits & allowed) &&
4126 !(bctl->sys.target & allowed)) ||
4127 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4128 (fs_info->avail_metadata_alloc_bits & allowed) &&
4129 !(bctl->meta.target & allowed)))
4130 reducing_integrity = true;
4132 reducing_integrity = false;
4134 /* if we're not converting, the target field is uninitialized */
4135 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4136 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4137 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4138 bctl->data.target : fs_info->avail_data_alloc_bits;
4139 } while (read_seqretry(&fs_info->profiles_lock, seq));
4141 if (reducing_integrity) {
4142 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4144 "balance: force reducing metadata integrity");
4147 "balance: reduces metadata integrity, use --force if you want this");
4153 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4154 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4155 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
4156 int data_index = btrfs_bg_flags_to_raid_index(data_target);
4159 "balance: metadata profile %s has lower redundancy than data profile %s",
4160 get_raid_name(meta_index), get_raid_name(data_index));
4163 ret = insert_balance_item(fs_info, bctl);
4164 if (ret && ret != -EEXIST)
4167 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4168 BUG_ON(ret == -EEXIST);
4169 BUG_ON(fs_info->balance_ctl);
4170 spin_lock(&fs_info->balance_lock);
4171 fs_info->balance_ctl = bctl;
4172 spin_unlock(&fs_info->balance_lock);
4174 BUG_ON(ret != -EEXIST);
4175 spin_lock(&fs_info->balance_lock);
4176 update_balance_args(bctl);
4177 spin_unlock(&fs_info->balance_lock);
4180 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4181 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4182 describe_balance_start_or_resume(fs_info);
4183 mutex_unlock(&fs_info->balance_mutex);
4185 ret = __btrfs_balance(fs_info);
4187 mutex_lock(&fs_info->balance_mutex);
4188 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4189 btrfs_info(fs_info, "balance: paused");
4190 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4191 btrfs_info(fs_info, "balance: canceled");
4193 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4195 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4198 memset(bargs, 0, sizeof(*bargs));
4199 btrfs_update_ioctl_balance_args(fs_info, bargs);
4202 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4203 balance_need_close(fs_info)) {
4204 reset_balance_state(fs_info);
4205 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4208 wake_up(&fs_info->balance_wait_q);
4212 if (bctl->flags & BTRFS_BALANCE_RESUME)
4213 reset_balance_state(fs_info);
4216 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4221 static int balance_kthread(void *data)
4223 struct btrfs_fs_info *fs_info = data;
4226 mutex_lock(&fs_info->balance_mutex);
4227 if (fs_info->balance_ctl)
4228 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4229 mutex_unlock(&fs_info->balance_mutex);
4234 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4236 struct task_struct *tsk;
4238 mutex_lock(&fs_info->balance_mutex);
4239 if (!fs_info->balance_ctl) {
4240 mutex_unlock(&fs_info->balance_mutex);
4243 mutex_unlock(&fs_info->balance_mutex);
4245 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4246 btrfs_info(fs_info, "balance: resume skipped");
4251 * A ro->rw remount sequence should continue with the paused balance
4252 * regardless of who pauses it, system or the user as of now, so set
4255 spin_lock(&fs_info->balance_lock);
4256 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4257 spin_unlock(&fs_info->balance_lock);
4259 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4260 return PTR_ERR_OR_ZERO(tsk);
4263 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4265 struct btrfs_balance_control *bctl;
4266 struct btrfs_balance_item *item;
4267 struct btrfs_disk_balance_args disk_bargs;
4268 struct btrfs_path *path;
4269 struct extent_buffer *leaf;
4270 struct btrfs_key key;
4273 path = btrfs_alloc_path();
4277 key.objectid = BTRFS_BALANCE_OBJECTID;
4278 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4281 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4284 if (ret > 0) { /* ret = -ENOENT; */
4289 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4295 leaf = path->nodes[0];
4296 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4298 bctl->flags = btrfs_balance_flags(leaf, item);
4299 bctl->flags |= BTRFS_BALANCE_RESUME;
4301 btrfs_balance_data(leaf, item, &disk_bargs);
4302 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4303 btrfs_balance_meta(leaf, item, &disk_bargs);
4304 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4305 btrfs_balance_sys(leaf, item, &disk_bargs);
4306 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4309 * This should never happen, as the paused balance state is recovered
4310 * during mount without any chance of other exclusive ops to collide.
4312 * This gives the exclusive op status to balance and keeps in paused
4313 * state until user intervention (cancel or umount). If the ownership
4314 * cannot be assigned, show a message but do not fail. The balance
4315 * is in a paused state and must have fs_info::balance_ctl properly
4318 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4320 "balance: cannot set exclusive op status, resume manually");
4322 mutex_lock(&fs_info->balance_mutex);
4323 BUG_ON(fs_info->balance_ctl);
4324 spin_lock(&fs_info->balance_lock);
4325 fs_info->balance_ctl = bctl;
4326 spin_unlock(&fs_info->balance_lock);
4327 mutex_unlock(&fs_info->balance_mutex);
4329 btrfs_free_path(path);
4333 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4337 mutex_lock(&fs_info->balance_mutex);
4338 if (!fs_info->balance_ctl) {
4339 mutex_unlock(&fs_info->balance_mutex);
4343 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4344 atomic_inc(&fs_info->balance_pause_req);
4345 mutex_unlock(&fs_info->balance_mutex);
4347 wait_event(fs_info->balance_wait_q,
4348 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4350 mutex_lock(&fs_info->balance_mutex);
4351 /* we are good with balance_ctl ripped off from under us */
4352 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4353 atomic_dec(&fs_info->balance_pause_req);
4358 mutex_unlock(&fs_info->balance_mutex);
4362 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4364 mutex_lock(&fs_info->balance_mutex);
4365 if (!fs_info->balance_ctl) {
4366 mutex_unlock(&fs_info->balance_mutex);
4371 * A paused balance with the item stored on disk can be resumed at
4372 * mount time if the mount is read-write. Otherwise it's still paused
4373 * and we must not allow cancelling as it deletes the item.
4375 if (sb_rdonly(fs_info->sb)) {
4376 mutex_unlock(&fs_info->balance_mutex);
4380 atomic_inc(&fs_info->balance_cancel_req);
4382 * if we are running just wait and return, balance item is
4383 * deleted in btrfs_balance in this case
4385 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4386 mutex_unlock(&fs_info->balance_mutex);
4387 wait_event(fs_info->balance_wait_q,
4388 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4389 mutex_lock(&fs_info->balance_mutex);
4391 mutex_unlock(&fs_info->balance_mutex);
4393 * Lock released to allow other waiters to continue, we'll
4394 * reexamine the status again.
4396 mutex_lock(&fs_info->balance_mutex);
4398 if (fs_info->balance_ctl) {
4399 reset_balance_state(fs_info);
4400 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4401 btrfs_info(fs_info, "balance: canceled");
4405 BUG_ON(fs_info->balance_ctl ||
4406 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4407 atomic_dec(&fs_info->balance_cancel_req);
4408 mutex_unlock(&fs_info->balance_mutex);
4412 static int btrfs_uuid_scan_kthread(void *data)
4414 struct btrfs_fs_info *fs_info = data;
4415 struct btrfs_root *root = fs_info->tree_root;
4416 struct btrfs_key key;
4417 struct btrfs_path *path = NULL;
4419 struct extent_buffer *eb;
4421 struct btrfs_root_item root_item;
4423 struct btrfs_trans_handle *trans = NULL;
4425 path = btrfs_alloc_path();
4432 key.type = BTRFS_ROOT_ITEM_KEY;
4436 ret = btrfs_search_forward(root, &key, path,
4437 BTRFS_OLDEST_GENERATION);
4444 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4445 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4446 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4447 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4450 eb = path->nodes[0];
4451 slot = path->slots[0];
4452 item_size = btrfs_item_size_nr(eb, slot);
4453 if (item_size < sizeof(root_item))
4456 read_extent_buffer(eb, &root_item,
4457 btrfs_item_ptr_offset(eb, slot),
4458 (int)sizeof(root_item));
4459 if (btrfs_root_refs(&root_item) == 0)
4462 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4463 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4467 btrfs_release_path(path);
4469 * 1 - subvol uuid item
4470 * 1 - received_subvol uuid item
4472 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4473 if (IS_ERR(trans)) {
4474 ret = PTR_ERR(trans);
4482 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4483 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4484 BTRFS_UUID_KEY_SUBVOL,
4487 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4493 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4494 ret = btrfs_uuid_tree_add(trans,
4495 root_item.received_uuid,
4496 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4499 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4507 ret = btrfs_end_transaction(trans);
4513 btrfs_release_path(path);
4514 if (key.offset < (u64)-1) {
4516 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4518 key.type = BTRFS_ROOT_ITEM_KEY;
4519 } else if (key.objectid < (u64)-1) {
4521 key.type = BTRFS_ROOT_ITEM_KEY;
4530 btrfs_free_path(path);
4531 if (trans && !IS_ERR(trans))
4532 btrfs_end_transaction(trans);
4534 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4536 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4537 up(&fs_info->uuid_tree_rescan_sem);
4542 * Callback for btrfs_uuid_tree_iterate().
4544 * 0 check succeeded, the entry is not outdated.
4545 * < 0 if an error occurred.
4546 * > 0 if the check failed, which means the caller shall remove the entry.
4548 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4549 u8 *uuid, u8 type, u64 subid)
4551 struct btrfs_key key;
4553 struct btrfs_root *subvol_root;
4555 if (type != BTRFS_UUID_KEY_SUBVOL &&
4556 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4559 key.objectid = subid;
4560 key.type = BTRFS_ROOT_ITEM_KEY;
4561 key.offset = (u64)-1;
4562 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4563 if (IS_ERR(subvol_root)) {
4564 ret = PTR_ERR(subvol_root);
4571 case BTRFS_UUID_KEY_SUBVOL:
4572 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4575 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4576 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4586 static int btrfs_uuid_rescan_kthread(void *data)
4588 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4592 * 1st step is to iterate through the existing UUID tree and
4593 * to delete all entries that contain outdated data.
4594 * 2nd step is to add all missing entries to the UUID tree.
4596 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4598 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4599 up(&fs_info->uuid_tree_rescan_sem);
4602 return btrfs_uuid_scan_kthread(data);
4605 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4607 struct btrfs_trans_handle *trans;
4608 struct btrfs_root *tree_root = fs_info->tree_root;
4609 struct btrfs_root *uuid_root;
4610 struct task_struct *task;
4617 trans = btrfs_start_transaction(tree_root, 2);
4619 return PTR_ERR(trans);
4621 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4622 if (IS_ERR(uuid_root)) {
4623 ret = PTR_ERR(uuid_root);
4624 btrfs_abort_transaction(trans, ret);
4625 btrfs_end_transaction(trans);
4629 fs_info->uuid_root = uuid_root;
4631 ret = btrfs_commit_transaction(trans);
4635 down(&fs_info->uuid_tree_rescan_sem);
4636 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4638 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4639 btrfs_warn(fs_info, "failed to start uuid_scan task");
4640 up(&fs_info->uuid_tree_rescan_sem);
4641 return PTR_ERR(task);
4647 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4649 struct task_struct *task;
4651 down(&fs_info->uuid_tree_rescan_sem);
4652 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4654 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4655 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4656 up(&fs_info->uuid_tree_rescan_sem);
4657 return PTR_ERR(task);
4664 * shrinking a device means finding all of the device extents past
4665 * the new size, and then following the back refs to the chunks.
4666 * The chunk relocation code actually frees the device extent
4668 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4670 struct btrfs_fs_info *fs_info = device->fs_info;
4671 struct btrfs_root *root = fs_info->dev_root;
4672 struct btrfs_trans_handle *trans;
4673 struct btrfs_dev_extent *dev_extent = NULL;
4674 struct btrfs_path *path;
4680 bool retried = false;
4681 struct extent_buffer *l;
4682 struct btrfs_key key;
4683 struct btrfs_super_block *super_copy = fs_info->super_copy;
4684 u64 old_total = btrfs_super_total_bytes(super_copy);
4685 u64 old_size = btrfs_device_get_total_bytes(device);
4689 new_size = round_down(new_size, fs_info->sectorsize);
4691 diff = round_down(old_size - new_size, fs_info->sectorsize);
4693 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4696 path = btrfs_alloc_path();
4700 path->reada = READA_BACK;
4702 trans = btrfs_start_transaction(root, 0);
4703 if (IS_ERR(trans)) {
4704 btrfs_free_path(path);
4705 return PTR_ERR(trans);
4708 mutex_lock(&fs_info->chunk_mutex);
4710 btrfs_device_set_total_bytes(device, new_size);
4711 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4712 device->fs_devices->total_rw_bytes -= diff;
4713 atomic64_sub(diff, &fs_info->free_chunk_space);
4717 * Once the device's size has been set to the new size, ensure all
4718 * in-memory chunks are synced to disk so that the loop below sees them
4719 * and relocates them accordingly.
4721 if (contains_pending_extent(device, &start, diff)) {
4722 mutex_unlock(&fs_info->chunk_mutex);
4723 ret = btrfs_commit_transaction(trans);
4727 mutex_unlock(&fs_info->chunk_mutex);
4728 btrfs_end_transaction(trans);
4732 key.objectid = device->devid;
4733 key.offset = (u64)-1;
4734 key.type = BTRFS_DEV_EXTENT_KEY;
4737 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4740 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4744 ret = btrfs_previous_item(root, path, 0, key.type);
4746 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4751 btrfs_release_path(path);
4756 slot = path->slots[0];
4757 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4759 if (key.objectid != device->devid) {
4760 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4761 btrfs_release_path(path);
4765 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4766 length = btrfs_dev_extent_length(l, dev_extent);
4768 if (key.offset + length <= new_size) {
4769 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4770 btrfs_release_path(path);
4774 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4775 btrfs_release_path(path);
4778 * We may be relocating the only data chunk we have,
4779 * which could potentially end up with losing data's
4780 * raid profile, so lets allocate an empty one in
4783 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4785 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4789 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4790 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4791 if (ret == -ENOSPC) {
4794 if (ret == -ETXTBSY) {
4796 "could not shrink block group %llu due to active swapfile",
4801 } while (key.offset-- > 0);
4803 if (failed && !retried) {
4807 } else if (failed && retried) {
4812 /* Shrinking succeeded, else we would be at "done". */
4813 trans = btrfs_start_transaction(root, 0);
4814 if (IS_ERR(trans)) {
4815 ret = PTR_ERR(trans);
4819 mutex_lock(&fs_info->chunk_mutex);
4820 btrfs_device_set_disk_total_bytes(device, new_size);
4821 if (list_empty(&device->post_commit_list))
4822 list_add_tail(&device->post_commit_list,
4823 &trans->transaction->dev_update_list);
4825 WARN_ON(diff > old_total);
4826 btrfs_set_super_total_bytes(super_copy,
4827 round_down(old_total - diff, fs_info->sectorsize));
4828 mutex_unlock(&fs_info->chunk_mutex);
4830 /* Now btrfs_update_device() will change the on-disk size. */
4831 ret = btrfs_update_device(trans, device);
4833 btrfs_abort_transaction(trans, ret);
4834 btrfs_end_transaction(trans);
4836 ret = btrfs_commit_transaction(trans);
4839 btrfs_free_path(path);
4841 mutex_lock(&fs_info->chunk_mutex);
4842 btrfs_device_set_total_bytes(device, old_size);
4843 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4844 device->fs_devices->total_rw_bytes += diff;
4845 atomic64_add(diff, &fs_info->free_chunk_space);
4846 mutex_unlock(&fs_info->chunk_mutex);
4851 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4852 struct btrfs_key *key,
4853 struct btrfs_chunk *chunk, int item_size)
4855 struct btrfs_super_block *super_copy = fs_info->super_copy;
4856 struct btrfs_disk_key disk_key;
4860 mutex_lock(&fs_info->chunk_mutex);
4861 array_size = btrfs_super_sys_array_size(super_copy);
4862 if (array_size + item_size + sizeof(disk_key)
4863 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4864 mutex_unlock(&fs_info->chunk_mutex);
4868 ptr = super_copy->sys_chunk_array + array_size;
4869 btrfs_cpu_key_to_disk(&disk_key, key);
4870 memcpy(ptr, &disk_key, sizeof(disk_key));
4871 ptr += sizeof(disk_key);
4872 memcpy(ptr, chunk, item_size);
4873 item_size += sizeof(disk_key);
4874 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4875 mutex_unlock(&fs_info->chunk_mutex);
4881 * sort the devices in descending order by max_avail, total_avail
4883 static int btrfs_cmp_device_info(const void *a, const void *b)
4885 const struct btrfs_device_info *di_a = a;
4886 const struct btrfs_device_info *di_b = b;
4888 if (di_a->max_avail > di_b->max_avail)
4890 if (di_a->max_avail < di_b->max_avail)
4892 if (di_a->total_avail > di_b->total_avail)
4894 if (di_a->total_avail < di_b->total_avail)
4899 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4901 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4904 btrfs_set_fs_incompat(info, RAID56);
4907 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4908 u64 start, u64 type)
4910 struct btrfs_fs_info *info = trans->fs_info;
4911 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4912 struct btrfs_device *device;
4913 struct map_lookup *map = NULL;
4914 struct extent_map_tree *em_tree;
4915 struct extent_map *em;
4916 struct btrfs_device_info *devices_info = NULL;
4918 int num_stripes; /* total number of stripes to allocate */
4919 int data_stripes; /* number of stripes that count for
4921 int sub_stripes; /* sub_stripes info for map */
4922 int dev_stripes; /* stripes per dev */
4923 int devs_max; /* max devs to use */
4924 int devs_min; /* min devs needed */
4925 int devs_increment; /* ndevs has to be a multiple of this */
4926 int ncopies; /* how many copies to data has */
4927 int nparity; /* number of stripes worth of bytes to
4928 store parity information */
4930 u64 max_stripe_size;
4939 BUG_ON(!alloc_profile_is_valid(type, 0));
4941 if (list_empty(&fs_devices->alloc_list)) {
4942 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4943 btrfs_debug(info, "%s: no writable device", __func__);
4947 index = btrfs_bg_flags_to_raid_index(type);
4949 sub_stripes = btrfs_raid_array[index].sub_stripes;
4950 dev_stripes = btrfs_raid_array[index].dev_stripes;
4951 devs_max = btrfs_raid_array[index].devs_max;
4952 devs_min = btrfs_raid_array[index].devs_min;
4953 devs_increment = btrfs_raid_array[index].devs_increment;
4954 ncopies = btrfs_raid_array[index].ncopies;
4955 nparity = btrfs_raid_array[index].nparity;
4957 if (type & BTRFS_BLOCK_GROUP_DATA) {
4958 max_stripe_size = SZ_1G;
4959 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4961 devs_max = BTRFS_MAX_DEVS(info);
4962 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4963 /* for larger filesystems, use larger metadata chunks */
4964 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4965 max_stripe_size = SZ_1G;
4967 max_stripe_size = SZ_256M;
4968 max_chunk_size = max_stripe_size;
4970 devs_max = BTRFS_MAX_DEVS(info);
4971 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4972 max_stripe_size = SZ_32M;
4973 max_chunk_size = 2 * max_stripe_size;
4975 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4977 btrfs_err(info, "invalid chunk type 0x%llx requested",
4982 /* We don't want a chunk larger than 10% of writable space */
4983 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4986 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4992 * in the first pass through the devices list, we gather information
4993 * about the available holes on each device.
4996 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5000 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5002 "BTRFS: read-only device in alloc_list\n");
5006 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5007 &device->dev_state) ||
5008 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5011 if (device->total_bytes > device->bytes_used)
5012 total_avail = device->total_bytes - device->bytes_used;
5016 /* If there is no space on this device, skip it. */
5017 if (total_avail == 0)
5020 ret = find_free_dev_extent(device,
5021 max_stripe_size * dev_stripes,
5022 &dev_offset, &max_avail);
5023 if (ret && ret != -ENOSPC)
5027 max_avail = max_stripe_size * dev_stripes;
5029 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5030 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5032 "%s: devid %llu has no free space, have=%llu want=%u",
5033 __func__, device->devid, max_avail,
5034 BTRFS_STRIPE_LEN * dev_stripes);
5038 if (ndevs == fs_devices->rw_devices) {
5039 WARN(1, "%s: found more than %llu devices\n",
5040 __func__, fs_devices->rw_devices);
5043 devices_info[ndevs].dev_offset = dev_offset;
5044 devices_info[ndevs].max_avail = max_avail;
5045 devices_info[ndevs].total_avail = total_avail;
5046 devices_info[ndevs].dev = device;
5051 * now sort the devices by hole size / available space
5053 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5054 btrfs_cmp_device_info, NULL);
5056 /* round down to number of usable stripes */
5057 ndevs = round_down(ndevs, devs_increment);
5059 if (ndevs < devs_min) {
5061 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5063 "%s: not enough devices with free space: have=%d minimum required=%d",
5064 __func__, ndevs, devs_min);
5069 ndevs = min(ndevs, devs_max);
5072 * The primary goal is to maximize the number of stripes, so use as
5073 * many devices as possible, even if the stripes are not maximum sized.
5075 * The DUP profile stores more than one stripe per device, the
5076 * max_avail is the total size so we have to adjust.
5078 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5079 num_stripes = ndevs * dev_stripes;
5082 * this will have to be fixed for RAID1 and RAID10 over
5085 data_stripes = (num_stripes - nparity) / ncopies;
5088 * Use the number of data stripes to figure out how big this chunk
5089 * is really going to be in terms of logical address space,
5090 * and compare that answer with the max chunk size. If it's higher,
5091 * we try to reduce stripe_size.
5093 if (stripe_size * data_stripes > max_chunk_size) {
5095 * Reduce stripe_size, round it up to a 16MB boundary again and
5096 * then use it, unless it ends up being even bigger than the
5097 * previous value we had already.
5099 stripe_size = min(round_up(div_u64(max_chunk_size,
5100 data_stripes), SZ_16M),
5104 /* align to BTRFS_STRIPE_LEN */
5105 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5107 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5112 map->num_stripes = num_stripes;
5114 for (i = 0; i < ndevs; ++i) {
5115 for (j = 0; j < dev_stripes; ++j) {
5116 int s = i * dev_stripes + j;
5117 map->stripes[s].dev = devices_info[i].dev;
5118 map->stripes[s].physical = devices_info[i].dev_offset +
5122 map->stripe_len = BTRFS_STRIPE_LEN;
5123 map->io_align = BTRFS_STRIPE_LEN;
5124 map->io_width = BTRFS_STRIPE_LEN;
5126 map->sub_stripes = sub_stripes;
5128 chunk_size = stripe_size * data_stripes;
5130 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5132 em = alloc_extent_map();
5138 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5139 em->map_lookup = map;
5141 em->len = chunk_size;
5142 em->block_start = 0;
5143 em->block_len = em->len;
5144 em->orig_block_len = stripe_size;
5146 em_tree = &info->mapping_tree.map_tree;
5147 write_lock(&em_tree->lock);
5148 ret = add_extent_mapping(em_tree, em, 0);
5150 write_unlock(&em_tree->lock);
5151 free_extent_map(em);
5154 write_unlock(&em_tree->lock);
5156 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5158 goto error_del_extent;
5160 for (i = 0; i < map->num_stripes; i++) {
5161 struct btrfs_device *dev = map->stripes[i].dev;
5163 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5164 if (list_empty(&dev->post_commit_list))
5165 list_add_tail(&dev->post_commit_list,
5166 &trans->transaction->dev_update_list);
5169 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5171 free_extent_map(em);
5172 check_raid56_incompat_flag(info, type);
5174 kfree(devices_info);
5178 write_lock(&em_tree->lock);
5179 remove_extent_mapping(em_tree, em);
5180 write_unlock(&em_tree->lock);
5182 /* One for our allocation */
5183 free_extent_map(em);
5184 /* One for the tree reference */
5185 free_extent_map(em);
5187 kfree(devices_info);
5191 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5192 u64 chunk_offset, u64 chunk_size)
5194 struct btrfs_fs_info *fs_info = trans->fs_info;
5195 struct btrfs_root *extent_root = fs_info->extent_root;
5196 struct btrfs_root *chunk_root = fs_info->chunk_root;
5197 struct btrfs_key key;
5198 struct btrfs_device *device;
5199 struct btrfs_chunk *chunk;
5200 struct btrfs_stripe *stripe;
5201 struct extent_map *em;
5202 struct map_lookup *map;
5209 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5213 map = em->map_lookup;
5214 item_size = btrfs_chunk_item_size(map->num_stripes);
5215 stripe_size = em->orig_block_len;
5217 chunk = kzalloc(item_size, GFP_NOFS);
5224 * Take the device list mutex to prevent races with the final phase of
5225 * a device replace operation that replaces the device object associated
5226 * with the map's stripes, because the device object's id can change
5227 * at any time during that final phase of the device replace operation
5228 * (dev-replace.c:btrfs_dev_replace_finishing()).
5230 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5231 for (i = 0; i < map->num_stripes; i++) {
5232 device = map->stripes[i].dev;
5233 dev_offset = map->stripes[i].physical;
5235 ret = btrfs_update_device(trans, device);
5238 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5239 dev_offset, stripe_size);
5244 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5248 stripe = &chunk->stripe;
5249 for (i = 0; i < map->num_stripes; i++) {
5250 device = map->stripes[i].dev;
5251 dev_offset = map->stripes[i].physical;
5253 btrfs_set_stack_stripe_devid(stripe, device->devid);
5254 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5255 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5258 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5260 btrfs_set_stack_chunk_length(chunk, chunk_size);
5261 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5262 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5263 btrfs_set_stack_chunk_type(chunk, map->type);
5264 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5265 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5266 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5267 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5268 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5270 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5271 key.type = BTRFS_CHUNK_ITEM_KEY;
5272 key.offset = chunk_offset;
5274 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5275 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5277 * TODO: Cleanup of inserted chunk root in case of
5280 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5285 free_extent_map(em);
5290 * Chunk allocation falls into two parts. The first part does work
5291 * that makes the new allocated chunk usable, but does not do any operation
5292 * that modifies the chunk tree. The second part does the work that
5293 * requires modifying the chunk tree. This division is important for the
5294 * bootstrap process of adding storage to a seed btrfs.
5296 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5300 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5301 chunk_offset = find_next_chunk(trans->fs_info);
5302 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5305 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5307 struct btrfs_fs_info *fs_info = trans->fs_info;
5309 u64 sys_chunk_offset;
5313 chunk_offset = find_next_chunk(fs_info);
5314 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5315 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5319 sys_chunk_offset = find_next_chunk(fs_info);
5320 alloc_profile = btrfs_system_alloc_profile(fs_info);
5321 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5325 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5329 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5330 BTRFS_BLOCK_GROUP_RAID10 |
5331 BTRFS_BLOCK_GROUP_RAID5 |
5332 BTRFS_BLOCK_GROUP_DUP)) {
5334 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5343 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5345 struct extent_map *em;
5346 struct map_lookup *map;
5351 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5355 map = em->map_lookup;
5356 for (i = 0; i < map->num_stripes; i++) {
5357 if (test_bit(BTRFS_DEV_STATE_MISSING,
5358 &map->stripes[i].dev->dev_state)) {
5362 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5363 &map->stripes[i].dev->dev_state)) {
5370 * If the number of missing devices is larger than max errors,
5371 * we can not write the data into that chunk successfully, so
5374 if (miss_ndevs > btrfs_chunk_max_errors(map))
5377 free_extent_map(em);
5381 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5383 extent_map_tree_init(&tree->map_tree);
5386 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5388 struct extent_map *em;
5391 write_lock(&tree->map_tree.lock);
5392 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5394 remove_extent_mapping(&tree->map_tree, em);
5395 write_unlock(&tree->map_tree.lock);
5399 free_extent_map(em);
5400 /* once for the tree */
5401 free_extent_map(em);
5405 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5407 struct extent_map *em;
5408 struct map_lookup *map;
5411 em = btrfs_get_chunk_map(fs_info, logical, len);
5414 * We could return errors for these cases, but that could get
5415 * ugly and we'd probably do the same thing which is just not do
5416 * anything else and exit, so return 1 so the callers don't try
5417 * to use other copies.
5421 map = em->map_lookup;
5422 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5423 ret = map->num_stripes;
5424 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5425 ret = map->sub_stripes;
5426 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5428 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5430 * There could be two corrupted data stripes, we need
5431 * to loop retry in order to rebuild the correct data.
5433 * Fail a stripe at a time on every retry except the
5434 * stripe under reconstruction.
5436 ret = map->num_stripes;
5439 free_extent_map(em);
5441 down_read(&fs_info->dev_replace.rwsem);
5442 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5443 fs_info->dev_replace.tgtdev)
5445 up_read(&fs_info->dev_replace.rwsem);
5450 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5453 struct extent_map *em;
5454 struct map_lookup *map;
5455 unsigned long len = fs_info->sectorsize;
5457 em = btrfs_get_chunk_map(fs_info, logical, len);
5459 if (!WARN_ON(IS_ERR(em))) {
5460 map = em->map_lookup;
5461 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5462 len = map->stripe_len * nr_data_stripes(map);
5463 free_extent_map(em);
5468 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5470 struct extent_map *em;
5471 struct map_lookup *map;
5474 em = btrfs_get_chunk_map(fs_info, logical, len);
5476 if(!WARN_ON(IS_ERR(em))) {
5477 map = em->map_lookup;
5478 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5480 free_extent_map(em);
5485 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5486 struct map_lookup *map, int first,
5487 int dev_replace_is_ongoing)
5491 int preferred_mirror;
5493 struct btrfs_device *srcdev;
5496 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5498 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5499 num_stripes = map->sub_stripes;
5501 num_stripes = map->num_stripes;
5503 preferred_mirror = first + current->pid % num_stripes;
5505 if (dev_replace_is_ongoing &&
5506 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5507 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5508 srcdev = fs_info->dev_replace.srcdev;
5513 * try to avoid the drive that is the source drive for a
5514 * dev-replace procedure, only choose it if no other non-missing
5515 * mirror is available
5517 for (tolerance = 0; tolerance < 2; tolerance++) {
5518 if (map->stripes[preferred_mirror].dev->bdev &&
5519 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5520 return preferred_mirror;
5521 for (i = first; i < first + num_stripes; i++) {
5522 if (map->stripes[i].dev->bdev &&
5523 (tolerance || map->stripes[i].dev != srcdev))
5528 /* we couldn't find one that doesn't fail. Just return something
5529 * and the io error handling code will clean up eventually
5531 return preferred_mirror;
5534 static inline int parity_smaller(u64 a, u64 b)
5539 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5540 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5542 struct btrfs_bio_stripe s;
5549 for (i = 0; i < num_stripes - 1; i++) {
5550 if (parity_smaller(bbio->raid_map[i],
5551 bbio->raid_map[i+1])) {
5552 s = bbio->stripes[i];
5553 l = bbio->raid_map[i];
5554 bbio->stripes[i] = bbio->stripes[i+1];
5555 bbio->raid_map[i] = bbio->raid_map[i+1];
5556 bbio->stripes[i+1] = s;
5557 bbio->raid_map[i+1] = l;
5565 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5567 struct btrfs_bio *bbio = kzalloc(
5568 /* the size of the btrfs_bio */
5569 sizeof(struct btrfs_bio) +
5570 /* plus the variable array for the stripes */
5571 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5572 /* plus the variable array for the tgt dev */
5573 sizeof(int) * (real_stripes) +
5575 * plus the raid_map, which includes both the tgt dev
5578 sizeof(u64) * (total_stripes),
5579 GFP_NOFS|__GFP_NOFAIL);
5581 atomic_set(&bbio->error, 0);
5582 refcount_set(&bbio->refs, 1);
5587 void btrfs_get_bbio(struct btrfs_bio *bbio)
5589 WARN_ON(!refcount_read(&bbio->refs));
5590 refcount_inc(&bbio->refs);
5593 void btrfs_put_bbio(struct btrfs_bio *bbio)
5597 if (refcount_dec_and_test(&bbio->refs))
5601 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5603 * Please note that, discard won't be sent to target device of device
5606 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5607 u64 logical, u64 length,
5608 struct btrfs_bio **bbio_ret)
5610 struct extent_map *em;
5611 struct map_lookup *map;
5612 struct btrfs_bio *bbio;
5616 u64 stripe_end_offset;
5623 u32 sub_stripes = 0;
5624 u64 stripes_per_dev = 0;
5625 u32 remaining_stripes = 0;
5626 u32 last_stripe = 0;
5630 /* discard always return a bbio */
5633 em = btrfs_get_chunk_map(fs_info, logical, length);
5637 map = em->map_lookup;
5638 /* we don't discard raid56 yet */
5639 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5644 offset = logical - em->start;
5645 length = min_t(u64, em->len - offset, length);
5647 stripe_len = map->stripe_len;
5649 * stripe_nr counts the total number of stripes we have to stride
5650 * to get to this block
5652 stripe_nr = div64_u64(offset, stripe_len);
5654 /* stripe_offset is the offset of this block in its stripe */
5655 stripe_offset = offset - stripe_nr * stripe_len;
5657 stripe_nr_end = round_up(offset + length, map->stripe_len);
5658 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5659 stripe_cnt = stripe_nr_end - stripe_nr;
5660 stripe_end_offset = stripe_nr_end * map->stripe_len -
5663 * after this, stripe_nr is the number of stripes on this
5664 * device we have to walk to find the data, and stripe_index is
5665 * the number of our device in the stripe array
5669 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5670 BTRFS_BLOCK_GROUP_RAID10)) {
5671 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5674 sub_stripes = map->sub_stripes;
5676 factor = map->num_stripes / sub_stripes;
5677 num_stripes = min_t(u64, map->num_stripes,
5678 sub_stripes * stripe_cnt);
5679 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5680 stripe_index *= sub_stripes;
5681 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5682 &remaining_stripes);
5683 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5684 last_stripe *= sub_stripes;
5685 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5686 BTRFS_BLOCK_GROUP_DUP)) {
5687 num_stripes = map->num_stripes;
5689 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5693 bbio = alloc_btrfs_bio(num_stripes, 0);
5699 for (i = 0; i < num_stripes; i++) {
5700 bbio->stripes[i].physical =
5701 map->stripes[stripe_index].physical +
5702 stripe_offset + stripe_nr * map->stripe_len;
5703 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5705 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5706 BTRFS_BLOCK_GROUP_RAID10)) {
5707 bbio->stripes[i].length = stripes_per_dev *
5710 if (i / sub_stripes < remaining_stripes)
5711 bbio->stripes[i].length +=
5715 * Special for the first stripe and
5718 * |-------|...|-------|
5722 if (i < sub_stripes)
5723 bbio->stripes[i].length -=
5726 if (stripe_index >= last_stripe &&
5727 stripe_index <= (last_stripe +
5729 bbio->stripes[i].length -=
5732 if (i == sub_stripes - 1)
5735 bbio->stripes[i].length = length;
5739 if (stripe_index == map->num_stripes) {
5746 bbio->map_type = map->type;
5747 bbio->num_stripes = num_stripes;
5749 free_extent_map(em);
5754 * In dev-replace case, for repair case (that's the only case where the mirror
5755 * is selected explicitly when calling btrfs_map_block), blocks left of the
5756 * left cursor can also be read from the target drive.
5758 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5760 * For READ, it also needs to be supported using the same mirror number.
5762 * If the requested block is not left of the left cursor, EIO is returned. This
5763 * can happen because btrfs_num_copies() returns one more in the dev-replace
5766 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5767 u64 logical, u64 length,
5768 u64 srcdev_devid, int *mirror_num,
5771 struct btrfs_bio *bbio = NULL;
5773 int index_srcdev = 0;
5775 u64 physical_of_found = 0;
5779 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5780 logical, &length, &bbio, 0, 0);
5782 ASSERT(bbio == NULL);
5786 num_stripes = bbio->num_stripes;
5787 if (*mirror_num > num_stripes) {
5789 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5790 * that means that the requested area is not left of the left
5793 btrfs_put_bbio(bbio);
5798 * process the rest of the function using the mirror_num of the source
5799 * drive. Therefore look it up first. At the end, patch the device
5800 * pointer to the one of the target drive.
5802 for (i = 0; i < num_stripes; i++) {
5803 if (bbio->stripes[i].dev->devid != srcdev_devid)
5807 * In case of DUP, in order to keep it simple, only add the
5808 * mirror with the lowest physical address
5811 physical_of_found <= bbio->stripes[i].physical)
5816 physical_of_found = bbio->stripes[i].physical;
5819 btrfs_put_bbio(bbio);
5825 *mirror_num = index_srcdev + 1;
5826 *physical = physical_of_found;
5830 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5831 struct btrfs_bio **bbio_ret,
5832 struct btrfs_dev_replace *dev_replace,
5833 int *num_stripes_ret, int *max_errors_ret)
5835 struct btrfs_bio *bbio = *bbio_ret;
5836 u64 srcdev_devid = dev_replace->srcdev->devid;
5837 int tgtdev_indexes = 0;
5838 int num_stripes = *num_stripes_ret;
5839 int max_errors = *max_errors_ret;
5842 if (op == BTRFS_MAP_WRITE) {
5843 int index_where_to_add;
5846 * duplicate the write operations while the dev replace
5847 * procedure is running. Since the copying of the old disk to
5848 * the new disk takes place at run time while the filesystem is
5849 * mounted writable, the regular write operations to the old
5850 * disk have to be duplicated to go to the new disk as well.
5852 * Note that device->missing is handled by the caller, and that
5853 * the write to the old disk is already set up in the stripes
5856 index_where_to_add = num_stripes;
5857 for (i = 0; i < num_stripes; i++) {
5858 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5859 /* write to new disk, too */
5860 struct btrfs_bio_stripe *new =
5861 bbio->stripes + index_where_to_add;
5862 struct btrfs_bio_stripe *old =
5865 new->physical = old->physical;
5866 new->length = old->length;
5867 new->dev = dev_replace->tgtdev;
5868 bbio->tgtdev_map[i] = index_where_to_add;
5869 index_where_to_add++;
5874 num_stripes = index_where_to_add;
5875 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5876 int index_srcdev = 0;
5878 u64 physical_of_found = 0;
5881 * During the dev-replace procedure, the target drive can also
5882 * be used to read data in case it is needed to repair a corrupt
5883 * block elsewhere. This is possible if the requested area is
5884 * left of the left cursor. In this area, the target drive is a
5885 * full copy of the source drive.
5887 for (i = 0; i < num_stripes; i++) {
5888 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5890 * In case of DUP, in order to keep it simple,
5891 * only add the mirror with the lowest physical
5895 physical_of_found <=
5896 bbio->stripes[i].physical)
5900 physical_of_found = bbio->stripes[i].physical;
5904 struct btrfs_bio_stripe *tgtdev_stripe =
5905 bbio->stripes + num_stripes;
5907 tgtdev_stripe->physical = physical_of_found;
5908 tgtdev_stripe->length =
5909 bbio->stripes[index_srcdev].length;
5910 tgtdev_stripe->dev = dev_replace->tgtdev;
5911 bbio->tgtdev_map[index_srcdev] = num_stripes;
5918 *num_stripes_ret = num_stripes;
5919 *max_errors_ret = max_errors;
5920 bbio->num_tgtdevs = tgtdev_indexes;
5924 static bool need_full_stripe(enum btrfs_map_op op)
5926 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5929 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5930 enum btrfs_map_op op,
5931 u64 logical, u64 *length,
5932 struct btrfs_bio **bbio_ret,
5933 int mirror_num, int need_raid_map)
5935 struct extent_map *em;
5936 struct map_lookup *map;
5946 int tgtdev_indexes = 0;
5947 struct btrfs_bio *bbio = NULL;
5948 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5949 int dev_replace_is_ongoing = 0;
5950 int num_alloc_stripes;
5951 int patch_the_first_stripe_for_dev_replace = 0;
5952 u64 physical_to_patch_in_first_stripe = 0;
5953 u64 raid56_full_stripe_start = (u64)-1;
5955 if (op == BTRFS_MAP_DISCARD)
5956 return __btrfs_map_block_for_discard(fs_info, logical,
5959 em = btrfs_get_chunk_map(fs_info, logical, *length);
5963 map = em->map_lookup;
5964 offset = logical - em->start;
5966 stripe_len = map->stripe_len;
5969 * stripe_nr counts the total number of stripes we have to stride
5970 * to get to this block
5972 stripe_nr = div64_u64(stripe_nr, stripe_len);
5974 stripe_offset = stripe_nr * stripe_len;
5975 if (offset < stripe_offset) {
5977 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5978 stripe_offset, offset, em->start, logical,
5980 free_extent_map(em);
5984 /* stripe_offset is the offset of this block in its stripe*/
5985 stripe_offset = offset - stripe_offset;
5987 /* if we're here for raid56, we need to know the stripe aligned start */
5988 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5989 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5990 raid56_full_stripe_start = offset;
5992 /* allow a write of a full stripe, but make sure we don't
5993 * allow straddling of stripes
5995 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5997 raid56_full_stripe_start *= full_stripe_len;
6000 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6002 /* For writes to RAID[56], allow a full stripeset across all disks.
6003 For other RAID types and for RAID[56] reads, just allow a single
6004 stripe (on a single disk). */
6005 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6006 (op == BTRFS_MAP_WRITE)) {
6007 max_len = stripe_len * nr_data_stripes(map) -
6008 (offset - raid56_full_stripe_start);
6010 /* we limit the length of each bio to what fits in a stripe */
6011 max_len = stripe_len - stripe_offset;
6013 *length = min_t(u64, em->len - offset, max_len);
6015 *length = em->len - offset;
6019 * This is for when we're called from btrfs_bio_fits_in_stripe and all
6020 * it cares about is the length
6025 down_read(&dev_replace->rwsem);
6026 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6028 * Hold the semaphore for read during the whole operation, write is
6029 * requested at commit time but must wait.
6031 if (!dev_replace_is_ongoing)
6032 up_read(&dev_replace->rwsem);
6034 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6035 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6036 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6037 dev_replace->srcdev->devid,
6039 &physical_to_patch_in_first_stripe);
6043 patch_the_first_stripe_for_dev_replace = 1;
6044 } else if (mirror_num > map->num_stripes) {
6050 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6051 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6053 if (!need_full_stripe(op))
6055 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
6056 if (need_full_stripe(op))
6057 num_stripes = map->num_stripes;
6058 else if (mirror_num)
6059 stripe_index = mirror_num - 1;
6061 stripe_index = find_live_mirror(fs_info, map, 0,
6062 dev_replace_is_ongoing);
6063 mirror_num = stripe_index + 1;
6066 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6067 if (need_full_stripe(op)) {
6068 num_stripes = map->num_stripes;
6069 } else if (mirror_num) {
6070 stripe_index = mirror_num - 1;
6075 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6076 u32 factor = map->num_stripes / map->sub_stripes;
6078 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6079 stripe_index *= map->sub_stripes;
6081 if (need_full_stripe(op))
6082 num_stripes = map->sub_stripes;
6083 else if (mirror_num)
6084 stripe_index += mirror_num - 1;
6086 int old_stripe_index = stripe_index;
6087 stripe_index = find_live_mirror(fs_info, map,
6089 dev_replace_is_ongoing);
6090 mirror_num = stripe_index - old_stripe_index + 1;
6093 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6094 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6095 /* push stripe_nr back to the start of the full stripe */
6096 stripe_nr = div64_u64(raid56_full_stripe_start,
6097 stripe_len * nr_data_stripes(map));
6099 /* RAID[56] write or recovery. Return all stripes */
6100 num_stripes = map->num_stripes;
6101 max_errors = nr_parity_stripes(map);
6103 *length = map->stripe_len;
6108 * Mirror #0 or #1 means the original data block.
6109 * Mirror #2 is RAID5 parity block.
6110 * Mirror #3 is RAID6 Q block.
6112 stripe_nr = div_u64_rem(stripe_nr,
6113 nr_data_stripes(map), &stripe_index);
6115 stripe_index = nr_data_stripes(map) +
6118 /* We distribute the parity blocks across stripes */
6119 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6121 if (!need_full_stripe(op) && mirror_num <= 1)
6126 * after this, stripe_nr is the number of stripes on this
6127 * device we have to walk to find the data, and stripe_index is
6128 * the number of our device in the stripe array
6130 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6132 mirror_num = stripe_index + 1;
6134 if (stripe_index >= map->num_stripes) {
6136 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6137 stripe_index, map->num_stripes);
6142 num_alloc_stripes = num_stripes;
6143 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6144 if (op == BTRFS_MAP_WRITE)
6145 num_alloc_stripes <<= 1;
6146 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6147 num_alloc_stripes++;
6148 tgtdev_indexes = num_stripes;
6151 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6156 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6157 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6159 /* build raid_map */
6160 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6161 (need_full_stripe(op) || mirror_num > 1)) {
6165 bbio->raid_map = (u64 *)((void *)bbio->stripes +
6166 sizeof(struct btrfs_bio_stripe) *
6168 sizeof(int) * tgtdev_indexes);
6170 /* Work out the disk rotation on this stripe-set */
6171 div_u64_rem(stripe_nr, num_stripes, &rot);
6173 /* Fill in the logical address of each stripe */
6174 tmp = stripe_nr * nr_data_stripes(map);
6175 for (i = 0; i < nr_data_stripes(map); i++)
6176 bbio->raid_map[(i+rot) % num_stripes] =
6177 em->start + (tmp + i) * map->stripe_len;
6179 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6180 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6181 bbio->raid_map[(i+rot+1) % num_stripes] =
6186 for (i = 0; i < num_stripes; i++) {
6187 bbio->stripes[i].physical =
6188 map->stripes[stripe_index].physical +
6190 stripe_nr * map->stripe_len;
6191 bbio->stripes[i].dev =
6192 map->stripes[stripe_index].dev;
6196 if (need_full_stripe(op))
6197 max_errors = btrfs_chunk_max_errors(map);
6200 sort_parity_stripes(bbio, num_stripes);
6202 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6203 need_full_stripe(op)) {
6204 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6209 bbio->map_type = map->type;
6210 bbio->num_stripes = num_stripes;
6211 bbio->max_errors = max_errors;
6212 bbio->mirror_num = mirror_num;
6215 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6216 * mirror_num == num_stripes + 1 && dev_replace target drive is
6217 * available as a mirror
6219 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6220 WARN_ON(num_stripes > 1);
6221 bbio->stripes[0].dev = dev_replace->tgtdev;
6222 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6223 bbio->mirror_num = map->num_stripes + 1;
6226 if (dev_replace_is_ongoing) {
6227 lockdep_assert_held(&dev_replace->rwsem);
6228 /* Unlock and let waiting writers proceed */
6229 up_read(&dev_replace->rwsem);
6231 free_extent_map(em);
6235 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6236 u64 logical, u64 *length,
6237 struct btrfs_bio **bbio_ret, int mirror_num)
6239 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6243 /* For Scrub/replace */
6244 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6245 u64 logical, u64 *length,
6246 struct btrfs_bio **bbio_ret)
6248 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6251 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
6252 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
6254 struct extent_map *em;
6255 struct map_lookup *map;
6263 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
6267 map = em->map_lookup;
6269 rmap_len = map->stripe_len;
6271 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6272 length = div_u64(length, map->num_stripes / map->sub_stripes);
6273 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6274 length = div_u64(length, map->num_stripes);
6275 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6276 length = div_u64(length, nr_data_stripes(map));
6277 rmap_len = map->stripe_len * nr_data_stripes(map);
6280 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6281 BUG_ON(!buf); /* -ENOMEM */
6283 for (i = 0; i < map->num_stripes; i++) {
6284 if (map->stripes[i].physical > physical ||
6285 map->stripes[i].physical + length <= physical)
6288 stripe_nr = physical - map->stripes[i].physical;
6289 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6291 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6292 stripe_nr = stripe_nr * map->num_stripes + i;
6293 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6294 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6295 stripe_nr = stripe_nr * map->num_stripes + i;
6296 } /* else if RAID[56], multiply by nr_data_stripes().
6297 * Alternatively, just use rmap_len below instead of
6298 * map->stripe_len */
6300 bytenr = chunk_start + stripe_nr * rmap_len;
6301 WARN_ON(nr >= map->num_stripes);
6302 for (j = 0; j < nr; j++) {
6303 if (buf[j] == bytenr)
6307 WARN_ON(nr >= map->num_stripes);
6314 *stripe_len = rmap_len;
6316 free_extent_map(em);
6320 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6322 bio->bi_private = bbio->private;
6323 bio->bi_end_io = bbio->end_io;
6326 btrfs_put_bbio(bbio);
6329 static void btrfs_end_bio(struct bio *bio)
6331 struct btrfs_bio *bbio = bio->bi_private;
6332 int is_orig_bio = 0;
6334 if (bio->bi_status) {
6335 atomic_inc(&bbio->error);
6336 if (bio->bi_status == BLK_STS_IOERR ||
6337 bio->bi_status == BLK_STS_TARGET) {
6338 unsigned int stripe_index =
6339 btrfs_io_bio(bio)->stripe_index;
6340 struct btrfs_device *dev;
6342 BUG_ON(stripe_index >= bbio->num_stripes);
6343 dev = bbio->stripes[stripe_index].dev;
6345 if (bio_op(bio) == REQ_OP_WRITE)
6346 btrfs_dev_stat_inc_and_print(dev,
6347 BTRFS_DEV_STAT_WRITE_ERRS);
6348 else if (!(bio->bi_opf & REQ_RAHEAD))
6349 btrfs_dev_stat_inc_and_print(dev,
6350 BTRFS_DEV_STAT_READ_ERRS);
6351 if (bio->bi_opf & REQ_PREFLUSH)
6352 btrfs_dev_stat_inc_and_print(dev,
6353 BTRFS_DEV_STAT_FLUSH_ERRS);
6358 if (bio == bbio->orig_bio)
6361 btrfs_bio_counter_dec(bbio->fs_info);
6363 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6366 bio = bbio->orig_bio;
6369 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6370 /* only send an error to the higher layers if it is
6371 * beyond the tolerance of the btrfs bio
6373 if (atomic_read(&bbio->error) > bbio->max_errors) {
6374 bio->bi_status = BLK_STS_IOERR;
6377 * this bio is actually up to date, we didn't
6378 * go over the max number of errors
6380 bio->bi_status = BLK_STS_OK;
6383 btrfs_end_bbio(bbio, bio);
6384 } else if (!is_orig_bio) {
6390 * see run_scheduled_bios for a description of why bios are collected for
6393 * This will add one bio to the pending list for a device and make sure
6394 * the work struct is scheduled.
6396 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6399 struct btrfs_fs_info *fs_info = device->fs_info;
6400 int should_queue = 1;
6401 struct btrfs_pending_bios *pending_bios;
6403 /* don't bother with additional async steps for reads, right now */
6404 if (bio_op(bio) == REQ_OP_READ) {
6405 btrfsic_submit_bio(bio);
6409 WARN_ON(bio->bi_next);
6410 bio->bi_next = NULL;
6412 spin_lock(&device->io_lock);
6413 if (op_is_sync(bio->bi_opf))
6414 pending_bios = &device->pending_sync_bios;
6416 pending_bios = &device->pending_bios;
6418 if (pending_bios->tail)
6419 pending_bios->tail->bi_next = bio;
6421 pending_bios->tail = bio;
6422 if (!pending_bios->head)
6423 pending_bios->head = bio;
6424 if (device->running_pending)
6427 spin_unlock(&device->io_lock);
6430 btrfs_queue_work(fs_info->submit_workers, &device->work);
6433 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6434 u64 physical, int dev_nr, int async)
6436 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6437 struct btrfs_fs_info *fs_info = bbio->fs_info;
6439 bio->bi_private = bbio;
6440 btrfs_io_bio(bio)->stripe_index = dev_nr;
6441 bio->bi_end_io = btrfs_end_bio;
6442 bio->bi_iter.bi_sector = physical >> 9;
6443 btrfs_debug_in_rcu(fs_info,
6444 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6445 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6446 (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid,
6447 bio->bi_iter.bi_size);
6448 bio_set_dev(bio, dev->bdev);
6450 btrfs_bio_counter_inc_noblocked(fs_info);
6453 btrfs_schedule_bio(dev, bio);
6455 btrfsic_submit_bio(bio);
6458 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6460 atomic_inc(&bbio->error);
6461 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6462 /* Should be the original bio. */
6463 WARN_ON(bio != bbio->orig_bio);
6465 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6466 bio->bi_iter.bi_sector = logical >> 9;
6467 if (atomic_read(&bbio->error) > bbio->max_errors)
6468 bio->bi_status = BLK_STS_IOERR;
6470 bio->bi_status = BLK_STS_OK;
6471 btrfs_end_bbio(bbio, bio);
6475 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6476 int mirror_num, int async_submit)
6478 struct btrfs_device *dev;
6479 struct bio *first_bio = bio;
6480 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6486 struct btrfs_bio *bbio = NULL;
6488 length = bio->bi_iter.bi_size;
6489 map_length = length;
6491 btrfs_bio_counter_inc_blocked(fs_info);
6492 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6493 &map_length, &bbio, mirror_num, 1);
6495 btrfs_bio_counter_dec(fs_info);
6496 return errno_to_blk_status(ret);
6499 total_devs = bbio->num_stripes;
6500 bbio->orig_bio = first_bio;
6501 bbio->private = first_bio->bi_private;
6502 bbio->end_io = first_bio->bi_end_io;
6503 bbio->fs_info = fs_info;
6504 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6506 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6507 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6508 /* In this case, map_length has been set to the length of
6509 a single stripe; not the whole write */
6510 if (bio_op(bio) == REQ_OP_WRITE) {
6511 ret = raid56_parity_write(fs_info, bio, bbio,
6514 ret = raid56_parity_recover(fs_info, bio, bbio,
6515 map_length, mirror_num, 1);
6518 btrfs_bio_counter_dec(fs_info);
6519 return errno_to_blk_status(ret);
6522 if (map_length < length) {
6524 "mapping failed logical %llu bio len %llu len %llu",
6525 logical, length, map_length);
6529 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6530 dev = bbio->stripes[dev_nr].dev;
6531 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6533 (bio_op(first_bio) == REQ_OP_WRITE &&
6534 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6535 bbio_error(bbio, first_bio, logical);
6539 if (dev_nr < total_devs - 1)
6540 bio = btrfs_bio_clone(first_bio);
6544 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6545 dev_nr, async_submit);
6547 btrfs_bio_counter_dec(fs_info);
6552 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6555 * If devid and uuid are both specified, the match must be exact, otherwise
6556 * only devid is used.
6558 * If @seed is true, traverse through the seed devices.
6560 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6561 u64 devid, u8 *uuid, u8 *fsid,
6564 struct btrfs_device *device;
6566 while (fs_devices) {
6568 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6569 list_for_each_entry(device, &fs_devices->devices,
6571 if (device->devid == devid &&
6572 (!uuid || memcmp(device->uuid, uuid,
6573 BTRFS_UUID_SIZE) == 0))
6578 fs_devices = fs_devices->seed;
6585 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6586 u64 devid, u8 *dev_uuid)
6588 struct btrfs_device *device;
6590 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6594 list_add(&device->dev_list, &fs_devices->devices);
6595 device->fs_devices = fs_devices;
6596 fs_devices->num_devices++;
6598 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6599 fs_devices->missing_devices++;
6605 * btrfs_alloc_device - allocate struct btrfs_device
6606 * @fs_info: used only for generating a new devid, can be NULL if
6607 * devid is provided (i.e. @devid != NULL).
6608 * @devid: a pointer to devid for this device. If NULL a new devid
6610 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6613 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6614 * on error. Returned struct is not linked onto any lists and must be
6615 * destroyed with btrfs_free_device.
6617 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6621 struct btrfs_device *dev;
6624 if (WARN_ON(!devid && !fs_info))
6625 return ERR_PTR(-EINVAL);
6627 dev = __alloc_device();
6636 ret = find_next_devid(fs_info, &tmp);
6638 btrfs_free_device(dev);
6639 return ERR_PTR(ret);
6645 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6647 generate_random_uuid(dev->uuid);
6649 btrfs_init_work(&dev->work, btrfs_submit_helper,
6650 pending_bios_fn, NULL, NULL);
6655 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6656 u64 devid, u8 *uuid, bool error)
6659 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6662 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6666 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6668 int index = btrfs_bg_flags_to_raid_index(type);
6669 int ncopies = btrfs_raid_array[index].ncopies;
6672 switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6673 case BTRFS_BLOCK_GROUP_RAID5:
6674 data_stripes = num_stripes - 1;
6676 case BTRFS_BLOCK_GROUP_RAID6:
6677 data_stripes = num_stripes - 2;
6680 data_stripes = num_stripes / ncopies;
6683 return div_u64(chunk_len, data_stripes);
6686 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6687 struct btrfs_chunk *chunk)
6689 struct btrfs_fs_info *fs_info = leaf->fs_info;
6690 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6691 struct map_lookup *map;
6692 struct extent_map *em;
6696 u8 uuid[BTRFS_UUID_SIZE];
6701 logical = key->offset;
6702 length = btrfs_chunk_length(leaf, chunk);
6703 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6706 * Only need to verify chunk item if we're reading from sys chunk array,
6707 * as chunk item in tree block is already verified by tree-checker.
6709 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6710 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6715 read_lock(&map_tree->map_tree.lock);
6716 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6717 read_unlock(&map_tree->map_tree.lock);
6719 /* already mapped? */
6720 if (em && em->start <= logical && em->start + em->len > logical) {
6721 free_extent_map(em);
6724 free_extent_map(em);
6727 em = alloc_extent_map();
6730 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6732 free_extent_map(em);
6736 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6737 em->map_lookup = map;
6738 em->start = logical;
6741 em->block_start = 0;
6742 em->block_len = em->len;
6744 map->num_stripes = num_stripes;
6745 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6746 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6747 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6748 map->type = btrfs_chunk_type(leaf, chunk);
6749 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6750 map->verified_stripes = 0;
6751 em->orig_block_len = calc_stripe_length(map->type, em->len,
6753 for (i = 0; i < num_stripes; i++) {
6754 map->stripes[i].physical =
6755 btrfs_stripe_offset_nr(leaf, chunk, i);
6756 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6757 read_extent_buffer(leaf, uuid, (unsigned long)
6758 btrfs_stripe_dev_uuid_nr(chunk, i),
6760 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6761 devid, uuid, NULL, true);
6762 if (!map->stripes[i].dev &&
6763 !btrfs_test_opt(fs_info, DEGRADED)) {
6764 free_extent_map(em);
6765 btrfs_report_missing_device(fs_info, devid, uuid, true);
6768 if (!map->stripes[i].dev) {
6769 map->stripes[i].dev =
6770 add_missing_dev(fs_info->fs_devices, devid,
6772 if (IS_ERR(map->stripes[i].dev)) {
6773 free_extent_map(em);
6775 "failed to init missing dev %llu: %ld",
6776 devid, PTR_ERR(map->stripes[i].dev));
6777 return PTR_ERR(map->stripes[i].dev);
6779 btrfs_report_missing_device(fs_info, devid, uuid, false);
6781 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6782 &(map->stripes[i].dev->dev_state));
6786 write_lock(&map_tree->map_tree.lock);
6787 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6788 write_unlock(&map_tree->map_tree.lock);
6791 "failed to add chunk map, start=%llu len=%llu: %d",
6792 em->start, em->len, ret);
6794 free_extent_map(em);
6799 static void fill_device_from_item(struct extent_buffer *leaf,
6800 struct btrfs_dev_item *dev_item,
6801 struct btrfs_device *device)
6805 device->devid = btrfs_device_id(leaf, dev_item);
6806 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6807 device->total_bytes = device->disk_total_bytes;
6808 device->commit_total_bytes = device->disk_total_bytes;
6809 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6810 device->commit_bytes_used = device->bytes_used;
6811 device->type = btrfs_device_type(leaf, dev_item);
6812 device->io_align = btrfs_device_io_align(leaf, dev_item);
6813 device->io_width = btrfs_device_io_width(leaf, dev_item);
6814 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6815 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6816 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6818 ptr = btrfs_device_uuid(dev_item);
6819 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6822 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6825 struct btrfs_fs_devices *fs_devices;
6828 lockdep_assert_held(&uuid_mutex);
6831 fs_devices = fs_info->fs_devices->seed;
6832 while (fs_devices) {
6833 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6836 fs_devices = fs_devices->seed;
6839 fs_devices = find_fsid(fsid, NULL);
6841 if (!btrfs_test_opt(fs_info, DEGRADED))
6842 return ERR_PTR(-ENOENT);
6844 fs_devices = alloc_fs_devices(fsid, NULL);
6845 if (IS_ERR(fs_devices))
6848 fs_devices->seeding = 1;
6849 fs_devices->opened = 1;
6853 fs_devices = clone_fs_devices(fs_devices);
6854 if (IS_ERR(fs_devices))
6857 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6859 free_fs_devices(fs_devices);
6860 fs_devices = ERR_PTR(ret);
6864 if (!fs_devices->seeding) {
6865 close_fs_devices(fs_devices);
6866 free_fs_devices(fs_devices);
6867 fs_devices = ERR_PTR(-EINVAL);
6871 fs_devices->seed = fs_info->fs_devices->seed;
6872 fs_info->fs_devices->seed = fs_devices;
6877 static int read_one_dev(struct extent_buffer *leaf,
6878 struct btrfs_dev_item *dev_item)
6880 struct btrfs_fs_info *fs_info = leaf->fs_info;
6881 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6882 struct btrfs_device *device;
6885 u8 fs_uuid[BTRFS_FSID_SIZE];
6886 u8 dev_uuid[BTRFS_UUID_SIZE];
6888 devid = btrfs_device_id(leaf, dev_item);
6889 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6891 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6894 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6895 fs_devices = open_seed_devices(fs_info, fs_uuid);
6896 if (IS_ERR(fs_devices))
6897 return PTR_ERR(fs_devices);
6900 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6903 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6904 btrfs_report_missing_device(fs_info, devid,
6909 device = add_missing_dev(fs_devices, devid, dev_uuid);
6910 if (IS_ERR(device)) {
6912 "failed to add missing dev %llu: %ld",
6913 devid, PTR_ERR(device));
6914 return PTR_ERR(device);
6916 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6918 if (!device->bdev) {
6919 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6920 btrfs_report_missing_device(fs_info,
6921 devid, dev_uuid, true);
6924 btrfs_report_missing_device(fs_info, devid,
6928 if (!device->bdev &&
6929 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6931 * this happens when a device that was properly setup
6932 * in the device info lists suddenly goes bad.
6933 * device->bdev is NULL, and so we have to set
6934 * device->missing to one here
6936 device->fs_devices->missing_devices++;
6937 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6940 /* Move the device to its own fs_devices */
6941 if (device->fs_devices != fs_devices) {
6942 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6943 &device->dev_state));
6945 list_move(&device->dev_list, &fs_devices->devices);
6946 device->fs_devices->num_devices--;
6947 fs_devices->num_devices++;
6949 device->fs_devices->missing_devices--;
6950 fs_devices->missing_devices++;
6952 device->fs_devices = fs_devices;
6956 if (device->fs_devices != fs_info->fs_devices) {
6957 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6958 if (device->generation !=
6959 btrfs_device_generation(leaf, dev_item))
6963 fill_device_from_item(leaf, dev_item, device);
6964 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6965 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6966 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6967 device->fs_devices->total_rw_bytes += device->total_bytes;
6968 atomic64_add(device->total_bytes - device->bytes_used,
6969 &fs_info->free_chunk_space);
6975 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6977 struct btrfs_root *root = fs_info->tree_root;
6978 struct btrfs_super_block *super_copy = fs_info->super_copy;
6979 struct extent_buffer *sb;
6980 struct btrfs_disk_key *disk_key;
6981 struct btrfs_chunk *chunk;
6983 unsigned long sb_array_offset;
6990 struct btrfs_key key;
6992 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6994 * This will create extent buffer of nodesize, superblock size is
6995 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6996 * overallocate but we can keep it as-is, only the first page is used.
6998 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
7001 set_extent_buffer_uptodate(sb);
7002 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
7004 * The sb extent buffer is artificial and just used to read the system array.
7005 * set_extent_buffer_uptodate() call does not properly mark all it's
7006 * pages up-to-date when the page is larger: extent does not cover the
7007 * whole page and consequently check_page_uptodate does not find all
7008 * the page's extents up-to-date (the hole beyond sb),
7009 * write_extent_buffer then triggers a WARN_ON.
7011 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7012 * but sb spans only this function. Add an explicit SetPageUptodate call
7013 * to silence the warning eg. on PowerPC 64.
7015 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7016 SetPageUptodate(sb->pages[0]);
7018 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7019 array_size = btrfs_super_sys_array_size(super_copy);
7021 array_ptr = super_copy->sys_chunk_array;
7022 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7025 while (cur_offset < array_size) {
7026 disk_key = (struct btrfs_disk_key *)array_ptr;
7027 len = sizeof(*disk_key);
7028 if (cur_offset + len > array_size)
7029 goto out_short_read;
7031 btrfs_disk_key_to_cpu(&key, disk_key);
7034 sb_array_offset += len;
7037 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7038 chunk = (struct btrfs_chunk *)sb_array_offset;
7040 * At least one btrfs_chunk with one stripe must be
7041 * present, exact stripe count check comes afterwards
7043 len = btrfs_chunk_item_size(1);
7044 if (cur_offset + len > array_size)
7045 goto out_short_read;
7047 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7050 "invalid number of stripes %u in sys_array at offset %u",
7051 num_stripes, cur_offset);
7056 type = btrfs_chunk_type(sb, chunk);
7057 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7059 "invalid chunk type %llu in sys_array at offset %u",
7065 len = btrfs_chunk_item_size(num_stripes);
7066 if (cur_offset + len > array_size)
7067 goto out_short_read;
7069 ret = read_one_chunk(&key, sb, chunk);
7074 "unexpected item type %u in sys_array at offset %u",
7075 (u32)key.type, cur_offset);
7080 sb_array_offset += len;
7083 clear_extent_buffer_uptodate(sb);
7084 free_extent_buffer_stale(sb);
7088 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7090 clear_extent_buffer_uptodate(sb);
7091 free_extent_buffer_stale(sb);
7096 * Check if all chunks in the fs are OK for read-write degraded mount
7098 * If the @failing_dev is specified, it's accounted as missing.
7100 * Return true if all chunks meet the minimal RW mount requirements.
7101 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7103 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7104 struct btrfs_device *failing_dev)
7106 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
7107 struct extent_map *em;
7111 read_lock(&map_tree->map_tree.lock);
7112 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
7113 read_unlock(&map_tree->map_tree.lock);
7114 /* No chunk at all? Return false anyway */
7120 struct map_lookup *map;
7125 map = em->map_lookup;
7127 btrfs_get_num_tolerated_disk_barrier_failures(
7129 for (i = 0; i < map->num_stripes; i++) {
7130 struct btrfs_device *dev = map->stripes[i].dev;
7132 if (!dev || !dev->bdev ||
7133 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7134 dev->last_flush_error)
7136 else if (failing_dev && failing_dev == dev)
7139 if (missing > max_tolerated) {
7142 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7143 em->start, missing, max_tolerated);
7144 free_extent_map(em);
7148 next_start = extent_map_end(em);
7149 free_extent_map(em);
7151 read_lock(&map_tree->map_tree.lock);
7152 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
7153 (u64)(-1) - next_start);
7154 read_unlock(&map_tree->map_tree.lock);
7160 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7162 struct btrfs_root *root = fs_info->chunk_root;
7163 struct btrfs_path *path;
7164 struct extent_buffer *leaf;
7165 struct btrfs_key key;
7166 struct btrfs_key found_key;
7171 path = btrfs_alloc_path();
7176 * uuid_mutex is needed only if we are mounting a sprout FS
7177 * otherwise we don't need it.
7179 mutex_lock(&uuid_mutex);
7180 mutex_lock(&fs_info->chunk_mutex);
7183 * Read all device items, and then all the chunk items. All
7184 * device items are found before any chunk item (their object id
7185 * is smaller than the lowest possible object id for a chunk
7186 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7188 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7195 leaf = path->nodes[0];
7196 slot = path->slots[0];
7197 if (slot >= btrfs_header_nritems(leaf)) {
7198 ret = btrfs_next_leaf(root, path);
7205 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7206 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7207 struct btrfs_dev_item *dev_item;
7208 dev_item = btrfs_item_ptr(leaf, slot,
7209 struct btrfs_dev_item);
7210 ret = read_one_dev(leaf, dev_item);
7214 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7215 struct btrfs_chunk *chunk;
7216 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7217 ret = read_one_chunk(&found_key, leaf, chunk);
7225 * After loading chunk tree, we've got all device information,
7226 * do another round of validation checks.
7228 if (total_dev != fs_info->fs_devices->total_devices) {
7230 "super_num_devices %llu mismatch with num_devices %llu found here",
7231 btrfs_super_num_devices(fs_info->super_copy),
7236 if (btrfs_super_total_bytes(fs_info->super_copy) <
7237 fs_info->fs_devices->total_rw_bytes) {
7239 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7240 btrfs_super_total_bytes(fs_info->super_copy),
7241 fs_info->fs_devices->total_rw_bytes);
7247 mutex_unlock(&fs_info->chunk_mutex);
7248 mutex_unlock(&uuid_mutex);
7250 btrfs_free_path(path);
7254 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7256 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7257 struct btrfs_device *device;
7259 while (fs_devices) {
7260 mutex_lock(&fs_devices->device_list_mutex);
7261 list_for_each_entry(device, &fs_devices->devices, dev_list)
7262 device->fs_info = fs_info;
7263 mutex_unlock(&fs_devices->device_list_mutex);
7265 fs_devices = fs_devices->seed;
7269 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7273 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7274 btrfs_dev_stat_reset(dev, i);
7277 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7279 struct btrfs_key key;
7280 struct btrfs_key found_key;
7281 struct btrfs_root *dev_root = fs_info->dev_root;
7282 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7283 struct extent_buffer *eb;
7286 struct btrfs_device *device;
7287 struct btrfs_path *path = NULL;
7290 path = btrfs_alloc_path();
7296 mutex_lock(&fs_devices->device_list_mutex);
7297 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7299 struct btrfs_dev_stats_item *ptr;
7301 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7302 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7303 key.offset = device->devid;
7304 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7306 __btrfs_reset_dev_stats(device);
7307 device->dev_stats_valid = 1;
7308 btrfs_release_path(path);
7311 slot = path->slots[0];
7312 eb = path->nodes[0];
7313 btrfs_item_key_to_cpu(eb, &found_key, slot);
7314 item_size = btrfs_item_size_nr(eb, slot);
7316 ptr = btrfs_item_ptr(eb, slot,
7317 struct btrfs_dev_stats_item);
7319 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7320 if (item_size >= (1 + i) * sizeof(__le64))
7321 btrfs_dev_stat_set(device, i,
7322 btrfs_dev_stats_value(eb, ptr, i));
7324 btrfs_dev_stat_reset(device, i);
7327 device->dev_stats_valid = 1;
7328 btrfs_dev_stat_print_on_load(device);
7329 btrfs_release_path(path);
7331 mutex_unlock(&fs_devices->device_list_mutex);
7334 btrfs_free_path(path);
7335 return ret < 0 ? ret : 0;
7338 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7339 struct btrfs_device *device)
7341 struct btrfs_fs_info *fs_info = trans->fs_info;
7342 struct btrfs_root *dev_root = fs_info->dev_root;
7343 struct btrfs_path *path;
7344 struct btrfs_key key;
7345 struct extent_buffer *eb;
7346 struct btrfs_dev_stats_item *ptr;
7350 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7351 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7352 key.offset = device->devid;
7354 path = btrfs_alloc_path();
7357 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7359 btrfs_warn_in_rcu(fs_info,
7360 "error %d while searching for dev_stats item for device %s",
7361 ret, rcu_str_deref(device->name));
7366 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7367 /* need to delete old one and insert a new one */
7368 ret = btrfs_del_item(trans, dev_root, path);
7370 btrfs_warn_in_rcu(fs_info,
7371 "delete too small dev_stats item for device %s failed %d",
7372 rcu_str_deref(device->name), ret);
7379 /* need to insert a new item */
7380 btrfs_release_path(path);
7381 ret = btrfs_insert_empty_item(trans, dev_root, path,
7382 &key, sizeof(*ptr));
7384 btrfs_warn_in_rcu(fs_info,
7385 "insert dev_stats item for device %s failed %d",
7386 rcu_str_deref(device->name), ret);
7391 eb = path->nodes[0];
7392 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7393 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7394 btrfs_set_dev_stats_value(eb, ptr, i,
7395 btrfs_dev_stat_read(device, i));
7396 btrfs_mark_buffer_dirty(eb);
7399 btrfs_free_path(path);
7404 * called from commit_transaction. Writes all changed device stats to disk.
7406 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7408 struct btrfs_fs_info *fs_info = trans->fs_info;
7409 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7410 struct btrfs_device *device;
7414 mutex_lock(&fs_devices->device_list_mutex);
7415 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7416 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7417 if (!device->dev_stats_valid || stats_cnt == 0)
7422 * There is a LOAD-LOAD control dependency between the value of
7423 * dev_stats_ccnt and updating the on-disk values which requires
7424 * reading the in-memory counters. Such control dependencies
7425 * require explicit read memory barriers.
7427 * This memory barriers pairs with smp_mb__before_atomic in
7428 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7429 * barrier implied by atomic_xchg in
7430 * btrfs_dev_stats_read_and_reset
7434 ret = update_dev_stat_item(trans, device);
7436 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7438 mutex_unlock(&fs_devices->device_list_mutex);
7443 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7445 btrfs_dev_stat_inc(dev, index);
7446 btrfs_dev_stat_print_on_error(dev);
7449 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7451 if (!dev->dev_stats_valid)
7453 btrfs_err_rl_in_rcu(dev->fs_info,
7454 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7455 rcu_str_deref(dev->name),
7456 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7457 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7458 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7459 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7460 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7463 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7467 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7468 if (btrfs_dev_stat_read(dev, i) != 0)
7470 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7471 return; /* all values == 0, suppress message */
7473 btrfs_info_in_rcu(dev->fs_info,
7474 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7475 rcu_str_deref(dev->name),
7476 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7477 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7478 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7479 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7480 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7483 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7484 struct btrfs_ioctl_get_dev_stats *stats)
7486 struct btrfs_device *dev;
7487 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7490 mutex_lock(&fs_devices->device_list_mutex);
7491 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7493 mutex_unlock(&fs_devices->device_list_mutex);
7496 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7498 } else if (!dev->dev_stats_valid) {
7499 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7501 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7502 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7503 if (stats->nr_items > i)
7505 btrfs_dev_stat_read_and_reset(dev, i);
7507 btrfs_dev_stat_reset(dev, i);
7510 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7511 if (stats->nr_items > i)
7512 stats->values[i] = btrfs_dev_stat_read(dev, i);
7514 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7515 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7519 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7521 struct buffer_head *bh;
7522 struct btrfs_super_block *disk_super;
7528 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7531 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7534 disk_super = (struct btrfs_super_block *)bh->b_data;
7536 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7537 set_buffer_dirty(bh);
7538 sync_dirty_buffer(bh);
7542 /* Notify udev that device has changed */
7543 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7545 /* Update ctime/mtime for device path for libblkid */
7546 update_dev_time(device_path);
7550 * Update the size and bytes used for each device where it changed. This is
7551 * delayed since we would otherwise get errors while writing out the
7554 * Must be invoked during transaction commit.
7556 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7558 struct btrfs_device *curr, *next;
7560 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7562 if (list_empty(&trans->dev_update_list))
7566 * We don't need the device_list_mutex here. This list is owned by the
7567 * transaction and the transaction must complete before the device is
7570 mutex_lock(&trans->fs_info->chunk_mutex);
7571 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7573 list_del_init(&curr->post_commit_list);
7574 curr->commit_total_bytes = curr->disk_total_bytes;
7575 curr->commit_bytes_used = curr->bytes_used;
7577 mutex_unlock(&trans->fs_info->chunk_mutex);
7580 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7582 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7583 while (fs_devices) {
7584 fs_devices->fs_info = fs_info;
7585 fs_devices = fs_devices->seed;
7589 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7591 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7592 while (fs_devices) {
7593 fs_devices->fs_info = NULL;
7594 fs_devices = fs_devices->seed;
7599 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7601 int btrfs_bg_type_to_factor(u64 flags)
7603 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
7604 BTRFS_BLOCK_GROUP_RAID10))
7611 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7612 u64 chunk_offset, u64 devid,
7613 u64 physical_offset, u64 physical_len)
7615 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7616 struct extent_map *em;
7617 struct map_lookup *map;
7618 struct btrfs_device *dev;
7624 read_lock(&em_tree->lock);
7625 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7626 read_unlock(&em_tree->lock);
7630 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7631 physical_offset, devid);
7636 map = em->map_lookup;
7637 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7638 if (physical_len != stripe_len) {
7640 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7641 physical_offset, devid, em->start, physical_len,
7647 for (i = 0; i < map->num_stripes; i++) {
7648 if (map->stripes[i].dev->devid == devid &&
7649 map->stripes[i].physical == physical_offset) {
7651 if (map->verified_stripes >= map->num_stripes) {
7653 "too many dev extents for chunk %llu found",
7658 map->verified_stripes++;
7664 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7665 physical_offset, devid);
7669 /* Make sure no dev extent is beyond device bondary */
7670 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7672 btrfs_err(fs_info, "failed to find devid %llu", devid);
7677 /* It's possible this device is a dummy for seed device */
7678 if (dev->disk_total_bytes == 0) {
7679 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7682 btrfs_err(fs_info, "failed to find seed devid %llu",
7689 if (physical_offset + physical_len > dev->disk_total_bytes) {
7691 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7692 devid, physical_offset, physical_len,
7693 dev->disk_total_bytes);
7698 free_extent_map(em);
7702 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7704 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
7705 struct extent_map *em;
7706 struct rb_node *node;
7709 read_lock(&em_tree->lock);
7710 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7711 em = rb_entry(node, struct extent_map, rb_node);
7712 if (em->map_lookup->num_stripes !=
7713 em->map_lookup->verified_stripes) {
7715 "chunk %llu has missing dev extent, have %d expect %d",
7716 em->start, em->map_lookup->verified_stripes,
7717 em->map_lookup->num_stripes);
7723 read_unlock(&em_tree->lock);
7728 * Ensure that all dev extents are mapped to correct chunk, otherwise
7729 * later chunk allocation/free would cause unexpected behavior.
7731 * NOTE: This will iterate through the whole device tree, which should be of
7732 * the same size level as the chunk tree. This slightly increases mount time.
7734 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7736 struct btrfs_path *path;
7737 struct btrfs_root *root = fs_info->dev_root;
7738 struct btrfs_key key;
7740 u64 prev_dev_ext_end = 0;
7744 key.type = BTRFS_DEV_EXTENT_KEY;
7747 path = btrfs_alloc_path();
7751 path->reada = READA_FORWARD;
7752 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7756 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7757 ret = btrfs_next_item(root, path);
7760 /* No dev extents at all? Not good */
7767 struct extent_buffer *leaf = path->nodes[0];
7768 struct btrfs_dev_extent *dext;
7769 int slot = path->slots[0];
7771 u64 physical_offset;
7775 btrfs_item_key_to_cpu(leaf, &key, slot);
7776 if (key.type != BTRFS_DEV_EXTENT_KEY)
7778 devid = key.objectid;
7779 physical_offset = key.offset;
7781 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7782 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7783 physical_len = btrfs_dev_extent_length(leaf, dext);
7785 /* Check if this dev extent overlaps with the previous one */
7786 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7788 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7789 devid, physical_offset, prev_dev_ext_end);
7794 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7795 physical_offset, physical_len);
7799 prev_dev_ext_end = physical_offset + physical_len;
7801 ret = btrfs_next_item(root, path);
7810 /* Ensure all chunks have corresponding dev extents */
7811 ret = verify_chunk_dev_extent_mapping(fs_info);
7813 btrfs_free_path(path);
7818 * Check whether the given block group or device is pinned by any inode being
7819 * used as a swapfile.
7821 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7823 struct btrfs_swapfile_pin *sp;
7824 struct rb_node *node;
7826 spin_lock(&fs_info->swapfile_pins_lock);
7827 node = fs_info->swapfile_pins.rb_node;
7829 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7831 node = node->rb_left;
7832 else if (ptr > sp->ptr)
7833 node = node->rb_right;
7837 spin_unlock(&fs_info->swapfile_pins_lock);
7838 return node != NULL;