2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 static void lock_chunks(struct btrfs_root *root)
58 mutex_lock(&root->fs_info->chunk_mutex);
61 static void unlock_chunks(struct btrfs_root *root)
63 mutex_unlock(&root->fs_info->chunk_mutex);
66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
68 struct btrfs_fs_devices *fs_devs;
70 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
72 return ERR_PTR(-ENOMEM);
74 mutex_init(&fs_devs->device_list_mutex);
76 INIT_LIST_HEAD(&fs_devs->devices);
77 INIT_LIST_HEAD(&fs_devs->alloc_list);
78 INIT_LIST_HEAD(&fs_devs->list);
84 * alloc_fs_devices - allocate struct btrfs_fs_devices
85 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
88 * Return: a pointer to a new &struct btrfs_fs_devices on success;
89 * ERR_PTR() on error. Returned struct is not linked onto any lists and
90 * can be destroyed with kfree() right away.
92 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
94 struct btrfs_fs_devices *fs_devs;
96 fs_devs = __alloc_fs_devices();
101 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
103 generate_random_uuid(fs_devs->fsid);
108 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
110 struct btrfs_device *device;
111 WARN_ON(fs_devices->opened);
112 while (!list_empty(&fs_devices->devices)) {
113 device = list_entry(fs_devices->devices.next,
114 struct btrfs_device, dev_list);
115 list_del(&device->dev_list);
116 rcu_string_free(device->name);
122 static void btrfs_kobject_uevent(struct block_device *bdev,
123 enum kobject_action action)
127 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
129 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
131 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
132 &disk_to_dev(bdev->bd_disk)->kobj);
135 void btrfs_cleanup_fs_uuids(void)
137 struct btrfs_fs_devices *fs_devices;
139 while (!list_empty(&fs_uuids)) {
140 fs_devices = list_entry(fs_uuids.next,
141 struct btrfs_fs_devices, list);
142 list_del(&fs_devices->list);
143 free_fs_devices(fs_devices);
147 static struct btrfs_device *__alloc_device(void)
149 struct btrfs_device *dev;
151 dev = kzalloc(sizeof(*dev), GFP_NOFS);
153 return ERR_PTR(-ENOMEM);
155 INIT_LIST_HEAD(&dev->dev_list);
156 INIT_LIST_HEAD(&dev->dev_alloc_list);
158 spin_lock_init(&dev->io_lock);
160 spin_lock_init(&dev->reada_lock);
161 atomic_set(&dev->reada_in_flight, 0);
162 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
163 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
168 static noinline struct btrfs_device *__find_device(struct list_head *head,
171 struct btrfs_device *dev;
173 list_for_each_entry(dev, head, dev_list) {
174 if (dev->devid == devid &&
175 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
182 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
184 struct btrfs_fs_devices *fs_devices;
186 list_for_each_entry(fs_devices, &fs_uuids, list) {
187 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
194 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
195 int flush, struct block_device **bdev,
196 struct buffer_head **bh)
200 *bdev = blkdev_get_by_path(device_path, flags, holder);
203 ret = PTR_ERR(*bdev);
204 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
209 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
210 ret = set_blocksize(*bdev, 4096);
212 blkdev_put(*bdev, flags);
215 invalidate_bdev(*bdev);
216 *bh = btrfs_read_dev_super(*bdev);
219 blkdev_put(*bdev, flags);
231 static void requeue_list(struct btrfs_pending_bios *pending_bios,
232 struct bio *head, struct bio *tail)
235 struct bio *old_head;
237 old_head = pending_bios->head;
238 pending_bios->head = head;
239 if (pending_bios->tail)
240 tail->bi_next = old_head;
242 pending_bios->tail = tail;
246 * we try to collect pending bios for a device so we don't get a large
247 * number of procs sending bios down to the same device. This greatly
248 * improves the schedulers ability to collect and merge the bios.
250 * But, it also turns into a long list of bios to process and that is sure
251 * to eventually make the worker thread block. The solution here is to
252 * make some progress and then put this work struct back at the end of
253 * the list if the block device is congested. This way, multiple devices
254 * can make progress from a single worker thread.
256 static noinline void run_scheduled_bios(struct btrfs_device *device)
259 struct backing_dev_info *bdi;
260 struct btrfs_fs_info *fs_info;
261 struct btrfs_pending_bios *pending_bios;
265 unsigned long num_run;
266 unsigned long batch_run = 0;
268 unsigned long last_waited = 0;
270 int sync_pending = 0;
271 struct blk_plug plug;
274 * this function runs all the bios we've collected for
275 * a particular device. We don't want to wander off to
276 * another device without first sending all of these down.
277 * So, setup a plug here and finish it off before we return
279 blk_start_plug(&plug);
281 bdi = blk_get_backing_dev_info(device->bdev);
282 fs_info = device->dev_root->fs_info;
283 limit = btrfs_async_submit_limit(fs_info);
284 limit = limit * 2 / 3;
287 spin_lock(&device->io_lock);
292 /* take all the bios off the list at once and process them
293 * later on (without the lock held). But, remember the
294 * tail and other pointers so the bios can be properly reinserted
295 * into the list if we hit congestion
297 if (!force_reg && device->pending_sync_bios.head) {
298 pending_bios = &device->pending_sync_bios;
301 pending_bios = &device->pending_bios;
305 pending = pending_bios->head;
306 tail = pending_bios->tail;
307 WARN_ON(pending && !tail);
310 * if pending was null this time around, no bios need processing
311 * at all and we can stop. Otherwise it'll loop back up again
312 * and do an additional check so no bios are missed.
314 * device->running_pending is used to synchronize with the
317 if (device->pending_sync_bios.head == NULL &&
318 device->pending_bios.head == NULL) {
320 device->running_pending = 0;
323 device->running_pending = 1;
326 pending_bios->head = NULL;
327 pending_bios->tail = NULL;
329 spin_unlock(&device->io_lock);
334 /* we want to work on both lists, but do more bios on the
335 * sync list than the regular list
338 pending_bios != &device->pending_sync_bios &&
339 device->pending_sync_bios.head) ||
340 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
341 device->pending_bios.head)) {
342 spin_lock(&device->io_lock);
343 requeue_list(pending_bios, pending, tail);
348 pending = pending->bi_next;
351 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
352 waitqueue_active(&fs_info->async_submit_wait))
353 wake_up(&fs_info->async_submit_wait);
355 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
358 * if we're doing the sync list, record that our
359 * plug has some sync requests on it
361 * If we're doing the regular list and there are
362 * sync requests sitting around, unplug before
365 if (pending_bios == &device->pending_sync_bios) {
367 } else if (sync_pending) {
368 blk_finish_plug(&plug);
369 blk_start_plug(&plug);
373 btrfsic_submit_bio(cur->bi_rw, cur);
380 * we made progress, there is more work to do and the bdi
381 * is now congested. Back off and let other work structs
384 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
385 fs_info->fs_devices->open_devices > 1) {
386 struct io_context *ioc;
388 ioc = current->io_context;
391 * the main goal here is that we don't want to
392 * block if we're going to be able to submit
393 * more requests without blocking.
395 * This code does two great things, it pokes into
396 * the elevator code from a filesystem _and_
397 * it makes assumptions about how batching works.
399 if (ioc && ioc->nr_batch_requests > 0 &&
400 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
402 ioc->last_waited == last_waited)) {
404 * we want to go through our batch of
405 * requests and stop. So, we copy out
406 * the ioc->last_waited time and test
407 * against it before looping
409 last_waited = ioc->last_waited;
414 spin_lock(&device->io_lock);
415 requeue_list(pending_bios, pending, tail);
416 device->running_pending = 1;
418 spin_unlock(&device->io_lock);
419 btrfs_queue_work(fs_info->submit_workers,
423 /* unplug every 64 requests just for good measure */
424 if (batch_run % 64 == 0) {
425 blk_finish_plug(&plug);
426 blk_start_plug(&plug);
435 spin_lock(&device->io_lock);
436 if (device->pending_bios.head || device->pending_sync_bios.head)
438 spin_unlock(&device->io_lock);
441 blk_finish_plug(&plug);
444 static void pending_bios_fn(struct btrfs_work *work)
446 struct btrfs_device *device;
448 device = container_of(work, struct btrfs_device, work);
449 run_scheduled_bios(device);
453 * Add new device to list of registered devices
456 * 1 - first time device is seen
457 * 0 - device already known
460 static noinline int device_list_add(const char *path,
461 struct btrfs_super_block *disk_super,
462 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
464 struct btrfs_device *device;
465 struct btrfs_fs_devices *fs_devices;
466 struct rcu_string *name;
468 u64 found_transid = btrfs_super_generation(disk_super);
470 fs_devices = find_fsid(disk_super->fsid);
472 fs_devices = alloc_fs_devices(disk_super->fsid);
473 if (IS_ERR(fs_devices))
474 return PTR_ERR(fs_devices);
476 list_add(&fs_devices->list, &fs_uuids);
477 fs_devices->latest_devid = devid;
478 fs_devices->latest_trans = found_transid;
482 device = __find_device(&fs_devices->devices, devid,
483 disk_super->dev_item.uuid);
486 if (fs_devices->opened)
489 device = btrfs_alloc_device(NULL, &devid,
490 disk_super->dev_item.uuid);
491 if (IS_ERR(device)) {
492 /* we can safely leave the fs_devices entry around */
493 return PTR_ERR(device);
496 name = rcu_string_strdup(path, GFP_NOFS);
501 rcu_assign_pointer(device->name, name);
503 mutex_lock(&fs_devices->device_list_mutex);
504 list_add_rcu(&device->dev_list, &fs_devices->devices);
505 fs_devices->num_devices++;
506 mutex_unlock(&fs_devices->device_list_mutex);
509 device->fs_devices = fs_devices;
510 } else if (!device->name || strcmp(device->name->str, path)) {
512 * When FS is already mounted.
513 * 1. If you are here and if the device->name is NULL that
514 * means this device was missing at time of FS mount.
515 * 2. If you are here and if the device->name is different
516 * from 'path' that means either
517 * a. The same device disappeared and reappeared with
519 * b. The missing-disk-which-was-replaced, has
522 * We must allow 1 and 2a above. But 2b would be a spurious
525 * Further in case of 1 and 2a above, the disk at 'path'
526 * would have missed some transaction when it was away and
527 * in case of 2a the stale bdev has to be updated as well.
528 * 2b must not be allowed at all time.
532 * As of now don't allow update to btrfs_fs_device through
533 * the btrfs dev scan cli, after FS has been mounted.
535 if (fs_devices->opened) {
539 * That is if the FS is _not_ mounted and if you
540 * are here, that means there is more than one
541 * disk with same uuid and devid.We keep the one
542 * with larger generation number or the last-in if
543 * generation are equal.
545 if (found_transid < device->generation)
549 name = rcu_string_strdup(path, GFP_NOFS);
552 rcu_string_free(device->name);
553 rcu_assign_pointer(device->name, name);
554 if (device->missing) {
555 fs_devices->missing_devices--;
561 * Unmount does not free the btrfs_device struct but would zero
562 * generation along with most of the other members. So just update
563 * it back. We need it to pick the disk with largest generation
566 if (!fs_devices->opened)
567 device->generation = found_transid;
569 if (found_transid > fs_devices->latest_trans) {
570 fs_devices->latest_devid = devid;
571 fs_devices->latest_trans = found_transid;
573 *fs_devices_ret = fs_devices;
578 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
580 struct btrfs_fs_devices *fs_devices;
581 struct btrfs_device *device;
582 struct btrfs_device *orig_dev;
584 fs_devices = alloc_fs_devices(orig->fsid);
585 if (IS_ERR(fs_devices))
588 fs_devices->latest_devid = orig->latest_devid;
589 fs_devices->latest_trans = orig->latest_trans;
590 fs_devices->total_devices = orig->total_devices;
592 /* We have held the volume lock, it is safe to get the devices. */
593 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
594 struct rcu_string *name;
596 device = btrfs_alloc_device(NULL, &orig_dev->devid,
602 * This is ok to do without rcu read locked because we hold the
603 * uuid mutex so nothing we touch in here is going to disappear.
605 if (orig_dev->name) {
606 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
611 rcu_assign_pointer(device->name, name);
614 list_add(&device->dev_list, &fs_devices->devices);
615 device->fs_devices = fs_devices;
616 fs_devices->num_devices++;
620 free_fs_devices(fs_devices);
621 return ERR_PTR(-ENOMEM);
624 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
625 struct btrfs_fs_devices *fs_devices, int step)
627 struct btrfs_device *device, *next;
629 struct block_device *latest_bdev = NULL;
630 u64 latest_devid = 0;
631 u64 latest_transid = 0;
633 mutex_lock(&uuid_mutex);
635 /* This is the initialized path, it is safe to release the devices. */
636 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
637 if (device->in_fs_metadata) {
638 if (!device->is_tgtdev_for_dev_replace &&
640 device->generation > latest_transid)) {
641 latest_devid = device->devid;
642 latest_transid = device->generation;
643 latest_bdev = device->bdev;
648 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
650 * In the first step, keep the device which has
651 * the correct fsid and the devid that is used
652 * for the dev_replace procedure.
653 * In the second step, the dev_replace state is
654 * read from the device tree and it is known
655 * whether the procedure is really active or
656 * not, which means whether this device is
657 * used or whether it should be removed.
659 if (step == 0 || device->is_tgtdev_for_dev_replace) {
664 blkdev_put(device->bdev, device->mode);
666 fs_devices->open_devices--;
668 if (device->writeable) {
669 list_del_init(&device->dev_alloc_list);
670 device->writeable = 0;
671 if (!device->is_tgtdev_for_dev_replace)
672 fs_devices->rw_devices--;
674 list_del_init(&device->dev_list);
675 fs_devices->num_devices--;
676 rcu_string_free(device->name);
680 if (fs_devices->seed) {
681 fs_devices = fs_devices->seed;
685 fs_devices->latest_bdev = latest_bdev;
686 fs_devices->latest_devid = latest_devid;
687 fs_devices->latest_trans = latest_transid;
689 mutex_unlock(&uuid_mutex);
692 static void __free_device(struct work_struct *work)
694 struct btrfs_device *device;
696 device = container_of(work, struct btrfs_device, rcu_work);
699 blkdev_put(device->bdev, device->mode);
701 rcu_string_free(device->name);
705 static void free_device(struct rcu_head *head)
707 struct btrfs_device *device;
709 device = container_of(head, struct btrfs_device, rcu);
711 INIT_WORK(&device->rcu_work, __free_device);
712 schedule_work(&device->rcu_work);
715 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
717 struct btrfs_device *device;
719 if (--fs_devices->opened > 0)
722 mutex_lock(&fs_devices->device_list_mutex);
723 list_for_each_entry(device, &fs_devices->devices, dev_list) {
724 struct btrfs_device *new_device;
725 struct rcu_string *name;
728 fs_devices->open_devices--;
730 if (device->writeable &&
731 device->devid != BTRFS_DEV_REPLACE_DEVID) {
732 list_del_init(&device->dev_alloc_list);
733 fs_devices->rw_devices--;
736 if (device->can_discard)
737 fs_devices->num_can_discard--;
739 fs_devices->missing_devices--;
741 new_device = btrfs_alloc_device(NULL, &device->devid,
743 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
745 /* Safe because we are under uuid_mutex */
747 name = rcu_string_strdup(device->name->str, GFP_NOFS);
748 BUG_ON(!name); /* -ENOMEM */
749 rcu_assign_pointer(new_device->name, name);
752 list_replace_rcu(&device->dev_list, &new_device->dev_list);
753 new_device->fs_devices = device->fs_devices;
755 call_rcu(&device->rcu, free_device);
757 mutex_unlock(&fs_devices->device_list_mutex);
759 WARN_ON(fs_devices->open_devices);
760 WARN_ON(fs_devices->rw_devices);
761 fs_devices->opened = 0;
762 fs_devices->seeding = 0;
767 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
769 struct btrfs_fs_devices *seed_devices = NULL;
772 mutex_lock(&uuid_mutex);
773 ret = __btrfs_close_devices(fs_devices);
774 if (!fs_devices->opened) {
775 seed_devices = fs_devices->seed;
776 fs_devices->seed = NULL;
778 mutex_unlock(&uuid_mutex);
780 while (seed_devices) {
781 fs_devices = seed_devices;
782 seed_devices = fs_devices->seed;
783 __btrfs_close_devices(fs_devices);
784 free_fs_devices(fs_devices);
787 * Wait for rcu kworkers under __btrfs_close_devices
788 * to finish all blkdev_puts so device is really
789 * free when umount is done.
795 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
796 fmode_t flags, void *holder)
798 struct request_queue *q;
799 struct block_device *bdev;
800 struct list_head *head = &fs_devices->devices;
801 struct btrfs_device *device;
802 struct block_device *latest_bdev = NULL;
803 struct buffer_head *bh;
804 struct btrfs_super_block *disk_super;
805 u64 latest_devid = 0;
806 u64 latest_transid = 0;
813 list_for_each_entry(device, head, dev_list) {
819 /* Just open everything we can; ignore failures here */
820 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
824 disk_super = (struct btrfs_super_block *)bh->b_data;
825 devid = btrfs_stack_device_id(&disk_super->dev_item);
826 if (devid != device->devid)
829 if (memcmp(device->uuid, disk_super->dev_item.uuid,
833 device->generation = btrfs_super_generation(disk_super);
834 if (!latest_transid || device->generation > latest_transid) {
835 latest_devid = devid;
836 latest_transid = device->generation;
840 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
841 device->writeable = 0;
843 device->writeable = !bdev_read_only(bdev);
847 q = bdev_get_queue(bdev);
848 if (blk_queue_discard(q)) {
849 device->can_discard = 1;
850 fs_devices->num_can_discard++;
854 device->in_fs_metadata = 0;
855 device->mode = flags;
857 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
858 fs_devices->rotating = 1;
860 fs_devices->open_devices++;
861 if (device->writeable &&
862 device->devid != BTRFS_DEV_REPLACE_DEVID) {
863 fs_devices->rw_devices++;
864 list_add(&device->dev_alloc_list,
865 &fs_devices->alloc_list);
872 blkdev_put(bdev, flags);
875 if (fs_devices->open_devices == 0) {
879 fs_devices->seeding = seeding;
880 fs_devices->opened = 1;
881 fs_devices->latest_bdev = latest_bdev;
882 fs_devices->latest_devid = latest_devid;
883 fs_devices->latest_trans = latest_transid;
884 fs_devices->total_rw_bytes = 0;
889 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
890 fmode_t flags, void *holder)
894 mutex_lock(&uuid_mutex);
895 if (fs_devices->opened) {
896 fs_devices->opened++;
899 ret = __btrfs_open_devices(fs_devices, flags, holder);
901 mutex_unlock(&uuid_mutex);
906 * Look for a btrfs signature on a device. This may be called out of the mount path
907 * and we are not allowed to call set_blocksize during the scan. The superblock
908 * is read via pagecache
910 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
911 struct btrfs_fs_devices **fs_devices_ret)
913 struct btrfs_super_block *disk_super;
914 struct block_device *bdev;
925 * we would like to check all the supers, but that would make
926 * a btrfs mount succeed after a mkfs from a different FS.
927 * So, we need to add a special mount option to scan for
928 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
930 bytenr = btrfs_sb_offset(0);
932 mutex_lock(&uuid_mutex);
934 bdev = blkdev_get_by_path(path, flags, holder);
941 /* make sure our super fits in the device */
942 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
945 /* make sure our super fits in the page */
946 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
949 /* make sure our super doesn't straddle pages on disk */
950 index = bytenr >> PAGE_CACHE_SHIFT;
951 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
954 /* pull in the page with our super */
955 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
958 if (IS_ERR_OR_NULL(page))
963 /* align our pointer to the offset of the super block */
964 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
966 if (btrfs_super_bytenr(disk_super) != bytenr ||
967 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
970 devid = btrfs_stack_device_id(&disk_super->dev_item);
971 transid = btrfs_super_generation(disk_super);
972 total_devices = btrfs_super_num_devices(disk_super);
974 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
976 if (disk_super->label[0]) {
977 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
978 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
979 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
981 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
984 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
987 if (!ret && fs_devices_ret)
988 (*fs_devices_ret)->total_devices = total_devices;
992 page_cache_release(page);
995 blkdev_put(bdev, flags);
997 mutex_unlock(&uuid_mutex);
1001 /* helper to account the used device space in the range */
1002 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1003 u64 end, u64 *length)
1005 struct btrfs_key key;
1006 struct btrfs_root *root = device->dev_root;
1007 struct btrfs_dev_extent *dev_extent;
1008 struct btrfs_path *path;
1012 struct extent_buffer *l;
1016 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1019 path = btrfs_alloc_path();
1024 key.objectid = device->devid;
1026 key.type = BTRFS_DEV_EXTENT_KEY;
1028 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1032 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1039 slot = path->slots[0];
1040 if (slot >= btrfs_header_nritems(l)) {
1041 ret = btrfs_next_leaf(root, path);
1049 btrfs_item_key_to_cpu(l, &key, slot);
1051 if (key.objectid < device->devid)
1054 if (key.objectid > device->devid)
1057 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1060 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1061 extent_end = key.offset + btrfs_dev_extent_length(l,
1063 if (key.offset <= start && extent_end > end) {
1064 *length = end - start + 1;
1066 } else if (key.offset <= start && extent_end > start)
1067 *length += extent_end - start;
1068 else if (key.offset > start && extent_end <= end)
1069 *length += extent_end - key.offset;
1070 else if (key.offset > start && key.offset <= end) {
1071 *length += end - key.offset + 1;
1073 } else if (key.offset > end)
1081 btrfs_free_path(path);
1085 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1086 struct btrfs_device *device,
1087 u64 *start, u64 len)
1089 struct extent_map *em;
1092 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1093 struct map_lookup *map;
1096 map = (struct map_lookup *)em->bdev;
1097 for (i = 0; i < map->num_stripes; i++) {
1098 if (map->stripes[i].dev != device)
1100 if (map->stripes[i].physical >= *start + len ||
1101 map->stripes[i].physical + em->orig_block_len <=
1104 *start = map->stripes[i].physical +
1115 * find_free_dev_extent - find free space in the specified device
1116 * @device: the device which we search the free space in
1117 * @num_bytes: the size of the free space that we need
1118 * @start: store the start of the free space.
1119 * @len: the size of the free space. that we find, or the size of the max
1120 * free space if we don't find suitable free space
1122 * this uses a pretty simple search, the expectation is that it is
1123 * called very infrequently and that a given device has a small number
1126 * @start is used to store the start of the free space if we find. But if we
1127 * don't find suitable free space, it will be used to store the start position
1128 * of the max free space.
1130 * @len is used to store the size of the free space that we find.
1131 * But if we don't find suitable free space, it is used to store the size of
1132 * the max free space.
1134 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1135 struct btrfs_device *device, u64 num_bytes,
1136 u64 *start, u64 *len)
1138 struct btrfs_key key;
1139 struct btrfs_root *root = device->dev_root;
1140 struct btrfs_dev_extent *dev_extent;
1141 struct btrfs_path *path;
1147 u64 search_end = device->total_bytes;
1150 struct extent_buffer *l;
1152 /* FIXME use last free of some kind */
1154 /* we don't want to overwrite the superblock on the drive,
1155 * so we make sure to start at an offset of at least 1MB
1157 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1159 path = btrfs_alloc_path();
1163 max_hole_start = search_start;
1167 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1173 path->search_commit_root = 1;
1174 path->skip_locking = 1;
1176 key.objectid = device->devid;
1177 key.offset = search_start;
1178 key.type = BTRFS_DEV_EXTENT_KEY;
1180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1184 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1191 slot = path->slots[0];
1192 if (slot >= btrfs_header_nritems(l)) {
1193 ret = btrfs_next_leaf(root, path);
1201 btrfs_item_key_to_cpu(l, &key, slot);
1203 if (key.objectid < device->devid)
1206 if (key.objectid > device->devid)
1209 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1212 if (key.offset > search_start) {
1213 hole_size = key.offset - search_start;
1216 * Have to check before we set max_hole_start, otherwise
1217 * we could end up sending back this offset anyway.
1219 if (contains_pending_extent(trans, device,
1224 if (hole_size > max_hole_size) {
1225 max_hole_start = search_start;
1226 max_hole_size = hole_size;
1230 * If this free space is greater than which we need,
1231 * it must be the max free space that we have found
1232 * until now, so max_hole_start must point to the start
1233 * of this free space and the length of this free space
1234 * is stored in max_hole_size. Thus, we return
1235 * max_hole_start and max_hole_size and go back to the
1238 if (hole_size >= num_bytes) {
1244 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1245 extent_end = key.offset + btrfs_dev_extent_length(l,
1247 if (extent_end > search_start)
1248 search_start = extent_end;
1255 * At this point, search_start should be the end of
1256 * allocated dev extents, and when shrinking the device,
1257 * search_end may be smaller than search_start.
1259 if (search_end > search_start)
1260 hole_size = search_end - search_start;
1262 if (hole_size > max_hole_size) {
1263 max_hole_start = search_start;
1264 max_hole_size = hole_size;
1267 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1268 btrfs_release_path(path);
1273 if (hole_size < num_bytes)
1279 btrfs_free_path(path);
1280 *start = max_hole_start;
1282 *len = max_hole_size;
1286 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1287 struct btrfs_device *device,
1291 struct btrfs_path *path;
1292 struct btrfs_root *root = device->dev_root;
1293 struct btrfs_key key;
1294 struct btrfs_key found_key;
1295 struct extent_buffer *leaf = NULL;
1296 struct btrfs_dev_extent *extent = NULL;
1298 path = btrfs_alloc_path();
1302 key.objectid = device->devid;
1304 key.type = BTRFS_DEV_EXTENT_KEY;
1306 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1308 ret = btrfs_previous_item(root, path, key.objectid,
1309 BTRFS_DEV_EXTENT_KEY);
1312 leaf = path->nodes[0];
1313 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1314 extent = btrfs_item_ptr(leaf, path->slots[0],
1315 struct btrfs_dev_extent);
1316 BUG_ON(found_key.offset > start || found_key.offset +
1317 btrfs_dev_extent_length(leaf, extent) < start);
1319 btrfs_release_path(path);
1321 } else if (ret == 0) {
1322 leaf = path->nodes[0];
1323 extent = btrfs_item_ptr(leaf, path->slots[0],
1324 struct btrfs_dev_extent);
1326 btrfs_error(root->fs_info, ret, "Slot search failed");
1330 if (device->bytes_used > 0) {
1331 u64 len = btrfs_dev_extent_length(leaf, extent);
1332 device->bytes_used -= len;
1333 spin_lock(&root->fs_info->free_chunk_lock);
1334 root->fs_info->free_chunk_space += len;
1335 spin_unlock(&root->fs_info->free_chunk_lock);
1337 ret = btrfs_del_item(trans, root, path);
1339 btrfs_error(root->fs_info, ret,
1340 "Failed to remove dev extent item");
1343 btrfs_free_path(path);
1347 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1348 struct btrfs_device *device,
1349 u64 chunk_tree, u64 chunk_objectid,
1350 u64 chunk_offset, u64 start, u64 num_bytes)
1353 struct btrfs_path *path;
1354 struct btrfs_root *root = device->dev_root;
1355 struct btrfs_dev_extent *extent;
1356 struct extent_buffer *leaf;
1357 struct btrfs_key key;
1359 WARN_ON(!device->in_fs_metadata);
1360 WARN_ON(device->is_tgtdev_for_dev_replace);
1361 path = btrfs_alloc_path();
1365 key.objectid = device->devid;
1367 key.type = BTRFS_DEV_EXTENT_KEY;
1368 ret = btrfs_insert_empty_item(trans, root, path, &key,
1373 leaf = path->nodes[0];
1374 extent = btrfs_item_ptr(leaf, path->slots[0],
1375 struct btrfs_dev_extent);
1376 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1377 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1378 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1380 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1381 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1383 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1384 btrfs_mark_buffer_dirty(leaf);
1386 btrfs_free_path(path);
1390 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1392 struct extent_map_tree *em_tree;
1393 struct extent_map *em;
1397 em_tree = &fs_info->mapping_tree.map_tree;
1398 read_lock(&em_tree->lock);
1399 n = rb_last(&em_tree->map);
1401 em = rb_entry(n, struct extent_map, rb_node);
1402 ret = em->start + em->len;
1404 read_unlock(&em_tree->lock);
1409 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1413 struct btrfs_key key;
1414 struct btrfs_key found_key;
1415 struct btrfs_path *path;
1417 path = btrfs_alloc_path();
1421 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1422 key.type = BTRFS_DEV_ITEM_KEY;
1423 key.offset = (u64)-1;
1425 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1429 BUG_ON(ret == 0); /* Corruption */
1431 ret = btrfs_previous_item(fs_info->chunk_root, path,
1432 BTRFS_DEV_ITEMS_OBJECTID,
1433 BTRFS_DEV_ITEM_KEY);
1437 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1439 *devid_ret = found_key.offset + 1;
1443 btrfs_free_path(path);
1448 * the device information is stored in the chunk root
1449 * the btrfs_device struct should be fully filled in
1451 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1452 struct btrfs_root *root,
1453 struct btrfs_device *device)
1456 struct btrfs_path *path;
1457 struct btrfs_dev_item *dev_item;
1458 struct extent_buffer *leaf;
1459 struct btrfs_key key;
1462 root = root->fs_info->chunk_root;
1464 path = btrfs_alloc_path();
1468 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1469 key.type = BTRFS_DEV_ITEM_KEY;
1470 key.offset = device->devid;
1472 ret = btrfs_insert_empty_item(trans, root, path, &key,
1477 leaf = path->nodes[0];
1478 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1480 btrfs_set_device_id(leaf, dev_item, device->devid);
1481 btrfs_set_device_generation(leaf, dev_item, 0);
1482 btrfs_set_device_type(leaf, dev_item, device->type);
1483 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1484 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1485 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1486 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1487 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1488 btrfs_set_device_group(leaf, dev_item, 0);
1489 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1490 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1491 btrfs_set_device_start_offset(leaf, dev_item, 0);
1493 ptr = btrfs_device_uuid(dev_item);
1494 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1495 ptr = btrfs_device_fsid(dev_item);
1496 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1497 btrfs_mark_buffer_dirty(leaf);
1501 btrfs_free_path(path);
1506 * Function to update ctime/mtime for a given device path.
1507 * Mainly used for ctime/mtime based probe like libblkid.
1509 static void update_dev_time(char *path_name)
1513 filp = filp_open(path_name, O_RDWR, 0);
1516 file_update_time(filp);
1517 filp_close(filp, NULL);
1521 static int btrfs_rm_dev_item(struct btrfs_root *root,
1522 struct btrfs_device *device)
1525 struct btrfs_path *path;
1526 struct btrfs_key key;
1527 struct btrfs_trans_handle *trans;
1529 root = root->fs_info->chunk_root;
1531 path = btrfs_alloc_path();
1535 trans = btrfs_start_transaction(root, 0);
1536 if (IS_ERR(trans)) {
1537 btrfs_free_path(path);
1538 return PTR_ERR(trans);
1540 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1541 key.type = BTRFS_DEV_ITEM_KEY;
1542 key.offset = device->devid;
1545 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1554 ret = btrfs_del_item(trans, root, path);
1558 btrfs_free_path(path);
1559 unlock_chunks(root);
1560 btrfs_commit_transaction(trans, root);
1564 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1566 struct btrfs_device *device;
1567 struct btrfs_device *next_device;
1568 struct block_device *bdev;
1569 struct buffer_head *bh = NULL;
1570 struct btrfs_super_block *disk_super;
1571 struct btrfs_fs_devices *cur_devices;
1578 bool clear_super = false;
1580 mutex_lock(&uuid_mutex);
1583 seq = read_seqbegin(&root->fs_info->profiles_lock);
1585 all_avail = root->fs_info->avail_data_alloc_bits |
1586 root->fs_info->avail_system_alloc_bits |
1587 root->fs_info->avail_metadata_alloc_bits;
1588 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1590 num_devices = root->fs_info->fs_devices->num_devices;
1591 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1592 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1593 WARN_ON(num_devices < 1);
1596 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1598 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1599 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1603 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1604 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1608 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1609 root->fs_info->fs_devices->rw_devices <= 2) {
1610 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1613 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1614 root->fs_info->fs_devices->rw_devices <= 3) {
1615 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1619 if (strcmp(device_path, "missing") == 0) {
1620 struct list_head *devices;
1621 struct btrfs_device *tmp;
1624 devices = &root->fs_info->fs_devices->devices;
1626 * It is safe to read the devices since the volume_mutex
1629 list_for_each_entry(tmp, devices, dev_list) {
1630 if (tmp->in_fs_metadata &&
1631 !tmp->is_tgtdev_for_dev_replace &&
1641 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1645 ret = btrfs_get_bdev_and_sb(device_path,
1646 FMODE_WRITE | FMODE_EXCL,
1647 root->fs_info->bdev_holder, 0,
1651 disk_super = (struct btrfs_super_block *)bh->b_data;
1652 devid = btrfs_stack_device_id(&disk_super->dev_item);
1653 dev_uuid = disk_super->dev_item.uuid;
1654 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1662 if (device->is_tgtdev_for_dev_replace) {
1663 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1667 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1668 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1672 if (device->writeable) {
1674 list_del_init(&device->dev_alloc_list);
1675 unlock_chunks(root);
1676 root->fs_info->fs_devices->rw_devices--;
1680 mutex_unlock(&uuid_mutex);
1681 ret = btrfs_shrink_device(device, 0);
1682 mutex_lock(&uuid_mutex);
1687 * TODO: the superblock still includes this device in its num_devices
1688 * counter although write_all_supers() is not locked out. This
1689 * could give a filesystem state which requires a degraded mount.
1691 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1695 spin_lock(&root->fs_info->free_chunk_lock);
1696 root->fs_info->free_chunk_space = device->total_bytes -
1698 spin_unlock(&root->fs_info->free_chunk_lock);
1700 device->in_fs_metadata = 0;
1701 btrfs_scrub_cancel_dev(root->fs_info, device);
1704 * the device list mutex makes sure that we don't change
1705 * the device list while someone else is writing out all
1706 * the device supers. Whoever is writing all supers, should
1707 * lock the device list mutex before getting the number of
1708 * devices in the super block (super_copy). Conversely,
1709 * whoever updates the number of devices in the super block
1710 * (super_copy) should hold the device list mutex.
1713 cur_devices = device->fs_devices;
1714 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1715 list_del_rcu(&device->dev_list);
1717 device->fs_devices->num_devices--;
1718 device->fs_devices->total_devices--;
1720 if (device->missing)
1721 device->fs_devices->missing_devices--;
1723 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1724 struct btrfs_device, dev_list);
1725 if (device->bdev == root->fs_info->sb->s_bdev)
1726 root->fs_info->sb->s_bdev = next_device->bdev;
1727 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1728 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1731 device->fs_devices->open_devices--;
1732 /* remove sysfs entry */
1733 btrfs_kobj_rm_device(root->fs_info, device);
1736 call_rcu(&device->rcu, free_device);
1738 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1739 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1740 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1742 if (cur_devices->open_devices == 0) {
1743 struct btrfs_fs_devices *fs_devices;
1744 fs_devices = root->fs_info->fs_devices;
1745 while (fs_devices) {
1746 if (fs_devices->seed == cur_devices) {
1747 fs_devices->seed = cur_devices->seed;
1750 fs_devices = fs_devices->seed;
1752 cur_devices->seed = NULL;
1754 __btrfs_close_devices(cur_devices);
1755 unlock_chunks(root);
1756 free_fs_devices(cur_devices);
1759 root->fs_info->num_tolerated_disk_barrier_failures =
1760 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1763 * at this point, the device is zero sized. We want to
1764 * remove it from the devices list and zero out the old super
1766 if (clear_super && disk_super) {
1770 /* make sure this device isn't detected as part of
1773 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1774 set_buffer_dirty(bh);
1775 sync_dirty_buffer(bh);
1777 /* clear the mirror copies of super block on the disk
1778 * being removed, 0th copy is been taken care above and
1779 * the below would take of the rest
1781 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1782 bytenr = btrfs_sb_offset(i);
1783 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1784 i_size_read(bdev->bd_inode))
1788 bh = __bread(bdev, bytenr / 4096,
1789 BTRFS_SUPER_INFO_SIZE);
1793 disk_super = (struct btrfs_super_block *)bh->b_data;
1795 if (btrfs_super_bytenr(disk_super) != bytenr ||
1796 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1799 memset(&disk_super->magic, 0,
1800 sizeof(disk_super->magic));
1801 set_buffer_dirty(bh);
1802 sync_dirty_buffer(bh);
1809 /* Notify udev that device has changed */
1810 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1812 /* Update ctime/mtime for device path for libblkid */
1813 update_dev_time(device_path);
1819 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1821 mutex_unlock(&uuid_mutex);
1824 if (device->writeable) {
1826 list_add(&device->dev_alloc_list,
1827 &root->fs_info->fs_devices->alloc_list);
1828 unlock_chunks(root);
1829 root->fs_info->fs_devices->rw_devices++;
1834 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1835 struct btrfs_device *srcdev)
1837 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1839 list_del_rcu(&srcdev->dev_list);
1840 list_del_rcu(&srcdev->dev_alloc_list);
1841 fs_info->fs_devices->num_devices--;
1842 if (srcdev->missing) {
1843 fs_info->fs_devices->missing_devices--;
1844 fs_info->fs_devices->rw_devices++;
1846 if (srcdev->can_discard)
1847 fs_info->fs_devices->num_can_discard--;
1849 fs_info->fs_devices->open_devices--;
1852 * zero out the old super if it is not writable
1853 * (e.g. seed device)
1855 if (srcdev->writeable)
1856 btrfs_scratch_superblock(srcdev);
1859 call_rcu(&srcdev->rcu, free_device);
1862 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1863 struct btrfs_device *tgtdev)
1865 struct btrfs_device *next_device;
1868 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1870 btrfs_scratch_superblock(tgtdev);
1871 fs_info->fs_devices->open_devices--;
1873 fs_info->fs_devices->num_devices--;
1874 if (tgtdev->can_discard)
1875 fs_info->fs_devices->num_can_discard++;
1877 next_device = list_entry(fs_info->fs_devices->devices.next,
1878 struct btrfs_device, dev_list);
1879 if (tgtdev->bdev == fs_info->sb->s_bdev)
1880 fs_info->sb->s_bdev = next_device->bdev;
1881 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1882 fs_info->fs_devices->latest_bdev = next_device->bdev;
1883 list_del_rcu(&tgtdev->dev_list);
1885 call_rcu(&tgtdev->rcu, free_device);
1887 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1890 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1891 struct btrfs_device **device)
1894 struct btrfs_super_block *disk_super;
1897 struct block_device *bdev;
1898 struct buffer_head *bh;
1901 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1902 root->fs_info->bdev_holder, 0, &bdev, &bh);
1905 disk_super = (struct btrfs_super_block *)bh->b_data;
1906 devid = btrfs_stack_device_id(&disk_super->dev_item);
1907 dev_uuid = disk_super->dev_item.uuid;
1908 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1913 blkdev_put(bdev, FMODE_READ);
1917 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1919 struct btrfs_device **device)
1922 if (strcmp(device_path, "missing") == 0) {
1923 struct list_head *devices;
1924 struct btrfs_device *tmp;
1926 devices = &root->fs_info->fs_devices->devices;
1928 * It is safe to read the devices since the volume_mutex
1929 * is held by the caller.
1931 list_for_each_entry(tmp, devices, dev_list) {
1932 if (tmp->in_fs_metadata && !tmp->bdev) {
1939 btrfs_err(root->fs_info, "no missing device found");
1945 return btrfs_find_device_by_path(root, device_path, device);
1950 * does all the dirty work required for changing file system's UUID.
1952 static int btrfs_prepare_sprout(struct btrfs_root *root)
1954 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1955 struct btrfs_fs_devices *old_devices;
1956 struct btrfs_fs_devices *seed_devices;
1957 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1958 struct btrfs_device *device;
1961 BUG_ON(!mutex_is_locked(&uuid_mutex));
1962 if (!fs_devices->seeding)
1965 seed_devices = __alloc_fs_devices();
1966 if (IS_ERR(seed_devices))
1967 return PTR_ERR(seed_devices);
1969 old_devices = clone_fs_devices(fs_devices);
1970 if (IS_ERR(old_devices)) {
1971 kfree(seed_devices);
1972 return PTR_ERR(old_devices);
1975 list_add(&old_devices->list, &fs_uuids);
1977 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1978 seed_devices->opened = 1;
1979 INIT_LIST_HEAD(&seed_devices->devices);
1980 INIT_LIST_HEAD(&seed_devices->alloc_list);
1981 mutex_init(&seed_devices->device_list_mutex);
1983 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1984 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1987 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1988 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1989 device->fs_devices = seed_devices;
1992 fs_devices->seeding = 0;
1993 fs_devices->num_devices = 0;
1994 fs_devices->open_devices = 0;
1995 fs_devices->missing_devices = 0;
1996 fs_devices->num_can_discard = 0;
1997 fs_devices->rotating = 0;
1998 fs_devices->seed = seed_devices;
2000 generate_random_uuid(fs_devices->fsid);
2001 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2002 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2003 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2005 super_flags = btrfs_super_flags(disk_super) &
2006 ~BTRFS_SUPER_FLAG_SEEDING;
2007 btrfs_set_super_flags(disk_super, super_flags);
2013 * strore the expected generation for seed devices in device items.
2015 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2016 struct btrfs_root *root)
2018 struct btrfs_path *path;
2019 struct extent_buffer *leaf;
2020 struct btrfs_dev_item *dev_item;
2021 struct btrfs_device *device;
2022 struct btrfs_key key;
2023 u8 fs_uuid[BTRFS_UUID_SIZE];
2024 u8 dev_uuid[BTRFS_UUID_SIZE];
2028 path = btrfs_alloc_path();
2032 root = root->fs_info->chunk_root;
2033 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2035 key.type = BTRFS_DEV_ITEM_KEY;
2038 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2042 leaf = path->nodes[0];
2044 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2045 ret = btrfs_next_leaf(root, path);
2050 leaf = path->nodes[0];
2051 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2052 btrfs_release_path(path);
2056 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2057 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2058 key.type != BTRFS_DEV_ITEM_KEY)
2061 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2062 struct btrfs_dev_item);
2063 devid = btrfs_device_id(leaf, dev_item);
2064 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2066 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2068 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2070 BUG_ON(!device); /* Logic error */
2072 if (device->fs_devices->seeding) {
2073 btrfs_set_device_generation(leaf, dev_item,
2074 device->generation);
2075 btrfs_mark_buffer_dirty(leaf);
2083 btrfs_free_path(path);
2087 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2089 struct request_queue *q;
2090 struct btrfs_trans_handle *trans;
2091 struct btrfs_device *device;
2092 struct block_device *bdev;
2093 struct list_head *devices;
2094 struct super_block *sb = root->fs_info->sb;
2095 struct rcu_string *name;
2097 int seeding_dev = 0;
2100 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2103 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2104 root->fs_info->bdev_holder);
2106 return PTR_ERR(bdev);
2108 if (root->fs_info->fs_devices->seeding) {
2110 down_write(&sb->s_umount);
2111 mutex_lock(&uuid_mutex);
2114 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2116 devices = &root->fs_info->fs_devices->devices;
2118 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2119 list_for_each_entry(device, devices, dev_list) {
2120 if (device->bdev == bdev) {
2123 &root->fs_info->fs_devices->device_list_mutex);
2127 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2129 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2130 if (IS_ERR(device)) {
2131 /* we can safely leave the fs_devices entry around */
2132 ret = PTR_ERR(device);
2136 name = rcu_string_strdup(device_path, GFP_NOFS);
2142 rcu_assign_pointer(device->name, name);
2144 trans = btrfs_start_transaction(root, 0);
2145 if (IS_ERR(trans)) {
2146 rcu_string_free(device->name);
2148 ret = PTR_ERR(trans);
2154 q = bdev_get_queue(bdev);
2155 if (blk_queue_discard(q))
2156 device->can_discard = 1;
2157 device->writeable = 1;
2158 device->generation = trans->transid;
2159 device->io_width = root->sectorsize;
2160 device->io_align = root->sectorsize;
2161 device->sector_size = root->sectorsize;
2162 device->total_bytes = i_size_read(bdev->bd_inode);
2163 device->disk_total_bytes = device->total_bytes;
2164 device->dev_root = root->fs_info->dev_root;
2165 device->bdev = bdev;
2166 device->in_fs_metadata = 1;
2167 device->is_tgtdev_for_dev_replace = 0;
2168 device->mode = FMODE_EXCL;
2169 device->dev_stats_valid = 1;
2170 set_blocksize(device->bdev, 4096);
2173 sb->s_flags &= ~MS_RDONLY;
2174 ret = btrfs_prepare_sprout(root);
2175 BUG_ON(ret); /* -ENOMEM */
2178 device->fs_devices = root->fs_info->fs_devices;
2180 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2181 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2182 list_add(&device->dev_alloc_list,
2183 &root->fs_info->fs_devices->alloc_list);
2184 root->fs_info->fs_devices->num_devices++;
2185 root->fs_info->fs_devices->open_devices++;
2186 root->fs_info->fs_devices->rw_devices++;
2187 root->fs_info->fs_devices->total_devices++;
2188 if (device->can_discard)
2189 root->fs_info->fs_devices->num_can_discard++;
2190 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2192 spin_lock(&root->fs_info->free_chunk_lock);
2193 root->fs_info->free_chunk_space += device->total_bytes;
2194 spin_unlock(&root->fs_info->free_chunk_lock);
2196 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2197 root->fs_info->fs_devices->rotating = 1;
2199 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2200 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2201 total_bytes + device->total_bytes);
2203 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2204 btrfs_set_super_num_devices(root->fs_info->super_copy,
2207 /* add sysfs device entry */
2208 btrfs_kobj_add_device(root->fs_info, device);
2210 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2213 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2214 ret = init_first_rw_device(trans, root, device);
2216 btrfs_abort_transaction(trans, root, ret);
2219 ret = btrfs_finish_sprout(trans, root);
2221 btrfs_abort_transaction(trans, root, ret);
2225 /* Sprouting would change fsid of the mounted root,
2226 * so rename the fsid on the sysfs
2228 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2229 root->fs_info->fsid);
2230 if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2233 ret = btrfs_add_device(trans, root, device);
2235 btrfs_abort_transaction(trans, root, ret);
2241 * we've got more storage, clear any full flags on the space
2244 btrfs_clear_space_info_full(root->fs_info);
2246 unlock_chunks(root);
2247 root->fs_info->num_tolerated_disk_barrier_failures =
2248 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2249 ret = btrfs_commit_transaction(trans, root);
2252 mutex_unlock(&uuid_mutex);
2253 up_write(&sb->s_umount);
2255 if (ret) /* transaction commit */
2258 ret = btrfs_relocate_sys_chunks(root);
2260 btrfs_error(root->fs_info, ret,
2261 "Failed to relocate sys chunks after "
2262 "device initialization. This can be fixed "
2263 "using the \"btrfs balance\" command.");
2264 trans = btrfs_attach_transaction(root);
2265 if (IS_ERR(trans)) {
2266 if (PTR_ERR(trans) == -ENOENT)
2268 return PTR_ERR(trans);
2270 ret = btrfs_commit_transaction(trans, root);
2273 /* Update ctime/mtime for libblkid */
2274 update_dev_time(device_path);
2278 unlock_chunks(root);
2279 btrfs_end_transaction(trans, root);
2280 rcu_string_free(device->name);
2281 btrfs_kobj_rm_device(root->fs_info, device);
2284 blkdev_put(bdev, FMODE_EXCL);
2286 mutex_unlock(&uuid_mutex);
2287 up_write(&sb->s_umount);
2292 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2293 struct btrfs_device **device_out)
2295 struct request_queue *q;
2296 struct btrfs_device *device;
2297 struct block_device *bdev;
2298 struct btrfs_fs_info *fs_info = root->fs_info;
2299 struct list_head *devices;
2300 struct rcu_string *name;
2301 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2305 if (fs_info->fs_devices->seeding)
2308 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2309 fs_info->bdev_holder);
2311 return PTR_ERR(bdev);
2313 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2315 devices = &fs_info->fs_devices->devices;
2316 list_for_each_entry(device, devices, dev_list) {
2317 if (device->bdev == bdev) {
2323 device = btrfs_alloc_device(NULL, &devid, NULL);
2324 if (IS_ERR(device)) {
2325 ret = PTR_ERR(device);
2329 name = rcu_string_strdup(device_path, GFP_NOFS);
2335 rcu_assign_pointer(device->name, name);
2337 q = bdev_get_queue(bdev);
2338 if (blk_queue_discard(q))
2339 device->can_discard = 1;
2340 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2341 device->writeable = 1;
2342 device->generation = 0;
2343 device->io_width = root->sectorsize;
2344 device->io_align = root->sectorsize;
2345 device->sector_size = root->sectorsize;
2346 device->total_bytes = i_size_read(bdev->bd_inode);
2347 device->disk_total_bytes = device->total_bytes;
2348 device->dev_root = fs_info->dev_root;
2349 device->bdev = bdev;
2350 device->in_fs_metadata = 1;
2351 device->is_tgtdev_for_dev_replace = 1;
2352 device->mode = FMODE_EXCL;
2353 device->dev_stats_valid = 1;
2354 set_blocksize(device->bdev, 4096);
2355 device->fs_devices = fs_info->fs_devices;
2356 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2357 fs_info->fs_devices->num_devices++;
2358 fs_info->fs_devices->open_devices++;
2359 if (device->can_discard)
2360 fs_info->fs_devices->num_can_discard++;
2361 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2363 *device_out = device;
2367 blkdev_put(bdev, FMODE_EXCL);
2371 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2372 struct btrfs_device *tgtdev)
2374 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2375 tgtdev->io_width = fs_info->dev_root->sectorsize;
2376 tgtdev->io_align = fs_info->dev_root->sectorsize;
2377 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2378 tgtdev->dev_root = fs_info->dev_root;
2379 tgtdev->in_fs_metadata = 1;
2382 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2383 struct btrfs_device *device)
2386 struct btrfs_path *path;
2387 struct btrfs_root *root;
2388 struct btrfs_dev_item *dev_item;
2389 struct extent_buffer *leaf;
2390 struct btrfs_key key;
2392 root = device->dev_root->fs_info->chunk_root;
2394 path = btrfs_alloc_path();
2398 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2399 key.type = BTRFS_DEV_ITEM_KEY;
2400 key.offset = device->devid;
2402 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2411 leaf = path->nodes[0];
2412 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2414 btrfs_set_device_id(leaf, dev_item, device->devid);
2415 btrfs_set_device_type(leaf, dev_item, device->type);
2416 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2417 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2418 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2419 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2420 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2421 btrfs_mark_buffer_dirty(leaf);
2424 btrfs_free_path(path);
2428 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2429 struct btrfs_device *device, u64 new_size)
2431 struct btrfs_super_block *super_copy =
2432 device->dev_root->fs_info->super_copy;
2433 u64 old_total = btrfs_super_total_bytes(super_copy);
2434 u64 diff = new_size - device->total_bytes;
2436 if (!device->writeable)
2438 if (new_size <= device->total_bytes ||
2439 device->is_tgtdev_for_dev_replace)
2442 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2443 device->fs_devices->total_rw_bytes += diff;
2445 device->total_bytes = new_size;
2446 device->disk_total_bytes = new_size;
2447 btrfs_clear_space_info_full(device->dev_root->fs_info);
2449 return btrfs_update_device(trans, device);
2452 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2453 struct btrfs_device *device, u64 new_size)
2456 lock_chunks(device->dev_root);
2457 ret = __btrfs_grow_device(trans, device, new_size);
2458 unlock_chunks(device->dev_root);
2462 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2463 struct btrfs_root *root,
2464 u64 chunk_tree, u64 chunk_objectid,
2468 struct btrfs_path *path;
2469 struct btrfs_key key;
2471 root = root->fs_info->chunk_root;
2472 path = btrfs_alloc_path();
2476 key.objectid = chunk_objectid;
2477 key.offset = chunk_offset;
2478 key.type = BTRFS_CHUNK_ITEM_KEY;
2480 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2483 else if (ret > 0) { /* Logic error or corruption */
2484 btrfs_error(root->fs_info, -ENOENT,
2485 "Failed lookup while freeing chunk.");
2490 ret = btrfs_del_item(trans, root, path);
2492 btrfs_error(root->fs_info, ret,
2493 "Failed to delete chunk item.");
2495 btrfs_free_path(path);
2499 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2502 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2503 struct btrfs_disk_key *disk_key;
2504 struct btrfs_chunk *chunk;
2511 struct btrfs_key key;
2513 array_size = btrfs_super_sys_array_size(super_copy);
2515 ptr = super_copy->sys_chunk_array;
2518 while (cur < array_size) {
2519 disk_key = (struct btrfs_disk_key *)ptr;
2520 btrfs_disk_key_to_cpu(&key, disk_key);
2522 len = sizeof(*disk_key);
2524 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2525 chunk = (struct btrfs_chunk *)(ptr + len);
2526 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2527 len += btrfs_chunk_item_size(num_stripes);
2532 if (key.objectid == chunk_objectid &&
2533 key.offset == chunk_offset) {
2534 memmove(ptr, ptr + len, array_size - (cur + len));
2536 btrfs_set_super_sys_array_size(super_copy, array_size);
2545 static int btrfs_relocate_chunk(struct btrfs_root *root,
2546 u64 chunk_tree, u64 chunk_objectid,
2549 struct extent_map_tree *em_tree;
2550 struct btrfs_root *extent_root;
2551 struct btrfs_trans_handle *trans;
2552 struct extent_map *em;
2553 struct map_lookup *map;
2557 root = root->fs_info->chunk_root;
2558 extent_root = root->fs_info->extent_root;
2559 em_tree = &root->fs_info->mapping_tree.map_tree;
2561 ret = btrfs_can_relocate(extent_root, chunk_offset);
2565 /* step one, relocate all the extents inside this chunk */
2566 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2570 trans = btrfs_start_transaction(root, 0);
2571 if (IS_ERR(trans)) {
2572 ret = PTR_ERR(trans);
2573 btrfs_std_error(root->fs_info, ret);
2580 * step two, delete the device extents and the
2581 * chunk tree entries
2583 read_lock(&em_tree->lock);
2584 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2585 read_unlock(&em_tree->lock);
2587 BUG_ON(!em || em->start > chunk_offset ||
2588 em->start + em->len < chunk_offset);
2589 map = (struct map_lookup *)em->bdev;
2591 for (i = 0; i < map->num_stripes; i++) {
2592 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2593 map->stripes[i].physical);
2596 if (map->stripes[i].dev) {
2597 ret = btrfs_update_device(trans, map->stripes[i].dev);
2601 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2606 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2608 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2609 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2613 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2616 write_lock(&em_tree->lock);
2617 remove_extent_mapping(em_tree, em);
2618 write_unlock(&em_tree->lock);
2620 /* once for the tree */
2621 free_extent_map(em);
2623 free_extent_map(em);
2625 unlock_chunks(root);
2626 btrfs_end_transaction(trans, root);
2630 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2632 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2633 struct btrfs_path *path;
2634 struct extent_buffer *leaf;
2635 struct btrfs_chunk *chunk;
2636 struct btrfs_key key;
2637 struct btrfs_key found_key;
2638 u64 chunk_tree = chunk_root->root_key.objectid;
2640 bool retried = false;
2644 path = btrfs_alloc_path();
2649 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2650 key.offset = (u64)-1;
2651 key.type = BTRFS_CHUNK_ITEM_KEY;
2654 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2657 BUG_ON(ret == 0); /* Corruption */
2659 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2666 leaf = path->nodes[0];
2667 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2669 chunk = btrfs_item_ptr(leaf, path->slots[0],
2670 struct btrfs_chunk);
2671 chunk_type = btrfs_chunk_type(leaf, chunk);
2672 btrfs_release_path(path);
2674 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2675 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2684 if (found_key.offset == 0)
2686 key.offset = found_key.offset - 1;
2689 if (failed && !retried) {
2693 } else if (WARN_ON(failed && retried)) {
2697 btrfs_free_path(path);
2701 static int insert_balance_item(struct btrfs_root *root,
2702 struct btrfs_balance_control *bctl)
2704 struct btrfs_trans_handle *trans;
2705 struct btrfs_balance_item *item;
2706 struct btrfs_disk_balance_args disk_bargs;
2707 struct btrfs_path *path;
2708 struct extent_buffer *leaf;
2709 struct btrfs_key key;
2712 path = btrfs_alloc_path();
2716 trans = btrfs_start_transaction(root, 0);
2717 if (IS_ERR(trans)) {
2718 btrfs_free_path(path);
2719 return PTR_ERR(trans);
2722 key.objectid = BTRFS_BALANCE_OBJECTID;
2723 key.type = BTRFS_BALANCE_ITEM_KEY;
2726 ret = btrfs_insert_empty_item(trans, root, path, &key,
2731 leaf = path->nodes[0];
2732 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2734 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2736 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2737 btrfs_set_balance_data(leaf, item, &disk_bargs);
2738 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2739 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2740 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2741 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2743 btrfs_set_balance_flags(leaf, item, bctl->flags);
2745 btrfs_mark_buffer_dirty(leaf);
2747 btrfs_free_path(path);
2748 err = btrfs_commit_transaction(trans, root);
2754 static int del_balance_item(struct btrfs_root *root)
2756 struct btrfs_trans_handle *trans;
2757 struct btrfs_path *path;
2758 struct btrfs_key key;
2761 path = btrfs_alloc_path();
2765 trans = btrfs_start_transaction(root, 0);
2766 if (IS_ERR(trans)) {
2767 btrfs_free_path(path);
2768 return PTR_ERR(trans);
2771 key.objectid = BTRFS_BALANCE_OBJECTID;
2772 key.type = BTRFS_BALANCE_ITEM_KEY;
2775 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2783 ret = btrfs_del_item(trans, root, path);
2785 btrfs_free_path(path);
2786 err = btrfs_commit_transaction(trans, root);
2793 * This is a heuristic used to reduce the number of chunks balanced on
2794 * resume after balance was interrupted.
2796 static void update_balance_args(struct btrfs_balance_control *bctl)
2799 * Turn on soft mode for chunk types that were being converted.
2801 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2802 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2803 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2804 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2805 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2806 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2809 * Turn on usage filter if is not already used. The idea is
2810 * that chunks that we have already balanced should be
2811 * reasonably full. Don't do it for chunks that are being
2812 * converted - that will keep us from relocating unconverted
2813 * (albeit full) chunks.
2815 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2816 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2817 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2818 bctl->data.usage = 90;
2820 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2821 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2822 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2823 bctl->sys.usage = 90;
2825 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2826 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2827 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2828 bctl->meta.usage = 90;
2833 * Should be called with both balance and volume mutexes held to
2834 * serialize other volume operations (add_dev/rm_dev/resize) with
2835 * restriper. Same goes for unset_balance_control.
2837 static void set_balance_control(struct btrfs_balance_control *bctl)
2839 struct btrfs_fs_info *fs_info = bctl->fs_info;
2841 BUG_ON(fs_info->balance_ctl);
2843 spin_lock(&fs_info->balance_lock);
2844 fs_info->balance_ctl = bctl;
2845 spin_unlock(&fs_info->balance_lock);
2848 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2850 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2852 BUG_ON(!fs_info->balance_ctl);
2854 spin_lock(&fs_info->balance_lock);
2855 fs_info->balance_ctl = NULL;
2856 spin_unlock(&fs_info->balance_lock);
2862 * Balance filters. Return 1 if chunk should be filtered out
2863 * (should not be balanced).
2865 static int chunk_profiles_filter(u64 chunk_type,
2866 struct btrfs_balance_args *bargs)
2868 chunk_type = chunk_to_extended(chunk_type) &
2869 BTRFS_EXTENDED_PROFILE_MASK;
2871 if (bargs->profiles & chunk_type)
2877 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2878 struct btrfs_balance_args *bargs)
2880 struct btrfs_block_group_cache *cache;
2881 u64 chunk_used, user_thresh;
2884 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2885 chunk_used = btrfs_block_group_used(&cache->item);
2887 if (bargs->usage == 0)
2889 else if (bargs->usage > 100)
2890 user_thresh = cache->key.offset;
2892 user_thresh = div_factor_fine(cache->key.offset,
2895 if (chunk_used < user_thresh)
2898 btrfs_put_block_group(cache);
2902 static int chunk_devid_filter(struct extent_buffer *leaf,
2903 struct btrfs_chunk *chunk,
2904 struct btrfs_balance_args *bargs)
2906 struct btrfs_stripe *stripe;
2907 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2910 for (i = 0; i < num_stripes; i++) {
2911 stripe = btrfs_stripe_nr(chunk, i);
2912 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2919 /* [pstart, pend) */
2920 static int chunk_drange_filter(struct extent_buffer *leaf,
2921 struct btrfs_chunk *chunk,
2923 struct btrfs_balance_args *bargs)
2925 struct btrfs_stripe *stripe;
2926 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2932 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2935 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2936 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2937 factor = num_stripes / 2;
2938 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2939 factor = num_stripes - 1;
2940 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2941 factor = num_stripes - 2;
2943 factor = num_stripes;
2946 for (i = 0; i < num_stripes; i++) {
2947 stripe = btrfs_stripe_nr(chunk, i);
2948 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2951 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2952 stripe_length = btrfs_chunk_length(leaf, chunk);
2953 do_div(stripe_length, factor);
2955 if (stripe_offset < bargs->pend &&
2956 stripe_offset + stripe_length > bargs->pstart)
2963 /* [vstart, vend) */
2964 static int chunk_vrange_filter(struct extent_buffer *leaf,
2965 struct btrfs_chunk *chunk,
2967 struct btrfs_balance_args *bargs)
2969 if (chunk_offset < bargs->vend &&
2970 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2971 /* at least part of the chunk is inside this vrange */
2977 static int chunk_soft_convert_filter(u64 chunk_type,
2978 struct btrfs_balance_args *bargs)
2980 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2983 chunk_type = chunk_to_extended(chunk_type) &
2984 BTRFS_EXTENDED_PROFILE_MASK;
2986 if (bargs->target == chunk_type)
2992 static int should_balance_chunk(struct btrfs_root *root,
2993 struct extent_buffer *leaf,
2994 struct btrfs_chunk *chunk, u64 chunk_offset)
2996 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2997 struct btrfs_balance_args *bargs = NULL;
2998 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3001 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3002 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3006 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3007 bargs = &bctl->data;
3008 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3010 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3011 bargs = &bctl->meta;
3013 /* profiles filter */
3014 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3015 chunk_profiles_filter(chunk_type, bargs)) {
3020 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3021 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3026 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3027 chunk_devid_filter(leaf, chunk, bargs)) {
3031 /* drange filter, makes sense only with devid filter */
3032 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3033 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3038 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3039 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3043 /* soft profile changing mode */
3044 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3045 chunk_soft_convert_filter(chunk_type, bargs)) {
3050 * limited by count, must be the last filter
3052 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3053 if (bargs->limit == 0)
3062 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3064 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3065 struct btrfs_root *chunk_root = fs_info->chunk_root;
3066 struct btrfs_root *dev_root = fs_info->dev_root;
3067 struct list_head *devices;
3068 struct btrfs_device *device;
3071 struct btrfs_chunk *chunk;
3072 struct btrfs_path *path;
3073 struct btrfs_key key;
3074 struct btrfs_key found_key;
3075 struct btrfs_trans_handle *trans;
3076 struct extent_buffer *leaf;
3079 int enospc_errors = 0;
3080 bool counting = true;
3081 u64 limit_data = bctl->data.limit;
3082 u64 limit_meta = bctl->meta.limit;
3083 u64 limit_sys = bctl->sys.limit;
3085 /* step one make some room on all the devices */
3086 devices = &fs_info->fs_devices->devices;
3087 list_for_each_entry(device, devices, dev_list) {
3088 old_size = device->total_bytes;
3089 size_to_free = div_factor(old_size, 1);
3090 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3091 if (!device->writeable ||
3092 device->total_bytes - device->bytes_used > size_to_free ||
3093 device->is_tgtdev_for_dev_replace)
3096 ret = btrfs_shrink_device(device, old_size - size_to_free);
3101 trans = btrfs_start_transaction(dev_root, 0);
3102 BUG_ON(IS_ERR(trans));
3104 ret = btrfs_grow_device(trans, device, old_size);
3107 btrfs_end_transaction(trans, dev_root);
3110 /* step two, relocate all the chunks */
3111 path = btrfs_alloc_path();
3117 /* zero out stat counters */
3118 spin_lock(&fs_info->balance_lock);
3119 memset(&bctl->stat, 0, sizeof(bctl->stat));
3120 spin_unlock(&fs_info->balance_lock);
3123 bctl->data.limit = limit_data;
3124 bctl->meta.limit = limit_meta;
3125 bctl->sys.limit = limit_sys;
3127 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3128 key.offset = (u64)-1;
3129 key.type = BTRFS_CHUNK_ITEM_KEY;
3132 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3133 atomic_read(&fs_info->balance_cancel_req)) {
3138 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3143 * this shouldn't happen, it means the last relocate
3147 BUG(); /* FIXME break ? */
3149 ret = btrfs_previous_item(chunk_root, path, 0,
3150 BTRFS_CHUNK_ITEM_KEY);
3156 leaf = path->nodes[0];
3157 slot = path->slots[0];
3158 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3160 if (found_key.objectid != key.objectid)
3163 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3166 spin_lock(&fs_info->balance_lock);
3167 bctl->stat.considered++;
3168 spin_unlock(&fs_info->balance_lock);
3171 ret = should_balance_chunk(chunk_root, leaf, chunk,
3173 btrfs_release_path(path);
3178 spin_lock(&fs_info->balance_lock);
3179 bctl->stat.expected++;
3180 spin_unlock(&fs_info->balance_lock);
3184 ret = btrfs_relocate_chunk(chunk_root,
3185 chunk_root->root_key.objectid,
3188 if (ret && ret != -ENOSPC)
3190 if (ret == -ENOSPC) {
3193 spin_lock(&fs_info->balance_lock);
3194 bctl->stat.completed++;
3195 spin_unlock(&fs_info->balance_lock);
3198 if (found_key.offset == 0)
3200 key.offset = found_key.offset - 1;
3204 btrfs_release_path(path);
3209 btrfs_free_path(path);
3210 if (enospc_errors) {
3211 btrfs_info(fs_info, "%d enospc errors during balance",
3221 * alloc_profile_is_valid - see if a given profile is valid and reduced
3222 * @flags: profile to validate
3223 * @extended: if true @flags is treated as an extended profile
3225 static int alloc_profile_is_valid(u64 flags, int extended)
3227 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3228 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3230 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3232 /* 1) check that all other bits are zeroed */
3236 /* 2) see if profile is reduced */
3238 return !extended; /* "0" is valid for usual profiles */
3240 /* true if exactly one bit set */
3241 return (flags & (flags - 1)) == 0;
3244 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3246 /* cancel requested || normal exit path */
3247 return atomic_read(&fs_info->balance_cancel_req) ||
3248 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3249 atomic_read(&fs_info->balance_cancel_req) == 0);
3252 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3256 unset_balance_control(fs_info);
3257 ret = del_balance_item(fs_info->tree_root);
3259 btrfs_std_error(fs_info, ret);
3261 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3265 * Should be called with both balance and volume mutexes held
3267 int btrfs_balance(struct btrfs_balance_control *bctl,
3268 struct btrfs_ioctl_balance_args *bargs)
3270 struct btrfs_fs_info *fs_info = bctl->fs_info;
3277 if (btrfs_fs_closing(fs_info) ||
3278 atomic_read(&fs_info->balance_pause_req) ||
3279 atomic_read(&fs_info->balance_cancel_req)) {
3284 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3285 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3289 * In case of mixed groups both data and meta should be picked,
3290 * and identical options should be given for both of them.
3292 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3293 if (mixed && (bctl->flags & allowed)) {
3294 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3295 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3296 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3297 btrfs_err(fs_info, "with mixed groups data and "
3298 "metadata balance options must be the same");
3304 num_devices = fs_info->fs_devices->num_devices;
3305 btrfs_dev_replace_lock(&fs_info->dev_replace);
3306 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3307 BUG_ON(num_devices < 1);
3310 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3311 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3312 if (num_devices == 1)
3313 allowed |= BTRFS_BLOCK_GROUP_DUP;
3314 else if (num_devices > 1)
3315 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3316 if (num_devices > 2)
3317 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3318 if (num_devices > 3)
3319 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3320 BTRFS_BLOCK_GROUP_RAID6);
3321 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3322 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3323 (bctl->data.target & ~allowed))) {
3324 btrfs_err(fs_info, "unable to start balance with target "
3325 "data profile %llu",
3330 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3331 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3332 (bctl->meta.target & ~allowed))) {
3334 "unable to start balance with target metadata profile %llu",
3339 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3340 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3341 (bctl->sys.target & ~allowed))) {
3343 "unable to start balance with target system profile %llu",
3349 /* allow dup'ed data chunks only in mixed mode */
3350 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3351 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3352 btrfs_err(fs_info, "dup for data is not allowed");
3357 /* allow to reduce meta or sys integrity only if force set */
3358 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3359 BTRFS_BLOCK_GROUP_RAID10 |
3360 BTRFS_BLOCK_GROUP_RAID5 |
3361 BTRFS_BLOCK_GROUP_RAID6;
3363 seq = read_seqbegin(&fs_info->profiles_lock);
3365 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3366 (fs_info->avail_system_alloc_bits & allowed) &&
3367 !(bctl->sys.target & allowed)) ||
3368 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3369 (fs_info->avail_metadata_alloc_bits & allowed) &&
3370 !(bctl->meta.target & allowed))) {
3371 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3372 btrfs_info(fs_info, "force reducing metadata integrity");
3374 btrfs_err(fs_info, "balance will reduce metadata "
3375 "integrity, use force if you want this");
3380 } while (read_seqretry(&fs_info->profiles_lock, seq));
3382 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3383 int num_tolerated_disk_barrier_failures;
3384 u64 target = bctl->sys.target;
3386 num_tolerated_disk_barrier_failures =
3387 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3388 if (num_tolerated_disk_barrier_failures > 0 &&
3390 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3391 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3392 num_tolerated_disk_barrier_failures = 0;
3393 else if (num_tolerated_disk_barrier_failures > 1 &&
3395 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3396 num_tolerated_disk_barrier_failures = 1;
3398 fs_info->num_tolerated_disk_barrier_failures =
3399 num_tolerated_disk_barrier_failures;
3402 ret = insert_balance_item(fs_info->tree_root, bctl);
3403 if (ret && ret != -EEXIST)
3406 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3407 BUG_ON(ret == -EEXIST);
3408 set_balance_control(bctl);
3410 BUG_ON(ret != -EEXIST);
3411 spin_lock(&fs_info->balance_lock);
3412 update_balance_args(bctl);
3413 spin_unlock(&fs_info->balance_lock);
3416 atomic_inc(&fs_info->balance_running);
3417 mutex_unlock(&fs_info->balance_mutex);
3419 ret = __btrfs_balance(fs_info);
3421 mutex_lock(&fs_info->balance_mutex);
3422 atomic_dec(&fs_info->balance_running);
3424 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3425 fs_info->num_tolerated_disk_barrier_failures =
3426 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3430 memset(bargs, 0, sizeof(*bargs));
3431 update_ioctl_balance_args(fs_info, 0, bargs);
3434 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3435 balance_need_close(fs_info)) {
3436 __cancel_balance(fs_info);
3439 wake_up(&fs_info->balance_wait_q);
3443 if (bctl->flags & BTRFS_BALANCE_RESUME)
3444 __cancel_balance(fs_info);
3447 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3452 static int balance_kthread(void *data)
3454 struct btrfs_fs_info *fs_info = data;
3457 mutex_lock(&fs_info->volume_mutex);
3458 mutex_lock(&fs_info->balance_mutex);
3460 if (fs_info->balance_ctl) {
3461 btrfs_info(fs_info, "continuing balance");
3462 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3465 mutex_unlock(&fs_info->balance_mutex);
3466 mutex_unlock(&fs_info->volume_mutex);
3471 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3473 struct task_struct *tsk;
3475 spin_lock(&fs_info->balance_lock);
3476 if (!fs_info->balance_ctl) {
3477 spin_unlock(&fs_info->balance_lock);
3480 spin_unlock(&fs_info->balance_lock);
3482 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3483 btrfs_info(fs_info, "force skipping balance");
3487 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3488 return PTR_ERR_OR_ZERO(tsk);
3491 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3493 struct btrfs_balance_control *bctl;
3494 struct btrfs_balance_item *item;
3495 struct btrfs_disk_balance_args disk_bargs;
3496 struct btrfs_path *path;
3497 struct extent_buffer *leaf;
3498 struct btrfs_key key;
3501 path = btrfs_alloc_path();
3505 key.objectid = BTRFS_BALANCE_OBJECTID;
3506 key.type = BTRFS_BALANCE_ITEM_KEY;
3509 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3512 if (ret > 0) { /* ret = -ENOENT; */
3517 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3523 leaf = path->nodes[0];
3524 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3526 bctl->fs_info = fs_info;
3527 bctl->flags = btrfs_balance_flags(leaf, item);
3528 bctl->flags |= BTRFS_BALANCE_RESUME;
3530 btrfs_balance_data(leaf, item, &disk_bargs);
3531 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3532 btrfs_balance_meta(leaf, item, &disk_bargs);
3533 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3534 btrfs_balance_sys(leaf, item, &disk_bargs);
3535 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3537 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3539 mutex_lock(&fs_info->volume_mutex);
3540 mutex_lock(&fs_info->balance_mutex);
3542 set_balance_control(bctl);
3544 mutex_unlock(&fs_info->balance_mutex);
3545 mutex_unlock(&fs_info->volume_mutex);
3547 btrfs_free_path(path);
3551 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3555 mutex_lock(&fs_info->balance_mutex);
3556 if (!fs_info->balance_ctl) {
3557 mutex_unlock(&fs_info->balance_mutex);
3561 if (atomic_read(&fs_info->balance_running)) {
3562 atomic_inc(&fs_info->balance_pause_req);
3563 mutex_unlock(&fs_info->balance_mutex);
3565 wait_event(fs_info->balance_wait_q,
3566 atomic_read(&fs_info->balance_running) == 0);
3568 mutex_lock(&fs_info->balance_mutex);
3569 /* we are good with balance_ctl ripped off from under us */
3570 BUG_ON(atomic_read(&fs_info->balance_running));
3571 atomic_dec(&fs_info->balance_pause_req);
3576 mutex_unlock(&fs_info->balance_mutex);
3580 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3582 if (fs_info->sb->s_flags & MS_RDONLY)
3585 mutex_lock(&fs_info->balance_mutex);
3586 if (!fs_info->balance_ctl) {
3587 mutex_unlock(&fs_info->balance_mutex);
3591 atomic_inc(&fs_info->balance_cancel_req);
3593 * if we are running just wait and return, balance item is
3594 * deleted in btrfs_balance in this case
3596 if (atomic_read(&fs_info->balance_running)) {
3597 mutex_unlock(&fs_info->balance_mutex);
3598 wait_event(fs_info->balance_wait_q,
3599 atomic_read(&fs_info->balance_running) == 0);
3600 mutex_lock(&fs_info->balance_mutex);
3602 /* __cancel_balance needs volume_mutex */
3603 mutex_unlock(&fs_info->balance_mutex);
3604 mutex_lock(&fs_info->volume_mutex);
3605 mutex_lock(&fs_info->balance_mutex);
3607 if (fs_info->balance_ctl)
3608 __cancel_balance(fs_info);
3610 mutex_unlock(&fs_info->volume_mutex);
3613 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3614 atomic_dec(&fs_info->balance_cancel_req);
3615 mutex_unlock(&fs_info->balance_mutex);
3619 static int btrfs_uuid_scan_kthread(void *data)
3621 struct btrfs_fs_info *fs_info = data;
3622 struct btrfs_root *root = fs_info->tree_root;
3623 struct btrfs_key key;
3624 struct btrfs_key max_key;
3625 struct btrfs_path *path = NULL;
3627 struct extent_buffer *eb;
3629 struct btrfs_root_item root_item;
3631 struct btrfs_trans_handle *trans = NULL;
3633 path = btrfs_alloc_path();
3640 key.type = BTRFS_ROOT_ITEM_KEY;
3643 max_key.objectid = (u64)-1;
3644 max_key.type = BTRFS_ROOT_ITEM_KEY;
3645 max_key.offset = (u64)-1;
3647 path->keep_locks = 1;
3650 ret = btrfs_search_forward(root, &key, path, 0);
3657 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3658 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3659 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3660 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3663 eb = path->nodes[0];
3664 slot = path->slots[0];
3665 item_size = btrfs_item_size_nr(eb, slot);
3666 if (item_size < sizeof(root_item))
3669 read_extent_buffer(eb, &root_item,
3670 btrfs_item_ptr_offset(eb, slot),
3671 (int)sizeof(root_item));
3672 if (btrfs_root_refs(&root_item) == 0)
3675 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3676 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3680 btrfs_release_path(path);
3682 * 1 - subvol uuid item
3683 * 1 - received_subvol uuid item
3685 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3686 if (IS_ERR(trans)) {
3687 ret = PTR_ERR(trans);
3695 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3696 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3698 BTRFS_UUID_KEY_SUBVOL,
3701 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3707 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3708 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3709 root_item.received_uuid,
3710 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3713 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3721 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3727 btrfs_release_path(path);
3728 if (key.offset < (u64)-1) {
3730 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3732 key.type = BTRFS_ROOT_ITEM_KEY;
3733 } else if (key.objectid < (u64)-1) {
3735 key.type = BTRFS_ROOT_ITEM_KEY;
3744 btrfs_free_path(path);
3745 if (trans && !IS_ERR(trans))
3746 btrfs_end_transaction(trans, fs_info->uuid_root);
3748 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3750 fs_info->update_uuid_tree_gen = 1;
3751 up(&fs_info->uuid_tree_rescan_sem);
3756 * Callback for btrfs_uuid_tree_iterate().
3758 * 0 check succeeded, the entry is not outdated.
3759 * < 0 if an error occured.
3760 * > 0 if the check failed, which means the caller shall remove the entry.
3762 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3763 u8 *uuid, u8 type, u64 subid)
3765 struct btrfs_key key;
3767 struct btrfs_root *subvol_root;
3769 if (type != BTRFS_UUID_KEY_SUBVOL &&
3770 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3773 key.objectid = subid;
3774 key.type = BTRFS_ROOT_ITEM_KEY;
3775 key.offset = (u64)-1;
3776 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3777 if (IS_ERR(subvol_root)) {
3778 ret = PTR_ERR(subvol_root);
3785 case BTRFS_UUID_KEY_SUBVOL:
3786 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3789 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3790 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3800 static int btrfs_uuid_rescan_kthread(void *data)
3802 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3806 * 1st step is to iterate through the existing UUID tree and
3807 * to delete all entries that contain outdated data.
3808 * 2nd step is to add all missing entries to the UUID tree.
3810 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3812 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3813 up(&fs_info->uuid_tree_rescan_sem);
3816 return btrfs_uuid_scan_kthread(data);
3819 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3821 struct btrfs_trans_handle *trans;
3822 struct btrfs_root *tree_root = fs_info->tree_root;
3823 struct btrfs_root *uuid_root;
3824 struct task_struct *task;
3831 trans = btrfs_start_transaction(tree_root, 2);
3833 return PTR_ERR(trans);
3835 uuid_root = btrfs_create_tree(trans, fs_info,
3836 BTRFS_UUID_TREE_OBJECTID);
3837 if (IS_ERR(uuid_root)) {
3838 btrfs_abort_transaction(trans, tree_root,
3839 PTR_ERR(uuid_root));
3840 return PTR_ERR(uuid_root);
3843 fs_info->uuid_root = uuid_root;
3845 ret = btrfs_commit_transaction(trans, tree_root);
3849 down(&fs_info->uuid_tree_rescan_sem);
3850 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3852 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3853 btrfs_warn(fs_info, "failed to start uuid_scan task");
3854 up(&fs_info->uuid_tree_rescan_sem);
3855 return PTR_ERR(task);
3861 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3863 struct task_struct *task;
3865 down(&fs_info->uuid_tree_rescan_sem);
3866 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3868 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3869 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3870 up(&fs_info->uuid_tree_rescan_sem);
3871 return PTR_ERR(task);
3878 * shrinking a device means finding all of the device extents past
3879 * the new size, and then following the back refs to the chunks.
3880 * The chunk relocation code actually frees the device extent
3882 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3884 struct btrfs_trans_handle *trans;
3885 struct btrfs_root *root = device->dev_root;
3886 struct btrfs_dev_extent *dev_extent = NULL;
3887 struct btrfs_path *path;
3895 bool retried = false;
3896 struct extent_buffer *l;
3897 struct btrfs_key key;
3898 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3899 u64 old_total = btrfs_super_total_bytes(super_copy);
3900 u64 old_size = device->total_bytes;
3901 u64 diff = device->total_bytes - new_size;
3903 if (device->is_tgtdev_for_dev_replace)
3906 path = btrfs_alloc_path();
3914 device->total_bytes = new_size;
3915 if (device->writeable) {
3916 device->fs_devices->total_rw_bytes -= diff;
3917 spin_lock(&root->fs_info->free_chunk_lock);
3918 root->fs_info->free_chunk_space -= diff;
3919 spin_unlock(&root->fs_info->free_chunk_lock);
3921 unlock_chunks(root);
3924 key.objectid = device->devid;
3925 key.offset = (u64)-1;
3926 key.type = BTRFS_DEV_EXTENT_KEY;
3929 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3933 ret = btrfs_previous_item(root, path, 0, key.type);
3938 btrfs_release_path(path);
3943 slot = path->slots[0];
3944 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3946 if (key.objectid != device->devid) {
3947 btrfs_release_path(path);
3951 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3952 length = btrfs_dev_extent_length(l, dev_extent);
3954 if (key.offset + length <= new_size) {
3955 btrfs_release_path(path);
3959 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3960 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3961 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3962 btrfs_release_path(path);
3964 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3966 if (ret && ret != -ENOSPC)
3970 } while (key.offset-- > 0);
3972 if (failed && !retried) {
3976 } else if (failed && retried) {
3980 device->total_bytes = old_size;
3981 if (device->writeable)
3982 device->fs_devices->total_rw_bytes += diff;
3983 spin_lock(&root->fs_info->free_chunk_lock);
3984 root->fs_info->free_chunk_space += diff;
3985 spin_unlock(&root->fs_info->free_chunk_lock);
3986 unlock_chunks(root);
3990 /* Shrinking succeeded, else we would be at "done". */
3991 trans = btrfs_start_transaction(root, 0);
3992 if (IS_ERR(trans)) {
3993 ret = PTR_ERR(trans);
3999 device->disk_total_bytes = new_size;
4000 /* Now btrfs_update_device() will change the on-disk size. */
4001 ret = btrfs_update_device(trans, device);
4003 unlock_chunks(root);
4004 btrfs_end_transaction(trans, root);
4007 WARN_ON(diff > old_total);
4008 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4009 unlock_chunks(root);
4010 btrfs_end_transaction(trans, root);
4012 btrfs_free_path(path);
4016 static int btrfs_add_system_chunk(struct btrfs_root *root,
4017 struct btrfs_key *key,
4018 struct btrfs_chunk *chunk, int item_size)
4020 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4021 struct btrfs_disk_key disk_key;
4025 array_size = btrfs_super_sys_array_size(super_copy);
4026 if (array_size + item_size + sizeof(disk_key)
4027 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4030 ptr = super_copy->sys_chunk_array + array_size;
4031 btrfs_cpu_key_to_disk(&disk_key, key);
4032 memcpy(ptr, &disk_key, sizeof(disk_key));
4033 ptr += sizeof(disk_key);
4034 memcpy(ptr, chunk, item_size);
4035 item_size += sizeof(disk_key);
4036 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4041 * sort the devices in descending order by max_avail, total_avail
4043 static int btrfs_cmp_device_info(const void *a, const void *b)
4045 const struct btrfs_device_info *di_a = a;
4046 const struct btrfs_device_info *di_b = b;
4048 if (di_a->max_avail > di_b->max_avail)
4050 if (di_a->max_avail < di_b->max_avail)
4052 if (di_a->total_avail > di_b->total_avail)
4054 if (di_a->total_avail < di_b->total_avail)
4059 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4060 [BTRFS_RAID_RAID10] = {
4063 .devs_max = 0, /* 0 == as many as possible */
4065 .devs_increment = 2,
4068 [BTRFS_RAID_RAID1] = {
4073 .devs_increment = 2,
4076 [BTRFS_RAID_DUP] = {
4081 .devs_increment = 1,
4084 [BTRFS_RAID_RAID0] = {
4089 .devs_increment = 1,
4092 [BTRFS_RAID_SINGLE] = {
4097 .devs_increment = 1,
4100 [BTRFS_RAID_RAID5] = {
4105 .devs_increment = 1,
4108 [BTRFS_RAID_RAID6] = {
4113 .devs_increment = 1,
4118 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4120 /* TODO allow them to set a preferred stripe size */
4124 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4126 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4129 btrfs_set_fs_incompat(info, RAID56);
4132 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4133 - sizeof(struct btrfs_item) \
4134 - sizeof(struct btrfs_chunk)) \
4135 / sizeof(struct btrfs_stripe) + 1)
4137 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4138 - 2 * sizeof(struct btrfs_disk_key) \
4139 - 2 * sizeof(struct btrfs_chunk)) \
4140 / sizeof(struct btrfs_stripe) + 1)
4142 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4143 struct btrfs_root *extent_root, u64 start,
4146 struct btrfs_fs_info *info = extent_root->fs_info;
4147 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4148 struct list_head *cur;
4149 struct map_lookup *map = NULL;
4150 struct extent_map_tree *em_tree;
4151 struct extent_map *em;
4152 struct btrfs_device_info *devices_info = NULL;
4154 int num_stripes; /* total number of stripes to allocate */
4155 int data_stripes; /* number of stripes that count for
4157 int sub_stripes; /* sub_stripes info for map */
4158 int dev_stripes; /* stripes per dev */
4159 int devs_max; /* max devs to use */
4160 int devs_min; /* min devs needed */
4161 int devs_increment; /* ndevs has to be a multiple of this */
4162 int ncopies; /* how many copies to data has */
4164 u64 max_stripe_size;
4168 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4174 BUG_ON(!alloc_profile_is_valid(type, 0));
4176 if (list_empty(&fs_devices->alloc_list))
4179 index = __get_raid_index(type);
4181 sub_stripes = btrfs_raid_array[index].sub_stripes;
4182 dev_stripes = btrfs_raid_array[index].dev_stripes;
4183 devs_max = btrfs_raid_array[index].devs_max;
4184 devs_min = btrfs_raid_array[index].devs_min;
4185 devs_increment = btrfs_raid_array[index].devs_increment;
4186 ncopies = btrfs_raid_array[index].ncopies;
4188 if (type & BTRFS_BLOCK_GROUP_DATA) {
4189 max_stripe_size = 1024 * 1024 * 1024;
4190 max_chunk_size = 10 * max_stripe_size;
4192 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4193 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4194 /* for larger filesystems, use larger metadata chunks */
4195 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4196 max_stripe_size = 1024 * 1024 * 1024;
4198 max_stripe_size = 256 * 1024 * 1024;
4199 max_chunk_size = max_stripe_size;
4201 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4202 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4203 max_stripe_size = 32 * 1024 * 1024;
4204 max_chunk_size = 2 * max_stripe_size;
4206 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4208 btrfs_err(info, "invalid chunk type 0x%llx requested",
4213 /* we don't want a chunk larger than 10% of writeable space */
4214 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4217 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4222 cur = fs_devices->alloc_list.next;
4225 * in the first pass through the devices list, we gather information
4226 * about the available holes on each device.
4229 while (cur != &fs_devices->alloc_list) {
4230 struct btrfs_device *device;
4234 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4238 if (!device->writeable) {
4240 "BTRFS: read-only device in alloc_list\n");
4244 if (!device->in_fs_metadata ||
4245 device->is_tgtdev_for_dev_replace)
4248 if (device->total_bytes > device->bytes_used)
4249 total_avail = device->total_bytes - device->bytes_used;
4253 /* If there is no space on this device, skip it. */
4254 if (total_avail == 0)
4257 ret = find_free_dev_extent(trans, device,
4258 max_stripe_size * dev_stripes,
4259 &dev_offset, &max_avail);
4260 if (ret && ret != -ENOSPC)
4264 max_avail = max_stripe_size * dev_stripes;
4266 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4269 if (ndevs == fs_devices->rw_devices) {
4270 WARN(1, "%s: found more than %llu devices\n",
4271 __func__, fs_devices->rw_devices);
4274 devices_info[ndevs].dev_offset = dev_offset;
4275 devices_info[ndevs].max_avail = max_avail;
4276 devices_info[ndevs].total_avail = total_avail;
4277 devices_info[ndevs].dev = device;
4282 * now sort the devices by hole size / available space
4284 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4285 btrfs_cmp_device_info, NULL);
4287 /* round down to number of usable stripes */
4288 ndevs -= ndevs % devs_increment;
4290 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4295 if (devs_max && ndevs > devs_max)
4298 * the primary goal is to maximize the number of stripes, so use as many
4299 * devices as possible, even if the stripes are not maximum sized.
4301 stripe_size = devices_info[ndevs-1].max_avail;
4302 num_stripes = ndevs * dev_stripes;
4305 * this will have to be fixed for RAID1 and RAID10 over
4308 data_stripes = num_stripes / ncopies;
4310 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4311 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4312 btrfs_super_stripesize(info->super_copy));
4313 data_stripes = num_stripes - 1;
4315 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4316 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4317 btrfs_super_stripesize(info->super_copy));
4318 data_stripes = num_stripes - 2;
4322 * Use the number of data stripes to figure out how big this chunk
4323 * is really going to be in terms of logical address space,
4324 * and compare that answer with the max chunk size
4326 if (stripe_size * data_stripes > max_chunk_size) {
4327 u64 mask = (1ULL << 24) - 1;
4328 stripe_size = max_chunk_size;
4329 do_div(stripe_size, data_stripes);
4331 /* bump the answer up to a 16MB boundary */
4332 stripe_size = (stripe_size + mask) & ~mask;
4334 /* but don't go higher than the limits we found
4335 * while searching for free extents
4337 if (stripe_size > devices_info[ndevs-1].max_avail)
4338 stripe_size = devices_info[ndevs-1].max_avail;
4341 do_div(stripe_size, dev_stripes);
4343 /* align to BTRFS_STRIPE_LEN */
4344 do_div(stripe_size, raid_stripe_len);
4345 stripe_size *= raid_stripe_len;
4347 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4352 map->num_stripes = num_stripes;
4354 for (i = 0; i < ndevs; ++i) {
4355 for (j = 0; j < dev_stripes; ++j) {
4356 int s = i * dev_stripes + j;
4357 map->stripes[s].dev = devices_info[i].dev;
4358 map->stripes[s].physical = devices_info[i].dev_offset +
4362 map->sector_size = extent_root->sectorsize;
4363 map->stripe_len = raid_stripe_len;
4364 map->io_align = raid_stripe_len;
4365 map->io_width = raid_stripe_len;
4367 map->sub_stripes = sub_stripes;
4369 num_bytes = stripe_size * data_stripes;
4371 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4373 em = alloc_extent_map();
4379 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4380 em->bdev = (struct block_device *)map;
4382 em->len = num_bytes;
4383 em->block_start = 0;
4384 em->block_len = em->len;
4385 em->orig_block_len = stripe_size;
4387 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4388 write_lock(&em_tree->lock);
4389 ret = add_extent_mapping(em_tree, em, 0);
4391 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4392 atomic_inc(&em->refs);
4394 write_unlock(&em_tree->lock);
4396 free_extent_map(em);
4400 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4401 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4404 goto error_del_extent;
4406 free_extent_map(em);
4407 check_raid56_incompat_flag(extent_root->fs_info, type);
4409 kfree(devices_info);
4413 write_lock(&em_tree->lock);
4414 remove_extent_mapping(em_tree, em);
4415 write_unlock(&em_tree->lock);
4417 /* One for our allocation */
4418 free_extent_map(em);
4419 /* One for the tree reference */
4420 free_extent_map(em);
4422 kfree(devices_info);
4426 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4427 struct btrfs_root *extent_root,
4428 u64 chunk_offset, u64 chunk_size)
4430 struct btrfs_key key;
4431 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4432 struct btrfs_device *device;
4433 struct btrfs_chunk *chunk;
4434 struct btrfs_stripe *stripe;
4435 struct extent_map_tree *em_tree;
4436 struct extent_map *em;
4437 struct map_lookup *map;
4444 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4445 read_lock(&em_tree->lock);
4446 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4447 read_unlock(&em_tree->lock);
4450 btrfs_crit(extent_root->fs_info, "unable to find logical "
4451 "%Lu len %Lu", chunk_offset, chunk_size);
4455 if (em->start != chunk_offset || em->len != chunk_size) {
4456 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4457 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4458 chunk_size, em->start, em->len);
4459 free_extent_map(em);
4463 map = (struct map_lookup *)em->bdev;
4464 item_size = btrfs_chunk_item_size(map->num_stripes);
4465 stripe_size = em->orig_block_len;
4467 chunk = kzalloc(item_size, GFP_NOFS);
4473 for (i = 0; i < map->num_stripes; i++) {
4474 device = map->stripes[i].dev;
4475 dev_offset = map->stripes[i].physical;
4477 device->bytes_used += stripe_size;
4478 ret = btrfs_update_device(trans, device);
4481 ret = btrfs_alloc_dev_extent(trans, device,
4482 chunk_root->root_key.objectid,
4483 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4484 chunk_offset, dev_offset,
4490 spin_lock(&extent_root->fs_info->free_chunk_lock);
4491 extent_root->fs_info->free_chunk_space -= (stripe_size *
4493 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4495 stripe = &chunk->stripe;
4496 for (i = 0; i < map->num_stripes; i++) {
4497 device = map->stripes[i].dev;
4498 dev_offset = map->stripes[i].physical;
4500 btrfs_set_stack_stripe_devid(stripe, device->devid);
4501 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4502 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4506 btrfs_set_stack_chunk_length(chunk, chunk_size);
4507 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4508 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4509 btrfs_set_stack_chunk_type(chunk, map->type);
4510 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4511 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4512 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4513 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4514 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4516 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4517 key.type = BTRFS_CHUNK_ITEM_KEY;
4518 key.offset = chunk_offset;
4520 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4521 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4523 * TODO: Cleanup of inserted chunk root in case of
4526 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4532 free_extent_map(em);
4537 * Chunk allocation falls into two parts. The first part does works
4538 * that make the new allocated chunk useable, but not do any operation
4539 * that modifies the chunk tree. The second part does the works that
4540 * require modifying the chunk tree. This division is important for the
4541 * bootstrap process of adding storage to a seed btrfs.
4543 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4544 struct btrfs_root *extent_root, u64 type)
4548 chunk_offset = find_next_chunk(extent_root->fs_info);
4549 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4552 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4553 struct btrfs_root *root,
4554 struct btrfs_device *device)
4557 u64 sys_chunk_offset;
4559 struct btrfs_fs_info *fs_info = root->fs_info;
4560 struct btrfs_root *extent_root = fs_info->extent_root;
4563 chunk_offset = find_next_chunk(fs_info);
4564 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4565 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4570 sys_chunk_offset = find_next_chunk(root->fs_info);
4571 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4572 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4575 btrfs_abort_transaction(trans, root, ret);
4579 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4581 btrfs_abort_transaction(trans, root, ret);
4586 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4588 struct extent_map *em;
4589 struct map_lookup *map;
4590 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4594 read_lock(&map_tree->map_tree.lock);
4595 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4596 read_unlock(&map_tree->map_tree.lock);
4600 if (btrfs_test_opt(root, DEGRADED)) {
4601 free_extent_map(em);
4605 map = (struct map_lookup *)em->bdev;
4606 for (i = 0; i < map->num_stripes; i++) {
4607 if (!map->stripes[i].dev->writeable) {
4612 free_extent_map(em);
4616 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4618 extent_map_tree_init(&tree->map_tree);
4621 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4623 struct extent_map *em;
4626 write_lock(&tree->map_tree.lock);
4627 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4629 remove_extent_mapping(&tree->map_tree, em);
4630 write_unlock(&tree->map_tree.lock);
4634 free_extent_map(em);
4635 /* once for the tree */
4636 free_extent_map(em);
4640 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4642 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4643 struct extent_map *em;
4644 struct map_lookup *map;
4645 struct extent_map_tree *em_tree = &map_tree->map_tree;
4648 read_lock(&em_tree->lock);
4649 em = lookup_extent_mapping(em_tree, logical, len);
4650 read_unlock(&em_tree->lock);
4653 * We could return errors for these cases, but that could get ugly and
4654 * we'd probably do the same thing which is just not do anything else
4655 * and exit, so return 1 so the callers don't try to use other copies.
4658 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4663 if (em->start > logical || em->start + em->len < logical) {
4664 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4665 "%Lu-%Lu", logical, logical+len, em->start,
4666 em->start + em->len);
4667 free_extent_map(em);
4671 map = (struct map_lookup *)em->bdev;
4672 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4673 ret = map->num_stripes;
4674 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4675 ret = map->sub_stripes;
4676 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4678 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4682 free_extent_map(em);
4684 btrfs_dev_replace_lock(&fs_info->dev_replace);
4685 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4687 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4692 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4693 struct btrfs_mapping_tree *map_tree,
4696 struct extent_map *em;
4697 struct map_lookup *map;
4698 struct extent_map_tree *em_tree = &map_tree->map_tree;
4699 unsigned long len = root->sectorsize;
4701 read_lock(&em_tree->lock);
4702 em = lookup_extent_mapping(em_tree, logical, len);
4703 read_unlock(&em_tree->lock);
4706 BUG_ON(em->start > logical || em->start + em->len < logical);
4707 map = (struct map_lookup *)em->bdev;
4708 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4709 BTRFS_BLOCK_GROUP_RAID6)) {
4710 len = map->stripe_len * nr_data_stripes(map);
4712 free_extent_map(em);
4716 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4717 u64 logical, u64 len, int mirror_num)
4719 struct extent_map *em;
4720 struct map_lookup *map;
4721 struct extent_map_tree *em_tree = &map_tree->map_tree;
4724 read_lock(&em_tree->lock);
4725 em = lookup_extent_mapping(em_tree, logical, len);
4726 read_unlock(&em_tree->lock);
4729 BUG_ON(em->start > logical || em->start + em->len < logical);
4730 map = (struct map_lookup *)em->bdev;
4731 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4732 BTRFS_BLOCK_GROUP_RAID6))
4734 free_extent_map(em);
4738 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4739 struct map_lookup *map, int first, int num,
4740 int optimal, int dev_replace_is_ongoing)
4744 struct btrfs_device *srcdev;
4746 if (dev_replace_is_ongoing &&
4747 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4748 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4749 srcdev = fs_info->dev_replace.srcdev;
4754 * try to avoid the drive that is the source drive for a
4755 * dev-replace procedure, only choose it if no other non-missing
4756 * mirror is available
4758 for (tolerance = 0; tolerance < 2; tolerance++) {
4759 if (map->stripes[optimal].dev->bdev &&
4760 (tolerance || map->stripes[optimal].dev != srcdev))
4762 for (i = first; i < first + num; i++) {
4763 if (map->stripes[i].dev->bdev &&
4764 (tolerance || map->stripes[i].dev != srcdev))
4769 /* we couldn't find one that doesn't fail. Just return something
4770 * and the io error handling code will clean up eventually
4775 static inline int parity_smaller(u64 a, u64 b)
4780 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4781 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4783 struct btrfs_bio_stripe s;
4790 for (i = 0; i < bbio->num_stripes - 1; i++) {
4791 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4792 s = bbio->stripes[i];
4794 bbio->stripes[i] = bbio->stripes[i+1];
4795 raid_map[i] = raid_map[i+1];
4796 bbio->stripes[i+1] = s;
4804 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4805 u64 logical, u64 *length,
4806 struct btrfs_bio **bbio_ret,
4807 int mirror_num, u64 **raid_map_ret)
4809 struct extent_map *em;
4810 struct map_lookup *map;
4811 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4812 struct extent_map_tree *em_tree = &map_tree->map_tree;
4815 u64 stripe_end_offset;
4820 u64 *raid_map = NULL;
4826 struct btrfs_bio *bbio = NULL;
4827 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4828 int dev_replace_is_ongoing = 0;
4829 int num_alloc_stripes;
4830 int patch_the_first_stripe_for_dev_replace = 0;
4831 u64 physical_to_patch_in_first_stripe = 0;
4832 u64 raid56_full_stripe_start = (u64)-1;
4834 read_lock(&em_tree->lock);
4835 em = lookup_extent_mapping(em_tree, logical, *length);
4836 read_unlock(&em_tree->lock);
4839 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4844 if (em->start > logical || em->start + em->len < logical) {
4845 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4846 "found %Lu-%Lu", logical, em->start,
4847 em->start + em->len);
4848 free_extent_map(em);
4852 map = (struct map_lookup *)em->bdev;
4853 offset = logical - em->start;
4855 stripe_len = map->stripe_len;
4858 * stripe_nr counts the total number of stripes we have to stride
4859 * to get to this block
4861 do_div(stripe_nr, stripe_len);
4863 stripe_offset = stripe_nr * stripe_len;
4864 BUG_ON(offset < stripe_offset);
4866 /* stripe_offset is the offset of this block in its stripe*/
4867 stripe_offset = offset - stripe_offset;
4869 /* if we're here for raid56, we need to know the stripe aligned start */
4870 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4871 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4872 raid56_full_stripe_start = offset;
4874 /* allow a write of a full stripe, but make sure we don't
4875 * allow straddling of stripes
4877 do_div(raid56_full_stripe_start, full_stripe_len);
4878 raid56_full_stripe_start *= full_stripe_len;
4881 if (rw & REQ_DISCARD) {
4882 /* we don't discard raid56 yet */
4884 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4888 *length = min_t(u64, em->len - offset, *length);
4889 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4891 /* For writes to RAID[56], allow a full stripeset across all disks.
4892 For other RAID types and for RAID[56] reads, just allow a single
4893 stripe (on a single disk). */
4894 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4896 max_len = stripe_len * nr_data_stripes(map) -
4897 (offset - raid56_full_stripe_start);
4899 /* we limit the length of each bio to what fits in a stripe */
4900 max_len = stripe_len - stripe_offset;
4902 *length = min_t(u64, em->len - offset, max_len);
4904 *length = em->len - offset;
4907 /* This is for when we're called from btrfs_merge_bio_hook() and all
4908 it cares about is the length */
4912 btrfs_dev_replace_lock(dev_replace);
4913 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4914 if (!dev_replace_is_ongoing)
4915 btrfs_dev_replace_unlock(dev_replace);
4917 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4918 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4919 dev_replace->tgtdev != NULL) {
4921 * in dev-replace case, for repair case (that's the only
4922 * case where the mirror is selected explicitly when
4923 * calling btrfs_map_block), blocks left of the left cursor
4924 * can also be read from the target drive.
4925 * For REQ_GET_READ_MIRRORS, the target drive is added as
4926 * the last one to the array of stripes. For READ, it also
4927 * needs to be supported using the same mirror number.
4928 * If the requested block is not left of the left cursor,
4929 * EIO is returned. This can happen because btrfs_num_copies()
4930 * returns one more in the dev-replace case.
4932 u64 tmp_length = *length;
4933 struct btrfs_bio *tmp_bbio = NULL;
4934 int tmp_num_stripes;
4935 u64 srcdev_devid = dev_replace->srcdev->devid;
4936 int index_srcdev = 0;
4938 u64 physical_of_found = 0;
4940 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4941 logical, &tmp_length, &tmp_bbio, 0, NULL);
4943 WARN_ON(tmp_bbio != NULL);
4947 tmp_num_stripes = tmp_bbio->num_stripes;
4948 if (mirror_num > tmp_num_stripes) {
4950 * REQ_GET_READ_MIRRORS does not contain this
4951 * mirror, that means that the requested area
4952 * is not left of the left cursor
4960 * process the rest of the function using the mirror_num
4961 * of the source drive. Therefore look it up first.
4962 * At the end, patch the device pointer to the one of the
4965 for (i = 0; i < tmp_num_stripes; i++) {
4966 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4968 * In case of DUP, in order to keep it
4969 * simple, only add the mirror with the
4970 * lowest physical address
4973 physical_of_found <=
4974 tmp_bbio->stripes[i].physical)
4979 tmp_bbio->stripes[i].physical;
4984 mirror_num = index_srcdev + 1;
4985 patch_the_first_stripe_for_dev_replace = 1;
4986 physical_to_patch_in_first_stripe = physical_of_found;
4995 } else if (mirror_num > map->num_stripes) {
5001 stripe_nr_orig = stripe_nr;
5002 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5003 do_div(stripe_nr_end, map->stripe_len);
5004 stripe_end_offset = stripe_nr_end * map->stripe_len -
5007 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5008 if (rw & REQ_DISCARD)
5009 num_stripes = min_t(u64, map->num_stripes,
5010 stripe_nr_end - stripe_nr_orig);
5011 stripe_index = do_div(stripe_nr, map->num_stripes);
5012 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5013 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5014 num_stripes = map->num_stripes;
5015 else if (mirror_num)
5016 stripe_index = mirror_num - 1;
5018 stripe_index = find_live_mirror(fs_info, map, 0,
5020 current->pid % map->num_stripes,
5021 dev_replace_is_ongoing);
5022 mirror_num = stripe_index + 1;
5025 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5026 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5027 num_stripes = map->num_stripes;
5028 } else if (mirror_num) {
5029 stripe_index = mirror_num - 1;
5034 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5035 int factor = map->num_stripes / map->sub_stripes;
5037 stripe_index = do_div(stripe_nr, factor);
5038 stripe_index *= map->sub_stripes;
5040 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5041 num_stripes = map->sub_stripes;
5042 else if (rw & REQ_DISCARD)
5043 num_stripes = min_t(u64, map->sub_stripes *
5044 (stripe_nr_end - stripe_nr_orig),
5046 else if (mirror_num)
5047 stripe_index += mirror_num - 1;
5049 int old_stripe_index = stripe_index;
5050 stripe_index = find_live_mirror(fs_info, map,
5052 map->sub_stripes, stripe_index +
5053 current->pid % map->sub_stripes,
5054 dev_replace_is_ongoing);
5055 mirror_num = stripe_index - old_stripe_index + 1;
5058 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5059 BTRFS_BLOCK_GROUP_RAID6)) {
5062 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5066 /* push stripe_nr back to the start of the full stripe */
5067 stripe_nr = raid56_full_stripe_start;
5068 do_div(stripe_nr, stripe_len);
5070 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5072 /* RAID[56] write or recovery. Return all stripes */
5073 num_stripes = map->num_stripes;
5074 max_errors = nr_parity_stripes(map);
5076 raid_map = kmalloc_array(num_stripes, sizeof(u64),
5083 /* Work out the disk rotation on this stripe-set */
5085 rot = do_div(tmp, num_stripes);
5087 /* Fill in the logical address of each stripe */
5088 tmp = stripe_nr * nr_data_stripes(map);
5089 for (i = 0; i < nr_data_stripes(map); i++)
5090 raid_map[(i+rot) % num_stripes] =
5091 em->start + (tmp + i) * map->stripe_len;
5093 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5094 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5095 raid_map[(i+rot+1) % num_stripes] =
5098 *length = map->stripe_len;
5103 * Mirror #0 or #1 means the original data block.
5104 * Mirror #2 is RAID5 parity block.
5105 * Mirror #3 is RAID6 Q block.
5107 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5109 stripe_index = nr_data_stripes(map) +
5112 /* We distribute the parity blocks across stripes */
5113 tmp = stripe_nr + stripe_index;
5114 stripe_index = do_div(tmp, map->num_stripes);
5118 * after this do_div call, stripe_nr is the number of stripes
5119 * on this device we have to walk to find the data, and
5120 * stripe_index is the number of our device in the stripe array
5122 stripe_index = do_div(stripe_nr, map->num_stripes);
5123 mirror_num = stripe_index + 1;
5125 BUG_ON(stripe_index >= map->num_stripes);
5127 num_alloc_stripes = num_stripes;
5128 if (dev_replace_is_ongoing) {
5129 if (rw & (REQ_WRITE | REQ_DISCARD))
5130 num_alloc_stripes <<= 1;
5131 if (rw & REQ_GET_READ_MIRRORS)
5132 num_alloc_stripes++;
5134 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5140 atomic_set(&bbio->error, 0);
5142 if (rw & REQ_DISCARD) {
5144 int sub_stripes = 0;
5145 u64 stripes_per_dev = 0;
5146 u32 remaining_stripes = 0;
5147 u32 last_stripe = 0;
5150 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5151 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5154 sub_stripes = map->sub_stripes;
5156 factor = map->num_stripes / sub_stripes;
5157 stripes_per_dev = div_u64_rem(stripe_nr_end -
5160 &remaining_stripes);
5161 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5162 last_stripe *= sub_stripes;
5165 for (i = 0; i < num_stripes; i++) {
5166 bbio->stripes[i].physical =
5167 map->stripes[stripe_index].physical +
5168 stripe_offset + stripe_nr * map->stripe_len;
5169 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5171 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5172 BTRFS_BLOCK_GROUP_RAID10)) {
5173 bbio->stripes[i].length = stripes_per_dev *
5176 if (i / sub_stripes < remaining_stripes)
5177 bbio->stripes[i].length +=
5181 * Special for the first stripe and
5184 * |-------|...|-------|
5188 if (i < sub_stripes)
5189 bbio->stripes[i].length -=
5192 if (stripe_index >= last_stripe &&
5193 stripe_index <= (last_stripe +
5195 bbio->stripes[i].length -=
5198 if (i == sub_stripes - 1)
5201 bbio->stripes[i].length = *length;
5204 if (stripe_index == map->num_stripes) {
5205 /* This could only happen for RAID0/10 */
5211 for (i = 0; i < num_stripes; i++) {
5212 bbio->stripes[i].physical =
5213 map->stripes[stripe_index].physical +
5215 stripe_nr * map->stripe_len;
5216 bbio->stripes[i].dev =
5217 map->stripes[stripe_index].dev;
5222 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5223 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5224 BTRFS_BLOCK_GROUP_RAID10 |
5225 BTRFS_BLOCK_GROUP_RAID5 |
5226 BTRFS_BLOCK_GROUP_DUP)) {
5228 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5233 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5234 dev_replace->tgtdev != NULL) {
5235 int index_where_to_add;
5236 u64 srcdev_devid = dev_replace->srcdev->devid;
5239 * duplicate the write operations while the dev replace
5240 * procedure is running. Since the copying of the old disk
5241 * to the new disk takes place at run time while the
5242 * filesystem is mounted writable, the regular write
5243 * operations to the old disk have to be duplicated to go
5244 * to the new disk as well.
5245 * Note that device->missing is handled by the caller, and
5246 * that the write to the old disk is already set up in the
5249 index_where_to_add = num_stripes;
5250 for (i = 0; i < num_stripes; i++) {
5251 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5252 /* write to new disk, too */
5253 struct btrfs_bio_stripe *new =
5254 bbio->stripes + index_where_to_add;
5255 struct btrfs_bio_stripe *old =
5258 new->physical = old->physical;
5259 new->length = old->length;
5260 new->dev = dev_replace->tgtdev;
5261 index_where_to_add++;
5265 num_stripes = index_where_to_add;
5266 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5267 dev_replace->tgtdev != NULL) {
5268 u64 srcdev_devid = dev_replace->srcdev->devid;
5269 int index_srcdev = 0;
5271 u64 physical_of_found = 0;
5274 * During the dev-replace procedure, the target drive can
5275 * also be used to read data in case it is needed to repair
5276 * a corrupt block elsewhere. This is possible if the
5277 * requested area is left of the left cursor. In this area,
5278 * the target drive is a full copy of the source drive.
5280 for (i = 0; i < num_stripes; i++) {
5281 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5283 * In case of DUP, in order to keep it
5284 * simple, only add the mirror with the
5285 * lowest physical address
5288 physical_of_found <=
5289 bbio->stripes[i].physical)
5293 physical_of_found = bbio->stripes[i].physical;
5297 u64 length = map->stripe_len;
5299 if (physical_of_found + length <=
5300 dev_replace->cursor_left) {
5301 struct btrfs_bio_stripe *tgtdev_stripe =
5302 bbio->stripes + num_stripes;
5304 tgtdev_stripe->physical = physical_of_found;
5305 tgtdev_stripe->length =
5306 bbio->stripes[index_srcdev].length;
5307 tgtdev_stripe->dev = dev_replace->tgtdev;
5315 bbio->num_stripes = num_stripes;
5316 bbio->max_errors = max_errors;
5317 bbio->mirror_num = mirror_num;
5320 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5321 * mirror_num == num_stripes + 1 && dev_replace target drive is
5322 * available as a mirror
5324 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5325 WARN_ON(num_stripes > 1);
5326 bbio->stripes[0].dev = dev_replace->tgtdev;
5327 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5328 bbio->mirror_num = map->num_stripes + 1;
5331 sort_parity_stripes(bbio, raid_map);
5332 *raid_map_ret = raid_map;
5335 if (dev_replace_is_ongoing)
5336 btrfs_dev_replace_unlock(dev_replace);
5337 free_extent_map(em);
5341 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5342 u64 logical, u64 *length,
5343 struct btrfs_bio **bbio_ret, int mirror_num)
5345 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5349 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5350 u64 chunk_start, u64 physical, u64 devid,
5351 u64 **logical, int *naddrs, int *stripe_len)
5353 struct extent_map_tree *em_tree = &map_tree->map_tree;
5354 struct extent_map *em;
5355 struct map_lookup *map;
5363 read_lock(&em_tree->lock);
5364 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5365 read_unlock(&em_tree->lock);
5368 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5373 if (em->start != chunk_start) {
5374 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5375 em->start, chunk_start);
5376 free_extent_map(em);
5379 map = (struct map_lookup *)em->bdev;
5382 rmap_len = map->stripe_len;
5384 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5385 do_div(length, map->num_stripes / map->sub_stripes);
5386 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5387 do_div(length, map->num_stripes);
5388 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5389 BTRFS_BLOCK_GROUP_RAID6)) {
5390 do_div(length, nr_data_stripes(map));
5391 rmap_len = map->stripe_len * nr_data_stripes(map);
5394 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5395 BUG_ON(!buf); /* -ENOMEM */
5397 for (i = 0; i < map->num_stripes; i++) {
5398 if (devid && map->stripes[i].dev->devid != devid)
5400 if (map->stripes[i].physical > physical ||
5401 map->stripes[i].physical + length <= physical)
5404 stripe_nr = physical - map->stripes[i].physical;
5405 do_div(stripe_nr, map->stripe_len);
5407 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5408 stripe_nr = stripe_nr * map->num_stripes + i;
5409 do_div(stripe_nr, map->sub_stripes);
5410 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5411 stripe_nr = stripe_nr * map->num_stripes + i;
5412 } /* else if RAID[56], multiply by nr_data_stripes().
5413 * Alternatively, just use rmap_len below instead of
5414 * map->stripe_len */
5416 bytenr = chunk_start + stripe_nr * rmap_len;
5417 WARN_ON(nr >= map->num_stripes);
5418 for (j = 0; j < nr; j++) {
5419 if (buf[j] == bytenr)
5423 WARN_ON(nr >= map->num_stripes);
5430 *stripe_len = rmap_len;
5432 free_extent_map(em);
5436 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5438 if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5439 bio_endio_nodec(bio, err);
5441 bio_endio(bio, err);
5445 static void btrfs_end_bio(struct bio *bio, int err)
5447 struct btrfs_bio *bbio = bio->bi_private;
5448 struct btrfs_device *dev = bbio->stripes[0].dev;
5449 int is_orig_bio = 0;
5452 atomic_inc(&bbio->error);
5453 if (err == -EIO || err == -EREMOTEIO) {
5454 unsigned int stripe_index =
5455 btrfs_io_bio(bio)->stripe_index;
5457 BUG_ON(stripe_index >= bbio->num_stripes);
5458 dev = bbio->stripes[stripe_index].dev;
5460 if (bio->bi_rw & WRITE)
5461 btrfs_dev_stat_inc(dev,
5462 BTRFS_DEV_STAT_WRITE_ERRS);
5464 btrfs_dev_stat_inc(dev,
5465 BTRFS_DEV_STAT_READ_ERRS);
5466 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5467 btrfs_dev_stat_inc(dev,
5468 BTRFS_DEV_STAT_FLUSH_ERRS);
5469 btrfs_dev_stat_print_on_error(dev);
5474 if (bio == bbio->orig_bio)
5477 btrfs_bio_counter_dec(bbio->fs_info);
5479 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5482 bio = bbio->orig_bio;
5485 bio->bi_private = bbio->private;
5486 bio->bi_end_io = bbio->end_io;
5487 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5488 /* only send an error to the higher layers if it is
5489 * beyond the tolerance of the btrfs bio
5491 if (atomic_read(&bbio->error) > bbio->max_errors) {
5495 * this bio is actually up to date, we didn't
5496 * go over the max number of errors
5498 set_bit(BIO_UPTODATE, &bio->bi_flags);
5502 btrfs_end_bbio(bbio, bio, err);
5503 } else if (!is_orig_bio) {
5509 * see run_scheduled_bios for a description of why bios are collected for
5512 * This will add one bio to the pending list for a device and make sure
5513 * the work struct is scheduled.
5515 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5516 struct btrfs_device *device,
5517 int rw, struct bio *bio)
5519 int should_queue = 1;
5520 struct btrfs_pending_bios *pending_bios;
5522 if (device->missing || !device->bdev) {
5523 bio_endio(bio, -EIO);
5527 /* don't bother with additional async steps for reads, right now */
5528 if (!(rw & REQ_WRITE)) {
5530 btrfsic_submit_bio(rw, bio);
5536 * nr_async_bios allows us to reliably return congestion to the
5537 * higher layers. Otherwise, the async bio makes it appear we have
5538 * made progress against dirty pages when we've really just put it
5539 * on a queue for later
5541 atomic_inc(&root->fs_info->nr_async_bios);
5542 WARN_ON(bio->bi_next);
5543 bio->bi_next = NULL;
5546 spin_lock(&device->io_lock);
5547 if (bio->bi_rw & REQ_SYNC)
5548 pending_bios = &device->pending_sync_bios;
5550 pending_bios = &device->pending_bios;
5552 if (pending_bios->tail)
5553 pending_bios->tail->bi_next = bio;
5555 pending_bios->tail = bio;
5556 if (!pending_bios->head)
5557 pending_bios->head = bio;
5558 if (device->running_pending)
5561 spin_unlock(&device->io_lock);
5564 btrfs_queue_work(root->fs_info->submit_workers,
5568 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5571 struct bio_vec *prev;
5572 struct request_queue *q = bdev_get_queue(bdev);
5573 unsigned int max_sectors = queue_max_sectors(q);
5574 struct bvec_merge_data bvm = {
5576 .bi_sector = sector,
5577 .bi_rw = bio->bi_rw,
5580 if (WARN_ON(bio->bi_vcnt == 0))
5583 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5584 if (bio_sectors(bio) > max_sectors)
5587 if (!q->merge_bvec_fn)
5590 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5591 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5596 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5597 struct bio *bio, u64 physical, int dev_nr,
5600 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5602 bio->bi_private = bbio;
5603 btrfs_io_bio(bio)->stripe_index = dev_nr;
5604 bio->bi_end_io = btrfs_end_bio;
5605 bio->bi_iter.bi_sector = physical >> 9;
5608 struct rcu_string *name;
5611 name = rcu_dereference(dev->name);
5612 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5613 "(%s id %llu), size=%u\n", rw,
5614 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5615 name->str, dev->devid, bio->bi_size);
5619 bio->bi_bdev = dev->bdev;
5621 btrfs_bio_counter_inc_noblocked(root->fs_info);
5624 btrfs_schedule_bio(root, dev, rw, bio);
5626 btrfsic_submit_bio(rw, bio);
5629 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5630 struct bio *first_bio, struct btrfs_device *dev,
5631 int dev_nr, int rw, int async)
5633 struct bio_vec *bvec = first_bio->bi_io_vec;
5635 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5636 u64 physical = bbio->stripes[dev_nr].physical;
5639 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5643 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5644 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5645 bvec->bv_offset) < bvec->bv_len) {
5646 u64 len = bio->bi_iter.bi_size;
5648 atomic_inc(&bbio->stripes_pending);
5649 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5657 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5661 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5663 atomic_inc(&bbio->error);
5664 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5665 /* Shoud be the original bio. */
5666 WARN_ON(bio != bbio->orig_bio);
5668 bio->bi_private = bbio->private;
5669 bio->bi_end_io = bbio->end_io;
5670 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5671 bio->bi_iter.bi_sector = logical >> 9;
5673 btrfs_end_bbio(bbio, bio, -EIO);
5677 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5678 int mirror_num, int async_submit)
5680 struct btrfs_device *dev;
5681 struct bio *first_bio = bio;
5682 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5685 u64 *raid_map = NULL;
5689 struct btrfs_bio *bbio = NULL;
5691 length = bio->bi_iter.bi_size;
5692 map_length = length;
5694 btrfs_bio_counter_inc_blocked(root->fs_info);
5695 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5696 mirror_num, &raid_map);
5698 btrfs_bio_counter_dec(root->fs_info);
5702 total_devs = bbio->num_stripes;
5703 bbio->orig_bio = first_bio;
5704 bbio->private = first_bio->bi_private;
5705 bbio->end_io = first_bio->bi_end_io;
5706 bbio->fs_info = root->fs_info;
5707 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5710 /* In this case, map_length has been set to the length of
5711 a single stripe; not the whole write */
5713 ret = raid56_parity_write(root, bio, bbio,
5714 raid_map, map_length);
5716 ret = raid56_parity_recover(root, bio, bbio,
5717 raid_map, map_length,
5721 * FIXME, replace dosen't support raid56 yet, please fix
5724 btrfs_bio_counter_dec(root->fs_info);
5728 if (map_length < length) {
5729 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5730 logical, length, map_length);
5734 while (dev_nr < total_devs) {
5735 dev = bbio->stripes[dev_nr].dev;
5736 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5737 bbio_error(bbio, first_bio, logical);
5743 * Check and see if we're ok with this bio based on it's size
5744 * and offset with the given device.
5746 if (!bio_size_ok(dev->bdev, first_bio,
5747 bbio->stripes[dev_nr].physical >> 9)) {
5748 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5749 dev_nr, rw, async_submit);
5755 if (dev_nr < total_devs - 1) {
5756 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5757 BUG_ON(!bio); /* -ENOMEM */
5760 bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5763 submit_stripe_bio(root, bbio, bio,
5764 bbio->stripes[dev_nr].physical, dev_nr, rw,
5768 btrfs_bio_counter_dec(root->fs_info);
5772 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5775 struct btrfs_device *device;
5776 struct btrfs_fs_devices *cur_devices;
5778 cur_devices = fs_info->fs_devices;
5779 while (cur_devices) {
5781 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5782 device = __find_device(&cur_devices->devices,
5787 cur_devices = cur_devices->seed;
5792 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5793 u64 devid, u8 *dev_uuid)
5795 struct btrfs_device *device;
5796 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5798 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5802 list_add(&device->dev_list, &fs_devices->devices);
5803 device->fs_devices = fs_devices;
5804 fs_devices->num_devices++;
5806 device->missing = 1;
5807 fs_devices->missing_devices++;
5813 * btrfs_alloc_device - allocate struct btrfs_device
5814 * @fs_info: used only for generating a new devid, can be NULL if
5815 * devid is provided (i.e. @devid != NULL).
5816 * @devid: a pointer to devid for this device. If NULL a new devid
5818 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5821 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5822 * on error. Returned struct is not linked onto any lists and can be
5823 * destroyed with kfree() right away.
5825 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5829 struct btrfs_device *dev;
5832 if (WARN_ON(!devid && !fs_info))
5833 return ERR_PTR(-EINVAL);
5835 dev = __alloc_device();
5844 ret = find_next_devid(fs_info, &tmp);
5847 return ERR_PTR(ret);
5853 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5855 generate_random_uuid(dev->uuid);
5857 btrfs_init_work(&dev->work, btrfs_submit_helper,
5858 pending_bios_fn, NULL, NULL);
5863 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5864 struct extent_buffer *leaf,
5865 struct btrfs_chunk *chunk)
5867 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5868 struct map_lookup *map;
5869 struct extent_map *em;
5873 u8 uuid[BTRFS_UUID_SIZE];
5878 logical = key->offset;
5879 length = btrfs_chunk_length(leaf, chunk);
5881 read_lock(&map_tree->map_tree.lock);
5882 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5883 read_unlock(&map_tree->map_tree.lock);
5885 /* already mapped? */
5886 if (em && em->start <= logical && em->start + em->len > logical) {
5887 free_extent_map(em);
5890 free_extent_map(em);
5893 em = alloc_extent_map();
5896 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5897 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5899 free_extent_map(em);
5903 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5904 em->bdev = (struct block_device *)map;
5905 em->start = logical;
5908 em->block_start = 0;
5909 em->block_len = em->len;
5911 map->num_stripes = num_stripes;
5912 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5913 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5914 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5915 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5916 map->type = btrfs_chunk_type(leaf, chunk);
5917 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5918 for (i = 0; i < num_stripes; i++) {
5919 map->stripes[i].physical =
5920 btrfs_stripe_offset_nr(leaf, chunk, i);
5921 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5922 read_extent_buffer(leaf, uuid, (unsigned long)
5923 btrfs_stripe_dev_uuid_nr(chunk, i),
5925 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5927 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5928 free_extent_map(em);
5931 if (!map->stripes[i].dev) {
5932 map->stripes[i].dev =
5933 add_missing_dev(root, devid, uuid);
5934 if (!map->stripes[i].dev) {
5935 free_extent_map(em);
5939 map->stripes[i].dev->in_fs_metadata = 1;
5942 write_lock(&map_tree->map_tree.lock);
5943 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5944 write_unlock(&map_tree->map_tree.lock);
5945 BUG_ON(ret); /* Tree corruption */
5946 free_extent_map(em);
5951 static void fill_device_from_item(struct extent_buffer *leaf,
5952 struct btrfs_dev_item *dev_item,
5953 struct btrfs_device *device)
5957 device->devid = btrfs_device_id(leaf, dev_item);
5958 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5959 device->total_bytes = device->disk_total_bytes;
5960 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5961 device->type = btrfs_device_type(leaf, dev_item);
5962 device->io_align = btrfs_device_io_align(leaf, dev_item);
5963 device->io_width = btrfs_device_io_width(leaf, dev_item);
5964 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5965 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5966 device->is_tgtdev_for_dev_replace = 0;
5968 ptr = btrfs_device_uuid(dev_item);
5969 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5972 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5974 struct btrfs_fs_devices *fs_devices;
5977 BUG_ON(!mutex_is_locked(&uuid_mutex));
5979 fs_devices = root->fs_info->fs_devices->seed;
5980 while (fs_devices) {
5981 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5985 fs_devices = fs_devices->seed;
5988 fs_devices = find_fsid(fsid);
5994 fs_devices = clone_fs_devices(fs_devices);
5995 if (IS_ERR(fs_devices)) {
5996 ret = PTR_ERR(fs_devices);
6000 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6001 root->fs_info->bdev_holder);
6003 free_fs_devices(fs_devices);
6007 if (!fs_devices->seeding) {
6008 __btrfs_close_devices(fs_devices);
6009 free_fs_devices(fs_devices);
6014 fs_devices->seed = root->fs_info->fs_devices->seed;
6015 root->fs_info->fs_devices->seed = fs_devices;
6020 static int read_one_dev(struct btrfs_root *root,
6021 struct extent_buffer *leaf,
6022 struct btrfs_dev_item *dev_item)
6024 struct btrfs_device *device;
6027 u8 fs_uuid[BTRFS_UUID_SIZE];
6028 u8 dev_uuid[BTRFS_UUID_SIZE];
6030 devid = btrfs_device_id(leaf, dev_item);
6031 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6033 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6036 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6037 ret = open_seed_devices(root, fs_uuid);
6038 if (ret && !btrfs_test_opt(root, DEGRADED))
6042 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6043 if (!device || !device->bdev) {
6044 if (!btrfs_test_opt(root, DEGRADED))
6048 btrfs_warn(root->fs_info, "devid %llu missing", devid);
6049 device = add_missing_dev(root, devid, dev_uuid);
6052 } else if (!device->missing) {
6054 * this happens when a device that was properly setup
6055 * in the device info lists suddenly goes bad.
6056 * device->bdev is NULL, and so we have to set
6057 * device->missing to one here
6059 root->fs_info->fs_devices->missing_devices++;
6060 device->missing = 1;
6064 if (device->fs_devices != root->fs_info->fs_devices) {
6065 BUG_ON(device->writeable);
6066 if (device->generation !=
6067 btrfs_device_generation(leaf, dev_item))
6071 fill_device_from_item(leaf, dev_item, device);
6072 device->in_fs_metadata = 1;
6073 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6074 device->fs_devices->total_rw_bytes += device->total_bytes;
6075 spin_lock(&root->fs_info->free_chunk_lock);
6076 root->fs_info->free_chunk_space += device->total_bytes -
6078 spin_unlock(&root->fs_info->free_chunk_lock);
6084 int btrfs_read_sys_array(struct btrfs_root *root)
6086 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6087 struct extent_buffer *sb;
6088 struct btrfs_disk_key *disk_key;
6089 struct btrfs_chunk *chunk;
6091 unsigned long sb_ptr;
6097 struct btrfs_key key;
6099 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6100 BTRFS_SUPER_INFO_SIZE);
6103 btrfs_set_buffer_uptodate(sb);
6104 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6106 * The sb extent buffer is artifical and just used to read the system array.
6107 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6108 * pages up-to-date when the page is larger: extent does not cover the
6109 * whole page and consequently check_page_uptodate does not find all
6110 * the page's extents up-to-date (the hole beyond sb),
6111 * write_extent_buffer then triggers a WARN_ON.
6113 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6114 * but sb spans only this function. Add an explicit SetPageUptodate call
6115 * to silence the warning eg. on PowerPC 64.
6117 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6118 SetPageUptodate(sb->pages[0]);
6120 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6121 array_size = btrfs_super_sys_array_size(super_copy);
6123 ptr = super_copy->sys_chunk_array;
6124 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6127 while (cur < array_size) {
6128 disk_key = (struct btrfs_disk_key *)ptr;
6129 btrfs_disk_key_to_cpu(&key, disk_key);
6131 len = sizeof(*disk_key); ptr += len;
6135 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6136 chunk = (struct btrfs_chunk *)sb_ptr;
6137 ret = read_one_chunk(root, &key, sb, chunk);
6140 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6141 len = btrfs_chunk_item_size(num_stripes);
6150 free_extent_buffer(sb);
6154 int btrfs_read_chunk_tree(struct btrfs_root *root)
6156 struct btrfs_path *path;
6157 struct extent_buffer *leaf;
6158 struct btrfs_key key;
6159 struct btrfs_key found_key;
6163 root = root->fs_info->chunk_root;
6165 path = btrfs_alloc_path();
6169 mutex_lock(&uuid_mutex);
6173 * Read all device items, and then all the chunk items. All
6174 * device items are found before any chunk item (their object id
6175 * is smaller than the lowest possible object id for a chunk
6176 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6178 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6181 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6185 leaf = path->nodes[0];
6186 slot = path->slots[0];
6187 if (slot >= btrfs_header_nritems(leaf)) {
6188 ret = btrfs_next_leaf(root, path);
6195 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6196 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6197 struct btrfs_dev_item *dev_item;
6198 dev_item = btrfs_item_ptr(leaf, slot,
6199 struct btrfs_dev_item);
6200 ret = read_one_dev(root, leaf, dev_item);
6203 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6204 struct btrfs_chunk *chunk;
6205 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6206 ret = read_one_chunk(root, &found_key, leaf, chunk);
6214 unlock_chunks(root);
6215 mutex_unlock(&uuid_mutex);
6217 btrfs_free_path(path);
6221 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6223 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6224 struct btrfs_device *device;
6226 while (fs_devices) {
6227 mutex_lock(&fs_devices->device_list_mutex);
6228 list_for_each_entry(device, &fs_devices->devices, dev_list)
6229 device->dev_root = fs_info->dev_root;
6230 mutex_unlock(&fs_devices->device_list_mutex);
6232 fs_devices = fs_devices->seed;
6236 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6240 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6241 btrfs_dev_stat_reset(dev, i);
6244 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6246 struct btrfs_key key;
6247 struct btrfs_key found_key;
6248 struct btrfs_root *dev_root = fs_info->dev_root;
6249 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6250 struct extent_buffer *eb;
6253 struct btrfs_device *device;
6254 struct btrfs_path *path = NULL;
6257 path = btrfs_alloc_path();
6263 mutex_lock(&fs_devices->device_list_mutex);
6264 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6266 struct btrfs_dev_stats_item *ptr;
6269 key.type = BTRFS_DEV_STATS_KEY;
6270 key.offset = device->devid;
6271 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6273 __btrfs_reset_dev_stats(device);
6274 device->dev_stats_valid = 1;
6275 btrfs_release_path(path);
6278 slot = path->slots[0];
6279 eb = path->nodes[0];
6280 btrfs_item_key_to_cpu(eb, &found_key, slot);
6281 item_size = btrfs_item_size_nr(eb, slot);
6283 ptr = btrfs_item_ptr(eb, slot,
6284 struct btrfs_dev_stats_item);
6286 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6287 if (item_size >= (1 + i) * sizeof(__le64))
6288 btrfs_dev_stat_set(device, i,
6289 btrfs_dev_stats_value(eb, ptr, i));
6291 btrfs_dev_stat_reset(device, i);
6294 device->dev_stats_valid = 1;
6295 btrfs_dev_stat_print_on_load(device);
6296 btrfs_release_path(path);
6298 mutex_unlock(&fs_devices->device_list_mutex);
6301 btrfs_free_path(path);
6302 return ret < 0 ? ret : 0;
6305 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6306 struct btrfs_root *dev_root,
6307 struct btrfs_device *device)
6309 struct btrfs_path *path;
6310 struct btrfs_key key;
6311 struct extent_buffer *eb;
6312 struct btrfs_dev_stats_item *ptr;
6317 key.type = BTRFS_DEV_STATS_KEY;
6318 key.offset = device->devid;
6320 path = btrfs_alloc_path();
6322 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6324 printk_in_rcu(KERN_WARNING "BTRFS: "
6325 "error %d while searching for dev_stats item for device %s!\n",
6326 ret, rcu_str_deref(device->name));
6331 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6332 /* need to delete old one and insert a new one */
6333 ret = btrfs_del_item(trans, dev_root, path);
6335 printk_in_rcu(KERN_WARNING "BTRFS: "
6336 "delete too small dev_stats item for device %s failed %d!\n",
6337 rcu_str_deref(device->name), ret);
6344 /* need to insert a new item */
6345 btrfs_release_path(path);
6346 ret = btrfs_insert_empty_item(trans, dev_root, path,
6347 &key, sizeof(*ptr));
6349 printk_in_rcu(KERN_WARNING "BTRFS: "
6350 "insert dev_stats item for device %s failed %d!\n",
6351 rcu_str_deref(device->name), ret);
6356 eb = path->nodes[0];
6357 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6358 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6359 btrfs_set_dev_stats_value(eb, ptr, i,
6360 btrfs_dev_stat_read(device, i));
6361 btrfs_mark_buffer_dirty(eb);
6364 btrfs_free_path(path);
6369 * called from commit_transaction. Writes all changed device stats to disk.
6371 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6372 struct btrfs_fs_info *fs_info)
6374 struct btrfs_root *dev_root = fs_info->dev_root;
6375 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6376 struct btrfs_device *device;
6379 mutex_lock(&fs_devices->device_list_mutex);
6380 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6381 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6384 ret = update_dev_stat_item(trans, dev_root, device);
6386 device->dev_stats_dirty = 0;
6388 mutex_unlock(&fs_devices->device_list_mutex);
6393 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6395 btrfs_dev_stat_inc(dev, index);
6396 btrfs_dev_stat_print_on_error(dev);
6399 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6401 if (!dev->dev_stats_valid)
6403 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6404 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6405 rcu_str_deref(dev->name),
6406 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6407 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6408 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6409 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6410 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6413 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6417 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6418 if (btrfs_dev_stat_read(dev, i) != 0)
6420 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6421 return; /* all values == 0, suppress message */
6423 printk_in_rcu(KERN_INFO "BTRFS: "
6424 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6425 rcu_str_deref(dev->name),
6426 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6427 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6428 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6429 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6430 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6433 int btrfs_get_dev_stats(struct btrfs_root *root,
6434 struct btrfs_ioctl_get_dev_stats *stats)
6436 struct btrfs_device *dev;
6437 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6440 mutex_lock(&fs_devices->device_list_mutex);
6441 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6442 mutex_unlock(&fs_devices->device_list_mutex);
6445 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6447 } else if (!dev->dev_stats_valid) {
6448 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6450 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6451 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6452 if (stats->nr_items > i)
6454 btrfs_dev_stat_read_and_reset(dev, i);
6456 btrfs_dev_stat_reset(dev, i);
6459 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6460 if (stats->nr_items > i)
6461 stats->values[i] = btrfs_dev_stat_read(dev, i);
6463 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6464 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6468 int btrfs_scratch_superblock(struct btrfs_device *device)
6470 struct buffer_head *bh;
6471 struct btrfs_super_block *disk_super;
6473 bh = btrfs_read_dev_super(device->bdev);
6476 disk_super = (struct btrfs_super_block *)bh->b_data;
6478 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6479 set_buffer_dirty(bh);
6480 sync_dirty_buffer(bh);