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>
33 #include "extent_map.h"
35 #include "transaction.h"
36 #include "print-tree.h"
39 #include "async-thread.h"
40 #include "check-integrity.h"
41 #include "rcu-string.h"
43 #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("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_requeue_work(&device->work);
422 /* unplug every 64 requests just for good measure */
423 if (batch_run % 64 == 0) {
424 blk_finish_plug(&plug);
425 blk_start_plug(&plug);
434 spin_lock(&device->io_lock);
435 if (device->pending_bios.head || device->pending_sync_bios.head)
437 spin_unlock(&device->io_lock);
440 blk_finish_plug(&plug);
443 static void pending_bios_fn(struct btrfs_work *work)
445 struct btrfs_device *device;
447 device = container_of(work, struct btrfs_device, work);
448 run_scheduled_bios(device);
451 static noinline int device_list_add(const char *path,
452 struct btrfs_super_block *disk_super,
453 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
455 struct btrfs_device *device;
456 struct btrfs_fs_devices *fs_devices;
457 struct rcu_string *name;
458 u64 found_transid = btrfs_super_generation(disk_super);
460 fs_devices = find_fsid(disk_super->fsid);
462 fs_devices = alloc_fs_devices(disk_super->fsid);
463 if (IS_ERR(fs_devices))
464 return PTR_ERR(fs_devices);
466 list_add(&fs_devices->list, &fs_uuids);
467 fs_devices->latest_devid = devid;
468 fs_devices->latest_trans = found_transid;
472 device = __find_device(&fs_devices->devices, devid,
473 disk_super->dev_item.uuid);
476 if (fs_devices->opened)
479 device = btrfs_alloc_device(NULL, &devid,
480 disk_super->dev_item.uuid);
481 if (IS_ERR(device)) {
482 /* we can safely leave the fs_devices entry around */
483 return PTR_ERR(device);
486 name = rcu_string_strdup(path, GFP_NOFS);
491 rcu_assign_pointer(device->name, name);
493 mutex_lock(&fs_devices->device_list_mutex);
494 list_add_rcu(&device->dev_list, &fs_devices->devices);
495 mutex_unlock(&fs_devices->device_list_mutex);
497 device->fs_devices = fs_devices;
498 fs_devices->num_devices++;
499 } else if (!device->name || strcmp(device->name->str, path)) {
500 name = rcu_string_strdup(path, GFP_NOFS);
503 rcu_string_free(device->name);
504 rcu_assign_pointer(device->name, name);
505 if (device->missing) {
506 fs_devices->missing_devices--;
511 if (found_transid > fs_devices->latest_trans) {
512 fs_devices->latest_devid = devid;
513 fs_devices->latest_trans = found_transid;
515 *fs_devices_ret = fs_devices;
519 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
521 struct btrfs_fs_devices *fs_devices;
522 struct btrfs_device *device;
523 struct btrfs_device *orig_dev;
525 fs_devices = alloc_fs_devices(orig->fsid);
526 if (IS_ERR(fs_devices))
529 fs_devices->latest_devid = orig->latest_devid;
530 fs_devices->latest_trans = orig->latest_trans;
531 fs_devices->total_devices = orig->total_devices;
533 /* We have held the volume lock, it is safe to get the devices. */
534 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
535 struct rcu_string *name;
537 device = btrfs_alloc_device(NULL, &orig_dev->devid,
543 * This is ok to do without rcu read locked because we hold the
544 * uuid mutex so nothing we touch in here is going to disappear.
546 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
551 rcu_assign_pointer(device->name, name);
553 list_add(&device->dev_list, &fs_devices->devices);
554 device->fs_devices = fs_devices;
555 fs_devices->num_devices++;
559 free_fs_devices(fs_devices);
560 return ERR_PTR(-ENOMEM);
563 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
564 struct btrfs_fs_devices *fs_devices, int step)
566 struct btrfs_device *device, *next;
568 struct block_device *latest_bdev = NULL;
569 u64 latest_devid = 0;
570 u64 latest_transid = 0;
572 mutex_lock(&uuid_mutex);
574 /* This is the initialized path, it is safe to release the devices. */
575 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
576 if (device->in_fs_metadata) {
577 if (!device->is_tgtdev_for_dev_replace &&
579 device->generation > latest_transid)) {
580 latest_devid = device->devid;
581 latest_transid = device->generation;
582 latest_bdev = device->bdev;
587 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
589 * In the first step, keep the device which has
590 * the correct fsid and the devid that is used
591 * for the dev_replace procedure.
592 * In the second step, the dev_replace state is
593 * read from the device tree and it is known
594 * whether the procedure is really active or
595 * not, which means whether this device is
596 * used or whether it should be removed.
598 if (step == 0 || device->is_tgtdev_for_dev_replace) {
603 blkdev_put(device->bdev, device->mode);
605 fs_devices->open_devices--;
607 if (device->writeable) {
608 list_del_init(&device->dev_alloc_list);
609 device->writeable = 0;
610 if (!device->is_tgtdev_for_dev_replace)
611 fs_devices->rw_devices--;
613 list_del_init(&device->dev_list);
614 fs_devices->num_devices--;
615 rcu_string_free(device->name);
619 if (fs_devices->seed) {
620 fs_devices = fs_devices->seed;
624 fs_devices->latest_bdev = latest_bdev;
625 fs_devices->latest_devid = latest_devid;
626 fs_devices->latest_trans = latest_transid;
628 mutex_unlock(&uuid_mutex);
631 static void __free_device(struct work_struct *work)
633 struct btrfs_device *device;
635 device = container_of(work, struct btrfs_device, rcu_work);
638 blkdev_put(device->bdev, device->mode);
640 rcu_string_free(device->name);
644 static void free_device(struct rcu_head *head)
646 struct btrfs_device *device;
648 device = container_of(head, struct btrfs_device, rcu);
650 INIT_WORK(&device->rcu_work, __free_device);
651 schedule_work(&device->rcu_work);
654 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
656 struct btrfs_device *device;
658 if (--fs_devices->opened > 0)
661 mutex_lock(&fs_devices->device_list_mutex);
662 list_for_each_entry(device, &fs_devices->devices, dev_list) {
663 struct btrfs_device *new_device;
664 struct rcu_string *name;
667 fs_devices->open_devices--;
669 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
670 list_del_init(&device->dev_alloc_list);
671 fs_devices->rw_devices--;
674 if (device->can_discard)
675 fs_devices->num_can_discard--;
677 new_device = btrfs_alloc_device(NULL, &device->devid,
679 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
681 /* Safe because we are under uuid_mutex */
683 name = rcu_string_strdup(device->name->str, GFP_NOFS);
684 BUG_ON(!name); /* -ENOMEM */
685 rcu_assign_pointer(new_device->name, name);
688 list_replace_rcu(&device->dev_list, &new_device->dev_list);
689 new_device->fs_devices = device->fs_devices;
691 call_rcu(&device->rcu, free_device);
693 mutex_unlock(&fs_devices->device_list_mutex);
695 WARN_ON(fs_devices->open_devices);
696 WARN_ON(fs_devices->rw_devices);
697 fs_devices->opened = 0;
698 fs_devices->seeding = 0;
703 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
705 struct btrfs_fs_devices *seed_devices = NULL;
708 mutex_lock(&uuid_mutex);
709 ret = __btrfs_close_devices(fs_devices);
710 if (!fs_devices->opened) {
711 seed_devices = fs_devices->seed;
712 fs_devices->seed = NULL;
714 mutex_unlock(&uuid_mutex);
716 while (seed_devices) {
717 fs_devices = seed_devices;
718 seed_devices = fs_devices->seed;
719 __btrfs_close_devices(fs_devices);
720 free_fs_devices(fs_devices);
723 * Wait for rcu kworkers under __btrfs_close_devices
724 * to finish all blkdev_puts so device is really
725 * free when umount is done.
731 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
732 fmode_t flags, void *holder)
734 struct request_queue *q;
735 struct block_device *bdev;
736 struct list_head *head = &fs_devices->devices;
737 struct btrfs_device *device;
738 struct block_device *latest_bdev = NULL;
739 struct buffer_head *bh;
740 struct btrfs_super_block *disk_super;
741 u64 latest_devid = 0;
742 u64 latest_transid = 0;
749 list_for_each_entry(device, head, dev_list) {
755 /* Just open everything we can; ignore failures here */
756 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
760 disk_super = (struct btrfs_super_block *)bh->b_data;
761 devid = btrfs_stack_device_id(&disk_super->dev_item);
762 if (devid != device->devid)
765 if (memcmp(device->uuid, disk_super->dev_item.uuid,
769 device->generation = btrfs_super_generation(disk_super);
770 if (!latest_transid || device->generation > latest_transid) {
771 latest_devid = devid;
772 latest_transid = device->generation;
776 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
777 device->writeable = 0;
779 device->writeable = !bdev_read_only(bdev);
783 q = bdev_get_queue(bdev);
784 if (blk_queue_discard(q)) {
785 device->can_discard = 1;
786 fs_devices->num_can_discard++;
790 device->in_fs_metadata = 0;
791 device->mode = flags;
793 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
794 fs_devices->rotating = 1;
796 fs_devices->open_devices++;
797 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
798 fs_devices->rw_devices++;
799 list_add(&device->dev_alloc_list,
800 &fs_devices->alloc_list);
807 blkdev_put(bdev, flags);
810 if (fs_devices->open_devices == 0) {
814 fs_devices->seeding = seeding;
815 fs_devices->opened = 1;
816 fs_devices->latest_bdev = latest_bdev;
817 fs_devices->latest_devid = latest_devid;
818 fs_devices->latest_trans = latest_transid;
819 fs_devices->total_rw_bytes = 0;
824 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
825 fmode_t flags, void *holder)
829 mutex_lock(&uuid_mutex);
830 if (fs_devices->opened) {
831 fs_devices->opened++;
834 ret = __btrfs_open_devices(fs_devices, flags, holder);
836 mutex_unlock(&uuid_mutex);
841 * Look for a btrfs signature on a device. This may be called out of the mount path
842 * and we are not allowed to call set_blocksize during the scan. The superblock
843 * is read via pagecache
845 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
846 struct btrfs_fs_devices **fs_devices_ret)
848 struct btrfs_super_block *disk_super;
849 struct block_device *bdev;
860 * we would like to check all the supers, but that would make
861 * a btrfs mount succeed after a mkfs from a different FS.
862 * So, we need to add a special mount option to scan for
863 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
865 bytenr = btrfs_sb_offset(0);
867 mutex_lock(&uuid_mutex);
869 bdev = blkdev_get_by_path(path, flags, holder);
876 /* make sure our super fits in the device */
877 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
880 /* make sure our super fits in the page */
881 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
884 /* make sure our super doesn't straddle pages on disk */
885 index = bytenr >> PAGE_CACHE_SHIFT;
886 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
889 /* pull in the page with our super */
890 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
893 if (IS_ERR_OR_NULL(page))
898 /* align our pointer to the offset of the super block */
899 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
901 if (btrfs_super_bytenr(disk_super) != bytenr ||
902 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
905 devid = btrfs_stack_device_id(&disk_super->dev_item);
906 transid = btrfs_super_generation(disk_super);
907 total_devices = btrfs_super_num_devices(disk_super);
909 if (disk_super->label[0]) {
910 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
911 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
912 printk(KERN_INFO "device label %s ", disk_super->label);
914 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
917 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
919 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
920 if (!ret && fs_devices_ret)
921 (*fs_devices_ret)->total_devices = total_devices;
925 page_cache_release(page);
928 blkdev_put(bdev, flags);
930 mutex_unlock(&uuid_mutex);
934 /* helper to account the used device space in the range */
935 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
936 u64 end, u64 *length)
938 struct btrfs_key key;
939 struct btrfs_root *root = device->dev_root;
940 struct btrfs_dev_extent *dev_extent;
941 struct btrfs_path *path;
945 struct extent_buffer *l;
949 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
952 path = btrfs_alloc_path();
957 key.objectid = device->devid;
959 key.type = BTRFS_DEV_EXTENT_KEY;
961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
965 ret = btrfs_previous_item(root, path, key.objectid, key.type);
972 slot = path->slots[0];
973 if (slot >= btrfs_header_nritems(l)) {
974 ret = btrfs_next_leaf(root, path);
982 btrfs_item_key_to_cpu(l, &key, slot);
984 if (key.objectid < device->devid)
987 if (key.objectid > device->devid)
990 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
993 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
994 extent_end = key.offset + btrfs_dev_extent_length(l,
996 if (key.offset <= start && extent_end > end) {
997 *length = end - start + 1;
999 } else if (key.offset <= start && extent_end > start)
1000 *length += extent_end - start;
1001 else if (key.offset > start && extent_end <= end)
1002 *length += extent_end - key.offset;
1003 else if (key.offset > start && key.offset <= end) {
1004 *length += end - key.offset + 1;
1006 } else if (key.offset > end)
1014 btrfs_free_path(path);
1018 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1019 struct btrfs_device *device,
1020 u64 *start, u64 len)
1022 struct extent_map *em;
1025 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1026 struct map_lookup *map;
1029 map = (struct map_lookup *)em->bdev;
1030 for (i = 0; i < map->num_stripes; i++) {
1031 if (map->stripes[i].dev != device)
1033 if (map->stripes[i].physical >= *start + len ||
1034 map->stripes[i].physical + em->orig_block_len <=
1037 *start = map->stripes[i].physical +
1048 * find_free_dev_extent - find free space in the specified device
1049 * @device: the device which we search the free space in
1050 * @num_bytes: the size of the free space that we need
1051 * @start: store the start of the free space.
1052 * @len: the size of the free space. that we find, or the size of the max
1053 * free space if we don't find suitable free space
1055 * this uses a pretty simple search, the expectation is that it is
1056 * called very infrequently and that a given device has a small number
1059 * @start is used to store the start of the free space if we find. But if we
1060 * don't find suitable free space, it will be used to store the start position
1061 * of the max free space.
1063 * @len is used to store the size of the free space that we find.
1064 * But if we don't find suitable free space, it is used to store the size of
1065 * the max free space.
1067 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1068 struct btrfs_device *device, u64 num_bytes,
1069 u64 *start, u64 *len)
1071 struct btrfs_key key;
1072 struct btrfs_root *root = device->dev_root;
1073 struct btrfs_dev_extent *dev_extent;
1074 struct btrfs_path *path;
1080 u64 search_end = device->total_bytes;
1083 struct extent_buffer *l;
1085 /* FIXME use last free of some kind */
1087 /* we don't want to overwrite the superblock on the drive,
1088 * so we make sure to start at an offset of at least 1MB
1090 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1092 path = btrfs_alloc_path();
1096 max_hole_start = search_start;
1100 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1106 path->search_commit_root = 1;
1107 path->skip_locking = 1;
1109 key.objectid = device->devid;
1110 key.offset = search_start;
1111 key.type = BTRFS_DEV_EXTENT_KEY;
1113 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1117 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1124 slot = path->slots[0];
1125 if (slot >= btrfs_header_nritems(l)) {
1126 ret = btrfs_next_leaf(root, path);
1134 btrfs_item_key_to_cpu(l, &key, slot);
1136 if (key.objectid < device->devid)
1139 if (key.objectid > device->devid)
1142 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1145 if (key.offset > search_start) {
1146 hole_size = key.offset - search_start;
1149 * Have to check before we set max_hole_start, otherwise
1150 * we could end up sending back this offset anyway.
1152 if (contains_pending_extent(trans, device,
1157 if (hole_size > max_hole_size) {
1158 max_hole_start = search_start;
1159 max_hole_size = hole_size;
1163 * If this free space is greater than which we need,
1164 * it must be the max free space that we have found
1165 * until now, so max_hole_start must point to the start
1166 * of this free space and the length of this free space
1167 * is stored in max_hole_size. Thus, we return
1168 * max_hole_start and max_hole_size and go back to the
1171 if (hole_size >= num_bytes) {
1177 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1178 extent_end = key.offset + btrfs_dev_extent_length(l,
1180 if (extent_end > search_start)
1181 search_start = extent_end;
1188 * At this point, search_start should be the end of
1189 * allocated dev extents, and when shrinking the device,
1190 * search_end may be smaller than search_start.
1192 if (search_end > search_start)
1193 hole_size = search_end - search_start;
1195 if (hole_size > max_hole_size) {
1196 max_hole_start = search_start;
1197 max_hole_size = hole_size;
1200 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1201 btrfs_release_path(path);
1206 if (hole_size < num_bytes)
1212 btrfs_free_path(path);
1213 *start = max_hole_start;
1215 *len = max_hole_size;
1219 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1220 struct btrfs_device *device,
1224 struct btrfs_path *path;
1225 struct btrfs_root *root = device->dev_root;
1226 struct btrfs_key key;
1227 struct btrfs_key found_key;
1228 struct extent_buffer *leaf = NULL;
1229 struct btrfs_dev_extent *extent = NULL;
1231 path = btrfs_alloc_path();
1235 key.objectid = device->devid;
1237 key.type = BTRFS_DEV_EXTENT_KEY;
1239 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1241 ret = btrfs_previous_item(root, path, key.objectid,
1242 BTRFS_DEV_EXTENT_KEY);
1245 leaf = path->nodes[0];
1246 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1247 extent = btrfs_item_ptr(leaf, path->slots[0],
1248 struct btrfs_dev_extent);
1249 BUG_ON(found_key.offset > start || found_key.offset +
1250 btrfs_dev_extent_length(leaf, extent) < start);
1252 btrfs_release_path(path);
1254 } else if (ret == 0) {
1255 leaf = path->nodes[0];
1256 extent = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_dev_extent);
1259 btrfs_error(root->fs_info, ret, "Slot search failed");
1263 if (device->bytes_used > 0) {
1264 u64 len = btrfs_dev_extent_length(leaf, extent);
1265 device->bytes_used -= len;
1266 spin_lock(&root->fs_info->free_chunk_lock);
1267 root->fs_info->free_chunk_space += len;
1268 spin_unlock(&root->fs_info->free_chunk_lock);
1270 ret = btrfs_del_item(trans, root, path);
1272 btrfs_error(root->fs_info, ret,
1273 "Failed to remove dev extent item");
1276 btrfs_free_path(path);
1280 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1281 struct btrfs_device *device,
1282 u64 chunk_tree, u64 chunk_objectid,
1283 u64 chunk_offset, u64 start, u64 num_bytes)
1286 struct btrfs_path *path;
1287 struct btrfs_root *root = device->dev_root;
1288 struct btrfs_dev_extent *extent;
1289 struct extent_buffer *leaf;
1290 struct btrfs_key key;
1292 WARN_ON(!device->in_fs_metadata);
1293 WARN_ON(device->is_tgtdev_for_dev_replace);
1294 path = btrfs_alloc_path();
1298 key.objectid = device->devid;
1300 key.type = BTRFS_DEV_EXTENT_KEY;
1301 ret = btrfs_insert_empty_item(trans, root, path, &key,
1306 leaf = path->nodes[0];
1307 extent = btrfs_item_ptr(leaf, path->slots[0],
1308 struct btrfs_dev_extent);
1309 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1310 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1311 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1313 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1314 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1317 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1318 btrfs_mark_buffer_dirty(leaf);
1320 btrfs_free_path(path);
1324 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1326 struct extent_map_tree *em_tree;
1327 struct extent_map *em;
1331 em_tree = &fs_info->mapping_tree.map_tree;
1332 read_lock(&em_tree->lock);
1333 n = rb_last(&em_tree->map);
1335 em = rb_entry(n, struct extent_map, rb_node);
1336 ret = em->start + em->len;
1338 read_unlock(&em_tree->lock);
1343 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1347 struct btrfs_key key;
1348 struct btrfs_key found_key;
1349 struct btrfs_path *path;
1351 path = btrfs_alloc_path();
1355 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1356 key.type = BTRFS_DEV_ITEM_KEY;
1357 key.offset = (u64)-1;
1359 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1363 BUG_ON(ret == 0); /* Corruption */
1365 ret = btrfs_previous_item(fs_info->chunk_root, path,
1366 BTRFS_DEV_ITEMS_OBJECTID,
1367 BTRFS_DEV_ITEM_KEY);
1371 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1373 *devid_ret = found_key.offset + 1;
1377 btrfs_free_path(path);
1382 * the device information is stored in the chunk root
1383 * the btrfs_device struct should be fully filled in
1385 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1386 struct btrfs_root *root,
1387 struct btrfs_device *device)
1390 struct btrfs_path *path;
1391 struct btrfs_dev_item *dev_item;
1392 struct extent_buffer *leaf;
1393 struct btrfs_key key;
1396 root = root->fs_info->chunk_root;
1398 path = btrfs_alloc_path();
1402 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1403 key.type = BTRFS_DEV_ITEM_KEY;
1404 key.offset = device->devid;
1406 ret = btrfs_insert_empty_item(trans, root, path, &key,
1411 leaf = path->nodes[0];
1412 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1414 btrfs_set_device_id(leaf, dev_item, device->devid);
1415 btrfs_set_device_generation(leaf, dev_item, 0);
1416 btrfs_set_device_type(leaf, dev_item, device->type);
1417 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1418 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1419 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1420 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1421 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1422 btrfs_set_device_group(leaf, dev_item, 0);
1423 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1424 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1425 btrfs_set_device_start_offset(leaf, dev_item, 0);
1427 ptr = btrfs_device_uuid(dev_item);
1428 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1429 ptr = btrfs_device_fsid(dev_item);
1430 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1431 btrfs_mark_buffer_dirty(leaf);
1435 btrfs_free_path(path);
1439 static int btrfs_rm_dev_item(struct btrfs_root *root,
1440 struct btrfs_device *device)
1443 struct btrfs_path *path;
1444 struct btrfs_key key;
1445 struct btrfs_trans_handle *trans;
1447 root = root->fs_info->chunk_root;
1449 path = btrfs_alloc_path();
1453 trans = btrfs_start_transaction(root, 0);
1454 if (IS_ERR(trans)) {
1455 btrfs_free_path(path);
1456 return PTR_ERR(trans);
1458 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1459 key.type = BTRFS_DEV_ITEM_KEY;
1460 key.offset = device->devid;
1463 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1472 ret = btrfs_del_item(trans, root, path);
1476 btrfs_free_path(path);
1477 unlock_chunks(root);
1478 btrfs_commit_transaction(trans, root);
1482 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1484 struct btrfs_device *device;
1485 struct btrfs_device *next_device;
1486 struct block_device *bdev;
1487 struct buffer_head *bh = NULL;
1488 struct btrfs_super_block *disk_super;
1489 struct btrfs_fs_devices *cur_devices;
1496 bool clear_super = false;
1498 mutex_lock(&uuid_mutex);
1501 seq = read_seqbegin(&root->fs_info->profiles_lock);
1503 all_avail = root->fs_info->avail_data_alloc_bits |
1504 root->fs_info->avail_system_alloc_bits |
1505 root->fs_info->avail_metadata_alloc_bits;
1506 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1508 num_devices = root->fs_info->fs_devices->num_devices;
1509 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1510 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1511 WARN_ON(num_devices < 1);
1514 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1516 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1517 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1521 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1522 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1526 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1527 root->fs_info->fs_devices->rw_devices <= 2) {
1528 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1531 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1532 root->fs_info->fs_devices->rw_devices <= 3) {
1533 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1537 if (strcmp(device_path, "missing") == 0) {
1538 struct list_head *devices;
1539 struct btrfs_device *tmp;
1542 devices = &root->fs_info->fs_devices->devices;
1544 * It is safe to read the devices since the volume_mutex
1547 list_for_each_entry(tmp, devices, dev_list) {
1548 if (tmp->in_fs_metadata &&
1549 !tmp->is_tgtdev_for_dev_replace &&
1559 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1563 ret = btrfs_get_bdev_and_sb(device_path,
1564 FMODE_WRITE | FMODE_EXCL,
1565 root->fs_info->bdev_holder, 0,
1569 disk_super = (struct btrfs_super_block *)bh->b_data;
1570 devid = btrfs_stack_device_id(&disk_super->dev_item);
1571 dev_uuid = disk_super->dev_item.uuid;
1572 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1580 if (device->is_tgtdev_for_dev_replace) {
1581 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1585 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1586 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1590 if (device->writeable) {
1592 list_del_init(&device->dev_alloc_list);
1593 unlock_chunks(root);
1594 root->fs_info->fs_devices->rw_devices--;
1598 mutex_unlock(&uuid_mutex);
1599 ret = btrfs_shrink_device(device, 0);
1600 mutex_lock(&uuid_mutex);
1605 * TODO: the superblock still includes this device in its num_devices
1606 * counter although write_all_supers() is not locked out. This
1607 * could give a filesystem state which requires a degraded mount.
1609 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1613 spin_lock(&root->fs_info->free_chunk_lock);
1614 root->fs_info->free_chunk_space = device->total_bytes -
1616 spin_unlock(&root->fs_info->free_chunk_lock);
1618 device->in_fs_metadata = 0;
1619 btrfs_scrub_cancel_dev(root->fs_info, device);
1622 * the device list mutex makes sure that we don't change
1623 * the device list while someone else is writing out all
1624 * the device supers.
1627 cur_devices = device->fs_devices;
1628 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1629 list_del_rcu(&device->dev_list);
1631 device->fs_devices->num_devices--;
1632 device->fs_devices->total_devices--;
1634 if (device->missing)
1635 root->fs_info->fs_devices->missing_devices--;
1637 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1638 struct btrfs_device, dev_list);
1639 if (device->bdev == root->fs_info->sb->s_bdev)
1640 root->fs_info->sb->s_bdev = next_device->bdev;
1641 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1642 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1645 device->fs_devices->open_devices--;
1647 call_rcu(&device->rcu, free_device);
1648 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1650 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1651 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1653 if (cur_devices->open_devices == 0) {
1654 struct btrfs_fs_devices *fs_devices;
1655 fs_devices = root->fs_info->fs_devices;
1656 while (fs_devices) {
1657 if (fs_devices->seed == cur_devices)
1659 fs_devices = fs_devices->seed;
1661 fs_devices->seed = cur_devices->seed;
1662 cur_devices->seed = NULL;
1664 __btrfs_close_devices(cur_devices);
1665 unlock_chunks(root);
1666 free_fs_devices(cur_devices);
1669 root->fs_info->num_tolerated_disk_barrier_failures =
1670 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1673 * at this point, the device is zero sized. We want to
1674 * remove it from the devices list and zero out the old super
1676 if (clear_super && disk_super) {
1677 /* make sure this device isn't detected as part of
1680 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1681 set_buffer_dirty(bh);
1682 sync_dirty_buffer(bh);
1687 /* Notify udev that device has changed */
1689 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1694 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1696 mutex_unlock(&uuid_mutex);
1699 if (device->writeable) {
1701 list_add(&device->dev_alloc_list,
1702 &root->fs_info->fs_devices->alloc_list);
1703 unlock_chunks(root);
1704 root->fs_info->fs_devices->rw_devices++;
1709 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1710 struct btrfs_device *srcdev)
1712 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1713 list_del_rcu(&srcdev->dev_list);
1714 list_del_rcu(&srcdev->dev_alloc_list);
1715 fs_info->fs_devices->num_devices--;
1716 if (srcdev->missing) {
1717 fs_info->fs_devices->missing_devices--;
1718 fs_info->fs_devices->rw_devices++;
1720 if (srcdev->can_discard)
1721 fs_info->fs_devices->num_can_discard--;
1723 fs_info->fs_devices->open_devices--;
1725 call_rcu(&srcdev->rcu, free_device);
1728 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1729 struct btrfs_device *tgtdev)
1731 struct btrfs_device *next_device;
1734 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1736 btrfs_scratch_superblock(tgtdev);
1737 fs_info->fs_devices->open_devices--;
1739 fs_info->fs_devices->num_devices--;
1740 if (tgtdev->can_discard)
1741 fs_info->fs_devices->num_can_discard++;
1743 next_device = list_entry(fs_info->fs_devices->devices.next,
1744 struct btrfs_device, dev_list);
1745 if (tgtdev->bdev == fs_info->sb->s_bdev)
1746 fs_info->sb->s_bdev = next_device->bdev;
1747 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1748 fs_info->fs_devices->latest_bdev = next_device->bdev;
1749 list_del_rcu(&tgtdev->dev_list);
1751 call_rcu(&tgtdev->rcu, free_device);
1753 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1756 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1757 struct btrfs_device **device)
1760 struct btrfs_super_block *disk_super;
1763 struct block_device *bdev;
1764 struct buffer_head *bh;
1767 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1768 root->fs_info->bdev_holder, 0, &bdev, &bh);
1771 disk_super = (struct btrfs_super_block *)bh->b_data;
1772 devid = btrfs_stack_device_id(&disk_super->dev_item);
1773 dev_uuid = disk_super->dev_item.uuid;
1774 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1779 blkdev_put(bdev, FMODE_READ);
1783 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1785 struct btrfs_device **device)
1788 if (strcmp(device_path, "missing") == 0) {
1789 struct list_head *devices;
1790 struct btrfs_device *tmp;
1792 devices = &root->fs_info->fs_devices->devices;
1794 * It is safe to read the devices since the volume_mutex
1795 * is held by the caller.
1797 list_for_each_entry(tmp, devices, dev_list) {
1798 if (tmp->in_fs_metadata && !tmp->bdev) {
1805 pr_err("btrfs: no missing device found\n");
1811 return btrfs_find_device_by_path(root, device_path, device);
1816 * does all the dirty work required for changing file system's UUID.
1818 static int btrfs_prepare_sprout(struct btrfs_root *root)
1820 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1821 struct btrfs_fs_devices *old_devices;
1822 struct btrfs_fs_devices *seed_devices;
1823 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1824 struct btrfs_device *device;
1827 BUG_ON(!mutex_is_locked(&uuid_mutex));
1828 if (!fs_devices->seeding)
1831 seed_devices = __alloc_fs_devices();
1832 if (IS_ERR(seed_devices))
1833 return PTR_ERR(seed_devices);
1835 old_devices = clone_fs_devices(fs_devices);
1836 if (IS_ERR(old_devices)) {
1837 kfree(seed_devices);
1838 return PTR_ERR(old_devices);
1841 list_add(&old_devices->list, &fs_uuids);
1843 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1844 seed_devices->opened = 1;
1845 INIT_LIST_HEAD(&seed_devices->devices);
1846 INIT_LIST_HEAD(&seed_devices->alloc_list);
1847 mutex_init(&seed_devices->device_list_mutex);
1849 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1850 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1852 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1854 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1855 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1856 device->fs_devices = seed_devices;
1859 fs_devices->seeding = 0;
1860 fs_devices->num_devices = 0;
1861 fs_devices->open_devices = 0;
1862 fs_devices->total_devices = 0;
1863 fs_devices->seed = seed_devices;
1865 generate_random_uuid(fs_devices->fsid);
1866 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1867 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1868 super_flags = btrfs_super_flags(disk_super) &
1869 ~BTRFS_SUPER_FLAG_SEEDING;
1870 btrfs_set_super_flags(disk_super, super_flags);
1876 * strore the expected generation for seed devices in device items.
1878 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1879 struct btrfs_root *root)
1881 struct btrfs_path *path;
1882 struct extent_buffer *leaf;
1883 struct btrfs_dev_item *dev_item;
1884 struct btrfs_device *device;
1885 struct btrfs_key key;
1886 u8 fs_uuid[BTRFS_UUID_SIZE];
1887 u8 dev_uuid[BTRFS_UUID_SIZE];
1891 path = btrfs_alloc_path();
1895 root = root->fs_info->chunk_root;
1896 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1898 key.type = BTRFS_DEV_ITEM_KEY;
1901 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1905 leaf = path->nodes[0];
1907 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1908 ret = btrfs_next_leaf(root, path);
1913 leaf = path->nodes[0];
1914 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1915 btrfs_release_path(path);
1919 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1920 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1921 key.type != BTRFS_DEV_ITEM_KEY)
1924 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1925 struct btrfs_dev_item);
1926 devid = btrfs_device_id(leaf, dev_item);
1927 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1929 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1931 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1933 BUG_ON(!device); /* Logic error */
1935 if (device->fs_devices->seeding) {
1936 btrfs_set_device_generation(leaf, dev_item,
1937 device->generation);
1938 btrfs_mark_buffer_dirty(leaf);
1946 btrfs_free_path(path);
1950 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1952 struct request_queue *q;
1953 struct btrfs_trans_handle *trans;
1954 struct btrfs_device *device;
1955 struct block_device *bdev;
1956 struct list_head *devices;
1957 struct super_block *sb = root->fs_info->sb;
1958 struct rcu_string *name;
1960 int seeding_dev = 0;
1963 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1966 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1967 root->fs_info->bdev_holder);
1969 return PTR_ERR(bdev);
1971 if (root->fs_info->fs_devices->seeding) {
1973 down_write(&sb->s_umount);
1974 mutex_lock(&uuid_mutex);
1977 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1979 devices = &root->fs_info->fs_devices->devices;
1981 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1982 list_for_each_entry(device, devices, dev_list) {
1983 if (device->bdev == bdev) {
1986 &root->fs_info->fs_devices->device_list_mutex);
1990 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1992 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1993 if (IS_ERR(device)) {
1994 /* we can safely leave the fs_devices entry around */
1995 ret = PTR_ERR(device);
1999 name = rcu_string_strdup(device_path, GFP_NOFS);
2005 rcu_assign_pointer(device->name, name);
2007 trans = btrfs_start_transaction(root, 0);
2008 if (IS_ERR(trans)) {
2009 rcu_string_free(device->name);
2011 ret = PTR_ERR(trans);
2017 q = bdev_get_queue(bdev);
2018 if (blk_queue_discard(q))
2019 device->can_discard = 1;
2020 device->writeable = 1;
2021 device->generation = trans->transid;
2022 device->io_width = root->sectorsize;
2023 device->io_align = root->sectorsize;
2024 device->sector_size = root->sectorsize;
2025 device->total_bytes = i_size_read(bdev->bd_inode);
2026 device->disk_total_bytes = device->total_bytes;
2027 device->dev_root = root->fs_info->dev_root;
2028 device->bdev = bdev;
2029 device->in_fs_metadata = 1;
2030 device->is_tgtdev_for_dev_replace = 0;
2031 device->mode = FMODE_EXCL;
2032 set_blocksize(device->bdev, 4096);
2035 sb->s_flags &= ~MS_RDONLY;
2036 ret = btrfs_prepare_sprout(root);
2037 BUG_ON(ret); /* -ENOMEM */
2040 device->fs_devices = root->fs_info->fs_devices;
2042 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2043 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2044 list_add(&device->dev_alloc_list,
2045 &root->fs_info->fs_devices->alloc_list);
2046 root->fs_info->fs_devices->num_devices++;
2047 root->fs_info->fs_devices->open_devices++;
2048 root->fs_info->fs_devices->rw_devices++;
2049 root->fs_info->fs_devices->total_devices++;
2050 if (device->can_discard)
2051 root->fs_info->fs_devices->num_can_discard++;
2052 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2054 spin_lock(&root->fs_info->free_chunk_lock);
2055 root->fs_info->free_chunk_space += device->total_bytes;
2056 spin_unlock(&root->fs_info->free_chunk_lock);
2058 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2059 root->fs_info->fs_devices->rotating = 1;
2061 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2062 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2063 total_bytes + device->total_bytes);
2065 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2066 btrfs_set_super_num_devices(root->fs_info->super_copy,
2068 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2071 ret = init_first_rw_device(trans, root, device);
2073 btrfs_abort_transaction(trans, root, ret);
2076 ret = btrfs_finish_sprout(trans, root);
2078 btrfs_abort_transaction(trans, root, ret);
2082 ret = btrfs_add_device(trans, root, device);
2084 btrfs_abort_transaction(trans, root, ret);
2090 * we've got more storage, clear any full flags on the space
2093 btrfs_clear_space_info_full(root->fs_info);
2095 unlock_chunks(root);
2096 root->fs_info->num_tolerated_disk_barrier_failures =
2097 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2098 ret = btrfs_commit_transaction(trans, root);
2101 mutex_unlock(&uuid_mutex);
2102 up_write(&sb->s_umount);
2104 if (ret) /* transaction commit */
2107 ret = btrfs_relocate_sys_chunks(root);
2109 btrfs_error(root->fs_info, ret,
2110 "Failed to relocate sys chunks after "
2111 "device initialization. This can be fixed "
2112 "using the \"btrfs balance\" command.");
2113 trans = btrfs_attach_transaction(root);
2114 if (IS_ERR(trans)) {
2115 if (PTR_ERR(trans) == -ENOENT)
2117 return PTR_ERR(trans);
2119 ret = btrfs_commit_transaction(trans, root);
2125 unlock_chunks(root);
2126 btrfs_end_transaction(trans, root);
2127 rcu_string_free(device->name);
2130 blkdev_put(bdev, FMODE_EXCL);
2132 mutex_unlock(&uuid_mutex);
2133 up_write(&sb->s_umount);
2138 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2139 struct btrfs_device **device_out)
2141 struct request_queue *q;
2142 struct btrfs_device *device;
2143 struct block_device *bdev;
2144 struct btrfs_fs_info *fs_info = root->fs_info;
2145 struct list_head *devices;
2146 struct rcu_string *name;
2147 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2151 if (fs_info->fs_devices->seeding)
2154 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2155 fs_info->bdev_holder);
2157 return PTR_ERR(bdev);
2159 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2161 devices = &fs_info->fs_devices->devices;
2162 list_for_each_entry(device, devices, dev_list) {
2163 if (device->bdev == bdev) {
2169 device = btrfs_alloc_device(NULL, &devid, NULL);
2170 if (IS_ERR(device)) {
2171 ret = PTR_ERR(device);
2175 name = rcu_string_strdup(device_path, GFP_NOFS);
2181 rcu_assign_pointer(device->name, name);
2183 q = bdev_get_queue(bdev);
2184 if (blk_queue_discard(q))
2185 device->can_discard = 1;
2186 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2187 device->writeable = 1;
2188 device->generation = 0;
2189 device->io_width = root->sectorsize;
2190 device->io_align = root->sectorsize;
2191 device->sector_size = root->sectorsize;
2192 device->total_bytes = i_size_read(bdev->bd_inode);
2193 device->disk_total_bytes = device->total_bytes;
2194 device->dev_root = fs_info->dev_root;
2195 device->bdev = bdev;
2196 device->in_fs_metadata = 1;
2197 device->is_tgtdev_for_dev_replace = 1;
2198 device->mode = FMODE_EXCL;
2199 set_blocksize(device->bdev, 4096);
2200 device->fs_devices = fs_info->fs_devices;
2201 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2202 fs_info->fs_devices->num_devices++;
2203 fs_info->fs_devices->open_devices++;
2204 if (device->can_discard)
2205 fs_info->fs_devices->num_can_discard++;
2206 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2208 *device_out = device;
2212 blkdev_put(bdev, FMODE_EXCL);
2216 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2217 struct btrfs_device *tgtdev)
2219 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2220 tgtdev->io_width = fs_info->dev_root->sectorsize;
2221 tgtdev->io_align = fs_info->dev_root->sectorsize;
2222 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2223 tgtdev->dev_root = fs_info->dev_root;
2224 tgtdev->in_fs_metadata = 1;
2227 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2228 struct btrfs_device *device)
2231 struct btrfs_path *path;
2232 struct btrfs_root *root;
2233 struct btrfs_dev_item *dev_item;
2234 struct extent_buffer *leaf;
2235 struct btrfs_key key;
2237 root = device->dev_root->fs_info->chunk_root;
2239 path = btrfs_alloc_path();
2243 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2244 key.type = BTRFS_DEV_ITEM_KEY;
2245 key.offset = device->devid;
2247 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2256 leaf = path->nodes[0];
2257 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2259 btrfs_set_device_id(leaf, dev_item, device->devid);
2260 btrfs_set_device_type(leaf, dev_item, device->type);
2261 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2262 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2263 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2264 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2265 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2266 btrfs_mark_buffer_dirty(leaf);
2269 btrfs_free_path(path);
2273 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2274 struct btrfs_device *device, u64 new_size)
2276 struct btrfs_super_block *super_copy =
2277 device->dev_root->fs_info->super_copy;
2278 u64 old_total = btrfs_super_total_bytes(super_copy);
2279 u64 diff = new_size - device->total_bytes;
2281 if (!device->writeable)
2283 if (new_size <= device->total_bytes ||
2284 device->is_tgtdev_for_dev_replace)
2287 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2288 device->fs_devices->total_rw_bytes += diff;
2290 device->total_bytes = new_size;
2291 device->disk_total_bytes = new_size;
2292 btrfs_clear_space_info_full(device->dev_root->fs_info);
2294 return btrfs_update_device(trans, device);
2297 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2298 struct btrfs_device *device, u64 new_size)
2301 lock_chunks(device->dev_root);
2302 ret = __btrfs_grow_device(trans, device, new_size);
2303 unlock_chunks(device->dev_root);
2307 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2308 struct btrfs_root *root,
2309 u64 chunk_tree, u64 chunk_objectid,
2313 struct btrfs_path *path;
2314 struct btrfs_key key;
2316 root = root->fs_info->chunk_root;
2317 path = btrfs_alloc_path();
2321 key.objectid = chunk_objectid;
2322 key.offset = chunk_offset;
2323 key.type = BTRFS_CHUNK_ITEM_KEY;
2325 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2328 else if (ret > 0) { /* Logic error or corruption */
2329 btrfs_error(root->fs_info, -ENOENT,
2330 "Failed lookup while freeing chunk.");
2335 ret = btrfs_del_item(trans, root, path);
2337 btrfs_error(root->fs_info, ret,
2338 "Failed to delete chunk item.");
2340 btrfs_free_path(path);
2344 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2347 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2348 struct btrfs_disk_key *disk_key;
2349 struct btrfs_chunk *chunk;
2356 struct btrfs_key key;
2358 array_size = btrfs_super_sys_array_size(super_copy);
2360 ptr = super_copy->sys_chunk_array;
2363 while (cur < array_size) {
2364 disk_key = (struct btrfs_disk_key *)ptr;
2365 btrfs_disk_key_to_cpu(&key, disk_key);
2367 len = sizeof(*disk_key);
2369 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2370 chunk = (struct btrfs_chunk *)(ptr + len);
2371 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2372 len += btrfs_chunk_item_size(num_stripes);
2377 if (key.objectid == chunk_objectid &&
2378 key.offset == chunk_offset) {
2379 memmove(ptr, ptr + len, array_size - (cur + len));
2381 btrfs_set_super_sys_array_size(super_copy, array_size);
2390 static int btrfs_relocate_chunk(struct btrfs_root *root,
2391 u64 chunk_tree, u64 chunk_objectid,
2394 struct extent_map_tree *em_tree;
2395 struct btrfs_root *extent_root;
2396 struct btrfs_trans_handle *trans;
2397 struct extent_map *em;
2398 struct map_lookup *map;
2402 root = root->fs_info->chunk_root;
2403 extent_root = root->fs_info->extent_root;
2404 em_tree = &root->fs_info->mapping_tree.map_tree;
2406 ret = btrfs_can_relocate(extent_root, chunk_offset);
2410 /* step one, relocate all the extents inside this chunk */
2411 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2415 trans = btrfs_start_transaction(root, 0);
2416 if (IS_ERR(trans)) {
2417 ret = PTR_ERR(trans);
2418 btrfs_std_error(root->fs_info, ret);
2425 * step two, delete the device extents and the
2426 * chunk tree entries
2428 read_lock(&em_tree->lock);
2429 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2430 read_unlock(&em_tree->lock);
2432 BUG_ON(!em || em->start > chunk_offset ||
2433 em->start + em->len < chunk_offset);
2434 map = (struct map_lookup *)em->bdev;
2436 for (i = 0; i < map->num_stripes; i++) {
2437 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2438 map->stripes[i].physical);
2441 if (map->stripes[i].dev) {
2442 ret = btrfs_update_device(trans, map->stripes[i].dev);
2446 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2451 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2453 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2454 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2458 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2461 write_lock(&em_tree->lock);
2462 remove_extent_mapping(em_tree, em);
2463 write_unlock(&em_tree->lock);
2468 /* once for the tree */
2469 free_extent_map(em);
2471 free_extent_map(em);
2473 unlock_chunks(root);
2474 btrfs_end_transaction(trans, root);
2478 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2480 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2481 struct btrfs_path *path;
2482 struct extent_buffer *leaf;
2483 struct btrfs_chunk *chunk;
2484 struct btrfs_key key;
2485 struct btrfs_key found_key;
2486 u64 chunk_tree = chunk_root->root_key.objectid;
2488 bool retried = false;
2492 path = btrfs_alloc_path();
2497 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2498 key.offset = (u64)-1;
2499 key.type = BTRFS_CHUNK_ITEM_KEY;
2502 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2505 BUG_ON(ret == 0); /* Corruption */
2507 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2514 leaf = path->nodes[0];
2515 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2517 chunk = btrfs_item_ptr(leaf, path->slots[0],
2518 struct btrfs_chunk);
2519 chunk_type = btrfs_chunk_type(leaf, chunk);
2520 btrfs_release_path(path);
2522 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2523 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2532 if (found_key.offset == 0)
2534 key.offset = found_key.offset - 1;
2537 if (failed && !retried) {
2541 } else if (failed && retried) {
2546 btrfs_free_path(path);
2550 static int insert_balance_item(struct btrfs_root *root,
2551 struct btrfs_balance_control *bctl)
2553 struct btrfs_trans_handle *trans;
2554 struct btrfs_balance_item *item;
2555 struct btrfs_disk_balance_args disk_bargs;
2556 struct btrfs_path *path;
2557 struct extent_buffer *leaf;
2558 struct btrfs_key key;
2561 path = btrfs_alloc_path();
2565 trans = btrfs_start_transaction(root, 0);
2566 if (IS_ERR(trans)) {
2567 btrfs_free_path(path);
2568 return PTR_ERR(trans);
2571 key.objectid = BTRFS_BALANCE_OBJECTID;
2572 key.type = BTRFS_BALANCE_ITEM_KEY;
2575 ret = btrfs_insert_empty_item(trans, root, path, &key,
2580 leaf = path->nodes[0];
2581 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2583 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2585 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2586 btrfs_set_balance_data(leaf, item, &disk_bargs);
2587 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2588 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2589 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2590 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2592 btrfs_set_balance_flags(leaf, item, bctl->flags);
2594 btrfs_mark_buffer_dirty(leaf);
2596 btrfs_free_path(path);
2597 err = btrfs_commit_transaction(trans, root);
2603 static int del_balance_item(struct btrfs_root *root)
2605 struct btrfs_trans_handle *trans;
2606 struct btrfs_path *path;
2607 struct btrfs_key key;
2610 path = btrfs_alloc_path();
2614 trans = btrfs_start_transaction(root, 0);
2615 if (IS_ERR(trans)) {
2616 btrfs_free_path(path);
2617 return PTR_ERR(trans);
2620 key.objectid = BTRFS_BALANCE_OBJECTID;
2621 key.type = BTRFS_BALANCE_ITEM_KEY;
2624 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2632 ret = btrfs_del_item(trans, root, path);
2634 btrfs_free_path(path);
2635 err = btrfs_commit_transaction(trans, root);
2642 * This is a heuristic used to reduce the number of chunks balanced on
2643 * resume after balance was interrupted.
2645 static void update_balance_args(struct btrfs_balance_control *bctl)
2648 * Turn on soft mode for chunk types that were being converted.
2650 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2651 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2652 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2653 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2654 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2655 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2658 * Turn on usage filter if is not already used. The idea is
2659 * that chunks that we have already balanced should be
2660 * reasonably full. Don't do it for chunks that are being
2661 * converted - that will keep us from relocating unconverted
2662 * (albeit full) chunks.
2664 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2665 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2666 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2667 bctl->data.usage = 90;
2669 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2670 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2671 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2672 bctl->sys.usage = 90;
2674 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2675 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2676 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2677 bctl->meta.usage = 90;
2682 * Should be called with both balance and volume mutexes held to
2683 * serialize other volume operations (add_dev/rm_dev/resize) with
2684 * restriper. Same goes for unset_balance_control.
2686 static void set_balance_control(struct btrfs_balance_control *bctl)
2688 struct btrfs_fs_info *fs_info = bctl->fs_info;
2690 BUG_ON(fs_info->balance_ctl);
2692 spin_lock(&fs_info->balance_lock);
2693 fs_info->balance_ctl = bctl;
2694 spin_unlock(&fs_info->balance_lock);
2697 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2699 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2701 BUG_ON(!fs_info->balance_ctl);
2703 spin_lock(&fs_info->balance_lock);
2704 fs_info->balance_ctl = NULL;
2705 spin_unlock(&fs_info->balance_lock);
2711 * Balance filters. Return 1 if chunk should be filtered out
2712 * (should not be balanced).
2714 static int chunk_profiles_filter(u64 chunk_type,
2715 struct btrfs_balance_args *bargs)
2717 chunk_type = chunk_to_extended(chunk_type) &
2718 BTRFS_EXTENDED_PROFILE_MASK;
2720 if (bargs->profiles & chunk_type)
2726 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2727 struct btrfs_balance_args *bargs)
2729 struct btrfs_block_group_cache *cache;
2730 u64 chunk_used, user_thresh;
2733 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2734 chunk_used = btrfs_block_group_used(&cache->item);
2736 if (bargs->usage == 0)
2738 else if (bargs->usage > 100)
2739 user_thresh = cache->key.offset;
2741 user_thresh = div_factor_fine(cache->key.offset,
2744 if (chunk_used < user_thresh)
2747 btrfs_put_block_group(cache);
2751 static int chunk_devid_filter(struct extent_buffer *leaf,
2752 struct btrfs_chunk *chunk,
2753 struct btrfs_balance_args *bargs)
2755 struct btrfs_stripe *stripe;
2756 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2759 for (i = 0; i < num_stripes; i++) {
2760 stripe = btrfs_stripe_nr(chunk, i);
2761 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2768 /* [pstart, pend) */
2769 static int chunk_drange_filter(struct extent_buffer *leaf,
2770 struct btrfs_chunk *chunk,
2772 struct btrfs_balance_args *bargs)
2774 struct btrfs_stripe *stripe;
2775 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2781 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2784 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2785 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2786 factor = num_stripes / 2;
2787 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2788 factor = num_stripes - 1;
2789 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2790 factor = num_stripes - 2;
2792 factor = num_stripes;
2795 for (i = 0; i < num_stripes; i++) {
2796 stripe = btrfs_stripe_nr(chunk, i);
2797 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2800 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2801 stripe_length = btrfs_chunk_length(leaf, chunk);
2802 do_div(stripe_length, factor);
2804 if (stripe_offset < bargs->pend &&
2805 stripe_offset + stripe_length > bargs->pstart)
2812 /* [vstart, vend) */
2813 static int chunk_vrange_filter(struct extent_buffer *leaf,
2814 struct btrfs_chunk *chunk,
2816 struct btrfs_balance_args *bargs)
2818 if (chunk_offset < bargs->vend &&
2819 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2820 /* at least part of the chunk is inside this vrange */
2826 static int chunk_soft_convert_filter(u64 chunk_type,
2827 struct btrfs_balance_args *bargs)
2829 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2832 chunk_type = chunk_to_extended(chunk_type) &
2833 BTRFS_EXTENDED_PROFILE_MASK;
2835 if (bargs->target == chunk_type)
2841 static int should_balance_chunk(struct btrfs_root *root,
2842 struct extent_buffer *leaf,
2843 struct btrfs_chunk *chunk, u64 chunk_offset)
2845 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2846 struct btrfs_balance_args *bargs = NULL;
2847 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2850 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2851 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2855 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2856 bargs = &bctl->data;
2857 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2859 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2860 bargs = &bctl->meta;
2862 /* profiles filter */
2863 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2864 chunk_profiles_filter(chunk_type, bargs)) {
2869 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2870 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2875 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2876 chunk_devid_filter(leaf, chunk, bargs)) {
2880 /* drange filter, makes sense only with devid filter */
2881 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2882 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2887 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2888 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2892 /* soft profile changing mode */
2893 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2894 chunk_soft_convert_filter(chunk_type, bargs)) {
2901 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2903 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2904 struct btrfs_root *chunk_root = fs_info->chunk_root;
2905 struct btrfs_root *dev_root = fs_info->dev_root;
2906 struct list_head *devices;
2907 struct btrfs_device *device;
2910 struct btrfs_chunk *chunk;
2911 struct btrfs_path *path;
2912 struct btrfs_key key;
2913 struct btrfs_key found_key;
2914 struct btrfs_trans_handle *trans;
2915 struct extent_buffer *leaf;
2918 int enospc_errors = 0;
2919 bool counting = true;
2921 /* step one make some room on all the devices */
2922 devices = &fs_info->fs_devices->devices;
2923 list_for_each_entry(device, devices, dev_list) {
2924 old_size = device->total_bytes;
2925 size_to_free = div_factor(old_size, 1);
2926 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2927 if (!device->writeable ||
2928 device->total_bytes - device->bytes_used > size_to_free ||
2929 device->is_tgtdev_for_dev_replace)
2932 ret = btrfs_shrink_device(device, old_size - size_to_free);
2937 trans = btrfs_start_transaction(dev_root, 0);
2938 BUG_ON(IS_ERR(trans));
2940 ret = btrfs_grow_device(trans, device, old_size);
2943 btrfs_end_transaction(trans, dev_root);
2946 /* step two, relocate all the chunks */
2947 path = btrfs_alloc_path();
2953 /* zero out stat counters */
2954 spin_lock(&fs_info->balance_lock);
2955 memset(&bctl->stat, 0, sizeof(bctl->stat));
2956 spin_unlock(&fs_info->balance_lock);
2958 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2959 key.offset = (u64)-1;
2960 key.type = BTRFS_CHUNK_ITEM_KEY;
2963 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2964 atomic_read(&fs_info->balance_cancel_req)) {
2969 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2974 * this shouldn't happen, it means the last relocate
2978 BUG(); /* FIXME break ? */
2980 ret = btrfs_previous_item(chunk_root, path, 0,
2981 BTRFS_CHUNK_ITEM_KEY);
2987 leaf = path->nodes[0];
2988 slot = path->slots[0];
2989 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2991 if (found_key.objectid != key.objectid)
2994 /* chunk zero is special */
2995 if (found_key.offset == 0)
2998 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3001 spin_lock(&fs_info->balance_lock);
3002 bctl->stat.considered++;
3003 spin_unlock(&fs_info->balance_lock);
3006 ret = should_balance_chunk(chunk_root, leaf, chunk,
3008 btrfs_release_path(path);
3013 spin_lock(&fs_info->balance_lock);
3014 bctl->stat.expected++;
3015 spin_unlock(&fs_info->balance_lock);
3019 ret = btrfs_relocate_chunk(chunk_root,
3020 chunk_root->root_key.objectid,
3023 if (ret && ret != -ENOSPC)
3025 if (ret == -ENOSPC) {
3028 spin_lock(&fs_info->balance_lock);
3029 bctl->stat.completed++;
3030 spin_unlock(&fs_info->balance_lock);
3033 key.offset = found_key.offset - 1;
3037 btrfs_release_path(path);
3042 btrfs_free_path(path);
3043 if (enospc_errors) {
3044 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3054 * alloc_profile_is_valid - see if a given profile is valid and reduced
3055 * @flags: profile to validate
3056 * @extended: if true @flags is treated as an extended profile
3058 static int alloc_profile_is_valid(u64 flags, int extended)
3060 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3061 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3063 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3065 /* 1) check that all other bits are zeroed */
3069 /* 2) see if profile is reduced */
3071 return !extended; /* "0" is valid for usual profiles */
3073 /* true if exactly one bit set */
3074 return (flags & (flags - 1)) == 0;
3077 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3079 /* cancel requested || normal exit path */
3080 return atomic_read(&fs_info->balance_cancel_req) ||
3081 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3082 atomic_read(&fs_info->balance_cancel_req) == 0);
3085 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3089 unset_balance_control(fs_info);
3090 ret = del_balance_item(fs_info->tree_root);
3092 btrfs_std_error(fs_info, ret);
3094 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3098 * Should be called with both balance and volume mutexes held
3100 int btrfs_balance(struct btrfs_balance_control *bctl,
3101 struct btrfs_ioctl_balance_args *bargs)
3103 struct btrfs_fs_info *fs_info = bctl->fs_info;
3110 if (btrfs_fs_closing(fs_info) ||
3111 atomic_read(&fs_info->balance_pause_req) ||
3112 atomic_read(&fs_info->balance_cancel_req)) {
3117 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3118 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3122 * In case of mixed groups both data and meta should be picked,
3123 * and identical options should be given for both of them.
3125 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3126 if (mixed && (bctl->flags & allowed)) {
3127 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3128 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3129 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3130 printk(KERN_ERR "btrfs: with mixed groups data and "
3131 "metadata balance options must be the same\n");
3137 num_devices = fs_info->fs_devices->num_devices;
3138 btrfs_dev_replace_lock(&fs_info->dev_replace);
3139 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3140 BUG_ON(num_devices < 1);
3143 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3144 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3145 if (num_devices == 1)
3146 allowed |= BTRFS_BLOCK_GROUP_DUP;
3147 else if (num_devices > 1)
3148 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3149 if (num_devices > 2)
3150 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3151 if (num_devices > 3)
3152 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3153 BTRFS_BLOCK_GROUP_RAID6);
3154 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3155 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3156 (bctl->data.target & ~allowed))) {
3157 printk(KERN_ERR "btrfs: unable to start balance with target "
3158 "data profile %llu\n",
3163 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3164 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3165 (bctl->meta.target & ~allowed))) {
3166 printk(KERN_ERR "btrfs: unable to start balance with target "
3167 "metadata profile %llu\n",
3172 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3173 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3174 (bctl->sys.target & ~allowed))) {
3175 printk(KERN_ERR "btrfs: unable to start balance with target "
3176 "system profile %llu\n",
3182 /* allow dup'ed data chunks only in mixed mode */
3183 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3184 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3185 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3190 /* allow to reduce meta or sys integrity only if force set */
3191 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3192 BTRFS_BLOCK_GROUP_RAID10 |
3193 BTRFS_BLOCK_GROUP_RAID5 |
3194 BTRFS_BLOCK_GROUP_RAID6;
3196 seq = read_seqbegin(&fs_info->profiles_lock);
3198 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3199 (fs_info->avail_system_alloc_bits & allowed) &&
3200 !(bctl->sys.target & allowed)) ||
3201 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3202 (fs_info->avail_metadata_alloc_bits & allowed) &&
3203 !(bctl->meta.target & allowed))) {
3204 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3205 printk(KERN_INFO "btrfs: force reducing metadata "
3208 printk(KERN_ERR "btrfs: balance will reduce metadata "
3209 "integrity, use force if you want this\n");
3214 } while (read_seqretry(&fs_info->profiles_lock, seq));
3216 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3217 int num_tolerated_disk_barrier_failures;
3218 u64 target = bctl->sys.target;
3220 num_tolerated_disk_barrier_failures =
3221 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3222 if (num_tolerated_disk_barrier_failures > 0 &&
3224 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3225 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3226 num_tolerated_disk_barrier_failures = 0;
3227 else if (num_tolerated_disk_barrier_failures > 1 &&
3229 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3230 num_tolerated_disk_barrier_failures = 1;
3232 fs_info->num_tolerated_disk_barrier_failures =
3233 num_tolerated_disk_barrier_failures;
3236 ret = insert_balance_item(fs_info->tree_root, bctl);
3237 if (ret && ret != -EEXIST)
3240 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3241 BUG_ON(ret == -EEXIST);
3242 set_balance_control(bctl);
3244 BUG_ON(ret != -EEXIST);
3245 spin_lock(&fs_info->balance_lock);
3246 update_balance_args(bctl);
3247 spin_unlock(&fs_info->balance_lock);
3250 atomic_inc(&fs_info->balance_running);
3251 mutex_unlock(&fs_info->balance_mutex);
3253 ret = __btrfs_balance(fs_info);
3255 mutex_lock(&fs_info->balance_mutex);
3256 atomic_dec(&fs_info->balance_running);
3258 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3259 fs_info->num_tolerated_disk_barrier_failures =
3260 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3264 memset(bargs, 0, sizeof(*bargs));
3265 update_ioctl_balance_args(fs_info, 0, bargs);
3268 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3269 balance_need_close(fs_info)) {
3270 __cancel_balance(fs_info);
3273 wake_up(&fs_info->balance_wait_q);
3277 if (bctl->flags & BTRFS_BALANCE_RESUME)
3278 __cancel_balance(fs_info);
3281 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3286 static int balance_kthread(void *data)
3288 struct btrfs_fs_info *fs_info = data;
3291 mutex_lock(&fs_info->volume_mutex);
3292 mutex_lock(&fs_info->balance_mutex);
3294 if (fs_info->balance_ctl) {
3295 printk(KERN_INFO "btrfs: continuing balance\n");
3296 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3299 mutex_unlock(&fs_info->balance_mutex);
3300 mutex_unlock(&fs_info->volume_mutex);
3305 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3307 struct task_struct *tsk;
3309 spin_lock(&fs_info->balance_lock);
3310 if (!fs_info->balance_ctl) {
3311 spin_unlock(&fs_info->balance_lock);
3314 spin_unlock(&fs_info->balance_lock);
3316 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3317 printk(KERN_INFO "btrfs: force skipping balance\n");
3321 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3322 return PTR_RET(tsk);
3325 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3327 struct btrfs_balance_control *bctl;
3328 struct btrfs_balance_item *item;
3329 struct btrfs_disk_balance_args disk_bargs;
3330 struct btrfs_path *path;
3331 struct extent_buffer *leaf;
3332 struct btrfs_key key;
3335 path = btrfs_alloc_path();
3339 key.objectid = BTRFS_BALANCE_OBJECTID;
3340 key.type = BTRFS_BALANCE_ITEM_KEY;
3343 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3346 if (ret > 0) { /* ret = -ENOENT; */
3351 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3357 leaf = path->nodes[0];
3358 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3360 bctl->fs_info = fs_info;
3361 bctl->flags = btrfs_balance_flags(leaf, item);
3362 bctl->flags |= BTRFS_BALANCE_RESUME;
3364 btrfs_balance_data(leaf, item, &disk_bargs);
3365 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3366 btrfs_balance_meta(leaf, item, &disk_bargs);
3367 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3368 btrfs_balance_sys(leaf, item, &disk_bargs);
3369 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3371 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3373 mutex_lock(&fs_info->volume_mutex);
3374 mutex_lock(&fs_info->balance_mutex);
3376 set_balance_control(bctl);
3378 mutex_unlock(&fs_info->balance_mutex);
3379 mutex_unlock(&fs_info->volume_mutex);
3381 btrfs_free_path(path);
3385 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3389 mutex_lock(&fs_info->balance_mutex);
3390 if (!fs_info->balance_ctl) {
3391 mutex_unlock(&fs_info->balance_mutex);
3395 if (atomic_read(&fs_info->balance_running)) {
3396 atomic_inc(&fs_info->balance_pause_req);
3397 mutex_unlock(&fs_info->balance_mutex);
3399 wait_event(fs_info->balance_wait_q,
3400 atomic_read(&fs_info->balance_running) == 0);
3402 mutex_lock(&fs_info->balance_mutex);
3403 /* we are good with balance_ctl ripped off from under us */
3404 BUG_ON(atomic_read(&fs_info->balance_running));
3405 atomic_dec(&fs_info->balance_pause_req);
3410 mutex_unlock(&fs_info->balance_mutex);
3414 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3416 mutex_lock(&fs_info->balance_mutex);
3417 if (!fs_info->balance_ctl) {
3418 mutex_unlock(&fs_info->balance_mutex);
3422 atomic_inc(&fs_info->balance_cancel_req);
3424 * if we are running just wait and return, balance item is
3425 * deleted in btrfs_balance in this case
3427 if (atomic_read(&fs_info->balance_running)) {
3428 mutex_unlock(&fs_info->balance_mutex);
3429 wait_event(fs_info->balance_wait_q,
3430 atomic_read(&fs_info->balance_running) == 0);
3431 mutex_lock(&fs_info->balance_mutex);
3433 /* __cancel_balance needs volume_mutex */
3434 mutex_unlock(&fs_info->balance_mutex);
3435 mutex_lock(&fs_info->volume_mutex);
3436 mutex_lock(&fs_info->balance_mutex);
3438 if (fs_info->balance_ctl)
3439 __cancel_balance(fs_info);
3441 mutex_unlock(&fs_info->volume_mutex);
3444 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3445 atomic_dec(&fs_info->balance_cancel_req);
3446 mutex_unlock(&fs_info->balance_mutex);
3450 static int btrfs_uuid_scan_kthread(void *data)
3452 struct btrfs_fs_info *fs_info = data;
3453 struct btrfs_root *root = fs_info->tree_root;
3454 struct btrfs_key key;
3455 struct btrfs_key max_key;
3456 struct btrfs_path *path = NULL;
3458 struct extent_buffer *eb;
3460 struct btrfs_root_item root_item;
3462 struct btrfs_trans_handle *trans;
3464 path = btrfs_alloc_path();
3471 key.type = BTRFS_ROOT_ITEM_KEY;
3474 max_key.objectid = (u64)-1;
3475 max_key.type = BTRFS_ROOT_ITEM_KEY;
3476 max_key.offset = (u64)-1;
3478 path->keep_locks = 1;
3481 ret = btrfs_search_forward(root, &key, &max_key, path, 0);
3488 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3489 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3490 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3491 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3494 eb = path->nodes[0];
3495 slot = path->slots[0];
3496 item_size = btrfs_item_size_nr(eb, slot);
3497 if (item_size < sizeof(root_item))
3501 read_extent_buffer(eb, &root_item,
3502 btrfs_item_ptr_offset(eb, slot),
3503 (int)sizeof(root_item));
3504 if (btrfs_root_refs(&root_item) == 0)
3506 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3508 * 1 - subvol uuid item
3509 * 1 - received_subvol uuid item
3511 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3512 if (IS_ERR(trans)) {
3513 ret = PTR_ERR(trans);
3516 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3518 BTRFS_UUID_KEY_SUBVOL,
3521 pr_warn("btrfs: uuid_tree_add failed %d\n",
3523 btrfs_end_transaction(trans,
3524 fs_info->uuid_root);
3529 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3531 /* 1 - received_subvol uuid item */
3532 trans = btrfs_start_transaction(
3533 fs_info->uuid_root, 1);
3534 if (IS_ERR(trans)) {
3535 ret = PTR_ERR(trans);
3539 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3540 root_item.received_uuid,
3541 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3544 pr_warn("btrfs: uuid_tree_add failed %d\n",
3546 btrfs_end_transaction(trans,
3547 fs_info->uuid_root);
3553 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3559 btrfs_release_path(path);
3560 if (key.offset < (u64)-1) {
3562 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3564 key.type = BTRFS_ROOT_ITEM_KEY;
3565 } else if (key.objectid < (u64)-1) {
3567 key.type = BTRFS_ROOT_ITEM_KEY;
3576 btrfs_free_path(path);
3578 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3580 fs_info->update_uuid_tree_gen = 1;
3581 up(&fs_info->uuid_tree_rescan_sem);
3586 * Callback for btrfs_uuid_tree_iterate().
3588 * 0 check succeeded, the entry is not outdated.
3589 * < 0 if an error occured.
3590 * > 0 if the check failed, which means the caller shall remove the entry.
3592 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3593 u8 *uuid, u8 type, u64 subid)
3595 struct btrfs_key key;
3597 struct btrfs_root *subvol_root;
3599 if (type != BTRFS_UUID_KEY_SUBVOL &&
3600 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3603 key.objectid = subid;
3604 key.type = BTRFS_ROOT_ITEM_KEY;
3605 key.offset = (u64)-1;
3606 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3607 if (IS_ERR(subvol_root)) {
3608 ret = PTR_ERR(subvol_root);
3615 case BTRFS_UUID_KEY_SUBVOL:
3616 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3619 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3620 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3630 static int btrfs_uuid_rescan_kthread(void *data)
3632 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3636 * 1st step is to iterate through the existing UUID tree and
3637 * to delete all entries that contain outdated data.
3638 * 2nd step is to add all missing entries to the UUID tree.
3640 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3642 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3643 up(&fs_info->uuid_tree_rescan_sem);
3646 return btrfs_uuid_scan_kthread(data);
3649 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3651 struct btrfs_trans_handle *trans;
3652 struct btrfs_root *tree_root = fs_info->tree_root;
3653 struct btrfs_root *uuid_root;
3654 struct task_struct *task;
3661 trans = btrfs_start_transaction(tree_root, 2);
3663 return PTR_ERR(trans);
3665 uuid_root = btrfs_create_tree(trans, fs_info,
3666 BTRFS_UUID_TREE_OBJECTID);
3667 if (IS_ERR(uuid_root)) {
3668 btrfs_abort_transaction(trans, tree_root,
3669 PTR_ERR(uuid_root));
3670 return PTR_ERR(uuid_root);
3673 fs_info->uuid_root = uuid_root;
3675 ret = btrfs_commit_transaction(trans, tree_root);
3679 down(&fs_info->uuid_tree_rescan_sem);
3680 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3682 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3683 pr_warn("btrfs: failed to start uuid_scan task\n");
3684 up(&fs_info->uuid_tree_rescan_sem);
3685 return PTR_ERR(task);
3691 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3693 struct task_struct *task;
3695 down(&fs_info->uuid_tree_rescan_sem);
3696 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3698 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3699 pr_warn("btrfs: failed to start uuid_rescan task\n");
3700 up(&fs_info->uuid_tree_rescan_sem);
3701 return PTR_ERR(task);
3708 * shrinking a device means finding all of the device extents past
3709 * the new size, and then following the back refs to the chunks.
3710 * The chunk relocation code actually frees the device extent
3712 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3714 struct btrfs_trans_handle *trans;
3715 struct btrfs_root *root = device->dev_root;
3716 struct btrfs_dev_extent *dev_extent = NULL;
3717 struct btrfs_path *path;
3725 bool retried = false;
3726 struct extent_buffer *l;
3727 struct btrfs_key key;
3728 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3729 u64 old_total = btrfs_super_total_bytes(super_copy);
3730 u64 old_size = device->total_bytes;
3731 u64 diff = device->total_bytes - new_size;
3733 if (device->is_tgtdev_for_dev_replace)
3736 path = btrfs_alloc_path();
3744 device->total_bytes = new_size;
3745 if (device->writeable) {
3746 device->fs_devices->total_rw_bytes -= diff;
3747 spin_lock(&root->fs_info->free_chunk_lock);
3748 root->fs_info->free_chunk_space -= diff;
3749 spin_unlock(&root->fs_info->free_chunk_lock);
3751 unlock_chunks(root);
3754 key.objectid = device->devid;
3755 key.offset = (u64)-1;
3756 key.type = BTRFS_DEV_EXTENT_KEY;
3759 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3763 ret = btrfs_previous_item(root, path, 0, key.type);
3768 btrfs_release_path(path);
3773 slot = path->slots[0];
3774 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3776 if (key.objectid != device->devid) {
3777 btrfs_release_path(path);
3781 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3782 length = btrfs_dev_extent_length(l, dev_extent);
3784 if (key.offset + length <= new_size) {
3785 btrfs_release_path(path);
3789 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3790 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3791 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3792 btrfs_release_path(path);
3794 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3796 if (ret && ret != -ENOSPC)
3800 } while (key.offset-- > 0);
3802 if (failed && !retried) {
3806 } else if (failed && retried) {
3810 device->total_bytes = old_size;
3811 if (device->writeable)
3812 device->fs_devices->total_rw_bytes += diff;
3813 spin_lock(&root->fs_info->free_chunk_lock);
3814 root->fs_info->free_chunk_space += diff;
3815 spin_unlock(&root->fs_info->free_chunk_lock);
3816 unlock_chunks(root);
3820 /* Shrinking succeeded, else we would be at "done". */
3821 trans = btrfs_start_transaction(root, 0);
3822 if (IS_ERR(trans)) {
3823 ret = PTR_ERR(trans);
3829 device->disk_total_bytes = new_size;
3830 /* Now btrfs_update_device() will change the on-disk size. */
3831 ret = btrfs_update_device(trans, device);
3833 unlock_chunks(root);
3834 btrfs_end_transaction(trans, root);
3837 WARN_ON(diff > old_total);
3838 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3839 unlock_chunks(root);
3840 btrfs_end_transaction(trans, root);
3842 btrfs_free_path(path);
3846 static int btrfs_add_system_chunk(struct btrfs_root *root,
3847 struct btrfs_key *key,
3848 struct btrfs_chunk *chunk, int item_size)
3850 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3851 struct btrfs_disk_key disk_key;
3855 array_size = btrfs_super_sys_array_size(super_copy);
3856 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3859 ptr = super_copy->sys_chunk_array + array_size;
3860 btrfs_cpu_key_to_disk(&disk_key, key);
3861 memcpy(ptr, &disk_key, sizeof(disk_key));
3862 ptr += sizeof(disk_key);
3863 memcpy(ptr, chunk, item_size);
3864 item_size += sizeof(disk_key);
3865 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3870 * sort the devices in descending order by max_avail, total_avail
3872 static int btrfs_cmp_device_info(const void *a, const void *b)
3874 const struct btrfs_device_info *di_a = a;
3875 const struct btrfs_device_info *di_b = b;
3877 if (di_a->max_avail > di_b->max_avail)
3879 if (di_a->max_avail < di_b->max_avail)
3881 if (di_a->total_avail > di_b->total_avail)
3883 if (di_a->total_avail < di_b->total_avail)
3888 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3889 [BTRFS_RAID_RAID10] = {
3892 .devs_max = 0, /* 0 == as many as possible */
3894 .devs_increment = 2,
3897 [BTRFS_RAID_RAID1] = {
3902 .devs_increment = 2,
3905 [BTRFS_RAID_DUP] = {
3910 .devs_increment = 1,
3913 [BTRFS_RAID_RAID0] = {
3918 .devs_increment = 1,
3921 [BTRFS_RAID_SINGLE] = {
3926 .devs_increment = 1,
3929 [BTRFS_RAID_RAID5] = {
3934 .devs_increment = 1,
3937 [BTRFS_RAID_RAID6] = {
3942 .devs_increment = 1,
3947 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3949 /* TODO allow them to set a preferred stripe size */
3953 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3955 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3958 btrfs_set_fs_incompat(info, RAID56);
3961 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3962 struct btrfs_root *extent_root, u64 start,
3965 struct btrfs_fs_info *info = extent_root->fs_info;
3966 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3967 struct list_head *cur;
3968 struct map_lookup *map = NULL;
3969 struct extent_map_tree *em_tree;
3970 struct extent_map *em;
3971 struct btrfs_device_info *devices_info = NULL;
3973 int num_stripes; /* total number of stripes to allocate */
3974 int data_stripes; /* number of stripes that count for
3976 int sub_stripes; /* sub_stripes info for map */
3977 int dev_stripes; /* stripes per dev */
3978 int devs_max; /* max devs to use */
3979 int devs_min; /* min devs needed */
3980 int devs_increment; /* ndevs has to be a multiple of this */
3981 int ncopies; /* how many copies to data has */
3983 u64 max_stripe_size;
3987 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3993 BUG_ON(!alloc_profile_is_valid(type, 0));
3995 if (list_empty(&fs_devices->alloc_list))
3998 index = __get_raid_index(type);
4000 sub_stripes = btrfs_raid_array[index].sub_stripes;
4001 dev_stripes = btrfs_raid_array[index].dev_stripes;
4002 devs_max = btrfs_raid_array[index].devs_max;
4003 devs_min = btrfs_raid_array[index].devs_min;
4004 devs_increment = btrfs_raid_array[index].devs_increment;
4005 ncopies = btrfs_raid_array[index].ncopies;
4007 if (type & BTRFS_BLOCK_GROUP_DATA) {
4008 max_stripe_size = 1024 * 1024 * 1024;
4009 max_chunk_size = 10 * max_stripe_size;
4010 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4011 /* for larger filesystems, use larger metadata chunks */
4012 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4013 max_stripe_size = 1024 * 1024 * 1024;
4015 max_stripe_size = 256 * 1024 * 1024;
4016 max_chunk_size = max_stripe_size;
4017 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4018 max_stripe_size = 32 * 1024 * 1024;
4019 max_chunk_size = 2 * max_stripe_size;
4021 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4026 /* we don't want a chunk larger than 10% of writeable space */
4027 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4030 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4035 cur = fs_devices->alloc_list.next;
4038 * in the first pass through the devices list, we gather information
4039 * about the available holes on each device.
4042 while (cur != &fs_devices->alloc_list) {
4043 struct btrfs_device *device;
4047 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4051 if (!device->writeable) {
4053 "btrfs: read-only device in alloc_list\n");
4057 if (!device->in_fs_metadata ||
4058 device->is_tgtdev_for_dev_replace)
4061 if (device->total_bytes > device->bytes_used)
4062 total_avail = device->total_bytes - device->bytes_used;
4066 /* If there is no space on this device, skip it. */
4067 if (total_avail == 0)
4070 ret = find_free_dev_extent(trans, device,
4071 max_stripe_size * dev_stripes,
4072 &dev_offset, &max_avail);
4073 if (ret && ret != -ENOSPC)
4077 max_avail = max_stripe_size * dev_stripes;
4079 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4082 if (ndevs == fs_devices->rw_devices) {
4083 WARN(1, "%s: found more than %llu devices\n",
4084 __func__, fs_devices->rw_devices);
4087 devices_info[ndevs].dev_offset = dev_offset;
4088 devices_info[ndevs].max_avail = max_avail;
4089 devices_info[ndevs].total_avail = total_avail;
4090 devices_info[ndevs].dev = device;
4095 * now sort the devices by hole size / available space
4097 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4098 btrfs_cmp_device_info, NULL);
4100 /* round down to number of usable stripes */
4101 ndevs -= ndevs % devs_increment;
4103 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4108 if (devs_max && ndevs > devs_max)
4111 * the primary goal is to maximize the number of stripes, so use as many
4112 * devices as possible, even if the stripes are not maximum sized.
4114 stripe_size = devices_info[ndevs-1].max_avail;
4115 num_stripes = ndevs * dev_stripes;
4118 * this will have to be fixed for RAID1 and RAID10 over
4121 data_stripes = num_stripes / ncopies;
4123 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4124 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4125 btrfs_super_stripesize(info->super_copy));
4126 data_stripes = num_stripes - 1;
4128 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4129 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4130 btrfs_super_stripesize(info->super_copy));
4131 data_stripes = num_stripes - 2;
4135 * Use the number of data stripes to figure out how big this chunk
4136 * is really going to be in terms of logical address space,
4137 * and compare that answer with the max chunk size
4139 if (stripe_size * data_stripes > max_chunk_size) {
4140 u64 mask = (1ULL << 24) - 1;
4141 stripe_size = max_chunk_size;
4142 do_div(stripe_size, data_stripes);
4144 /* bump the answer up to a 16MB boundary */
4145 stripe_size = (stripe_size + mask) & ~mask;
4147 /* but don't go higher than the limits we found
4148 * while searching for free extents
4150 if (stripe_size > devices_info[ndevs-1].max_avail)
4151 stripe_size = devices_info[ndevs-1].max_avail;
4154 do_div(stripe_size, dev_stripes);
4156 /* align to BTRFS_STRIPE_LEN */
4157 do_div(stripe_size, raid_stripe_len);
4158 stripe_size *= raid_stripe_len;
4160 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4165 map->num_stripes = num_stripes;
4167 for (i = 0; i < ndevs; ++i) {
4168 for (j = 0; j < dev_stripes; ++j) {
4169 int s = i * dev_stripes + j;
4170 map->stripes[s].dev = devices_info[i].dev;
4171 map->stripes[s].physical = devices_info[i].dev_offset +
4175 map->sector_size = extent_root->sectorsize;
4176 map->stripe_len = raid_stripe_len;
4177 map->io_align = raid_stripe_len;
4178 map->io_width = raid_stripe_len;
4180 map->sub_stripes = sub_stripes;
4182 num_bytes = stripe_size * data_stripes;
4184 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4186 em = alloc_extent_map();
4191 em->bdev = (struct block_device *)map;
4193 em->len = num_bytes;
4194 em->block_start = 0;
4195 em->block_len = em->len;
4196 em->orig_block_len = stripe_size;
4198 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4199 write_lock(&em_tree->lock);
4200 ret = add_extent_mapping(em_tree, em, 0);
4202 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4203 atomic_inc(&em->refs);
4205 write_unlock(&em_tree->lock);
4207 free_extent_map(em);
4211 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4212 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4215 goto error_del_extent;
4217 free_extent_map(em);
4218 check_raid56_incompat_flag(extent_root->fs_info, type);
4220 kfree(devices_info);
4224 write_lock(&em_tree->lock);
4225 remove_extent_mapping(em_tree, em);
4226 write_unlock(&em_tree->lock);
4228 /* One for our allocation */
4229 free_extent_map(em);
4230 /* One for the tree reference */
4231 free_extent_map(em);
4234 kfree(devices_info);
4238 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4239 struct btrfs_root *extent_root,
4240 u64 chunk_offset, u64 chunk_size)
4242 struct btrfs_key key;
4243 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4244 struct btrfs_device *device;
4245 struct btrfs_chunk *chunk;
4246 struct btrfs_stripe *stripe;
4247 struct extent_map_tree *em_tree;
4248 struct extent_map *em;
4249 struct map_lookup *map;
4256 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4257 read_lock(&em_tree->lock);
4258 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4259 read_unlock(&em_tree->lock);
4262 btrfs_crit(extent_root->fs_info, "unable to find logical "
4263 "%Lu len %Lu", chunk_offset, chunk_size);
4267 if (em->start != chunk_offset || em->len != chunk_size) {
4268 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4269 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4270 chunk_size, em->start, em->len);
4271 free_extent_map(em);
4275 map = (struct map_lookup *)em->bdev;
4276 item_size = btrfs_chunk_item_size(map->num_stripes);
4277 stripe_size = em->orig_block_len;
4279 chunk = kzalloc(item_size, GFP_NOFS);
4285 for (i = 0; i < map->num_stripes; i++) {
4286 device = map->stripes[i].dev;
4287 dev_offset = map->stripes[i].physical;
4289 device->bytes_used += stripe_size;
4290 ret = btrfs_update_device(trans, device);
4293 ret = btrfs_alloc_dev_extent(trans, device,
4294 chunk_root->root_key.objectid,
4295 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4296 chunk_offset, dev_offset,
4302 spin_lock(&extent_root->fs_info->free_chunk_lock);
4303 extent_root->fs_info->free_chunk_space -= (stripe_size *
4305 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4307 stripe = &chunk->stripe;
4308 for (i = 0; i < map->num_stripes; i++) {
4309 device = map->stripes[i].dev;
4310 dev_offset = map->stripes[i].physical;
4312 btrfs_set_stack_stripe_devid(stripe, device->devid);
4313 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4314 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4318 btrfs_set_stack_chunk_length(chunk, chunk_size);
4319 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4320 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4321 btrfs_set_stack_chunk_type(chunk, map->type);
4322 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4323 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4324 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4325 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4326 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4328 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4329 key.type = BTRFS_CHUNK_ITEM_KEY;
4330 key.offset = chunk_offset;
4332 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4333 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4335 * TODO: Cleanup of inserted chunk root in case of
4338 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4344 free_extent_map(em);
4349 * Chunk allocation falls into two parts. The first part does works
4350 * that make the new allocated chunk useable, but not do any operation
4351 * that modifies the chunk tree. The second part does the works that
4352 * require modifying the chunk tree. This division is important for the
4353 * bootstrap process of adding storage to a seed btrfs.
4355 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4356 struct btrfs_root *extent_root, u64 type)
4360 chunk_offset = find_next_chunk(extent_root->fs_info);
4361 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4364 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4365 struct btrfs_root *root,
4366 struct btrfs_device *device)
4369 u64 sys_chunk_offset;
4371 struct btrfs_fs_info *fs_info = root->fs_info;
4372 struct btrfs_root *extent_root = fs_info->extent_root;
4375 chunk_offset = find_next_chunk(fs_info);
4376 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4377 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4382 sys_chunk_offset = find_next_chunk(root->fs_info);
4383 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4384 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4387 btrfs_abort_transaction(trans, root, ret);
4391 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4393 btrfs_abort_transaction(trans, root, ret);
4398 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4400 struct extent_map *em;
4401 struct map_lookup *map;
4402 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4406 read_lock(&map_tree->map_tree.lock);
4407 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4408 read_unlock(&map_tree->map_tree.lock);
4412 if (btrfs_test_opt(root, DEGRADED)) {
4413 free_extent_map(em);
4417 map = (struct map_lookup *)em->bdev;
4418 for (i = 0; i < map->num_stripes; i++) {
4419 if (!map->stripes[i].dev->writeable) {
4424 free_extent_map(em);
4428 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4430 extent_map_tree_init(&tree->map_tree);
4433 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4435 struct extent_map *em;
4438 write_lock(&tree->map_tree.lock);
4439 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4441 remove_extent_mapping(&tree->map_tree, em);
4442 write_unlock(&tree->map_tree.lock);
4447 free_extent_map(em);
4448 /* once for the tree */
4449 free_extent_map(em);
4453 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4455 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4456 struct extent_map *em;
4457 struct map_lookup *map;
4458 struct extent_map_tree *em_tree = &map_tree->map_tree;
4461 read_lock(&em_tree->lock);
4462 em = lookup_extent_mapping(em_tree, logical, len);
4463 read_unlock(&em_tree->lock);
4466 * We could return errors for these cases, but that could get ugly and
4467 * we'd probably do the same thing which is just not do anything else
4468 * and exit, so return 1 so the callers don't try to use other copies.
4471 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4476 if (em->start > logical || em->start + em->len < logical) {
4477 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4478 "%Lu-%Lu\n", logical, logical+len, em->start,
4479 em->start + em->len);
4483 map = (struct map_lookup *)em->bdev;
4484 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4485 ret = map->num_stripes;
4486 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4487 ret = map->sub_stripes;
4488 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4490 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4494 free_extent_map(em);
4496 btrfs_dev_replace_lock(&fs_info->dev_replace);
4497 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4499 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4504 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4505 struct btrfs_mapping_tree *map_tree,
4508 struct extent_map *em;
4509 struct map_lookup *map;
4510 struct extent_map_tree *em_tree = &map_tree->map_tree;
4511 unsigned long len = root->sectorsize;
4513 read_lock(&em_tree->lock);
4514 em = lookup_extent_mapping(em_tree, logical, len);
4515 read_unlock(&em_tree->lock);
4518 BUG_ON(em->start > logical || em->start + em->len < logical);
4519 map = (struct map_lookup *)em->bdev;
4520 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4521 BTRFS_BLOCK_GROUP_RAID6)) {
4522 len = map->stripe_len * nr_data_stripes(map);
4524 free_extent_map(em);
4528 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4529 u64 logical, u64 len, int mirror_num)
4531 struct extent_map *em;
4532 struct map_lookup *map;
4533 struct extent_map_tree *em_tree = &map_tree->map_tree;
4536 read_lock(&em_tree->lock);
4537 em = lookup_extent_mapping(em_tree, logical, len);
4538 read_unlock(&em_tree->lock);
4541 BUG_ON(em->start > logical || em->start + em->len < logical);
4542 map = (struct map_lookup *)em->bdev;
4543 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4544 BTRFS_BLOCK_GROUP_RAID6))
4546 free_extent_map(em);
4550 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4551 struct map_lookup *map, int first, int num,
4552 int optimal, int dev_replace_is_ongoing)
4556 struct btrfs_device *srcdev;
4558 if (dev_replace_is_ongoing &&
4559 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4560 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4561 srcdev = fs_info->dev_replace.srcdev;
4566 * try to avoid the drive that is the source drive for a
4567 * dev-replace procedure, only choose it if no other non-missing
4568 * mirror is available
4570 for (tolerance = 0; tolerance < 2; tolerance++) {
4571 if (map->stripes[optimal].dev->bdev &&
4572 (tolerance || map->stripes[optimal].dev != srcdev))
4574 for (i = first; i < first + num; i++) {
4575 if (map->stripes[i].dev->bdev &&
4576 (tolerance || map->stripes[i].dev != srcdev))
4581 /* we couldn't find one that doesn't fail. Just return something
4582 * and the io error handling code will clean up eventually
4587 static inline int parity_smaller(u64 a, u64 b)
4592 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4593 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4595 struct btrfs_bio_stripe s;
4602 for (i = 0; i < bbio->num_stripes - 1; i++) {
4603 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4604 s = bbio->stripes[i];
4606 bbio->stripes[i] = bbio->stripes[i+1];
4607 raid_map[i] = raid_map[i+1];
4608 bbio->stripes[i+1] = s;
4616 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4617 u64 logical, u64 *length,
4618 struct btrfs_bio **bbio_ret,
4619 int mirror_num, u64 **raid_map_ret)
4621 struct extent_map *em;
4622 struct map_lookup *map;
4623 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4624 struct extent_map_tree *em_tree = &map_tree->map_tree;
4627 u64 stripe_end_offset;
4632 u64 *raid_map = NULL;
4638 struct btrfs_bio *bbio = NULL;
4639 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4640 int dev_replace_is_ongoing = 0;
4641 int num_alloc_stripes;
4642 int patch_the_first_stripe_for_dev_replace = 0;
4643 u64 physical_to_patch_in_first_stripe = 0;
4644 u64 raid56_full_stripe_start = (u64)-1;
4646 read_lock(&em_tree->lock);
4647 em = lookup_extent_mapping(em_tree, logical, *length);
4648 read_unlock(&em_tree->lock);
4651 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4656 if (em->start > logical || em->start + em->len < logical) {
4657 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4658 "found %Lu-%Lu\n", logical, em->start,
4659 em->start + em->len);
4663 map = (struct map_lookup *)em->bdev;
4664 offset = logical - em->start;
4666 stripe_len = map->stripe_len;
4669 * stripe_nr counts the total number of stripes we have to stride
4670 * to get to this block
4672 do_div(stripe_nr, stripe_len);
4674 stripe_offset = stripe_nr * stripe_len;
4675 BUG_ON(offset < stripe_offset);
4677 /* stripe_offset is the offset of this block in its stripe*/
4678 stripe_offset = offset - stripe_offset;
4680 /* if we're here for raid56, we need to know the stripe aligned start */
4681 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4682 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4683 raid56_full_stripe_start = offset;
4685 /* allow a write of a full stripe, but make sure we don't
4686 * allow straddling of stripes
4688 do_div(raid56_full_stripe_start, full_stripe_len);
4689 raid56_full_stripe_start *= full_stripe_len;
4692 if (rw & REQ_DISCARD) {
4693 /* we don't discard raid56 yet */
4695 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4699 *length = min_t(u64, em->len - offset, *length);
4700 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4702 /* For writes to RAID[56], allow a full stripeset across all disks.
4703 For other RAID types and for RAID[56] reads, just allow a single
4704 stripe (on a single disk). */
4705 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4707 max_len = stripe_len * nr_data_stripes(map) -
4708 (offset - raid56_full_stripe_start);
4710 /* we limit the length of each bio to what fits in a stripe */
4711 max_len = stripe_len - stripe_offset;
4713 *length = min_t(u64, em->len - offset, max_len);
4715 *length = em->len - offset;
4718 /* This is for when we're called from btrfs_merge_bio_hook() and all
4719 it cares about is the length */
4723 btrfs_dev_replace_lock(dev_replace);
4724 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4725 if (!dev_replace_is_ongoing)
4726 btrfs_dev_replace_unlock(dev_replace);
4728 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4729 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4730 dev_replace->tgtdev != NULL) {
4732 * in dev-replace case, for repair case (that's the only
4733 * case where the mirror is selected explicitly when
4734 * calling btrfs_map_block), blocks left of the left cursor
4735 * can also be read from the target drive.
4736 * For REQ_GET_READ_MIRRORS, the target drive is added as
4737 * the last one to the array of stripes. For READ, it also
4738 * needs to be supported using the same mirror number.
4739 * If the requested block is not left of the left cursor,
4740 * EIO is returned. This can happen because btrfs_num_copies()
4741 * returns one more in the dev-replace case.
4743 u64 tmp_length = *length;
4744 struct btrfs_bio *tmp_bbio = NULL;
4745 int tmp_num_stripes;
4746 u64 srcdev_devid = dev_replace->srcdev->devid;
4747 int index_srcdev = 0;
4749 u64 physical_of_found = 0;
4751 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4752 logical, &tmp_length, &tmp_bbio, 0, NULL);
4754 WARN_ON(tmp_bbio != NULL);
4758 tmp_num_stripes = tmp_bbio->num_stripes;
4759 if (mirror_num > tmp_num_stripes) {
4761 * REQ_GET_READ_MIRRORS does not contain this
4762 * mirror, that means that the requested area
4763 * is not left of the left cursor
4771 * process the rest of the function using the mirror_num
4772 * of the source drive. Therefore look it up first.
4773 * At the end, patch the device pointer to the one of the
4776 for (i = 0; i < tmp_num_stripes; i++) {
4777 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4779 * In case of DUP, in order to keep it
4780 * simple, only add the mirror with the
4781 * lowest physical address
4784 physical_of_found <=
4785 tmp_bbio->stripes[i].physical)
4790 tmp_bbio->stripes[i].physical;
4795 mirror_num = index_srcdev + 1;
4796 patch_the_first_stripe_for_dev_replace = 1;
4797 physical_to_patch_in_first_stripe = physical_of_found;
4806 } else if (mirror_num > map->num_stripes) {
4812 stripe_nr_orig = stripe_nr;
4813 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4814 do_div(stripe_nr_end, map->stripe_len);
4815 stripe_end_offset = stripe_nr_end * map->stripe_len -
4818 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4819 if (rw & REQ_DISCARD)
4820 num_stripes = min_t(u64, map->num_stripes,
4821 stripe_nr_end - stripe_nr_orig);
4822 stripe_index = do_div(stripe_nr, map->num_stripes);
4823 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4824 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4825 num_stripes = map->num_stripes;
4826 else if (mirror_num)
4827 stripe_index = mirror_num - 1;
4829 stripe_index = find_live_mirror(fs_info, map, 0,
4831 current->pid % map->num_stripes,
4832 dev_replace_is_ongoing);
4833 mirror_num = stripe_index + 1;
4836 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4837 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4838 num_stripes = map->num_stripes;
4839 } else if (mirror_num) {
4840 stripe_index = mirror_num - 1;
4845 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4846 int factor = map->num_stripes / map->sub_stripes;
4848 stripe_index = do_div(stripe_nr, factor);
4849 stripe_index *= map->sub_stripes;
4851 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4852 num_stripes = map->sub_stripes;
4853 else if (rw & REQ_DISCARD)
4854 num_stripes = min_t(u64, map->sub_stripes *
4855 (stripe_nr_end - stripe_nr_orig),
4857 else if (mirror_num)
4858 stripe_index += mirror_num - 1;
4860 int old_stripe_index = stripe_index;
4861 stripe_index = find_live_mirror(fs_info, map,
4863 map->sub_stripes, stripe_index +
4864 current->pid % map->sub_stripes,
4865 dev_replace_is_ongoing);
4866 mirror_num = stripe_index - old_stripe_index + 1;
4869 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4870 BTRFS_BLOCK_GROUP_RAID6)) {
4873 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4877 /* push stripe_nr back to the start of the full stripe */
4878 stripe_nr = raid56_full_stripe_start;
4879 do_div(stripe_nr, stripe_len);
4881 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4883 /* RAID[56] write or recovery. Return all stripes */
4884 num_stripes = map->num_stripes;
4885 max_errors = nr_parity_stripes(map);
4887 raid_map = kmalloc(sizeof(u64) * num_stripes,
4894 /* Work out the disk rotation on this stripe-set */
4896 rot = do_div(tmp, num_stripes);
4898 /* Fill in the logical address of each stripe */
4899 tmp = stripe_nr * nr_data_stripes(map);
4900 for (i = 0; i < nr_data_stripes(map); i++)
4901 raid_map[(i+rot) % num_stripes] =
4902 em->start + (tmp + i) * map->stripe_len;
4904 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4905 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4906 raid_map[(i+rot+1) % num_stripes] =
4909 *length = map->stripe_len;
4914 * Mirror #0 or #1 means the original data block.
4915 * Mirror #2 is RAID5 parity block.
4916 * Mirror #3 is RAID6 Q block.
4918 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4920 stripe_index = nr_data_stripes(map) +
4923 /* We distribute the parity blocks across stripes */
4924 tmp = stripe_nr + stripe_index;
4925 stripe_index = do_div(tmp, map->num_stripes);
4929 * after this do_div call, stripe_nr is the number of stripes
4930 * on this device we have to walk to find the data, and
4931 * stripe_index is the number of our device in the stripe array
4933 stripe_index = do_div(stripe_nr, map->num_stripes);
4934 mirror_num = stripe_index + 1;
4936 BUG_ON(stripe_index >= map->num_stripes);
4938 num_alloc_stripes = num_stripes;
4939 if (dev_replace_is_ongoing) {
4940 if (rw & (REQ_WRITE | REQ_DISCARD))
4941 num_alloc_stripes <<= 1;
4942 if (rw & REQ_GET_READ_MIRRORS)
4943 num_alloc_stripes++;
4945 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4951 atomic_set(&bbio->error, 0);
4953 if (rw & REQ_DISCARD) {
4955 int sub_stripes = 0;
4956 u64 stripes_per_dev = 0;
4957 u32 remaining_stripes = 0;
4958 u32 last_stripe = 0;
4961 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4962 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4965 sub_stripes = map->sub_stripes;
4967 factor = map->num_stripes / sub_stripes;
4968 stripes_per_dev = div_u64_rem(stripe_nr_end -
4971 &remaining_stripes);
4972 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4973 last_stripe *= sub_stripes;
4976 for (i = 0; i < num_stripes; i++) {
4977 bbio->stripes[i].physical =
4978 map->stripes[stripe_index].physical +
4979 stripe_offset + stripe_nr * map->stripe_len;
4980 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4982 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4983 BTRFS_BLOCK_GROUP_RAID10)) {
4984 bbio->stripes[i].length = stripes_per_dev *
4987 if (i / sub_stripes < remaining_stripes)
4988 bbio->stripes[i].length +=
4992 * Special for the first stripe and
4995 * |-------|...|-------|
4999 if (i < sub_stripes)
5000 bbio->stripes[i].length -=
5003 if (stripe_index >= last_stripe &&
5004 stripe_index <= (last_stripe +
5006 bbio->stripes[i].length -=
5009 if (i == sub_stripes - 1)
5012 bbio->stripes[i].length = *length;
5015 if (stripe_index == map->num_stripes) {
5016 /* This could only happen for RAID0/10 */
5022 for (i = 0; i < num_stripes; i++) {
5023 bbio->stripes[i].physical =
5024 map->stripes[stripe_index].physical +
5026 stripe_nr * map->stripe_len;
5027 bbio->stripes[i].dev =
5028 map->stripes[stripe_index].dev;
5033 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5034 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5035 BTRFS_BLOCK_GROUP_RAID10 |
5036 BTRFS_BLOCK_GROUP_RAID5 |
5037 BTRFS_BLOCK_GROUP_DUP)) {
5039 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5044 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5045 dev_replace->tgtdev != NULL) {
5046 int index_where_to_add;
5047 u64 srcdev_devid = dev_replace->srcdev->devid;
5050 * duplicate the write operations while the dev replace
5051 * procedure is running. Since the copying of the old disk
5052 * to the new disk takes place at run time while the
5053 * filesystem is mounted writable, the regular write
5054 * operations to the old disk have to be duplicated to go
5055 * to the new disk as well.
5056 * Note that device->missing is handled by the caller, and
5057 * that the write to the old disk is already set up in the
5060 index_where_to_add = num_stripes;
5061 for (i = 0; i < num_stripes; i++) {
5062 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5063 /* write to new disk, too */
5064 struct btrfs_bio_stripe *new =
5065 bbio->stripes + index_where_to_add;
5066 struct btrfs_bio_stripe *old =
5069 new->physical = old->physical;
5070 new->length = old->length;
5071 new->dev = dev_replace->tgtdev;
5072 index_where_to_add++;
5076 num_stripes = index_where_to_add;
5077 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5078 dev_replace->tgtdev != NULL) {
5079 u64 srcdev_devid = dev_replace->srcdev->devid;
5080 int index_srcdev = 0;
5082 u64 physical_of_found = 0;
5085 * During the dev-replace procedure, the target drive can
5086 * also be used to read data in case it is needed to repair
5087 * a corrupt block elsewhere. This is possible if the
5088 * requested area is left of the left cursor. In this area,
5089 * the target drive is a full copy of the source drive.
5091 for (i = 0; i < num_stripes; i++) {
5092 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5094 * In case of DUP, in order to keep it
5095 * simple, only add the mirror with the
5096 * lowest physical address
5099 physical_of_found <=
5100 bbio->stripes[i].physical)
5104 physical_of_found = bbio->stripes[i].physical;
5108 u64 length = map->stripe_len;
5110 if (physical_of_found + length <=
5111 dev_replace->cursor_left) {
5112 struct btrfs_bio_stripe *tgtdev_stripe =
5113 bbio->stripes + num_stripes;
5115 tgtdev_stripe->physical = physical_of_found;
5116 tgtdev_stripe->length =
5117 bbio->stripes[index_srcdev].length;
5118 tgtdev_stripe->dev = dev_replace->tgtdev;
5126 bbio->num_stripes = num_stripes;
5127 bbio->max_errors = max_errors;
5128 bbio->mirror_num = mirror_num;
5131 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5132 * mirror_num == num_stripes + 1 && dev_replace target drive is
5133 * available as a mirror
5135 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5136 WARN_ON(num_stripes > 1);
5137 bbio->stripes[0].dev = dev_replace->tgtdev;
5138 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5139 bbio->mirror_num = map->num_stripes + 1;
5142 sort_parity_stripes(bbio, raid_map);
5143 *raid_map_ret = raid_map;
5146 if (dev_replace_is_ongoing)
5147 btrfs_dev_replace_unlock(dev_replace);
5148 free_extent_map(em);
5152 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5153 u64 logical, u64 *length,
5154 struct btrfs_bio **bbio_ret, int mirror_num)
5156 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5160 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5161 u64 chunk_start, u64 physical, u64 devid,
5162 u64 **logical, int *naddrs, int *stripe_len)
5164 struct extent_map_tree *em_tree = &map_tree->map_tree;
5165 struct extent_map *em;
5166 struct map_lookup *map;
5174 read_lock(&em_tree->lock);
5175 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5176 read_unlock(&em_tree->lock);
5179 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5184 if (em->start != chunk_start) {
5185 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5186 em->start, chunk_start);
5187 free_extent_map(em);
5190 map = (struct map_lookup *)em->bdev;
5193 rmap_len = map->stripe_len;
5195 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5196 do_div(length, map->num_stripes / map->sub_stripes);
5197 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5198 do_div(length, map->num_stripes);
5199 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5200 BTRFS_BLOCK_GROUP_RAID6)) {
5201 do_div(length, nr_data_stripes(map));
5202 rmap_len = map->stripe_len * nr_data_stripes(map);
5205 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5206 BUG_ON(!buf); /* -ENOMEM */
5208 for (i = 0; i < map->num_stripes; i++) {
5209 if (devid && map->stripes[i].dev->devid != devid)
5211 if (map->stripes[i].physical > physical ||
5212 map->stripes[i].physical + length <= physical)
5215 stripe_nr = physical - map->stripes[i].physical;
5216 do_div(stripe_nr, map->stripe_len);
5218 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5219 stripe_nr = stripe_nr * map->num_stripes + i;
5220 do_div(stripe_nr, map->sub_stripes);
5221 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5222 stripe_nr = stripe_nr * map->num_stripes + i;
5223 } /* else if RAID[56], multiply by nr_data_stripes().
5224 * Alternatively, just use rmap_len below instead of
5225 * map->stripe_len */
5227 bytenr = chunk_start + stripe_nr * rmap_len;
5228 WARN_ON(nr >= map->num_stripes);
5229 for (j = 0; j < nr; j++) {
5230 if (buf[j] == bytenr)
5234 WARN_ON(nr >= map->num_stripes);
5241 *stripe_len = rmap_len;
5243 free_extent_map(em);
5247 static void btrfs_end_bio(struct bio *bio, int err)
5249 struct btrfs_bio *bbio = bio->bi_private;
5250 int is_orig_bio = 0;
5253 atomic_inc(&bbio->error);
5254 if (err == -EIO || err == -EREMOTEIO) {
5255 unsigned int stripe_index =
5256 btrfs_io_bio(bio)->stripe_index;
5257 struct btrfs_device *dev;
5259 BUG_ON(stripe_index >= bbio->num_stripes);
5260 dev = bbio->stripes[stripe_index].dev;
5262 if (bio->bi_rw & WRITE)
5263 btrfs_dev_stat_inc(dev,
5264 BTRFS_DEV_STAT_WRITE_ERRS);
5266 btrfs_dev_stat_inc(dev,
5267 BTRFS_DEV_STAT_READ_ERRS);
5268 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5269 btrfs_dev_stat_inc(dev,
5270 BTRFS_DEV_STAT_FLUSH_ERRS);
5271 btrfs_dev_stat_print_on_error(dev);
5276 if (bio == bbio->orig_bio)
5279 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5282 bio = bbio->orig_bio;
5284 bio->bi_private = bbio->private;
5285 bio->bi_end_io = bbio->end_io;
5286 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5287 /* only send an error to the higher layers if it is
5288 * beyond the tolerance of the btrfs bio
5290 if (atomic_read(&bbio->error) > bbio->max_errors) {
5294 * this bio is actually up to date, we didn't
5295 * go over the max number of errors
5297 set_bit(BIO_UPTODATE, &bio->bi_flags);
5302 bio_endio(bio, err);
5303 } else if (!is_orig_bio) {
5308 struct async_sched {
5311 struct btrfs_fs_info *info;
5312 struct btrfs_work work;
5316 * see run_scheduled_bios for a description of why bios are collected for
5319 * This will add one bio to the pending list for a device and make sure
5320 * the work struct is scheduled.
5322 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5323 struct btrfs_device *device,
5324 int rw, struct bio *bio)
5326 int should_queue = 1;
5327 struct btrfs_pending_bios *pending_bios;
5329 if (device->missing || !device->bdev) {
5330 bio_endio(bio, -EIO);
5334 /* don't bother with additional async steps for reads, right now */
5335 if (!(rw & REQ_WRITE)) {
5337 btrfsic_submit_bio(rw, bio);
5343 * nr_async_bios allows us to reliably return congestion to the
5344 * higher layers. Otherwise, the async bio makes it appear we have
5345 * made progress against dirty pages when we've really just put it
5346 * on a queue for later
5348 atomic_inc(&root->fs_info->nr_async_bios);
5349 WARN_ON(bio->bi_next);
5350 bio->bi_next = NULL;
5353 spin_lock(&device->io_lock);
5354 if (bio->bi_rw & REQ_SYNC)
5355 pending_bios = &device->pending_sync_bios;
5357 pending_bios = &device->pending_bios;
5359 if (pending_bios->tail)
5360 pending_bios->tail->bi_next = bio;
5362 pending_bios->tail = bio;
5363 if (!pending_bios->head)
5364 pending_bios->head = bio;
5365 if (device->running_pending)
5368 spin_unlock(&device->io_lock);
5371 btrfs_queue_worker(&root->fs_info->submit_workers,
5375 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5378 struct bio_vec *prev;
5379 struct request_queue *q = bdev_get_queue(bdev);
5380 unsigned short max_sectors = queue_max_sectors(q);
5381 struct bvec_merge_data bvm = {
5383 .bi_sector = sector,
5384 .bi_rw = bio->bi_rw,
5387 if (bio->bi_vcnt == 0) {
5392 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5393 if (bio_sectors(bio) > max_sectors)
5396 if (!q->merge_bvec_fn)
5399 bvm.bi_size = bio->bi_size - prev->bv_len;
5400 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5405 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5406 struct bio *bio, u64 physical, int dev_nr,
5409 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5411 bio->bi_private = bbio;
5412 btrfs_io_bio(bio)->stripe_index = dev_nr;
5413 bio->bi_end_io = btrfs_end_bio;
5414 bio->bi_sector = physical >> 9;
5417 struct rcu_string *name;
5420 name = rcu_dereference(dev->name);
5421 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5422 "(%s id %llu), size=%u\n", rw,
5423 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5424 name->str, dev->devid, bio->bi_size);
5428 bio->bi_bdev = dev->bdev;
5430 btrfs_schedule_bio(root, dev, rw, bio);
5432 btrfsic_submit_bio(rw, bio);
5435 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5436 struct bio *first_bio, struct btrfs_device *dev,
5437 int dev_nr, int rw, int async)
5439 struct bio_vec *bvec = first_bio->bi_io_vec;
5441 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5442 u64 physical = bbio->stripes[dev_nr].physical;
5445 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5449 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5450 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5451 bvec->bv_offset) < bvec->bv_len) {
5452 u64 len = bio->bi_size;
5454 atomic_inc(&bbio->stripes_pending);
5455 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5463 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5467 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5469 atomic_inc(&bbio->error);
5470 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5471 bio->bi_private = bbio->private;
5472 bio->bi_end_io = bbio->end_io;
5473 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5474 bio->bi_sector = logical >> 9;
5476 bio_endio(bio, -EIO);
5480 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5481 int mirror_num, int async_submit)
5483 struct btrfs_device *dev;
5484 struct bio *first_bio = bio;
5485 u64 logical = (u64)bio->bi_sector << 9;
5488 u64 *raid_map = NULL;
5492 struct btrfs_bio *bbio = NULL;
5494 length = bio->bi_size;
5495 map_length = length;
5497 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5498 mirror_num, &raid_map);
5499 if (ret) /* -ENOMEM */
5502 total_devs = bbio->num_stripes;
5503 bbio->orig_bio = first_bio;
5504 bbio->private = first_bio->bi_private;
5505 bbio->end_io = first_bio->bi_end_io;
5506 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5509 /* In this case, map_length has been set to the length of
5510 a single stripe; not the whole write */
5512 return raid56_parity_write(root, bio, bbio,
5513 raid_map, map_length);
5515 return raid56_parity_recover(root, bio, bbio,
5516 raid_map, map_length,
5521 if (map_length < length) {
5522 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5523 logical, length, map_length);
5527 while (dev_nr < total_devs) {
5528 dev = bbio->stripes[dev_nr].dev;
5529 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5530 bbio_error(bbio, first_bio, logical);
5536 * Check and see if we're ok with this bio based on it's size
5537 * and offset with the given device.
5539 if (!bio_size_ok(dev->bdev, first_bio,
5540 bbio->stripes[dev_nr].physical >> 9)) {
5541 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5542 dev_nr, rw, async_submit);
5548 if (dev_nr < total_devs - 1) {
5549 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5550 BUG_ON(!bio); /* -ENOMEM */
5555 submit_stripe_bio(root, bbio, bio,
5556 bbio->stripes[dev_nr].physical, dev_nr, rw,
5563 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5566 struct btrfs_device *device;
5567 struct btrfs_fs_devices *cur_devices;
5569 cur_devices = fs_info->fs_devices;
5570 while (cur_devices) {
5572 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5573 device = __find_device(&cur_devices->devices,
5578 cur_devices = cur_devices->seed;
5583 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5584 u64 devid, u8 *dev_uuid)
5586 struct btrfs_device *device;
5587 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5589 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5593 list_add(&device->dev_list, &fs_devices->devices);
5594 device->fs_devices = fs_devices;
5595 fs_devices->num_devices++;
5597 device->missing = 1;
5598 fs_devices->missing_devices++;
5604 * btrfs_alloc_device - allocate struct btrfs_device
5605 * @fs_info: used only for generating a new devid, can be NULL if
5606 * devid is provided (i.e. @devid != NULL).
5607 * @devid: a pointer to devid for this device. If NULL a new devid
5609 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5612 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5613 * on error. Returned struct is not linked onto any lists and can be
5614 * destroyed with kfree() right away.
5616 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5620 struct btrfs_device *dev;
5623 if (!devid && !fs_info) {
5625 return ERR_PTR(-EINVAL);
5628 dev = __alloc_device();
5637 ret = find_next_devid(fs_info, &tmp);
5640 return ERR_PTR(ret);
5646 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5648 generate_random_uuid(dev->uuid);
5650 dev->work.func = pending_bios_fn;
5655 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5656 struct extent_buffer *leaf,
5657 struct btrfs_chunk *chunk)
5659 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5660 struct map_lookup *map;
5661 struct extent_map *em;
5665 u8 uuid[BTRFS_UUID_SIZE];
5670 logical = key->offset;
5671 length = btrfs_chunk_length(leaf, chunk);
5673 read_lock(&map_tree->map_tree.lock);
5674 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5675 read_unlock(&map_tree->map_tree.lock);
5677 /* already mapped? */
5678 if (em && em->start <= logical && em->start + em->len > logical) {
5679 free_extent_map(em);
5682 free_extent_map(em);
5685 em = alloc_extent_map();
5688 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5689 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5691 free_extent_map(em);
5695 em->bdev = (struct block_device *)map;
5696 em->start = logical;
5699 em->block_start = 0;
5700 em->block_len = em->len;
5702 map->num_stripes = num_stripes;
5703 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5704 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5705 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5706 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5707 map->type = btrfs_chunk_type(leaf, chunk);
5708 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5709 for (i = 0; i < num_stripes; i++) {
5710 map->stripes[i].physical =
5711 btrfs_stripe_offset_nr(leaf, chunk, i);
5712 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5713 read_extent_buffer(leaf, uuid, (unsigned long)
5714 btrfs_stripe_dev_uuid_nr(chunk, i),
5716 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5718 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5720 free_extent_map(em);
5723 if (!map->stripes[i].dev) {
5724 map->stripes[i].dev =
5725 add_missing_dev(root, devid, uuid);
5726 if (!map->stripes[i].dev) {
5728 free_extent_map(em);
5732 map->stripes[i].dev->in_fs_metadata = 1;
5735 write_lock(&map_tree->map_tree.lock);
5736 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5737 write_unlock(&map_tree->map_tree.lock);
5738 BUG_ON(ret); /* Tree corruption */
5739 free_extent_map(em);
5744 static void fill_device_from_item(struct extent_buffer *leaf,
5745 struct btrfs_dev_item *dev_item,
5746 struct btrfs_device *device)
5750 device->devid = btrfs_device_id(leaf, dev_item);
5751 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5752 device->total_bytes = device->disk_total_bytes;
5753 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5754 device->type = btrfs_device_type(leaf, dev_item);
5755 device->io_align = btrfs_device_io_align(leaf, dev_item);
5756 device->io_width = btrfs_device_io_width(leaf, dev_item);
5757 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5758 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5759 device->is_tgtdev_for_dev_replace = 0;
5761 ptr = btrfs_device_uuid(dev_item);
5762 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5765 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5767 struct btrfs_fs_devices *fs_devices;
5770 BUG_ON(!mutex_is_locked(&uuid_mutex));
5772 fs_devices = root->fs_info->fs_devices->seed;
5773 while (fs_devices) {
5774 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5778 fs_devices = fs_devices->seed;
5781 fs_devices = find_fsid(fsid);
5787 fs_devices = clone_fs_devices(fs_devices);
5788 if (IS_ERR(fs_devices)) {
5789 ret = PTR_ERR(fs_devices);
5793 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5794 root->fs_info->bdev_holder);
5796 free_fs_devices(fs_devices);
5800 if (!fs_devices->seeding) {
5801 __btrfs_close_devices(fs_devices);
5802 free_fs_devices(fs_devices);
5807 fs_devices->seed = root->fs_info->fs_devices->seed;
5808 root->fs_info->fs_devices->seed = fs_devices;
5813 static int read_one_dev(struct btrfs_root *root,
5814 struct extent_buffer *leaf,
5815 struct btrfs_dev_item *dev_item)
5817 struct btrfs_device *device;
5820 u8 fs_uuid[BTRFS_UUID_SIZE];
5821 u8 dev_uuid[BTRFS_UUID_SIZE];
5823 devid = btrfs_device_id(leaf, dev_item);
5824 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5826 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5829 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5830 ret = open_seed_devices(root, fs_uuid);
5831 if (ret && !btrfs_test_opt(root, DEGRADED))
5835 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5836 if (!device || !device->bdev) {
5837 if (!btrfs_test_opt(root, DEGRADED))
5841 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5842 device = add_missing_dev(root, devid, dev_uuid);
5845 } else if (!device->missing) {
5847 * this happens when a device that was properly setup
5848 * in the device info lists suddenly goes bad.
5849 * device->bdev is NULL, and so we have to set
5850 * device->missing to one here
5852 root->fs_info->fs_devices->missing_devices++;
5853 device->missing = 1;
5857 if (device->fs_devices != root->fs_info->fs_devices) {
5858 BUG_ON(device->writeable);
5859 if (device->generation !=
5860 btrfs_device_generation(leaf, dev_item))
5864 fill_device_from_item(leaf, dev_item, device);
5865 device->in_fs_metadata = 1;
5866 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5867 device->fs_devices->total_rw_bytes += device->total_bytes;
5868 spin_lock(&root->fs_info->free_chunk_lock);
5869 root->fs_info->free_chunk_space += device->total_bytes -
5871 spin_unlock(&root->fs_info->free_chunk_lock);
5877 int btrfs_read_sys_array(struct btrfs_root *root)
5879 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5880 struct extent_buffer *sb;
5881 struct btrfs_disk_key *disk_key;
5882 struct btrfs_chunk *chunk;
5884 unsigned long sb_ptr;
5890 struct btrfs_key key;
5892 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5893 BTRFS_SUPER_INFO_SIZE);
5896 btrfs_set_buffer_uptodate(sb);
5897 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5899 * The sb extent buffer is artifical and just used to read the system array.
5900 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5901 * pages up-to-date when the page is larger: extent does not cover the
5902 * whole page and consequently check_page_uptodate does not find all
5903 * the page's extents up-to-date (the hole beyond sb),
5904 * write_extent_buffer then triggers a WARN_ON.
5906 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5907 * but sb spans only this function. Add an explicit SetPageUptodate call
5908 * to silence the warning eg. on PowerPC 64.
5910 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5911 SetPageUptodate(sb->pages[0]);
5913 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5914 array_size = btrfs_super_sys_array_size(super_copy);
5916 ptr = super_copy->sys_chunk_array;
5917 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5920 while (cur < array_size) {
5921 disk_key = (struct btrfs_disk_key *)ptr;
5922 btrfs_disk_key_to_cpu(&key, disk_key);
5924 len = sizeof(*disk_key); ptr += len;
5928 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5929 chunk = (struct btrfs_chunk *)sb_ptr;
5930 ret = read_one_chunk(root, &key, sb, chunk);
5933 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5934 len = btrfs_chunk_item_size(num_stripes);
5943 free_extent_buffer(sb);
5947 int btrfs_read_chunk_tree(struct btrfs_root *root)
5949 struct btrfs_path *path;
5950 struct extent_buffer *leaf;
5951 struct btrfs_key key;
5952 struct btrfs_key found_key;
5956 root = root->fs_info->chunk_root;
5958 path = btrfs_alloc_path();
5962 mutex_lock(&uuid_mutex);
5966 * Read all device items, and then all the chunk items. All
5967 * device items are found before any chunk item (their object id
5968 * is smaller than the lowest possible object id for a chunk
5969 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5971 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5974 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5978 leaf = path->nodes[0];
5979 slot = path->slots[0];
5980 if (slot >= btrfs_header_nritems(leaf)) {
5981 ret = btrfs_next_leaf(root, path);
5988 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5989 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5990 struct btrfs_dev_item *dev_item;
5991 dev_item = btrfs_item_ptr(leaf, slot,
5992 struct btrfs_dev_item);
5993 ret = read_one_dev(root, leaf, dev_item);
5996 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5997 struct btrfs_chunk *chunk;
5998 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5999 ret = read_one_chunk(root, &found_key, leaf, chunk);
6007 unlock_chunks(root);
6008 mutex_unlock(&uuid_mutex);
6010 btrfs_free_path(path);
6014 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6016 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6017 struct btrfs_device *device;
6019 mutex_lock(&fs_devices->device_list_mutex);
6020 list_for_each_entry(device, &fs_devices->devices, dev_list)
6021 device->dev_root = fs_info->dev_root;
6022 mutex_unlock(&fs_devices->device_list_mutex);
6025 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6029 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6030 btrfs_dev_stat_reset(dev, i);
6033 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6035 struct btrfs_key key;
6036 struct btrfs_key found_key;
6037 struct btrfs_root *dev_root = fs_info->dev_root;
6038 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6039 struct extent_buffer *eb;
6042 struct btrfs_device *device;
6043 struct btrfs_path *path = NULL;
6046 path = btrfs_alloc_path();
6052 mutex_lock(&fs_devices->device_list_mutex);
6053 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6055 struct btrfs_dev_stats_item *ptr;
6058 key.type = BTRFS_DEV_STATS_KEY;
6059 key.offset = device->devid;
6060 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6062 __btrfs_reset_dev_stats(device);
6063 device->dev_stats_valid = 1;
6064 btrfs_release_path(path);
6067 slot = path->slots[0];
6068 eb = path->nodes[0];
6069 btrfs_item_key_to_cpu(eb, &found_key, slot);
6070 item_size = btrfs_item_size_nr(eb, slot);
6072 ptr = btrfs_item_ptr(eb, slot,
6073 struct btrfs_dev_stats_item);
6075 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6076 if (item_size >= (1 + i) * sizeof(__le64))
6077 btrfs_dev_stat_set(device, i,
6078 btrfs_dev_stats_value(eb, ptr, i));
6080 btrfs_dev_stat_reset(device, i);
6083 device->dev_stats_valid = 1;
6084 btrfs_dev_stat_print_on_load(device);
6085 btrfs_release_path(path);
6087 mutex_unlock(&fs_devices->device_list_mutex);
6090 btrfs_free_path(path);
6091 return ret < 0 ? ret : 0;
6094 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6095 struct btrfs_root *dev_root,
6096 struct btrfs_device *device)
6098 struct btrfs_path *path;
6099 struct btrfs_key key;
6100 struct extent_buffer *eb;
6101 struct btrfs_dev_stats_item *ptr;
6106 key.type = BTRFS_DEV_STATS_KEY;
6107 key.offset = device->devid;
6109 path = btrfs_alloc_path();
6111 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6113 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6114 ret, rcu_str_deref(device->name));
6119 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6120 /* need to delete old one and insert a new one */
6121 ret = btrfs_del_item(trans, dev_root, path);
6123 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6124 rcu_str_deref(device->name), ret);
6131 /* need to insert a new item */
6132 btrfs_release_path(path);
6133 ret = btrfs_insert_empty_item(trans, dev_root, path,
6134 &key, sizeof(*ptr));
6136 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6137 rcu_str_deref(device->name), ret);
6142 eb = path->nodes[0];
6143 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6144 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6145 btrfs_set_dev_stats_value(eb, ptr, i,
6146 btrfs_dev_stat_read(device, i));
6147 btrfs_mark_buffer_dirty(eb);
6150 btrfs_free_path(path);
6155 * called from commit_transaction. Writes all changed device stats to disk.
6157 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6158 struct btrfs_fs_info *fs_info)
6160 struct btrfs_root *dev_root = fs_info->dev_root;
6161 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6162 struct btrfs_device *device;
6165 mutex_lock(&fs_devices->device_list_mutex);
6166 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6167 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6170 ret = update_dev_stat_item(trans, dev_root, device);
6172 device->dev_stats_dirty = 0;
6174 mutex_unlock(&fs_devices->device_list_mutex);
6179 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6181 btrfs_dev_stat_inc(dev, index);
6182 btrfs_dev_stat_print_on_error(dev);
6185 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6187 if (!dev->dev_stats_valid)
6189 printk_ratelimited_in_rcu(KERN_ERR
6190 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6191 rcu_str_deref(dev->name),
6192 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6193 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6194 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6195 btrfs_dev_stat_read(dev,
6196 BTRFS_DEV_STAT_CORRUPTION_ERRS),
6197 btrfs_dev_stat_read(dev,
6198 BTRFS_DEV_STAT_GENERATION_ERRS));
6201 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6205 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6206 if (btrfs_dev_stat_read(dev, i) != 0)
6208 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6209 return; /* all values == 0, suppress message */
6211 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6212 rcu_str_deref(dev->name),
6213 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6214 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6215 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6216 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6217 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6220 int btrfs_get_dev_stats(struct btrfs_root *root,
6221 struct btrfs_ioctl_get_dev_stats *stats)
6223 struct btrfs_device *dev;
6224 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6227 mutex_lock(&fs_devices->device_list_mutex);
6228 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6229 mutex_unlock(&fs_devices->device_list_mutex);
6233 "btrfs: get dev_stats failed, device not found\n");
6235 } else if (!dev->dev_stats_valid) {
6237 "btrfs: get dev_stats failed, not yet valid\n");
6239 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6240 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6241 if (stats->nr_items > i)
6243 btrfs_dev_stat_read_and_reset(dev, i);
6245 btrfs_dev_stat_reset(dev, i);
6248 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6249 if (stats->nr_items > i)
6250 stats->values[i] = btrfs_dev_stat_read(dev, i);
6252 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6253 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6257 int btrfs_scratch_superblock(struct btrfs_device *device)
6259 struct buffer_head *bh;
6260 struct btrfs_super_block *disk_super;
6262 bh = btrfs_read_dev_super(device->bdev);
6265 disk_super = (struct btrfs_super_block *)bh->b_data;
6267 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6268 set_buffer_dirty(bh);
6269 sync_dirty_buffer(bh);