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 = (unsigned long)btrfs_device_uuid(dev_item);
1428 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1429 ptr = (unsigned long)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,
1928 (unsigned long)btrfs_device_uuid(dev_item),
1930 read_extent_buffer(leaf, fs_uuid,
1931 (unsigned long)btrfs_device_fsid(dev_item),
1933 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1935 BUG_ON(!device); /* Logic error */
1937 if (device->fs_devices->seeding) {
1938 btrfs_set_device_generation(leaf, dev_item,
1939 device->generation);
1940 btrfs_mark_buffer_dirty(leaf);
1948 btrfs_free_path(path);
1952 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1954 struct request_queue *q;
1955 struct btrfs_trans_handle *trans;
1956 struct btrfs_device *device;
1957 struct block_device *bdev;
1958 struct list_head *devices;
1959 struct super_block *sb = root->fs_info->sb;
1960 struct rcu_string *name;
1962 int seeding_dev = 0;
1965 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1968 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1969 root->fs_info->bdev_holder);
1971 return PTR_ERR(bdev);
1973 if (root->fs_info->fs_devices->seeding) {
1975 down_write(&sb->s_umount);
1976 mutex_lock(&uuid_mutex);
1979 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1981 devices = &root->fs_info->fs_devices->devices;
1983 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1984 list_for_each_entry(device, devices, dev_list) {
1985 if (device->bdev == bdev) {
1988 &root->fs_info->fs_devices->device_list_mutex);
1992 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1994 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1995 if (IS_ERR(device)) {
1996 /* we can safely leave the fs_devices entry around */
1997 ret = PTR_ERR(device);
2001 name = rcu_string_strdup(device_path, GFP_NOFS);
2007 rcu_assign_pointer(device->name, name);
2009 trans = btrfs_start_transaction(root, 0);
2010 if (IS_ERR(trans)) {
2011 rcu_string_free(device->name);
2013 ret = PTR_ERR(trans);
2019 q = bdev_get_queue(bdev);
2020 if (blk_queue_discard(q))
2021 device->can_discard = 1;
2022 device->writeable = 1;
2023 device->generation = trans->transid;
2024 device->io_width = root->sectorsize;
2025 device->io_align = root->sectorsize;
2026 device->sector_size = root->sectorsize;
2027 device->total_bytes = i_size_read(bdev->bd_inode);
2028 device->disk_total_bytes = device->total_bytes;
2029 device->dev_root = root->fs_info->dev_root;
2030 device->bdev = bdev;
2031 device->in_fs_metadata = 1;
2032 device->is_tgtdev_for_dev_replace = 0;
2033 device->mode = FMODE_EXCL;
2034 set_blocksize(device->bdev, 4096);
2037 sb->s_flags &= ~MS_RDONLY;
2038 ret = btrfs_prepare_sprout(root);
2039 BUG_ON(ret); /* -ENOMEM */
2042 device->fs_devices = root->fs_info->fs_devices;
2044 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2045 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2046 list_add(&device->dev_alloc_list,
2047 &root->fs_info->fs_devices->alloc_list);
2048 root->fs_info->fs_devices->num_devices++;
2049 root->fs_info->fs_devices->open_devices++;
2050 root->fs_info->fs_devices->rw_devices++;
2051 root->fs_info->fs_devices->total_devices++;
2052 if (device->can_discard)
2053 root->fs_info->fs_devices->num_can_discard++;
2054 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2056 spin_lock(&root->fs_info->free_chunk_lock);
2057 root->fs_info->free_chunk_space += device->total_bytes;
2058 spin_unlock(&root->fs_info->free_chunk_lock);
2060 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2061 root->fs_info->fs_devices->rotating = 1;
2063 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2064 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2065 total_bytes + device->total_bytes);
2067 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2068 btrfs_set_super_num_devices(root->fs_info->super_copy,
2070 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2073 ret = init_first_rw_device(trans, root, device);
2075 btrfs_abort_transaction(trans, root, ret);
2078 ret = btrfs_finish_sprout(trans, root);
2080 btrfs_abort_transaction(trans, root, ret);
2084 ret = btrfs_add_device(trans, root, device);
2086 btrfs_abort_transaction(trans, root, ret);
2092 * we've got more storage, clear any full flags on the space
2095 btrfs_clear_space_info_full(root->fs_info);
2097 unlock_chunks(root);
2098 root->fs_info->num_tolerated_disk_barrier_failures =
2099 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2100 ret = btrfs_commit_transaction(trans, root);
2103 mutex_unlock(&uuid_mutex);
2104 up_write(&sb->s_umount);
2106 if (ret) /* transaction commit */
2109 ret = btrfs_relocate_sys_chunks(root);
2111 btrfs_error(root->fs_info, ret,
2112 "Failed to relocate sys chunks after "
2113 "device initialization. This can be fixed "
2114 "using the \"btrfs balance\" command.");
2115 trans = btrfs_attach_transaction(root);
2116 if (IS_ERR(trans)) {
2117 if (PTR_ERR(trans) == -ENOENT)
2119 return PTR_ERR(trans);
2121 ret = btrfs_commit_transaction(trans, root);
2127 unlock_chunks(root);
2128 btrfs_end_transaction(trans, root);
2129 rcu_string_free(device->name);
2132 blkdev_put(bdev, FMODE_EXCL);
2134 mutex_unlock(&uuid_mutex);
2135 up_write(&sb->s_umount);
2140 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2141 struct btrfs_device **device_out)
2143 struct request_queue *q;
2144 struct btrfs_device *device;
2145 struct block_device *bdev;
2146 struct btrfs_fs_info *fs_info = root->fs_info;
2147 struct list_head *devices;
2148 struct rcu_string *name;
2149 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2153 if (fs_info->fs_devices->seeding)
2156 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2157 fs_info->bdev_holder);
2159 return PTR_ERR(bdev);
2161 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2163 devices = &fs_info->fs_devices->devices;
2164 list_for_each_entry(device, devices, dev_list) {
2165 if (device->bdev == bdev) {
2171 device = btrfs_alloc_device(NULL, &devid, NULL);
2172 if (IS_ERR(device)) {
2173 ret = PTR_ERR(device);
2177 name = rcu_string_strdup(device_path, GFP_NOFS);
2183 rcu_assign_pointer(device->name, name);
2185 q = bdev_get_queue(bdev);
2186 if (blk_queue_discard(q))
2187 device->can_discard = 1;
2188 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2189 device->writeable = 1;
2190 device->generation = 0;
2191 device->io_width = root->sectorsize;
2192 device->io_align = root->sectorsize;
2193 device->sector_size = root->sectorsize;
2194 device->total_bytes = i_size_read(bdev->bd_inode);
2195 device->disk_total_bytes = device->total_bytes;
2196 device->dev_root = fs_info->dev_root;
2197 device->bdev = bdev;
2198 device->in_fs_metadata = 1;
2199 device->is_tgtdev_for_dev_replace = 1;
2200 device->mode = FMODE_EXCL;
2201 set_blocksize(device->bdev, 4096);
2202 device->fs_devices = fs_info->fs_devices;
2203 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2204 fs_info->fs_devices->num_devices++;
2205 fs_info->fs_devices->open_devices++;
2206 if (device->can_discard)
2207 fs_info->fs_devices->num_can_discard++;
2208 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2210 *device_out = device;
2214 blkdev_put(bdev, FMODE_EXCL);
2218 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2219 struct btrfs_device *tgtdev)
2221 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2222 tgtdev->io_width = fs_info->dev_root->sectorsize;
2223 tgtdev->io_align = fs_info->dev_root->sectorsize;
2224 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2225 tgtdev->dev_root = fs_info->dev_root;
2226 tgtdev->in_fs_metadata = 1;
2229 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2230 struct btrfs_device *device)
2233 struct btrfs_path *path;
2234 struct btrfs_root *root;
2235 struct btrfs_dev_item *dev_item;
2236 struct extent_buffer *leaf;
2237 struct btrfs_key key;
2239 root = device->dev_root->fs_info->chunk_root;
2241 path = btrfs_alloc_path();
2245 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2246 key.type = BTRFS_DEV_ITEM_KEY;
2247 key.offset = device->devid;
2249 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2258 leaf = path->nodes[0];
2259 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2261 btrfs_set_device_id(leaf, dev_item, device->devid);
2262 btrfs_set_device_type(leaf, dev_item, device->type);
2263 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2264 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2265 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2266 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2267 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2268 btrfs_mark_buffer_dirty(leaf);
2271 btrfs_free_path(path);
2275 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2276 struct btrfs_device *device, u64 new_size)
2278 struct btrfs_super_block *super_copy =
2279 device->dev_root->fs_info->super_copy;
2280 u64 old_total = btrfs_super_total_bytes(super_copy);
2281 u64 diff = new_size - device->total_bytes;
2283 if (!device->writeable)
2285 if (new_size <= device->total_bytes ||
2286 device->is_tgtdev_for_dev_replace)
2289 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2290 device->fs_devices->total_rw_bytes += diff;
2292 device->total_bytes = new_size;
2293 device->disk_total_bytes = new_size;
2294 btrfs_clear_space_info_full(device->dev_root->fs_info);
2296 return btrfs_update_device(trans, device);
2299 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2300 struct btrfs_device *device, u64 new_size)
2303 lock_chunks(device->dev_root);
2304 ret = __btrfs_grow_device(trans, device, new_size);
2305 unlock_chunks(device->dev_root);
2309 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2310 struct btrfs_root *root,
2311 u64 chunk_tree, u64 chunk_objectid,
2315 struct btrfs_path *path;
2316 struct btrfs_key key;
2318 root = root->fs_info->chunk_root;
2319 path = btrfs_alloc_path();
2323 key.objectid = chunk_objectid;
2324 key.offset = chunk_offset;
2325 key.type = BTRFS_CHUNK_ITEM_KEY;
2327 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2330 else if (ret > 0) { /* Logic error or corruption */
2331 btrfs_error(root->fs_info, -ENOENT,
2332 "Failed lookup while freeing chunk.");
2337 ret = btrfs_del_item(trans, root, path);
2339 btrfs_error(root->fs_info, ret,
2340 "Failed to delete chunk item.");
2342 btrfs_free_path(path);
2346 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2349 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2350 struct btrfs_disk_key *disk_key;
2351 struct btrfs_chunk *chunk;
2358 struct btrfs_key key;
2360 array_size = btrfs_super_sys_array_size(super_copy);
2362 ptr = super_copy->sys_chunk_array;
2365 while (cur < array_size) {
2366 disk_key = (struct btrfs_disk_key *)ptr;
2367 btrfs_disk_key_to_cpu(&key, disk_key);
2369 len = sizeof(*disk_key);
2371 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2372 chunk = (struct btrfs_chunk *)(ptr + len);
2373 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2374 len += btrfs_chunk_item_size(num_stripes);
2379 if (key.objectid == chunk_objectid &&
2380 key.offset == chunk_offset) {
2381 memmove(ptr, ptr + len, array_size - (cur + len));
2383 btrfs_set_super_sys_array_size(super_copy, array_size);
2392 static int btrfs_relocate_chunk(struct btrfs_root *root,
2393 u64 chunk_tree, u64 chunk_objectid,
2396 struct extent_map_tree *em_tree;
2397 struct btrfs_root *extent_root;
2398 struct btrfs_trans_handle *trans;
2399 struct extent_map *em;
2400 struct map_lookup *map;
2404 root = root->fs_info->chunk_root;
2405 extent_root = root->fs_info->extent_root;
2406 em_tree = &root->fs_info->mapping_tree.map_tree;
2408 ret = btrfs_can_relocate(extent_root, chunk_offset);
2412 /* step one, relocate all the extents inside this chunk */
2413 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2417 trans = btrfs_start_transaction(root, 0);
2418 if (IS_ERR(trans)) {
2419 ret = PTR_ERR(trans);
2420 btrfs_std_error(root->fs_info, ret);
2427 * step two, delete the device extents and the
2428 * chunk tree entries
2430 read_lock(&em_tree->lock);
2431 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2432 read_unlock(&em_tree->lock);
2434 BUG_ON(!em || em->start > chunk_offset ||
2435 em->start + em->len < chunk_offset);
2436 map = (struct map_lookup *)em->bdev;
2438 for (i = 0; i < map->num_stripes; i++) {
2439 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2440 map->stripes[i].physical);
2443 if (map->stripes[i].dev) {
2444 ret = btrfs_update_device(trans, map->stripes[i].dev);
2448 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2453 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2455 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2456 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2460 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2463 write_lock(&em_tree->lock);
2464 remove_extent_mapping(em_tree, em);
2465 write_unlock(&em_tree->lock);
2470 /* once for the tree */
2471 free_extent_map(em);
2473 free_extent_map(em);
2475 unlock_chunks(root);
2476 btrfs_end_transaction(trans, root);
2480 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2482 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2483 struct btrfs_path *path;
2484 struct extent_buffer *leaf;
2485 struct btrfs_chunk *chunk;
2486 struct btrfs_key key;
2487 struct btrfs_key found_key;
2488 u64 chunk_tree = chunk_root->root_key.objectid;
2490 bool retried = false;
2494 path = btrfs_alloc_path();
2499 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2500 key.offset = (u64)-1;
2501 key.type = BTRFS_CHUNK_ITEM_KEY;
2504 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2507 BUG_ON(ret == 0); /* Corruption */
2509 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2516 leaf = path->nodes[0];
2517 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2519 chunk = btrfs_item_ptr(leaf, path->slots[0],
2520 struct btrfs_chunk);
2521 chunk_type = btrfs_chunk_type(leaf, chunk);
2522 btrfs_release_path(path);
2524 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2525 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2534 if (found_key.offset == 0)
2536 key.offset = found_key.offset - 1;
2539 if (failed && !retried) {
2543 } else if (failed && retried) {
2548 btrfs_free_path(path);
2552 static int insert_balance_item(struct btrfs_root *root,
2553 struct btrfs_balance_control *bctl)
2555 struct btrfs_trans_handle *trans;
2556 struct btrfs_balance_item *item;
2557 struct btrfs_disk_balance_args disk_bargs;
2558 struct btrfs_path *path;
2559 struct extent_buffer *leaf;
2560 struct btrfs_key key;
2563 path = btrfs_alloc_path();
2567 trans = btrfs_start_transaction(root, 0);
2568 if (IS_ERR(trans)) {
2569 btrfs_free_path(path);
2570 return PTR_ERR(trans);
2573 key.objectid = BTRFS_BALANCE_OBJECTID;
2574 key.type = BTRFS_BALANCE_ITEM_KEY;
2577 ret = btrfs_insert_empty_item(trans, root, path, &key,
2582 leaf = path->nodes[0];
2583 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2585 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2587 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2588 btrfs_set_balance_data(leaf, item, &disk_bargs);
2589 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2590 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2591 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2592 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2594 btrfs_set_balance_flags(leaf, item, bctl->flags);
2596 btrfs_mark_buffer_dirty(leaf);
2598 btrfs_free_path(path);
2599 err = btrfs_commit_transaction(trans, root);
2605 static int del_balance_item(struct btrfs_root *root)
2607 struct btrfs_trans_handle *trans;
2608 struct btrfs_path *path;
2609 struct btrfs_key key;
2612 path = btrfs_alloc_path();
2616 trans = btrfs_start_transaction(root, 0);
2617 if (IS_ERR(trans)) {
2618 btrfs_free_path(path);
2619 return PTR_ERR(trans);
2622 key.objectid = BTRFS_BALANCE_OBJECTID;
2623 key.type = BTRFS_BALANCE_ITEM_KEY;
2626 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2634 ret = btrfs_del_item(trans, root, path);
2636 btrfs_free_path(path);
2637 err = btrfs_commit_transaction(trans, root);
2644 * This is a heuristic used to reduce the number of chunks balanced on
2645 * resume after balance was interrupted.
2647 static void update_balance_args(struct btrfs_balance_control *bctl)
2650 * Turn on soft mode for chunk types that were being converted.
2652 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2653 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2654 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2655 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2656 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2657 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2660 * Turn on usage filter if is not already used. The idea is
2661 * that chunks that we have already balanced should be
2662 * reasonably full. Don't do it for chunks that are being
2663 * converted - that will keep us from relocating unconverted
2664 * (albeit full) chunks.
2666 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2667 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2668 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2669 bctl->data.usage = 90;
2671 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2672 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2673 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2674 bctl->sys.usage = 90;
2676 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2677 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2678 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2679 bctl->meta.usage = 90;
2684 * Should be called with both balance and volume mutexes held to
2685 * serialize other volume operations (add_dev/rm_dev/resize) with
2686 * restriper. Same goes for unset_balance_control.
2688 static void set_balance_control(struct btrfs_balance_control *bctl)
2690 struct btrfs_fs_info *fs_info = bctl->fs_info;
2692 BUG_ON(fs_info->balance_ctl);
2694 spin_lock(&fs_info->balance_lock);
2695 fs_info->balance_ctl = bctl;
2696 spin_unlock(&fs_info->balance_lock);
2699 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2701 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2703 BUG_ON(!fs_info->balance_ctl);
2705 spin_lock(&fs_info->balance_lock);
2706 fs_info->balance_ctl = NULL;
2707 spin_unlock(&fs_info->balance_lock);
2713 * Balance filters. Return 1 if chunk should be filtered out
2714 * (should not be balanced).
2716 static int chunk_profiles_filter(u64 chunk_type,
2717 struct btrfs_balance_args *bargs)
2719 chunk_type = chunk_to_extended(chunk_type) &
2720 BTRFS_EXTENDED_PROFILE_MASK;
2722 if (bargs->profiles & chunk_type)
2728 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2729 struct btrfs_balance_args *bargs)
2731 struct btrfs_block_group_cache *cache;
2732 u64 chunk_used, user_thresh;
2735 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2736 chunk_used = btrfs_block_group_used(&cache->item);
2738 if (bargs->usage == 0)
2740 else if (bargs->usage > 100)
2741 user_thresh = cache->key.offset;
2743 user_thresh = div_factor_fine(cache->key.offset,
2746 if (chunk_used < user_thresh)
2749 btrfs_put_block_group(cache);
2753 static int chunk_devid_filter(struct extent_buffer *leaf,
2754 struct btrfs_chunk *chunk,
2755 struct btrfs_balance_args *bargs)
2757 struct btrfs_stripe *stripe;
2758 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2761 for (i = 0; i < num_stripes; i++) {
2762 stripe = btrfs_stripe_nr(chunk, i);
2763 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2770 /* [pstart, pend) */
2771 static int chunk_drange_filter(struct extent_buffer *leaf,
2772 struct btrfs_chunk *chunk,
2774 struct btrfs_balance_args *bargs)
2776 struct btrfs_stripe *stripe;
2777 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2783 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2786 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2787 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2788 factor = num_stripes / 2;
2789 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2790 factor = num_stripes - 1;
2791 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2792 factor = num_stripes - 2;
2794 factor = num_stripes;
2797 for (i = 0; i < num_stripes; i++) {
2798 stripe = btrfs_stripe_nr(chunk, i);
2799 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2802 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2803 stripe_length = btrfs_chunk_length(leaf, chunk);
2804 do_div(stripe_length, factor);
2806 if (stripe_offset < bargs->pend &&
2807 stripe_offset + stripe_length > bargs->pstart)
2814 /* [vstart, vend) */
2815 static int chunk_vrange_filter(struct extent_buffer *leaf,
2816 struct btrfs_chunk *chunk,
2818 struct btrfs_balance_args *bargs)
2820 if (chunk_offset < bargs->vend &&
2821 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2822 /* at least part of the chunk is inside this vrange */
2828 static int chunk_soft_convert_filter(u64 chunk_type,
2829 struct btrfs_balance_args *bargs)
2831 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2834 chunk_type = chunk_to_extended(chunk_type) &
2835 BTRFS_EXTENDED_PROFILE_MASK;
2837 if (bargs->target == chunk_type)
2843 static int should_balance_chunk(struct btrfs_root *root,
2844 struct extent_buffer *leaf,
2845 struct btrfs_chunk *chunk, u64 chunk_offset)
2847 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2848 struct btrfs_balance_args *bargs = NULL;
2849 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2852 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2853 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2857 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2858 bargs = &bctl->data;
2859 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2861 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2862 bargs = &bctl->meta;
2864 /* profiles filter */
2865 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2866 chunk_profiles_filter(chunk_type, bargs)) {
2871 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2872 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2877 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2878 chunk_devid_filter(leaf, chunk, bargs)) {
2882 /* drange filter, makes sense only with devid filter */
2883 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2884 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2889 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2890 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2894 /* soft profile changing mode */
2895 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2896 chunk_soft_convert_filter(chunk_type, bargs)) {
2903 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2905 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2906 struct btrfs_root *chunk_root = fs_info->chunk_root;
2907 struct btrfs_root *dev_root = fs_info->dev_root;
2908 struct list_head *devices;
2909 struct btrfs_device *device;
2912 struct btrfs_chunk *chunk;
2913 struct btrfs_path *path;
2914 struct btrfs_key key;
2915 struct btrfs_key found_key;
2916 struct btrfs_trans_handle *trans;
2917 struct extent_buffer *leaf;
2920 int enospc_errors = 0;
2921 bool counting = true;
2923 /* step one make some room on all the devices */
2924 devices = &fs_info->fs_devices->devices;
2925 list_for_each_entry(device, devices, dev_list) {
2926 old_size = device->total_bytes;
2927 size_to_free = div_factor(old_size, 1);
2928 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2929 if (!device->writeable ||
2930 device->total_bytes - device->bytes_used > size_to_free ||
2931 device->is_tgtdev_for_dev_replace)
2934 ret = btrfs_shrink_device(device, old_size - size_to_free);
2939 trans = btrfs_start_transaction(dev_root, 0);
2940 BUG_ON(IS_ERR(trans));
2942 ret = btrfs_grow_device(trans, device, old_size);
2945 btrfs_end_transaction(trans, dev_root);
2948 /* step two, relocate all the chunks */
2949 path = btrfs_alloc_path();
2955 /* zero out stat counters */
2956 spin_lock(&fs_info->balance_lock);
2957 memset(&bctl->stat, 0, sizeof(bctl->stat));
2958 spin_unlock(&fs_info->balance_lock);
2960 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2961 key.offset = (u64)-1;
2962 key.type = BTRFS_CHUNK_ITEM_KEY;
2965 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2966 atomic_read(&fs_info->balance_cancel_req)) {
2971 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2976 * this shouldn't happen, it means the last relocate
2980 BUG(); /* FIXME break ? */
2982 ret = btrfs_previous_item(chunk_root, path, 0,
2983 BTRFS_CHUNK_ITEM_KEY);
2989 leaf = path->nodes[0];
2990 slot = path->slots[0];
2991 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2993 if (found_key.objectid != key.objectid)
2996 /* chunk zero is special */
2997 if (found_key.offset == 0)
3000 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3003 spin_lock(&fs_info->balance_lock);
3004 bctl->stat.considered++;
3005 spin_unlock(&fs_info->balance_lock);
3008 ret = should_balance_chunk(chunk_root, leaf, chunk,
3010 btrfs_release_path(path);
3015 spin_lock(&fs_info->balance_lock);
3016 bctl->stat.expected++;
3017 spin_unlock(&fs_info->balance_lock);
3021 ret = btrfs_relocate_chunk(chunk_root,
3022 chunk_root->root_key.objectid,
3025 if (ret && ret != -ENOSPC)
3027 if (ret == -ENOSPC) {
3030 spin_lock(&fs_info->balance_lock);
3031 bctl->stat.completed++;
3032 spin_unlock(&fs_info->balance_lock);
3035 key.offset = found_key.offset - 1;
3039 btrfs_release_path(path);
3044 btrfs_free_path(path);
3045 if (enospc_errors) {
3046 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3056 * alloc_profile_is_valid - see if a given profile is valid and reduced
3057 * @flags: profile to validate
3058 * @extended: if true @flags is treated as an extended profile
3060 static int alloc_profile_is_valid(u64 flags, int extended)
3062 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3063 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3065 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3067 /* 1) check that all other bits are zeroed */
3071 /* 2) see if profile is reduced */
3073 return !extended; /* "0" is valid for usual profiles */
3075 /* true if exactly one bit set */
3076 return (flags & (flags - 1)) == 0;
3079 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3081 /* cancel requested || normal exit path */
3082 return atomic_read(&fs_info->balance_cancel_req) ||
3083 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3084 atomic_read(&fs_info->balance_cancel_req) == 0);
3087 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3091 unset_balance_control(fs_info);
3092 ret = del_balance_item(fs_info->tree_root);
3094 btrfs_std_error(fs_info, ret);
3096 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3100 * Should be called with both balance and volume mutexes held
3102 int btrfs_balance(struct btrfs_balance_control *bctl,
3103 struct btrfs_ioctl_balance_args *bargs)
3105 struct btrfs_fs_info *fs_info = bctl->fs_info;
3112 if (btrfs_fs_closing(fs_info) ||
3113 atomic_read(&fs_info->balance_pause_req) ||
3114 atomic_read(&fs_info->balance_cancel_req)) {
3119 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3120 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3124 * In case of mixed groups both data and meta should be picked,
3125 * and identical options should be given for both of them.
3127 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3128 if (mixed && (bctl->flags & allowed)) {
3129 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3130 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3131 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3132 printk(KERN_ERR "btrfs: with mixed groups data and "
3133 "metadata balance options must be the same\n");
3139 num_devices = fs_info->fs_devices->num_devices;
3140 btrfs_dev_replace_lock(&fs_info->dev_replace);
3141 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3142 BUG_ON(num_devices < 1);
3145 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3146 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3147 if (num_devices == 1)
3148 allowed |= BTRFS_BLOCK_GROUP_DUP;
3149 else if (num_devices > 1)
3150 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3151 if (num_devices > 2)
3152 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3153 if (num_devices > 3)
3154 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3155 BTRFS_BLOCK_GROUP_RAID6);
3156 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3157 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3158 (bctl->data.target & ~allowed))) {
3159 printk(KERN_ERR "btrfs: unable to start balance with target "
3160 "data profile %llu\n",
3165 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3166 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3167 (bctl->meta.target & ~allowed))) {
3168 printk(KERN_ERR "btrfs: unable to start balance with target "
3169 "metadata profile %llu\n",
3174 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3175 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3176 (bctl->sys.target & ~allowed))) {
3177 printk(KERN_ERR "btrfs: unable to start balance with target "
3178 "system profile %llu\n",
3184 /* allow dup'ed data chunks only in mixed mode */
3185 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3186 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3187 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3192 /* allow to reduce meta or sys integrity only if force set */
3193 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3194 BTRFS_BLOCK_GROUP_RAID10 |
3195 BTRFS_BLOCK_GROUP_RAID5 |
3196 BTRFS_BLOCK_GROUP_RAID6;
3198 seq = read_seqbegin(&fs_info->profiles_lock);
3200 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3201 (fs_info->avail_system_alloc_bits & allowed) &&
3202 !(bctl->sys.target & allowed)) ||
3203 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3204 (fs_info->avail_metadata_alloc_bits & allowed) &&
3205 !(bctl->meta.target & allowed))) {
3206 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3207 printk(KERN_INFO "btrfs: force reducing metadata "
3210 printk(KERN_ERR "btrfs: balance will reduce metadata "
3211 "integrity, use force if you want this\n");
3216 } while (read_seqretry(&fs_info->profiles_lock, seq));
3218 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3219 int num_tolerated_disk_barrier_failures;
3220 u64 target = bctl->sys.target;
3222 num_tolerated_disk_barrier_failures =
3223 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3224 if (num_tolerated_disk_barrier_failures > 0 &&
3226 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3227 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3228 num_tolerated_disk_barrier_failures = 0;
3229 else if (num_tolerated_disk_barrier_failures > 1 &&
3231 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3232 num_tolerated_disk_barrier_failures = 1;
3234 fs_info->num_tolerated_disk_barrier_failures =
3235 num_tolerated_disk_barrier_failures;
3238 ret = insert_balance_item(fs_info->tree_root, bctl);
3239 if (ret && ret != -EEXIST)
3242 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3243 BUG_ON(ret == -EEXIST);
3244 set_balance_control(bctl);
3246 BUG_ON(ret != -EEXIST);
3247 spin_lock(&fs_info->balance_lock);
3248 update_balance_args(bctl);
3249 spin_unlock(&fs_info->balance_lock);
3252 atomic_inc(&fs_info->balance_running);
3253 mutex_unlock(&fs_info->balance_mutex);
3255 ret = __btrfs_balance(fs_info);
3257 mutex_lock(&fs_info->balance_mutex);
3258 atomic_dec(&fs_info->balance_running);
3260 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3261 fs_info->num_tolerated_disk_barrier_failures =
3262 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3266 memset(bargs, 0, sizeof(*bargs));
3267 update_ioctl_balance_args(fs_info, 0, bargs);
3270 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3271 balance_need_close(fs_info)) {
3272 __cancel_balance(fs_info);
3275 wake_up(&fs_info->balance_wait_q);
3279 if (bctl->flags & BTRFS_BALANCE_RESUME)
3280 __cancel_balance(fs_info);
3283 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3288 static int balance_kthread(void *data)
3290 struct btrfs_fs_info *fs_info = data;
3293 mutex_lock(&fs_info->volume_mutex);
3294 mutex_lock(&fs_info->balance_mutex);
3296 if (fs_info->balance_ctl) {
3297 printk(KERN_INFO "btrfs: continuing balance\n");
3298 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3301 mutex_unlock(&fs_info->balance_mutex);
3302 mutex_unlock(&fs_info->volume_mutex);
3307 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3309 struct task_struct *tsk;
3311 spin_lock(&fs_info->balance_lock);
3312 if (!fs_info->balance_ctl) {
3313 spin_unlock(&fs_info->balance_lock);
3316 spin_unlock(&fs_info->balance_lock);
3318 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3319 printk(KERN_INFO "btrfs: force skipping balance\n");
3323 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3324 return PTR_RET(tsk);
3327 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3329 struct btrfs_balance_control *bctl;
3330 struct btrfs_balance_item *item;
3331 struct btrfs_disk_balance_args disk_bargs;
3332 struct btrfs_path *path;
3333 struct extent_buffer *leaf;
3334 struct btrfs_key key;
3337 path = btrfs_alloc_path();
3341 key.objectid = BTRFS_BALANCE_OBJECTID;
3342 key.type = BTRFS_BALANCE_ITEM_KEY;
3345 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3348 if (ret > 0) { /* ret = -ENOENT; */
3353 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3359 leaf = path->nodes[0];
3360 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3362 bctl->fs_info = fs_info;
3363 bctl->flags = btrfs_balance_flags(leaf, item);
3364 bctl->flags |= BTRFS_BALANCE_RESUME;
3366 btrfs_balance_data(leaf, item, &disk_bargs);
3367 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3368 btrfs_balance_meta(leaf, item, &disk_bargs);
3369 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3370 btrfs_balance_sys(leaf, item, &disk_bargs);
3371 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3373 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3375 mutex_lock(&fs_info->volume_mutex);
3376 mutex_lock(&fs_info->balance_mutex);
3378 set_balance_control(bctl);
3380 mutex_unlock(&fs_info->balance_mutex);
3381 mutex_unlock(&fs_info->volume_mutex);
3383 btrfs_free_path(path);
3387 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3391 mutex_lock(&fs_info->balance_mutex);
3392 if (!fs_info->balance_ctl) {
3393 mutex_unlock(&fs_info->balance_mutex);
3397 if (atomic_read(&fs_info->balance_running)) {
3398 atomic_inc(&fs_info->balance_pause_req);
3399 mutex_unlock(&fs_info->balance_mutex);
3401 wait_event(fs_info->balance_wait_q,
3402 atomic_read(&fs_info->balance_running) == 0);
3404 mutex_lock(&fs_info->balance_mutex);
3405 /* we are good with balance_ctl ripped off from under us */
3406 BUG_ON(atomic_read(&fs_info->balance_running));
3407 atomic_dec(&fs_info->balance_pause_req);
3412 mutex_unlock(&fs_info->balance_mutex);
3416 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3418 mutex_lock(&fs_info->balance_mutex);
3419 if (!fs_info->balance_ctl) {
3420 mutex_unlock(&fs_info->balance_mutex);
3424 atomic_inc(&fs_info->balance_cancel_req);
3426 * if we are running just wait and return, balance item is
3427 * deleted in btrfs_balance in this case
3429 if (atomic_read(&fs_info->balance_running)) {
3430 mutex_unlock(&fs_info->balance_mutex);
3431 wait_event(fs_info->balance_wait_q,
3432 atomic_read(&fs_info->balance_running) == 0);
3433 mutex_lock(&fs_info->balance_mutex);
3435 /* __cancel_balance needs volume_mutex */
3436 mutex_unlock(&fs_info->balance_mutex);
3437 mutex_lock(&fs_info->volume_mutex);
3438 mutex_lock(&fs_info->balance_mutex);
3440 if (fs_info->balance_ctl)
3441 __cancel_balance(fs_info);
3443 mutex_unlock(&fs_info->volume_mutex);
3446 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3447 atomic_dec(&fs_info->balance_cancel_req);
3448 mutex_unlock(&fs_info->balance_mutex);
3452 static int btrfs_uuid_scan_kthread(void *data)
3454 struct btrfs_fs_info *fs_info = data;
3455 struct btrfs_root *root = fs_info->tree_root;
3456 struct btrfs_key key;
3457 struct btrfs_key max_key;
3458 struct btrfs_path *path = NULL;
3460 struct extent_buffer *eb;
3462 struct btrfs_root_item root_item;
3464 struct btrfs_trans_handle *trans;
3466 path = btrfs_alloc_path();
3473 key.type = BTRFS_ROOT_ITEM_KEY;
3476 max_key.objectid = (u64)-1;
3477 max_key.type = BTRFS_ROOT_ITEM_KEY;
3478 max_key.offset = (u64)-1;
3480 path->keep_locks = 1;
3483 ret = btrfs_search_forward(root, &key, &max_key, path, 0);
3490 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3491 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3492 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3493 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3496 eb = path->nodes[0];
3497 slot = path->slots[0];
3498 item_size = btrfs_item_size_nr(eb, slot);
3499 if (item_size < sizeof(root_item))
3503 read_extent_buffer(eb, &root_item,
3504 btrfs_item_ptr_offset(eb, slot),
3505 (int)sizeof(root_item));
3506 if (btrfs_root_refs(&root_item) == 0)
3508 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3510 * 1 - subvol uuid item
3511 * 1 - received_subvol uuid item
3513 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3514 if (IS_ERR(trans)) {
3515 ret = PTR_ERR(trans);
3518 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3520 BTRFS_UUID_KEY_SUBVOL,
3523 pr_warn("btrfs: uuid_tree_add failed %d\n",
3525 btrfs_end_transaction(trans,
3526 fs_info->uuid_root);
3531 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3533 /* 1 - received_subvol uuid item */
3534 trans = btrfs_start_transaction(
3535 fs_info->uuid_root, 1);
3536 if (IS_ERR(trans)) {
3537 ret = PTR_ERR(trans);
3541 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3542 root_item.received_uuid,
3543 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3546 pr_warn("btrfs: uuid_tree_add failed %d\n",
3548 btrfs_end_transaction(trans,
3549 fs_info->uuid_root);
3555 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3561 btrfs_release_path(path);
3562 if (key.offset < (u64)-1) {
3564 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3566 key.type = BTRFS_ROOT_ITEM_KEY;
3567 } else if (key.objectid < (u64)-1) {
3569 key.type = BTRFS_ROOT_ITEM_KEY;
3578 btrfs_free_path(path);
3580 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3582 fs_info->update_uuid_tree_gen = 1;
3583 up(&fs_info->uuid_tree_rescan_sem);
3588 * Callback for btrfs_uuid_tree_iterate().
3590 * 0 check succeeded, the entry is not outdated.
3591 * < 0 if an error occured.
3592 * > 0 if the check failed, which means the caller shall remove the entry.
3594 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3595 u8 *uuid, u8 type, u64 subid)
3597 struct btrfs_key key;
3599 struct btrfs_root *subvol_root;
3601 if (type != BTRFS_UUID_KEY_SUBVOL &&
3602 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3605 key.objectid = subid;
3606 key.type = BTRFS_ROOT_ITEM_KEY;
3607 key.offset = (u64)-1;
3608 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3609 if (IS_ERR(subvol_root)) {
3610 ret = PTR_ERR(subvol_root);
3617 case BTRFS_UUID_KEY_SUBVOL:
3618 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3621 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3622 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3632 static int btrfs_uuid_rescan_kthread(void *data)
3634 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3638 * 1st step is to iterate through the existing UUID tree and
3639 * to delete all entries that contain outdated data.
3640 * 2nd step is to add all missing entries to the UUID tree.
3642 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3644 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3645 up(&fs_info->uuid_tree_rescan_sem);
3648 return btrfs_uuid_scan_kthread(data);
3651 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3653 struct btrfs_trans_handle *trans;
3654 struct btrfs_root *tree_root = fs_info->tree_root;
3655 struct btrfs_root *uuid_root;
3656 struct task_struct *task;
3663 trans = btrfs_start_transaction(tree_root, 2);
3665 return PTR_ERR(trans);
3667 uuid_root = btrfs_create_tree(trans, fs_info,
3668 BTRFS_UUID_TREE_OBJECTID);
3669 if (IS_ERR(uuid_root)) {
3670 btrfs_abort_transaction(trans, tree_root,
3671 PTR_ERR(uuid_root));
3672 return PTR_ERR(uuid_root);
3675 fs_info->uuid_root = uuid_root;
3677 ret = btrfs_commit_transaction(trans, tree_root);
3681 down(&fs_info->uuid_tree_rescan_sem);
3682 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3684 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3685 pr_warn("btrfs: failed to start uuid_scan task\n");
3686 up(&fs_info->uuid_tree_rescan_sem);
3687 return PTR_ERR(task);
3693 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3695 struct task_struct *task;
3697 down(&fs_info->uuid_tree_rescan_sem);
3698 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3700 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3701 pr_warn("btrfs: failed to start uuid_rescan task\n");
3702 up(&fs_info->uuid_tree_rescan_sem);
3703 return PTR_ERR(task);
3710 * shrinking a device means finding all of the device extents past
3711 * the new size, and then following the back refs to the chunks.
3712 * The chunk relocation code actually frees the device extent
3714 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3716 struct btrfs_trans_handle *trans;
3717 struct btrfs_root *root = device->dev_root;
3718 struct btrfs_dev_extent *dev_extent = NULL;
3719 struct btrfs_path *path;
3727 bool retried = false;
3728 struct extent_buffer *l;
3729 struct btrfs_key key;
3730 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3731 u64 old_total = btrfs_super_total_bytes(super_copy);
3732 u64 old_size = device->total_bytes;
3733 u64 diff = device->total_bytes - new_size;
3735 if (device->is_tgtdev_for_dev_replace)
3738 path = btrfs_alloc_path();
3746 device->total_bytes = new_size;
3747 if (device->writeable) {
3748 device->fs_devices->total_rw_bytes -= diff;
3749 spin_lock(&root->fs_info->free_chunk_lock);
3750 root->fs_info->free_chunk_space -= diff;
3751 spin_unlock(&root->fs_info->free_chunk_lock);
3753 unlock_chunks(root);
3756 key.objectid = device->devid;
3757 key.offset = (u64)-1;
3758 key.type = BTRFS_DEV_EXTENT_KEY;
3761 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3765 ret = btrfs_previous_item(root, path, 0, key.type);
3770 btrfs_release_path(path);
3775 slot = path->slots[0];
3776 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3778 if (key.objectid != device->devid) {
3779 btrfs_release_path(path);
3783 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3784 length = btrfs_dev_extent_length(l, dev_extent);
3786 if (key.offset + length <= new_size) {
3787 btrfs_release_path(path);
3791 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3792 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3793 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3794 btrfs_release_path(path);
3796 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3798 if (ret && ret != -ENOSPC)
3802 } while (key.offset-- > 0);
3804 if (failed && !retried) {
3808 } else if (failed && retried) {
3812 device->total_bytes = old_size;
3813 if (device->writeable)
3814 device->fs_devices->total_rw_bytes += diff;
3815 spin_lock(&root->fs_info->free_chunk_lock);
3816 root->fs_info->free_chunk_space += diff;
3817 spin_unlock(&root->fs_info->free_chunk_lock);
3818 unlock_chunks(root);
3822 /* Shrinking succeeded, else we would be at "done". */
3823 trans = btrfs_start_transaction(root, 0);
3824 if (IS_ERR(trans)) {
3825 ret = PTR_ERR(trans);
3831 device->disk_total_bytes = new_size;
3832 /* Now btrfs_update_device() will change the on-disk size. */
3833 ret = btrfs_update_device(trans, device);
3835 unlock_chunks(root);
3836 btrfs_end_transaction(trans, root);
3839 WARN_ON(diff > old_total);
3840 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3841 unlock_chunks(root);
3842 btrfs_end_transaction(trans, root);
3844 btrfs_free_path(path);
3848 static int btrfs_add_system_chunk(struct btrfs_root *root,
3849 struct btrfs_key *key,
3850 struct btrfs_chunk *chunk, int item_size)
3852 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3853 struct btrfs_disk_key disk_key;
3857 array_size = btrfs_super_sys_array_size(super_copy);
3858 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3861 ptr = super_copy->sys_chunk_array + array_size;
3862 btrfs_cpu_key_to_disk(&disk_key, key);
3863 memcpy(ptr, &disk_key, sizeof(disk_key));
3864 ptr += sizeof(disk_key);
3865 memcpy(ptr, chunk, item_size);
3866 item_size += sizeof(disk_key);
3867 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3872 * sort the devices in descending order by max_avail, total_avail
3874 static int btrfs_cmp_device_info(const void *a, const void *b)
3876 const struct btrfs_device_info *di_a = a;
3877 const struct btrfs_device_info *di_b = b;
3879 if (di_a->max_avail > di_b->max_avail)
3881 if (di_a->max_avail < di_b->max_avail)
3883 if (di_a->total_avail > di_b->total_avail)
3885 if (di_a->total_avail < di_b->total_avail)
3890 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3891 [BTRFS_RAID_RAID10] = {
3894 .devs_max = 0, /* 0 == as many as possible */
3896 .devs_increment = 2,
3899 [BTRFS_RAID_RAID1] = {
3904 .devs_increment = 2,
3907 [BTRFS_RAID_DUP] = {
3912 .devs_increment = 1,
3915 [BTRFS_RAID_RAID0] = {
3920 .devs_increment = 1,
3923 [BTRFS_RAID_SINGLE] = {
3928 .devs_increment = 1,
3931 [BTRFS_RAID_RAID5] = {
3936 .devs_increment = 1,
3939 [BTRFS_RAID_RAID6] = {
3944 .devs_increment = 1,
3949 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3951 /* TODO allow them to set a preferred stripe size */
3955 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3957 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3960 btrfs_set_fs_incompat(info, RAID56);
3963 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3964 struct btrfs_root *extent_root, u64 start,
3967 struct btrfs_fs_info *info = extent_root->fs_info;
3968 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3969 struct list_head *cur;
3970 struct map_lookup *map = NULL;
3971 struct extent_map_tree *em_tree;
3972 struct extent_map *em;
3973 struct btrfs_device_info *devices_info = NULL;
3975 int num_stripes; /* total number of stripes to allocate */
3976 int data_stripes; /* number of stripes that count for
3978 int sub_stripes; /* sub_stripes info for map */
3979 int dev_stripes; /* stripes per dev */
3980 int devs_max; /* max devs to use */
3981 int devs_min; /* min devs needed */
3982 int devs_increment; /* ndevs has to be a multiple of this */
3983 int ncopies; /* how many copies to data has */
3985 u64 max_stripe_size;
3989 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3995 BUG_ON(!alloc_profile_is_valid(type, 0));
3997 if (list_empty(&fs_devices->alloc_list))
4000 index = __get_raid_index(type);
4002 sub_stripes = btrfs_raid_array[index].sub_stripes;
4003 dev_stripes = btrfs_raid_array[index].dev_stripes;
4004 devs_max = btrfs_raid_array[index].devs_max;
4005 devs_min = btrfs_raid_array[index].devs_min;
4006 devs_increment = btrfs_raid_array[index].devs_increment;
4007 ncopies = btrfs_raid_array[index].ncopies;
4009 if (type & BTRFS_BLOCK_GROUP_DATA) {
4010 max_stripe_size = 1024 * 1024 * 1024;
4011 max_chunk_size = 10 * max_stripe_size;
4012 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4013 /* for larger filesystems, use larger metadata chunks */
4014 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4015 max_stripe_size = 1024 * 1024 * 1024;
4017 max_stripe_size = 256 * 1024 * 1024;
4018 max_chunk_size = max_stripe_size;
4019 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4020 max_stripe_size = 32 * 1024 * 1024;
4021 max_chunk_size = 2 * max_stripe_size;
4023 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4028 /* we don't want a chunk larger than 10% of writeable space */
4029 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4032 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4037 cur = fs_devices->alloc_list.next;
4040 * in the first pass through the devices list, we gather information
4041 * about the available holes on each device.
4044 while (cur != &fs_devices->alloc_list) {
4045 struct btrfs_device *device;
4049 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4053 if (!device->writeable) {
4055 "btrfs: read-only device in alloc_list\n");
4059 if (!device->in_fs_metadata ||
4060 device->is_tgtdev_for_dev_replace)
4063 if (device->total_bytes > device->bytes_used)
4064 total_avail = device->total_bytes - device->bytes_used;
4068 /* If there is no space on this device, skip it. */
4069 if (total_avail == 0)
4072 ret = find_free_dev_extent(trans, device,
4073 max_stripe_size * dev_stripes,
4074 &dev_offset, &max_avail);
4075 if (ret && ret != -ENOSPC)
4079 max_avail = max_stripe_size * dev_stripes;
4081 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4084 if (ndevs == fs_devices->rw_devices) {
4085 WARN(1, "%s: found more than %llu devices\n",
4086 __func__, fs_devices->rw_devices);
4089 devices_info[ndevs].dev_offset = dev_offset;
4090 devices_info[ndevs].max_avail = max_avail;
4091 devices_info[ndevs].total_avail = total_avail;
4092 devices_info[ndevs].dev = device;
4097 * now sort the devices by hole size / available space
4099 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4100 btrfs_cmp_device_info, NULL);
4102 /* round down to number of usable stripes */
4103 ndevs -= ndevs % devs_increment;
4105 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4110 if (devs_max && ndevs > devs_max)
4113 * the primary goal is to maximize the number of stripes, so use as many
4114 * devices as possible, even if the stripes are not maximum sized.
4116 stripe_size = devices_info[ndevs-1].max_avail;
4117 num_stripes = ndevs * dev_stripes;
4120 * this will have to be fixed for RAID1 and RAID10 over
4123 data_stripes = num_stripes / ncopies;
4125 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4126 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4127 btrfs_super_stripesize(info->super_copy));
4128 data_stripes = num_stripes - 1;
4130 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4131 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4132 btrfs_super_stripesize(info->super_copy));
4133 data_stripes = num_stripes - 2;
4137 * Use the number of data stripes to figure out how big this chunk
4138 * is really going to be in terms of logical address space,
4139 * and compare that answer with the max chunk size
4141 if (stripe_size * data_stripes > max_chunk_size) {
4142 u64 mask = (1ULL << 24) - 1;
4143 stripe_size = max_chunk_size;
4144 do_div(stripe_size, data_stripes);
4146 /* bump the answer up to a 16MB boundary */
4147 stripe_size = (stripe_size + mask) & ~mask;
4149 /* but don't go higher than the limits we found
4150 * while searching for free extents
4152 if (stripe_size > devices_info[ndevs-1].max_avail)
4153 stripe_size = devices_info[ndevs-1].max_avail;
4156 do_div(stripe_size, dev_stripes);
4158 /* align to BTRFS_STRIPE_LEN */
4159 do_div(stripe_size, raid_stripe_len);
4160 stripe_size *= raid_stripe_len;
4162 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4167 map->num_stripes = num_stripes;
4169 for (i = 0; i < ndevs; ++i) {
4170 for (j = 0; j < dev_stripes; ++j) {
4171 int s = i * dev_stripes + j;
4172 map->stripes[s].dev = devices_info[i].dev;
4173 map->stripes[s].physical = devices_info[i].dev_offset +
4177 map->sector_size = extent_root->sectorsize;
4178 map->stripe_len = raid_stripe_len;
4179 map->io_align = raid_stripe_len;
4180 map->io_width = raid_stripe_len;
4182 map->sub_stripes = sub_stripes;
4184 num_bytes = stripe_size * data_stripes;
4186 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4188 em = alloc_extent_map();
4193 em->bdev = (struct block_device *)map;
4195 em->len = num_bytes;
4196 em->block_start = 0;
4197 em->block_len = em->len;
4198 em->orig_block_len = stripe_size;
4200 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4201 write_lock(&em_tree->lock);
4202 ret = add_extent_mapping(em_tree, em, 0);
4204 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4205 atomic_inc(&em->refs);
4207 write_unlock(&em_tree->lock);
4209 free_extent_map(em);
4213 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4214 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4217 goto error_del_extent;
4219 free_extent_map(em);
4220 check_raid56_incompat_flag(extent_root->fs_info, type);
4222 kfree(devices_info);
4226 write_lock(&em_tree->lock);
4227 remove_extent_mapping(em_tree, em);
4228 write_unlock(&em_tree->lock);
4230 /* One for our allocation */
4231 free_extent_map(em);
4232 /* One for the tree reference */
4233 free_extent_map(em);
4236 kfree(devices_info);
4240 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4241 struct btrfs_root *extent_root,
4242 u64 chunk_offset, u64 chunk_size)
4244 struct btrfs_key key;
4245 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4246 struct btrfs_device *device;
4247 struct btrfs_chunk *chunk;
4248 struct btrfs_stripe *stripe;
4249 struct extent_map_tree *em_tree;
4250 struct extent_map *em;
4251 struct map_lookup *map;
4258 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4259 read_lock(&em_tree->lock);
4260 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4261 read_unlock(&em_tree->lock);
4264 btrfs_crit(extent_root->fs_info, "unable to find logical "
4265 "%Lu len %Lu", chunk_offset, chunk_size);
4269 if (em->start != chunk_offset || em->len != chunk_size) {
4270 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4271 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4272 chunk_size, em->start, em->len);
4273 free_extent_map(em);
4277 map = (struct map_lookup *)em->bdev;
4278 item_size = btrfs_chunk_item_size(map->num_stripes);
4279 stripe_size = em->orig_block_len;
4281 chunk = kzalloc(item_size, GFP_NOFS);
4287 for (i = 0; i < map->num_stripes; i++) {
4288 device = map->stripes[i].dev;
4289 dev_offset = map->stripes[i].physical;
4291 device->bytes_used += stripe_size;
4292 ret = btrfs_update_device(trans, device);
4295 ret = btrfs_alloc_dev_extent(trans, device,
4296 chunk_root->root_key.objectid,
4297 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4298 chunk_offset, dev_offset,
4304 spin_lock(&extent_root->fs_info->free_chunk_lock);
4305 extent_root->fs_info->free_chunk_space -= (stripe_size *
4307 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4309 stripe = &chunk->stripe;
4310 for (i = 0; i < map->num_stripes; i++) {
4311 device = map->stripes[i].dev;
4312 dev_offset = map->stripes[i].physical;
4314 btrfs_set_stack_stripe_devid(stripe, device->devid);
4315 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4316 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4320 btrfs_set_stack_chunk_length(chunk, chunk_size);
4321 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4322 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4323 btrfs_set_stack_chunk_type(chunk, map->type);
4324 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4325 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4326 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4327 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4328 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4330 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4331 key.type = BTRFS_CHUNK_ITEM_KEY;
4332 key.offset = chunk_offset;
4334 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4335 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4337 * TODO: Cleanup of inserted chunk root in case of
4340 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4346 free_extent_map(em);
4351 * Chunk allocation falls into two parts. The first part does works
4352 * that make the new allocated chunk useable, but not do any operation
4353 * that modifies the chunk tree. The second part does the works that
4354 * require modifying the chunk tree. This division is important for the
4355 * bootstrap process of adding storage to a seed btrfs.
4357 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4358 struct btrfs_root *extent_root, u64 type)
4362 chunk_offset = find_next_chunk(extent_root->fs_info);
4363 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4366 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4367 struct btrfs_root *root,
4368 struct btrfs_device *device)
4371 u64 sys_chunk_offset;
4373 struct btrfs_fs_info *fs_info = root->fs_info;
4374 struct btrfs_root *extent_root = fs_info->extent_root;
4377 chunk_offset = find_next_chunk(fs_info);
4378 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4379 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4384 sys_chunk_offset = find_next_chunk(root->fs_info);
4385 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4386 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4389 btrfs_abort_transaction(trans, root, ret);
4393 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4395 btrfs_abort_transaction(trans, root, ret);
4400 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4402 struct extent_map *em;
4403 struct map_lookup *map;
4404 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4408 read_lock(&map_tree->map_tree.lock);
4409 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4410 read_unlock(&map_tree->map_tree.lock);
4414 if (btrfs_test_opt(root, DEGRADED)) {
4415 free_extent_map(em);
4419 map = (struct map_lookup *)em->bdev;
4420 for (i = 0; i < map->num_stripes; i++) {
4421 if (!map->stripes[i].dev->writeable) {
4426 free_extent_map(em);
4430 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4432 extent_map_tree_init(&tree->map_tree);
4435 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4437 struct extent_map *em;
4440 write_lock(&tree->map_tree.lock);
4441 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4443 remove_extent_mapping(&tree->map_tree, em);
4444 write_unlock(&tree->map_tree.lock);
4449 free_extent_map(em);
4450 /* once for the tree */
4451 free_extent_map(em);
4455 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4457 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4458 struct extent_map *em;
4459 struct map_lookup *map;
4460 struct extent_map_tree *em_tree = &map_tree->map_tree;
4463 read_lock(&em_tree->lock);
4464 em = lookup_extent_mapping(em_tree, logical, len);
4465 read_unlock(&em_tree->lock);
4468 * We could return errors for these cases, but that could get ugly and
4469 * we'd probably do the same thing which is just not do anything else
4470 * and exit, so return 1 so the callers don't try to use other copies.
4473 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4478 if (em->start > logical || em->start + em->len < logical) {
4479 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4480 "%Lu-%Lu\n", logical, logical+len, em->start,
4481 em->start + em->len);
4485 map = (struct map_lookup *)em->bdev;
4486 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4487 ret = map->num_stripes;
4488 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4489 ret = map->sub_stripes;
4490 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4492 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4496 free_extent_map(em);
4498 btrfs_dev_replace_lock(&fs_info->dev_replace);
4499 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4501 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4506 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4507 struct btrfs_mapping_tree *map_tree,
4510 struct extent_map *em;
4511 struct map_lookup *map;
4512 struct extent_map_tree *em_tree = &map_tree->map_tree;
4513 unsigned long len = root->sectorsize;
4515 read_lock(&em_tree->lock);
4516 em = lookup_extent_mapping(em_tree, logical, len);
4517 read_unlock(&em_tree->lock);
4520 BUG_ON(em->start > logical || em->start + em->len < logical);
4521 map = (struct map_lookup *)em->bdev;
4522 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4523 BTRFS_BLOCK_GROUP_RAID6)) {
4524 len = map->stripe_len * nr_data_stripes(map);
4526 free_extent_map(em);
4530 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4531 u64 logical, u64 len, int mirror_num)
4533 struct extent_map *em;
4534 struct map_lookup *map;
4535 struct extent_map_tree *em_tree = &map_tree->map_tree;
4538 read_lock(&em_tree->lock);
4539 em = lookup_extent_mapping(em_tree, logical, len);
4540 read_unlock(&em_tree->lock);
4543 BUG_ON(em->start > logical || em->start + em->len < logical);
4544 map = (struct map_lookup *)em->bdev;
4545 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4546 BTRFS_BLOCK_GROUP_RAID6))
4548 free_extent_map(em);
4552 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4553 struct map_lookup *map, int first, int num,
4554 int optimal, int dev_replace_is_ongoing)
4558 struct btrfs_device *srcdev;
4560 if (dev_replace_is_ongoing &&
4561 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4562 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4563 srcdev = fs_info->dev_replace.srcdev;
4568 * try to avoid the drive that is the source drive for a
4569 * dev-replace procedure, only choose it if no other non-missing
4570 * mirror is available
4572 for (tolerance = 0; tolerance < 2; tolerance++) {
4573 if (map->stripes[optimal].dev->bdev &&
4574 (tolerance || map->stripes[optimal].dev != srcdev))
4576 for (i = first; i < first + num; i++) {
4577 if (map->stripes[i].dev->bdev &&
4578 (tolerance || map->stripes[i].dev != srcdev))
4583 /* we couldn't find one that doesn't fail. Just return something
4584 * and the io error handling code will clean up eventually
4589 static inline int parity_smaller(u64 a, u64 b)
4594 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4595 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4597 struct btrfs_bio_stripe s;
4604 for (i = 0; i < bbio->num_stripes - 1; i++) {
4605 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4606 s = bbio->stripes[i];
4608 bbio->stripes[i] = bbio->stripes[i+1];
4609 raid_map[i] = raid_map[i+1];
4610 bbio->stripes[i+1] = s;
4618 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4619 u64 logical, u64 *length,
4620 struct btrfs_bio **bbio_ret,
4621 int mirror_num, u64 **raid_map_ret)
4623 struct extent_map *em;
4624 struct map_lookup *map;
4625 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4626 struct extent_map_tree *em_tree = &map_tree->map_tree;
4629 u64 stripe_end_offset;
4634 u64 *raid_map = NULL;
4640 struct btrfs_bio *bbio = NULL;
4641 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4642 int dev_replace_is_ongoing = 0;
4643 int num_alloc_stripes;
4644 int patch_the_first_stripe_for_dev_replace = 0;
4645 u64 physical_to_patch_in_first_stripe = 0;
4646 u64 raid56_full_stripe_start = (u64)-1;
4648 read_lock(&em_tree->lock);
4649 em = lookup_extent_mapping(em_tree, logical, *length);
4650 read_unlock(&em_tree->lock);
4653 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4658 if (em->start > logical || em->start + em->len < logical) {
4659 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4660 "found %Lu-%Lu\n", logical, em->start,
4661 em->start + em->len);
4665 map = (struct map_lookup *)em->bdev;
4666 offset = logical - em->start;
4668 stripe_len = map->stripe_len;
4671 * stripe_nr counts the total number of stripes we have to stride
4672 * to get to this block
4674 do_div(stripe_nr, stripe_len);
4676 stripe_offset = stripe_nr * stripe_len;
4677 BUG_ON(offset < stripe_offset);
4679 /* stripe_offset is the offset of this block in its stripe*/
4680 stripe_offset = offset - stripe_offset;
4682 /* if we're here for raid56, we need to know the stripe aligned start */
4683 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4684 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4685 raid56_full_stripe_start = offset;
4687 /* allow a write of a full stripe, but make sure we don't
4688 * allow straddling of stripes
4690 do_div(raid56_full_stripe_start, full_stripe_len);
4691 raid56_full_stripe_start *= full_stripe_len;
4694 if (rw & REQ_DISCARD) {
4695 /* we don't discard raid56 yet */
4697 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4701 *length = min_t(u64, em->len - offset, *length);
4702 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4704 /* For writes to RAID[56], allow a full stripeset across all disks.
4705 For other RAID types and for RAID[56] reads, just allow a single
4706 stripe (on a single disk). */
4707 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4709 max_len = stripe_len * nr_data_stripes(map) -
4710 (offset - raid56_full_stripe_start);
4712 /* we limit the length of each bio to what fits in a stripe */
4713 max_len = stripe_len - stripe_offset;
4715 *length = min_t(u64, em->len - offset, max_len);
4717 *length = em->len - offset;
4720 /* This is for when we're called from btrfs_merge_bio_hook() and all
4721 it cares about is the length */
4725 btrfs_dev_replace_lock(dev_replace);
4726 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4727 if (!dev_replace_is_ongoing)
4728 btrfs_dev_replace_unlock(dev_replace);
4730 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4731 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4732 dev_replace->tgtdev != NULL) {
4734 * in dev-replace case, for repair case (that's the only
4735 * case where the mirror is selected explicitly when
4736 * calling btrfs_map_block), blocks left of the left cursor
4737 * can also be read from the target drive.
4738 * For REQ_GET_READ_MIRRORS, the target drive is added as
4739 * the last one to the array of stripes. For READ, it also
4740 * needs to be supported using the same mirror number.
4741 * If the requested block is not left of the left cursor,
4742 * EIO is returned. This can happen because btrfs_num_copies()
4743 * returns one more in the dev-replace case.
4745 u64 tmp_length = *length;
4746 struct btrfs_bio *tmp_bbio = NULL;
4747 int tmp_num_stripes;
4748 u64 srcdev_devid = dev_replace->srcdev->devid;
4749 int index_srcdev = 0;
4751 u64 physical_of_found = 0;
4753 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4754 logical, &tmp_length, &tmp_bbio, 0, NULL);
4756 WARN_ON(tmp_bbio != NULL);
4760 tmp_num_stripes = tmp_bbio->num_stripes;
4761 if (mirror_num > tmp_num_stripes) {
4763 * REQ_GET_READ_MIRRORS does not contain this
4764 * mirror, that means that the requested area
4765 * is not left of the left cursor
4773 * process the rest of the function using the mirror_num
4774 * of the source drive. Therefore look it up first.
4775 * At the end, patch the device pointer to the one of the
4778 for (i = 0; i < tmp_num_stripes; i++) {
4779 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4781 * In case of DUP, in order to keep it
4782 * simple, only add the mirror with the
4783 * lowest physical address
4786 physical_of_found <=
4787 tmp_bbio->stripes[i].physical)
4792 tmp_bbio->stripes[i].physical;
4797 mirror_num = index_srcdev + 1;
4798 patch_the_first_stripe_for_dev_replace = 1;
4799 physical_to_patch_in_first_stripe = physical_of_found;
4808 } else if (mirror_num > map->num_stripes) {
4814 stripe_nr_orig = stripe_nr;
4815 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4816 do_div(stripe_nr_end, map->stripe_len);
4817 stripe_end_offset = stripe_nr_end * map->stripe_len -
4820 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4821 if (rw & REQ_DISCARD)
4822 num_stripes = min_t(u64, map->num_stripes,
4823 stripe_nr_end - stripe_nr_orig);
4824 stripe_index = do_div(stripe_nr, map->num_stripes);
4825 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4826 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4827 num_stripes = map->num_stripes;
4828 else if (mirror_num)
4829 stripe_index = mirror_num - 1;
4831 stripe_index = find_live_mirror(fs_info, map, 0,
4833 current->pid % map->num_stripes,
4834 dev_replace_is_ongoing);
4835 mirror_num = stripe_index + 1;
4838 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4839 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4840 num_stripes = map->num_stripes;
4841 } else if (mirror_num) {
4842 stripe_index = mirror_num - 1;
4847 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4848 int factor = map->num_stripes / map->sub_stripes;
4850 stripe_index = do_div(stripe_nr, factor);
4851 stripe_index *= map->sub_stripes;
4853 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4854 num_stripes = map->sub_stripes;
4855 else if (rw & REQ_DISCARD)
4856 num_stripes = min_t(u64, map->sub_stripes *
4857 (stripe_nr_end - stripe_nr_orig),
4859 else if (mirror_num)
4860 stripe_index += mirror_num - 1;
4862 int old_stripe_index = stripe_index;
4863 stripe_index = find_live_mirror(fs_info, map,
4865 map->sub_stripes, stripe_index +
4866 current->pid % map->sub_stripes,
4867 dev_replace_is_ongoing);
4868 mirror_num = stripe_index - old_stripe_index + 1;
4871 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4872 BTRFS_BLOCK_GROUP_RAID6)) {
4875 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4879 /* push stripe_nr back to the start of the full stripe */
4880 stripe_nr = raid56_full_stripe_start;
4881 do_div(stripe_nr, stripe_len);
4883 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4885 /* RAID[56] write or recovery. Return all stripes */
4886 num_stripes = map->num_stripes;
4887 max_errors = nr_parity_stripes(map);
4889 raid_map = kmalloc(sizeof(u64) * num_stripes,
4896 /* Work out the disk rotation on this stripe-set */
4898 rot = do_div(tmp, num_stripes);
4900 /* Fill in the logical address of each stripe */
4901 tmp = stripe_nr * nr_data_stripes(map);
4902 for (i = 0; i < nr_data_stripes(map); i++)
4903 raid_map[(i+rot) % num_stripes] =
4904 em->start + (tmp + i) * map->stripe_len;
4906 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4907 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4908 raid_map[(i+rot+1) % num_stripes] =
4911 *length = map->stripe_len;
4916 * Mirror #0 or #1 means the original data block.
4917 * Mirror #2 is RAID5 parity block.
4918 * Mirror #3 is RAID6 Q block.
4920 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4922 stripe_index = nr_data_stripes(map) +
4925 /* We distribute the parity blocks across stripes */
4926 tmp = stripe_nr + stripe_index;
4927 stripe_index = do_div(tmp, map->num_stripes);
4931 * after this do_div call, stripe_nr is the number of stripes
4932 * on this device we have to walk to find the data, and
4933 * stripe_index is the number of our device in the stripe array
4935 stripe_index = do_div(stripe_nr, map->num_stripes);
4936 mirror_num = stripe_index + 1;
4938 BUG_ON(stripe_index >= map->num_stripes);
4940 num_alloc_stripes = num_stripes;
4941 if (dev_replace_is_ongoing) {
4942 if (rw & (REQ_WRITE | REQ_DISCARD))
4943 num_alloc_stripes <<= 1;
4944 if (rw & REQ_GET_READ_MIRRORS)
4945 num_alloc_stripes++;
4947 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4953 atomic_set(&bbio->error, 0);
4955 if (rw & REQ_DISCARD) {
4957 int sub_stripes = 0;
4958 u64 stripes_per_dev = 0;
4959 u32 remaining_stripes = 0;
4960 u32 last_stripe = 0;
4963 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4964 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4967 sub_stripes = map->sub_stripes;
4969 factor = map->num_stripes / sub_stripes;
4970 stripes_per_dev = div_u64_rem(stripe_nr_end -
4973 &remaining_stripes);
4974 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4975 last_stripe *= sub_stripes;
4978 for (i = 0; i < num_stripes; i++) {
4979 bbio->stripes[i].physical =
4980 map->stripes[stripe_index].physical +
4981 stripe_offset + stripe_nr * map->stripe_len;
4982 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4984 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4985 BTRFS_BLOCK_GROUP_RAID10)) {
4986 bbio->stripes[i].length = stripes_per_dev *
4989 if (i / sub_stripes < remaining_stripes)
4990 bbio->stripes[i].length +=
4994 * Special for the first stripe and
4997 * |-------|...|-------|
5001 if (i < sub_stripes)
5002 bbio->stripes[i].length -=
5005 if (stripe_index >= last_stripe &&
5006 stripe_index <= (last_stripe +
5008 bbio->stripes[i].length -=
5011 if (i == sub_stripes - 1)
5014 bbio->stripes[i].length = *length;
5017 if (stripe_index == map->num_stripes) {
5018 /* This could only happen for RAID0/10 */
5024 for (i = 0; i < num_stripes; i++) {
5025 bbio->stripes[i].physical =
5026 map->stripes[stripe_index].physical +
5028 stripe_nr * map->stripe_len;
5029 bbio->stripes[i].dev =
5030 map->stripes[stripe_index].dev;
5035 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5036 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5037 BTRFS_BLOCK_GROUP_RAID10 |
5038 BTRFS_BLOCK_GROUP_RAID5 |
5039 BTRFS_BLOCK_GROUP_DUP)) {
5041 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5046 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5047 dev_replace->tgtdev != NULL) {
5048 int index_where_to_add;
5049 u64 srcdev_devid = dev_replace->srcdev->devid;
5052 * duplicate the write operations while the dev replace
5053 * procedure is running. Since the copying of the old disk
5054 * to the new disk takes place at run time while the
5055 * filesystem is mounted writable, the regular write
5056 * operations to the old disk have to be duplicated to go
5057 * to the new disk as well.
5058 * Note that device->missing is handled by the caller, and
5059 * that the write to the old disk is already set up in the
5062 index_where_to_add = num_stripes;
5063 for (i = 0; i < num_stripes; i++) {
5064 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5065 /* write to new disk, too */
5066 struct btrfs_bio_stripe *new =
5067 bbio->stripes + index_where_to_add;
5068 struct btrfs_bio_stripe *old =
5071 new->physical = old->physical;
5072 new->length = old->length;
5073 new->dev = dev_replace->tgtdev;
5074 index_where_to_add++;
5078 num_stripes = index_where_to_add;
5079 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5080 dev_replace->tgtdev != NULL) {
5081 u64 srcdev_devid = dev_replace->srcdev->devid;
5082 int index_srcdev = 0;
5084 u64 physical_of_found = 0;
5087 * During the dev-replace procedure, the target drive can
5088 * also be used to read data in case it is needed to repair
5089 * a corrupt block elsewhere. This is possible if the
5090 * requested area is left of the left cursor. In this area,
5091 * the target drive is a full copy of the source drive.
5093 for (i = 0; i < num_stripes; i++) {
5094 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5096 * In case of DUP, in order to keep it
5097 * simple, only add the mirror with the
5098 * lowest physical address
5101 physical_of_found <=
5102 bbio->stripes[i].physical)
5106 physical_of_found = bbio->stripes[i].physical;
5110 u64 length = map->stripe_len;
5112 if (physical_of_found + length <=
5113 dev_replace->cursor_left) {
5114 struct btrfs_bio_stripe *tgtdev_stripe =
5115 bbio->stripes + num_stripes;
5117 tgtdev_stripe->physical = physical_of_found;
5118 tgtdev_stripe->length =
5119 bbio->stripes[index_srcdev].length;
5120 tgtdev_stripe->dev = dev_replace->tgtdev;
5128 bbio->num_stripes = num_stripes;
5129 bbio->max_errors = max_errors;
5130 bbio->mirror_num = mirror_num;
5133 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5134 * mirror_num == num_stripes + 1 && dev_replace target drive is
5135 * available as a mirror
5137 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5138 WARN_ON(num_stripes > 1);
5139 bbio->stripes[0].dev = dev_replace->tgtdev;
5140 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5141 bbio->mirror_num = map->num_stripes + 1;
5144 sort_parity_stripes(bbio, raid_map);
5145 *raid_map_ret = raid_map;
5148 if (dev_replace_is_ongoing)
5149 btrfs_dev_replace_unlock(dev_replace);
5150 free_extent_map(em);
5154 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5155 u64 logical, u64 *length,
5156 struct btrfs_bio **bbio_ret, int mirror_num)
5158 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5162 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5163 u64 chunk_start, u64 physical, u64 devid,
5164 u64 **logical, int *naddrs, int *stripe_len)
5166 struct extent_map_tree *em_tree = &map_tree->map_tree;
5167 struct extent_map *em;
5168 struct map_lookup *map;
5176 read_lock(&em_tree->lock);
5177 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5178 read_unlock(&em_tree->lock);
5181 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5186 if (em->start != chunk_start) {
5187 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5188 em->start, chunk_start);
5189 free_extent_map(em);
5192 map = (struct map_lookup *)em->bdev;
5195 rmap_len = map->stripe_len;
5197 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5198 do_div(length, map->num_stripes / map->sub_stripes);
5199 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5200 do_div(length, map->num_stripes);
5201 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5202 BTRFS_BLOCK_GROUP_RAID6)) {
5203 do_div(length, nr_data_stripes(map));
5204 rmap_len = map->stripe_len * nr_data_stripes(map);
5207 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5208 BUG_ON(!buf); /* -ENOMEM */
5210 for (i = 0; i < map->num_stripes; i++) {
5211 if (devid && map->stripes[i].dev->devid != devid)
5213 if (map->stripes[i].physical > physical ||
5214 map->stripes[i].physical + length <= physical)
5217 stripe_nr = physical - map->stripes[i].physical;
5218 do_div(stripe_nr, map->stripe_len);
5220 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5221 stripe_nr = stripe_nr * map->num_stripes + i;
5222 do_div(stripe_nr, map->sub_stripes);
5223 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5224 stripe_nr = stripe_nr * map->num_stripes + i;
5225 } /* else if RAID[56], multiply by nr_data_stripes().
5226 * Alternatively, just use rmap_len below instead of
5227 * map->stripe_len */
5229 bytenr = chunk_start + stripe_nr * rmap_len;
5230 WARN_ON(nr >= map->num_stripes);
5231 for (j = 0; j < nr; j++) {
5232 if (buf[j] == bytenr)
5236 WARN_ON(nr >= map->num_stripes);
5243 *stripe_len = rmap_len;
5245 free_extent_map(em);
5249 static void btrfs_end_bio(struct bio *bio, int err)
5251 struct btrfs_bio *bbio = bio->bi_private;
5252 int is_orig_bio = 0;
5255 atomic_inc(&bbio->error);
5256 if (err == -EIO || err == -EREMOTEIO) {
5257 unsigned int stripe_index =
5258 btrfs_io_bio(bio)->stripe_index;
5259 struct btrfs_device *dev;
5261 BUG_ON(stripe_index >= bbio->num_stripes);
5262 dev = bbio->stripes[stripe_index].dev;
5264 if (bio->bi_rw & WRITE)
5265 btrfs_dev_stat_inc(dev,
5266 BTRFS_DEV_STAT_WRITE_ERRS);
5268 btrfs_dev_stat_inc(dev,
5269 BTRFS_DEV_STAT_READ_ERRS);
5270 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5271 btrfs_dev_stat_inc(dev,
5272 BTRFS_DEV_STAT_FLUSH_ERRS);
5273 btrfs_dev_stat_print_on_error(dev);
5278 if (bio == bbio->orig_bio)
5281 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5284 bio = bbio->orig_bio;
5286 bio->bi_private = bbio->private;
5287 bio->bi_end_io = bbio->end_io;
5288 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5289 /* only send an error to the higher layers if it is
5290 * beyond the tolerance of the btrfs bio
5292 if (atomic_read(&bbio->error) > bbio->max_errors) {
5296 * this bio is actually up to date, we didn't
5297 * go over the max number of errors
5299 set_bit(BIO_UPTODATE, &bio->bi_flags);
5304 bio_endio(bio, err);
5305 } else if (!is_orig_bio) {
5310 struct async_sched {
5313 struct btrfs_fs_info *info;
5314 struct btrfs_work work;
5318 * see run_scheduled_bios for a description of why bios are collected for
5321 * This will add one bio to the pending list for a device and make sure
5322 * the work struct is scheduled.
5324 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5325 struct btrfs_device *device,
5326 int rw, struct bio *bio)
5328 int should_queue = 1;
5329 struct btrfs_pending_bios *pending_bios;
5331 if (device->missing || !device->bdev) {
5332 bio_endio(bio, -EIO);
5336 /* don't bother with additional async steps for reads, right now */
5337 if (!(rw & REQ_WRITE)) {
5339 btrfsic_submit_bio(rw, bio);
5345 * nr_async_bios allows us to reliably return congestion to the
5346 * higher layers. Otherwise, the async bio makes it appear we have
5347 * made progress against dirty pages when we've really just put it
5348 * on a queue for later
5350 atomic_inc(&root->fs_info->nr_async_bios);
5351 WARN_ON(bio->bi_next);
5352 bio->bi_next = NULL;
5355 spin_lock(&device->io_lock);
5356 if (bio->bi_rw & REQ_SYNC)
5357 pending_bios = &device->pending_sync_bios;
5359 pending_bios = &device->pending_bios;
5361 if (pending_bios->tail)
5362 pending_bios->tail->bi_next = bio;
5364 pending_bios->tail = bio;
5365 if (!pending_bios->head)
5366 pending_bios->head = bio;
5367 if (device->running_pending)
5370 spin_unlock(&device->io_lock);
5373 btrfs_queue_worker(&root->fs_info->submit_workers,
5377 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5380 struct bio_vec *prev;
5381 struct request_queue *q = bdev_get_queue(bdev);
5382 unsigned short max_sectors = queue_max_sectors(q);
5383 struct bvec_merge_data bvm = {
5385 .bi_sector = sector,
5386 .bi_rw = bio->bi_rw,
5389 if (bio->bi_vcnt == 0) {
5394 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5395 if (bio_sectors(bio) > max_sectors)
5398 if (!q->merge_bvec_fn)
5401 bvm.bi_size = bio->bi_size - prev->bv_len;
5402 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5407 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5408 struct bio *bio, u64 physical, int dev_nr,
5411 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5413 bio->bi_private = bbio;
5414 btrfs_io_bio(bio)->stripe_index = dev_nr;
5415 bio->bi_end_io = btrfs_end_bio;
5416 bio->bi_sector = physical >> 9;
5419 struct rcu_string *name;
5422 name = rcu_dereference(dev->name);
5423 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5424 "(%s id %llu), size=%u\n", rw,
5425 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5426 name->str, dev->devid, bio->bi_size);
5430 bio->bi_bdev = dev->bdev;
5432 btrfs_schedule_bio(root, dev, rw, bio);
5434 btrfsic_submit_bio(rw, bio);
5437 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5438 struct bio *first_bio, struct btrfs_device *dev,
5439 int dev_nr, int rw, int async)
5441 struct bio_vec *bvec = first_bio->bi_io_vec;
5443 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5444 u64 physical = bbio->stripes[dev_nr].physical;
5447 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5451 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5452 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5453 bvec->bv_offset) < bvec->bv_len) {
5454 u64 len = bio->bi_size;
5456 atomic_inc(&bbio->stripes_pending);
5457 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5465 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5469 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5471 atomic_inc(&bbio->error);
5472 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5473 bio->bi_private = bbio->private;
5474 bio->bi_end_io = bbio->end_io;
5475 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5476 bio->bi_sector = logical >> 9;
5478 bio_endio(bio, -EIO);
5482 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5483 int mirror_num, int async_submit)
5485 struct btrfs_device *dev;
5486 struct bio *first_bio = bio;
5487 u64 logical = (u64)bio->bi_sector << 9;
5490 u64 *raid_map = NULL;
5494 struct btrfs_bio *bbio = NULL;
5496 length = bio->bi_size;
5497 map_length = length;
5499 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5500 mirror_num, &raid_map);
5501 if (ret) /* -ENOMEM */
5504 total_devs = bbio->num_stripes;
5505 bbio->orig_bio = first_bio;
5506 bbio->private = first_bio->bi_private;
5507 bbio->end_io = first_bio->bi_end_io;
5508 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5511 /* In this case, map_length has been set to the length of
5512 a single stripe; not the whole write */
5514 return raid56_parity_write(root, bio, bbio,
5515 raid_map, map_length);
5517 return raid56_parity_recover(root, bio, bbio,
5518 raid_map, map_length,
5523 if (map_length < length) {
5524 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5525 logical, length, map_length);
5529 while (dev_nr < total_devs) {
5530 dev = bbio->stripes[dev_nr].dev;
5531 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5532 bbio_error(bbio, first_bio, logical);
5538 * Check and see if we're ok with this bio based on it's size
5539 * and offset with the given device.
5541 if (!bio_size_ok(dev->bdev, first_bio,
5542 bbio->stripes[dev_nr].physical >> 9)) {
5543 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5544 dev_nr, rw, async_submit);
5550 if (dev_nr < total_devs - 1) {
5551 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5552 BUG_ON(!bio); /* -ENOMEM */
5557 submit_stripe_bio(root, bbio, bio,
5558 bbio->stripes[dev_nr].physical, dev_nr, rw,
5565 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5568 struct btrfs_device *device;
5569 struct btrfs_fs_devices *cur_devices;
5571 cur_devices = fs_info->fs_devices;
5572 while (cur_devices) {
5574 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5575 device = __find_device(&cur_devices->devices,
5580 cur_devices = cur_devices->seed;
5585 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5586 u64 devid, u8 *dev_uuid)
5588 struct btrfs_device *device;
5589 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5591 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5595 list_add(&device->dev_list, &fs_devices->devices);
5596 device->fs_devices = fs_devices;
5597 fs_devices->num_devices++;
5599 device->missing = 1;
5600 fs_devices->missing_devices++;
5606 * btrfs_alloc_device - allocate struct btrfs_device
5607 * @fs_info: used only for generating a new devid, can be NULL if
5608 * devid is provided (i.e. @devid != NULL).
5609 * @devid: a pointer to devid for this device. If NULL a new devid
5611 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5614 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5615 * on error. Returned struct is not linked onto any lists and can be
5616 * destroyed with kfree() right away.
5618 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5622 struct btrfs_device *dev;
5625 if (!devid && !fs_info) {
5627 return ERR_PTR(-EINVAL);
5630 dev = __alloc_device();
5639 ret = find_next_devid(fs_info, &tmp);
5642 return ERR_PTR(ret);
5648 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5650 generate_random_uuid(dev->uuid);
5652 dev->work.func = pending_bios_fn;
5657 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5658 struct extent_buffer *leaf,
5659 struct btrfs_chunk *chunk)
5661 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5662 struct map_lookup *map;
5663 struct extent_map *em;
5667 u8 uuid[BTRFS_UUID_SIZE];
5672 logical = key->offset;
5673 length = btrfs_chunk_length(leaf, chunk);
5675 read_lock(&map_tree->map_tree.lock);
5676 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5677 read_unlock(&map_tree->map_tree.lock);
5679 /* already mapped? */
5680 if (em && em->start <= logical && em->start + em->len > logical) {
5681 free_extent_map(em);
5684 free_extent_map(em);
5687 em = alloc_extent_map();
5690 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5691 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5693 free_extent_map(em);
5697 em->bdev = (struct block_device *)map;
5698 em->start = logical;
5701 em->block_start = 0;
5702 em->block_len = em->len;
5704 map->num_stripes = num_stripes;
5705 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5706 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5707 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5708 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5709 map->type = btrfs_chunk_type(leaf, chunk);
5710 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5711 for (i = 0; i < num_stripes; i++) {
5712 map->stripes[i].physical =
5713 btrfs_stripe_offset_nr(leaf, chunk, i);
5714 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5715 read_extent_buffer(leaf, uuid, (unsigned long)
5716 btrfs_stripe_dev_uuid_nr(chunk, i),
5718 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5720 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5722 free_extent_map(em);
5725 if (!map->stripes[i].dev) {
5726 map->stripes[i].dev =
5727 add_missing_dev(root, devid, uuid);
5728 if (!map->stripes[i].dev) {
5730 free_extent_map(em);
5734 map->stripes[i].dev->in_fs_metadata = 1;
5737 write_lock(&map_tree->map_tree.lock);
5738 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5739 write_unlock(&map_tree->map_tree.lock);
5740 BUG_ON(ret); /* Tree corruption */
5741 free_extent_map(em);
5746 static void fill_device_from_item(struct extent_buffer *leaf,
5747 struct btrfs_dev_item *dev_item,
5748 struct btrfs_device *device)
5752 device->devid = btrfs_device_id(leaf, dev_item);
5753 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5754 device->total_bytes = device->disk_total_bytes;
5755 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5756 device->type = btrfs_device_type(leaf, dev_item);
5757 device->io_align = btrfs_device_io_align(leaf, dev_item);
5758 device->io_width = btrfs_device_io_width(leaf, dev_item);
5759 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5760 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5761 device->is_tgtdev_for_dev_replace = 0;
5763 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5764 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5767 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5769 struct btrfs_fs_devices *fs_devices;
5772 BUG_ON(!mutex_is_locked(&uuid_mutex));
5774 fs_devices = root->fs_info->fs_devices->seed;
5775 while (fs_devices) {
5776 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5780 fs_devices = fs_devices->seed;
5783 fs_devices = find_fsid(fsid);
5789 fs_devices = clone_fs_devices(fs_devices);
5790 if (IS_ERR(fs_devices)) {
5791 ret = PTR_ERR(fs_devices);
5795 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5796 root->fs_info->bdev_holder);
5798 free_fs_devices(fs_devices);
5802 if (!fs_devices->seeding) {
5803 __btrfs_close_devices(fs_devices);
5804 free_fs_devices(fs_devices);
5809 fs_devices->seed = root->fs_info->fs_devices->seed;
5810 root->fs_info->fs_devices->seed = fs_devices;
5815 static int read_one_dev(struct btrfs_root *root,
5816 struct extent_buffer *leaf,
5817 struct btrfs_dev_item *dev_item)
5819 struct btrfs_device *device;
5822 u8 fs_uuid[BTRFS_UUID_SIZE];
5823 u8 dev_uuid[BTRFS_UUID_SIZE];
5825 devid = btrfs_device_id(leaf, dev_item);
5826 read_extent_buffer(leaf, dev_uuid,
5827 (unsigned long)btrfs_device_uuid(dev_item),
5829 read_extent_buffer(leaf, fs_uuid,
5830 (unsigned long)btrfs_device_fsid(dev_item),
5833 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5834 ret = open_seed_devices(root, fs_uuid);
5835 if (ret && !btrfs_test_opt(root, DEGRADED))
5839 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5840 if (!device || !device->bdev) {
5841 if (!btrfs_test_opt(root, DEGRADED))
5845 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5846 device = add_missing_dev(root, devid, dev_uuid);
5849 } else if (!device->missing) {
5851 * this happens when a device that was properly setup
5852 * in the device info lists suddenly goes bad.
5853 * device->bdev is NULL, and so we have to set
5854 * device->missing to one here
5856 root->fs_info->fs_devices->missing_devices++;
5857 device->missing = 1;
5861 if (device->fs_devices != root->fs_info->fs_devices) {
5862 BUG_ON(device->writeable);
5863 if (device->generation !=
5864 btrfs_device_generation(leaf, dev_item))
5868 fill_device_from_item(leaf, dev_item, device);
5869 device->in_fs_metadata = 1;
5870 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5871 device->fs_devices->total_rw_bytes += device->total_bytes;
5872 spin_lock(&root->fs_info->free_chunk_lock);
5873 root->fs_info->free_chunk_space += device->total_bytes -
5875 spin_unlock(&root->fs_info->free_chunk_lock);
5881 int btrfs_read_sys_array(struct btrfs_root *root)
5883 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5884 struct extent_buffer *sb;
5885 struct btrfs_disk_key *disk_key;
5886 struct btrfs_chunk *chunk;
5888 unsigned long sb_ptr;
5894 struct btrfs_key key;
5896 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5897 BTRFS_SUPER_INFO_SIZE);
5900 btrfs_set_buffer_uptodate(sb);
5901 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5903 * The sb extent buffer is artifical and just used to read the system array.
5904 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5905 * pages up-to-date when the page is larger: extent does not cover the
5906 * whole page and consequently check_page_uptodate does not find all
5907 * the page's extents up-to-date (the hole beyond sb),
5908 * write_extent_buffer then triggers a WARN_ON.
5910 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5911 * but sb spans only this function. Add an explicit SetPageUptodate call
5912 * to silence the warning eg. on PowerPC 64.
5914 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5915 SetPageUptodate(sb->pages[0]);
5917 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5918 array_size = btrfs_super_sys_array_size(super_copy);
5920 ptr = super_copy->sys_chunk_array;
5921 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5924 while (cur < array_size) {
5925 disk_key = (struct btrfs_disk_key *)ptr;
5926 btrfs_disk_key_to_cpu(&key, disk_key);
5928 len = sizeof(*disk_key); ptr += len;
5932 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5933 chunk = (struct btrfs_chunk *)sb_ptr;
5934 ret = read_one_chunk(root, &key, sb, chunk);
5937 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5938 len = btrfs_chunk_item_size(num_stripes);
5947 free_extent_buffer(sb);
5951 int btrfs_read_chunk_tree(struct btrfs_root *root)
5953 struct btrfs_path *path;
5954 struct extent_buffer *leaf;
5955 struct btrfs_key key;
5956 struct btrfs_key found_key;
5960 root = root->fs_info->chunk_root;
5962 path = btrfs_alloc_path();
5966 mutex_lock(&uuid_mutex);
5970 * Read all device items, and then all the chunk items. All
5971 * device items are found before any chunk item (their object id
5972 * is smaller than the lowest possible object id for a chunk
5973 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5975 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5982 leaf = path->nodes[0];
5983 slot = path->slots[0];
5984 if (slot >= btrfs_header_nritems(leaf)) {
5985 ret = btrfs_next_leaf(root, path);
5992 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5993 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5994 struct btrfs_dev_item *dev_item;
5995 dev_item = btrfs_item_ptr(leaf, slot,
5996 struct btrfs_dev_item);
5997 ret = read_one_dev(root, leaf, dev_item);
6000 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6001 struct btrfs_chunk *chunk;
6002 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6003 ret = read_one_chunk(root, &found_key, leaf, chunk);
6011 unlock_chunks(root);
6012 mutex_unlock(&uuid_mutex);
6014 btrfs_free_path(path);
6018 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6020 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6021 struct btrfs_device *device;
6023 mutex_lock(&fs_devices->device_list_mutex);
6024 list_for_each_entry(device, &fs_devices->devices, dev_list)
6025 device->dev_root = fs_info->dev_root;
6026 mutex_unlock(&fs_devices->device_list_mutex);
6029 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6033 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6034 btrfs_dev_stat_reset(dev, i);
6037 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6039 struct btrfs_key key;
6040 struct btrfs_key found_key;
6041 struct btrfs_root *dev_root = fs_info->dev_root;
6042 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6043 struct extent_buffer *eb;
6046 struct btrfs_device *device;
6047 struct btrfs_path *path = NULL;
6050 path = btrfs_alloc_path();
6056 mutex_lock(&fs_devices->device_list_mutex);
6057 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6059 struct btrfs_dev_stats_item *ptr;
6062 key.type = BTRFS_DEV_STATS_KEY;
6063 key.offset = device->devid;
6064 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6066 __btrfs_reset_dev_stats(device);
6067 device->dev_stats_valid = 1;
6068 btrfs_release_path(path);
6071 slot = path->slots[0];
6072 eb = path->nodes[0];
6073 btrfs_item_key_to_cpu(eb, &found_key, slot);
6074 item_size = btrfs_item_size_nr(eb, slot);
6076 ptr = btrfs_item_ptr(eb, slot,
6077 struct btrfs_dev_stats_item);
6079 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6080 if (item_size >= (1 + i) * sizeof(__le64))
6081 btrfs_dev_stat_set(device, i,
6082 btrfs_dev_stats_value(eb, ptr, i));
6084 btrfs_dev_stat_reset(device, i);
6087 device->dev_stats_valid = 1;
6088 btrfs_dev_stat_print_on_load(device);
6089 btrfs_release_path(path);
6091 mutex_unlock(&fs_devices->device_list_mutex);
6094 btrfs_free_path(path);
6095 return ret < 0 ? ret : 0;
6098 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6099 struct btrfs_root *dev_root,
6100 struct btrfs_device *device)
6102 struct btrfs_path *path;
6103 struct btrfs_key key;
6104 struct extent_buffer *eb;
6105 struct btrfs_dev_stats_item *ptr;
6110 key.type = BTRFS_DEV_STATS_KEY;
6111 key.offset = device->devid;
6113 path = btrfs_alloc_path();
6115 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6117 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6118 ret, rcu_str_deref(device->name));
6123 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6124 /* need to delete old one and insert a new one */
6125 ret = btrfs_del_item(trans, dev_root, path);
6127 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6128 rcu_str_deref(device->name), ret);
6135 /* need to insert a new item */
6136 btrfs_release_path(path);
6137 ret = btrfs_insert_empty_item(trans, dev_root, path,
6138 &key, sizeof(*ptr));
6140 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6141 rcu_str_deref(device->name), ret);
6146 eb = path->nodes[0];
6147 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6148 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6149 btrfs_set_dev_stats_value(eb, ptr, i,
6150 btrfs_dev_stat_read(device, i));
6151 btrfs_mark_buffer_dirty(eb);
6154 btrfs_free_path(path);
6159 * called from commit_transaction. Writes all changed device stats to disk.
6161 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6162 struct btrfs_fs_info *fs_info)
6164 struct btrfs_root *dev_root = fs_info->dev_root;
6165 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6166 struct btrfs_device *device;
6169 mutex_lock(&fs_devices->device_list_mutex);
6170 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6171 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6174 ret = update_dev_stat_item(trans, dev_root, device);
6176 device->dev_stats_dirty = 0;
6178 mutex_unlock(&fs_devices->device_list_mutex);
6183 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6185 btrfs_dev_stat_inc(dev, index);
6186 btrfs_dev_stat_print_on_error(dev);
6189 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6191 if (!dev->dev_stats_valid)
6193 printk_ratelimited_in_rcu(KERN_ERR
6194 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6195 rcu_str_deref(dev->name),
6196 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6197 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6198 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6199 btrfs_dev_stat_read(dev,
6200 BTRFS_DEV_STAT_CORRUPTION_ERRS),
6201 btrfs_dev_stat_read(dev,
6202 BTRFS_DEV_STAT_GENERATION_ERRS));
6205 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6209 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6210 if (btrfs_dev_stat_read(dev, i) != 0)
6212 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6213 return; /* all values == 0, suppress message */
6215 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6216 rcu_str_deref(dev->name),
6217 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6218 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6219 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6220 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6221 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6224 int btrfs_get_dev_stats(struct btrfs_root *root,
6225 struct btrfs_ioctl_get_dev_stats *stats)
6227 struct btrfs_device *dev;
6228 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6231 mutex_lock(&fs_devices->device_list_mutex);
6232 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6233 mutex_unlock(&fs_devices->device_list_mutex);
6237 "btrfs: get dev_stats failed, device not found\n");
6239 } else if (!dev->dev_stats_valid) {
6241 "btrfs: get dev_stats failed, not yet valid\n");
6243 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6244 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6245 if (stats->nr_items > i)
6247 btrfs_dev_stat_read_and_reset(dev, i);
6249 btrfs_dev_stat_reset(dev, i);
6252 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6253 if (stats->nr_items > i)
6254 stats->values[i] = btrfs_dev_stat_read(dev, i);
6256 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6257 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6261 int btrfs_scratch_superblock(struct btrfs_device *device)
6263 struct buffer_head *bh;
6264 struct btrfs_super_block *disk_super;
6266 bh = btrfs_read_dev_super(device->bdev);
6269 disk_super = (struct btrfs_super_block *)bh->b_data;
6271 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6272 set_buffer_dirty(bh);
6273 sync_dirty_buffer(bh);