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 fs_devices->num_devices++;
496 mutex_unlock(&fs_devices->device_list_mutex);
498 device->fs_devices = fs_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 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1316 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1317 btrfs_mark_buffer_dirty(leaf);
1319 btrfs_free_path(path);
1323 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1325 struct extent_map_tree *em_tree;
1326 struct extent_map *em;
1330 em_tree = &fs_info->mapping_tree.map_tree;
1331 read_lock(&em_tree->lock);
1332 n = rb_last(&em_tree->map);
1334 em = rb_entry(n, struct extent_map, rb_node);
1335 ret = em->start + em->len;
1337 read_unlock(&em_tree->lock);
1342 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1346 struct btrfs_key key;
1347 struct btrfs_key found_key;
1348 struct btrfs_path *path;
1350 path = btrfs_alloc_path();
1354 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1355 key.type = BTRFS_DEV_ITEM_KEY;
1356 key.offset = (u64)-1;
1358 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1362 BUG_ON(ret == 0); /* Corruption */
1364 ret = btrfs_previous_item(fs_info->chunk_root, path,
1365 BTRFS_DEV_ITEMS_OBJECTID,
1366 BTRFS_DEV_ITEM_KEY);
1370 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1372 *devid_ret = found_key.offset + 1;
1376 btrfs_free_path(path);
1381 * the device information is stored in the chunk root
1382 * the btrfs_device struct should be fully filled in
1384 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1385 struct btrfs_root *root,
1386 struct btrfs_device *device)
1389 struct btrfs_path *path;
1390 struct btrfs_dev_item *dev_item;
1391 struct extent_buffer *leaf;
1392 struct btrfs_key key;
1395 root = root->fs_info->chunk_root;
1397 path = btrfs_alloc_path();
1401 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1402 key.type = BTRFS_DEV_ITEM_KEY;
1403 key.offset = device->devid;
1405 ret = btrfs_insert_empty_item(trans, root, path, &key,
1410 leaf = path->nodes[0];
1411 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1413 btrfs_set_device_id(leaf, dev_item, device->devid);
1414 btrfs_set_device_generation(leaf, dev_item, 0);
1415 btrfs_set_device_type(leaf, dev_item, device->type);
1416 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1417 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1418 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1419 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1420 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1421 btrfs_set_device_group(leaf, dev_item, 0);
1422 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1423 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1424 btrfs_set_device_start_offset(leaf, dev_item, 0);
1426 ptr = btrfs_device_uuid(dev_item);
1427 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1428 ptr = btrfs_device_fsid(dev_item);
1429 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1430 btrfs_mark_buffer_dirty(leaf);
1434 btrfs_free_path(path);
1438 static int btrfs_rm_dev_item(struct btrfs_root *root,
1439 struct btrfs_device *device)
1442 struct btrfs_path *path;
1443 struct btrfs_key key;
1444 struct btrfs_trans_handle *trans;
1446 root = root->fs_info->chunk_root;
1448 path = btrfs_alloc_path();
1452 trans = btrfs_start_transaction(root, 0);
1453 if (IS_ERR(trans)) {
1454 btrfs_free_path(path);
1455 return PTR_ERR(trans);
1457 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1458 key.type = BTRFS_DEV_ITEM_KEY;
1459 key.offset = device->devid;
1462 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1471 ret = btrfs_del_item(trans, root, path);
1475 btrfs_free_path(path);
1476 unlock_chunks(root);
1477 btrfs_commit_transaction(trans, root);
1481 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1483 struct btrfs_device *device;
1484 struct btrfs_device *next_device;
1485 struct block_device *bdev;
1486 struct buffer_head *bh = NULL;
1487 struct btrfs_super_block *disk_super;
1488 struct btrfs_fs_devices *cur_devices;
1495 bool clear_super = false;
1497 mutex_lock(&uuid_mutex);
1500 seq = read_seqbegin(&root->fs_info->profiles_lock);
1502 all_avail = root->fs_info->avail_data_alloc_bits |
1503 root->fs_info->avail_system_alloc_bits |
1504 root->fs_info->avail_metadata_alloc_bits;
1505 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1507 num_devices = root->fs_info->fs_devices->num_devices;
1508 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1509 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1510 WARN_ON(num_devices < 1);
1513 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1515 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1516 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1520 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1521 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1525 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1526 root->fs_info->fs_devices->rw_devices <= 2) {
1527 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1530 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1531 root->fs_info->fs_devices->rw_devices <= 3) {
1532 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1536 if (strcmp(device_path, "missing") == 0) {
1537 struct list_head *devices;
1538 struct btrfs_device *tmp;
1541 devices = &root->fs_info->fs_devices->devices;
1543 * It is safe to read the devices since the volume_mutex
1546 list_for_each_entry(tmp, devices, dev_list) {
1547 if (tmp->in_fs_metadata &&
1548 !tmp->is_tgtdev_for_dev_replace &&
1558 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1562 ret = btrfs_get_bdev_and_sb(device_path,
1563 FMODE_WRITE | FMODE_EXCL,
1564 root->fs_info->bdev_holder, 0,
1568 disk_super = (struct btrfs_super_block *)bh->b_data;
1569 devid = btrfs_stack_device_id(&disk_super->dev_item);
1570 dev_uuid = disk_super->dev_item.uuid;
1571 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1579 if (device->is_tgtdev_for_dev_replace) {
1580 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1584 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1585 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1589 if (device->writeable) {
1591 list_del_init(&device->dev_alloc_list);
1592 unlock_chunks(root);
1593 root->fs_info->fs_devices->rw_devices--;
1597 mutex_unlock(&uuid_mutex);
1598 ret = btrfs_shrink_device(device, 0);
1599 mutex_lock(&uuid_mutex);
1604 * TODO: the superblock still includes this device in its num_devices
1605 * counter although write_all_supers() is not locked out. This
1606 * could give a filesystem state which requires a degraded mount.
1608 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1612 spin_lock(&root->fs_info->free_chunk_lock);
1613 root->fs_info->free_chunk_space = device->total_bytes -
1615 spin_unlock(&root->fs_info->free_chunk_lock);
1617 device->in_fs_metadata = 0;
1618 btrfs_scrub_cancel_dev(root->fs_info, device);
1621 * the device list mutex makes sure that we don't change
1622 * the device list while someone else is writing out all
1623 * the device supers. Whoever is writing all supers, should
1624 * lock the device list mutex before getting the number of
1625 * devices in the super block (super_copy). Conversely,
1626 * whoever updates the number of devices in the super block
1627 * (super_copy) should hold the device list mutex.
1630 cur_devices = device->fs_devices;
1631 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1632 list_del_rcu(&device->dev_list);
1634 device->fs_devices->num_devices--;
1635 device->fs_devices->total_devices--;
1637 if (device->missing)
1638 root->fs_info->fs_devices->missing_devices--;
1640 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1641 struct btrfs_device, dev_list);
1642 if (device->bdev == root->fs_info->sb->s_bdev)
1643 root->fs_info->sb->s_bdev = next_device->bdev;
1644 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1645 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1648 device->fs_devices->open_devices--;
1650 call_rcu(&device->rcu, free_device);
1652 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1653 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1654 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1656 if (cur_devices->open_devices == 0) {
1657 struct btrfs_fs_devices *fs_devices;
1658 fs_devices = root->fs_info->fs_devices;
1659 while (fs_devices) {
1660 if (fs_devices->seed == cur_devices)
1662 fs_devices = fs_devices->seed;
1664 fs_devices->seed = cur_devices->seed;
1665 cur_devices->seed = NULL;
1667 __btrfs_close_devices(cur_devices);
1668 unlock_chunks(root);
1669 free_fs_devices(cur_devices);
1672 root->fs_info->num_tolerated_disk_barrier_failures =
1673 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1676 * at this point, the device is zero sized. We want to
1677 * remove it from the devices list and zero out the old super
1679 if (clear_super && disk_super) {
1680 /* make sure this device isn't detected as part of
1683 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1684 set_buffer_dirty(bh);
1685 sync_dirty_buffer(bh);
1690 /* Notify udev that device has changed */
1692 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1697 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1699 mutex_unlock(&uuid_mutex);
1702 if (device->writeable) {
1704 list_add(&device->dev_alloc_list,
1705 &root->fs_info->fs_devices->alloc_list);
1706 unlock_chunks(root);
1707 root->fs_info->fs_devices->rw_devices++;
1712 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1713 struct btrfs_device *srcdev)
1715 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1716 list_del_rcu(&srcdev->dev_list);
1717 list_del_rcu(&srcdev->dev_alloc_list);
1718 fs_info->fs_devices->num_devices--;
1719 if (srcdev->missing) {
1720 fs_info->fs_devices->missing_devices--;
1721 fs_info->fs_devices->rw_devices++;
1723 if (srcdev->can_discard)
1724 fs_info->fs_devices->num_can_discard--;
1726 fs_info->fs_devices->open_devices--;
1728 call_rcu(&srcdev->rcu, free_device);
1731 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1732 struct btrfs_device *tgtdev)
1734 struct btrfs_device *next_device;
1737 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1739 btrfs_scratch_superblock(tgtdev);
1740 fs_info->fs_devices->open_devices--;
1742 fs_info->fs_devices->num_devices--;
1743 if (tgtdev->can_discard)
1744 fs_info->fs_devices->num_can_discard++;
1746 next_device = list_entry(fs_info->fs_devices->devices.next,
1747 struct btrfs_device, dev_list);
1748 if (tgtdev->bdev == fs_info->sb->s_bdev)
1749 fs_info->sb->s_bdev = next_device->bdev;
1750 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1751 fs_info->fs_devices->latest_bdev = next_device->bdev;
1752 list_del_rcu(&tgtdev->dev_list);
1754 call_rcu(&tgtdev->rcu, free_device);
1756 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1759 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1760 struct btrfs_device **device)
1763 struct btrfs_super_block *disk_super;
1766 struct block_device *bdev;
1767 struct buffer_head *bh;
1770 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1771 root->fs_info->bdev_holder, 0, &bdev, &bh);
1774 disk_super = (struct btrfs_super_block *)bh->b_data;
1775 devid = btrfs_stack_device_id(&disk_super->dev_item);
1776 dev_uuid = disk_super->dev_item.uuid;
1777 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1782 blkdev_put(bdev, FMODE_READ);
1786 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1788 struct btrfs_device **device)
1791 if (strcmp(device_path, "missing") == 0) {
1792 struct list_head *devices;
1793 struct btrfs_device *tmp;
1795 devices = &root->fs_info->fs_devices->devices;
1797 * It is safe to read the devices since the volume_mutex
1798 * is held by the caller.
1800 list_for_each_entry(tmp, devices, dev_list) {
1801 if (tmp->in_fs_metadata && !tmp->bdev) {
1808 pr_err("btrfs: no missing device found\n");
1814 return btrfs_find_device_by_path(root, device_path, device);
1819 * does all the dirty work required for changing file system's UUID.
1821 static int btrfs_prepare_sprout(struct btrfs_root *root)
1823 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1824 struct btrfs_fs_devices *old_devices;
1825 struct btrfs_fs_devices *seed_devices;
1826 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1827 struct btrfs_device *device;
1830 BUG_ON(!mutex_is_locked(&uuid_mutex));
1831 if (!fs_devices->seeding)
1834 seed_devices = __alloc_fs_devices();
1835 if (IS_ERR(seed_devices))
1836 return PTR_ERR(seed_devices);
1838 old_devices = clone_fs_devices(fs_devices);
1839 if (IS_ERR(old_devices)) {
1840 kfree(seed_devices);
1841 return PTR_ERR(old_devices);
1844 list_add(&old_devices->list, &fs_uuids);
1846 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1847 seed_devices->opened = 1;
1848 INIT_LIST_HEAD(&seed_devices->devices);
1849 INIT_LIST_HEAD(&seed_devices->alloc_list);
1850 mutex_init(&seed_devices->device_list_mutex);
1852 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1853 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1856 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1857 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1858 device->fs_devices = seed_devices;
1861 fs_devices->seeding = 0;
1862 fs_devices->num_devices = 0;
1863 fs_devices->open_devices = 0;
1864 fs_devices->total_devices = 0;
1865 fs_devices->seed = seed_devices;
1867 generate_random_uuid(fs_devices->fsid);
1868 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1869 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1870 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1872 super_flags = btrfs_super_flags(disk_super) &
1873 ~BTRFS_SUPER_FLAG_SEEDING;
1874 btrfs_set_super_flags(disk_super, super_flags);
1880 * strore the expected generation for seed devices in device items.
1882 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1883 struct btrfs_root *root)
1885 struct btrfs_path *path;
1886 struct extent_buffer *leaf;
1887 struct btrfs_dev_item *dev_item;
1888 struct btrfs_device *device;
1889 struct btrfs_key key;
1890 u8 fs_uuid[BTRFS_UUID_SIZE];
1891 u8 dev_uuid[BTRFS_UUID_SIZE];
1895 path = btrfs_alloc_path();
1899 root = root->fs_info->chunk_root;
1900 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1902 key.type = BTRFS_DEV_ITEM_KEY;
1905 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1909 leaf = path->nodes[0];
1911 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1912 ret = btrfs_next_leaf(root, path);
1917 leaf = path->nodes[0];
1918 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1919 btrfs_release_path(path);
1923 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1924 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1925 key.type != BTRFS_DEV_ITEM_KEY)
1928 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1929 struct btrfs_dev_item);
1930 devid = btrfs_device_id(leaf, dev_item);
1931 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1933 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1935 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1937 BUG_ON(!device); /* Logic error */
1939 if (device->fs_devices->seeding) {
1940 btrfs_set_device_generation(leaf, dev_item,
1941 device->generation);
1942 btrfs_mark_buffer_dirty(leaf);
1950 btrfs_free_path(path);
1954 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1956 struct request_queue *q;
1957 struct btrfs_trans_handle *trans;
1958 struct btrfs_device *device;
1959 struct block_device *bdev;
1960 struct list_head *devices;
1961 struct super_block *sb = root->fs_info->sb;
1962 struct rcu_string *name;
1964 int seeding_dev = 0;
1967 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1970 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1971 root->fs_info->bdev_holder);
1973 return PTR_ERR(bdev);
1975 if (root->fs_info->fs_devices->seeding) {
1977 down_write(&sb->s_umount);
1978 mutex_lock(&uuid_mutex);
1981 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1983 devices = &root->fs_info->fs_devices->devices;
1985 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1986 list_for_each_entry(device, devices, dev_list) {
1987 if (device->bdev == bdev) {
1990 &root->fs_info->fs_devices->device_list_mutex);
1994 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1996 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1997 if (IS_ERR(device)) {
1998 /* we can safely leave the fs_devices entry around */
1999 ret = PTR_ERR(device);
2003 name = rcu_string_strdup(device_path, GFP_NOFS);
2009 rcu_assign_pointer(device->name, name);
2011 trans = btrfs_start_transaction(root, 0);
2012 if (IS_ERR(trans)) {
2013 rcu_string_free(device->name);
2015 ret = PTR_ERR(trans);
2021 q = bdev_get_queue(bdev);
2022 if (blk_queue_discard(q))
2023 device->can_discard = 1;
2024 device->writeable = 1;
2025 device->generation = trans->transid;
2026 device->io_width = root->sectorsize;
2027 device->io_align = root->sectorsize;
2028 device->sector_size = root->sectorsize;
2029 device->total_bytes = i_size_read(bdev->bd_inode);
2030 device->disk_total_bytes = device->total_bytes;
2031 device->dev_root = root->fs_info->dev_root;
2032 device->bdev = bdev;
2033 device->in_fs_metadata = 1;
2034 device->is_tgtdev_for_dev_replace = 0;
2035 device->mode = FMODE_EXCL;
2036 set_blocksize(device->bdev, 4096);
2039 sb->s_flags &= ~MS_RDONLY;
2040 ret = btrfs_prepare_sprout(root);
2041 BUG_ON(ret); /* -ENOMEM */
2044 device->fs_devices = root->fs_info->fs_devices;
2046 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2047 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2048 list_add(&device->dev_alloc_list,
2049 &root->fs_info->fs_devices->alloc_list);
2050 root->fs_info->fs_devices->num_devices++;
2051 root->fs_info->fs_devices->open_devices++;
2052 root->fs_info->fs_devices->rw_devices++;
2053 root->fs_info->fs_devices->total_devices++;
2054 if (device->can_discard)
2055 root->fs_info->fs_devices->num_can_discard++;
2056 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2058 spin_lock(&root->fs_info->free_chunk_lock);
2059 root->fs_info->free_chunk_space += device->total_bytes;
2060 spin_unlock(&root->fs_info->free_chunk_lock);
2062 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2063 root->fs_info->fs_devices->rotating = 1;
2065 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2066 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2067 total_bytes + device->total_bytes);
2069 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2070 btrfs_set_super_num_devices(root->fs_info->super_copy,
2072 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2075 ret = init_first_rw_device(trans, root, device);
2077 btrfs_abort_transaction(trans, root, ret);
2080 ret = btrfs_finish_sprout(trans, root);
2082 btrfs_abort_transaction(trans, root, ret);
2086 ret = btrfs_add_device(trans, root, device);
2088 btrfs_abort_transaction(trans, root, ret);
2094 * we've got more storage, clear any full flags on the space
2097 btrfs_clear_space_info_full(root->fs_info);
2099 unlock_chunks(root);
2100 root->fs_info->num_tolerated_disk_barrier_failures =
2101 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2102 ret = btrfs_commit_transaction(trans, root);
2105 mutex_unlock(&uuid_mutex);
2106 up_write(&sb->s_umount);
2108 if (ret) /* transaction commit */
2111 ret = btrfs_relocate_sys_chunks(root);
2113 btrfs_error(root->fs_info, ret,
2114 "Failed to relocate sys chunks after "
2115 "device initialization. This can be fixed "
2116 "using the \"btrfs balance\" command.");
2117 trans = btrfs_attach_transaction(root);
2118 if (IS_ERR(trans)) {
2119 if (PTR_ERR(trans) == -ENOENT)
2121 return PTR_ERR(trans);
2123 ret = btrfs_commit_transaction(trans, root);
2129 unlock_chunks(root);
2130 btrfs_end_transaction(trans, root);
2131 rcu_string_free(device->name);
2134 blkdev_put(bdev, FMODE_EXCL);
2136 mutex_unlock(&uuid_mutex);
2137 up_write(&sb->s_umount);
2142 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2143 struct btrfs_device **device_out)
2145 struct request_queue *q;
2146 struct btrfs_device *device;
2147 struct block_device *bdev;
2148 struct btrfs_fs_info *fs_info = root->fs_info;
2149 struct list_head *devices;
2150 struct rcu_string *name;
2151 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2155 if (fs_info->fs_devices->seeding)
2158 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2159 fs_info->bdev_holder);
2161 return PTR_ERR(bdev);
2163 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2165 devices = &fs_info->fs_devices->devices;
2166 list_for_each_entry(device, devices, dev_list) {
2167 if (device->bdev == bdev) {
2173 device = btrfs_alloc_device(NULL, &devid, NULL);
2174 if (IS_ERR(device)) {
2175 ret = PTR_ERR(device);
2179 name = rcu_string_strdup(device_path, GFP_NOFS);
2185 rcu_assign_pointer(device->name, name);
2187 q = bdev_get_queue(bdev);
2188 if (blk_queue_discard(q))
2189 device->can_discard = 1;
2190 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2191 device->writeable = 1;
2192 device->generation = 0;
2193 device->io_width = root->sectorsize;
2194 device->io_align = root->sectorsize;
2195 device->sector_size = root->sectorsize;
2196 device->total_bytes = i_size_read(bdev->bd_inode);
2197 device->disk_total_bytes = device->total_bytes;
2198 device->dev_root = fs_info->dev_root;
2199 device->bdev = bdev;
2200 device->in_fs_metadata = 1;
2201 device->is_tgtdev_for_dev_replace = 1;
2202 device->mode = FMODE_EXCL;
2203 set_blocksize(device->bdev, 4096);
2204 device->fs_devices = fs_info->fs_devices;
2205 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2206 fs_info->fs_devices->num_devices++;
2207 fs_info->fs_devices->open_devices++;
2208 if (device->can_discard)
2209 fs_info->fs_devices->num_can_discard++;
2210 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2212 *device_out = device;
2216 blkdev_put(bdev, FMODE_EXCL);
2220 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2221 struct btrfs_device *tgtdev)
2223 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2224 tgtdev->io_width = fs_info->dev_root->sectorsize;
2225 tgtdev->io_align = fs_info->dev_root->sectorsize;
2226 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2227 tgtdev->dev_root = fs_info->dev_root;
2228 tgtdev->in_fs_metadata = 1;
2231 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2232 struct btrfs_device *device)
2235 struct btrfs_path *path;
2236 struct btrfs_root *root;
2237 struct btrfs_dev_item *dev_item;
2238 struct extent_buffer *leaf;
2239 struct btrfs_key key;
2241 root = device->dev_root->fs_info->chunk_root;
2243 path = btrfs_alloc_path();
2247 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2248 key.type = BTRFS_DEV_ITEM_KEY;
2249 key.offset = device->devid;
2251 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2260 leaf = path->nodes[0];
2261 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2263 btrfs_set_device_id(leaf, dev_item, device->devid);
2264 btrfs_set_device_type(leaf, dev_item, device->type);
2265 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2266 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2267 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2268 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2269 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2270 btrfs_mark_buffer_dirty(leaf);
2273 btrfs_free_path(path);
2277 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2278 struct btrfs_device *device, u64 new_size)
2280 struct btrfs_super_block *super_copy =
2281 device->dev_root->fs_info->super_copy;
2282 u64 old_total = btrfs_super_total_bytes(super_copy);
2283 u64 diff = new_size - device->total_bytes;
2285 if (!device->writeable)
2287 if (new_size <= device->total_bytes ||
2288 device->is_tgtdev_for_dev_replace)
2291 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2292 device->fs_devices->total_rw_bytes += diff;
2294 device->total_bytes = new_size;
2295 device->disk_total_bytes = new_size;
2296 btrfs_clear_space_info_full(device->dev_root->fs_info);
2298 return btrfs_update_device(trans, device);
2301 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2302 struct btrfs_device *device, u64 new_size)
2305 lock_chunks(device->dev_root);
2306 ret = __btrfs_grow_device(trans, device, new_size);
2307 unlock_chunks(device->dev_root);
2311 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2312 struct btrfs_root *root,
2313 u64 chunk_tree, u64 chunk_objectid,
2317 struct btrfs_path *path;
2318 struct btrfs_key key;
2320 root = root->fs_info->chunk_root;
2321 path = btrfs_alloc_path();
2325 key.objectid = chunk_objectid;
2326 key.offset = chunk_offset;
2327 key.type = BTRFS_CHUNK_ITEM_KEY;
2329 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2332 else if (ret > 0) { /* Logic error or corruption */
2333 btrfs_error(root->fs_info, -ENOENT,
2334 "Failed lookup while freeing chunk.");
2339 ret = btrfs_del_item(trans, root, path);
2341 btrfs_error(root->fs_info, ret,
2342 "Failed to delete chunk item.");
2344 btrfs_free_path(path);
2348 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2351 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2352 struct btrfs_disk_key *disk_key;
2353 struct btrfs_chunk *chunk;
2360 struct btrfs_key key;
2362 array_size = btrfs_super_sys_array_size(super_copy);
2364 ptr = super_copy->sys_chunk_array;
2367 while (cur < array_size) {
2368 disk_key = (struct btrfs_disk_key *)ptr;
2369 btrfs_disk_key_to_cpu(&key, disk_key);
2371 len = sizeof(*disk_key);
2373 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2374 chunk = (struct btrfs_chunk *)(ptr + len);
2375 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2376 len += btrfs_chunk_item_size(num_stripes);
2381 if (key.objectid == chunk_objectid &&
2382 key.offset == chunk_offset) {
2383 memmove(ptr, ptr + len, array_size - (cur + len));
2385 btrfs_set_super_sys_array_size(super_copy, array_size);
2394 static int btrfs_relocate_chunk(struct btrfs_root *root,
2395 u64 chunk_tree, u64 chunk_objectid,
2398 struct extent_map_tree *em_tree;
2399 struct btrfs_root *extent_root;
2400 struct btrfs_trans_handle *trans;
2401 struct extent_map *em;
2402 struct map_lookup *map;
2406 root = root->fs_info->chunk_root;
2407 extent_root = root->fs_info->extent_root;
2408 em_tree = &root->fs_info->mapping_tree.map_tree;
2410 ret = btrfs_can_relocate(extent_root, chunk_offset);
2414 /* step one, relocate all the extents inside this chunk */
2415 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2419 trans = btrfs_start_transaction(root, 0);
2420 if (IS_ERR(trans)) {
2421 ret = PTR_ERR(trans);
2422 btrfs_std_error(root->fs_info, ret);
2429 * step two, delete the device extents and the
2430 * chunk tree entries
2432 read_lock(&em_tree->lock);
2433 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2434 read_unlock(&em_tree->lock);
2436 BUG_ON(!em || em->start > chunk_offset ||
2437 em->start + em->len < chunk_offset);
2438 map = (struct map_lookup *)em->bdev;
2440 for (i = 0; i < map->num_stripes; i++) {
2441 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2442 map->stripes[i].physical);
2445 if (map->stripes[i].dev) {
2446 ret = btrfs_update_device(trans, map->stripes[i].dev);
2450 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2455 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2457 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2458 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2462 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2465 write_lock(&em_tree->lock);
2466 remove_extent_mapping(em_tree, em);
2467 write_unlock(&em_tree->lock);
2472 /* once for the tree */
2473 free_extent_map(em);
2475 free_extent_map(em);
2477 unlock_chunks(root);
2478 btrfs_end_transaction(trans, root);
2482 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2484 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2485 struct btrfs_path *path;
2486 struct extent_buffer *leaf;
2487 struct btrfs_chunk *chunk;
2488 struct btrfs_key key;
2489 struct btrfs_key found_key;
2490 u64 chunk_tree = chunk_root->root_key.objectid;
2492 bool retried = false;
2496 path = btrfs_alloc_path();
2501 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2502 key.offset = (u64)-1;
2503 key.type = BTRFS_CHUNK_ITEM_KEY;
2506 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2509 BUG_ON(ret == 0); /* Corruption */
2511 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2518 leaf = path->nodes[0];
2519 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2521 chunk = btrfs_item_ptr(leaf, path->slots[0],
2522 struct btrfs_chunk);
2523 chunk_type = btrfs_chunk_type(leaf, chunk);
2524 btrfs_release_path(path);
2526 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2527 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2536 if (found_key.offset == 0)
2538 key.offset = found_key.offset - 1;
2541 if (failed && !retried) {
2545 } else if (failed && retried) {
2550 btrfs_free_path(path);
2554 static int insert_balance_item(struct btrfs_root *root,
2555 struct btrfs_balance_control *bctl)
2557 struct btrfs_trans_handle *trans;
2558 struct btrfs_balance_item *item;
2559 struct btrfs_disk_balance_args disk_bargs;
2560 struct btrfs_path *path;
2561 struct extent_buffer *leaf;
2562 struct btrfs_key key;
2565 path = btrfs_alloc_path();
2569 trans = btrfs_start_transaction(root, 0);
2570 if (IS_ERR(trans)) {
2571 btrfs_free_path(path);
2572 return PTR_ERR(trans);
2575 key.objectid = BTRFS_BALANCE_OBJECTID;
2576 key.type = BTRFS_BALANCE_ITEM_KEY;
2579 ret = btrfs_insert_empty_item(trans, root, path, &key,
2584 leaf = path->nodes[0];
2585 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2587 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2589 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2590 btrfs_set_balance_data(leaf, item, &disk_bargs);
2591 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2592 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2593 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2594 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2596 btrfs_set_balance_flags(leaf, item, bctl->flags);
2598 btrfs_mark_buffer_dirty(leaf);
2600 btrfs_free_path(path);
2601 err = btrfs_commit_transaction(trans, root);
2607 static int del_balance_item(struct btrfs_root *root)
2609 struct btrfs_trans_handle *trans;
2610 struct btrfs_path *path;
2611 struct btrfs_key key;
2614 path = btrfs_alloc_path();
2618 trans = btrfs_start_transaction(root, 0);
2619 if (IS_ERR(trans)) {
2620 btrfs_free_path(path);
2621 return PTR_ERR(trans);
2624 key.objectid = BTRFS_BALANCE_OBJECTID;
2625 key.type = BTRFS_BALANCE_ITEM_KEY;
2628 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2636 ret = btrfs_del_item(trans, root, path);
2638 btrfs_free_path(path);
2639 err = btrfs_commit_transaction(trans, root);
2646 * This is a heuristic used to reduce the number of chunks balanced on
2647 * resume after balance was interrupted.
2649 static void update_balance_args(struct btrfs_balance_control *bctl)
2652 * Turn on soft mode for chunk types that were being converted.
2654 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2655 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2656 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2657 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2658 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2659 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2662 * Turn on usage filter if is not already used. The idea is
2663 * that chunks that we have already balanced should be
2664 * reasonably full. Don't do it for chunks that are being
2665 * converted - that will keep us from relocating unconverted
2666 * (albeit full) chunks.
2668 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2669 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2670 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2671 bctl->data.usage = 90;
2673 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2674 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2675 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2676 bctl->sys.usage = 90;
2678 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2679 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2680 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2681 bctl->meta.usage = 90;
2686 * Should be called with both balance and volume mutexes held to
2687 * serialize other volume operations (add_dev/rm_dev/resize) with
2688 * restriper. Same goes for unset_balance_control.
2690 static void set_balance_control(struct btrfs_balance_control *bctl)
2692 struct btrfs_fs_info *fs_info = bctl->fs_info;
2694 BUG_ON(fs_info->balance_ctl);
2696 spin_lock(&fs_info->balance_lock);
2697 fs_info->balance_ctl = bctl;
2698 spin_unlock(&fs_info->balance_lock);
2701 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2703 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2705 BUG_ON(!fs_info->balance_ctl);
2707 spin_lock(&fs_info->balance_lock);
2708 fs_info->balance_ctl = NULL;
2709 spin_unlock(&fs_info->balance_lock);
2715 * Balance filters. Return 1 if chunk should be filtered out
2716 * (should not be balanced).
2718 static int chunk_profiles_filter(u64 chunk_type,
2719 struct btrfs_balance_args *bargs)
2721 chunk_type = chunk_to_extended(chunk_type) &
2722 BTRFS_EXTENDED_PROFILE_MASK;
2724 if (bargs->profiles & chunk_type)
2730 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2731 struct btrfs_balance_args *bargs)
2733 struct btrfs_block_group_cache *cache;
2734 u64 chunk_used, user_thresh;
2737 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2738 chunk_used = btrfs_block_group_used(&cache->item);
2740 if (bargs->usage == 0)
2742 else if (bargs->usage > 100)
2743 user_thresh = cache->key.offset;
2745 user_thresh = div_factor_fine(cache->key.offset,
2748 if (chunk_used < user_thresh)
2751 btrfs_put_block_group(cache);
2755 static int chunk_devid_filter(struct extent_buffer *leaf,
2756 struct btrfs_chunk *chunk,
2757 struct btrfs_balance_args *bargs)
2759 struct btrfs_stripe *stripe;
2760 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2763 for (i = 0; i < num_stripes; i++) {
2764 stripe = btrfs_stripe_nr(chunk, i);
2765 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2772 /* [pstart, pend) */
2773 static int chunk_drange_filter(struct extent_buffer *leaf,
2774 struct btrfs_chunk *chunk,
2776 struct btrfs_balance_args *bargs)
2778 struct btrfs_stripe *stripe;
2779 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2785 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2788 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2789 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2790 factor = num_stripes / 2;
2791 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2792 factor = num_stripes - 1;
2793 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2794 factor = num_stripes - 2;
2796 factor = num_stripes;
2799 for (i = 0; i < num_stripes; i++) {
2800 stripe = btrfs_stripe_nr(chunk, i);
2801 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2804 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2805 stripe_length = btrfs_chunk_length(leaf, chunk);
2806 do_div(stripe_length, factor);
2808 if (stripe_offset < bargs->pend &&
2809 stripe_offset + stripe_length > bargs->pstart)
2816 /* [vstart, vend) */
2817 static int chunk_vrange_filter(struct extent_buffer *leaf,
2818 struct btrfs_chunk *chunk,
2820 struct btrfs_balance_args *bargs)
2822 if (chunk_offset < bargs->vend &&
2823 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2824 /* at least part of the chunk is inside this vrange */
2830 static int chunk_soft_convert_filter(u64 chunk_type,
2831 struct btrfs_balance_args *bargs)
2833 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2836 chunk_type = chunk_to_extended(chunk_type) &
2837 BTRFS_EXTENDED_PROFILE_MASK;
2839 if (bargs->target == chunk_type)
2845 static int should_balance_chunk(struct btrfs_root *root,
2846 struct extent_buffer *leaf,
2847 struct btrfs_chunk *chunk, u64 chunk_offset)
2849 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2850 struct btrfs_balance_args *bargs = NULL;
2851 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2854 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2855 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2859 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2860 bargs = &bctl->data;
2861 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2863 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2864 bargs = &bctl->meta;
2866 /* profiles filter */
2867 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2868 chunk_profiles_filter(chunk_type, bargs)) {
2873 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2874 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2879 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2880 chunk_devid_filter(leaf, chunk, bargs)) {
2884 /* drange filter, makes sense only with devid filter */
2885 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2886 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2891 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2892 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2896 /* soft profile changing mode */
2897 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2898 chunk_soft_convert_filter(chunk_type, bargs)) {
2905 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2907 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2908 struct btrfs_root *chunk_root = fs_info->chunk_root;
2909 struct btrfs_root *dev_root = fs_info->dev_root;
2910 struct list_head *devices;
2911 struct btrfs_device *device;
2914 struct btrfs_chunk *chunk;
2915 struct btrfs_path *path;
2916 struct btrfs_key key;
2917 struct btrfs_key found_key;
2918 struct btrfs_trans_handle *trans;
2919 struct extent_buffer *leaf;
2922 int enospc_errors = 0;
2923 bool counting = true;
2925 /* step one make some room on all the devices */
2926 devices = &fs_info->fs_devices->devices;
2927 list_for_each_entry(device, devices, dev_list) {
2928 old_size = device->total_bytes;
2929 size_to_free = div_factor(old_size, 1);
2930 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2931 if (!device->writeable ||
2932 device->total_bytes - device->bytes_used > size_to_free ||
2933 device->is_tgtdev_for_dev_replace)
2936 ret = btrfs_shrink_device(device, old_size - size_to_free);
2941 trans = btrfs_start_transaction(dev_root, 0);
2942 BUG_ON(IS_ERR(trans));
2944 ret = btrfs_grow_device(trans, device, old_size);
2947 btrfs_end_transaction(trans, dev_root);
2950 /* step two, relocate all the chunks */
2951 path = btrfs_alloc_path();
2957 /* zero out stat counters */
2958 spin_lock(&fs_info->balance_lock);
2959 memset(&bctl->stat, 0, sizeof(bctl->stat));
2960 spin_unlock(&fs_info->balance_lock);
2962 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2963 key.offset = (u64)-1;
2964 key.type = BTRFS_CHUNK_ITEM_KEY;
2967 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2968 atomic_read(&fs_info->balance_cancel_req)) {
2973 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2978 * this shouldn't happen, it means the last relocate
2982 BUG(); /* FIXME break ? */
2984 ret = btrfs_previous_item(chunk_root, path, 0,
2985 BTRFS_CHUNK_ITEM_KEY);
2991 leaf = path->nodes[0];
2992 slot = path->slots[0];
2993 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2995 if (found_key.objectid != key.objectid)
2998 /* chunk zero is special */
2999 if (found_key.offset == 0)
3002 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3005 spin_lock(&fs_info->balance_lock);
3006 bctl->stat.considered++;
3007 spin_unlock(&fs_info->balance_lock);
3010 ret = should_balance_chunk(chunk_root, leaf, chunk,
3012 btrfs_release_path(path);
3017 spin_lock(&fs_info->balance_lock);
3018 bctl->stat.expected++;
3019 spin_unlock(&fs_info->balance_lock);
3023 ret = btrfs_relocate_chunk(chunk_root,
3024 chunk_root->root_key.objectid,
3027 if (ret && ret != -ENOSPC)
3029 if (ret == -ENOSPC) {
3032 spin_lock(&fs_info->balance_lock);
3033 bctl->stat.completed++;
3034 spin_unlock(&fs_info->balance_lock);
3037 key.offset = found_key.offset - 1;
3041 btrfs_release_path(path);
3046 btrfs_free_path(path);
3047 if (enospc_errors) {
3048 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3058 * alloc_profile_is_valid - see if a given profile is valid and reduced
3059 * @flags: profile to validate
3060 * @extended: if true @flags is treated as an extended profile
3062 static int alloc_profile_is_valid(u64 flags, int extended)
3064 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3065 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3067 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3069 /* 1) check that all other bits are zeroed */
3073 /* 2) see if profile is reduced */
3075 return !extended; /* "0" is valid for usual profiles */
3077 /* true if exactly one bit set */
3078 return (flags & (flags - 1)) == 0;
3081 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3083 /* cancel requested || normal exit path */
3084 return atomic_read(&fs_info->balance_cancel_req) ||
3085 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3086 atomic_read(&fs_info->balance_cancel_req) == 0);
3089 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3093 unset_balance_control(fs_info);
3094 ret = del_balance_item(fs_info->tree_root);
3096 btrfs_std_error(fs_info, ret);
3098 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3102 * Should be called with both balance and volume mutexes held
3104 int btrfs_balance(struct btrfs_balance_control *bctl,
3105 struct btrfs_ioctl_balance_args *bargs)
3107 struct btrfs_fs_info *fs_info = bctl->fs_info;
3114 if (btrfs_fs_closing(fs_info) ||
3115 atomic_read(&fs_info->balance_pause_req) ||
3116 atomic_read(&fs_info->balance_cancel_req)) {
3121 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3122 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3126 * In case of mixed groups both data and meta should be picked,
3127 * and identical options should be given for both of them.
3129 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3130 if (mixed && (bctl->flags & allowed)) {
3131 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3132 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3133 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3134 printk(KERN_ERR "btrfs: with mixed groups data and "
3135 "metadata balance options must be the same\n");
3141 num_devices = fs_info->fs_devices->num_devices;
3142 btrfs_dev_replace_lock(&fs_info->dev_replace);
3143 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3144 BUG_ON(num_devices < 1);
3147 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3148 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3149 if (num_devices == 1)
3150 allowed |= BTRFS_BLOCK_GROUP_DUP;
3151 else if (num_devices > 1)
3152 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3153 if (num_devices > 2)
3154 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3155 if (num_devices > 3)
3156 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3157 BTRFS_BLOCK_GROUP_RAID6);
3158 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3159 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3160 (bctl->data.target & ~allowed))) {
3161 printk(KERN_ERR "btrfs: unable to start balance with target "
3162 "data profile %llu\n",
3167 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3168 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3169 (bctl->meta.target & ~allowed))) {
3170 printk(KERN_ERR "btrfs: unable to start balance with target "
3171 "metadata profile %llu\n",
3176 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3177 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3178 (bctl->sys.target & ~allowed))) {
3179 printk(KERN_ERR "btrfs: unable to start balance with target "
3180 "system profile %llu\n",
3186 /* allow dup'ed data chunks only in mixed mode */
3187 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3188 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3189 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3194 /* allow to reduce meta or sys integrity only if force set */
3195 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3196 BTRFS_BLOCK_GROUP_RAID10 |
3197 BTRFS_BLOCK_GROUP_RAID5 |
3198 BTRFS_BLOCK_GROUP_RAID6;
3200 seq = read_seqbegin(&fs_info->profiles_lock);
3202 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3203 (fs_info->avail_system_alloc_bits & allowed) &&
3204 !(bctl->sys.target & allowed)) ||
3205 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3206 (fs_info->avail_metadata_alloc_bits & allowed) &&
3207 !(bctl->meta.target & allowed))) {
3208 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3209 printk(KERN_INFO "btrfs: force reducing metadata "
3212 printk(KERN_ERR "btrfs: balance will reduce metadata "
3213 "integrity, use force if you want this\n");
3218 } while (read_seqretry(&fs_info->profiles_lock, seq));
3220 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3221 int num_tolerated_disk_barrier_failures;
3222 u64 target = bctl->sys.target;
3224 num_tolerated_disk_barrier_failures =
3225 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3226 if (num_tolerated_disk_barrier_failures > 0 &&
3228 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3229 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3230 num_tolerated_disk_barrier_failures = 0;
3231 else if (num_tolerated_disk_barrier_failures > 1 &&
3233 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3234 num_tolerated_disk_barrier_failures = 1;
3236 fs_info->num_tolerated_disk_barrier_failures =
3237 num_tolerated_disk_barrier_failures;
3240 ret = insert_balance_item(fs_info->tree_root, bctl);
3241 if (ret && ret != -EEXIST)
3244 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3245 BUG_ON(ret == -EEXIST);
3246 set_balance_control(bctl);
3248 BUG_ON(ret != -EEXIST);
3249 spin_lock(&fs_info->balance_lock);
3250 update_balance_args(bctl);
3251 spin_unlock(&fs_info->balance_lock);
3254 atomic_inc(&fs_info->balance_running);
3255 mutex_unlock(&fs_info->balance_mutex);
3257 ret = __btrfs_balance(fs_info);
3259 mutex_lock(&fs_info->balance_mutex);
3260 atomic_dec(&fs_info->balance_running);
3262 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3263 fs_info->num_tolerated_disk_barrier_failures =
3264 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3268 memset(bargs, 0, sizeof(*bargs));
3269 update_ioctl_balance_args(fs_info, 0, bargs);
3272 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3273 balance_need_close(fs_info)) {
3274 __cancel_balance(fs_info);
3277 wake_up(&fs_info->balance_wait_q);
3281 if (bctl->flags & BTRFS_BALANCE_RESUME)
3282 __cancel_balance(fs_info);
3285 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3290 static int balance_kthread(void *data)
3292 struct btrfs_fs_info *fs_info = data;
3295 mutex_lock(&fs_info->volume_mutex);
3296 mutex_lock(&fs_info->balance_mutex);
3298 if (fs_info->balance_ctl) {
3299 printk(KERN_INFO "btrfs: continuing balance\n");
3300 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3303 mutex_unlock(&fs_info->balance_mutex);
3304 mutex_unlock(&fs_info->volume_mutex);
3309 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3311 struct task_struct *tsk;
3313 spin_lock(&fs_info->balance_lock);
3314 if (!fs_info->balance_ctl) {
3315 spin_unlock(&fs_info->balance_lock);
3318 spin_unlock(&fs_info->balance_lock);
3320 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3321 printk(KERN_INFO "btrfs: force skipping balance\n");
3325 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3326 return PTR_RET(tsk);
3329 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3331 struct btrfs_balance_control *bctl;
3332 struct btrfs_balance_item *item;
3333 struct btrfs_disk_balance_args disk_bargs;
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
3336 struct btrfs_key key;
3339 path = btrfs_alloc_path();
3343 key.objectid = BTRFS_BALANCE_OBJECTID;
3344 key.type = BTRFS_BALANCE_ITEM_KEY;
3347 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3350 if (ret > 0) { /* ret = -ENOENT; */
3355 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3361 leaf = path->nodes[0];
3362 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3364 bctl->fs_info = fs_info;
3365 bctl->flags = btrfs_balance_flags(leaf, item);
3366 bctl->flags |= BTRFS_BALANCE_RESUME;
3368 btrfs_balance_data(leaf, item, &disk_bargs);
3369 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3370 btrfs_balance_meta(leaf, item, &disk_bargs);
3371 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3372 btrfs_balance_sys(leaf, item, &disk_bargs);
3373 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3375 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3377 mutex_lock(&fs_info->volume_mutex);
3378 mutex_lock(&fs_info->balance_mutex);
3380 set_balance_control(bctl);
3382 mutex_unlock(&fs_info->balance_mutex);
3383 mutex_unlock(&fs_info->volume_mutex);
3385 btrfs_free_path(path);
3389 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3393 mutex_lock(&fs_info->balance_mutex);
3394 if (!fs_info->balance_ctl) {
3395 mutex_unlock(&fs_info->balance_mutex);
3399 if (atomic_read(&fs_info->balance_running)) {
3400 atomic_inc(&fs_info->balance_pause_req);
3401 mutex_unlock(&fs_info->balance_mutex);
3403 wait_event(fs_info->balance_wait_q,
3404 atomic_read(&fs_info->balance_running) == 0);
3406 mutex_lock(&fs_info->balance_mutex);
3407 /* we are good with balance_ctl ripped off from under us */
3408 BUG_ON(atomic_read(&fs_info->balance_running));
3409 atomic_dec(&fs_info->balance_pause_req);
3414 mutex_unlock(&fs_info->balance_mutex);
3418 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3420 mutex_lock(&fs_info->balance_mutex);
3421 if (!fs_info->balance_ctl) {
3422 mutex_unlock(&fs_info->balance_mutex);
3426 atomic_inc(&fs_info->balance_cancel_req);
3428 * if we are running just wait and return, balance item is
3429 * deleted in btrfs_balance in this case
3431 if (atomic_read(&fs_info->balance_running)) {
3432 mutex_unlock(&fs_info->balance_mutex);
3433 wait_event(fs_info->balance_wait_q,
3434 atomic_read(&fs_info->balance_running) == 0);
3435 mutex_lock(&fs_info->balance_mutex);
3437 /* __cancel_balance needs volume_mutex */
3438 mutex_unlock(&fs_info->balance_mutex);
3439 mutex_lock(&fs_info->volume_mutex);
3440 mutex_lock(&fs_info->balance_mutex);
3442 if (fs_info->balance_ctl)
3443 __cancel_balance(fs_info);
3445 mutex_unlock(&fs_info->volume_mutex);
3448 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3449 atomic_dec(&fs_info->balance_cancel_req);
3450 mutex_unlock(&fs_info->balance_mutex);
3454 static int btrfs_uuid_scan_kthread(void *data)
3456 struct btrfs_fs_info *fs_info = data;
3457 struct btrfs_root *root = fs_info->tree_root;
3458 struct btrfs_key key;
3459 struct btrfs_key max_key;
3460 struct btrfs_path *path = NULL;
3462 struct extent_buffer *eb;
3464 struct btrfs_root_item root_item;
3466 struct btrfs_trans_handle *trans;
3468 path = btrfs_alloc_path();
3475 key.type = BTRFS_ROOT_ITEM_KEY;
3478 max_key.objectid = (u64)-1;
3479 max_key.type = BTRFS_ROOT_ITEM_KEY;
3480 max_key.offset = (u64)-1;
3482 path->keep_locks = 1;
3485 ret = btrfs_search_forward(root, &key, &max_key, path, 0);
3492 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3493 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3494 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3495 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3498 eb = path->nodes[0];
3499 slot = path->slots[0];
3500 item_size = btrfs_item_size_nr(eb, slot);
3501 if (item_size < sizeof(root_item))
3505 read_extent_buffer(eb, &root_item,
3506 btrfs_item_ptr_offset(eb, slot),
3507 (int)sizeof(root_item));
3508 if (btrfs_root_refs(&root_item) == 0)
3510 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3512 * 1 - subvol uuid item
3513 * 1 - received_subvol uuid item
3515 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3516 if (IS_ERR(trans)) {
3517 ret = PTR_ERR(trans);
3520 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3522 BTRFS_UUID_KEY_SUBVOL,
3525 pr_warn("btrfs: uuid_tree_add failed %d\n",
3527 btrfs_end_transaction(trans,
3528 fs_info->uuid_root);
3533 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3535 /* 1 - received_subvol uuid item */
3536 trans = btrfs_start_transaction(
3537 fs_info->uuid_root, 1);
3538 if (IS_ERR(trans)) {
3539 ret = PTR_ERR(trans);
3543 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3544 root_item.received_uuid,
3545 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3548 pr_warn("btrfs: uuid_tree_add failed %d\n",
3550 btrfs_end_transaction(trans,
3551 fs_info->uuid_root);
3557 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3563 btrfs_release_path(path);
3564 if (key.offset < (u64)-1) {
3566 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3568 key.type = BTRFS_ROOT_ITEM_KEY;
3569 } else if (key.objectid < (u64)-1) {
3571 key.type = BTRFS_ROOT_ITEM_KEY;
3580 btrfs_free_path(path);
3582 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3584 fs_info->update_uuid_tree_gen = 1;
3585 up(&fs_info->uuid_tree_rescan_sem);
3590 * Callback for btrfs_uuid_tree_iterate().
3592 * 0 check succeeded, the entry is not outdated.
3593 * < 0 if an error occured.
3594 * > 0 if the check failed, which means the caller shall remove the entry.
3596 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3597 u8 *uuid, u8 type, u64 subid)
3599 struct btrfs_key key;
3601 struct btrfs_root *subvol_root;
3603 if (type != BTRFS_UUID_KEY_SUBVOL &&
3604 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3607 key.objectid = subid;
3608 key.type = BTRFS_ROOT_ITEM_KEY;
3609 key.offset = (u64)-1;
3610 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3611 if (IS_ERR(subvol_root)) {
3612 ret = PTR_ERR(subvol_root);
3619 case BTRFS_UUID_KEY_SUBVOL:
3620 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3623 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3624 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3634 static int btrfs_uuid_rescan_kthread(void *data)
3636 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3640 * 1st step is to iterate through the existing UUID tree and
3641 * to delete all entries that contain outdated data.
3642 * 2nd step is to add all missing entries to the UUID tree.
3644 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3646 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3647 up(&fs_info->uuid_tree_rescan_sem);
3650 return btrfs_uuid_scan_kthread(data);
3653 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3655 struct btrfs_trans_handle *trans;
3656 struct btrfs_root *tree_root = fs_info->tree_root;
3657 struct btrfs_root *uuid_root;
3658 struct task_struct *task;
3665 trans = btrfs_start_transaction(tree_root, 2);
3667 return PTR_ERR(trans);
3669 uuid_root = btrfs_create_tree(trans, fs_info,
3670 BTRFS_UUID_TREE_OBJECTID);
3671 if (IS_ERR(uuid_root)) {
3672 btrfs_abort_transaction(trans, tree_root,
3673 PTR_ERR(uuid_root));
3674 return PTR_ERR(uuid_root);
3677 fs_info->uuid_root = uuid_root;
3679 ret = btrfs_commit_transaction(trans, tree_root);
3683 down(&fs_info->uuid_tree_rescan_sem);
3684 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3686 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3687 pr_warn("btrfs: failed to start uuid_scan task\n");
3688 up(&fs_info->uuid_tree_rescan_sem);
3689 return PTR_ERR(task);
3695 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3697 struct task_struct *task;
3699 down(&fs_info->uuid_tree_rescan_sem);
3700 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3702 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3703 pr_warn("btrfs: failed to start uuid_rescan task\n");
3704 up(&fs_info->uuid_tree_rescan_sem);
3705 return PTR_ERR(task);
3712 * shrinking a device means finding all of the device extents past
3713 * the new size, and then following the back refs to the chunks.
3714 * The chunk relocation code actually frees the device extent
3716 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3718 struct btrfs_trans_handle *trans;
3719 struct btrfs_root *root = device->dev_root;
3720 struct btrfs_dev_extent *dev_extent = NULL;
3721 struct btrfs_path *path;
3729 bool retried = false;
3730 struct extent_buffer *l;
3731 struct btrfs_key key;
3732 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3733 u64 old_total = btrfs_super_total_bytes(super_copy);
3734 u64 old_size = device->total_bytes;
3735 u64 diff = device->total_bytes - new_size;
3737 if (device->is_tgtdev_for_dev_replace)
3740 path = btrfs_alloc_path();
3748 device->total_bytes = new_size;
3749 if (device->writeable) {
3750 device->fs_devices->total_rw_bytes -= diff;
3751 spin_lock(&root->fs_info->free_chunk_lock);
3752 root->fs_info->free_chunk_space -= diff;
3753 spin_unlock(&root->fs_info->free_chunk_lock);
3755 unlock_chunks(root);
3758 key.objectid = device->devid;
3759 key.offset = (u64)-1;
3760 key.type = BTRFS_DEV_EXTENT_KEY;
3763 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3767 ret = btrfs_previous_item(root, path, 0, key.type);
3772 btrfs_release_path(path);
3777 slot = path->slots[0];
3778 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3780 if (key.objectid != device->devid) {
3781 btrfs_release_path(path);
3785 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3786 length = btrfs_dev_extent_length(l, dev_extent);
3788 if (key.offset + length <= new_size) {
3789 btrfs_release_path(path);
3793 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3794 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3795 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3796 btrfs_release_path(path);
3798 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3800 if (ret && ret != -ENOSPC)
3804 } while (key.offset-- > 0);
3806 if (failed && !retried) {
3810 } else if (failed && retried) {
3814 device->total_bytes = old_size;
3815 if (device->writeable)
3816 device->fs_devices->total_rw_bytes += diff;
3817 spin_lock(&root->fs_info->free_chunk_lock);
3818 root->fs_info->free_chunk_space += diff;
3819 spin_unlock(&root->fs_info->free_chunk_lock);
3820 unlock_chunks(root);
3824 /* Shrinking succeeded, else we would be at "done". */
3825 trans = btrfs_start_transaction(root, 0);
3826 if (IS_ERR(trans)) {
3827 ret = PTR_ERR(trans);
3833 device->disk_total_bytes = new_size;
3834 /* Now btrfs_update_device() will change the on-disk size. */
3835 ret = btrfs_update_device(trans, device);
3837 unlock_chunks(root);
3838 btrfs_end_transaction(trans, root);
3841 WARN_ON(diff > old_total);
3842 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3843 unlock_chunks(root);
3844 btrfs_end_transaction(trans, root);
3846 btrfs_free_path(path);
3850 static int btrfs_add_system_chunk(struct btrfs_root *root,
3851 struct btrfs_key *key,
3852 struct btrfs_chunk *chunk, int item_size)
3854 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3855 struct btrfs_disk_key disk_key;
3859 array_size = btrfs_super_sys_array_size(super_copy);
3860 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3863 ptr = super_copy->sys_chunk_array + array_size;
3864 btrfs_cpu_key_to_disk(&disk_key, key);
3865 memcpy(ptr, &disk_key, sizeof(disk_key));
3866 ptr += sizeof(disk_key);
3867 memcpy(ptr, chunk, item_size);
3868 item_size += sizeof(disk_key);
3869 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3874 * sort the devices in descending order by max_avail, total_avail
3876 static int btrfs_cmp_device_info(const void *a, const void *b)
3878 const struct btrfs_device_info *di_a = a;
3879 const struct btrfs_device_info *di_b = b;
3881 if (di_a->max_avail > di_b->max_avail)
3883 if (di_a->max_avail < di_b->max_avail)
3885 if (di_a->total_avail > di_b->total_avail)
3887 if (di_a->total_avail < di_b->total_avail)
3892 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3893 [BTRFS_RAID_RAID10] = {
3896 .devs_max = 0, /* 0 == as many as possible */
3898 .devs_increment = 2,
3901 [BTRFS_RAID_RAID1] = {
3906 .devs_increment = 2,
3909 [BTRFS_RAID_DUP] = {
3914 .devs_increment = 1,
3917 [BTRFS_RAID_RAID0] = {
3922 .devs_increment = 1,
3925 [BTRFS_RAID_SINGLE] = {
3930 .devs_increment = 1,
3933 [BTRFS_RAID_RAID5] = {
3938 .devs_increment = 1,
3941 [BTRFS_RAID_RAID6] = {
3946 .devs_increment = 1,
3951 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3953 /* TODO allow them to set a preferred stripe size */
3957 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3959 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3962 btrfs_set_fs_incompat(info, RAID56);
3965 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3966 struct btrfs_root *extent_root, u64 start,
3969 struct btrfs_fs_info *info = extent_root->fs_info;
3970 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3971 struct list_head *cur;
3972 struct map_lookup *map = NULL;
3973 struct extent_map_tree *em_tree;
3974 struct extent_map *em;
3975 struct btrfs_device_info *devices_info = NULL;
3977 int num_stripes; /* total number of stripes to allocate */
3978 int data_stripes; /* number of stripes that count for
3980 int sub_stripes; /* sub_stripes info for map */
3981 int dev_stripes; /* stripes per dev */
3982 int devs_max; /* max devs to use */
3983 int devs_min; /* min devs needed */
3984 int devs_increment; /* ndevs has to be a multiple of this */
3985 int ncopies; /* how many copies to data has */
3987 u64 max_stripe_size;
3991 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3997 BUG_ON(!alloc_profile_is_valid(type, 0));
3999 if (list_empty(&fs_devices->alloc_list))
4002 index = __get_raid_index(type);
4004 sub_stripes = btrfs_raid_array[index].sub_stripes;
4005 dev_stripes = btrfs_raid_array[index].dev_stripes;
4006 devs_max = btrfs_raid_array[index].devs_max;
4007 devs_min = btrfs_raid_array[index].devs_min;
4008 devs_increment = btrfs_raid_array[index].devs_increment;
4009 ncopies = btrfs_raid_array[index].ncopies;
4011 if (type & BTRFS_BLOCK_GROUP_DATA) {
4012 max_stripe_size = 1024 * 1024 * 1024;
4013 max_chunk_size = 10 * max_stripe_size;
4014 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4015 /* for larger filesystems, use larger metadata chunks */
4016 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4017 max_stripe_size = 1024 * 1024 * 1024;
4019 max_stripe_size = 256 * 1024 * 1024;
4020 max_chunk_size = max_stripe_size;
4021 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4022 max_stripe_size = 32 * 1024 * 1024;
4023 max_chunk_size = 2 * max_stripe_size;
4025 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4030 /* we don't want a chunk larger than 10% of writeable space */
4031 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4034 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4039 cur = fs_devices->alloc_list.next;
4042 * in the first pass through the devices list, we gather information
4043 * about the available holes on each device.
4046 while (cur != &fs_devices->alloc_list) {
4047 struct btrfs_device *device;
4051 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4055 if (!device->writeable) {
4057 "btrfs: read-only device in alloc_list\n");
4061 if (!device->in_fs_metadata ||
4062 device->is_tgtdev_for_dev_replace)
4065 if (device->total_bytes > device->bytes_used)
4066 total_avail = device->total_bytes - device->bytes_used;
4070 /* If there is no space on this device, skip it. */
4071 if (total_avail == 0)
4074 ret = find_free_dev_extent(trans, device,
4075 max_stripe_size * dev_stripes,
4076 &dev_offset, &max_avail);
4077 if (ret && ret != -ENOSPC)
4081 max_avail = max_stripe_size * dev_stripes;
4083 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4086 if (ndevs == fs_devices->rw_devices) {
4087 WARN(1, "%s: found more than %llu devices\n",
4088 __func__, fs_devices->rw_devices);
4091 devices_info[ndevs].dev_offset = dev_offset;
4092 devices_info[ndevs].max_avail = max_avail;
4093 devices_info[ndevs].total_avail = total_avail;
4094 devices_info[ndevs].dev = device;
4099 * now sort the devices by hole size / available space
4101 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4102 btrfs_cmp_device_info, NULL);
4104 /* round down to number of usable stripes */
4105 ndevs -= ndevs % devs_increment;
4107 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4112 if (devs_max && ndevs > devs_max)
4115 * the primary goal is to maximize the number of stripes, so use as many
4116 * devices as possible, even if the stripes are not maximum sized.
4118 stripe_size = devices_info[ndevs-1].max_avail;
4119 num_stripes = ndevs * dev_stripes;
4122 * this will have to be fixed for RAID1 and RAID10 over
4125 data_stripes = num_stripes / ncopies;
4127 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4128 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4129 btrfs_super_stripesize(info->super_copy));
4130 data_stripes = num_stripes - 1;
4132 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4133 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4134 btrfs_super_stripesize(info->super_copy));
4135 data_stripes = num_stripes - 2;
4139 * Use the number of data stripes to figure out how big this chunk
4140 * is really going to be in terms of logical address space,
4141 * and compare that answer with the max chunk size
4143 if (stripe_size * data_stripes > max_chunk_size) {
4144 u64 mask = (1ULL << 24) - 1;
4145 stripe_size = max_chunk_size;
4146 do_div(stripe_size, data_stripes);
4148 /* bump the answer up to a 16MB boundary */
4149 stripe_size = (stripe_size + mask) & ~mask;
4151 /* but don't go higher than the limits we found
4152 * while searching for free extents
4154 if (stripe_size > devices_info[ndevs-1].max_avail)
4155 stripe_size = devices_info[ndevs-1].max_avail;
4158 do_div(stripe_size, dev_stripes);
4160 /* align to BTRFS_STRIPE_LEN */
4161 do_div(stripe_size, raid_stripe_len);
4162 stripe_size *= raid_stripe_len;
4164 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4169 map->num_stripes = num_stripes;
4171 for (i = 0; i < ndevs; ++i) {
4172 for (j = 0; j < dev_stripes; ++j) {
4173 int s = i * dev_stripes + j;
4174 map->stripes[s].dev = devices_info[i].dev;
4175 map->stripes[s].physical = devices_info[i].dev_offset +
4179 map->sector_size = extent_root->sectorsize;
4180 map->stripe_len = raid_stripe_len;
4181 map->io_align = raid_stripe_len;
4182 map->io_width = raid_stripe_len;
4184 map->sub_stripes = sub_stripes;
4186 num_bytes = stripe_size * data_stripes;
4188 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4190 em = alloc_extent_map();
4195 em->bdev = (struct block_device *)map;
4197 em->len = num_bytes;
4198 em->block_start = 0;
4199 em->block_len = em->len;
4200 em->orig_block_len = stripe_size;
4202 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4203 write_lock(&em_tree->lock);
4204 ret = add_extent_mapping(em_tree, em, 0);
4206 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4207 atomic_inc(&em->refs);
4209 write_unlock(&em_tree->lock);
4211 free_extent_map(em);
4215 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4216 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4219 goto error_del_extent;
4221 free_extent_map(em);
4222 check_raid56_incompat_flag(extent_root->fs_info, type);
4224 kfree(devices_info);
4228 write_lock(&em_tree->lock);
4229 remove_extent_mapping(em_tree, em);
4230 write_unlock(&em_tree->lock);
4232 /* One for our allocation */
4233 free_extent_map(em);
4234 /* One for the tree reference */
4235 free_extent_map(em);
4238 kfree(devices_info);
4242 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4243 struct btrfs_root *extent_root,
4244 u64 chunk_offset, u64 chunk_size)
4246 struct btrfs_key key;
4247 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4248 struct btrfs_device *device;
4249 struct btrfs_chunk *chunk;
4250 struct btrfs_stripe *stripe;
4251 struct extent_map_tree *em_tree;
4252 struct extent_map *em;
4253 struct map_lookup *map;
4260 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4261 read_lock(&em_tree->lock);
4262 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4263 read_unlock(&em_tree->lock);
4266 btrfs_crit(extent_root->fs_info, "unable to find logical "
4267 "%Lu len %Lu", chunk_offset, chunk_size);
4271 if (em->start != chunk_offset || em->len != chunk_size) {
4272 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4273 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4274 chunk_size, em->start, em->len);
4275 free_extent_map(em);
4279 map = (struct map_lookup *)em->bdev;
4280 item_size = btrfs_chunk_item_size(map->num_stripes);
4281 stripe_size = em->orig_block_len;
4283 chunk = kzalloc(item_size, GFP_NOFS);
4289 for (i = 0; i < map->num_stripes; i++) {
4290 device = map->stripes[i].dev;
4291 dev_offset = map->stripes[i].physical;
4293 device->bytes_used += stripe_size;
4294 ret = btrfs_update_device(trans, device);
4297 ret = btrfs_alloc_dev_extent(trans, device,
4298 chunk_root->root_key.objectid,
4299 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4300 chunk_offset, dev_offset,
4306 spin_lock(&extent_root->fs_info->free_chunk_lock);
4307 extent_root->fs_info->free_chunk_space -= (stripe_size *
4309 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4311 stripe = &chunk->stripe;
4312 for (i = 0; i < map->num_stripes; i++) {
4313 device = map->stripes[i].dev;
4314 dev_offset = map->stripes[i].physical;
4316 btrfs_set_stack_stripe_devid(stripe, device->devid);
4317 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4318 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4322 btrfs_set_stack_chunk_length(chunk, chunk_size);
4323 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4324 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4325 btrfs_set_stack_chunk_type(chunk, map->type);
4326 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4327 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4328 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4329 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4330 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4332 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4333 key.type = BTRFS_CHUNK_ITEM_KEY;
4334 key.offset = chunk_offset;
4336 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4337 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4339 * TODO: Cleanup of inserted chunk root in case of
4342 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4348 free_extent_map(em);
4353 * Chunk allocation falls into two parts. The first part does works
4354 * that make the new allocated chunk useable, but not do any operation
4355 * that modifies the chunk tree. The second part does the works that
4356 * require modifying the chunk tree. This division is important for the
4357 * bootstrap process of adding storage to a seed btrfs.
4359 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4360 struct btrfs_root *extent_root, u64 type)
4364 chunk_offset = find_next_chunk(extent_root->fs_info);
4365 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4368 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4369 struct btrfs_root *root,
4370 struct btrfs_device *device)
4373 u64 sys_chunk_offset;
4375 struct btrfs_fs_info *fs_info = root->fs_info;
4376 struct btrfs_root *extent_root = fs_info->extent_root;
4379 chunk_offset = find_next_chunk(fs_info);
4380 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4381 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4386 sys_chunk_offset = find_next_chunk(root->fs_info);
4387 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4388 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4391 btrfs_abort_transaction(trans, root, ret);
4395 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4397 btrfs_abort_transaction(trans, root, ret);
4402 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4404 struct extent_map *em;
4405 struct map_lookup *map;
4406 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4410 read_lock(&map_tree->map_tree.lock);
4411 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4412 read_unlock(&map_tree->map_tree.lock);
4416 if (btrfs_test_opt(root, DEGRADED)) {
4417 free_extent_map(em);
4421 map = (struct map_lookup *)em->bdev;
4422 for (i = 0; i < map->num_stripes; i++) {
4423 if (!map->stripes[i].dev->writeable) {
4428 free_extent_map(em);
4432 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4434 extent_map_tree_init(&tree->map_tree);
4437 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4439 struct extent_map *em;
4442 write_lock(&tree->map_tree.lock);
4443 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4445 remove_extent_mapping(&tree->map_tree, em);
4446 write_unlock(&tree->map_tree.lock);
4451 free_extent_map(em);
4452 /* once for the tree */
4453 free_extent_map(em);
4457 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4459 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4460 struct extent_map *em;
4461 struct map_lookup *map;
4462 struct extent_map_tree *em_tree = &map_tree->map_tree;
4465 read_lock(&em_tree->lock);
4466 em = lookup_extent_mapping(em_tree, logical, len);
4467 read_unlock(&em_tree->lock);
4470 * We could return errors for these cases, but that could get ugly and
4471 * we'd probably do the same thing which is just not do anything else
4472 * and exit, so return 1 so the callers don't try to use other copies.
4475 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4480 if (em->start > logical || em->start + em->len < logical) {
4481 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4482 "%Lu-%Lu\n", logical, logical+len, em->start,
4483 em->start + em->len);
4487 map = (struct map_lookup *)em->bdev;
4488 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4489 ret = map->num_stripes;
4490 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4491 ret = map->sub_stripes;
4492 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4494 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4498 free_extent_map(em);
4500 btrfs_dev_replace_lock(&fs_info->dev_replace);
4501 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4503 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4508 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4509 struct btrfs_mapping_tree *map_tree,
4512 struct extent_map *em;
4513 struct map_lookup *map;
4514 struct extent_map_tree *em_tree = &map_tree->map_tree;
4515 unsigned long len = root->sectorsize;
4517 read_lock(&em_tree->lock);
4518 em = lookup_extent_mapping(em_tree, logical, len);
4519 read_unlock(&em_tree->lock);
4522 BUG_ON(em->start > logical || em->start + em->len < logical);
4523 map = (struct map_lookup *)em->bdev;
4524 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4525 BTRFS_BLOCK_GROUP_RAID6)) {
4526 len = map->stripe_len * nr_data_stripes(map);
4528 free_extent_map(em);
4532 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4533 u64 logical, u64 len, int mirror_num)
4535 struct extent_map *em;
4536 struct map_lookup *map;
4537 struct extent_map_tree *em_tree = &map_tree->map_tree;
4540 read_lock(&em_tree->lock);
4541 em = lookup_extent_mapping(em_tree, logical, len);
4542 read_unlock(&em_tree->lock);
4545 BUG_ON(em->start > logical || em->start + em->len < logical);
4546 map = (struct map_lookup *)em->bdev;
4547 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4548 BTRFS_BLOCK_GROUP_RAID6))
4550 free_extent_map(em);
4554 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4555 struct map_lookup *map, int first, int num,
4556 int optimal, int dev_replace_is_ongoing)
4560 struct btrfs_device *srcdev;
4562 if (dev_replace_is_ongoing &&
4563 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4564 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4565 srcdev = fs_info->dev_replace.srcdev;
4570 * try to avoid the drive that is the source drive for a
4571 * dev-replace procedure, only choose it if no other non-missing
4572 * mirror is available
4574 for (tolerance = 0; tolerance < 2; tolerance++) {
4575 if (map->stripes[optimal].dev->bdev &&
4576 (tolerance || map->stripes[optimal].dev != srcdev))
4578 for (i = first; i < first + num; i++) {
4579 if (map->stripes[i].dev->bdev &&
4580 (tolerance || map->stripes[i].dev != srcdev))
4585 /* we couldn't find one that doesn't fail. Just return something
4586 * and the io error handling code will clean up eventually
4591 static inline int parity_smaller(u64 a, u64 b)
4596 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4597 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4599 struct btrfs_bio_stripe s;
4606 for (i = 0; i < bbio->num_stripes - 1; i++) {
4607 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4608 s = bbio->stripes[i];
4610 bbio->stripes[i] = bbio->stripes[i+1];
4611 raid_map[i] = raid_map[i+1];
4612 bbio->stripes[i+1] = s;
4620 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4621 u64 logical, u64 *length,
4622 struct btrfs_bio **bbio_ret,
4623 int mirror_num, u64 **raid_map_ret)
4625 struct extent_map *em;
4626 struct map_lookup *map;
4627 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4628 struct extent_map_tree *em_tree = &map_tree->map_tree;
4631 u64 stripe_end_offset;
4636 u64 *raid_map = NULL;
4642 struct btrfs_bio *bbio = NULL;
4643 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4644 int dev_replace_is_ongoing = 0;
4645 int num_alloc_stripes;
4646 int patch_the_first_stripe_for_dev_replace = 0;
4647 u64 physical_to_patch_in_first_stripe = 0;
4648 u64 raid56_full_stripe_start = (u64)-1;
4650 read_lock(&em_tree->lock);
4651 em = lookup_extent_mapping(em_tree, logical, *length);
4652 read_unlock(&em_tree->lock);
4655 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4660 if (em->start > logical || em->start + em->len < logical) {
4661 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4662 "found %Lu-%Lu\n", logical, em->start,
4663 em->start + em->len);
4667 map = (struct map_lookup *)em->bdev;
4668 offset = logical - em->start;
4670 stripe_len = map->stripe_len;
4673 * stripe_nr counts the total number of stripes we have to stride
4674 * to get to this block
4676 do_div(stripe_nr, stripe_len);
4678 stripe_offset = stripe_nr * stripe_len;
4679 BUG_ON(offset < stripe_offset);
4681 /* stripe_offset is the offset of this block in its stripe*/
4682 stripe_offset = offset - stripe_offset;
4684 /* if we're here for raid56, we need to know the stripe aligned start */
4685 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4686 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4687 raid56_full_stripe_start = offset;
4689 /* allow a write of a full stripe, but make sure we don't
4690 * allow straddling of stripes
4692 do_div(raid56_full_stripe_start, full_stripe_len);
4693 raid56_full_stripe_start *= full_stripe_len;
4696 if (rw & REQ_DISCARD) {
4697 /* we don't discard raid56 yet */
4699 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4703 *length = min_t(u64, em->len - offset, *length);
4704 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4706 /* For writes to RAID[56], allow a full stripeset across all disks.
4707 For other RAID types and for RAID[56] reads, just allow a single
4708 stripe (on a single disk). */
4709 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4711 max_len = stripe_len * nr_data_stripes(map) -
4712 (offset - raid56_full_stripe_start);
4714 /* we limit the length of each bio to what fits in a stripe */
4715 max_len = stripe_len - stripe_offset;
4717 *length = min_t(u64, em->len - offset, max_len);
4719 *length = em->len - offset;
4722 /* This is for when we're called from btrfs_merge_bio_hook() and all
4723 it cares about is the length */
4727 btrfs_dev_replace_lock(dev_replace);
4728 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4729 if (!dev_replace_is_ongoing)
4730 btrfs_dev_replace_unlock(dev_replace);
4732 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4733 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4734 dev_replace->tgtdev != NULL) {
4736 * in dev-replace case, for repair case (that's the only
4737 * case where the mirror is selected explicitly when
4738 * calling btrfs_map_block), blocks left of the left cursor
4739 * can also be read from the target drive.
4740 * For REQ_GET_READ_MIRRORS, the target drive is added as
4741 * the last one to the array of stripes. For READ, it also
4742 * needs to be supported using the same mirror number.
4743 * If the requested block is not left of the left cursor,
4744 * EIO is returned. This can happen because btrfs_num_copies()
4745 * returns one more in the dev-replace case.
4747 u64 tmp_length = *length;
4748 struct btrfs_bio *tmp_bbio = NULL;
4749 int tmp_num_stripes;
4750 u64 srcdev_devid = dev_replace->srcdev->devid;
4751 int index_srcdev = 0;
4753 u64 physical_of_found = 0;
4755 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4756 logical, &tmp_length, &tmp_bbio, 0, NULL);
4758 WARN_ON(tmp_bbio != NULL);
4762 tmp_num_stripes = tmp_bbio->num_stripes;
4763 if (mirror_num > tmp_num_stripes) {
4765 * REQ_GET_READ_MIRRORS does not contain this
4766 * mirror, that means that the requested area
4767 * is not left of the left cursor
4775 * process the rest of the function using the mirror_num
4776 * of the source drive. Therefore look it up first.
4777 * At the end, patch the device pointer to the one of the
4780 for (i = 0; i < tmp_num_stripes; i++) {
4781 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4783 * In case of DUP, in order to keep it
4784 * simple, only add the mirror with the
4785 * lowest physical address
4788 physical_of_found <=
4789 tmp_bbio->stripes[i].physical)
4794 tmp_bbio->stripes[i].physical;
4799 mirror_num = index_srcdev + 1;
4800 patch_the_first_stripe_for_dev_replace = 1;
4801 physical_to_patch_in_first_stripe = physical_of_found;
4810 } else if (mirror_num > map->num_stripes) {
4816 stripe_nr_orig = stripe_nr;
4817 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4818 do_div(stripe_nr_end, map->stripe_len);
4819 stripe_end_offset = stripe_nr_end * map->stripe_len -
4822 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4823 if (rw & REQ_DISCARD)
4824 num_stripes = min_t(u64, map->num_stripes,
4825 stripe_nr_end - stripe_nr_orig);
4826 stripe_index = do_div(stripe_nr, map->num_stripes);
4827 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4828 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4829 num_stripes = map->num_stripes;
4830 else if (mirror_num)
4831 stripe_index = mirror_num - 1;
4833 stripe_index = find_live_mirror(fs_info, map, 0,
4835 current->pid % map->num_stripes,
4836 dev_replace_is_ongoing);
4837 mirror_num = stripe_index + 1;
4840 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4841 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4842 num_stripes = map->num_stripes;
4843 } else if (mirror_num) {
4844 stripe_index = mirror_num - 1;
4849 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4850 int factor = map->num_stripes / map->sub_stripes;
4852 stripe_index = do_div(stripe_nr, factor);
4853 stripe_index *= map->sub_stripes;
4855 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4856 num_stripes = map->sub_stripes;
4857 else if (rw & REQ_DISCARD)
4858 num_stripes = min_t(u64, map->sub_stripes *
4859 (stripe_nr_end - stripe_nr_orig),
4861 else if (mirror_num)
4862 stripe_index += mirror_num - 1;
4864 int old_stripe_index = stripe_index;
4865 stripe_index = find_live_mirror(fs_info, map,
4867 map->sub_stripes, stripe_index +
4868 current->pid % map->sub_stripes,
4869 dev_replace_is_ongoing);
4870 mirror_num = stripe_index - old_stripe_index + 1;
4873 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4874 BTRFS_BLOCK_GROUP_RAID6)) {
4877 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4881 /* push stripe_nr back to the start of the full stripe */
4882 stripe_nr = raid56_full_stripe_start;
4883 do_div(stripe_nr, stripe_len);
4885 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4887 /* RAID[56] write or recovery. Return all stripes */
4888 num_stripes = map->num_stripes;
4889 max_errors = nr_parity_stripes(map);
4891 raid_map = kmalloc(sizeof(u64) * num_stripes,
4898 /* Work out the disk rotation on this stripe-set */
4900 rot = do_div(tmp, num_stripes);
4902 /* Fill in the logical address of each stripe */
4903 tmp = stripe_nr * nr_data_stripes(map);
4904 for (i = 0; i < nr_data_stripes(map); i++)
4905 raid_map[(i+rot) % num_stripes] =
4906 em->start + (tmp + i) * map->stripe_len;
4908 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4909 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4910 raid_map[(i+rot+1) % num_stripes] =
4913 *length = map->stripe_len;
4918 * Mirror #0 or #1 means the original data block.
4919 * Mirror #2 is RAID5 parity block.
4920 * Mirror #3 is RAID6 Q block.
4922 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4924 stripe_index = nr_data_stripes(map) +
4927 /* We distribute the parity blocks across stripes */
4928 tmp = stripe_nr + stripe_index;
4929 stripe_index = do_div(tmp, map->num_stripes);
4933 * after this do_div call, stripe_nr is the number of stripes
4934 * on this device we have to walk to find the data, and
4935 * stripe_index is the number of our device in the stripe array
4937 stripe_index = do_div(stripe_nr, map->num_stripes);
4938 mirror_num = stripe_index + 1;
4940 BUG_ON(stripe_index >= map->num_stripes);
4942 num_alloc_stripes = num_stripes;
4943 if (dev_replace_is_ongoing) {
4944 if (rw & (REQ_WRITE | REQ_DISCARD))
4945 num_alloc_stripes <<= 1;
4946 if (rw & REQ_GET_READ_MIRRORS)
4947 num_alloc_stripes++;
4949 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4955 atomic_set(&bbio->error, 0);
4957 if (rw & REQ_DISCARD) {
4959 int sub_stripes = 0;
4960 u64 stripes_per_dev = 0;
4961 u32 remaining_stripes = 0;
4962 u32 last_stripe = 0;
4965 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4966 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4969 sub_stripes = map->sub_stripes;
4971 factor = map->num_stripes / sub_stripes;
4972 stripes_per_dev = div_u64_rem(stripe_nr_end -
4975 &remaining_stripes);
4976 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4977 last_stripe *= sub_stripes;
4980 for (i = 0; i < num_stripes; i++) {
4981 bbio->stripes[i].physical =
4982 map->stripes[stripe_index].physical +
4983 stripe_offset + stripe_nr * map->stripe_len;
4984 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4986 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4987 BTRFS_BLOCK_GROUP_RAID10)) {
4988 bbio->stripes[i].length = stripes_per_dev *
4991 if (i / sub_stripes < remaining_stripes)
4992 bbio->stripes[i].length +=
4996 * Special for the first stripe and
4999 * |-------|...|-------|
5003 if (i < sub_stripes)
5004 bbio->stripes[i].length -=
5007 if (stripe_index >= last_stripe &&
5008 stripe_index <= (last_stripe +
5010 bbio->stripes[i].length -=
5013 if (i == sub_stripes - 1)
5016 bbio->stripes[i].length = *length;
5019 if (stripe_index == map->num_stripes) {
5020 /* This could only happen for RAID0/10 */
5026 for (i = 0; i < num_stripes; i++) {
5027 bbio->stripes[i].physical =
5028 map->stripes[stripe_index].physical +
5030 stripe_nr * map->stripe_len;
5031 bbio->stripes[i].dev =
5032 map->stripes[stripe_index].dev;
5037 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5038 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5039 BTRFS_BLOCK_GROUP_RAID10 |
5040 BTRFS_BLOCK_GROUP_RAID5 |
5041 BTRFS_BLOCK_GROUP_DUP)) {
5043 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5048 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5049 dev_replace->tgtdev != NULL) {
5050 int index_where_to_add;
5051 u64 srcdev_devid = dev_replace->srcdev->devid;
5054 * duplicate the write operations while the dev replace
5055 * procedure is running. Since the copying of the old disk
5056 * to the new disk takes place at run time while the
5057 * filesystem is mounted writable, the regular write
5058 * operations to the old disk have to be duplicated to go
5059 * to the new disk as well.
5060 * Note that device->missing is handled by the caller, and
5061 * that the write to the old disk is already set up in the
5064 index_where_to_add = num_stripes;
5065 for (i = 0; i < num_stripes; i++) {
5066 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5067 /* write to new disk, too */
5068 struct btrfs_bio_stripe *new =
5069 bbio->stripes + index_where_to_add;
5070 struct btrfs_bio_stripe *old =
5073 new->physical = old->physical;
5074 new->length = old->length;
5075 new->dev = dev_replace->tgtdev;
5076 index_where_to_add++;
5080 num_stripes = index_where_to_add;
5081 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5082 dev_replace->tgtdev != NULL) {
5083 u64 srcdev_devid = dev_replace->srcdev->devid;
5084 int index_srcdev = 0;
5086 u64 physical_of_found = 0;
5089 * During the dev-replace procedure, the target drive can
5090 * also be used to read data in case it is needed to repair
5091 * a corrupt block elsewhere. This is possible if the
5092 * requested area is left of the left cursor. In this area,
5093 * the target drive is a full copy of the source drive.
5095 for (i = 0; i < num_stripes; i++) {
5096 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5098 * In case of DUP, in order to keep it
5099 * simple, only add the mirror with the
5100 * lowest physical address
5103 physical_of_found <=
5104 bbio->stripes[i].physical)
5108 physical_of_found = bbio->stripes[i].physical;
5112 u64 length = map->stripe_len;
5114 if (physical_of_found + length <=
5115 dev_replace->cursor_left) {
5116 struct btrfs_bio_stripe *tgtdev_stripe =
5117 bbio->stripes + num_stripes;
5119 tgtdev_stripe->physical = physical_of_found;
5120 tgtdev_stripe->length =
5121 bbio->stripes[index_srcdev].length;
5122 tgtdev_stripe->dev = dev_replace->tgtdev;
5130 bbio->num_stripes = num_stripes;
5131 bbio->max_errors = max_errors;
5132 bbio->mirror_num = mirror_num;
5135 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5136 * mirror_num == num_stripes + 1 && dev_replace target drive is
5137 * available as a mirror
5139 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5140 WARN_ON(num_stripes > 1);
5141 bbio->stripes[0].dev = dev_replace->tgtdev;
5142 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5143 bbio->mirror_num = map->num_stripes + 1;
5146 sort_parity_stripes(bbio, raid_map);
5147 *raid_map_ret = raid_map;
5150 if (dev_replace_is_ongoing)
5151 btrfs_dev_replace_unlock(dev_replace);
5152 free_extent_map(em);
5156 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5157 u64 logical, u64 *length,
5158 struct btrfs_bio **bbio_ret, int mirror_num)
5160 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5164 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5165 u64 chunk_start, u64 physical, u64 devid,
5166 u64 **logical, int *naddrs, int *stripe_len)
5168 struct extent_map_tree *em_tree = &map_tree->map_tree;
5169 struct extent_map *em;
5170 struct map_lookup *map;
5178 read_lock(&em_tree->lock);
5179 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5180 read_unlock(&em_tree->lock);
5183 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5188 if (em->start != chunk_start) {
5189 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5190 em->start, chunk_start);
5191 free_extent_map(em);
5194 map = (struct map_lookup *)em->bdev;
5197 rmap_len = map->stripe_len;
5199 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5200 do_div(length, map->num_stripes / map->sub_stripes);
5201 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5202 do_div(length, map->num_stripes);
5203 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5204 BTRFS_BLOCK_GROUP_RAID6)) {
5205 do_div(length, nr_data_stripes(map));
5206 rmap_len = map->stripe_len * nr_data_stripes(map);
5209 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5210 BUG_ON(!buf); /* -ENOMEM */
5212 for (i = 0; i < map->num_stripes; i++) {
5213 if (devid && map->stripes[i].dev->devid != devid)
5215 if (map->stripes[i].physical > physical ||
5216 map->stripes[i].physical + length <= physical)
5219 stripe_nr = physical - map->stripes[i].physical;
5220 do_div(stripe_nr, map->stripe_len);
5222 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5223 stripe_nr = stripe_nr * map->num_stripes + i;
5224 do_div(stripe_nr, map->sub_stripes);
5225 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5226 stripe_nr = stripe_nr * map->num_stripes + i;
5227 } /* else if RAID[56], multiply by nr_data_stripes().
5228 * Alternatively, just use rmap_len below instead of
5229 * map->stripe_len */
5231 bytenr = chunk_start + stripe_nr * rmap_len;
5232 WARN_ON(nr >= map->num_stripes);
5233 for (j = 0; j < nr; j++) {
5234 if (buf[j] == bytenr)
5238 WARN_ON(nr >= map->num_stripes);
5245 *stripe_len = rmap_len;
5247 free_extent_map(em);
5251 static void btrfs_end_bio(struct bio *bio, int err)
5253 struct btrfs_bio *bbio = bio->bi_private;
5254 int is_orig_bio = 0;
5257 atomic_inc(&bbio->error);
5258 if (err == -EIO || err == -EREMOTEIO) {
5259 unsigned int stripe_index =
5260 btrfs_io_bio(bio)->stripe_index;
5261 struct btrfs_device *dev;
5263 BUG_ON(stripe_index >= bbio->num_stripes);
5264 dev = bbio->stripes[stripe_index].dev;
5266 if (bio->bi_rw & WRITE)
5267 btrfs_dev_stat_inc(dev,
5268 BTRFS_DEV_STAT_WRITE_ERRS);
5270 btrfs_dev_stat_inc(dev,
5271 BTRFS_DEV_STAT_READ_ERRS);
5272 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5273 btrfs_dev_stat_inc(dev,
5274 BTRFS_DEV_STAT_FLUSH_ERRS);
5275 btrfs_dev_stat_print_on_error(dev);
5280 if (bio == bbio->orig_bio)
5283 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5286 bio = bbio->orig_bio;
5288 bio->bi_private = bbio->private;
5289 bio->bi_end_io = bbio->end_io;
5290 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5291 /* only send an error to the higher layers if it is
5292 * beyond the tolerance of the btrfs bio
5294 if (atomic_read(&bbio->error) > bbio->max_errors) {
5298 * this bio is actually up to date, we didn't
5299 * go over the max number of errors
5301 set_bit(BIO_UPTODATE, &bio->bi_flags);
5306 bio_endio(bio, err);
5307 } else if (!is_orig_bio) {
5312 struct async_sched {
5315 struct btrfs_fs_info *info;
5316 struct btrfs_work work;
5320 * see run_scheduled_bios for a description of why bios are collected for
5323 * This will add one bio to the pending list for a device and make sure
5324 * the work struct is scheduled.
5326 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5327 struct btrfs_device *device,
5328 int rw, struct bio *bio)
5330 int should_queue = 1;
5331 struct btrfs_pending_bios *pending_bios;
5333 if (device->missing || !device->bdev) {
5334 bio_endio(bio, -EIO);
5338 /* don't bother with additional async steps for reads, right now */
5339 if (!(rw & REQ_WRITE)) {
5341 btrfsic_submit_bio(rw, bio);
5347 * nr_async_bios allows us to reliably return congestion to the
5348 * higher layers. Otherwise, the async bio makes it appear we have
5349 * made progress against dirty pages when we've really just put it
5350 * on a queue for later
5352 atomic_inc(&root->fs_info->nr_async_bios);
5353 WARN_ON(bio->bi_next);
5354 bio->bi_next = NULL;
5357 spin_lock(&device->io_lock);
5358 if (bio->bi_rw & REQ_SYNC)
5359 pending_bios = &device->pending_sync_bios;
5361 pending_bios = &device->pending_bios;
5363 if (pending_bios->tail)
5364 pending_bios->tail->bi_next = bio;
5366 pending_bios->tail = bio;
5367 if (!pending_bios->head)
5368 pending_bios->head = bio;
5369 if (device->running_pending)
5372 spin_unlock(&device->io_lock);
5375 btrfs_queue_worker(&root->fs_info->submit_workers,
5379 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5382 struct bio_vec *prev;
5383 struct request_queue *q = bdev_get_queue(bdev);
5384 unsigned short max_sectors = queue_max_sectors(q);
5385 struct bvec_merge_data bvm = {
5387 .bi_sector = sector,
5388 .bi_rw = bio->bi_rw,
5391 if (bio->bi_vcnt == 0) {
5396 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5397 if (bio_sectors(bio) > max_sectors)
5400 if (!q->merge_bvec_fn)
5403 bvm.bi_size = bio->bi_size - prev->bv_len;
5404 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5409 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5410 struct bio *bio, u64 physical, int dev_nr,
5413 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5415 bio->bi_private = bbio;
5416 btrfs_io_bio(bio)->stripe_index = dev_nr;
5417 bio->bi_end_io = btrfs_end_bio;
5418 bio->bi_sector = physical >> 9;
5421 struct rcu_string *name;
5424 name = rcu_dereference(dev->name);
5425 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5426 "(%s id %llu), size=%u\n", rw,
5427 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5428 name->str, dev->devid, bio->bi_size);
5432 bio->bi_bdev = dev->bdev;
5434 btrfs_schedule_bio(root, dev, rw, bio);
5436 btrfsic_submit_bio(rw, bio);
5439 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5440 struct bio *first_bio, struct btrfs_device *dev,
5441 int dev_nr, int rw, int async)
5443 struct bio_vec *bvec = first_bio->bi_io_vec;
5445 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5446 u64 physical = bbio->stripes[dev_nr].physical;
5449 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5453 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5454 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5455 bvec->bv_offset) < bvec->bv_len) {
5456 u64 len = bio->bi_size;
5458 atomic_inc(&bbio->stripes_pending);
5459 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5467 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5471 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5473 atomic_inc(&bbio->error);
5474 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5475 bio->bi_private = bbio->private;
5476 bio->bi_end_io = bbio->end_io;
5477 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5478 bio->bi_sector = logical >> 9;
5480 bio_endio(bio, -EIO);
5484 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5485 int mirror_num, int async_submit)
5487 struct btrfs_device *dev;
5488 struct bio *first_bio = bio;
5489 u64 logical = (u64)bio->bi_sector << 9;
5492 u64 *raid_map = NULL;
5496 struct btrfs_bio *bbio = NULL;
5498 length = bio->bi_size;
5499 map_length = length;
5501 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5502 mirror_num, &raid_map);
5503 if (ret) /* -ENOMEM */
5506 total_devs = bbio->num_stripes;
5507 bbio->orig_bio = first_bio;
5508 bbio->private = first_bio->bi_private;
5509 bbio->end_io = first_bio->bi_end_io;
5510 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5513 /* In this case, map_length has been set to the length of
5514 a single stripe; not the whole write */
5516 return raid56_parity_write(root, bio, bbio,
5517 raid_map, map_length);
5519 return raid56_parity_recover(root, bio, bbio,
5520 raid_map, map_length,
5525 if (map_length < length) {
5526 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5527 logical, length, map_length);
5531 while (dev_nr < total_devs) {
5532 dev = bbio->stripes[dev_nr].dev;
5533 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5534 bbio_error(bbio, first_bio, logical);
5540 * Check and see if we're ok with this bio based on it's size
5541 * and offset with the given device.
5543 if (!bio_size_ok(dev->bdev, first_bio,
5544 bbio->stripes[dev_nr].physical >> 9)) {
5545 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5546 dev_nr, rw, async_submit);
5552 if (dev_nr < total_devs - 1) {
5553 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5554 BUG_ON(!bio); /* -ENOMEM */
5559 submit_stripe_bio(root, bbio, bio,
5560 bbio->stripes[dev_nr].physical, dev_nr, rw,
5567 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5570 struct btrfs_device *device;
5571 struct btrfs_fs_devices *cur_devices;
5573 cur_devices = fs_info->fs_devices;
5574 while (cur_devices) {
5576 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5577 device = __find_device(&cur_devices->devices,
5582 cur_devices = cur_devices->seed;
5587 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5588 u64 devid, u8 *dev_uuid)
5590 struct btrfs_device *device;
5591 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5593 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5597 list_add(&device->dev_list, &fs_devices->devices);
5598 device->fs_devices = fs_devices;
5599 fs_devices->num_devices++;
5601 device->missing = 1;
5602 fs_devices->missing_devices++;
5608 * btrfs_alloc_device - allocate struct btrfs_device
5609 * @fs_info: used only for generating a new devid, can be NULL if
5610 * devid is provided (i.e. @devid != NULL).
5611 * @devid: a pointer to devid for this device. If NULL a new devid
5613 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5616 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5617 * on error. Returned struct is not linked onto any lists and can be
5618 * destroyed with kfree() right away.
5620 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5624 struct btrfs_device *dev;
5627 if (!devid && !fs_info) {
5629 return ERR_PTR(-EINVAL);
5632 dev = __alloc_device();
5641 ret = find_next_devid(fs_info, &tmp);
5644 return ERR_PTR(ret);
5650 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5652 generate_random_uuid(dev->uuid);
5654 dev->work.func = pending_bios_fn;
5659 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5660 struct extent_buffer *leaf,
5661 struct btrfs_chunk *chunk)
5663 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5664 struct map_lookup *map;
5665 struct extent_map *em;
5669 u8 uuid[BTRFS_UUID_SIZE];
5674 logical = key->offset;
5675 length = btrfs_chunk_length(leaf, chunk);
5677 read_lock(&map_tree->map_tree.lock);
5678 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5679 read_unlock(&map_tree->map_tree.lock);
5681 /* already mapped? */
5682 if (em && em->start <= logical && em->start + em->len > logical) {
5683 free_extent_map(em);
5686 free_extent_map(em);
5689 em = alloc_extent_map();
5692 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5693 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5695 free_extent_map(em);
5699 em->bdev = (struct block_device *)map;
5700 em->start = logical;
5703 em->block_start = 0;
5704 em->block_len = em->len;
5706 map->num_stripes = num_stripes;
5707 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5708 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5709 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5710 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5711 map->type = btrfs_chunk_type(leaf, chunk);
5712 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5713 for (i = 0; i < num_stripes; i++) {
5714 map->stripes[i].physical =
5715 btrfs_stripe_offset_nr(leaf, chunk, i);
5716 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5717 read_extent_buffer(leaf, uuid, (unsigned long)
5718 btrfs_stripe_dev_uuid_nr(chunk, i),
5720 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5722 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5724 free_extent_map(em);
5727 if (!map->stripes[i].dev) {
5728 map->stripes[i].dev =
5729 add_missing_dev(root, devid, uuid);
5730 if (!map->stripes[i].dev) {
5732 free_extent_map(em);
5736 map->stripes[i].dev->in_fs_metadata = 1;
5739 write_lock(&map_tree->map_tree.lock);
5740 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5741 write_unlock(&map_tree->map_tree.lock);
5742 BUG_ON(ret); /* Tree corruption */
5743 free_extent_map(em);
5748 static void fill_device_from_item(struct extent_buffer *leaf,
5749 struct btrfs_dev_item *dev_item,
5750 struct btrfs_device *device)
5754 device->devid = btrfs_device_id(leaf, dev_item);
5755 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5756 device->total_bytes = device->disk_total_bytes;
5757 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5758 device->type = btrfs_device_type(leaf, dev_item);
5759 device->io_align = btrfs_device_io_align(leaf, dev_item);
5760 device->io_width = btrfs_device_io_width(leaf, dev_item);
5761 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5762 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5763 device->is_tgtdev_for_dev_replace = 0;
5765 ptr = btrfs_device_uuid(dev_item);
5766 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5769 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5771 struct btrfs_fs_devices *fs_devices;
5774 BUG_ON(!mutex_is_locked(&uuid_mutex));
5776 fs_devices = root->fs_info->fs_devices->seed;
5777 while (fs_devices) {
5778 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5782 fs_devices = fs_devices->seed;
5785 fs_devices = find_fsid(fsid);
5791 fs_devices = clone_fs_devices(fs_devices);
5792 if (IS_ERR(fs_devices)) {
5793 ret = PTR_ERR(fs_devices);
5797 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5798 root->fs_info->bdev_holder);
5800 free_fs_devices(fs_devices);
5804 if (!fs_devices->seeding) {
5805 __btrfs_close_devices(fs_devices);
5806 free_fs_devices(fs_devices);
5811 fs_devices->seed = root->fs_info->fs_devices->seed;
5812 root->fs_info->fs_devices->seed = fs_devices;
5817 static int read_one_dev(struct btrfs_root *root,
5818 struct extent_buffer *leaf,
5819 struct btrfs_dev_item *dev_item)
5821 struct btrfs_device *device;
5824 u8 fs_uuid[BTRFS_UUID_SIZE];
5825 u8 dev_uuid[BTRFS_UUID_SIZE];
5827 devid = btrfs_device_id(leaf, dev_item);
5828 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5830 read_extent_buffer(leaf, fs_uuid, 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);