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 fs_devices->missing_devices--;
679 new_device = btrfs_alloc_device(NULL, &device->devid,
681 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
683 /* Safe because we are under uuid_mutex */
685 name = rcu_string_strdup(device->name->str, GFP_NOFS);
686 BUG_ON(!name); /* -ENOMEM */
687 rcu_assign_pointer(new_device->name, name);
690 list_replace_rcu(&device->dev_list, &new_device->dev_list);
691 new_device->fs_devices = device->fs_devices;
693 call_rcu(&device->rcu, free_device);
695 mutex_unlock(&fs_devices->device_list_mutex);
697 WARN_ON(fs_devices->open_devices);
698 WARN_ON(fs_devices->rw_devices);
699 fs_devices->opened = 0;
700 fs_devices->seeding = 0;
705 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
707 struct btrfs_fs_devices *seed_devices = NULL;
710 mutex_lock(&uuid_mutex);
711 ret = __btrfs_close_devices(fs_devices);
712 if (!fs_devices->opened) {
713 seed_devices = fs_devices->seed;
714 fs_devices->seed = NULL;
716 mutex_unlock(&uuid_mutex);
718 while (seed_devices) {
719 fs_devices = seed_devices;
720 seed_devices = fs_devices->seed;
721 __btrfs_close_devices(fs_devices);
722 free_fs_devices(fs_devices);
725 * Wait for rcu kworkers under __btrfs_close_devices
726 * to finish all blkdev_puts so device is really
727 * free when umount is done.
733 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
734 fmode_t flags, void *holder)
736 struct request_queue *q;
737 struct block_device *bdev;
738 struct list_head *head = &fs_devices->devices;
739 struct btrfs_device *device;
740 struct block_device *latest_bdev = NULL;
741 struct buffer_head *bh;
742 struct btrfs_super_block *disk_super;
743 u64 latest_devid = 0;
744 u64 latest_transid = 0;
751 list_for_each_entry(device, head, dev_list) {
757 /* Just open everything we can; ignore failures here */
758 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
762 disk_super = (struct btrfs_super_block *)bh->b_data;
763 devid = btrfs_stack_device_id(&disk_super->dev_item);
764 if (devid != device->devid)
767 if (memcmp(device->uuid, disk_super->dev_item.uuid,
771 device->generation = btrfs_super_generation(disk_super);
772 if (!latest_transid || device->generation > latest_transid) {
773 latest_devid = devid;
774 latest_transid = device->generation;
778 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
779 device->writeable = 0;
781 device->writeable = !bdev_read_only(bdev);
785 q = bdev_get_queue(bdev);
786 if (blk_queue_discard(q)) {
787 device->can_discard = 1;
788 fs_devices->num_can_discard++;
792 device->in_fs_metadata = 0;
793 device->mode = flags;
795 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
796 fs_devices->rotating = 1;
798 fs_devices->open_devices++;
799 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
800 fs_devices->rw_devices++;
801 list_add(&device->dev_alloc_list,
802 &fs_devices->alloc_list);
809 blkdev_put(bdev, flags);
812 if (fs_devices->open_devices == 0) {
816 fs_devices->seeding = seeding;
817 fs_devices->opened = 1;
818 fs_devices->latest_bdev = latest_bdev;
819 fs_devices->latest_devid = latest_devid;
820 fs_devices->latest_trans = latest_transid;
821 fs_devices->total_rw_bytes = 0;
826 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
827 fmode_t flags, void *holder)
831 mutex_lock(&uuid_mutex);
832 if (fs_devices->opened) {
833 fs_devices->opened++;
836 ret = __btrfs_open_devices(fs_devices, flags, holder);
838 mutex_unlock(&uuid_mutex);
843 * Look for a btrfs signature on a device. This may be called out of the mount path
844 * and we are not allowed to call set_blocksize during the scan. The superblock
845 * is read via pagecache
847 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
848 struct btrfs_fs_devices **fs_devices_ret)
850 struct btrfs_super_block *disk_super;
851 struct block_device *bdev;
862 * we would like to check all the supers, but that would make
863 * a btrfs mount succeed after a mkfs from a different FS.
864 * So, we need to add a special mount option to scan for
865 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
867 bytenr = btrfs_sb_offset(0);
869 mutex_lock(&uuid_mutex);
871 bdev = blkdev_get_by_path(path, flags, holder);
878 /* make sure our super fits in the device */
879 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
882 /* make sure our super fits in the page */
883 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
886 /* make sure our super doesn't straddle pages on disk */
887 index = bytenr >> PAGE_CACHE_SHIFT;
888 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
891 /* pull in the page with our super */
892 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
895 if (IS_ERR_OR_NULL(page))
900 /* align our pointer to the offset of the super block */
901 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
903 if (btrfs_super_bytenr(disk_super) != bytenr ||
904 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
907 devid = btrfs_stack_device_id(&disk_super->dev_item);
908 transid = btrfs_super_generation(disk_super);
909 total_devices = btrfs_super_num_devices(disk_super);
911 if (disk_super->label[0]) {
912 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
913 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
914 printk(KERN_INFO "device label %s ", disk_super->label);
916 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
919 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
921 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
922 if (!ret && fs_devices_ret)
923 (*fs_devices_ret)->total_devices = total_devices;
927 page_cache_release(page);
930 blkdev_put(bdev, flags);
932 mutex_unlock(&uuid_mutex);
936 /* helper to account the used device space in the range */
937 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
938 u64 end, u64 *length)
940 struct btrfs_key key;
941 struct btrfs_root *root = device->dev_root;
942 struct btrfs_dev_extent *dev_extent;
943 struct btrfs_path *path;
947 struct extent_buffer *l;
951 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
954 path = btrfs_alloc_path();
959 key.objectid = device->devid;
961 key.type = BTRFS_DEV_EXTENT_KEY;
963 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
967 ret = btrfs_previous_item(root, path, key.objectid, key.type);
974 slot = path->slots[0];
975 if (slot >= btrfs_header_nritems(l)) {
976 ret = btrfs_next_leaf(root, path);
984 btrfs_item_key_to_cpu(l, &key, slot);
986 if (key.objectid < device->devid)
989 if (key.objectid > device->devid)
992 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
995 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
996 extent_end = key.offset + btrfs_dev_extent_length(l,
998 if (key.offset <= start && extent_end > end) {
999 *length = end - start + 1;
1001 } else if (key.offset <= start && extent_end > start)
1002 *length += extent_end - start;
1003 else if (key.offset > start && extent_end <= end)
1004 *length += extent_end - key.offset;
1005 else if (key.offset > start && key.offset <= end) {
1006 *length += end - key.offset + 1;
1008 } else if (key.offset > end)
1016 btrfs_free_path(path);
1020 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1021 struct btrfs_device *device,
1022 u64 *start, u64 len)
1024 struct extent_map *em;
1027 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1028 struct map_lookup *map;
1031 map = (struct map_lookup *)em->bdev;
1032 for (i = 0; i < map->num_stripes; i++) {
1033 if (map->stripes[i].dev != device)
1035 if (map->stripes[i].physical >= *start + len ||
1036 map->stripes[i].physical + em->orig_block_len <=
1039 *start = map->stripes[i].physical +
1050 * find_free_dev_extent - find free space in the specified device
1051 * @device: the device which we search the free space in
1052 * @num_bytes: the size of the free space that we need
1053 * @start: store the start of the free space.
1054 * @len: the size of the free space. that we find, or the size of the max
1055 * free space if we don't find suitable free space
1057 * this uses a pretty simple search, the expectation is that it is
1058 * called very infrequently and that a given device has a small number
1061 * @start is used to store the start of the free space if we find. But if we
1062 * don't find suitable free space, it will be used to store the start position
1063 * of the max free space.
1065 * @len is used to store the size of the free space that we find.
1066 * But if we don't find suitable free space, it is used to store the size of
1067 * the max free space.
1069 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1070 struct btrfs_device *device, u64 num_bytes,
1071 u64 *start, u64 *len)
1073 struct btrfs_key key;
1074 struct btrfs_root *root = device->dev_root;
1075 struct btrfs_dev_extent *dev_extent;
1076 struct btrfs_path *path;
1082 u64 search_end = device->total_bytes;
1085 struct extent_buffer *l;
1087 /* FIXME use last free of some kind */
1089 /* we don't want to overwrite the superblock on the drive,
1090 * so we make sure to start at an offset of at least 1MB
1092 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1094 path = btrfs_alloc_path();
1098 max_hole_start = search_start;
1102 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1108 path->search_commit_root = 1;
1109 path->skip_locking = 1;
1111 key.objectid = device->devid;
1112 key.offset = search_start;
1113 key.type = BTRFS_DEV_EXTENT_KEY;
1115 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1119 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1126 slot = path->slots[0];
1127 if (slot >= btrfs_header_nritems(l)) {
1128 ret = btrfs_next_leaf(root, path);
1136 btrfs_item_key_to_cpu(l, &key, slot);
1138 if (key.objectid < device->devid)
1141 if (key.objectid > device->devid)
1144 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1147 if (key.offset > search_start) {
1148 hole_size = key.offset - search_start;
1151 * Have to check before we set max_hole_start, otherwise
1152 * we could end up sending back this offset anyway.
1154 if (contains_pending_extent(trans, device,
1159 if (hole_size > max_hole_size) {
1160 max_hole_start = search_start;
1161 max_hole_size = hole_size;
1165 * If this free space is greater than which we need,
1166 * it must be the max free space that we have found
1167 * until now, so max_hole_start must point to the start
1168 * of this free space and the length of this free space
1169 * is stored in max_hole_size. Thus, we return
1170 * max_hole_start and max_hole_size and go back to the
1173 if (hole_size >= num_bytes) {
1179 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180 extent_end = key.offset + btrfs_dev_extent_length(l,
1182 if (extent_end > search_start)
1183 search_start = extent_end;
1190 * At this point, search_start should be the end of
1191 * allocated dev extents, and when shrinking the device,
1192 * search_end may be smaller than search_start.
1194 if (search_end > search_start)
1195 hole_size = search_end - search_start;
1197 if (hole_size > max_hole_size) {
1198 max_hole_start = search_start;
1199 max_hole_size = hole_size;
1202 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1203 btrfs_release_path(path);
1208 if (hole_size < num_bytes)
1214 btrfs_free_path(path);
1215 *start = max_hole_start;
1217 *len = max_hole_size;
1221 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1222 struct btrfs_device *device,
1226 struct btrfs_path *path;
1227 struct btrfs_root *root = device->dev_root;
1228 struct btrfs_key key;
1229 struct btrfs_key found_key;
1230 struct extent_buffer *leaf = NULL;
1231 struct btrfs_dev_extent *extent = NULL;
1233 path = btrfs_alloc_path();
1237 key.objectid = device->devid;
1239 key.type = BTRFS_DEV_EXTENT_KEY;
1241 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1243 ret = btrfs_previous_item(root, path, key.objectid,
1244 BTRFS_DEV_EXTENT_KEY);
1247 leaf = path->nodes[0];
1248 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1249 extent = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_dev_extent);
1251 BUG_ON(found_key.offset > start || found_key.offset +
1252 btrfs_dev_extent_length(leaf, extent) < start);
1254 btrfs_release_path(path);
1256 } else if (ret == 0) {
1257 leaf = path->nodes[0];
1258 extent = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_dev_extent);
1261 btrfs_error(root->fs_info, ret, "Slot search failed");
1265 if (device->bytes_used > 0) {
1266 u64 len = btrfs_dev_extent_length(leaf, extent);
1267 device->bytes_used -= len;
1268 spin_lock(&root->fs_info->free_chunk_lock);
1269 root->fs_info->free_chunk_space += len;
1270 spin_unlock(&root->fs_info->free_chunk_lock);
1272 ret = btrfs_del_item(trans, root, path);
1274 btrfs_error(root->fs_info, ret,
1275 "Failed to remove dev extent item");
1278 btrfs_free_path(path);
1282 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1283 struct btrfs_device *device,
1284 u64 chunk_tree, u64 chunk_objectid,
1285 u64 chunk_offset, u64 start, u64 num_bytes)
1288 struct btrfs_path *path;
1289 struct btrfs_root *root = device->dev_root;
1290 struct btrfs_dev_extent *extent;
1291 struct extent_buffer *leaf;
1292 struct btrfs_key key;
1294 WARN_ON(!device->in_fs_metadata);
1295 WARN_ON(device->is_tgtdev_for_dev_replace);
1296 path = btrfs_alloc_path();
1300 key.objectid = device->devid;
1302 key.type = BTRFS_DEV_EXTENT_KEY;
1303 ret = btrfs_insert_empty_item(trans, root, path, &key,
1308 leaf = path->nodes[0];
1309 extent = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_dev_extent);
1311 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1312 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1313 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1315 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1316 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1318 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1319 btrfs_mark_buffer_dirty(leaf);
1321 btrfs_free_path(path);
1325 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1327 struct extent_map_tree *em_tree;
1328 struct extent_map *em;
1332 em_tree = &fs_info->mapping_tree.map_tree;
1333 read_lock(&em_tree->lock);
1334 n = rb_last(&em_tree->map);
1336 em = rb_entry(n, struct extent_map, rb_node);
1337 ret = em->start + em->len;
1339 read_unlock(&em_tree->lock);
1344 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1348 struct btrfs_key key;
1349 struct btrfs_key found_key;
1350 struct btrfs_path *path;
1352 path = btrfs_alloc_path();
1356 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1357 key.type = BTRFS_DEV_ITEM_KEY;
1358 key.offset = (u64)-1;
1360 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1364 BUG_ON(ret == 0); /* Corruption */
1366 ret = btrfs_previous_item(fs_info->chunk_root, path,
1367 BTRFS_DEV_ITEMS_OBJECTID,
1368 BTRFS_DEV_ITEM_KEY);
1372 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1374 *devid_ret = found_key.offset + 1;
1378 btrfs_free_path(path);
1383 * the device information is stored in the chunk root
1384 * the btrfs_device struct should be fully filled in
1386 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root,
1388 struct btrfs_device *device)
1391 struct btrfs_path *path;
1392 struct btrfs_dev_item *dev_item;
1393 struct extent_buffer *leaf;
1394 struct btrfs_key key;
1397 root = root->fs_info->chunk_root;
1399 path = btrfs_alloc_path();
1403 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1404 key.type = BTRFS_DEV_ITEM_KEY;
1405 key.offset = device->devid;
1407 ret = btrfs_insert_empty_item(trans, root, path, &key,
1412 leaf = path->nodes[0];
1413 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1415 btrfs_set_device_id(leaf, dev_item, device->devid);
1416 btrfs_set_device_generation(leaf, dev_item, 0);
1417 btrfs_set_device_type(leaf, dev_item, device->type);
1418 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1419 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1420 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1421 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1422 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1423 btrfs_set_device_group(leaf, dev_item, 0);
1424 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1425 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1426 btrfs_set_device_start_offset(leaf, dev_item, 0);
1428 ptr = btrfs_device_uuid(dev_item);
1429 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1430 ptr = btrfs_device_fsid(dev_item);
1431 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1432 btrfs_mark_buffer_dirty(leaf);
1436 btrfs_free_path(path);
1440 static int btrfs_rm_dev_item(struct btrfs_root *root,
1441 struct btrfs_device *device)
1444 struct btrfs_path *path;
1445 struct btrfs_key key;
1446 struct btrfs_trans_handle *trans;
1448 root = root->fs_info->chunk_root;
1450 path = btrfs_alloc_path();
1454 trans = btrfs_start_transaction(root, 0);
1455 if (IS_ERR(trans)) {
1456 btrfs_free_path(path);
1457 return PTR_ERR(trans);
1459 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1460 key.type = BTRFS_DEV_ITEM_KEY;
1461 key.offset = device->devid;
1464 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1473 ret = btrfs_del_item(trans, root, path);
1477 btrfs_free_path(path);
1478 unlock_chunks(root);
1479 btrfs_commit_transaction(trans, root);
1483 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1485 struct btrfs_device *device;
1486 struct btrfs_device *next_device;
1487 struct block_device *bdev;
1488 struct buffer_head *bh = NULL;
1489 struct btrfs_super_block *disk_super;
1490 struct btrfs_fs_devices *cur_devices;
1497 bool clear_super = false;
1499 mutex_lock(&uuid_mutex);
1502 seq = read_seqbegin(&root->fs_info->profiles_lock);
1504 all_avail = root->fs_info->avail_data_alloc_bits |
1505 root->fs_info->avail_system_alloc_bits |
1506 root->fs_info->avail_metadata_alloc_bits;
1507 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1509 num_devices = root->fs_info->fs_devices->num_devices;
1510 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1511 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1512 WARN_ON(num_devices < 1);
1515 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1517 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1518 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1522 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1523 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1527 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1528 root->fs_info->fs_devices->rw_devices <= 2) {
1529 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1532 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1533 root->fs_info->fs_devices->rw_devices <= 3) {
1534 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1538 if (strcmp(device_path, "missing") == 0) {
1539 struct list_head *devices;
1540 struct btrfs_device *tmp;
1543 devices = &root->fs_info->fs_devices->devices;
1545 * It is safe to read the devices since the volume_mutex
1548 list_for_each_entry(tmp, devices, dev_list) {
1549 if (tmp->in_fs_metadata &&
1550 !tmp->is_tgtdev_for_dev_replace &&
1560 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1564 ret = btrfs_get_bdev_and_sb(device_path,
1565 FMODE_WRITE | FMODE_EXCL,
1566 root->fs_info->bdev_holder, 0,
1570 disk_super = (struct btrfs_super_block *)bh->b_data;
1571 devid = btrfs_stack_device_id(&disk_super->dev_item);
1572 dev_uuid = disk_super->dev_item.uuid;
1573 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1581 if (device->is_tgtdev_for_dev_replace) {
1582 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1586 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1587 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1591 if (device->writeable) {
1593 list_del_init(&device->dev_alloc_list);
1594 unlock_chunks(root);
1595 root->fs_info->fs_devices->rw_devices--;
1599 mutex_unlock(&uuid_mutex);
1600 ret = btrfs_shrink_device(device, 0);
1601 mutex_lock(&uuid_mutex);
1606 * TODO: the superblock still includes this device in its num_devices
1607 * counter although write_all_supers() is not locked out. This
1608 * could give a filesystem state which requires a degraded mount.
1610 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1614 spin_lock(&root->fs_info->free_chunk_lock);
1615 root->fs_info->free_chunk_space = device->total_bytes -
1617 spin_unlock(&root->fs_info->free_chunk_lock);
1619 device->in_fs_metadata = 0;
1620 btrfs_scrub_cancel_dev(root->fs_info, device);
1623 * the device list mutex makes sure that we don't change
1624 * the device list while someone else is writing out all
1625 * the device supers. Whoever is writing all supers, should
1626 * lock the device list mutex before getting the number of
1627 * devices in the super block (super_copy). Conversely,
1628 * whoever updates the number of devices in the super block
1629 * (super_copy) should hold the device list mutex.
1632 cur_devices = device->fs_devices;
1633 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1634 list_del_rcu(&device->dev_list);
1636 device->fs_devices->num_devices--;
1637 device->fs_devices->total_devices--;
1639 if (device->missing)
1640 root->fs_info->fs_devices->missing_devices--;
1642 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1643 struct btrfs_device, dev_list);
1644 if (device->bdev == root->fs_info->sb->s_bdev)
1645 root->fs_info->sb->s_bdev = next_device->bdev;
1646 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1647 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1650 device->fs_devices->open_devices--;
1652 call_rcu(&device->rcu, free_device);
1654 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1655 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1656 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1658 if (cur_devices->open_devices == 0) {
1659 struct btrfs_fs_devices *fs_devices;
1660 fs_devices = root->fs_info->fs_devices;
1661 while (fs_devices) {
1662 if (fs_devices->seed == cur_devices)
1664 fs_devices = fs_devices->seed;
1666 fs_devices->seed = cur_devices->seed;
1667 cur_devices->seed = NULL;
1669 __btrfs_close_devices(cur_devices);
1670 unlock_chunks(root);
1671 free_fs_devices(cur_devices);
1674 root->fs_info->num_tolerated_disk_barrier_failures =
1675 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1678 * at this point, the device is zero sized. We want to
1679 * remove it from the devices list and zero out the old super
1681 if (clear_super && disk_super) {
1682 /* make sure this device isn't detected as part of
1685 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1686 set_buffer_dirty(bh);
1687 sync_dirty_buffer(bh);
1692 /* Notify udev that device has changed */
1694 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1699 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1701 mutex_unlock(&uuid_mutex);
1704 if (device->writeable) {
1706 list_add(&device->dev_alloc_list,
1707 &root->fs_info->fs_devices->alloc_list);
1708 unlock_chunks(root);
1709 root->fs_info->fs_devices->rw_devices++;
1714 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1715 struct btrfs_device *srcdev)
1717 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1718 list_del_rcu(&srcdev->dev_list);
1719 list_del_rcu(&srcdev->dev_alloc_list);
1720 fs_info->fs_devices->num_devices--;
1721 if (srcdev->missing) {
1722 fs_info->fs_devices->missing_devices--;
1723 fs_info->fs_devices->rw_devices++;
1725 if (srcdev->can_discard)
1726 fs_info->fs_devices->num_can_discard--;
1728 fs_info->fs_devices->open_devices--;
1730 call_rcu(&srcdev->rcu, free_device);
1733 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1734 struct btrfs_device *tgtdev)
1736 struct btrfs_device *next_device;
1739 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1741 btrfs_scratch_superblock(tgtdev);
1742 fs_info->fs_devices->open_devices--;
1744 fs_info->fs_devices->num_devices--;
1745 if (tgtdev->can_discard)
1746 fs_info->fs_devices->num_can_discard++;
1748 next_device = list_entry(fs_info->fs_devices->devices.next,
1749 struct btrfs_device, dev_list);
1750 if (tgtdev->bdev == fs_info->sb->s_bdev)
1751 fs_info->sb->s_bdev = next_device->bdev;
1752 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1753 fs_info->fs_devices->latest_bdev = next_device->bdev;
1754 list_del_rcu(&tgtdev->dev_list);
1756 call_rcu(&tgtdev->rcu, free_device);
1758 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1761 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1762 struct btrfs_device **device)
1765 struct btrfs_super_block *disk_super;
1768 struct block_device *bdev;
1769 struct buffer_head *bh;
1772 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1773 root->fs_info->bdev_holder, 0, &bdev, &bh);
1776 disk_super = (struct btrfs_super_block *)bh->b_data;
1777 devid = btrfs_stack_device_id(&disk_super->dev_item);
1778 dev_uuid = disk_super->dev_item.uuid;
1779 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1784 blkdev_put(bdev, FMODE_READ);
1788 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1790 struct btrfs_device **device)
1793 if (strcmp(device_path, "missing") == 0) {
1794 struct list_head *devices;
1795 struct btrfs_device *tmp;
1797 devices = &root->fs_info->fs_devices->devices;
1799 * It is safe to read the devices since the volume_mutex
1800 * is held by the caller.
1802 list_for_each_entry(tmp, devices, dev_list) {
1803 if (tmp->in_fs_metadata && !tmp->bdev) {
1810 pr_err("btrfs: no missing device found\n");
1816 return btrfs_find_device_by_path(root, device_path, device);
1821 * does all the dirty work required for changing file system's UUID.
1823 static int btrfs_prepare_sprout(struct btrfs_root *root)
1825 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1826 struct btrfs_fs_devices *old_devices;
1827 struct btrfs_fs_devices *seed_devices;
1828 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1829 struct btrfs_device *device;
1832 BUG_ON(!mutex_is_locked(&uuid_mutex));
1833 if (!fs_devices->seeding)
1836 seed_devices = __alloc_fs_devices();
1837 if (IS_ERR(seed_devices))
1838 return PTR_ERR(seed_devices);
1840 old_devices = clone_fs_devices(fs_devices);
1841 if (IS_ERR(old_devices)) {
1842 kfree(seed_devices);
1843 return PTR_ERR(old_devices);
1846 list_add(&old_devices->list, &fs_uuids);
1848 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1849 seed_devices->opened = 1;
1850 INIT_LIST_HEAD(&seed_devices->devices);
1851 INIT_LIST_HEAD(&seed_devices->alloc_list);
1852 mutex_init(&seed_devices->device_list_mutex);
1854 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1855 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1858 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1859 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1860 device->fs_devices = seed_devices;
1863 fs_devices->seeding = 0;
1864 fs_devices->num_devices = 0;
1865 fs_devices->open_devices = 0;
1866 fs_devices->total_devices = 0;
1867 fs_devices->seed = seed_devices;
1869 generate_random_uuid(fs_devices->fsid);
1870 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1871 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1872 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1874 super_flags = btrfs_super_flags(disk_super) &
1875 ~BTRFS_SUPER_FLAG_SEEDING;
1876 btrfs_set_super_flags(disk_super, super_flags);
1882 * strore the expected generation for seed devices in device items.
1884 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root)
1887 struct btrfs_path *path;
1888 struct extent_buffer *leaf;
1889 struct btrfs_dev_item *dev_item;
1890 struct btrfs_device *device;
1891 struct btrfs_key key;
1892 u8 fs_uuid[BTRFS_UUID_SIZE];
1893 u8 dev_uuid[BTRFS_UUID_SIZE];
1897 path = btrfs_alloc_path();
1901 root = root->fs_info->chunk_root;
1902 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1904 key.type = BTRFS_DEV_ITEM_KEY;
1907 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1911 leaf = path->nodes[0];
1913 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1914 ret = btrfs_next_leaf(root, path);
1919 leaf = path->nodes[0];
1920 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1921 btrfs_release_path(path);
1925 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1926 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1927 key.type != BTRFS_DEV_ITEM_KEY)
1930 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1931 struct btrfs_dev_item);
1932 devid = btrfs_device_id(leaf, dev_item);
1933 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1935 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1937 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1939 BUG_ON(!device); /* Logic error */
1941 if (device->fs_devices->seeding) {
1942 btrfs_set_device_generation(leaf, dev_item,
1943 device->generation);
1944 btrfs_mark_buffer_dirty(leaf);
1952 btrfs_free_path(path);
1956 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1958 struct request_queue *q;
1959 struct btrfs_trans_handle *trans;
1960 struct btrfs_device *device;
1961 struct block_device *bdev;
1962 struct list_head *devices;
1963 struct super_block *sb = root->fs_info->sb;
1964 struct rcu_string *name;
1966 int seeding_dev = 0;
1969 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1972 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1973 root->fs_info->bdev_holder);
1975 return PTR_ERR(bdev);
1977 if (root->fs_info->fs_devices->seeding) {
1979 down_write(&sb->s_umount);
1980 mutex_lock(&uuid_mutex);
1983 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1985 devices = &root->fs_info->fs_devices->devices;
1987 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1988 list_for_each_entry(device, devices, dev_list) {
1989 if (device->bdev == bdev) {
1992 &root->fs_info->fs_devices->device_list_mutex);
1996 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1998 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1999 if (IS_ERR(device)) {
2000 /* we can safely leave the fs_devices entry around */
2001 ret = PTR_ERR(device);
2005 name = rcu_string_strdup(device_path, GFP_NOFS);
2011 rcu_assign_pointer(device->name, name);
2013 trans = btrfs_start_transaction(root, 0);
2014 if (IS_ERR(trans)) {
2015 rcu_string_free(device->name);
2017 ret = PTR_ERR(trans);
2023 q = bdev_get_queue(bdev);
2024 if (blk_queue_discard(q))
2025 device->can_discard = 1;
2026 device->writeable = 1;
2027 device->generation = trans->transid;
2028 device->io_width = root->sectorsize;
2029 device->io_align = root->sectorsize;
2030 device->sector_size = root->sectorsize;
2031 device->total_bytes = i_size_read(bdev->bd_inode);
2032 device->disk_total_bytes = device->total_bytes;
2033 device->dev_root = root->fs_info->dev_root;
2034 device->bdev = bdev;
2035 device->in_fs_metadata = 1;
2036 device->is_tgtdev_for_dev_replace = 0;
2037 device->mode = FMODE_EXCL;
2038 set_blocksize(device->bdev, 4096);
2041 sb->s_flags &= ~MS_RDONLY;
2042 ret = btrfs_prepare_sprout(root);
2043 BUG_ON(ret); /* -ENOMEM */
2046 device->fs_devices = root->fs_info->fs_devices;
2048 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2049 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2050 list_add(&device->dev_alloc_list,
2051 &root->fs_info->fs_devices->alloc_list);
2052 root->fs_info->fs_devices->num_devices++;
2053 root->fs_info->fs_devices->open_devices++;
2054 root->fs_info->fs_devices->rw_devices++;
2055 root->fs_info->fs_devices->total_devices++;
2056 if (device->can_discard)
2057 root->fs_info->fs_devices->num_can_discard++;
2058 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2060 spin_lock(&root->fs_info->free_chunk_lock);
2061 root->fs_info->free_chunk_space += device->total_bytes;
2062 spin_unlock(&root->fs_info->free_chunk_lock);
2064 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2065 root->fs_info->fs_devices->rotating = 1;
2067 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2068 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2069 total_bytes + device->total_bytes);
2071 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2072 btrfs_set_super_num_devices(root->fs_info->super_copy,
2074 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2077 ret = init_first_rw_device(trans, root, device);
2079 btrfs_abort_transaction(trans, root, ret);
2082 ret = btrfs_finish_sprout(trans, root);
2084 btrfs_abort_transaction(trans, root, ret);
2088 ret = btrfs_add_device(trans, root, device);
2090 btrfs_abort_transaction(trans, root, ret);
2096 * we've got more storage, clear any full flags on the space
2099 btrfs_clear_space_info_full(root->fs_info);
2101 unlock_chunks(root);
2102 root->fs_info->num_tolerated_disk_barrier_failures =
2103 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2104 ret = btrfs_commit_transaction(trans, root);
2107 mutex_unlock(&uuid_mutex);
2108 up_write(&sb->s_umount);
2110 if (ret) /* transaction commit */
2113 ret = btrfs_relocate_sys_chunks(root);
2115 btrfs_error(root->fs_info, ret,
2116 "Failed to relocate sys chunks after "
2117 "device initialization. This can be fixed "
2118 "using the \"btrfs balance\" command.");
2119 trans = btrfs_attach_transaction(root);
2120 if (IS_ERR(trans)) {
2121 if (PTR_ERR(trans) == -ENOENT)
2123 return PTR_ERR(trans);
2125 ret = btrfs_commit_transaction(trans, root);
2131 unlock_chunks(root);
2132 btrfs_end_transaction(trans, root);
2133 rcu_string_free(device->name);
2136 blkdev_put(bdev, FMODE_EXCL);
2138 mutex_unlock(&uuid_mutex);
2139 up_write(&sb->s_umount);
2144 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2145 struct btrfs_device **device_out)
2147 struct request_queue *q;
2148 struct btrfs_device *device;
2149 struct block_device *bdev;
2150 struct btrfs_fs_info *fs_info = root->fs_info;
2151 struct list_head *devices;
2152 struct rcu_string *name;
2153 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2157 if (fs_info->fs_devices->seeding)
2160 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2161 fs_info->bdev_holder);
2163 return PTR_ERR(bdev);
2165 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2167 devices = &fs_info->fs_devices->devices;
2168 list_for_each_entry(device, devices, dev_list) {
2169 if (device->bdev == bdev) {
2175 device = btrfs_alloc_device(NULL, &devid, NULL);
2176 if (IS_ERR(device)) {
2177 ret = PTR_ERR(device);
2181 name = rcu_string_strdup(device_path, GFP_NOFS);
2187 rcu_assign_pointer(device->name, name);
2189 q = bdev_get_queue(bdev);
2190 if (blk_queue_discard(q))
2191 device->can_discard = 1;
2192 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2193 device->writeable = 1;
2194 device->generation = 0;
2195 device->io_width = root->sectorsize;
2196 device->io_align = root->sectorsize;
2197 device->sector_size = root->sectorsize;
2198 device->total_bytes = i_size_read(bdev->bd_inode);
2199 device->disk_total_bytes = device->total_bytes;
2200 device->dev_root = fs_info->dev_root;
2201 device->bdev = bdev;
2202 device->in_fs_metadata = 1;
2203 device->is_tgtdev_for_dev_replace = 1;
2204 device->mode = FMODE_EXCL;
2205 set_blocksize(device->bdev, 4096);
2206 device->fs_devices = fs_info->fs_devices;
2207 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2208 fs_info->fs_devices->num_devices++;
2209 fs_info->fs_devices->open_devices++;
2210 if (device->can_discard)
2211 fs_info->fs_devices->num_can_discard++;
2212 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2214 *device_out = device;
2218 blkdev_put(bdev, FMODE_EXCL);
2222 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2223 struct btrfs_device *tgtdev)
2225 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2226 tgtdev->io_width = fs_info->dev_root->sectorsize;
2227 tgtdev->io_align = fs_info->dev_root->sectorsize;
2228 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2229 tgtdev->dev_root = fs_info->dev_root;
2230 tgtdev->in_fs_metadata = 1;
2233 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2234 struct btrfs_device *device)
2237 struct btrfs_path *path;
2238 struct btrfs_root *root;
2239 struct btrfs_dev_item *dev_item;
2240 struct extent_buffer *leaf;
2241 struct btrfs_key key;
2243 root = device->dev_root->fs_info->chunk_root;
2245 path = btrfs_alloc_path();
2249 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2250 key.type = BTRFS_DEV_ITEM_KEY;
2251 key.offset = device->devid;
2253 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2262 leaf = path->nodes[0];
2263 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2265 btrfs_set_device_id(leaf, dev_item, device->devid);
2266 btrfs_set_device_type(leaf, dev_item, device->type);
2267 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2268 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2269 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2270 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2271 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2272 btrfs_mark_buffer_dirty(leaf);
2275 btrfs_free_path(path);
2279 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2280 struct btrfs_device *device, u64 new_size)
2282 struct btrfs_super_block *super_copy =
2283 device->dev_root->fs_info->super_copy;
2284 u64 old_total = btrfs_super_total_bytes(super_copy);
2285 u64 diff = new_size - device->total_bytes;
2287 if (!device->writeable)
2289 if (new_size <= device->total_bytes ||
2290 device->is_tgtdev_for_dev_replace)
2293 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2294 device->fs_devices->total_rw_bytes += diff;
2296 device->total_bytes = new_size;
2297 device->disk_total_bytes = new_size;
2298 btrfs_clear_space_info_full(device->dev_root->fs_info);
2300 return btrfs_update_device(trans, device);
2303 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2304 struct btrfs_device *device, u64 new_size)
2307 lock_chunks(device->dev_root);
2308 ret = __btrfs_grow_device(trans, device, new_size);
2309 unlock_chunks(device->dev_root);
2313 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2314 struct btrfs_root *root,
2315 u64 chunk_tree, u64 chunk_objectid,
2319 struct btrfs_path *path;
2320 struct btrfs_key key;
2322 root = root->fs_info->chunk_root;
2323 path = btrfs_alloc_path();
2327 key.objectid = chunk_objectid;
2328 key.offset = chunk_offset;
2329 key.type = BTRFS_CHUNK_ITEM_KEY;
2331 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2334 else if (ret > 0) { /* Logic error or corruption */
2335 btrfs_error(root->fs_info, -ENOENT,
2336 "Failed lookup while freeing chunk.");
2341 ret = btrfs_del_item(trans, root, path);
2343 btrfs_error(root->fs_info, ret,
2344 "Failed to delete chunk item.");
2346 btrfs_free_path(path);
2350 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2353 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2354 struct btrfs_disk_key *disk_key;
2355 struct btrfs_chunk *chunk;
2362 struct btrfs_key key;
2364 array_size = btrfs_super_sys_array_size(super_copy);
2366 ptr = super_copy->sys_chunk_array;
2369 while (cur < array_size) {
2370 disk_key = (struct btrfs_disk_key *)ptr;
2371 btrfs_disk_key_to_cpu(&key, disk_key);
2373 len = sizeof(*disk_key);
2375 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2376 chunk = (struct btrfs_chunk *)(ptr + len);
2377 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2378 len += btrfs_chunk_item_size(num_stripes);
2383 if (key.objectid == chunk_objectid &&
2384 key.offset == chunk_offset) {
2385 memmove(ptr, ptr + len, array_size - (cur + len));
2387 btrfs_set_super_sys_array_size(super_copy, array_size);
2396 static int btrfs_relocate_chunk(struct btrfs_root *root,
2397 u64 chunk_tree, u64 chunk_objectid,
2400 struct extent_map_tree *em_tree;
2401 struct btrfs_root *extent_root;
2402 struct btrfs_trans_handle *trans;
2403 struct extent_map *em;
2404 struct map_lookup *map;
2408 root = root->fs_info->chunk_root;
2409 extent_root = root->fs_info->extent_root;
2410 em_tree = &root->fs_info->mapping_tree.map_tree;
2412 ret = btrfs_can_relocate(extent_root, chunk_offset);
2416 /* step one, relocate all the extents inside this chunk */
2417 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2421 trans = btrfs_start_transaction(root, 0);
2422 if (IS_ERR(trans)) {
2423 ret = PTR_ERR(trans);
2424 btrfs_std_error(root->fs_info, ret);
2431 * step two, delete the device extents and the
2432 * chunk tree entries
2434 read_lock(&em_tree->lock);
2435 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2436 read_unlock(&em_tree->lock);
2438 BUG_ON(!em || em->start > chunk_offset ||
2439 em->start + em->len < chunk_offset);
2440 map = (struct map_lookup *)em->bdev;
2442 for (i = 0; i < map->num_stripes; i++) {
2443 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2444 map->stripes[i].physical);
2447 if (map->stripes[i].dev) {
2448 ret = btrfs_update_device(trans, map->stripes[i].dev);
2452 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2457 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2459 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2460 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2464 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2467 write_lock(&em_tree->lock);
2468 remove_extent_mapping(em_tree, em);
2469 write_unlock(&em_tree->lock);
2474 /* once for the tree */
2475 free_extent_map(em);
2477 free_extent_map(em);
2479 unlock_chunks(root);
2480 btrfs_end_transaction(trans, root);
2484 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2486 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2487 struct btrfs_path *path;
2488 struct extent_buffer *leaf;
2489 struct btrfs_chunk *chunk;
2490 struct btrfs_key key;
2491 struct btrfs_key found_key;
2492 u64 chunk_tree = chunk_root->root_key.objectid;
2494 bool retried = false;
2498 path = btrfs_alloc_path();
2503 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2504 key.offset = (u64)-1;
2505 key.type = BTRFS_CHUNK_ITEM_KEY;
2508 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2511 BUG_ON(ret == 0); /* Corruption */
2513 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2520 leaf = path->nodes[0];
2521 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2523 chunk = btrfs_item_ptr(leaf, path->slots[0],
2524 struct btrfs_chunk);
2525 chunk_type = btrfs_chunk_type(leaf, chunk);
2526 btrfs_release_path(path);
2528 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2529 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2538 if (found_key.offset == 0)
2540 key.offset = found_key.offset - 1;
2543 if (failed && !retried) {
2547 } else if (failed && retried) {
2552 btrfs_free_path(path);
2556 static int insert_balance_item(struct btrfs_root *root,
2557 struct btrfs_balance_control *bctl)
2559 struct btrfs_trans_handle *trans;
2560 struct btrfs_balance_item *item;
2561 struct btrfs_disk_balance_args disk_bargs;
2562 struct btrfs_path *path;
2563 struct extent_buffer *leaf;
2564 struct btrfs_key key;
2567 path = btrfs_alloc_path();
2571 trans = btrfs_start_transaction(root, 0);
2572 if (IS_ERR(trans)) {
2573 btrfs_free_path(path);
2574 return PTR_ERR(trans);
2577 key.objectid = BTRFS_BALANCE_OBJECTID;
2578 key.type = BTRFS_BALANCE_ITEM_KEY;
2581 ret = btrfs_insert_empty_item(trans, root, path, &key,
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2589 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2591 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2592 btrfs_set_balance_data(leaf, item, &disk_bargs);
2593 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2594 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2595 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2596 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2598 btrfs_set_balance_flags(leaf, item, bctl->flags);
2600 btrfs_mark_buffer_dirty(leaf);
2602 btrfs_free_path(path);
2603 err = btrfs_commit_transaction(trans, root);
2609 static int del_balance_item(struct btrfs_root *root)
2611 struct btrfs_trans_handle *trans;
2612 struct btrfs_path *path;
2613 struct btrfs_key key;
2616 path = btrfs_alloc_path();
2620 trans = btrfs_start_transaction(root, 0);
2621 if (IS_ERR(trans)) {
2622 btrfs_free_path(path);
2623 return PTR_ERR(trans);
2626 key.objectid = BTRFS_BALANCE_OBJECTID;
2627 key.type = BTRFS_BALANCE_ITEM_KEY;
2630 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2638 ret = btrfs_del_item(trans, root, path);
2640 btrfs_free_path(path);
2641 err = btrfs_commit_transaction(trans, root);
2648 * This is a heuristic used to reduce the number of chunks balanced on
2649 * resume after balance was interrupted.
2651 static void update_balance_args(struct btrfs_balance_control *bctl)
2654 * Turn on soft mode for chunk types that were being converted.
2656 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2657 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2658 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2659 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2660 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2661 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2664 * Turn on usage filter if is not already used. The idea is
2665 * that chunks that we have already balanced should be
2666 * reasonably full. Don't do it for chunks that are being
2667 * converted - that will keep us from relocating unconverted
2668 * (albeit full) chunks.
2670 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2671 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2672 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2673 bctl->data.usage = 90;
2675 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2676 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2677 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2678 bctl->sys.usage = 90;
2680 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2681 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2682 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2683 bctl->meta.usage = 90;
2688 * Should be called with both balance and volume mutexes held to
2689 * serialize other volume operations (add_dev/rm_dev/resize) with
2690 * restriper. Same goes for unset_balance_control.
2692 static void set_balance_control(struct btrfs_balance_control *bctl)
2694 struct btrfs_fs_info *fs_info = bctl->fs_info;
2696 BUG_ON(fs_info->balance_ctl);
2698 spin_lock(&fs_info->balance_lock);
2699 fs_info->balance_ctl = bctl;
2700 spin_unlock(&fs_info->balance_lock);
2703 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2705 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2707 BUG_ON(!fs_info->balance_ctl);
2709 spin_lock(&fs_info->balance_lock);
2710 fs_info->balance_ctl = NULL;
2711 spin_unlock(&fs_info->balance_lock);
2717 * Balance filters. Return 1 if chunk should be filtered out
2718 * (should not be balanced).
2720 static int chunk_profiles_filter(u64 chunk_type,
2721 struct btrfs_balance_args *bargs)
2723 chunk_type = chunk_to_extended(chunk_type) &
2724 BTRFS_EXTENDED_PROFILE_MASK;
2726 if (bargs->profiles & chunk_type)
2732 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2733 struct btrfs_balance_args *bargs)
2735 struct btrfs_block_group_cache *cache;
2736 u64 chunk_used, user_thresh;
2739 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2740 chunk_used = btrfs_block_group_used(&cache->item);
2742 if (bargs->usage == 0)
2744 else if (bargs->usage > 100)
2745 user_thresh = cache->key.offset;
2747 user_thresh = div_factor_fine(cache->key.offset,
2750 if (chunk_used < user_thresh)
2753 btrfs_put_block_group(cache);
2757 static int chunk_devid_filter(struct extent_buffer *leaf,
2758 struct btrfs_chunk *chunk,
2759 struct btrfs_balance_args *bargs)
2761 struct btrfs_stripe *stripe;
2762 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2765 for (i = 0; i < num_stripes; i++) {
2766 stripe = btrfs_stripe_nr(chunk, i);
2767 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2774 /* [pstart, pend) */
2775 static int chunk_drange_filter(struct extent_buffer *leaf,
2776 struct btrfs_chunk *chunk,
2778 struct btrfs_balance_args *bargs)
2780 struct btrfs_stripe *stripe;
2781 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2787 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2790 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2791 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2792 factor = num_stripes / 2;
2793 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2794 factor = num_stripes - 1;
2795 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2796 factor = num_stripes - 2;
2798 factor = num_stripes;
2801 for (i = 0; i < num_stripes; i++) {
2802 stripe = btrfs_stripe_nr(chunk, i);
2803 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2806 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2807 stripe_length = btrfs_chunk_length(leaf, chunk);
2808 do_div(stripe_length, factor);
2810 if (stripe_offset < bargs->pend &&
2811 stripe_offset + stripe_length > bargs->pstart)
2818 /* [vstart, vend) */
2819 static int chunk_vrange_filter(struct extent_buffer *leaf,
2820 struct btrfs_chunk *chunk,
2822 struct btrfs_balance_args *bargs)
2824 if (chunk_offset < bargs->vend &&
2825 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2826 /* at least part of the chunk is inside this vrange */
2832 static int chunk_soft_convert_filter(u64 chunk_type,
2833 struct btrfs_balance_args *bargs)
2835 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2838 chunk_type = chunk_to_extended(chunk_type) &
2839 BTRFS_EXTENDED_PROFILE_MASK;
2841 if (bargs->target == chunk_type)
2847 static int should_balance_chunk(struct btrfs_root *root,
2848 struct extent_buffer *leaf,
2849 struct btrfs_chunk *chunk, u64 chunk_offset)
2851 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2852 struct btrfs_balance_args *bargs = NULL;
2853 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2856 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2857 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2861 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2862 bargs = &bctl->data;
2863 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2865 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2866 bargs = &bctl->meta;
2868 /* profiles filter */
2869 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2870 chunk_profiles_filter(chunk_type, bargs)) {
2875 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2876 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2881 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2882 chunk_devid_filter(leaf, chunk, bargs)) {
2886 /* drange filter, makes sense only with devid filter */
2887 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2888 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2893 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2894 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2898 /* soft profile changing mode */
2899 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2900 chunk_soft_convert_filter(chunk_type, bargs)) {
2907 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2909 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2910 struct btrfs_root *chunk_root = fs_info->chunk_root;
2911 struct btrfs_root *dev_root = fs_info->dev_root;
2912 struct list_head *devices;
2913 struct btrfs_device *device;
2916 struct btrfs_chunk *chunk;
2917 struct btrfs_path *path;
2918 struct btrfs_key key;
2919 struct btrfs_key found_key;
2920 struct btrfs_trans_handle *trans;
2921 struct extent_buffer *leaf;
2924 int enospc_errors = 0;
2925 bool counting = true;
2927 /* step one make some room on all the devices */
2928 devices = &fs_info->fs_devices->devices;
2929 list_for_each_entry(device, devices, dev_list) {
2930 old_size = device->total_bytes;
2931 size_to_free = div_factor(old_size, 1);
2932 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2933 if (!device->writeable ||
2934 device->total_bytes - device->bytes_used > size_to_free ||
2935 device->is_tgtdev_for_dev_replace)
2938 ret = btrfs_shrink_device(device, old_size - size_to_free);
2943 trans = btrfs_start_transaction(dev_root, 0);
2944 BUG_ON(IS_ERR(trans));
2946 ret = btrfs_grow_device(trans, device, old_size);
2949 btrfs_end_transaction(trans, dev_root);
2952 /* step two, relocate all the chunks */
2953 path = btrfs_alloc_path();
2959 /* zero out stat counters */
2960 spin_lock(&fs_info->balance_lock);
2961 memset(&bctl->stat, 0, sizeof(bctl->stat));
2962 spin_unlock(&fs_info->balance_lock);
2964 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2965 key.offset = (u64)-1;
2966 key.type = BTRFS_CHUNK_ITEM_KEY;
2969 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2970 atomic_read(&fs_info->balance_cancel_req)) {
2975 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2980 * this shouldn't happen, it means the last relocate
2984 BUG(); /* FIXME break ? */
2986 ret = btrfs_previous_item(chunk_root, path, 0,
2987 BTRFS_CHUNK_ITEM_KEY);
2993 leaf = path->nodes[0];
2994 slot = path->slots[0];
2995 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2997 if (found_key.objectid != key.objectid)
3000 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3003 spin_lock(&fs_info->balance_lock);
3004 bctl->stat.considered++;
3005 spin_unlock(&fs_info->balance_lock);
3008 ret = should_balance_chunk(chunk_root, leaf, chunk,
3010 btrfs_release_path(path);
3015 spin_lock(&fs_info->balance_lock);
3016 bctl->stat.expected++;
3017 spin_unlock(&fs_info->balance_lock);
3021 ret = btrfs_relocate_chunk(chunk_root,
3022 chunk_root->root_key.objectid,
3025 if (ret && ret != -ENOSPC)
3027 if (ret == -ENOSPC) {
3030 spin_lock(&fs_info->balance_lock);
3031 bctl->stat.completed++;
3032 spin_unlock(&fs_info->balance_lock);
3035 if (found_key.offset == 0)
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_ERR_OR_ZERO(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 = NULL;
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))
3504 read_extent_buffer(eb, &root_item,
3505 btrfs_item_ptr_offset(eb, slot),
3506 (int)sizeof(root_item));
3507 if (btrfs_root_refs(&root_item) == 0)
3510 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3511 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3515 btrfs_release_path(path);
3517 * 1 - subvol uuid item
3518 * 1 - received_subvol uuid item
3520 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3521 if (IS_ERR(trans)) {
3522 ret = PTR_ERR(trans);
3530 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3531 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3533 BTRFS_UUID_KEY_SUBVOL,
3536 pr_warn("btrfs: uuid_tree_add failed %d\n",
3542 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
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",
3556 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3562 btrfs_release_path(path);
3563 if (key.offset < (u64)-1) {
3565 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3567 key.type = BTRFS_ROOT_ITEM_KEY;
3568 } else if (key.objectid < (u64)-1) {
3570 key.type = BTRFS_ROOT_ITEM_KEY;
3579 btrfs_free_path(path);
3580 if (trans && !IS_ERR(trans))
3581 btrfs_end_transaction(trans, fs_info->uuid_root);
3583 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3585 fs_info->update_uuid_tree_gen = 1;
3586 up(&fs_info->uuid_tree_rescan_sem);
3591 * Callback for btrfs_uuid_tree_iterate().
3593 * 0 check succeeded, the entry is not outdated.
3594 * < 0 if an error occured.
3595 * > 0 if the check failed, which means the caller shall remove the entry.
3597 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3598 u8 *uuid, u8 type, u64 subid)
3600 struct btrfs_key key;
3602 struct btrfs_root *subvol_root;
3604 if (type != BTRFS_UUID_KEY_SUBVOL &&
3605 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3608 key.objectid = subid;
3609 key.type = BTRFS_ROOT_ITEM_KEY;
3610 key.offset = (u64)-1;
3611 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3612 if (IS_ERR(subvol_root)) {
3613 ret = PTR_ERR(subvol_root);
3620 case BTRFS_UUID_KEY_SUBVOL:
3621 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3624 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3625 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3635 static int btrfs_uuid_rescan_kthread(void *data)
3637 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3641 * 1st step is to iterate through the existing UUID tree and
3642 * to delete all entries that contain outdated data.
3643 * 2nd step is to add all missing entries to the UUID tree.
3645 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3647 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3648 up(&fs_info->uuid_tree_rescan_sem);
3651 return btrfs_uuid_scan_kthread(data);
3654 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3656 struct btrfs_trans_handle *trans;
3657 struct btrfs_root *tree_root = fs_info->tree_root;
3658 struct btrfs_root *uuid_root;
3659 struct task_struct *task;
3666 trans = btrfs_start_transaction(tree_root, 2);
3668 return PTR_ERR(trans);
3670 uuid_root = btrfs_create_tree(trans, fs_info,
3671 BTRFS_UUID_TREE_OBJECTID);
3672 if (IS_ERR(uuid_root)) {
3673 btrfs_abort_transaction(trans, tree_root,
3674 PTR_ERR(uuid_root));
3675 return PTR_ERR(uuid_root);
3678 fs_info->uuid_root = uuid_root;
3680 ret = btrfs_commit_transaction(trans, tree_root);
3684 down(&fs_info->uuid_tree_rescan_sem);
3685 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3687 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3688 pr_warn("btrfs: failed to start uuid_scan task\n");
3689 up(&fs_info->uuid_tree_rescan_sem);
3690 return PTR_ERR(task);
3696 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3698 struct task_struct *task;
3700 down(&fs_info->uuid_tree_rescan_sem);
3701 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3703 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3704 pr_warn("btrfs: failed to start uuid_rescan task\n");
3705 up(&fs_info->uuid_tree_rescan_sem);
3706 return PTR_ERR(task);
3713 * shrinking a device means finding all of the device extents past
3714 * the new size, and then following the back refs to the chunks.
3715 * The chunk relocation code actually frees the device extent
3717 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3719 struct btrfs_trans_handle *trans;
3720 struct btrfs_root *root = device->dev_root;
3721 struct btrfs_dev_extent *dev_extent = NULL;
3722 struct btrfs_path *path;
3730 bool retried = false;
3731 struct extent_buffer *l;
3732 struct btrfs_key key;
3733 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3734 u64 old_total = btrfs_super_total_bytes(super_copy);
3735 u64 old_size = device->total_bytes;
3736 u64 diff = device->total_bytes - new_size;
3738 if (device->is_tgtdev_for_dev_replace)
3741 path = btrfs_alloc_path();
3749 device->total_bytes = new_size;
3750 if (device->writeable) {
3751 device->fs_devices->total_rw_bytes -= diff;
3752 spin_lock(&root->fs_info->free_chunk_lock);
3753 root->fs_info->free_chunk_space -= diff;
3754 spin_unlock(&root->fs_info->free_chunk_lock);
3756 unlock_chunks(root);
3759 key.objectid = device->devid;
3760 key.offset = (u64)-1;
3761 key.type = BTRFS_DEV_EXTENT_KEY;
3764 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3768 ret = btrfs_previous_item(root, path, 0, key.type);
3773 btrfs_release_path(path);
3778 slot = path->slots[0];
3779 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3781 if (key.objectid != device->devid) {
3782 btrfs_release_path(path);
3786 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3787 length = btrfs_dev_extent_length(l, dev_extent);
3789 if (key.offset + length <= new_size) {
3790 btrfs_release_path(path);
3794 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3795 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3796 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3797 btrfs_release_path(path);
3799 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3801 if (ret && ret != -ENOSPC)
3805 } while (key.offset-- > 0);
3807 if (failed && !retried) {
3811 } else if (failed && retried) {
3815 device->total_bytes = old_size;
3816 if (device->writeable)
3817 device->fs_devices->total_rw_bytes += diff;
3818 spin_lock(&root->fs_info->free_chunk_lock);
3819 root->fs_info->free_chunk_space += diff;
3820 spin_unlock(&root->fs_info->free_chunk_lock);
3821 unlock_chunks(root);
3825 /* Shrinking succeeded, else we would be at "done". */
3826 trans = btrfs_start_transaction(root, 0);
3827 if (IS_ERR(trans)) {
3828 ret = PTR_ERR(trans);
3834 device->disk_total_bytes = new_size;
3835 /* Now btrfs_update_device() will change the on-disk size. */
3836 ret = btrfs_update_device(trans, device);
3838 unlock_chunks(root);
3839 btrfs_end_transaction(trans, root);
3842 WARN_ON(diff > old_total);
3843 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3844 unlock_chunks(root);
3845 btrfs_end_transaction(trans, root);
3847 btrfs_free_path(path);
3851 static int btrfs_add_system_chunk(struct btrfs_root *root,
3852 struct btrfs_key *key,
3853 struct btrfs_chunk *chunk, int item_size)
3855 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3856 struct btrfs_disk_key disk_key;
3860 array_size = btrfs_super_sys_array_size(super_copy);
3861 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3864 ptr = super_copy->sys_chunk_array + array_size;
3865 btrfs_cpu_key_to_disk(&disk_key, key);
3866 memcpy(ptr, &disk_key, sizeof(disk_key));
3867 ptr += sizeof(disk_key);
3868 memcpy(ptr, chunk, item_size);
3869 item_size += sizeof(disk_key);
3870 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3875 * sort the devices in descending order by max_avail, total_avail
3877 static int btrfs_cmp_device_info(const void *a, const void *b)
3879 const struct btrfs_device_info *di_a = a;
3880 const struct btrfs_device_info *di_b = b;
3882 if (di_a->max_avail > di_b->max_avail)
3884 if (di_a->max_avail < di_b->max_avail)
3886 if (di_a->total_avail > di_b->total_avail)
3888 if (di_a->total_avail < di_b->total_avail)
3893 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3894 [BTRFS_RAID_RAID10] = {
3897 .devs_max = 0, /* 0 == as many as possible */
3899 .devs_increment = 2,
3902 [BTRFS_RAID_RAID1] = {
3907 .devs_increment = 2,
3910 [BTRFS_RAID_DUP] = {
3915 .devs_increment = 1,
3918 [BTRFS_RAID_RAID0] = {
3923 .devs_increment = 1,
3926 [BTRFS_RAID_SINGLE] = {
3931 .devs_increment = 1,
3934 [BTRFS_RAID_RAID5] = {
3939 .devs_increment = 1,
3942 [BTRFS_RAID_RAID6] = {
3947 .devs_increment = 1,
3952 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3954 /* TODO allow them to set a preferred stripe size */
3958 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3960 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3963 btrfs_set_fs_incompat(info, RAID56);
3966 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3967 struct btrfs_root *extent_root, u64 start,
3970 struct btrfs_fs_info *info = extent_root->fs_info;
3971 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3972 struct list_head *cur;
3973 struct map_lookup *map = NULL;
3974 struct extent_map_tree *em_tree;
3975 struct extent_map *em;
3976 struct btrfs_device_info *devices_info = NULL;
3978 int num_stripes; /* total number of stripes to allocate */
3979 int data_stripes; /* number of stripes that count for
3981 int sub_stripes; /* sub_stripes info for map */
3982 int dev_stripes; /* stripes per dev */
3983 int devs_max; /* max devs to use */
3984 int devs_min; /* min devs needed */
3985 int devs_increment; /* ndevs has to be a multiple of this */
3986 int ncopies; /* how many copies to data has */
3988 u64 max_stripe_size;
3992 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3998 BUG_ON(!alloc_profile_is_valid(type, 0));
4000 if (list_empty(&fs_devices->alloc_list))
4003 index = __get_raid_index(type);
4005 sub_stripes = btrfs_raid_array[index].sub_stripes;
4006 dev_stripes = btrfs_raid_array[index].dev_stripes;
4007 devs_max = btrfs_raid_array[index].devs_max;
4008 devs_min = btrfs_raid_array[index].devs_min;
4009 devs_increment = btrfs_raid_array[index].devs_increment;
4010 ncopies = btrfs_raid_array[index].ncopies;
4012 if (type & BTRFS_BLOCK_GROUP_DATA) {
4013 max_stripe_size = 1024 * 1024 * 1024;
4014 max_chunk_size = 10 * max_stripe_size;
4015 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4016 /* for larger filesystems, use larger metadata chunks */
4017 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4018 max_stripe_size = 1024 * 1024 * 1024;
4020 max_stripe_size = 256 * 1024 * 1024;
4021 max_chunk_size = max_stripe_size;
4022 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4023 max_stripe_size = 32 * 1024 * 1024;
4024 max_chunk_size = 2 * max_stripe_size;
4026 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4031 /* we don't want a chunk larger than 10% of writeable space */
4032 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4035 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4040 cur = fs_devices->alloc_list.next;
4043 * in the first pass through the devices list, we gather information
4044 * about the available holes on each device.
4047 while (cur != &fs_devices->alloc_list) {
4048 struct btrfs_device *device;
4052 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4056 if (!device->writeable) {
4058 "btrfs: read-only device in alloc_list\n");
4062 if (!device->in_fs_metadata ||
4063 device->is_tgtdev_for_dev_replace)
4066 if (device->total_bytes > device->bytes_used)
4067 total_avail = device->total_bytes - device->bytes_used;
4071 /* If there is no space on this device, skip it. */
4072 if (total_avail == 0)
4075 ret = find_free_dev_extent(trans, device,
4076 max_stripe_size * dev_stripes,
4077 &dev_offset, &max_avail);
4078 if (ret && ret != -ENOSPC)
4082 max_avail = max_stripe_size * dev_stripes;
4084 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4087 if (ndevs == fs_devices->rw_devices) {
4088 WARN(1, "%s: found more than %llu devices\n",
4089 __func__, fs_devices->rw_devices);
4092 devices_info[ndevs].dev_offset = dev_offset;
4093 devices_info[ndevs].max_avail = max_avail;
4094 devices_info[ndevs].total_avail = total_avail;
4095 devices_info[ndevs].dev = device;
4100 * now sort the devices by hole size / available space
4102 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4103 btrfs_cmp_device_info, NULL);
4105 /* round down to number of usable stripes */
4106 ndevs -= ndevs % devs_increment;
4108 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4113 if (devs_max && ndevs > devs_max)
4116 * the primary goal is to maximize the number of stripes, so use as many
4117 * devices as possible, even if the stripes are not maximum sized.
4119 stripe_size = devices_info[ndevs-1].max_avail;
4120 num_stripes = ndevs * dev_stripes;
4123 * this will have to be fixed for RAID1 and RAID10 over
4126 data_stripes = num_stripes / ncopies;
4128 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4129 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4130 btrfs_super_stripesize(info->super_copy));
4131 data_stripes = num_stripes - 1;
4133 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4134 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4135 btrfs_super_stripesize(info->super_copy));
4136 data_stripes = num_stripes - 2;
4140 * Use the number of data stripes to figure out how big this chunk
4141 * is really going to be in terms of logical address space,
4142 * and compare that answer with the max chunk size
4144 if (stripe_size * data_stripes > max_chunk_size) {
4145 u64 mask = (1ULL << 24) - 1;
4146 stripe_size = max_chunk_size;
4147 do_div(stripe_size, data_stripes);
4149 /* bump the answer up to a 16MB boundary */
4150 stripe_size = (stripe_size + mask) & ~mask;
4152 /* but don't go higher than the limits we found
4153 * while searching for free extents
4155 if (stripe_size > devices_info[ndevs-1].max_avail)
4156 stripe_size = devices_info[ndevs-1].max_avail;
4159 do_div(stripe_size, dev_stripes);
4161 /* align to BTRFS_STRIPE_LEN */
4162 do_div(stripe_size, raid_stripe_len);
4163 stripe_size *= raid_stripe_len;
4165 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4170 map->num_stripes = num_stripes;
4172 for (i = 0; i < ndevs; ++i) {
4173 for (j = 0; j < dev_stripes; ++j) {
4174 int s = i * dev_stripes + j;
4175 map->stripes[s].dev = devices_info[i].dev;
4176 map->stripes[s].physical = devices_info[i].dev_offset +
4180 map->sector_size = extent_root->sectorsize;
4181 map->stripe_len = raid_stripe_len;
4182 map->io_align = raid_stripe_len;
4183 map->io_width = raid_stripe_len;
4185 map->sub_stripes = sub_stripes;
4187 num_bytes = stripe_size * data_stripes;
4189 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4191 em = alloc_extent_map();
4196 em->bdev = (struct block_device *)map;
4198 em->len = num_bytes;
4199 em->block_start = 0;
4200 em->block_len = em->len;
4201 em->orig_block_len = stripe_size;
4203 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4204 write_lock(&em_tree->lock);
4205 ret = add_extent_mapping(em_tree, em, 0);
4207 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4208 atomic_inc(&em->refs);
4210 write_unlock(&em_tree->lock);
4212 free_extent_map(em);
4216 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4217 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4220 goto error_del_extent;
4222 free_extent_map(em);
4223 check_raid56_incompat_flag(extent_root->fs_info, type);
4225 kfree(devices_info);
4229 write_lock(&em_tree->lock);
4230 remove_extent_mapping(em_tree, em);
4231 write_unlock(&em_tree->lock);
4233 /* One for our allocation */
4234 free_extent_map(em);
4235 /* One for the tree reference */
4236 free_extent_map(em);
4239 kfree(devices_info);
4243 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4244 struct btrfs_root *extent_root,
4245 u64 chunk_offset, u64 chunk_size)
4247 struct btrfs_key key;
4248 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4249 struct btrfs_device *device;
4250 struct btrfs_chunk *chunk;
4251 struct btrfs_stripe *stripe;
4252 struct extent_map_tree *em_tree;
4253 struct extent_map *em;
4254 struct map_lookup *map;
4261 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4262 read_lock(&em_tree->lock);
4263 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4264 read_unlock(&em_tree->lock);
4267 btrfs_crit(extent_root->fs_info, "unable to find logical "
4268 "%Lu len %Lu", chunk_offset, chunk_size);
4272 if (em->start != chunk_offset || em->len != chunk_size) {
4273 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4274 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4275 chunk_size, em->start, em->len);
4276 free_extent_map(em);
4280 map = (struct map_lookup *)em->bdev;
4281 item_size = btrfs_chunk_item_size(map->num_stripes);
4282 stripe_size = em->orig_block_len;
4284 chunk = kzalloc(item_size, GFP_NOFS);
4290 for (i = 0; i < map->num_stripes; i++) {
4291 device = map->stripes[i].dev;
4292 dev_offset = map->stripes[i].physical;
4294 device->bytes_used += stripe_size;
4295 ret = btrfs_update_device(trans, device);
4298 ret = btrfs_alloc_dev_extent(trans, device,
4299 chunk_root->root_key.objectid,
4300 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4301 chunk_offset, dev_offset,
4307 spin_lock(&extent_root->fs_info->free_chunk_lock);
4308 extent_root->fs_info->free_chunk_space -= (stripe_size *
4310 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4312 stripe = &chunk->stripe;
4313 for (i = 0; i < map->num_stripes; i++) {
4314 device = map->stripes[i].dev;
4315 dev_offset = map->stripes[i].physical;
4317 btrfs_set_stack_stripe_devid(stripe, device->devid);
4318 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4319 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4323 btrfs_set_stack_chunk_length(chunk, chunk_size);
4324 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4325 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4326 btrfs_set_stack_chunk_type(chunk, map->type);
4327 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4328 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4329 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4330 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4331 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4333 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4334 key.type = BTRFS_CHUNK_ITEM_KEY;
4335 key.offset = chunk_offset;
4337 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4338 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4340 * TODO: Cleanup of inserted chunk root in case of
4343 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4349 free_extent_map(em);
4354 * Chunk allocation falls into two parts. The first part does works
4355 * that make the new allocated chunk useable, but not do any operation
4356 * that modifies the chunk tree. The second part does the works that
4357 * require modifying the chunk tree. This division is important for the
4358 * bootstrap process of adding storage to a seed btrfs.
4360 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4361 struct btrfs_root *extent_root, u64 type)
4365 chunk_offset = find_next_chunk(extent_root->fs_info);
4366 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4369 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4370 struct btrfs_root *root,
4371 struct btrfs_device *device)
4374 u64 sys_chunk_offset;
4376 struct btrfs_fs_info *fs_info = root->fs_info;
4377 struct btrfs_root *extent_root = fs_info->extent_root;
4380 chunk_offset = find_next_chunk(fs_info);
4381 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4382 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4387 sys_chunk_offset = find_next_chunk(root->fs_info);
4388 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4389 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4392 btrfs_abort_transaction(trans, root, ret);
4396 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4398 btrfs_abort_transaction(trans, root, ret);
4403 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4405 struct extent_map *em;
4406 struct map_lookup *map;
4407 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4411 read_lock(&map_tree->map_tree.lock);
4412 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4413 read_unlock(&map_tree->map_tree.lock);
4417 if (btrfs_test_opt(root, DEGRADED)) {
4418 free_extent_map(em);
4422 map = (struct map_lookup *)em->bdev;
4423 for (i = 0; i < map->num_stripes; i++) {
4424 if (!map->stripes[i].dev->writeable) {
4429 free_extent_map(em);
4433 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4435 extent_map_tree_init(&tree->map_tree);
4438 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4440 struct extent_map *em;
4443 write_lock(&tree->map_tree.lock);
4444 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4446 remove_extent_mapping(&tree->map_tree, em);
4447 write_unlock(&tree->map_tree.lock);
4452 free_extent_map(em);
4453 /* once for the tree */
4454 free_extent_map(em);
4458 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4460 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4461 struct extent_map *em;
4462 struct map_lookup *map;
4463 struct extent_map_tree *em_tree = &map_tree->map_tree;
4466 read_lock(&em_tree->lock);
4467 em = lookup_extent_mapping(em_tree, logical, len);
4468 read_unlock(&em_tree->lock);
4471 * We could return errors for these cases, but that could get ugly and
4472 * we'd probably do the same thing which is just not do anything else
4473 * and exit, so return 1 so the callers don't try to use other copies.
4476 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4481 if (em->start > logical || em->start + em->len < logical) {
4482 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4483 "%Lu-%Lu\n", logical, logical+len, em->start,
4484 em->start + em->len);
4488 map = (struct map_lookup *)em->bdev;
4489 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4490 ret = map->num_stripes;
4491 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4492 ret = map->sub_stripes;
4493 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4495 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4499 free_extent_map(em);
4501 btrfs_dev_replace_lock(&fs_info->dev_replace);
4502 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4504 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4509 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4510 struct btrfs_mapping_tree *map_tree,
4513 struct extent_map *em;
4514 struct map_lookup *map;
4515 struct extent_map_tree *em_tree = &map_tree->map_tree;
4516 unsigned long len = root->sectorsize;
4518 read_lock(&em_tree->lock);
4519 em = lookup_extent_mapping(em_tree, logical, len);
4520 read_unlock(&em_tree->lock);
4523 BUG_ON(em->start > logical || em->start + em->len < logical);
4524 map = (struct map_lookup *)em->bdev;
4525 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4526 BTRFS_BLOCK_GROUP_RAID6)) {
4527 len = map->stripe_len * nr_data_stripes(map);
4529 free_extent_map(em);
4533 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4534 u64 logical, u64 len, int mirror_num)
4536 struct extent_map *em;
4537 struct map_lookup *map;
4538 struct extent_map_tree *em_tree = &map_tree->map_tree;
4541 read_lock(&em_tree->lock);
4542 em = lookup_extent_mapping(em_tree, logical, len);
4543 read_unlock(&em_tree->lock);
4546 BUG_ON(em->start > logical || em->start + em->len < logical);
4547 map = (struct map_lookup *)em->bdev;
4548 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4549 BTRFS_BLOCK_GROUP_RAID6))
4551 free_extent_map(em);
4555 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4556 struct map_lookup *map, int first, int num,
4557 int optimal, int dev_replace_is_ongoing)
4561 struct btrfs_device *srcdev;
4563 if (dev_replace_is_ongoing &&
4564 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4565 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4566 srcdev = fs_info->dev_replace.srcdev;
4571 * try to avoid the drive that is the source drive for a
4572 * dev-replace procedure, only choose it if no other non-missing
4573 * mirror is available
4575 for (tolerance = 0; tolerance < 2; tolerance++) {
4576 if (map->stripes[optimal].dev->bdev &&
4577 (tolerance || map->stripes[optimal].dev != srcdev))
4579 for (i = first; i < first + num; i++) {
4580 if (map->stripes[i].dev->bdev &&
4581 (tolerance || map->stripes[i].dev != srcdev))
4586 /* we couldn't find one that doesn't fail. Just return something
4587 * and the io error handling code will clean up eventually
4592 static inline int parity_smaller(u64 a, u64 b)
4597 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4598 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4600 struct btrfs_bio_stripe s;
4607 for (i = 0; i < bbio->num_stripes - 1; i++) {
4608 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4609 s = bbio->stripes[i];
4611 bbio->stripes[i] = bbio->stripes[i+1];
4612 raid_map[i] = raid_map[i+1];
4613 bbio->stripes[i+1] = s;
4621 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4622 u64 logical, u64 *length,
4623 struct btrfs_bio **bbio_ret,
4624 int mirror_num, u64 **raid_map_ret)
4626 struct extent_map *em;
4627 struct map_lookup *map;
4628 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4629 struct extent_map_tree *em_tree = &map_tree->map_tree;
4632 u64 stripe_end_offset;
4637 u64 *raid_map = NULL;
4643 struct btrfs_bio *bbio = NULL;
4644 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4645 int dev_replace_is_ongoing = 0;
4646 int num_alloc_stripes;
4647 int patch_the_first_stripe_for_dev_replace = 0;
4648 u64 physical_to_patch_in_first_stripe = 0;
4649 u64 raid56_full_stripe_start = (u64)-1;
4651 read_lock(&em_tree->lock);
4652 em = lookup_extent_mapping(em_tree, logical, *length);
4653 read_unlock(&em_tree->lock);
4656 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4661 if (em->start > logical || em->start + em->len < logical) {
4662 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4663 "found %Lu-%Lu\n", logical, em->start,
4664 em->start + em->len);
4668 map = (struct map_lookup *)em->bdev;
4669 offset = logical - em->start;
4671 stripe_len = map->stripe_len;
4674 * stripe_nr counts the total number of stripes we have to stride
4675 * to get to this block
4677 do_div(stripe_nr, stripe_len);
4679 stripe_offset = stripe_nr * stripe_len;
4680 BUG_ON(offset < stripe_offset);
4682 /* stripe_offset is the offset of this block in its stripe*/
4683 stripe_offset = offset - stripe_offset;
4685 /* if we're here for raid56, we need to know the stripe aligned start */
4686 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4687 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4688 raid56_full_stripe_start = offset;
4690 /* allow a write of a full stripe, but make sure we don't
4691 * allow straddling of stripes
4693 do_div(raid56_full_stripe_start, full_stripe_len);
4694 raid56_full_stripe_start *= full_stripe_len;
4697 if (rw & REQ_DISCARD) {
4698 /* we don't discard raid56 yet */
4700 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4704 *length = min_t(u64, em->len - offset, *length);
4705 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4707 /* For writes to RAID[56], allow a full stripeset across all disks.
4708 For other RAID types and for RAID[56] reads, just allow a single
4709 stripe (on a single disk). */
4710 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4712 max_len = stripe_len * nr_data_stripes(map) -
4713 (offset - raid56_full_stripe_start);
4715 /* we limit the length of each bio to what fits in a stripe */
4716 max_len = stripe_len - stripe_offset;
4718 *length = min_t(u64, em->len - offset, max_len);
4720 *length = em->len - offset;
4723 /* This is for when we're called from btrfs_merge_bio_hook() and all
4724 it cares about is the length */
4728 btrfs_dev_replace_lock(dev_replace);
4729 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4730 if (!dev_replace_is_ongoing)
4731 btrfs_dev_replace_unlock(dev_replace);
4733 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4734 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4735 dev_replace->tgtdev != NULL) {
4737 * in dev-replace case, for repair case (that's the only
4738 * case where the mirror is selected explicitly when
4739 * calling btrfs_map_block), blocks left of the left cursor
4740 * can also be read from the target drive.
4741 * For REQ_GET_READ_MIRRORS, the target drive is added as
4742 * the last one to the array of stripes. For READ, it also
4743 * needs to be supported using the same mirror number.
4744 * If the requested block is not left of the left cursor,
4745 * EIO is returned. This can happen because btrfs_num_copies()
4746 * returns one more in the dev-replace case.
4748 u64 tmp_length = *length;
4749 struct btrfs_bio *tmp_bbio = NULL;
4750 int tmp_num_stripes;
4751 u64 srcdev_devid = dev_replace->srcdev->devid;
4752 int index_srcdev = 0;
4754 u64 physical_of_found = 0;
4756 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4757 logical, &tmp_length, &tmp_bbio, 0, NULL);
4759 WARN_ON(tmp_bbio != NULL);
4763 tmp_num_stripes = tmp_bbio->num_stripes;
4764 if (mirror_num > tmp_num_stripes) {
4766 * REQ_GET_READ_MIRRORS does not contain this
4767 * mirror, that means that the requested area
4768 * is not left of the left cursor
4776 * process the rest of the function using the mirror_num
4777 * of the source drive. Therefore look it up first.
4778 * At the end, patch the device pointer to the one of the
4781 for (i = 0; i < tmp_num_stripes; i++) {
4782 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4784 * In case of DUP, in order to keep it
4785 * simple, only add the mirror with the
4786 * lowest physical address
4789 physical_of_found <=
4790 tmp_bbio->stripes[i].physical)
4795 tmp_bbio->stripes[i].physical;
4800 mirror_num = index_srcdev + 1;
4801 patch_the_first_stripe_for_dev_replace = 1;
4802 physical_to_patch_in_first_stripe = physical_of_found;
4811 } else if (mirror_num > map->num_stripes) {
4817 stripe_nr_orig = stripe_nr;
4818 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4819 do_div(stripe_nr_end, map->stripe_len);
4820 stripe_end_offset = stripe_nr_end * map->stripe_len -
4823 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4824 if (rw & REQ_DISCARD)
4825 num_stripes = min_t(u64, map->num_stripes,
4826 stripe_nr_end - stripe_nr_orig);
4827 stripe_index = do_div(stripe_nr, map->num_stripes);
4828 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4829 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4830 num_stripes = map->num_stripes;
4831 else if (mirror_num)
4832 stripe_index = mirror_num - 1;
4834 stripe_index = find_live_mirror(fs_info, map, 0,
4836 current->pid % map->num_stripes,
4837 dev_replace_is_ongoing);
4838 mirror_num = stripe_index + 1;
4841 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4842 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4843 num_stripes = map->num_stripes;
4844 } else if (mirror_num) {
4845 stripe_index = mirror_num - 1;
4850 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4851 int factor = map->num_stripes / map->sub_stripes;
4853 stripe_index = do_div(stripe_nr, factor);
4854 stripe_index *= map->sub_stripes;
4856 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4857 num_stripes = map->sub_stripes;
4858 else if (rw & REQ_DISCARD)
4859 num_stripes = min_t(u64, map->sub_stripes *
4860 (stripe_nr_end - stripe_nr_orig),
4862 else if (mirror_num)
4863 stripe_index += mirror_num - 1;
4865 int old_stripe_index = stripe_index;
4866 stripe_index = find_live_mirror(fs_info, map,
4868 map->sub_stripes, stripe_index +
4869 current->pid % map->sub_stripes,
4870 dev_replace_is_ongoing);
4871 mirror_num = stripe_index - old_stripe_index + 1;
4874 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4875 BTRFS_BLOCK_GROUP_RAID6)) {
4878 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4882 /* push stripe_nr back to the start of the full stripe */
4883 stripe_nr = raid56_full_stripe_start;
4884 do_div(stripe_nr, stripe_len);
4886 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4888 /* RAID[56] write or recovery. Return all stripes */
4889 num_stripes = map->num_stripes;
4890 max_errors = nr_parity_stripes(map);
4892 raid_map = kmalloc(sizeof(u64) * num_stripes,
4899 /* Work out the disk rotation on this stripe-set */
4901 rot = do_div(tmp, num_stripes);
4903 /* Fill in the logical address of each stripe */
4904 tmp = stripe_nr * nr_data_stripes(map);
4905 for (i = 0; i < nr_data_stripes(map); i++)
4906 raid_map[(i+rot) % num_stripes] =
4907 em->start + (tmp + i) * map->stripe_len;
4909 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4910 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4911 raid_map[(i+rot+1) % num_stripes] =
4914 *length = map->stripe_len;
4919 * Mirror #0 or #1 means the original data block.
4920 * Mirror #2 is RAID5 parity block.
4921 * Mirror #3 is RAID6 Q block.
4923 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4925 stripe_index = nr_data_stripes(map) +
4928 /* We distribute the parity blocks across stripes */
4929 tmp = stripe_nr + stripe_index;
4930 stripe_index = do_div(tmp, map->num_stripes);
4934 * after this do_div call, stripe_nr is the number of stripes
4935 * on this device we have to walk to find the data, and
4936 * stripe_index is the number of our device in the stripe array
4938 stripe_index = do_div(stripe_nr, map->num_stripes);
4939 mirror_num = stripe_index + 1;
4941 BUG_ON(stripe_index >= map->num_stripes);
4943 num_alloc_stripes = num_stripes;
4944 if (dev_replace_is_ongoing) {
4945 if (rw & (REQ_WRITE | REQ_DISCARD))
4946 num_alloc_stripes <<= 1;
4947 if (rw & REQ_GET_READ_MIRRORS)
4948 num_alloc_stripes++;
4950 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4956 atomic_set(&bbio->error, 0);
4958 if (rw & REQ_DISCARD) {
4960 int sub_stripes = 0;
4961 u64 stripes_per_dev = 0;
4962 u32 remaining_stripes = 0;
4963 u32 last_stripe = 0;
4966 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4967 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4970 sub_stripes = map->sub_stripes;
4972 factor = map->num_stripes / sub_stripes;
4973 stripes_per_dev = div_u64_rem(stripe_nr_end -
4976 &remaining_stripes);
4977 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4978 last_stripe *= sub_stripes;
4981 for (i = 0; i < num_stripes; i++) {
4982 bbio->stripes[i].physical =
4983 map->stripes[stripe_index].physical +
4984 stripe_offset + stripe_nr * map->stripe_len;
4985 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4987 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4988 BTRFS_BLOCK_GROUP_RAID10)) {
4989 bbio->stripes[i].length = stripes_per_dev *
4992 if (i / sub_stripes < remaining_stripes)
4993 bbio->stripes[i].length +=
4997 * Special for the first stripe and
5000 * |-------|...|-------|
5004 if (i < sub_stripes)
5005 bbio->stripes[i].length -=
5008 if (stripe_index >= last_stripe &&
5009 stripe_index <= (last_stripe +
5011 bbio->stripes[i].length -=
5014 if (i == sub_stripes - 1)
5017 bbio->stripes[i].length = *length;
5020 if (stripe_index == map->num_stripes) {
5021 /* This could only happen for RAID0/10 */
5027 for (i = 0; i < num_stripes; i++) {
5028 bbio->stripes[i].physical =
5029 map->stripes[stripe_index].physical +
5031 stripe_nr * map->stripe_len;
5032 bbio->stripes[i].dev =
5033 map->stripes[stripe_index].dev;
5038 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5039 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5040 BTRFS_BLOCK_GROUP_RAID10 |
5041 BTRFS_BLOCK_GROUP_RAID5 |
5042 BTRFS_BLOCK_GROUP_DUP)) {
5044 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5049 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5050 dev_replace->tgtdev != NULL) {
5051 int index_where_to_add;
5052 u64 srcdev_devid = dev_replace->srcdev->devid;
5055 * duplicate the write operations while the dev replace
5056 * procedure is running. Since the copying of the old disk
5057 * to the new disk takes place at run time while the
5058 * filesystem is mounted writable, the regular write
5059 * operations to the old disk have to be duplicated to go
5060 * to the new disk as well.
5061 * Note that device->missing is handled by the caller, and
5062 * that the write to the old disk is already set up in the
5065 index_where_to_add = num_stripes;
5066 for (i = 0; i < num_stripes; i++) {
5067 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5068 /* write to new disk, too */
5069 struct btrfs_bio_stripe *new =
5070 bbio->stripes + index_where_to_add;
5071 struct btrfs_bio_stripe *old =
5074 new->physical = old->physical;
5075 new->length = old->length;
5076 new->dev = dev_replace->tgtdev;
5077 index_where_to_add++;
5081 num_stripes = index_where_to_add;
5082 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5083 dev_replace->tgtdev != NULL) {
5084 u64 srcdev_devid = dev_replace->srcdev->devid;
5085 int index_srcdev = 0;
5087 u64 physical_of_found = 0;
5090 * During the dev-replace procedure, the target drive can
5091 * also be used to read data in case it is needed to repair
5092 * a corrupt block elsewhere. This is possible if the
5093 * requested area is left of the left cursor. In this area,
5094 * the target drive is a full copy of the source drive.
5096 for (i = 0; i < num_stripes; i++) {
5097 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5099 * In case of DUP, in order to keep it
5100 * simple, only add the mirror with the
5101 * lowest physical address
5104 physical_of_found <=
5105 bbio->stripes[i].physical)
5109 physical_of_found = bbio->stripes[i].physical;
5113 u64 length = map->stripe_len;
5115 if (physical_of_found + length <=
5116 dev_replace->cursor_left) {
5117 struct btrfs_bio_stripe *tgtdev_stripe =
5118 bbio->stripes + num_stripes;
5120 tgtdev_stripe->physical = physical_of_found;
5121 tgtdev_stripe->length =
5122 bbio->stripes[index_srcdev].length;
5123 tgtdev_stripe->dev = dev_replace->tgtdev;
5131 bbio->num_stripes = num_stripes;
5132 bbio->max_errors = max_errors;
5133 bbio->mirror_num = mirror_num;
5136 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5137 * mirror_num == num_stripes + 1 && dev_replace target drive is
5138 * available as a mirror
5140 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5141 WARN_ON(num_stripes > 1);
5142 bbio->stripes[0].dev = dev_replace->tgtdev;
5143 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5144 bbio->mirror_num = map->num_stripes + 1;
5147 sort_parity_stripes(bbio, raid_map);
5148 *raid_map_ret = raid_map;
5151 if (dev_replace_is_ongoing)
5152 btrfs_dev_replace_unlock(dev_replace);
5153 free_extent_map(em);
5157 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5158 u64 logical, u64 *length,
5159 struct btrfs_bio **bbio_ret, int mirror_num)
5161 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5165 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5166 u64 chunk_start, u64 physical, u64 devid,
5167 u64 **logical, int *naddrs, int *stripe_len)
5169 struct extent_map_tree *em_tree = &map_tree->map_tree;
5170 struct extent_map *em;
5171 struct map_lookup *map;
5179 read_lock(&em_tree->lock);
5180 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5181 read_unlock(&em_tree->lock);
5184 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5189 if (em->start != chunk_start) {
5190 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5191 em->start, chunk_start);
5192 free_extent_map(em);
5195 map = (struct map_lookup *)em->bdev;
5198 rmap_len = map->stripe_len;
5200 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5201 do_div(length, map->num_stripes / map->sub_stripes);
5202 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5203 do_div(length, map->num_stripes);
5204 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5205 BTRFS_BLOCK_GROUP_RAID6)) {
5206 do_div(length, nr_data_stripes(map));
5207 rmap_len = map->stripe_len * nr_data_stripes(map);
5210 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5211 BUG_ON(!buf); /* -ENOMEM */
5213 for (i = 0; i < map->num_stripes; i++) {
5214 if (devid && map->stripes[i].dev->devid != devid)
5216 if (map->stripes[i].physical > physical ||
5217 map->stripes[i].physical + length <= physical)
5220 stripe_nr = physical - map->stripes[i].physical;
5221 do_div(stripe_nr, map->stripe_len);
5223 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5224 stripe_nr = stripe_nr * map->num_stripes + i;
5225 do_div(stripe_nr, map->sub_stripes);
5226 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5227 stripe_nr = stripe_nr * map->num_stripes + i;
5228 } /* else if RAID[56], multiply by nr_data_stripes().
5229 * Alternatively, just use rmap_len below instead of
5230 * map->stripe_len */
5232 bytenr = chunk_start + stripe_nr * rmap_len;
5233 WARN_ON(nr >= map->num_stripes);
5234 for (j = 0; j < nr; j++) {
5235 if (buf[j] == bytenr)
5239 WARN_ON(nr >= map->num_stripes);
5246 *stripe_len = rmap_len;
5248 free_extent_map(em);
5252 static void btrfs_end_bio(struct bio *bio, int err)
5254 struct btrfs_bio *bbio = bio->bi_private;
5255 int is_orig_bio = 0;
5258 atomic_inc(&bbio->error);
5259 if (err == -EIO || err == -EREMOTEIO) {
5260 unsigned int stripe_index =
5261 btrfs_io_bio(bio)->stripe_index;
5262 struct btrfs_device *dev;
5264 BUG_ON(stripe_index >= bbio->num_stripes);
5265 dev = bbio->stripes[stripe_index].dev;
5267 if (bio->bi_rw & WRITE)
5268 btrfs_dev_stat_inc(dev,
5269 BTRFS_DEV_STAT_WRITE_ERRS);
5271 btrfs_dev_stat_inc(dev,
5272 BTRFS_DEV_STAT_READ_ERRS);
5273 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5274 btrfs_dev_stat_inc(dev,
5275 BTRFS_DEV_STAT_FLUSH_ERRS);
5276 btrfs_dev_stat_print_on_error(dev);
5281 if (bio == bbio->orig_bio)
5284 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5287 bio = bbio->orig_bio;
5289 bio->bi_private = bbio->private;
5290 bio->bi_end_io = bbio->end_io;
5291 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5292 /* only send an error to the higher layers if it is
5293 * beyond the tolerance of the btrfs bio
5295 if (atomic_read(&bbio->error) > bbio->max_errors) {
5299 * this bio is actually up to date, we didn't
5300 * go over the max number of errors
5302 set_bit(BIO_UPTODATE, &bio->bi_flags);
5307 bio_endio(bio, err);
5308 } else if (!is_orig_bio) {
5313 struct async_sched {
5316 struct btrfs_fs_info *info;
5317 struct btrfs_work work;
5321 * see run_scheduled_bios for a description of why bios are collected for
5324 * This will add one bio to the pending list for a device and make sure
5325 * the work struct is scheduled.
5327 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5328 struct btrfs_device *device,
5329 int rw, struct bio *bio)
5331 int should_queue = 1;
5332 struct btrfs_pending_bios *pending_bios;
5334 if (device->missing || !device->bdev) {
5335 bio_endio(bio, -EIO);
5339 /* don't bother with additional async steps for reads, right now */
5340 if (!(rw & REQ_WRITE)) {
5342 btrfsic_submit_bio(rw, bio);
5348 * nr_async_bios allows us to reliably return congestion to the
5349 * higher layers. Otherwise, the async bio makes it appear we have
5350 * made progress against dirty pages when we've really just put it
5351 * on a queue for later
5353 atomic_inc(&root->fs_info->nr_async_bios);
5354 WARN_ON(bio->bi_next);
5355 bio->bi_next = NULL;
5358 spin_lock(&device->io_lock);
5359 if (bio->bi_rw & REQ_SYNC)
5360 pending_bios = &device->pending_sync_bios;
5362 pending_bios = &device->pending_bios;
5364 if (pending_bios->tail)
5365 pending_bios->tail->bi_next = bio;
5367 pending_bios->tail = bio;
5368 if (!pending_bios->head)
5369 pending_bios->head = bio;
5370 if (device->running_pending)
5373 spin_unlock(&device->io_lock);
5376 btrfs_queue_worker(&root->fs_info->submit_workers,
5380 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5383 struct bio_vec *prev;
5384 struct request_queue *q = bdev_get_queue(bdev);
5385 unsigned short max_sectors = queue_max_sectors(q);
5386 struct bvec_merge_data bvm = {
5388 .bi_sector = sector,
5389 .bi_rw = bio->bi_rw,
5392 if (bio->bi_vcnt == 0) {
5397 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5398 if (bio_sectors(bio) > max_sectors)
5401 if (!q->merge_bvec_fn)
5404 bvm.bi_size = bio->bi_size - prev->bv_len;
5405 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5410 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5411 struct bio *bio, u64 physical, int dev_nr,
5414 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5416 bio->bi_private = bbio;
5417 btrfs_io_bio(bio)->stripe_index = dev_nr;
5418 bio->bi_end_io = btrfs_end_bio;
5419 bio->bi_sector = physical >> 9;
5422 struct rcu_string *name;
5425 name = rcu_dereference(dev->name);
5426 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5427 "(%s id %llu), size=%u\n", rw,
5428 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5429 name->str, dev->devid, bio->bi_size);
5433 bio->bi_bdev = dev->bdev;
5435 btrfs_schedule_bio(root, dev, rw, bio);
5437 btrfsic_submit_bio(rw, bio);
5440 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5441 struct bio *first_bio, struct btrfs_device *dev,
5442 int dev_nr, int rw, int async)
5444 struct bio_vec *bvec = first_bio->bi_io_vec;
5446 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5447 u64 physical = bbio->stripes[dev_nr].physical;
5450 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5454 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5455 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5456 bvec->bv_offset) < bvec->bv_len) {
5457 u64 len = bio->bi_size;
5459 atomic_inc(&bbio->stripes_pending);
5460 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5468 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5472 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5474 atomic_inc(&bbio->error);
5475 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5476 bio->bi_private = bbio->private;
5477 bio->bi_end_io = bbio->end_io;
5478 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5479 bio->bi_sector = logical >> 9;
5481 bio_endio(bio, -EIO);
5485 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5486 int mirror_num, int async_submit)
5488 struct btrfs_device *dev;
5489 struct bio *first_bio = bio;
5490 u64 logical = (u64)bio->bi_sector << 9;
5493 u64 *raid_map = NULL;
5497 struct btrfs_bio *bbio = NULL;
5499 length = bio->bi_size;
5500 map_length = length;
5502 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5503 mirror_num, &raid_map);
5504 if (ret) /* -ENOMEM */
5507 total_devs = bbio->num_stripes;
5508 bbio->orig_bio = first_bio;
5509 bbio->private = first_bio->bi_private;
5510 bbio->end_io = first_bio->bi_end_io;
5511 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5514 /* In this case, map_length has been set to the length of
5515 a single stripe; not the whole write */
5517 return raid56_parity_write(root, bio, bbio,
5518 raid_map, map_length);
5520 return raid56_parity_recover(root, bio, bbio,
5521 raid_map, map_length,
5526 if (map_length < length) {
5527 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5528 logical, length, map_length);
5532 while (dev_nr < total_devs) {
5533 dev = bbio->stripes[dev_nr].dev;
5534 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5535 bbio_error(bbio, first_bio, logical);
5541 * Check and see if we're ok with this bio based on it's size
5542 * and offset with the given device.
5544 if (!bio_size_ok(dev->bdev, first_bio,
5545 bbio->stripes[dev_nr].physical >> 9)) {
5546 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5547 dev_nr, rw, async_submit);
5553 if (dev_nr < total_devs - 1) {
5554 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5555 BUG_ON(!bio); /* -ENOMEM */
5560 submit_stripe_bio(root, bbio, bio,
5561 bbio->stripes[dev_nr].physical, dev_nr, rw,
5568 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5571 struct btrfs_device *device;
5572 struct btrfs_fs_devices *cur_devices;
5574 cur_devices = fs_info->fs_devices;
5575 while (cur_devices) {
5577 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5578 device = __find_device(&cur_devices->devices,
5583 cur_devices = cur_devices->seed;
5588 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5589 u64 devid, u8 *dev_uuid)
5591 struct btrfs_device *device;
5592 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5594 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5598 list_add(&device->dev_list, &fs_devices->devices);
5599 device->fs_devices = fs_devices;
5600 fs_devices->num_devices++;
5602 device->missing = 1;
5603 fs_devices->missing_devices++;
5609 * btrfs_alloc_device - allocate struct btrfs_device
5610 * @fs_info: used only for generating a new devid, can be NULL if
5611 * devid is provided (i.e. @devid != NULL).
5612 * @devid: a pointer to devid for this device. If NULL a new devid
5614 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5617 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5618 * on error. Returned struct is not linked onto any lists and can be
5619 * destroyed with kfree() right away.
5621 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5625 struct btrfs_device *dev;
5628 if (!devid && !fs_info) {
5630 return ERR_PTR(-EINVAL);
5633 dev = __alloc_device();
5642 ret = find_next_devid(fs_info, &tmp);
5645 return ERR_PTR(ret);
5651 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5653 generate_random_uuid(dev->uuid);
5655 dev->work.func = pending_bios_fn;
5660 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5661 struct extent_buffer *leaf,
5662 struct btrfs_chunk *chunk)
5664 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5665 struct map_lookup *map;
5666 struct extent_map *em;
5670 u8 uuid[BTRFS_UUID_SIZE];
5675 logical = key->offset;
5676 length = btrfs_chunk_length(leaf, chunk);
5678 read_lock(&map_tree->map_tree.lock);
5679 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5680 read_unlock(&map_tree->map_tree.lock);
5682 /* already mapped? */
5683 if (em && em->start <= logical && em->start + em->len > logical) {
5684 free_extent_map(em);
5687 free_extent_map(em);
5690 em = alloc_extent_map();
5693 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5694 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5696 free_extent_map(em);
5700 em->bdev = (struct block_device *)map;
5701 em->start = logical;
5704 em->block_start = 0;
5705 em->block_len = em->len;
5707 map->num_stripes = num_stripes;
5708 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5709 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5710 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5711 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5712 map->type = btrfs_chunk_type(leaf, chunk);
5713 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5714 for (i = 0; i < num_stripes; i++) {
5715 map->stripes[i].physical =
5716 btrfs_stripe_offset_nr(leaf, chunk, i);
5717 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5718 read_extent_buffer(leaf, uuid, (unsigned long)
5719 btrfs_stripe_dev_uuid_nr(chunk, i),
5721 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5723 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5725 free_extent_map(em);
5728 if (!map->stripes[i].dev) {
5729 map->stripes[i].dev =
5730 add_missing_dev(root, devid, uuid);
5731 if (!map->stripes[i].dev) {
5733 free_extent_map(em);
5737 map->stripes[i].dev->in_fs_metadata = 1;
5740 write_lock(&map_tree->map_tree.lock);
5741 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5742 write_unlock(&map_tree->map_tree.lock);
5743 BUG_ON(ret); /* Tree corruption */
5744 free_extent_map(em);
5749 static void fill_device_from_item(struct extent_buffer *leaf,
5750 struct btrfs_dev_item *dev_item,
5751 struct btrfs_device *device)
5755 device->devid = btrfs_device_id(leaf, dev_item);
5756 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5757 device->total_bytes = device->disk_total_bytes;
5758 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5759 device->type = btrfs_device_type(leaf, dev_item);
5760 device->io_align = btrfs_device_io_align(leaf, dev_item);
5761 device->io_width = btrfs_device_io_width(leaf, dev_item);
5762 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5763 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5764 device->is_tgtdev_for_dev_replace = 0;
5766 ptr = btrfs_device_uuid(dev_item);
5767 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5770 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5772 struct btrfs_fs_devices *fs_devices;
5775 BUG_ON(!mutex_is_locked(&uuid_mutex));
5777 fs_devices = root->fs_info->fs_devices->seed;
5778 while (fs_devices) {
5779 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5783 fs_devices = fs_devices->seed;
5786 fs_devices = find_fsid(fsid);
5792 fs_devices = clone_fs_devices(fs_devices);
5793 if (IS_ERR(fs_devices)) {
5794 ret = PTR_ERR(fs_devices);
5798 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5799 root->fs_info->bdev_holder);
5801 free_fs_devices(fs_devices);
5805 if (!fs_devices->seeding) {
5806 __btrfs_close_devices(fs_devices);
5807 free_fs_devices(fs_devices);
5812 fs_devices->seed = root->fs_info->fs_devices->seed;
5813 root->fs_info->fs_devices->seed = fs_devices;
5818 static int read_one_dev(struct btrfs_root *root,
5819 struct extent_buffer *leaf,
5820 struct btrfs_dev_item *dev_item)
5822 struct btrfs_device *device;
5825 u8 fs_uuid[BTRFS_UUID_SIZE];
5826 u8 dev_uuid[BTRFS_UUID_SIZE];
5828 devid = btrfs_device_id(leaf, dev_item);
5829 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5831 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5834 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5835 ret = open_seed_devices(root, fs_uuid);
5836 if (ret && !btrfs_test_opt(root, DEGRADED))
5840 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5841 if (!device || !device->bdev) {
5842 if (!btrfs_test_opt(root, DEGRADED))
5846 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5847 device = add_missing_dev(root, devid, dev_uuid);
5850 } else if (!device->missing) {
5852 * this happens when a device that was properly setup
5853 * in the device info lists suddenly goes bad.
5854 * device->bdev is NULL, and so we have to set
5855 * device->missing to one here
5857 root->fs_info->fs_devices->missing_devices++;
5858 device->missing = 1;
5862 if (device->fs_devices != root->fs_info->fs_devices) {
5863 BUG_ON(device->writeable);
5864 if (device->generation !=
5865 btrfs_device_generation(leaf, dev_item))
5869 fill_device_from_item(leaf, dev_item, device);
5870 device->in_fs_metadata = 1;
5871 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5872 device->fs_devices->total_rw_bytes += device->total_bytes;
5873 spin_lock(&root->fs_info->free_chunk_lock);
5874 root->fs_info->free_chunk_space += device->total_bytes -
5876 spin_unlock(&root->fs_info->free_chunk_lock);
5882 int btrfs_read_sys_array(struct btrfs_root *root)
5884 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5885 struct extent_buffer *sb;
5886 struct btrfs_disk_key *disk_key;
5887 struct btrfs_chunk *chunk;
5889 unsigned long sb_ptr;
5895 struct btrfs_key key;
5897 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5898 BTRFS_SUPER_INFO_SIZE);
5901 btrfs_set_buffer_uptodate(sb);
5902 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5904 * The sb extent buffer is artifical and just used to read the system array.
5905 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5906 * pages up-to-date when the page is larger: extent does not cover the
5907 * whole page and consequently check_page_uptodate does not find all
5908 * the page's extents up-to-date (the hole beyond sb),
5909 * write_extent_buffer then triggers a WARN_ON.
5911 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5912 * but sb spans only this function. Add an explicit SetPageUptodate call
5913 * to silence the warning eg. on PowerPC 64.
5915 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5916 SetPageUptodate(sb->pages[0]);
5918 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5919 array_size = btrfs_super_sys_array_size(super_copy);
5921 ptr = super_copy->sys_chunk_array;
5922 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5925 while (cur < array_size) {
5926 disk_key = (struct btrfs_disk_key *)ptr;
5927 btrfs_disk_key_to_cpu(&key, disk_key);
5929 len = sizeof(*disk_key); ptr += len;
5933 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5934 chunk = (struct btrfs_chunk *)sb_ptr;
5935 ret = read_one_chunk(root, &key, sb, chunk);
5938 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5939 len = btrfs_chunk_item_size(num_stripes);
5948 free_extent_buffer(sb);
5952 int btrfs_read_chunk_tree(struct btrfs_root *root)
5954 struct btrfs_path *path;
5955 struct extent_buffer *leaf;
5956 struct btrfs_key key;
5957 struct btrfs_key found_key;
5961 root = root->fs_info->chunk_root;
5963 path = btrfs_alloc_path();
5967 mutex_lock(&uuid_mutex);
5971 * Read all device items, and then all the chunk items. All
5972 * device items are found before any chunk item (their object id
5973 * is smaller than the lowest possible object id for a chunk
5974 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5976 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5979 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5983 leaf = path->nodes[0];
5984 slot = path->slots[0];
5985 if (slot >= btrfs_header_nritems(leaf)) {
5986 ret = btrfs_next_leaf(root, path);
5993 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5994 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5995 struct btrfs_dev_item *dev_item;
5996 dev_item = btrfs_item_ptr(leaf, slot,
5997 struct btrfs_dev_item);
5998 ret = read_one_dev(root, leaf, dev_item);
6001 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6002 struct btrfs_chunk *chunk;
6003 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6004 ret = read_one_chunk(root, &found_key, leaf, chunk);
6012 unlock_chunks(root);
6013 mutex_unlock(&uuid_mutex);
6015 btrfs_free_path(path);
6019 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6021 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6022 struct btrfs_device *device;
6024 mutex_lock(&fs_devices->device_list_mutex);
6025 list_for_each_entry(device, &fs_devices->devices, dev_list)
6026 device->dev_root = fs_info->dev_root;
6027 mutex_unlock(&fs_devices->device_list_mutex);
6030 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6034 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6035 btrfs_dev_stat_reset(dev, i);
6038 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6040 struct btrfs_key key;
6041 struct btrfs_key found_key;
6042 struct btrfs_root *dev_root = fs_info->dev_root;
6043 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6044 struct extent_buffer *eb;
6047 struct btrfs_device *device;
6048 struct btrfs_path *path = NULL;
6051 path = btrfs_alloc_path();
6057 mutex_lock(&fs_devices->device_list_mutex);
6058 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6060 struct btrfs_dev_stats_item *ptr;
6063 key.type = BTRFS_DEV_STATS_KEY;
6064 key.offset = device->devid;
6065 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6067 __btrfs_reset_dev_stats(device);
6068 device->dev_stats_valid = 1;
6069 btrfs_release_path(path);
6072 slot = path->slots[0];
6073 eb = path->nodes[0];
6074 btrfs_item_key_to_cpu(eb, &found_key, slot);
6075 item_size = btrfs_item_size_nr(eb, slot);
6077 ptr = btrfs_item_ptr(eb, slot,
6078 struct btrfs_dev_stats_item);
6080 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6081 if (item_size >= (1 + i) * sizeof(__le64))
6082 btrfs_dev_stat_set(device, i,
6083 btrfs_dev_stats_value(eb, ptr, i));
6085 btrfs_dev_stat_reset(device, i);
6088 device->dev_stats_valid = 1;
6089 btrfs_dev_stat_print_on_load(device);
6090 btrfs_release_path(path);
6092 mutex_unlock(&fs_devices->device_list_mutex);
6095 btrfs_free_path(path);
6096 return ret < 0 ? ret : 0;
6099 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6100 struct btrfs_root *dev_root,
6101 struct btrfs_device *device)
6103 struct btrfs_path *path;
6104 struct btrfs_key key;
6105 struct extent_buffer *eb;
6106 struct btrfs_dev_stats_item *ptr;
6111 key.type = BTRFS_DEV_STATS_KEY;
6112 key.offset = device->devid;
6114 path = btrfs_alloc_path();
6116 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6118 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6119 ret, rcu_str_deref(device->name));
6124 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6125 /* need to delete old one and insert a new one */
6126 ret = btrfs_del_item(trans, dev_root, path);
6128 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6129 rcu_str_deref(device->name), ret);
6136 /* need to insert a new item */
6137 btrfs_release_path(path);
6138 ret = btrfs_insert_empty_item(trans, dev_root, path,
6139 &key, sizeof(*ptr));
6141 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6142 rcu_str_deref(device->name), ret);
6147 eb = path->nodes[0];
6148 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6149 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6150 btrfs_set_dev_stats_value(eb, ptr, i,
6151 btrfs_dev_stat_read(device, i));
6152 btrfs_mark_buffer_dirty(eb);
6155 btrfs_free_path(path);
6160 * called from commit_transaction. Writes all changed device stats to disk.
6162 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6163 struct btrfs_fs_info *fs_info)
6165 struct btrfs_root *dev_root = fs_info->dev_root;
6166 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6167 struct btrfs_device *device;
6170 mutex_lock(&fs_devices->device_list_mutex);
6171 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6172 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6175 ret = update_dev_stat_item(trans, dev_root, device);
6177 device->dev_stats_dirty = 0;
6179 mutex_unlock(&fs_devices->device_list_mutex);
6184 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6186 btrfs_dev_stat_inc(dev, index);
6187 btrfs_dev_stat_print_on_error(dev);
6190 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6192 if (!dev->dev_stats_valid)
6194 printk_ratelimited_in_rcu(KERN_ERR
6195 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6196 rcu_str_deref(dev->name),
6197 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6198 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6199 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6200 btrfs_dev_stat_read(dev,
6201 BTRFS_DEV_STAT_CORRUPTION_ERRS),
6202 btrfs_dev_stat_read(dev,
6203 BTRFS_DEV_STAT_GENERATION_ERRS));
6206 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6210 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6211 if (btrfs_dev_stat_read(dev, i) != 0)
6213 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6214 return; /* all values == 0, suppress message */
6216 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6217 rcu_str_deref(dev->name),
6218 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6219 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6220 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6221 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6222 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6225 int btrfs_get_dev_stats(struct btrfs_root *root,
6226 struct btrfs_ioctl_get_dev_stats *stats)
6228 struct btrfs_device *dev;
6229 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6232 mutex_lock(&fs_devices->device_list_mutex);
6233 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6234 mutex_unlock(&fs_devices->device_list_mutex);
6238 "btrfs: get dev_stats failed, device not found\n");
6240 } else if (!dev->dev_stats_valid) {
6242 "btrfs: get dev_stats failed, not yet valid\n");
6244 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6245 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6246 if (stats->nr_items > i)
6248 btrfs_dev_stat_read_and_reset(dev, i);
6250 btrfs_dev_stat_reset(dev, i);
6253 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6254 if (stats->nr_items > i)
6255 stats->values[i] = btrfs_dev_stat_read(dev, i);
6257 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6258 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6262 int btrfs_scratch_superblock(struct btrfs_device *device)
6264 struct buffer_head *bh;
6265 struct btrfs_super_block *disk_super;
6267 bh = btrfs_read_dev_super(device->bdev);
6270 disk_super = (struct btrfs_super_block *)bh->b_data;
6272 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6273 set_buffer_dirty(bh);
6274 sync_dirty_buffer(bh);