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 <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 struct btrfs_root *root,
46 struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
49 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
51 static DEFINE_MUTEX(uuid_mutex);
52 static LIST_HEAD(fs_uuids);
54 static void lock_chunks(struct btrfs_root *root)
56 mutex_lock(&root->fs_info->chunk_mutex);
59 static void unlock_chunks(struct btrfs_root *root)
61 mutex_unlock(&root->fs_info->chunk_mutex);
64 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
66 struct btrfs_device *device;
67 WARN_ON(fs_devices->opened);
68 while (!list_empty(&fs_devices->devices)) {
69 device = list_entry(fs_devices->devices.next,
70 struct btrfs_device, dev_list);
71 list_del(&device->dev_list);
72 rcu_string_free(device->name);
78 static void btrfs_kobject_uevent(struct block_device *bdev,
79 enum kobject_action action)
83 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
85 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
87 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
88 &disk_to_dev(bdev->bd_disk)->kobj);
91 void btrfs_cleanup_fs_uuids(void)
93 struct btrfs_fs_devices *fs_devices;
95 while (!list_empty(&fs_uuids)) {
96 fs_devices = list_entry(fs_uuids.next,
97 struct btrfs_fs_devices, list);
98 list_del(&fs_devices->list);
99 free_fs_devices(fs_devices);
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
106 struct btrfs_device *dev;
108 list_for_each_entry(dev, head, dev_list) {
109 if (dev->devid == devid &&
110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 struct btrfs_fs_devices *fs_devices;
121 list_for_each_entry(fs_devices, &fs_uuids, list) {
122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
129 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
130 int flush, struct block_device **bdev,
131 struct buffer_head **bh)
135 *bdev = blkdev_get_by_path(device_path, flags, holder);
138 ret = PTR_ERR(*bdev);
139 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
144 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
145 ret = set_blocksize(*bdev, 4096);
147 blkdev_put(*bdev, flags);
150 invalidate_bdev(*bdev);
151 *bh = btrfs_read_dev_super(*bdev);
154 blkdev_put(*bdev, flags);
166 static void requeue_list(struct btrfs_pending_bios *pending_bios,
167 struct bio *head, struct bio *tail)
170 struct bio *old_head;
172 old_head = pending_bios->head;
173 pending_bios->head = head;
174 if (pending_bios->tail)
175 tail->bi_next = old_head;
177 pending_bios->tail = tail;
181 * we try to collect pending bios for a device so we don't get a large
182 * number of procs sending bios down to the same device. This greatly
183 * improves the schedulers ability to collect and merge the bios.
185 * But, it also turns into a long list of bios to process and that is sure
186 * to eventually make the worker thread block. The solution here is to
187 * make some progress and then put this work struct back at the end of
188 * the list if the block device is congested. This way, multiple devices
189 * can make progress from a single worker thread.
191 static noinline void run_scheduled_bios(struct btrfs_device *device)
194 struct backing_dev_info *bdi;
195 struct btrfs_fs_info *fs_info;
196 struct btrfs_pending_bios *pending_bios;
200 unsigned long num_run;
201 unsigned long batch_run = 0;
203 unsigned long last_waited = 0;
205 int sync_pending = 0;
206 struct blk_plug plug;
209 * this function runs all the bios we've collected for
210 * a particular device. We don't want to wander off to
211 * another device without first sending all of these down.
212 * So, setup a plug here and finish it off before we return
214 blk_start_plug(&plug);
216 bdi = blk_get_backing_dev_info(device->bdev);
217 fs_info = device->dev_root->fs_info;
218 limit = btrfs_async_submit_limit(fs_info);
219 limit = limit * 2 / 3;
222 spin_lock(&device->io_lock);
227 /* take all the bios off the list at once and process them
228 * later on (without the lock held). But, remember the
229 * tail and other pointers so the bios can be properly reinserted
230 * into the list if we hit congestion
232 if (!force_reg && device->pending_sync_bios.head) {
233 pending_bios = &device->pending_sync_bios;
236 pending_bios = &device->pending_bios;
240 pending = pending_bios->head;
241 tail = pending_bios->tail;
242 WARN_ON(pending && !tail);
245 * if pending was null this time around, no bios need processing
246 * at all and we can stop. Otherwise it'll loop back up again
247 * and do an additional check so no bios are missed.
249 * device->running_pending is used to synchronize with the
252 if (device->pending_sync_bios.head == NULL &&
253 device->pending_bios.head == NULL) {
255 device->running_pending = 0;
258 device->running_pending = 1;
261 pending_bios->head = NULL;
262 pending_bios->tail = NULL;
264 spin_unlock(&device->io_lock);
269 /* we want to work on both lists, but do more bios on the
270 * sync list than the regular list
273 pending_bios != &device->pending_sync_bios &&
274 device->pending_sync_bios.head) ||
275 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
276 device->pending_bios.head)) {
277 spin_lock(&device->io_lock);
278 requeue_list(pending_bios, pending, tail);
283 pending = pending->bi_next;
286 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
287 waitqueue_active(&fs_info->async_submit_wait))
288 wake_up(&fs_info->async_submit_wait);
290 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
293 * if we're doing the sync list, record that our
294 * plug has some sync requests on it
296 * If we're doing the regular list and there are
297 * sync requests sitting around, unplug before
300 if (pending_bios == &device->pending_sync_bios) {
302 } else if (sync_pending) {
303 blk_finish_plug(&plug);
304 blk_start_plug(&plug);
308 btrfsic_submit_bio(cur->bi_rw, cur);
315 * we made progress, there is more work to do and the bdi
316 * is now congested. Back off and let other work structs
319 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
320 fs_info->fs_devices->open_devices > 1) {
321 struct io_context *ioc;
323 ioc = current->io_context;
326 * the main goal here is that we don't want to
327 * block if we're going to be able to submit
328 * more requests without blocking.
330 * This code does two great things, it pokes into
331 * the elevator code from a filesystem _and_
332 * it makes assumptions about how batching works.
334 if (ioc && ioc->nr_batch_requests > 0 &&
335 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
337 ioc->last_waited == last_waited)) {
339 * we want to go through our batch of
340 * requests and stop. So, we copy out
341 * the ioc->last_waited time and test
342 * against it before looping
344 last_waited = ioc->last_waited;
349 spin_lock(&device->io_lock);
350 requeue_list(pending_bios, pending, tail);
351 device->running_pending = 1;
353 spin_unlock(&device->io_lock);
354 btrfs_requeue_work(&device->work);
357 /* unplug every 64 requests just for good measure */
358 if (batch_run % 64 == 0) {
359 blk_finish_plug(&plug);
360 blk_start_plug(&plug);
369 spin_lock(&device->io_lock);
370 if (device->pending_bios.head || device->pending_sync_bios.head)
372 spin_unlock(&device->io_lock);
375 blk_finish_plug(&plug);
378 static void pending_bios_fn(struct btrfs_work *work)
380 struct btrfs_device *device;
382 device = container_of(work, struct btrfs_device, work);
383 run_scheduled_bios(device);
386 static noinline int device_list_add(const char *path,
387 struct btrfs_super_block *disk_super,
388 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
390 struct btrfs_device *device;
391 struct btrfs_fs_devices *fs_devices;
392 struct rcu_string *name;
393 u64 found_transid = btrfs_super_generation(disk_super);
395 fs_devices = find_fsid(disk_super->fsid);
397 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
400 INIT_LIST_HEAD(&fs_devices->devices);
401 INIT_LIST_HEAD(&fs_devices->alloc_list);
402 list_add(&fs_devices->list, &fs_uuids);
403 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
404 fs_devices->latest_devid = devid;
405 fs_devices->latest_trans = found_transid;
406 mutex_init(&fs_devices->device_list_mutex);
409 device = __find_device(&fs_devices->devices, devid,
410 disk_super->dev_item.uuid);
413 if (fs_devices->opened)
416 device = kzalloc(sizeof(*device), GFP_NOFS);
418 /* we can safely leave the fs_devices entry around */
421 device->devid = devid;
422 device->dev_stats_valid = 0;
423 device->work.func = pending_bios_fn;
424 memcpy(device->uuid, disk_super->dev_item.uuid,
426 spin_lock_init(&device->io_lock);
428 name = rcu_string_strdup(path, GFP_NOFS);
433 rcu_assign_pointer(device->name, name);
434 INIT_LIST_HEAD(&device->dev_alloc_list);
436 /* init readahead state */
437 spin_lock_init(&device->reada_lock);
438 device->reada_curr_zone = NULL;
439 atomic_set(&device->reada_in_flight, 0);
440 device->reada_next = 0;
441 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
442 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
444 mutex_lock(&fs_devices->device_list_mutex);
445 list_add_rcu(&device->dev_list, &fs_devices->devices);
446 mutex_unlock(&fs_devices->device_list_mutex);
448 device->fs_devices = fs_devices;
449 fs_devices->num_devices++;
450 } else if (!device->name || strcmp(device->name->str, path)) {
451 name = rcu_string_strdup(path, GFP_NOFS);
454 rcu_string_free(device->name);
455 rcu_assign_pointer(device->name, name);
456 if (device->missing) {
457 fs_devices->missing_devices--;
462 if (found_transid > fs_devices->latest_trans) {
463 fs_devices->latest_devid = devid;
464 fs_devices->latest_trans = found_transid;
466 *fs_devices_ret = fs_devices;
470 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
472 struct btrfs_fs_devices *fs_devices;
473 struct btrfs_device *device;
474 struct btrfs_device *orig_dev;
476 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
478 return ERR_PTR(-ENOMEM);
480 INIT_LIST_HEAD(&fs_devices->devices);
481 INIT_LIST_HEAD(&fs_devices->alloc_list);
482 INIT_LIST_HEAD(&fs_devices->list);
483 mutex_init(&fs_devices->device_list_mutex);
484 fs_devices->latest_devid = orig->latest_devid;
485 fs_devices->latest_trans = orig->latest_trans;
486 fs_devices->total_devices = orig->total_devices;
487 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
489 /* We have held the volume lock, it is safe to get the devices. */
490 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
491 struct rcu_string *name;
493 device = kzalloc(sizeof(*device), GFP_NOFS);
498 * This is ok to do without rcu read locked because we hold the
499 * uuid mutex so nothing we touch in here is going to disappear.
501 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
506 rcu_assign_pointer(device->name, name);
508 device->devid = orig_dev->devid;
509 device->work.func = pending_bios_fn;
510 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
511 spin_lock_init(&device->io_lock);
512 INIT_LIST_HEAD(&device->dev_list);
513 INIT_LIST_HEAD(&device->dev_alloc_list);
515 list_add(&device->dev_list, &fs_devices->devices);
516 device->fs_devices = fs_devices;
517 fs_devices->num_devices++;
521 free_fs_devices(fs_devices);
522 return ERR_PTR(-ENOMEM);
525 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
526 struct btrfs_fs_devices *fs_devices, int step)
528 struct btrfs_device *device, *next;
530 struct block_device *latest_bdev = NULL;
531 u64 latest_devid = 0;
532 u64 latest_transid = 0;
534 mutex_lock(&uuid_mutex);
536 /* This is the initialized path, it is safe to release the devices. */
537 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
538 if (device->in_fs_metadata) {
539 if (!device->is_tgtdev_for_dev_replace &&
541 device->generation > latest_transid)) {
542 latest_devid = device->devid;
543 latest_transid = device->generation;
544 latest_bdev = device->bdev;
549 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
551 * In the first step, keep the device which has
552 * the correct fsid and the devid that is used
553 * for the dev_replace procedure.
554 * In the second step, the dev_replace state is
555 * read from the device tree and it is known
556 * whether the procedure is really active or
557 * not, which means whether this device is
558 * used or whether it should be removed.
560 if (step == 0 || device->is_tgtdev_for_dev_replace) {
565 blkdev_put(device->bdev, device->mode);
567 fs_devices->open_devices--;
569 if (device->writeable) {
570 list_del_init(&device->dev_alloc_list);
571 device->writeable = 0;
572 if (!device->is_tgtdev_for_dev_replace)
573 fs_devices->rw_devices--;
575 list_del_init(&device->dev_list);
576 fs_devices->num_devices--;
577 rcu_string_free(device->name);
581 if (fs_devices->seed) {
582 fs_devices = fs_devices->seed;
586 fs_devices->latest_bdev = latest_bdev;
587 fs_devices->latest_devid = latest_devid;
588 fs_devices->latest_trans = latest_transid;
590 mutex_unlock(&uuid_mutex);
593 static void __free_device(struct work_struct *work)
595 struct btrfs_device *device;
597 device = container_of(work, struct btrfs_device, rcu_work);
600 blkdev_put(device->bdev, device->mode);
602 rcu_string_free(device->name);
606 static void free_device(struct rcu_head *head)
608 struct btrfs_device *device;
610 device = container_of(head, struct btrfs_device, rcu);
612 INIT_WORK(&device->rcu_work, __free_device);
613 schedule_work(&device->rcu_work);
616 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
618 struct btrfs_device *device;
620 if (--fs_devices->opened > 0)
623 mutex_lock(&fs_devices->device_list_mutex);
624 list_for_each_entry(device, &fs_devices->devices, dev_list) {
625 struct btrfs_device *new_device;
626 struct rcu_string *name;
629 fs_devices->open_devices--;
631 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
632 list_del_init(&device->dev_alloc_list);
633 fs_devices->rw_devices--;
636 if (device->can_discard)
637 fs_devices->num_can_discard--;
639 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
640 BUG_ON(!new_device); /* -ENOMEM */
641 memcpy(new_device, device, sizeof(*new_device));
643 /* Safe because we are under uuid_mutex */
645 name = rcu_string_strdup(device->name->str, GFP_NOFS);
646 BUG_ON(device->name && !name); /* -ENOMEM */
647 rcu_assign_pointer(new_device->name, name);
649 new_device->bdev = NULL;
650 new_device->writeable = 0;
651 new_device->in_fs_metadata = 0;
652 new_device->can_discard = 0;
653 spin_lock_init(&new_device->io_lock);
654 list_replace_rcu(&device->dev_list, &new_device->dev_list);
656 call_rcu(&device->rcu, free_device);
658 mutex_unlock(&fs_devices->device_list_mutex);
660 WARN_ON(fs_devices->open_devices);
661 WARN_ON(fs_devices->rw_devices);
662 fs_devices->opened = 0;
663 fs_devices->seeding = 0;
668 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
670 struct btrfs_fs_devices *seed_devices = NULL;
673 mutex_lock(&uuid_mutex);
674 ret = __btrfs_close_devices(fs_devices);
675 if (!fs_devices->opened) {
676 seed_devices = fs_devices->seed;
677 fs_devices->seed = NULL;
679 mutex_unlock(&uuid_mutex);
681 while (seed_devices) {
682 fs_devices = seed_devices;
683 seed_devices = fs_devices->seed;
684 __btrfs_close_devices(fs_devices);
685 free_fs_devices(fs_devices);
690 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
691 fmode_t flags, void *holder)
693 struct request_queue *q;
694 struct block_device *bdev;
695 struct list_head *head = &fs_devices->devices;
696 struct btrfs_device *device;
697 struct block_device *latest_bdev = NULL;
698 struct buffer_head *bh;
699 struct btrfs_super_block *disk_super;
700 u64 latest_devid = 0;
701 u64 latest_transid = 0;
708 list_for_each_entry(device, head, dev_list) {
714 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
719 disk_super = (struct btrfs_super_block *)bh->b_data;
720 devid = btrfs_stack_device_id(&disk_super->dev_item);
721 if (devid != device->devid)
724 if (memcmp(device->uuid, disk_super->dev_item.uuid,
728 device->generation = btrfs_super_generation(disk_super);
729 if (!latest_transid || device->generation > latest_transid) {
730 latest_devid = devid;
731 latest_transid = device->generation;
735 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
736 device->writeable = 0;
738 device->writeable = !bdev_read_only(bdev);
742 q = bdev_get_queue(bdev);
743 if (blk_queue_discard(q)) {
744 device->can_discard = 1;
745 fs_devices->num_can_discard++;
749 device->in_fs_metadata = 0;
750 device->mode = flags;
752 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
753 fs_devices->rotating = 1;
755 fs_devices->open_devices++;
756 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
757 fs_devices->rw_devices++;
758 list_add(&device->dev_alloc_list,
759 &fs_devices->alloc_list);
766 blkdev_put(bdev, flags);
769 if (fs_devices->open_devices == 0) {
773 fs_devices->seeding = seeding;
774 fs_devices->opened = 1;
775 fs_devices->latest_bdev = latest_bdev;
776 fs_devices->latest_devid = latest_devid;
777 fs_devices->latest_trans = latest_transid;
778 fs_devices->total_rw_bytes = 0;
783 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
784 fmode_t flags, void *holder)
788 mutex_lock(&uuid_mutex);
789 if (fs_devices->opened) {
790 fs_devices->opened++;
793 ret = __btrfs_open_devices(fs_devices, flags, holder);
795 mutex_unlock(&uuid_mutex);
800 * Look for a btrfs signature on a device. This may be called out of the mount path
801 * and we are not allowed to call set_blocksize during the scan. The superblock
802 * is read via pagecache
804 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
805 struct btrfs_fs_devices **fs_devices_ret)
807 struct btrfs_super_block *disk_super;
808 struct block_device *bdev;
819 * we would like to check all the supers, but that would make
820 * a btrfs mount succeed after a mkfs from a different FS.
821 * So, we need to add a special mount option to scan for
822 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
824 bytenr = btrfs_sb_offset(0);
826 mutex_lock(&uuid_mutex);
828 bdev = blkdev_get_by_path(path, flags, holder);
835 /* make sure our super fits in the device */
836 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
839 /* make sure our super fits in the page */
840 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
843 /* make sure our super doesn't straddle pages on disk */
844 index = bytenr >> PAGE_CACHE_SHIFT;
845 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
848 /* pull in the page with our super */
849 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
852 if (IS_ERR_OR_NULL(page))
857 /* align our pointer to the offset of the super block */
858 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
860 if (btrfs_super_bytenr(disk_super) != bytenr ||
861 disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
864 devid = btrfs_stack_device_id(&disk_super->dev_item);
865 transid = btrfs_super_generation(disk_super);
866 total_devices = btrfs_super_num_devices(disk_super);
868 if (disk_super->label[0]) {
869 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
870 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
871 printk(KERN_INFO "device label %s ", disk_super->label);
873 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
876 printk(KERN_CONT "devid %llu transid %llu %s\n",
877 (unsigned long long)devid, (unsigned long long)transid, path);
879 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
880 if (!ret && fs_devices_ret)
881 (*fs_devices_ret)->total_devices = total_devices;
885 page_cache_release(page);
888 blkdev_put(bdev, flags);
890 mutex_unlock(&uuid_mutex);
894 /* helper to account the used device space in the range */
895 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
896 u64 end, u64 *length)
898 struct btrfs_key key;
899 struct btrfs_root *root = device->dev_root;
900 struct btrfs_dev_extent *dev_extent;
901 struct btrfs_path *path;
905 struct extent_buffer *l;
909 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
912 path = btrfs_alloc_path();
917 key.objectid = device->devid;
919 key.type = BTRFS_DEV_EXTENT_KEY;
921 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
925 ret = btrfs_previous_item(root, path, key.objectid, key.type);
932 slot = path->slots[0];
933 if (slot >= btrfs_header_nritems(l)) {
934 ret = btrfs_next_leaf(root, path);
942 btrfs_item_key_to_cpu(l, &key, slot);
944 if (key.objectid < device->devid)
947 if (key.objectid > device->devid)
950 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
953 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
954 extent_end = key.offset + btrfs_dev_extent_length(l,
956 if (key.offset <= start && extent_end > end) {
957 *length = end - start + 1;
959 } else if (key.offset <= start && extent_end > start)
960 *length += extent_end - start;
961 else if (key.offset > start && extent_end <= end)
962 *length += extent_end - key.offset;
963 else if (key.offset > start && key.offset <= end) {
964 *length += end - key.offset + 1;
966 } else if (key.offset > end)
974 btrfs_free_path(path);
979 * find_free_dev_extent - find free space in the specified device
980 * @device: the device which we search the free space in
981 * @num_bytes: the size of the free space that we need
982 * @start: store the start of the free space.
983 * @len: the size of the free space. that we find, or the size of the max
984 * free space if we don't find suitable free space
986 * this uses a pretty simple search, the expectation is that it is
987 * called very infrequently and that a given device has a small number
990 * @start is used to store the start of the free space if we find. But if we
991 * don't find suitable free space, it will be used to store the start position
992 * of the max free space.
994 * @len is used to store the size of the free space that we find.
995 * But if we don't find suitable free space, it is used to store the size of
996 * the max free space.
998 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
999 u64 *start, u64 *len)
1001 struct btrfs_key key;
1002 struct btrfs_root *root = device->dev_root;
1003 struct btrfs_dev_extent *dev_extent;
1004 struct btrfs_path *path;
1010 u64 search_end = device->total_bytes;
1013 struct extent_buffer *l;
1015 /* FIXME use last free of some kind */
1017 /* we don't want to overwrite the superblock on the drive,
1018 * so we make sure to start at an offset of at least 1MB
1020 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1022 max_hole_start = search_start;
1026 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1031 path = btrfs_alloc_path();
1038 key.objectid = device->devid;
1039 key.offset = search_start;
1040 key.type = BTRFS_DEV_EXTENT_KEY;
1042 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1046 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1053 slot = path->slots[0];
1054 if (slot >= btrfs_header_nritems(l)) {
1055 ret = btrfs_next_leaf(root, path);
1063 btrfs_item_key_to_cpu(l, &key, slot);
1065 if (key.objectid < device->devid)
1068 if (key.objectid > device->devid)
1071 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1074 if (key.offset > search_start) {
1075 hole_size = key.offset - search_start;
1077 if (hole_size > max_hole_size) {
1078 max_hole_start = search_start;
1079 max_hole_size = hole_size;
1083 * If this free space is greater than which we need,
1084 * it must be the max free space that we have found
1085 * until now, so max_hole_start must point to the start
1086 * of this free space and the length of this free space
1087 * is stored in max_hole_size. Thus, we return
1088 * max_hole_start and max_hole_size and go back to the
1091 if (hole_size >= num_bytes) {
1097 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1098 extent_end = key.offset + btrfs_dev_extent_length(l,
1100 if (extent_end > search_start)
1101 search_start = extent_end;
1108 * At this point, search_start should be the end of
1109 * allocated dev extents, and when shrinking the device,
1110 * search_end may be smaller than search_start.
1112 if (search_end > search_start)
1113 hole_size = search_end - search_start;
1115 if (hole_size > max_hole_size) {
1116 max_hole_start = search_start;
1117 max_hole_size = hole_size;
1121 if (hole_size < num_bytes)
1127 btrfs_free_path(path);
1129 *start = max_hole_start;
1131 *len = max_hole_size;
1135 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1136 struct btrfs_device *device,
1140 struct btrfs_path *path;
1141 struct btrfs_root *root = device->dev_root;
1142 struct btrfs_key key;
1143 struct btrfs_key found_key;
1144 struct extent_buffer *leaf = NULL;
1145 struct btrfs_dev_extent *extent = NULL;
1147 path = btrfs_alloc_path();
1151 key.objectid = device->devid;
1153 key.type = BTRFS_DEV_EXTENT_KEY;
1155 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1157 ret = btrfs_previous_item(root, path, key.objectid,
1158 BTRFS_DEV_EXTENT_KEY);
1161 leaf = path->nodes[0];
1162 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1163 extent = btrfs_item_ptr(leaf, path->slots[0],
1164 struct btrfs_dev_extent);
1165 BUG_ON(found_key.offset > start || found_key.offset +
1166 btrfs_dev_extent_length(leaf, extent) < start);
1168 btrfs_release_path(path);
1170 } else if (ret == 0) {
1171 leaf = path->nodes[0];
1172 extent = btrfs_item_ptr(leaf, path->slots[0],
1173 struct btrfs_dev_extent);
1175 btrfs_error(root->fs_info, ret, "Slot search failed");
1179 if (device->bytes_used > 0) {
1180 u64 len = btrfs_dev_extent_length(leaf, extent);
1181 device->bytes_used -= len;
1182 spin_lock(&root->fs_info->free_chunk_lock);
1183 root->fs_info->free_chunk_space += len;
1184 spin_unlock(&root->fs_info->free_chunk_lock);
1186 ret = btrfs_del_item(trans, root, path);
1188 btrfs_error(root->fs_info, ret,
1189 "Failed to remove dev extent item");
1192 btrfs_free_path(path);
1196 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1197 struct btrfs_device *device,
1198 u64 chunk_tree, u64 chunk_objectid,
1199 u64 chunk_offset, u64 start, u64 num_bytes)
1202 struct btrfs_path *path;
1203 struct btrfs_root *root = device->dev_root;
1204 struct btrfs_dev_extent *extent;
1205 struct extent_buffer *leaf;
1206 struct btrfs_key key;
1208 WARN_ON(!device->in_fs_metadata);
1209 WARN_ON(device->is_tgtdev_for_dev_replace);
1210 path = btrfs_alloc_path();
1214 key.objectid = device->devid;
1216 key.type = BTRFS_DEV_EXTENT_KEY;
1217 ret = btrfs_insert_empty_item(trans, root, path, &key,
1222 leaf = path->nodes[0];
1223 extent = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_dev_extent);
1225 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1226 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1227 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1229 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1230 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1233 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1234 btrfs_mark_buffer_dirty(leaf);
1236 btrfs_free_path(path);
1240 static noinline int find_next_chunk(struct btrfs_root *root,
1241 u64 objectid, u64 *offset)
1243 struct btrfs_path *path;
1245 struct btrfs_key key;
1246 struct btrfs_chunk *chunk;
1247 struct btrfs_key found_key;
1249 path = btrfs_alloc_path();
1253 key.objectid = objectid;
1254 key.offset = (u64)-1;
1255 key.type = BTRFS_CHUNK_ITEM_KEY;
1257 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1261 BUG_ON(ret == 0); /* Corruption */
1263 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1267 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1269 if (found_key.objectid != objectid)
1272 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1273 struct btrfs_chunk);
1274 *offset = found_key.offset +
1275 btrfs_chunk_length(path->nodes[0], chunk);
1280 btrfs_free_path(path);
1284 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1287 struct btrfs_key key;
1288 struct btrfs_key found_key;
1289 struct btrfs_path *path;
1291 root = root->fs_info->chunk_root;
1293 path = btrfs_alloc_path();
1297 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1298 key.type = BTRFS_DEV_ITEM_KEY;
1299 key.offset = (u64)-1;
1301 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1305 BUG_ON(ret == 0); /* Corruption */
1307 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1308 BTRFS_DEV_ITEM_KEY);
1312 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1314 *objectid = found_key.offset + 1;
1318 btrfs_free_path(path);
1323 * the device information is stored in the chunk root
1324 * the btrfs_device struct should be fully filled in
1326 int btrfs_add_device(struct btrfs_trans_handle *trans,
1327 struct btrfs_root *root,
1328 struct btrfs_device *device)
1331 struct btrfs_path *path;
1332 struct btrfs_dev_item *dev_item;
1333 struct extent_buffer *leaf;
1334 struct btrfs_key key;
1337 root = root->fs_info->chunk_root;
1339 path = btrfs_alloc_path();
1343 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1344 key.type = BTRFS_DEV_ITEM_KEY;
1345 key.offset = device->devid;
1347 ret = btrfs_insert_empty_item(trans, root, path, &key,
1352 leaf = path->nodes[0];
1353 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1355 btrfs_set_device_id(leaf, dev_item, device->devid);
1356 btrfs_set_device_generation(leaf, dev_item, 0);
1357 btrfs_set_device_type(leaf, dev_item, device->type);
1358 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1359 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1360 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1361 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1362 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1363 btrfs_set_device_group(leaf, dev_item, 0);
1364 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1365 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1366 btrfs_set_device_start_offset(leaf, dev_item, 0);
1368 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1369 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1370 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1371 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1372 btrfs_mark_buffer_dirty(leaf);
1376 btrfs_free_path(path);
1380 static int btrfs_rm_dev_item(struct btrfs_root *root,
1381 struct btrfs_device *device)
1384 struct btrfs_path *path;
1385 struct btrfs_key key;
1386 struct btrfs_trans_handle *trans;
1388 root = root->fs_info->chunk_root;
1390 path = btrfs_alloc_path();
1394 trans = btrfs_start_transaction(root, 0);
1395 if (IS_ERR(trans)) {
1396 btrfs_free_path(path);
1397 return PTR_ERR(trans);
1399 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1400 key.type = BTRFS_DEV_ITEM_KEY;
1401 key.offset = device->devid;
1404 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1413 ret = btrfs_del_item(trans, root, path);
1417 btrfs_free_path(path);
1418 unlock_chunks(root);
1419 btrfs_commit_transaction(trans, root);
1423 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1425 struct btrfs_device *device;
1426 struct btrfs_device *next_device;
1427 struct block_device *bdev;
1428 struct buffer_head *bh = NULL;
1429 struct btrfs_super_block *disk_super;
1430 struct btrfs_fs_devices *cur_devices;
1437 bool clear_super = false;
1439 mutex_lock(&uuid_mutex);
1442 seq = read_seqbegin(&root->fs_info->profiles_lock);
1444 all_avail = root->fs_info->avail_data_alloc_bits |
1445 root->fs_info->avail_system_alloc_bits |
1446 root->fs_info->avail_metadata_alloc_bits;
1447 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1449 num_devices = root->fs_info->fs_devices->num_devices;
1450 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1451 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1452 WARN_ON(num_devices < 1);
1455 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1457 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1458 printk(KERN_ERR "btrfs: unable to go below four devices "
1464 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1465 printk(KERN_ERR "btrfs: unable to go below two "
1466 "devices on raid1\n");
1471 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1472 root->fs_info->fs_devices->rw_devices <= 2) {
1473 printk(KERN_ERR "btrfs: unable to go below two "
1474 "devices on raid5\n");
1478 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1479 root->fs_info->fs_devices->rw_devices <= 3) {
1480 printk(KERN_ERR "btrfs: unable to go below three "
1481 "devices on raid6\n");
1486 if (strcmp(device_path, "missing") == 0) {
1487 struct list_head *devices;
1488 struct btrfs_device *tmp;
1491 devices = &root->fs_info->fs_devices->devices;
1493 * It is safe to read the devices since the volume_mutex
1496 list_for_each_entry(tmp, devices, dev_list) {
1497 if (tmp->in_fs_metadata &&
1498 !tmp->is_tgtdev_for_dev_replace &&
1508 printk(KERN_ERR "btrfs: no missing devices found to "
1513 ret = btrfs_get_bdev_and_sb(device_path,
1514 FMODE_WRITE | FMODE_EXCL,
1515 root->fs_info->bdev_holder, 0,
1519 disk_super = (struct btrfs_super_block *)bh->b_data;
1520 devid = btrfs_stack_device_id(&disk_super->dev_item);
1521 dev_uuid = disk_super->dev_item.uuid;
1522 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1530 if (device->is_tgtdev_for_dev_replace) {
1531 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1536 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1537 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1543 if (device->writeable) {
1545 list_del_init(&device->dev_alloc_list);
1546 unlock_chunks(root);
1547 root->fs_info->fs_devices->rw_devices--;
1551 ret = btrfs_shrink_device(device, 0);
1556 * TODO: the superblock still includes this device in its num_devices
1557 * counter although write_all_supers() is not locked out. This
1558 * could give a filesystem state which requires a degraded mount.
1560 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1564 spin_lock(&root->fs_info->free_chunk_lock);
1565 root->fs_info->free_chunk_space = device->total_bytes -
1567 spin_unlock(&root->fs_info->free_chunk_lock);
1569 device->in_fs_metadata = 0;
1570 btrfs_scrub_cancel_dev(root->fs_info, device);
1573 * the device list mutex makes sure that we don't change
1574 * the device list while someone else is writing out all
1575 * the device supers.
1578 cur_devices = device->fs_devices;
1579 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1580 list_del_rcu(&device->dev_list);
1582 device->fs_devices->num_devices--;
1583 device->fs_devices->total_devices--;
1585 if (device->missing)
1586 root->fs_info->fs_devices->missing_devices--;
1588 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1589 struct btrfs_device, dev_list);
1590 if (device->bdev == root->fs_info->sb->s_bdev)
1591 root->fs_info->sb->s_bdev = next_device->bdev;
1592 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1593 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1596 device->fs_devices->open_devices--;
1598 call_rcu(&device->rcu, free_device);
1599 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1601 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1602 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1604 if (cur_devices->open_devices == 0) {
1605 struct btrfs_fs_devices *fs_devices;
1606 fs_devices = root->fs_info->fs_devices;
1607 while (fs_devices) {
1608 if (fs_devices->seed == cur_devices)
1610 fs_devices = fs_devices->seed;
1612 fs_devices->seed = cur_devices->seed;
1613 cur_devices->seed = NULL;
1615 __btrfs_close_devices(cur_devices);
1616 unlock_chunks(root);
1617 free_fs_devices(cur_devices);
1620 root->fs_info->num_tolerated_disk_barrier_failures =
1621 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1624 * at this point, the device is zero sized. We want to
1625 * remove it from the devices list and zero out the old super
1627 if (clear_super && disk_super) {
1628 /* make sure this device isn't detected as part of
1631 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1632 set_buffer_dirty(bh);
1633 sync_dirty_buffer(bh);
1638 /* Notify udev that device has changed */
1640 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1645 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1647 mutex_unlock(&uuid_mutex);
1650 if (device->writeable) {
1652 list_add(&device->dev_alloc_list,
1653 &root->fs_info->fs_devices->alloc_list);
1654 unlock_chunks(root);
1655 root->fs_info->fs_devices->rw_devices++;
1660 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1661 struct btrfs_device *srcdev)
1663 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1664 list_del_rcu(&srcdev->dev_list);
1665 list_del_rcu(&srcdev->dev_alloc_list);
1666 fs_info->fs_devices->num_devices--;
1667 if (srcdev->missing) {
1668 fs_info->fs_devices->missing_devices--;
1669 fs_info->fs_devices->rw_devices++;
1671 if (srcdev->can_discard)
1672 fs_info->fs_devices->num_can_discard--;
1674 fs_info->fs_devices->open_devices--;
1676 call_rcu(&srcdev->rcu, free_device);
1679 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1680 struct btrfs_device *tgtdev)
1682 struct btrfs_device *next_device;
1685 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1687 btrfs_scratch_superblock(tgtdev);
1688 fs_info->fs_devices->open_devices--;
1690 fs_info->fs_devices->num_devices--;
1691 if (tgtdev->can_discard)
1692 fs_info->fs_devices->num_can_discard++;
1694 next_device = list_entry(fs_info->fs_devices->devices.next,
1695 struct btrfs_device, dev_list);
1696 if (tgtdev->bdev == fs_info->sb->s_bdev)
1697 fs_info->sb->s_bdev = next_device->bdev;
1698 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1699 fs_info->fs_devices->latest_bdev = next_device->bdev;
1700 list_del_rcu(&tgtdev->dev_list);
1702 call_rcu(&tgtdev->rcu, free_device);
1704 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1707 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1708 struct btrfs_device **device)
1711 struct btrfs_super_block *disk_super;
1714 struct block_device *bdev;
1715 struct buffer_head *bh;
1718 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1719 root->fs_info->bdev_holder, 0, &bdev, &bh);
1722 disk_super = (struct btrfs_super_block *)bh->b_data;
1723 devid = btrfs_stack_device_id(&disk_super->dev_item);
1724 dev_uuid = disk_super->dev_item.uuid;
1725 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1730 blkdev_put(bdev, FMODE_READ);
1734 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1736 struct btrfs_device **device)
1739 if (strcmp(device_path, "missing") == 0) {
1740 struct list_head *devices;
1741 struct btrfs_device *tmp;
1743 devices = &root->fs_info->fs_devices->devices;
1745 * It is safe to read the devices since the volume_mutex
1746 * is held by the caller.
1748 list_for_each_entry(tmp, devices, dev_list) {
1749 if (tmp->in_fs_metadata && !tmp->bdev) {
1756 pr_err("btrfs: no missing device found\n");
1762 return btrfs_find_device_by_path(root, device_path, device);
1767 * does all the dirty work required for changing file system's UUID.
1769 static int btrfs_prepare_sprout(struct btrfs_root *root)
1771 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1772 struct btrfs_fs_devices *old_devices;
1773 struct btrfs_fs_devices *seed_devices;
1774 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1775 struct btrfs_device *device;
1778 BUG_ON(!mutex_is_locked(&uuid_mutex));
1779 if (!fs_devices->seeding)
1782 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1786 old_devices = clone_fs_devices(fs_devices);
1787 if (IS_ERR(old_devices)) {
1788 kfree(seed_devices);
1789 return PTR_ERR(old_devices);
1792 list_add(&old_devices->list, &fs_uuids);
1794 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1795 seed_devices->opened = 1;
1796 INIT_LIST_HEAD(&seed_devices->devices);
1797 INIT_LIST_HEAD(&seed_devices->alloc_list);
1798 mutex_init(&seed_devices->device_list_mutex);
1800 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1801 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1803 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1805 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1806 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1807 device->fs_devices = seed_devices;
1810 fs_devices->seeding = 0;
1811 fs_devices->num_devices = 0;
1812 fs_devices->open_devices = 0;
1813 fs_devices->total_devices = 0;
1814 fs_devices->seed = seed_devices;
1816 generate_random_uuid(fs_devices->fsid);
1817 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1818 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1819 super_flags = btrfs_super_flags(disk_super) &
1820 ~BTRFS_SUPER_FLAG_SEEDING;
1821 btrfs_set_super_flags(disk_super, super_flags);
1827 * strore the expected generation for seed devices in device items.
1829 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1830 struct btrfs_root *root)
1832 struct btrfs_path *path;
1833 struct extent_buffer *leaf;
1834 struct btrfs_dev_item *dev_item;
1835 struct btrfs_device *device;
1836 struct btrfs_key key;
1837 u8 fs_uuid[BTRFS_UUID_SIZE];
1838 u8 dev_uuid[BTRFS_UUID_SIZE];
1842 path = btrfs_alloc_path();
1846 root = root->fs_info->chunk_root;
1847 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1849 key.type = BTRFS_DEV_ITEM_KEY;
1852 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1856 leaf = path->nodes[0];
1858 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1859 ret = btrfs_next_leaf(root, path);
1864 leaf = path->nodes[0];
1865 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1866 btrfs_release_path(path);
1870 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1871 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1872 key.type != BTRFS_DEV_ITEM_KEY)
1875 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1876 struct btrfs_dev_item);
1877 devid = btrfs_device_id(leaf, dev_item);
1878 read_extent_buffer(leaf, dev_uuid,
1879 (unsigned long)btrfs_device_uuid(dev_item),
1881 read_extent_buffer(leaf, fs_uuid,
1882 (unsigned long)btrfs_device_fsid(dev_item),
1884 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1886 BUG_ON(!device); /* Logic error */
1888 if (device->fs_devices->seeding) {
1889 btrfs_set_device_generation(leaf, dev_item,
1890 device->generation);
1891 btrfs_mark_buffer_dirty(leaf);
1899 btrfs_free_path(path);
1903 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1905 struct request_queue *q;
1906 struct btrfs_trans_handle *trans;
1907 struct btrfs_device *device;
1908 struct block_device *bdev;
1909 struct list_head *devices;
1910 struct super_block *sb = root->fs_info->sb;
1911 struct rcu_string *name;
1913 int seeding_dev = 0;
1916 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1919 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1920 root->fs_info->bdev_holder);
1922 return PTR_ERR(bdev);
1924 if (root->fs_info->fs_devices->seeding) {
1926 down_write(&sb->s_umount);
1927 mutex_lock(&uuid_mutex);
1930 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1932 devices = &root->fs_info->fs_devices->devices;
1934 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1935 list_for_each_entry(device, devices, dev_list) {
1936 if (device->bdev == bdev) {
1939 &root->fs_info->fs_devices->device_list_mutex);
1943 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1945 device = kzalloc(sizeof(*device), GFP_NOFS);
1947 /* we can safely leave the fs_devices entry around */
1952 name = rcu_string_strdup(device_path, GFP_NOFS);
1958 rcu_assign_pointer(device->name, name);
1960 ret = find_next_devid(root, &device->devid);
1962 rcu_string_free(device->name);
1967 trans = btrfs_start_transaction(root, 0);
1968 if (IS_ERR(trans)) {
1969 rcu_string_free(device->name);
1971 ret = PTR_ERR(trans);
1977 q = bdev_get_queue(bdev);
1978 if (blk_queue_discard(q))
1979 device->can_discard = 1;
1980 device->writeable = 1;
1981 device->work.func = pending_bios_fn;
1982 generate_random_uuid(device->uuid);
1983 spin_lock_init(&device->io_lock);
1984 device->generation = trans->transid;
1985 device->io_width = root->sectorsize;
1986 device->io_align = root->sectorsize;
1987 device->sector_size = root->sectorsize;
1988 device->total_bytes = i_size_read(bdev->bd_inode);
1989 device->disk_total_bytes = device->total_bytes;
1990 device->dev_root = root->fs_info->dev_root;
1991 device->bdev = bdev;
1992 device->in_fs_metadata = 1;
1993 device->is_tgtdev_for_dev_replace = 0;
1994 device->mode = FMODE_EXCL;
1995 set_blocksize(device->bdev, 4096);
1998 sb->s_flags &= ~MS_RDONLY;
1999 ret = btrfs_prepare_sprout(root);
2000 BUG_ON(ret); /* -ENOMEM */
2003 device->fs_devices = root->fs_info->fs_devices;
2005 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2006 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2007 list_add(&device->dev_alloc_list,
2008 &root->fs_info->fs_devices->alloc_list);
2009 root->fs_info->fs_devices->num_devices++;
2010 root->fs_info->fs_devices->open_devices++;
2011 root->fs_info->fs_devices->rw_devices++;
2012 root->fs_info->fs_devices->total_devices++;
2013 if (device->can_discard)
2014 root->fs_info->fs_devices->num_can_discard++;
2015 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2017 spin_lock(&root->fs_info->free_chunk_lock);
2018 root->fs_info->free_chunk_space += device->total_bytes;
2019 spin_unlock(&root->fs_info->free_chunk_lock);
2021 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2022 root->fs_info->fs_devices->rotating = 1;
2024 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2025 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2026 total_bytes + device->total_bytes);
2028 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2029 btrfs_set_super_num_devices(root->fs_info->super_copy,
2031 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2034 ret = init_first_rw_device(trans, root, device);
2036 btrfs_abort_transaction(trans, root, ret);
2039 ret = btrfs_finish_sprout(trans, root);
2041 btrfs_abort_transaction(trans, root, ret);
2045 ret = btrfs_add_device(trans, root, device);
2047 btrfs_abort_transaction(trans, root, ret);
2053 * we've got more storage, clear any full flags on the space
2056 btrfs_clear_space_info_full(root->fs_info);
2058 unlock_chunks(root);
2059 root->fs_info->num_tolerated_disk_barrier_failures =
2060 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2061 ret = btrfs_commit_transaction(trans, root);
2064 mutex_unlock(&uuid_mutex);
2065 up_write(&sb->s_umount);
2067 if (ret) /* transaction commit */
2070 ret = btrfs_relocate_sys_chunks(root);
2072 btrfs_error(root->fs_info, ret,
2073 "Failed to relocate sys chunks after "
2074 "device initialization. This can be fixed "
2075 "using the \"btrfs balance\" command.");
2076 trans = btrfs_attach_transaction(root);
2077 if (IS_ERR(trans)) {
2078 if (PTR_ERR(trans) == -ENOENT)
2080 return PTR_ERR(trans);
2082 ret = btrfs_commit_transaction(trans, root);
2088 unlock_chunks(root);
2089 btrfs_end_transaction(trans, root);
2090 rcu_string_free(device->name);
2093 blkdev_put(bdev, FMODE_EXCL);
2095 mutex_unlock(&uuid_mutex);
2096 up_write(&sb->s_umount);
2101 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2102 struct btrfs_device **device_out)
2104 struct request_queue *q;
2105 struct btrfs_device *device;
2106 struct block_device *bdev;
2107 struct btrfs_fs_info *fs_info = root->fs_info;
2108 struct list_head *devices;
2109 struct rcu_string *name;
2113 if (fs_info->fs_devices->seeding)
2116 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2117 fs_info->bdev_holder);
2119 return PTR_ERR(bdev);
2121 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2123 devices = &fs_info->fs_devices->devices;
2124 list_for_each_entry(device, devices, dev_list) {
2125 if (device->bdev == bdev) {
2131 device = kzalloc(sizeof(*device), GFP_NOFS);
2137 name = rcu_string_strdup(device_path, GFP_NOFS);
2143 rcu_assign_pointer(device->name, name);
2145 q = bdev_get_queue(bdev);
2146 if (blk_queue_discard(q))
2147 device->can_discard = 1;
2148 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2149 device->writeable = 1;
2150 device->work.func = pending_bios_fn;
2151 generate_random_uuid(device->uuid);
2152 device->devid = BTRFS_DEV_REPLACE_DEVID;
2153 spin_lock_init(&device->io_lock);
2154 device->generation = 0;
2155 device->io_width = root->sectorsize;
2156 device->io_align = root->sectorsize;
2157 device->sector_size = root->sectorsize;
2158 device->total_bytes = i_size_read(bdev->bd_inode);
2159 device->disk_total_bytes = device->total_bytes;
2160 device->dev_root = fs_info->dev_root;
2161 device->bdev = bdev;
2162 device->in_fs_metadata = 1;
2163 device->is_tgtdev_for_dev_replace = 1;
2164 device->mode = FMODE_EXCL;
2165 set_blocksize(device->bdev, 4096);
2166 device->fs_devices = fs_info->fs_devices;
2167 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2168 fs_info->fs_devices->num_devices++;
2169 fs_info->fs_devices->open_devices++;
2170 if (device->can_discard)
2171 fs_info->fs_devices->num_can_discard++;
2172 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2174 *device_out = device;
2178 blkdev_put(bdev, FMODE_EXCL);
2182 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2183 struct btrfs_device *tgtdev)
2185 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2186 tgtdev->io_width = fs_info->dev_root->sectorsize;
2187 tgtdev->io_align = fs_info->dev_root->sectorsize;
2188 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2189 tgtdev->dev_root = fs_info->dev_root;
2190 tgtdev->in_fs_metadata = 1;
2193 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2194 struct btrfs_device *device)
2197 struct btrfs_path *path;
2198 struct btrfs_root *root;
2199 struct btrfs_dev_item *dev_item;
2200 struct extent_buffer *leaf;
2201 struct btrfs_key key;
2203 root = device->dev_root->fs_info->chunk_root;
2205 path = btrfs_alloc_path();
2209 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2210 key.type = BTRFS_DEV_ITEM_KEY;
2211 key.offset = device->devid;
2213 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2222 leaf = path->nodes[0];
2223 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2225 btrfs_set_device_id(leaf, dev_item, device->devid);
2226 btrfs_set_device_type(leaf, dev_item, device->type);
2227 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2228 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2229 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2230 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2231 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2232 btrfs_mark_buffer_dirty(leaf);
2235 btrfs_free_path(path);
2239 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2240 struct btrfs_device *device, u64 new_size)
2242 struct btrfs_super_block *super_copy =
2243 device->dev_root->fs_info->super_copy;
2244 u64 old_total = btrfs_super_total_bytes(super_copy);
2245 u64 diff = new_size - device->total_bytes;
2247 if (!device->writeable)
2249 if (new_size <= device->total_bytes ||
2250 device->is_tgtdev_for_dev_replace)
2253 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2254 device->fs_devices->total_rw_bytes += diff;
2256 device->total_bytes = new_size;
2257 device->disk_total_bytes = new_size;
2258 btrfs_clear_space_info_full(device->dev_root->fs_info);
2260 return btrfs_update_device(trans, device);
2263 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2264 struct btrfs_device *device, u64 new_size)
2267 lock_chunks(device->dev_root);
2268 ret = __btrfs_grow_device(trans, device, new_size);
2269 unlock_chunks(device->dev_root);
2273 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2274 struct btrfs_root *root,
2275 u64 chunk_tree, u64 chunk_objectid,
2279 struct btrfs_path *path;
2280 struct btrfs_key key;
2282 root = root->fs_info->chunk_root;
2283 path = btrfs_alloc_path();
2287 key.objectid = chunk_objectid;
2288 key.offset = chunk_offset;
2289 key.type = BTRFS_CHUNK_ITEM_KEY;
2291 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2294 else if (ret > 0) { /* Logic error or corruption */
2295 btrfs_error(root->fs_info, -ENOENT,
2296 "Failed lookup while freeing chunk.");
2301 ret = btrfs_del_item(trans, root, path);
2303 btrfs_error(root->fs_info, ret,
2304 "Failed to delete chunk item.");
2306 btrfs_free_path(path);
2310 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2313 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2314 struct btrfs_disk_key *disk_key;
2315 struct btrfs_chunk *chunk;
2322 struct btrfs_key key;
2324 array_size = btrfs_super_sys_array_size(super_copy);
2326 ptr = super_copy->sys_chunk_array;
2329 while (cur < array_size) {
2330 disk_key = (struct btrfs_disk_key *)ptr;
2331 btrfs_disk_key_to_cpu(&key, disk_key);
2333 len = sizeof(*disk_key);
2335 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2336 chunk = (struct btrfs_chunk *)(ptr + len);
2337 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2338 len += btrfs_chunk_item_size(num_stripes);
2343 if (key.objectid == chunk_objectid &&
2344 key.offset == chunk_offset) {
2345 memmove(ptr, ptr + len, array_size - (cur + len));
2347 btrfs_set_super_sys_array_size(super_copy, array_size);
2356 static int btrfs_relocate_chunk(struct btrfs_root *root,
2357 u64 chunk_tree, u64 chunk_objectid,
2360 struct extent_map_tree *em_tree;
2361 struct btrfs_root *extent_root;
2362 struct btrfs_trans_handle *trans;
2363 struct extent_map *em;
2364 struct map_lookup *map;
2368 root = root->fs_info->chunk_root;
2369 extent_root = root->fs_info->extent_root;
2370 em_tree = &root->fs_info->mapping_tree.map_tree;
2372 ret = btrfs_can_relocate(extent_root, chunk_offset);
2376 /* step one, relocate all the extents inside this chunk */
2377 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2381 trans = btrfs_start_transaction(root, 0);
2382 if (IS_ERR(trans)) {
2383 ret = PTR_ERR(trans);
2384 btrfs_std_error(root->fs_info, ret);
2391 * step two, delete the device extents and the
2392 * chunk tree entries
2394 read_lock(&em_tree->lock);
2395 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2396 read_unlock(&em_tree->lock);
2398 BUG_ON(!em || em->start > chunk_offset ||
2399 em->start + em->len < chunk_offset);
2400 map = (struct map_lookup *)em->bdev;
2402 for (i = 0; i < map->num_stripes; i++) {
2403 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2404 map->stripes[i].physical);
2407 if (map->stripes[i].dev) {
2408 ret = btrfs_update_device(trans, map->stripes[i].dev);
2412 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2417 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2419 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2420 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2424 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2427 write_lock(&em_tree->lock);
2428 remove_extent_mapping(em_tree, em);
2429 write_unlock(&em_tree->lock);
2434 /* once for the tree */
2435 free_extent_map(em);
2437 free_extent_map(em);
2439 unlock_chunks(root);
2440 btrfs_end_transaction(trans, root);
2444 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2446 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2447 struct btrfs_path *path;
2448 struct extent_buffer *leaf;
2449 struct btrfs_chunk *chunk;
2450 struct btrfs_key key;
2451 struct btrfs_key found_key;
2452 u64 chunk_tree = chunk_root->root_key.objectid;
2454 bool retried = false;
2458 path = btrfs_alloc_path();
2463 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2464 key.offset = (u64)-1;
2465 key.type = BTRFS_CHUNK_ITEM_KEY;
2468 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2471 BUG_ON(ret == 0); /* Corruption */
2473 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2480 leaf = path->nodes[0];
2481 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2483 chunk = btrfs_item_ptr(leaf, path->slots[0],
2484 struct btrfs_chunk);
2485 chunk_type = btrfs_chunk_type(leaf, chunk);
2486 btrfs_release_path(path);
2488 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2489 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2498 if (found_key.offset == 0)
2500 key.offset = found_key.offset - 1;
2503 if (failed && !retried) {
2507 } else if (failed && retried) {
2512 btrfs_free_path(path);
2516 static int insert_balance_item(struct btrfs_root *root,
2517 struct btrfs_balance_control *bctl)
2519 struct btrfs_trans_handle *trans;
2520 struct btrfs_balance_item *item;
2521 struct btrfs_disk_balance_args disk_bargs;
2522 struct btrfs_path *path;
2523 struct extent_buffer *leaf;
2524 struct btrfs_key key;
2527 path = btrfs_alloc_path();
2531 trans = btrfs_start_transaction(root, 0);
2532 if (IS_ERR(trans)) {
2533 btrfs_free_path(path);
2534 return PTR_ERR(trans);
2537 key.objectid = BTRFS_BALANCE_OBJECTID;
2538 key.type = BTRFS_BALANCE_ITEM_KEY;
2541 ret = btrfs_insert_empty_item(trans, root, path, &key,
2546 leaf = path->nodes[0];
2547 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2549 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2551 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2552 btrfs_set_balance_data(leaf, item, &disk_bargs);
2553 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2554 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2555 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2556 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2558 btrfs_set_balance_flags(leaf, item, bctl->flags);
2560 btrfs_mark_buffer_dirty(leaf);
2562 btrfs_free_path(path);
2563 err = btrfs_commit_transaction(trans, root);
2569 static int del_balance_item(struct btrfs_root *root)
2571 struct btrfs_trans_handle *trans;
2572 struct btrfs_path *path;
2573 struct btrfs_key key;
2576 path = btrfs_alloc_path();
2580 trans = btrfs_start_transaction(root, 0);
2581 if (IS_ERR(trans)) {
2582 btrfs_free_path(path);
2583 return PTR_ERR(trans);
2586 key.objectid = BTRFS_BALANCE_OBJECTID;
2587 key.type = BTRFS_BALANCE_ITEM_KEY;
2590 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2598 ret = btrfs_del_item(trans, root, path);
2600 btrfs_free_path(path);
2601 err = btrfs_commit_transaction(trans, root);
2608 * This is a heuristic used to reduce the number of chunks balanced on
2609 * resume after balance was interrupted.
2611 static void update_balance_args(struct btrfs_balance_control *bctl)
2614 * Turn on soft mode for chunk types that were being converted.
2616 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2617 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2618 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2619 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2620 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2621 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2624 * Turn on usage filter if is not already used. The idea is
2625 * that chunks that we have already balanced should be
2626 * reasonably full. Don't do it for chunks that are being
2627 * converted - that will keep us from relocating unconverted
2628 * (albeit full) chunks.
2630 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2631 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2632 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2633 bctl->data.usage = 90;
2635 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2636 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2637 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2638 bctl->sys.usage = 90;
2640 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2641 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2642 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2643 bctl->meta.usage = 90;
2648 * Should be called with both balance and volume mutexes held to
2649 * serialize other volume operations (add_dev/rm_dev/resize) with
2650 * restriper. Same goes for unset_balance_control.
2652 static void set_balance_control(struct btrfs_balance_control *bctl)
2654 struct btrfs_fs_info *fs_info = bctl->fs_info;
2656 BUG_ON(fs_info->balance_ctl);
2658 spin_lock(&fs_info->balance_lock);
2659 fs_info->balance_ctl = bctl;
2660 spin_unlock(&fs_info->balance_lock);
2663 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2665 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2667 BUG_ON(!fs_info->balance_ctl);
2669 spin_lock(&fs_info->balance_lock);
2670 fs_info->balance_ctl = NULL;
2671 spin_unlock(&fs_info->balance_lock);
2677 * Balance filters. Return 1 if chunk should be filtered out
2678 * (should not be balanced).
2680 static int chunk_profiles_filter(u64 chunk_type,
2681 struct btrfs_balance_args *bargs)
2683 chunk_type = chunk_to_extended(chunk_type) &
2684 BTRFS_EXTENDED_PROFILE_MASK;
2686 if (bargs->profiles & chunk_type)
2692 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2693 struct btrfs_balance_args *bargs)
2695 struct btrfs_block_group_cache *cache;
2696 u64 chunk_used, user_thresh;
2699 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2700 chunk_used = btrfs_block_group_used(&cache->item);
2702 if (bargs->usage == 0)
2704 else if (bargs->usage > 100)
2705 user_thresh = cache->key.offset;
2707 user_thresh = div_factor_fine(cache->key.offset,
2710 if (chunk_used < user_thresh)
2713 btrfs_put_block_group(cache);
2717 static int chunk_devid_filter(struct extent_buffer *leaf,
2718 struct btrfs_chunk *chunk,
2719 struct btrfs_balance_args *bargs)
2721 struct btrfs_stripe *stripe;
2722 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2725 for (i = 0; i < num_stripes; i++) {
2726 stripe = btrfs_stripe_nr(chunk, i);
2727 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2734 /* [pstart, pend) */
2735 static int chunk_drange_filter(struct extent_buffer *leaf,
2736 struct btrfs_chunk *chunk,
2738 struct btrfs_balance_args *bargs)
2740 struct btrfs_stripe *stripe;
2741 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2747 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2750 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2751 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2752 factor = num_stripes / 2;
2753 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2754 factor = num_stripes - 1;
2755 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2756 factor = num_stripes - 2;
2758 factor = num_stripes;
2761 for (i = 0; i < num_stripes; i++) {
2762 stripe = btrfs_stripe_nr(chunk, i);
2763 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2766 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2767 stripe_length = btrfs_chunk_length(leaf, chunk);
2768 do_div(stripe_length, factor);
2770 if (stripe_offset < bargs->pend &&
2771 stripe_offset + stripe_length > bargs->pstart)
2778 /* [vstart, vend) */
2779 static int chunk_vrange_filter(struct extent_buffer *leaf,
2780 struct btrfs_chunk *chunk,
2782 struct btrfs_balance_args *bargs)
2784 if (chunk_offset < bargs->vend &&
2785 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2786 /* at least part of the chunk is inside this vrange */
2792 static int chunk_soft_convert_filter(u64 chunk_type,
2793 struct btrfs_balance_args *bargs)
2795 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2798 chunk_type = chunk_to_extended(chunk_type) &
2799 BTRFS_EXTENDED_PROFILE_MASK;
2801 if (bargs->target == chunk_type)
2807 static int should_balance_chunk(struct btrfs_root *root,
2808 struct extent_buffer *leaf,
2809 struct btrfs_chunk *chunk, u64 chunk_offset)
2811 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2812 struct btrfs_balance_args *bargs = NULL;
2813 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2816 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2817 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2821 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2822 bargs = &bctl->data;
2823 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2825 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2826 bargs = &bctl->meta;
2828 /* profiles filter */
2829 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2830 chunk_profiles_filter(chunk_type, bargs)) {
2835 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2836 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2841 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2842 chunk_devid_filter(leaf, chunk, bargs)) {
2846 /* drange filter, makes sense only with devid filter */
2847 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2848 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2853 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2854 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2858 /* soft profile changing mode */
2859 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2860 chunk_soft_convert_filter(chunk_type, bargs)) {
2867 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2869 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2870 struct btrfs_root *chunk_root = fs_info->chunk_root;
2871 struct btrfs_root *dev_root = fs_info->dev_root;
2872 struct list_head *devices;
2873 struct btrfs_device *device;
2876 struct btrfs_chunk *chunk;
2877 struct btrfs_path *path;
2878 struct btrfs_key key;
2879 struct btrfs_key found_key;
2880 struct btrfs_trans_handle *trans;
2881 struct extent_buffer *leaf;
2884 int enospc_errors = 0;
2885 bool counting = true;
2887 /* step one make some room on all the devices */
2888 devices = &fs_info->fs_devices->devices;
2889 list_for_each_entry(device, devices, dev_list) {
2890 old_size = device->total_bytes;
2891 size_to_free = div_factor(old_size, 1);
2892 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2893 if (!device->writeable ||
2894 device->total_bytes - device->bytes_used > size_to_free ||
2895 device->is_tgtdev_for_dev_replace)
2898 ret = btrfs_shrink_device(device, old_size - size_to_free);
2903 trans = btrfs_start_transaction(dev_root, 0);
2904 BUG_ON(IS_ERR(trans));
2906 ret = btrfs_grow_device(trans, device, old_size);
2909 btrfs_end_transaction(trans, dev_root);
2912 /* step two, relocate all the chunks */
2913 path = btrfs_alloc_path();
2919 /* zero out stat counters */
2920 spin_lock(&fs_info->balance_lock);
2921 memset(&bctl->stat, 0, sizeof(bctl->stat));
2922 spin_unlock(&fs_info->balance_lock);
2924 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2925 key.offset = (u64)-1;
2926 key.type = BTRFS_CHUNK_ITEM_KEY;
2929 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2930 atomic_read(&fs_info->balance_cancel_req)) {
2935 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2940 * this shouldn't happen, it means the last relocate
2944 BUG(); /* FIXME break ? */
2946 ret = btrfs_previous_item(chunk_root, path, 0,
2947 BTRFS_CHUNK_ITEM_KEY);
2953 leaf = path->nodes[0];
2954 slot = path->slots[0];
2955 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2957 if (found_key.objectid != key.objectid)
2960 /* chunk zero is special */
2961 if (found_key.offset == 0)
2964 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2967 spin_lock(&fs_info->balance_lock);
2968 bctl->stat.considered++;
2969 spin_unlock(&fs_info->balance_lock);
2972 ret = should_balance_chunk(chunk_root, leaf, chunk,
2974 btrfs_release_path(path);
2979 spin_lock(&fs_info->balance_lock);
2980 bctl->stat.expected++;
2981 spin_unlock(&fs_info->balance_lock);
2985 ret = btrfs_relocate_chunk(chunk_root,
2986 chunk_root->root_key.objectid,
2989 if (ret && ret != -ENOSPC)
2991 if (ret == -ENOSPC) {
2994 spin_lock(&fs_info->balance_lock);
2995 bctl->stat.completed++;
2996 spin_unlock(&fs_info->balance_lock);
2999 key.offset = found_key.offset - 1;
3003 btrfs_release_path(path);
3008 btrfs_free_path(path);
3009 if (enospc_errors) {
3010 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3020 * alloc_profile_is_valid - see if a given profile is valid and reduced
3021 * @flags: profile to validate
3022 * @extended: if true @flags is treated as an extended profile
3024 static int alloc_profile_is_valid(u64 flags, int extended)
3026 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3027 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3029 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3031 /* 1) check that all other bits are zeroed */
3035 /* 2) see if profile is reduced */
3037 return !extended; /* "0" is valid for usual profiles */
3039 /* true if exactly one bit set */
3040 return (flags & (flags - 1)) == 0;
3043 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3045 /* cancel requested || normal exit path */
3046 return atomic_read(&fs_info->balance_cancel_req) ||
3047 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3048 atomic_read(&fs_info->balance_cancel_req) == 0);
3051 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3055 unset_balance_control(fs_info);
3056 ret = del_balance_item(fs_info->tree_root);
3058 btrfs_std_error(fs_info, ret);
3060 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3063 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3064 struct btrfs_ioctl_balance_args *bargs);
3067 * Should be called with both balance and volume mutexes held
3069 int btrfs_balance(struct btrfs_balance_control *bctl,
3070 struct btrfs_ioctl_balance_args *bargs)
3072 struct btrfs_fs_info *fs_info = bctl->fs_info;
3079 if (btrfs_fs_closing(fs_info) ||
3080 atomic_read(&fs_info->balance_pause_req) ||
3081 atomic_read(&fs_info->balance_cancel_req)) {
3086 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3087 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3091 * In case of mixed groups both data and meta should be picked,
3092 * and identical options should be given for both of them.
3094 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3095 if (mixed && (bctl->flags & allowed)) {
3096 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3097 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3098 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3099 printk(KERN_ERR "btrfs: with mixed groups data and "
3100 "metadata balance options must be the same\n");
3106 num_devices = fs_info->fs_devices->num_devices;
3107 btrfs_dev_replace_lock(&fs_info->dev_replace);
3108 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3109 BUG_ON(num_devices < 1);
3112 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3113 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3114 if (num_devices == 1)
3115 allowed |= BTRFS_BLOCK_GROUP_DUP;
3116 else if (num_devices < 4)
3117 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3119 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3120 BTRFS_BLOCK_GROUP_RAID10 |
3121 BTRFS_BLOCK_GROUP_RAID5 |
3122 BTRFS_BLOCK_GROUP_RAID6);
3124 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3125 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3126 (bctl->data.target & ~allowed))) {
3127 printk(KERN_ERR "btrfs: unable to start balance with target "
3128 "data profile %llu\n",
3129 (unsigned long long)bctl->data.target);
3133 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3134 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3135 (bctl->meta.target & ~allowed))) {
3136 printk(KERN_ERR "btrfs: unable to start balance with target "
3137 "metadata profile %llu\n",
3138 (unsigned long long)bctl->meta.target);
3142 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3143 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3144 (bctl->sys.target & ~allowed))) {
3145 printk(KERN_ERR "btrfs: unable to start balance with target "
3146 "system profile %llu\n",
3147 (unsigned long long)bctl->sys.target);
3152 /* allow dup'ed data chunks only in mixed mode */
3153 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3154 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3155 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3160 /* allow to reduce meta or sys integrity only if force set */
3161 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3162 BTRFS_BLOCK_GROUP_RAID10 |
3163 BTRFS_BLOCK_GROUP_RAID5 |
3164 BTRFS_BLOCK_GROUP_RAID6;
3166 seq = read_seqbegin(&fs_info->profiles_lock);
3168 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3169 (fs_info->avail_system_alloc_bits & allowed) &&
3170 !(bctl->sys.target & allowed)) ||
3171 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3172 (fs_info->avail_metadata_alloc_bits & allowed) &&
3173 !(bctl->meta.target & allowed))) {
3174 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3175 printk(KERN_INFO "btrfs: force reducing metadata "
3178 printk(KERN_ERR "btrfs: balance will reduce metadata "
3179 "integrity, use force if you want this\n");
3184 } while (read_seqretry(&fs_info->profiles_lock, seq));
3186 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3187 int num_tolerated_disk_barrier_failures;
3188 u64 target = bctl->sys.target;
3190 num_tolerated_disk_barrier_failures =
3191 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3192 if (num_tolerated_disk_barrier_failures > 0 &&
3194 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3195 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3196 num_tolerated_disk_barrier_failures = 0;
3197 else if (num_tolerated_disk_barrier_failures > 1 &&
3199 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3200 num_tolerated_disk_barrier_failures = 1;
3202 fs_info->num_tolerated_disk_barrier_failures =
3203 num_tolerated_disk_barrier_failures;
3206 ret = insert_balance_item(fs_info->tree_root, bctl);
3207 if (ret && ret != -EEXIST)
3210 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3211 BUG_ON(ret == -EEXIST);
3212 set_balance_control(bctl);
3214 BUG_ON(ret != -EEXIST);
3215 spin_lock(&fs_info->balance_lock);
3216 update_balance_args(bctl);
3217 spin_unlock(&fs_info->balance_lock);
3220 atomic_inc(&fs_info->balance_running);
3221 mutex_unlock(&fs_info->balance_mutex);
3223 ret = __btrfs_balance(fs_info);
3225 mutex_lock(&fs_info->balance_mutex);
3226 atomic_dec(&fs_info->balance_running);
3228 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3229 fs_info->num_tolerated_disk_barrier_failures =
3230 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3234 memset(bargs, 0, sizeof(*bargs));
3235 update_ioctl_balance_args(fs_info, 0, bargs);
3238 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3239 balance_need_close(fs_info)) {
3240 __cancel_balance(fs_info);
3243 wake_up(&fs_info->balance_wait_q);
3247 if (bctl->flags & BTRFS_BALANCE_RESUME)
3248 __cancel_balance(fs_info);
3251 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3256 static int balance_kthread(void *data)
3258 struct btrfs_fs_info *fs_info = data;
3261 mutex_lock(&fs_info->volume_mutex);
3262 mutex_lock(&fs_info->balance_mutex);
3264 if (fs_info->balance_ctl) {
3265 printk(KERN_INFO "btrfs: continuing balance\n");
3266 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3269 mutex_unlock(&fs_info->balance_mutex);
3270 mutex_unlock(&fs_info->volume_mutex);
3275 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3277 struct task_struct *tsk;
3279 spin_lock(&fs_info->balance_lock);
3280 if (!fs_info->balance_ctl) {
3281 spin_unlock(&fs_info->balance_lock);
3284 spin_unlock(&fs_info->balance_lock);
3286 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3287 printk(KERN_INFO "btrfs: force skipping balance\n");
3291 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3293 return PTR_ERR(tsk);
3298 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3300 struct btrfs_balance_control *bctl;
3301 struct btrfs_balance_item *item;
3302 struct btrfs_disk_balance_args disk_bargs;
3303 struct btrfs_path *path;
3304 struct extent_buffer *leaf;
3305 struct btrfs_key key;
3308 path = btrfs_alloc_path();
3312 key.objectid = BTRFS_BALANCE_OBJECTID;
3313 key.type = BTRFS_BALANCE_ITEM_KEY;
3316 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3319 if (ret > 0) { /* ret = -ENOENT; */
3324 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3330 leaf = path->nodes[0];
3331 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3333 bctl->fs_info = fs_info;
3334 bctl->flags = btrfs_balance_flags(leaf, item);
3335 bctl->flags |= BTRFS_BALANCE_RESUME;
3337 btrfs_balance_data(leaf, item, &disk_bargs);
3338 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3339 btrfs_balance_meta(leaf, item, &disk_bargs);
3340 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3341 btrfs_balance_sys(leaf, item, &disk_bargs);
3342 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3344 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3346 mutex_lock(&fs_info->volume_mutex);
3347 mutex_lock(&fs_info->balance_mutex);
3349 set_balance_control(bctl);
3351 mutex_unlock(&fs_info->balance_mutex);
3352 mutex_unlock(&fs_info->volume_mutex);
3354 btrfs_free_path(path);
3358 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3362 mutex_lock(&fs_info->balance_mutex);
3363 if (!fs_info->balance_ctl) {
3364 mutex_unlock(&fs_info->balance_mutex);
3368 if (atomic_read(&fs_info->balance_running)) {
3369 atomic_inc(&fs_info->balance_pause_req);
3370 mutex_unlock(&fs_info->balance_mutex);
3372 wait_event(fs_info->balance_wait_q,
3373 atomic_read(&fs_info->balance_running) == 0);
3375 mutex_lock(&fs_info->balance_mutex);
3376 /* we are good with balance_ctl ripped off from under us */
3377 BUG_ON(atomic_read(&fs_info->balance_running));
3378 atomic_dec(&fs_info->balance_pause_req);
3383 mutex_unlock(&fs_info->balance_mutex);
3387 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3389 mutex_lock(&fs_info->balance_mutex);
3390 if (!fs_info->balance_ctl) {
3391 mutex_unlock(&fs_info->balance_mutex);
3395 atomic_inc(&fs_info->balance_cancel_req);
3397 * if we are running just wait and return, balance item is
3398 * deleted in btrfs_balance in this case
3400 if (atomic_read(&fs_info->balance_running)) {
3401 mutex_unlock(&fs_info->balance_mutex);
3402 wait_event(fs_info->balance_wait_q,
3403 atomic_read(&fs_info->balance_running) == 0);
3404 mutex_lock(&fs_info->balance_mutex);
3406 /* __cancel_balance needs volume_mutex */
3407 mutex_unlock(&fs_info->balance_mutex);
3408 mutex_lock(&fs_info->volume_mutex);
3409 mutex_lock(&fs_info->balance_mutex);
3411 if (fs_info->balance_ctl)
3412 __cancel_balance(fs_info);
3414 mutex_unlock(&fs_info->volume_mutex);
3417 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3418 atomic_dec(&fs_info->balance_cancel_req);
3419 mutex_unlock(&fs_info->balance_mutex);
3424 * shrinking a device means finding all of the device extents past
3425 * the new size, and then following the back refs to the chunks.
3426 * The chunk relocation code actually frees the device extent
3428 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3430 struct btrfs_trans_handle *trans;
3431 struct btrfs_root *root = device->dev_root;
3432 struct btrfs_dev_extent *dev_extent = NULL;
3433 struct btrfs_path *path;
3441 bool retried = false;
3442 struct extent_buffer *l;
3443 struct btrfs_key key;
3444 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3445 u64 old_total = btrfs_super_total_bytes(super_copy);
3446 u64 old_size = device->total_bytes;
3447 u64 diff = device->total_bytes - new_size;
3449 if (device->is_tgtdev_for_dev_replace)
3452 path = btrfs_alloc_path();
3460 device->total_bytes = new_size;
3461 if (device->writeable) {
3462 device->fs_devices->total_rw_bytes -= diff;
3463 spin_lock(&root->fs_info->free_chunk_lock);
3464 root->fs_info->free_chunk_space -= diff;
3465 spin_unlock(&root->fs_info->free_chunk_lock);
3467 unlock_chunks(root);
3470 key.objectid = device->devid;
3471 key.offset = (u64)-1;
3472 key.type = BTRFS_DEV_EXTENT_KEY;
3475 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3479 ret = btrfs_previous_item(root, path, 0, key.type);
3484 btrfs_release_path(path);
3489 slot = path->slots[0];
3490 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3492 if (key.objectid != device->devid) {
3493 btrfs_release_path(path);
3497 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3498 length = btrfs_dev_extent_length(l, dev_extent);
3500 if (key.offset + length <= new_size) {
3501 btrfs_release_path(path);
3505 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3506 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3507 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3508 btrfs_release_path(path);
3510 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3512 if (ret && ret != -ENOSPC)
3516 } while (key.offset-- > 0);
3518 if (failed && !retried) {
3522 } else if (failed && retried) {
3526 device->total_bytes = old_size;
3527 if (device->writeable)
3528 device->fs_devices->total_rw_bytes += diff;
3529 spin_lock(&root->fs_info->free_chunk_lock);
3530 root->fs_info->free_chunk_space += diff;
3531 spin_unlock(&root->fs_info->free_chunk_lock);
3532 unlock_chunks(root);
3536 /* Shrinking succeeded, else we would be at "done". */
3537 trans = btrfs_start_transaction(root, 0);
3538 if (IS_ERR(trans)) {
3539 ret = PTR_ERR(trans);
3545 device->disk_total_bytes = new_size;
3546 /* Now btrfs_update_device() will change the on-disk size. */
3547 ret = btrfs_update_device(trans, device);
3549 unlock_chunks(root);
3550 btrfs_end_transaction(trans, root);
3553 WARN_ON(diff > old_total);
3554 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3555 unlock_chunks(root);
3556 btrfs_end_transaction(trans, root);
3558 btrfs_free_path(path);
3562 static int btrfs_add_system_chunk(struct btrfs_root *root,
3563 struct btrfs_key *key,
3564 struct btrfs_chunk *chunk, int item_size)
3566 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3567 struct btrfs_disk_key disk_key;
3571 array_size = btrfs_super_sys_array_size(super_copy);
3572 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3575 ptr = super_copy->sys_chunk_array + array_size;
3576 btrfs_cpu_key_to_disk(&disk_key, key);
3577 memcpy(ptr, &disk_key, sizeof(disk_key));
3578 ptr += sizeof(disk_key);
3579 memcpy(ptr, chunk, item_size);
3580 item_size += sizeof(disk_key);
3581 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3586 * sort the devices in descending order by max_avail, total_avail
3588 static int btrfs_cmp_device_info(const void *a, const void *b)
3590 const struct btrfs_device_info *di_a = a;
3591 const struct btrfs_device_info *di_b = b;
3593 if (di_a->max_avail > di_b->max_avail)
3595 if (di_a->max_avail < di_b->max_avail)
3597 if (di_a->total_avail > di_b->total_avail)
3599 if (di_a->total_avail < di_b->total_avail)
3604 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3605 [BTRFS_RAID_RAID10] = {
3608 .devs_max = 0, /* 0 == as many as possible */
3610 .devs_increment = 2,
3613 [BTRFS_RAID_RAID1] = {
3618 .devs_increment = 2,
3621 [BTRFS_RAID_DUP] = {
3626 .devs_increment = 1,
3629 [BTRFS_RAID_RAID0] = {
3634 .devs_increment = 1,
3637 [BTRFS_RAID_SINGLE] = {
3642 .devs_increment = 1,
3645 [BTRFS_RAID_RAID5] = {
3650 .devs_increment = 1,
3653 [BTRFS_RAID_RAID6] = {
3658 .devs_increment = 1,
3663 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3665 /* TODO allow them to set a preferred stripe size */
3669 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3673 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3676 features = btrfs_super_incompat_flags(info->super_copy);
3677 if (features & BTRFS_FEATURE_INCOMPAT_RAID56)
3680 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
3681 btrfs_set_super_incompat_flags(info->super_copy, features);
3682 printk(KERN_INFO "btrfs: setting RAID5/6 feature flag\n");
3685 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3686 struct btrfs_root *extent_root,
3687 struct map_lookup **map_ret,
3688 u64 *num_bytes_out, u64 *stripe_size_out,
3689 u64 start, u64 type)
3691 struct btrfs_fs_info *info = extent_root->fs_info;
3692 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3693 struct list_head *cur;
3694 struct map_lookup *map = NULL;
3695 struct extent_map_tree *em_tree;
3696 struct extent_map *em;
3697 struct btrfs_device_info *devices_info = NULL;
3699 int num_stripes; /* total number of stripes to allocate */
3700 int data_stripes; /* number of stripes that count for
3702 int sub_stripes; /* sub_stripes info for map */
3703 int dev_stripes; /* stripes per dev */
3704 int devs_max; /* max devs to use */
3705 int devs_min; /* min devs needed */
3706 int devs_increment; /* ndevs has to be a multiple of this */
3707 int ncopies; /* how many copies to data has */
3709 u64 max_stripe_size;
3713 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3719 BUG_ON(!alloc_profile_is_valid(type, 0));
3721 if (list_empty(&fs_devices->alloc_list))
3724 index = __get_raid_index(type);
3726 sub_stripes = btrfs_raid_array[index].sub_stripes;
3727 dev_stripes = btrfs_raid_array[index].dev_stripes;
3728 devs_max = btrfs_raid_array[index].devs_max;
3729 devs_min = btrfs_raid_array[index].devs_min;
3730 devs_increment = btrfs_raid_array[index].devs_increment;
3731 ncopies = btrfs_raid_array[index].ncopies;
3733 if (type & BTRFS_BLOCK_GROUP_DATA) {
3734 max_stripe_size = 1024 * 1024 * 1024;
3735 max_chunk_size = 10 * max_stripe_size;
3736 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3737 /* for larger filesystems, use larger metadata chunks */
3738 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3739 max_stripe_size = 1024 * 1024 * 1024;
3741 max_stripe_size = 256 * 1024 * 1024;
3742 max_chunk_size = max_stripe_size;
3743 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3744 max_stripe_size = 32 * 1024 * 1024;
3745 max_chunk_size = 2 * max_stripe_size;
3747 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3752 /* we don't want a chunk larger than 10% of writeable space */
3753 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3756 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3761 cur = fs_devices->alloc_list.next;
3764 * in the first pass through the devices list, we gather information
3765 * about the available holes on each device.
3768 while (cur != &fs_devices->alloc_list) {
3769 struct btrfs_device *device;
3773 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3777 if (!device->writeable) {
3779 "btrfs: read-only device in alloc_list\n");
3783 if (!device->in_fs_metadata ||
3784 device->is_tgtdev_for_dev_replace)
3787 if (device->total_bytes > device->bytes_used)
3788 total_avail = device->total_bytes - device->bytes_used;
3792 /* If there is no space on this device, skip it. */
3793 if (total_avail == 0)
3796 ret = find_free_dev_extent(device,
3797 max_stripe_size * dev_stripes,
3798 &dev_offset, &max_avail);
3799 if (ret && ret != -ENOSPC)
3803 max_avail = max_stripe_size * dev_stripes;
3805 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3808 if (ndevs == fs_devices->rw_devices) {
3809 WARN(1, "%s: found more than %llu devices\n",
3810 __func__, fs_devices->rw_devices);
3813 devices_info[ndevs].dev_offset = dev_offset;
3814 devices_info[ndevs].max_avail = max_avail;
3815 devices_info[ndevs].total_avail = total_avail;
3816 devices_info[ndevs].dev = device;
3821 * now sort the devices by hole size / available space
3823 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3824 btrfs_cmp_device_info, NULL);
3826 /* round down to number of usable stripes */
3827 ndevs -= ndevs % devs_increment;
3829 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3834 if (devs_max && ndevs > devs_max)
3837 * the primary goal is to maximize the number of stripes, so use as many
3838 * devices as possible, even if the stripes are not maximum sized.
3840 stripe_size = devices_info[ndevs-1].max_avail;
3841 num_stripes = ndevs * dev_stripes;
3844 * this will have to be fixed for RAID1 and RAID10 over
3847 data_stripes = num_stripes / ncopies;
3849 if (type & BTRFS_BLOCK_GROUP_RAID5) {
3850 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3851 btrfs_super_stripesize(info->super_copy));
3852 data_stripes = num_stripes - 1;
3854 if (type & BTRFS_BLOCK_GROUP_RAID6) {
3855 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3856 btrfs_super_stripesize(info->super_copy));
3857 data_stripes = num_stripes - 2;
3861 * Use the number of data stripes to figure out how big this chunk
3862 * is really going to be in terms of logical address space,
3863 * and compare that answer with the max chunk size
3865 if (stripe_size * data_stripes > max_chunk_size) {
3866 u64 mask = (1ULL << 24) - 1;
3867 stripe_size = max_chunk_size;
3868 do_div(stripe_size, data_stripes);
3870 /* bump the answer up to a 16MB boundary */
3871 stripe_size = (stripe_size + mask) & ~mask;
3873 /* but don't go higher than the limits we found
3874 * while searching for free extents
3876 if (stripe_size > devices_info[ndevs-1].max_avail)
3877 stripe_size = devices_info[ndevs-1].max_avail;
3880 do_div(stripe_size, dev_stripes);
3882 /* align to BTRFS_STRIPE_LEN */
3883 do_div(stripe_size, raid_stripe_len);
3884 stripe_size *= raid_stripe_len;
3886 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3891 map->num_stripes = num_stripes;
3893 for (i = 0; i < ndevs; ++i) {
3894 for (j = 0; j < dev_stripes; ++j) {
3895 int s = i * dev_stripes + j;
3896 map->stripes[s].dev = devices_info[i].dev;
3897 map->stripes[s].physical = devices_info[i].dev_offset +
3901 map->sector_size = extent_root->sectorsize;
3902 map->stripe_len = raid_stripe_len;
3903 map->io_align = raid_stripe_len;
3904 map->io_width = raid_stripe_len;
3906 map->sub_stripes = sub_stripes;
3909 num_bytes = stripe_size * data_stripes;
3911 *stripe_size_out = stripe_size;
3912 *num_bytes_out = num_bytes;
3914 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3916 em = alloc_extent_map();
3921 em->bdev = (struct block_device *)map;
3923 em->len = num_bytes;
3924 em->block_start = 0;
3925 em->block_len = em->len;
3927 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3928 write_lock(&em_tree->lock);
3929 ret = add_extent_mapping(em_tree, em);
3930 write_unlock(&em_tree->lock);
3932 free_extent_map(em);
3936 for (i = 0; i < map->num_stripes; ++i) {
3937 struct btrfs_device *device;
3940 device = map->stripes[i].dev;
3941 dev_offset = map->stripes[i].physical;
3943 ret = btrfs_alloc_dev_extent(trans, device,
3944 info->chunk_root->root_key.objectid,
3945 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3946 start, dev_offset, stripe_size);
3948 goto error_dev_extent;
3951 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3952 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3955 i = map->num_stripes - 1;
3956 goto error_dev_extent;
3959 free_extent_map(em);
3960 check_raid56_incompat_flag(extent_root->fs_info, type);
3962 kfree(devices_info);
3966 for (; i >= 0; i--) {
3967 struct btrfs_device *device;
3970 device = map->stripes[i].dev;
3971 err = btrfs_free_dev_extent(trans, device, start);
3973 btrfs_abort_transaction(trans, extent_root, err);
3977 write_lock(&em_tree->lock);
3978 remove_extent_mapping(em_tree, em);
3979 write_unlock(&em_tree->lock);
3981 /* One for our allocation */
3982 free_extent_map(em);
3983 /* One for the tree reference */
3984 free_extent_map(em);
3987 kfree(devices_info);
3991 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3992 struct btrfs_root *extent_root,
3993 struct map_lookup *map, u64 chunk_offset,
3994 u64 chunk_size, u64 stripe_size)
3997 struct btrfs_key key;
3998 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3999 struct btrfs_device *device;
4000 struct btrfs_chunk *chunk;
4001 struct btrfs_stripe *stripe;
4002 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
4006 chunk = kzalloc(item_size, GFP_NOFS);
4011 while (index < map->num_stripes) {
4012 device = map->stripes[index].dev;
4013 device->bytes_used += stripe_size;
4014 ret = btrfs_update_device(trans, device);
4020 spin_lock(&extent_root->fs_info->free_chunk_lock);
4021 extent_root->fs_info->free_chunk_space -= (stripe_size *
4023 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4026 stripe = &chunk->stripe;
4027 while (index < map->num_stripes) {
4028 device = map->stripes[index].dev;
4029 dev_offset = map->stripes[index].physical;
4031 btrfs_set_stack_stripe_devid(stripe, device->devid);
4032 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4033 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4038 btrfs_set_stack_chunk_length(chunk, chunk_size);
4039 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4040 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4041 btrfs_set_stack_chunk_type(chunk, map->type);
4042 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4043 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4044 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4045 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4046 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4048 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4049 key.type = BTRFS_CHUNK_ITEM_KEY;
4050 key.offset = chunk_offset;
4052 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4054 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4056 * TODO: Cleanup of inserted chunk root in case of
4059 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4069 * Chunk allocation falls into two parts. The first part does works
4070 * that make the new allocated chunk useable, but not do any operation
4071 * that modifies the chunk tree. The second part does the works that
4072 * require modifying the chunk tree. This division is important for the
4073 * bootstrap process of adding storage to a seed btrfs.
4075 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4076 struct btrfs_root *extent_root, u64 type)
4081 struct map_lookup *map;
4082 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4085 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4090 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4091 &stripe_size, chunk_offset, type);
4095 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4096 chunk_size, stripe_size);
4102 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4103 struct btrfs_root *root,
4104 struct btrfs_device *device)
4107 u64 sys_chunk_offset;
4111 u64 sys_stripe_size;
4113 struct map_lookup *map;
4114 struct map_lookup *sys_map;
4115 struct btrfs_fs_info *fs_info = root->fs_info;
4116 struct btrfs_root *extent_root = fs_info->extent_root;
4119 ret = find_next_chunk(fs_info->chunk_root,
4120 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4124 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4125 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4126 &stripe_size, chunk_offset, alloc_profile);
4130 sys_chunk_offset = chunk_offset + chunk_size;
4132 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4133 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4134 &sys_chunk_size, &sys_stripe_size,
4135 sys_chunk_offset, alloc_profile);
4137 btrfs_abort_transaction(trans, root, ret);
4141 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4143 btrfs_abort_transaction(trans, root, ret);
4148 * Modifying chunk tree needs allocating new blocks from both
4149 * system block group and metadata block group. So we only can
4150 * do operations require modifying the chunk tree after both
4151 * block groups were created.
4153 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4154 chunk_size, stripe_size);
4156 btrfs_abort_transaction(trans, root, ret);
4160 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4161 sys_chunk_offset, sys_chunk_size,
4164 btrfs_abort_transaction(trans, root, ret);
4171 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4173 struct extent_map *em;
4174 struct map_lookup *map;
4175 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4179 read_lock(&map_tree->map_tree.lock);
4180 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4181 read_unlock(&map_tree->map_tree.lock);
4185 if (btrfs_test_opt(root, DEGRADED)) {
4186 free_extent_map(em);
4190 map = (struct map_lookup *)em->bdev;
4191 for (i = 0; i < map->num_stripes; i++) {
4192 if (!map->stripes[i].dev->writeable) {
4197 free_extent_map(em);
4201 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4203 extent_map_tree_init(&tree->map_tree);
4206 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4208 struct extent_map *em;
4211 write_lock(&tree->map_tree.lock);
4212 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4214 remove_extent_mapping(&tree->map_tree, em);
4215 write_unlock(&tree->map_tree.lock);
4220 free_extent_map(em);
4221 /* once for the tree */
4222 free_extent_map(em);
4226 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4228 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4229 struct extent_map *em;
4230 struct map_lookup *map;
4231 struct extent_map_tree *em_tree = &map_tree->map_tree;
4234 read_lock(&em_tree->lock);
4235 em = lookup_extent_mapping(em_tree, logical, len);
4236 read_unlock(&em_tree->lock);
4239 BUG_ON(em->start > logical || em->start + em->len < logical);
4240 map = (struct map_lookup *)em->bdev;
4241 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4242 ret = map->num_stripes;
4243 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4244 ret = map->sub_stripes;
4245 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4247 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4251 free_extent_map(em);
4253 btrfs_dev_replace_lock(&fs_info->dev_replace);
4254 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4256 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4261 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4262 struct btrfs_mapping_tree *map_tree,
4265 struct extent_map *em;
4266 struct map_lookup *map;
4267 struct extent_map_tree *em_tree = &map_tree->map_tree;
4268 unsigned long len = root->sectorsize;
4270 read_lock(&em_tree->lock);
4271 em = lookup_extent_mapping(em_tree, logical, len);
4272 read_unlock(&em_tree->lock);
4275 BUG_ON(em->start > logical || em->start + em->len < logical);
4276 map = (struct map_lookup *)em->bdev;
4277 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4278 BTRFS_BLOCK_GROUP_RAID6)) {
4279 len = map->stripe_len * nr_data_stripes(map);
4281 free_extent_map(em);
4285 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4286 u64 logical, u64 len, int mirror_num)
4288 struct extent_map *em;
4289 struct map_lookup *map;
4290 struct extent_map_tree *em_tree = &map_tree->map_tree;
4293 read_lock(&em_tree->lock);
4294 em = lookup_extent_mapping(em_tree, logical, len);
4295 read_unlock(&em_tree->lock);
4298 BUG_ON(em->start > logical || em->start + em->len < logical);
4299 map = (struct map_lookup *)em->bdev;
4300 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4301 BTRFS_BLOCK_GROUP_RAID6))
4303 free_extent_map(em);
4307 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4308 struct map_lookup *map, int first, int num,
4309 int optimal, int dev_replace_is_ongoing)
4313 struct btrfs_device *srcdev;
4315 if (dev_replace_is_ongoing &&
4316 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4317 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4318 srcdev = fs_info->dev_replace.srcdev;
4323 * try to avoid the drive that is the source drive for a
4324 * dev-replace procedure, only choose it if no other non-missing
4325 * mirror is available
4327 for (tolerance = 0; tolerance < 2; tolerance++) {
4328 if (map->stripes[optimal].dev->bdev &&
4329 (tolerance || map->stripes[optimal].dev != srcdev))
4331 for (i = first; i < first + num; i++) {
4332 if (map->stripes[i].dev->bdev &&
4333 (tolerance || map->stripes[i].dev != srcdev))
4338 /* we couldn't find one that doesn't fail. Just return something
4339 * and the io error handling code will clean up eventually
4344 static inline int parity_smaller(u64 a, u64 b)
4349 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4350 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4352 struct btrfs_bio_stripe s;
4359 for (i = 0; i < bbio->num_stripes - 1; i++) {
4360 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4361 s = bbio->stripes[i];
4363 bbio->stripes[i] = bbio->stripes[i+1];
4364 raid_map[i] = raid_map[i+1];
4365 bbio->stripes[i+1] = s;
4373 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4374 u64 logical, u64 *length,
4375 struct btrfs_bio **bbio_ret,
4376 int mirror_num, u64 **raid_map_ret)
4378 struct extent_map *em;
4379 struct map_lookup *map;
4380 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4381 struct extent_map_tree *em_tree = &map_tree->map_tree;
4384 u64 stripe_end_offset;
4389 u64 *raid_map = NULL;
4395 struct btrfs_bio *bbio = NULL;
4396 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4397 int dev_replace_is_ongoing = 0;
4398 int num_alloc_stripes;
4399 int patch_the_first_stripe_for_dev_replace = 0;
4400 u64 physical_to_patch_in_first_stripe = 0;
4401 u64 raid56_full_stripe_start = (u64)-1;
4403 read_lock(&em_tree->lock);
4404 em = lookup_extent_mapping(em_tree, logical, *length);
4405 read_unlock(&em_tree->lock);
4408 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4409 (unsigned long long)logical,
4410 (unsigned long long)*length);
4414 BUG_ON(em->start > logical || em->start + em->len < logical);
4415 map = (struct map_lookup *)em->bdev;
4416 offset = logical - em->start;
4418 if (mirror_num > map->num_stripes)
4421 stripe_len = map->stripe_len;
4424 * stripe_nr counts the total number of stripes we have to stride
4425 * to get to this block
4427 do_div(stripe_nr, stripe_len);
4429 stripe_offset = stripe_nr * stripe_len;
4430 BUG_ON(offset < stripe_offset);
4432 /* stripe_offset is the offset of this block in its stripe*/
4433 stripe_offset = offset - stripe_offset;
4435 /* if we're here for raid56, we need to know the stripe aligned start */
4436 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4437 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4438 raid56_full_stripe_start = offset;
4440 /* allow a write of a full stripe, but make sure we don't
4441 * allow straddling of stripes
4443 do_div(raid56_full_stripe_start, full_stripe_len);
4444 raid56_full_stripe_start *= full_stripe_len;
4447 if (rw & REQ_DISCARD) {
4448 /* we don't discard raid56 yet */
4450 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4454 *length = min_t(u64, em->len - offset, *length);
4455 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4457 /* For writes to RAID[56], allow a full stripeset across all disks.
4458 For other RAID types and for RAID[56] reads, just allow a single
4459 stripe (on a single disk). */
4460 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4462 max_len = stripe_len * nr_data_stripes(map) -
4463 (offset - raid56_full_stripe_start);
4465 /* we limit the length of each bio to what fits in a stripe */
4466 max_len = stripe_len - stripe_offset;
4468 *length = min_t(u64, em->len - offset, max_len);
4470 *length = em->len - offset;
4473 /* This is for when we're called from btrfs_merge_bio_hook() and all
4474 it cares about is the length */
4478 btrfs_dev_replace_lock(dev_replace);
4479 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4480 if (!dev_replace_is_ongoing)
4481 btrfs_dev_replace_unlock(dev_replace);
4483 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4484 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4485 dev_replace->tgtdev != NULL) {
4487 * in dev-replace case, for repair case (that's the only
4488 * case where the mirror is selected explicitly when
4489 * calling btrfs_map_block), blocks left of the left cursor
4490 * can also be read from the target drive.
4491 * For REQ_GET_READ_MIRRORS, the target drive is added as
4492 * the last one to the array of stripes. For READ, it also
4493 * needs to be supported using the same mirror number.
4494 * If the requested block is not left of the left cursor,
4495 * EIO is returned. This can happen because btrfs_num_copies()
4496 * returns one more in the dev-replace case.
4498 u64 tmp_length = *length;
4499 struct btrfs_bio *tmp_bbio = NULL;
4500 int tmp_num_stripes;
4501 u64 srcdev_devid = dev_replace->srcdev->devid;
4502 int index_srcdev = 0;
4504 u64 physical_of_found = 0;
4506 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4507 logical, &tmp_length, &tmp_bbio, 0, NULL);
4509 WARN_ON(tmp_bbio != NULL);
4513 tmp_num_stripes = tmp_bbio->num_stripes;
4514 if (mirror_num > tmp_num_stripes) {
4516 * REQ_GET_READ_MIRRORS does not contain this
4517 * mirror, that means that the requested area
4518 * is not left of the left cursor
4526 * process the rest of the function using the mirror_num
4527 * of the source drive. Therefore look it up first.
4528 * At the end, patch the device pointer to the one of the
4531 for (i = 0; i < tmp_num_stripes; i++) {
4532 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4534 * In case of DUP, in order to keep it
4535 * simple, only add the mirror with the
4536 * lowest physical address
4539 physical_of_found <=
4540 tmp_bbio->stripes[i].physical)
4545 tmp_bbio->stripes[i].physical;
4550 mirror_num = index_srcdev + 1;
4551 patch_the_first_stripe_for_dev_replace = 1;
4552 physical_to_patch_in_first_stripe = physical_of_found;
4561 } else if (mirror_num > map->num_stripes) {
4567 stripe_nr_orig = stripe_nr;
4568 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4569 do_div(stripe_nr_end, map->stripe_len);
4570 stripe_end_offset = stripe_nr_end * map->stripe_len -
4573 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4574 if (rw & REQ_DISCARD)
4575 num_stripes = min_t(u64, map->num_stripes,
4576 stripe_nr_end - stripe_nr_orig);
4577 stripe_index = do_div(stripe_nr, map->num_stripes);
4578 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4579 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4580 num_stripes = map->num_stripes;
4581 else if (mirror_num)
4582 stripe_index = mirror_num - 1;
4584 stripe_index = find_live_mirror(fs_info, map, 0,
4586 current->pid % map->num_stripes,
4587 dev_replace_is_ongoing);
4588 mirror_num = stripe_index + 1;
4591 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4592 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4593 num_stripes = map->num_stripes;
4594 } else if (mirror_num) {
4595 stripe_index = mirror_num - 1;
4600 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4601 int factor = map->num_stripes / map->sub_stripes;
4603 stripe_index = do_div(stripe_nr, factor);
4604 stripe_index *= map->sub_stripes;
4606 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4607 num_stripes = map->sub_stripes;
4608 else if (rw & REQ_DISCARD)
4609 num_stripes = min_t(u64, map->sub_stripes *
4610 (stripe_nr_end - stripe_nr_orig),
4612 else if (mirror_num)
4613 stripe_index += mirror_num - 1;
4615 int old_stripe_index = stripe_index;
4616 stripe_index = find_live_mirror(fs_info, map,
4618 map->sub_stripes, stripe_index +
4619 current->pid % map->sub_stripes,
4620 dev_replace_is_ongoing);
4621 mirror_num = stripe_index - old_stripe_index + 1;
4624 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4625 BTRFS_BLOCK_GROUP_RAID6)) {
4628 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4632 /* push stripe_nr back to the start of the full stripe */
4633 stripe_nr = raid56_full_stripe_start;
4634 do_div(stripe_nr, stripe_len);
4636 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4638 /* RAID[56] write or recovery. Return all stripes */
4639 num_stripes = map->num_stripes;
4640 max_errors = nr_parity_stripes(map);
4642 raid_map = kmalloc(sizeof(u64) * num_stripes,
4649 /* Work out the disk rotation on this stripe-set */
4651 rot = do_div(tmp, num_stripes);
4653 /* Fill in the logical address of each stripe */
4654 tmp = stripe_nr * nr_data_stripes(map);
4655 for (i = 0; i < nr_data_stripes(map); i++)
4656 raid_map[(i+rot) % num_stripes] =
4657 em->start + (tmp + i) * map->stripe_len;
4659 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4660 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4661 raid_map[(i+rot+1) % num_stripes] =
4664 *length = map->stripe_len;
4669 * Mirror #0 or #1 means the original data block.
4670 * Mirror #2 is RAID5 parity block.
4671 * Mirror #3 is RAID6 Q block.
4673 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4675 stripe_index = nr_data_stripes(map) +
4678 /* We distribute the parity blocks across stripes */
4679 tmp = stripe_nr + stripe_index;
4680 stripe_index = do_div(tmp, map->num_stripes);
4684 * after this do_div call, stripe_nr is the number of stripes
4685 * on this device we have to walk to find the data, and
4686 * stripe_index is the number of our device in the stripe array
4688 stripe_index = do_div(stripe_nr, map->num_stripes);
4689 mirror_num = stripe_index + 1;
4691 BUG_ON(stripe_index >= map->num_stripes);
4693 num_alloc_stripes = num_stripes;
4694 if (dev_replace_is_ongoing) {
4695 if (rw & (REQ_WRITE | REQ_DISCARD))
4696 num_alloc_stripes <<= 1;
4697 if (rw & REQ_GET_READ_MIRRORS)
4698 num_alloc_stripes++;
4700 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4705 atomic_set(&bbio->error, 0);
4707 if (rw & REQ_DISCARD) {
4709 int sub_stripes = 0;
4710 u64 stripes_per_dev = 0;
4711 u32 remaining_stripes = 0;
4712 u32 last_stripe = 0;
4715 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4716 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4719 sub_stripes = map->sub_stripes;
4721 factor = map->num_stripes / sub_stripes;
4722 stripes_per_dev = div_u64_rem(stripe_nr_end -
4725 &remaining_stripes);
4726 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4727 last_stripe *= sub_stripes;
4730 for (i = 0; i < num_stripes; i++) {
4731 bbio->stripes[i].physical =
4732 map->stripes[stripe_index].physical +
4733 stripe_offset + stripe_nr * map->stripe_len;
4734 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4736 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4737 BTRFS_BLOCK_GROUP_RAID10)) {
4738 bbio->stripes[i].length = stripes_per_dev *
4741 if (i / sub_stripes < remaining_stripes)
4742 bbio->stripes[i].length +=
4746 * Special for the first stripe and
4749 * |-------|...|-------|
4753 if (i < sub_stripes)
4754 bbio->stripes[i].length -=
4757 if (stripe_index >= last_stripe &&
4758 stripe_index <= (last_stripe +
4760 bbio->stripes[i].length -=
4763 if (i == sub_stripes - 1)
4766 bbio->stripes[i].length = *length;
4769 if (stripe_index == map->num_stripes) {
4770 /* This could only happen for RAID0/10 */
4776 for (i = 0; i < num_stripes; i++) {
4777 bbio->stripes[i].physical =
4778 map->stripes[stripe_index].physical +
4780 stripe_nr * map->stripe_len;
4781 bbio->stripes[i].dev =
4782 map->stripes[stripe_index].dev;
4787 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4788 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4789 BTRFS_BLOCK_GROUP_RAID10 |
4790 BTRFS_BLOCK_GROUP_RAID5 |
4791 BTRFS_BLOCK_GROUP_DUP)) {
4793 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4798 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4799 dev_replace->tgtdev != NULL) {
4800 int index_where_to_add;
4801 u64 srcdev_devid = dev_replace->srcdev->devid;
4804 * duplicate the write operations while the dev replace
4805 * procedure is running. Since the copying of the old disk
4806 * to the new disk takes place at run time while the
4807 * filesystem is mounted writable, the regular write
4808 * operations to the old disk have to be duplicated to go
4809 * to the new disk as well.
4810 * Note that device->missing is handled by the caller, and
4811 * that the write to the old disk is already set up in the
4814 index_where_to_add = num_stripes;
4815 for (i = 0; i < num_stripes; i++) {
4816 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4817 /* write to new disk, too */
4818 struct btrfs_bio_stripe *new =
4819 bbio->stripes + index_where_to_add;
4820 struct btrfs_bio_stripe *old =
4823 new->physical = old->physical;
4824 new->length = old->length;
4825 new->dev = dev_replace->tgtdev;
4826 index_where_to_add++;
4830 num_stripes = index_where_to_add;
4831 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4832 dev_replace->tgtdev != NULL) {
4833 u64 srcdev_devid = dev_replace->srcdev->devid;
4834 int index_srcdev = 0;
4836 u64 physical_of_found = 0;
4839 * During the dev-replace procedure, the target drive can
4840 * also be used to read data in case it is needed to repair
4841 * a corrupt block elsewhere. This is possible if the
4842 * requested area is left of the left cursor. In this area,
4843 * the target drive is a full copy of the source drive.
4845 for (i = 0; i < num_stripes; i++) {
4846 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4848 * In case of DUP, in order to keep it
4849 * simple, only add the mirror with the
4850 * lowest physical address
4853 physical_of_found <=
4854 bbio->stripes[i].physical)
4858 physical_of_found = bbio->stripes[i].physical;
4862 u64 length = map->stripe_len;
4864 if (physical_of_found + length <=
4865 dev_replace->cursor_left) {
4866 struct btrfs_bio_stripe *tgtdev_stripe =
4867 bbio->stripes + num_stripes;
4869 tgtdev_stripe->physical = physical_of_found;
4870 tgtdev_stripe->length =
4871 bbio->stripes[index_srcdev].length;
4872 tgtdev_stripe->dev = dev_replace->tgtdev;
4880 bbio->num_stripes = num_stripes;
4881 bbio->max_errors = max_errors;
4882 bbio->mirror_num = mirror_num;
4885 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4886 * mirror_num == num_stripes + 1 && dev_replace target drive is
4887 * available as a mirror
4889 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4890 WARN_ON(num_stripes > 1);
4891 bbio->stripes[0].dev = dev_replace->tgtdev;
4892 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4893 bbio->mirror_num = map->num_stripes + 1;
4896 sort_parity_stripes(bbio, raid_map);
4897 *raid_map_ret = raid_map;
4900 if (dev_replace_is_ongoing)
4901 btrfs_dev_replace_unlock(dev_replace);
4902 free_extent_map(em);
4906 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4907 u64 logical, u64 *length,
4908 struct btrfs_bio **bbio_ret, int mirror_num)
4910 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4914 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4915 u64 chunk_start, u64 physical, u64 devid,
4916 u64 **logical, int *naddrs, int *stripe_len)
4918 struct extent_map_tree *em_tree = &map_tree->map_tree;
4919 struct extent_map *em;
4920 struct map_lookup *map;
4928 read_lock(&em_tree->lock);
4929 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4930 read_unlock(&em_tree->lock);
4932 BUG_ON(!em || em->start != chunk_start);
4933 map = (struct map_lookup *)em->bdev;
4936 rmap_len = map->stripe_len;
4938 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4939 do_div(length, map->num_stripes / map->sub_stripes);
4940 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4941 do_div(length, map->num_stripes);
4942 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4943 BTRFS_BLOCK_GROUP_RAID6)) {
4944 do_div(length, nr_data_stripes(map));
4945 rmap_len = map->stripe_len * nr_data_stripes(map);
4948 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4949 BUG_ON(!buf); /* -ENOMEM */
4951 for (i = 0; i < map->num_stripes; i++) {
4952 if (devid && map->stripes[i].dev->devid != devid)
4954 if (map->stripes[i].physical > physical ||
4955 map->stripes[i].physical + length <= physical)
4958 stripe_nr = physical - map->stripes[i].physical;
4959 do_div(stripe_nr, map->stripe_len);
4961 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4962 stripe_nr = stripe_nr * map->num_stripes + i;
4963 do_div(stripe_nr, map->sub_stripes);
4964 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4965 stripe_nr = stripe_nr * map->num_stripes + i;
4966 } /* else if RAID[56], multiply by nr_data_stripes().
4967 * Alternatively, just use rmap_len below instead of
4968 * map->stripe_len */
4970 bytenr = chunk_start + stripe_nr * rmap_len;
4971 WARN_ON(nr >= map->num_stripes);
4972 for (j = 0; j < nr; j++) {
4973 if (buf[j] == bytenr)
4977 WARN_ON(nr >= map->num_stripes);
4984 *stripe_len = rmap_len;
4986 free_extent_map(em);
4990 static void *merge_stripe_index_into_bio_private(void *bi_private,
4991 unsigned int stripe_index)
4994 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4996 * The alternative solution (instead of stealing bits from the
4997 * pointer) would be to allocate an intermediate structure
4998 * that contains the old private pointer plus the stripe_index.
5000 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
5001 BUG_ON(stripe_index > 3);
5002 return (void *)(((uintptr_t)bi_private) | stripe_index);
5005 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
5007 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
5010 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
5012 return (unsigned int)((uintptr_t)bi_private) & 3;
5015 static void btrfs_end_bio(struct bio *bio, int err)
5017 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
5018 int is_orig_bio = 0;
5021 atomic_inc(&bbio->error);
5022 if (err == -EIO || err == -EREMOTEIO) {
5023 unsigned int stripe_index =
5024 extract_stripe_index_from_bio_private(
5026 struct btrfs_device *dev;
5028 BUG_ON(stripe_index >= bbio->num_stripes);
5029 dev = bbio->stripes[stripe_index].dev;
5031 if (bio->bi_rw & WRITE)
5032 btrfs_dev_stat_inc(dev,
5033 BTRFS_DEV_STAT_WRITE_ERRS);
5035 btrfs_dev_stat_inc(dev,
5036 BTRFS_DEV_STAT_READ_ERRS);
5037 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5038 btrfs_dev_stat_inc(dev,
5039 BTRFS_DEV_STAT_FLUSH_ERRS);
5040 btrfs_dev_stat_print_on_error(dev);
5045 if (bio == bbio->orig_bio)
5048 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5051 bio = bbio->orig_bio;
5053 bio->bi_private = bbio->private;
5054 bio->bi_end_io = bbio->end_io;
5055 bio->bi_bdev = (struct block_device *)
5056 (unsigned long)bbio->mirror_num;
5057 /* only send an error to the higher layers if it is
5058 * beyond the tolerance of the btrfs bio
5060 if (atomic_read(&bbio->error) > bbio->max_errors) {
5064 * this bio is actually up to date, we didn't
5065 * go over the max number of errors
5067 set_bit(BIO_UPTODATE, &bio->bi_flags);
5072 bio_endio(bio, err);
5073 } else if (!is_orig_bio) {
5078 struct async_sched {
5081 struct btrfs_fs_info *info;
5082 struct btrfs_work work;
5086 * see run_scheduled_bios for a description of why bios are collected for
5089 * This will add one bio to the pending list for a device and make sure
5090 * the work struct is scheduled.
5092 noinline void btrfs_schedule_bio(struct btrfs_root *root,
5093 struct btrfs_device *device,
5094 int rw, struct bio *bio)
5096 int should_queue = 1;
5097 struct btrfs_pending_bios *pending_bios;
5099 if (device->missing || !device->bdev) {
5100 bio_endio(bio, -EIO);
5104 /* don't bother with additional async steps for reads, right now */
5105 if (!(rw & REQ_WRITE)) {
5107 btrfsic_submit_bio(rw, bio);
5113 * nr_async_bios allows us to reliably return congestion to the
5114 * higher layers. Otherwise, the async bio makes it appear we have
5115 * made progress against dirty pages when we've really just put it
5116 * on a queue for later
5118 atomic_inc(&root->fs_info->nr_async_bios);
5119 WARN_ON(bio->bi_next);
5120 bio->bi_next = NULL;
5123 spin_lock(&device->io_lock);
5124 if (bio->bi_rw & REQ_SYNC)
5125 pending_bios = &device->pending_sync_bios;
5127 pending_bios = &device->pending_bios;
5129 if (pending_bios->tail)
5130 pending_bios->tail->bi_next = bio;
5132 pending_bios->tail = bio;
5133 if (!pending_bios->head)
5134 pending_bios->head = bio;
5135 if (device->running_pending)
5138 spin_unlock(&device->io_lock);
5141 btrfs_queue_worker(&root->fs_info->submit_workers,
5145 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5148 struct bio_vec *prev;
5149 struct request_queue *q = bdev_get_queue(bdev);
5150 unsigned short max_sectors = queue_max_sectors(q);
5151 struct bvec_merge_data bvm = {
5153 .bi_sector = sector,
5154 .bi_rw = bio->bi_rw,
5157 if (bio->bi_vcnt == 0) {
5162 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5163 if ((bio->bi_size >> 9) > max_sectors)
5166 if (!q->merge_bvec_fn)
5169 bvm.bi_size = bio->bi_size - prev->bv_len;
5170 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5175 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5176 struct bio *bio, u64 physical, int dev_nr,
5179 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5181 bio->bi_private = bbio;
5182 bio->bi_private = merge_stripe_index_into_bio_private(
5183 bio->bi_private, (unsigned int)dev_nr);
5184 bio->bi_end_io = btrfs_end_bio;
5185 bio->bi_sector = physical >> 9;
5188 struct rcu_string *name;
5191 name = rcu_dereference(dev->name);
5192 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5193 "(%s id %llu), size=%u\n", rw,
5194 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5195 name->str, dev->devid, bio->bi_size);
5199 bio->bi_bdev = dev->bdev;
5201 btrfs_schedule_bio(root, dev, rw, bio);
5203 btrfsic_submit_bio(rw, bio);
5206 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5207 struct bio *first_bio, struct btrfs_device *dev,
5208 int dev_nr, int rw, int async)
5210 struct bio_vec *bvec = first_bio->bi_io_vec;
5212 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5213 u64 physical = bbio->stripes[dev_nr].physical;
5216 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5220 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5221 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5222 bvec->bv_offset) < bvec->bv_len) {
5223 u64 len = bio->bi_size;
5225 atomic_inc(&bbio->stripes_pending);
5226 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5234 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5238 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5240 atomic_inc(&bbio->error);
5241 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5242 bio->bi_private = bbio->private;
5243 bio->bi_end_io = bbio->end_io;
5244 bio->bi_bdev = (struct block_device *)
5245 (unsigned long)bbio->mirror_num;
5246 bio->bi_sector = logical >> 9;
5248 bio_endio(bio, -EIO);
5252 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5253 int mirror_num, int async_submit)
5255 struct btrfs_device *dev;
5256 struct bio *first_bio = bio;
5257 u64 logical = (u64)bio->bi_sector << 9;
5260 u64 *raid_map = NULL;
5264 struct btrfs_bio *bbio = NULL;
5266 length = bio->bi_size;
5267 map_length = length;
5269 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5270 mirror_num, &raid_map);
5271 if (ret) /* -ENOMEM */
5274 total_devs = bbio->num_stripes;
5275 bbio->orig_bio = first_bio;
5276 bbio->private = first_bio->bi_private;
5277 bbio->end_io = first_bio->bi_end_io;
5278 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5281 /* In this case, map_length has been set to the length of
5282 a single stripe; not the whole write */
5284 return raid56_parity_write(root, bio, bbio,
5285 raid_map, map_length);
5287 return raid56_parity_recover(root, bio, bbio,
5288 raid_map, map_length,
5293 if (map_length < length) {
5294 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
5295 "len %llu\n", (unsigned long long)logical,
5296 (unsigned long long)length,
5297 (unsigned long long)map_length);
5301 while (dev_nr < total_devs) {
5302 dev = bbio->stripes[dev_nr].dev;
5303 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5304 bbio_error(bbio, first_bio, logical);
5310 * Check and see if we're ok with this bio based on it's size
5311 * and offset with the given device.
5313 if (!bio_size_ok(dev->bdev, first_bio,
5314 bbio->stripes[dev_nr].physical >> 9)) {
5315 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5316 dev_nr, rw, async_submit);
5322 if (dev_nr < total_devs - 1) {
5323 bio = bio_clone(first_bio, GFP_NOFS);
5324 BUG_ON(!bio); /* -ENOMEM */
5329 submit_stripe_bio(root, bbio, bio,
5330 bbio->stripes[dev_nr].physical, dev_nr, rw,
5337 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5340 struct btrfs_device *device;
5341 struct btrfs_fs_devices *cur_devices;
5343 cur_devices = fs_info->fs_devices;
5344 while (cur_devices) {
5346 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5347 device = __find_device(&cur_devices->devices,
5352 cur_devices = cur_devices->seed;
5357 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5358 u64 devid, u8 *dev_uuid)
5360 struct btrfs_device *device;
5361 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5363 device = kzalloc(sizeof(*device), GFP_NOFS);
5366 list_add(&device->dev_list,
5367 &fs_devices->devices);
5368 device->dev_root = root->fs_info->dev_root;
5369 device->devid = devid;
5370 device->work.func = pending_bios_fn;
5371 device->fs_devices = fs_devices;
5372 device->missing = 1;
5373 fs_devices->num_devices++;
5374 fs_devices->missing_devices++;
5375 spin_lock_init(&device->io_lock);
5376 INIT_LIST_HEAD(&device->dev_alloc_list);
5377 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5381 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5382 struct extent_buffer *leaf,
5383 struct btrfs_chunk *chunk)
5385 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5386 struct map_lookup *map;
5387 struct extent_map *em;
5391 u8 uuid[BTRFS_UUID_SIZE];
5396 logical = key->offset;
5397 length = btrfs_chunk_length(leaf, chunk);
5399 read_lock(&map_tree->map_tree.lock);
5400 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5401 read_unlock(&map_tree->map_tree.lock);
5403 /* already mapped? */
5404 if (em && em->start <= logical && em->start + em->len > logical) {
5405 free_extent_map(em);
5408 free_extent_map(em);
5411 em = alloc_extent_map();
5414 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5415 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5417 free_extent_map(em);
5421 em->bdev = (struct block_device *)map;
5422 em->start = logical;
5425 em->block_start = 0;
5426 em->block_len = em->len;
5428 map->num_stripes = num_stripes;
5429 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5430 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5431 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5432 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5433 map->type = btrfs_chunk_type(leaf, chunk);
5434 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5435 for (i = 0; i < num_stripes; i++) {
5436 map->stripes[i].physical =
5437 btrfs_stripe_offset_nr(leaf, chunk, i);
5438 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5439 read_extent_buffer(leaf, uuid, (unsigned long)
5440 btrfs_stripe_dev_uuid_nr(chunk, i),
5442 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5444 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5446 free_extent_map(em);
5449 if (!map->stripes[i].dev) {
5450 map->stripes[i].dev =
5451 add_missing_dev(root, devid, uuid);
5452 if (!map->stripes[i].dev) {
5454 free_extent_map(em);
5458 map->stripes[i].dev->in_fs_metadata = 1;
5461 write_lock(&map_tree->map_tree.lock);
5462 ret = add_extent_mapping(&map_tree->map_tree, em);
5463 write_unlock(&map_tree->map_tree.lock);
5464 BUG_ON(ret); /* Tree corruption */
5465 free_extent_map(em);
5470 static void fill_device_from_item(struct extent_buffer *leaf,
5471 struct btrfs_dev_item *dev_item,
5472 struct btrfs_device *device)
5476 device->devid = btrfs_device_id(leaf, dev_item);
5477 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5478 device->total_bytes = device->disk_total_bytes;
5479 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5480 device->type = btrfs_device_type(leaf, dev_item);
5481 device->io_align = btrfs_device_io_align(leaf, dev_item);
5482 device->io_width = btrfs_device_io_width(leaf, dev_item);
5483 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5484 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5485 device->is_tgtdev_for_dev_replace = 0;
5487 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5488 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5491 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5493 struct btrfs_fs_devices *fs_devices;
5496 BUG_ON(!mutex_is_locked(&uuid_mutex));
5498 fs_devices = root->fs_info->fs_devices->seed;
5499 while (fs_devices) {
5500 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5504 fs_devices = fs_devices->seed;
5507 fs_devices = find_fsid(fsid);
5513 fs_devices = clone_fs_devices(fs_devices);
5514 if (IS_ERR(fs_devices)) {
5515 ret = PTR_ERR(fs_devices);
5519 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5520 root->fs_info->bdev_holder);
5522 free_fs_devices(fs_devices);
5526 if (!fs_devices->seeding) {
5527 __btrfs_close_devices(fs_devices);
5528 free_fs_devices(fs_devices);
5533 fs_devices->seed = root->fs_info->fs_devices->seed;
5534 root->fs_info->fs_devices->seed = fs_devices;
5539 static int read_one_dev(struct btrfs_root *root,
5540 struct extent_buffer *leaf,
5541 struct btrfs_dev_item *dev_item)
5543 struct btrfs_device *device;
5546 u8 fs_uuid[BTRFS_UUID_SIZE];
5547 u8 dev_uuid[BTRFS_UUID_SIZE];
5549 devid = btrfs_device_id(leaf, dev_item);
5550 read_extent_buffer(leaf, dev_uuid,
5551 (unsigned long)btrfs_device_uuid(dev_item),
5553 read_extent_buffer(leaf, fs_uuid,
5554 (unsigned long)btrfs_device_fsid(dev_item),
5557 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5558 ret = open_seed_devices(root, fs_uuid);
5559 if (ret && !btrfs_test_opt(root, DEGRADED))
5563 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5564 if (!device || !device->bdev) {
5565 if (!btrfs_test_opt(root, DEGRADED))
5569 printk(KERN_WARNING "warning devid %llu missing\n",
5570 (unsigned long long)devid);
5571 device = add_missing_dev(root, devid, dev_uuid);
5574 } else if (!device->missing) {
5576 * this happens when a device that was properly setup
5577 * in the device info lists suddenly goes bad.
5578 * device->bdev is NULL, and so we have to set
5579 * device->missing to one here
5581 root->fs_info->fs_devices->missing_devices++;
5582 device->missing = 1;
5586 if (device->fs_devices != root->fs_info->fs_devices) {
5587 BUG_ON(device->writeable);
5588 if (device->generation !=
5589 btrfs_device_generation(leaf, dev_item))
5593 fill_device_from_item(leaf, dev_item, device);
5594 device->dev_root = root->fs_info->dev_root;
5595 device->in_fs_metadata = 1;
5596 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5597 device->fs_devices->total_rw_bytes += device->total_bytes;
5598 spin_lock(&root->fs_info->free_chunk_lock);
5599 root->fs_info->free_chunk_space += device->total_bytes -
5601 spin_unlock(&root->fs_info->free_chunk_lock);
5607 int btrfs_read_sys_array(struct btrfs_root *root)
5609 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5610 struct extent_buffer *sb;
5611 struct btrfs_disk_key *disk_key;
5612 struct btrfs_chunk *chunk;
5614 unsigned long sb_ptr;
5620 struct btrfs_key key;
5622 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5623 BTRFS_SUPER_INFO_SIZE);
5626 btrfs_set_buffer_uptodate(sb);
5627 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5629 * The sb extent buffer is artifical and just used to read the system array.
5630 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5631 * pages up-to-date when the page is larger: extent does not cover the
5632 * whole page and consequently check_page_uptodate does not find all
5633 * the page's extents up-to-date (the hole beyond sb),
5634 * write_extent_buffer then triggers a WARN_ON.
5636 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5637 * but sb spans only this function. Add an explicit SetPageUptodate call
5638 * to silence the warning eg. on PowerPC 64.
5640 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5641 SetPageUptodate(sb->pages[0]);
5643 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5644 array_size = btrfs_super_sys_array_size(super_copy);
5646 ptr = super_copy->sys_chunk_array;
5647 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5650 while (cur < array_size) {
5651 disk_key = (struct btrfs_disk_key *)ptr;
5652 btrfs_disk_key_to_cpu(&key, disk_key);
5654 len = sizeof(*disk_key); ptr += len;
5658 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5659 chunk = (struct btrfs_chunk *)sb_ptr;
5660 ret = read_one_chunk(root, &key, sb, chunk);
5663 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5664 len = btrfs_chunk_item_size(num_stripes);
5673 free_extent_buffer(sb);
5677 int btrfs_read_chunk_tree(struct btrfs_root *root)
5679 struct btrfs_path *path;
5680 struct extent_buffer *leaf;
5681 struct btrfs_key key;
5682 struct btrfs_key found_key;
5686 root = root->fs_info->chunk_root;
5688 path = btrfs_alloc_path();
5692 mutex_lock(&uuid_mutex);
5695 /* first we search for all of the device items, and then we
5696 * read in all of the chunk items. This way we can create chunk
5697 * mappings that reference all of the devices that are afound
5699 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5703 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5707 leaf = path->nodes[0];
5708 slot = path->slots[0];
5709 if (slot >= btrfs_header_nritems(leaf)) {
5710 ret = btrfs_next_leaf(root, path);
5717 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5718 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5719 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5721 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5722 struct btrfs_dev_item *dev_item;
5723 dev_item = btrfs_item_ptr(leaf, slot,
5724 struct btrfs_dev_item);
5725 ret = read_one_dev(root, leaf, dev_item);
5729 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5730 struct btrfs_chunk *chunk;
5731 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5732 ret = read_one_chunk(root, &found_key, leaf, chunk);
5738 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5740 btrfs_release_path(path);
5745 unlock_chunks(root);
5746 mutex_unlock(&uuid_mutex);
5748 btrfs_free_path(path);
5752 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5756 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5757 btrfs_dev_stat_reset(dev, i);
5760 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5762 struct btrfs_key key;
5763 struct btrfs_key found_key;
5764 struct btrfs_root *dev_root = fs_info->dev_root;
5765 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5766 struct extent_buffer *eb;
5769 struct btrfs_device *device;
5770 struct btrfs_path *path = NULL;
5773 path = btrfs_alloc_path();
5779 mutex_lock(&fs_devices->device_list_mutex);
5780 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5782 struct btrfs_dev_stats_item *ptr;
5785 key.type = BTRFS_DEV_STATS_KEY;
5786 key.offset = device->devid;
5787 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5789 __btrfs_reset_dev_stats(device);
5790 device->dev_stats_valid = 1;
5791 btrfs_release_path(path);
5794 slot = path->slots[0];
5795 eb = path->nodes[0];
5796 btrfs_item_key_to_cpu(eb, &found_key, slot);
5797 item_size = btrfs_item_size_nr(eb, slot);
5799 ptr = btrfs_item_ptr(eb, slot,
5800 struct btrfs_dev_stats_item);
5802 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5803 if (item_size >= (1 + i) * sizeof(__le64))
5804 btrfs_dev_stat_set(device, i,
5805 btrfs_dev_stats_value(eb, ptr, i));
5807 btrfs_dev_stat_reset(device, i);
5810 device->dev_stats_valid = 1;
5811 btrfs_dev_stat_print_on_load(device);
5812 btrfs_release_path(path);
5814 mutex_unlock(&fs_devices->device_list_mutex);
5817 btrfs_free_path(path);
5818 return ret < 0 ? ret : 0;
5821 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5822 struct btrfs_root *dev_root,
5823 struct btrfs_device *device)
5825 struct btrfs_path *path;
5826 struct btrfs_key key;
5827 struct extent_buffer *eb;
5828 struct btrfs_dev_stats_item *ptr;
5833 key.type = BTRFS_DEV_STATS_KEY;
5834 key.offset = device->devid;
5836 path = btrfs_alloc_path();
5838 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5840 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5841 ret, rcu_str_deref(device->name));
5846 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5847 /* need to delete old one and insert a new one */
5848 ret = btrfs_del_item(trans, dev_root, path);
5850 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5851 rcu_str_deref(device->name), ret);
5858 /* need to insert a new item */
5859 btrfs_release_path(path);
5860 ret = btrfs_insert_empty_item(trans, dev_root, path,
5861 &key, sizeof(*ptr));
5863 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5864 rcu_str_deref(device->name), ret);
5869 eb = path->nodes[0];
5870 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5871 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5872 btrfs_set_dev_stats_value(eb, ptr, i,
5873 btrfs_dev_stat_read(device, i));
5874 btrfs_mark_buffer_dirty(eb);
5877 btrfs_free_path(path);
5882 * called from commit_transaction. Writes all changed device stats to disk.
5884 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5885 struct btrfs_fs_info *fs_info)
5887 struct btrfs_root *dev_root = fs_info->dev_root;
5888 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5889 struct btrfs_device *device;
5892 mutex_lock(&fs_devices->device_list_mutex);
5893 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5894 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5897 ret = update_dev_stat_item(trans, dev_root, device);
5899 device->dev_stats_dirty = 0;
5901 mutex_unlock(&fs_devices->device_list_mutex);
5906 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5908 btrfs_dev_stat_inc(dev, index);
5909 btrfs_dev_stat_print_on_error(dev);
5912 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5914 if (!dev->dev_stats_valid)
5916 printk_ratelimited_in_rcu(KERN_ERR
5917 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5918 rcu_str_deref(dev->name),
5919 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5920 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5921 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5922 btrfs_dev_stat_read(dev,
5923 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5924 btrfs_dev_stat_read(dev,
5925 BTRFS_DEV_STAT_GENERATION_ERRS));
5928 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5932 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5933 if (btrfs_dev_stat_read(dev, i) != 0)
5935 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5936 return; /* all values == 0, suppress message */
5938 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5939 rcu_str_deref(dev->name),
5940 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5941 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5942 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5943 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5944 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5947 int btrfs_get_dev_stats(struct btrfs_root *root,
5948 struct btrfs_ioctl_get_dev_stats *stats)
5950 struct btrfs_device *dev;
5951 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5954 mutex_lock(&fs_devices->device_list_mutex);
5955 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5956 mutex_unlock(&fs_devices->device_list_mutex);
5960 "btrfs: get dev_stats failed, device not found\n");
5962 } else if (!dev->dev_stats_valid) {
5964 "btrfs: get dev_stats failed, not yet valid\n");
5966 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5967 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5968 if (stats->nr_items > i)
5970 btrfs_dev_stat_read_and_reset(dev, i);
5972 btrfs_dev_stat_reset(dev, i);
5975 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5976 if (stats->nr_items > i)
5977 stats->values[i] = btrfs_dev_stat_read(dev, i);
5979 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5980 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5984 int btrfs_scratch_superblock(struct btrfs_device *device)
5986 struct buffer_head *bh;
5987 struct btrfs_super_block *disk_super;
5989 bh = btrfs_read_dev_super(device->bdev);
5992 disk_super = (struct btrfs_super_block *)bh->b_data;
5994 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5995 set_buffer_dirty(bh);
5996 sync_dirty_buffer(bh);