2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
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_error(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
55 static void lock_chunks(struct btrfs_root *root)
57 mutex_lock(&root->fs_info->chunk_mutex);
60 static void unlock_chunks(struct btrfs_root *root)
62 mutex_unlock(&root->fs_info->chunk_mutex);
65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
67 struct btrfs_fs_devices *fs_devs;
69 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
71 return ERR_PTR(-ENOMEM);
73 mutex_init(&fs_devs->device_list_mutex);
75 INIT_LIST_HEAD(&fs_devs->devices);
76 INIT_LIST_HEAD(&fs_devs->alloc_list);
77 INIT_LIST_HEAD(&fs_devs->list);
83 * alloc_fs_devices - allocate struct btrfs_fs_devices
84 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
87 * Return: a pointer to a new &struct btrfs_fs_devices on success;
88 * ERR_PTR() on error. Returned struct is not linked onto any lists and
89 * can be destroyed with kfree() right away.
91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
93 struct btrfs_fs_devices *fs_devs;
95 fs_devs = __alloc_fs_devices();
100 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
102 generate_random_uuid(fs_devs->fsid);
107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
109 struct btrfs_device *device;
110 WARN_ON(fs_devices->opened);
111 while (!list_empty(&fs_devices->devices)) {
112 device = list_entry(fs_devices->devices.next,
113 struct btrfs_device, dev_list);
114 list_del(&device->dev_list);
115 rcu_string_free(device->name);
121 static void btrfs_kobject_uevent(struct block_device *bdev,
122 enum kobject_action action)
126 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
128 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
130 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
131 &disk_to_dev(bdev->bd_disk)->kobj);
134 void btrfs_cleanup_fs_uuids(void)
136 struct btrfs_fs_devices *fs_devices;
138 while (!list_empty(&fs_uuids)) {
139 fs_devices = list_entry(fs_uuids.next,
140 struct btrfs_fs_devices, list);
141 list_del(&fs_devices->list);
142 free_fs_devices(fs_devices);
146 static struct btrfs_device *__alloc_device(void)
148 struct btrfs_device *dev;
150 dev = kzalloc(sizeof(*dev), GFP_NOFS);
152 return ERR_PTR(-ENOMEM);
154 INIT_LIST_HEAD(&dev->dev_list);
155 INIT_LIST_HEAD(&dev->dev_alloc_list);
157 spin_lock_init(&dev->io_lock);
159 spin_lock_init(&dev->reada_lock);
160 atomic_set(&dev->reada_in_flight, 0);
161 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
162 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
167 static noinline struct btrfs_device *__find_device(struct list_head *head,
170 struct btrfs_device *dev;
172 list_for_each_entry(dev, head, dev_list) {
173 if (dev->devid == devid &&
174 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
183 struct btrfs_fs_devices *fs_devices;
185 list_for_each_entry(fs_devices, &fs_uuids, list) {
186 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
194 int flush, struct block_device **bdev,
195 struct buffer_head **bh)
199 *bdev = blkdev_get_by_path(device_path, flags, holder);
202 ret = PTR_ERR(*bdev);
203 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
208 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
209 ret = set_blocksize(*bdev, 4096);
211 blkdev_put(*bdev, flags);
214 invalidate_bdev(*bdev);
215 *bh = btrfs_read_dev_super(*bdev);
218 blkdev_put(*bdev, flags);
230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
231 struct bio *head, struct bio *tail)
234 struct bio *old_head;
236 old_head = pending_bios->head;
237 pending_bios->head = head;
238 if (pending_bios->tail)
239 tail->bi_next = old_head;
241 pending_bios->tail = tail;
245 * we try to collect pending bios for a device so we don't get a large
246 * number of procs sending bios down to the same device. This greatly
247 * improves the schedulers ability to collect and merge the bios.
249 * But, it also turns into a long list of bios to process and that is sure
250 * to eventually make the worker thread block. The solution here is to
251 * make some progress and then put this work struct back at the end of
252 * the list if the block device is congested. This way, multiple devices
253 * can make progress from a single worker thread.
255 static noinline void run_scheduled_bios(struct btrfs_device *device)
258 struct backing_dev_info *bdi;
259 struct btrfs_fs_info *fs_info;
260 struct btrfs_pending_bios *pending_bios;
264 unsigned long num_run;
265 unsigned long batch_run = 0;
267 unsigned long last_waited = 0;
269 int sync_pending = 0;
270 struct blk_plug plug;
273 * this function runs all the bios we've collected for
274 * a particular device. We don't want to wander off to
275 * another device without first sending all of these down.
276 * So, setup a plug here and finish it off before we return
278 blk_start_plug(&plug);
280 bdi = blk_get_backing_dev_info(device->bdev);
281 fs_info = device->dev_root->fs_info;
282 limit = btrfs_async_submit_limit(fs_info);
283 limit = limit * 2 / 3;
286 spin_lock(&device->io_lock);
291 /* take all the bios off the list at once and process them
292 * later on (without the lock held). But, remember the
293 * tail and other pointers so the bios can be properly reinserted
294 * into the list if we hit congestion
296 if (!force_reg && device->pending_sync_bios.head) {
297 pending_bios = &device->pending_sync_bios;
300 pending_bios = &device->pending_bios;
304 pending = pending_bios->head;
305 tail = pending_bios->tail;
306 WARN_ON(pending && !tail);
309 * if pending was null this time around, no bios need processing
310 * at all and we can stop. Otherwise it'll loop back up again
311 * and do an additional check so no bios are missed.
313 * device->running_pending is used to synchronize with the
316 if (device->pending_sync_bios.head == NULL &&
317 device->pending_bios.head == NULL) {
319 device->running_pending = 0;
322 device->running_pending = 1;
325 pending_bios->head = NULL;
326 pending_bios->tail = NULL;
328 spin_unlock(&device->io_lock);
333 /* we want to work on both lists, but do more bios on the
334 * sync list than the regular list
337 pending_bios != &device->pending_sync_bios &&
338 device->pending_sync_bios.head) ||
339 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
340 device->pending_bios.head)) {
341 spin_lock(&device->io_lock);
342 requeue_list(pending_bios, pending, tail);
347 pending = pending->bi_next;
350 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
351 waitqueue_active(&fs_info->async_submit_wait))
352 wake_up(&fs_info->async_submit_wait);
354 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
357 * if we're doing the sync list, record that our
358 * plug has some sync requests on it
360 * If we're doing the regular list and there are
361 * sync requests sitting around, unplug before
364 if (pending_bios == &device->pending_sync_bios) {
366 } else if (sync_pending) {
367 blk_finish_plug(&plug);
368 blk_start_plug(&plug);
372 btrfsic_submit_bio(cur->bi_rw, cur);
379 * we made progress, there is more work to do and the bdi
380 * is now congested. Back off and let other work structs
383 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
384 fs_info->fs_devices->open_devices > 1) {
385 struct io_context *ioc;
387 ioc = current->io_context;
390 * the main goal here is that we don't want to
391 * block if we're going to be able to submit
392 * more requests without blocking.
394 * This code does two great things, it pokes into
395 * the elevator code from a filesystem _and_
396 * it makes assumptions about how batching works.
398 if (ioc && ioc->nr_batch_requests > 0 &&
399 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
401 ioc->last_waited == last_waited)) {
403 * we want to go through our batch of
404 * requests and stop. So, we copy out
405 * the ioc->last_waited time and test
406 * against it before looping
408 last_waited = ioc->last_waited;
413 spin_lock(&device->io_lock);
414 requeue_list(pending_bios, pending, tail);
415 device->running_pending = 1;
417 spin_unlock(&device->io_lock);
418 btrfs_queue_work(fs_info->submit_workers,
422 /* unplug every 64 requests just for good measure */
423 if (batch_run % 64 == 0) {
424 blk_finish_plug(&plug);
425 blk_start_plug(&plug);
434 spin_lock(&device->io_lock);
435 if (device->pending_bios.head || device->pending_sync_bios.head)
437 spin_unlock(&device->io_lock);
440 blk_finish_plug(&plug);
443 static void pending_bios_fn(struct btrfs_work *work)
445 struct btrfs_device *device;
447 device = container_of(work, struct btrfs_device, work);
448 run_scheduled_bios(device);
452 * Add new device to list of registered devices
455 * 1 - first time device is seen
456 * 0 - device already known
459 static noinline int device_list_add(const char *path,
460 struct btrfs_super_block *disk_super,
461 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
463 struct btrfs_device *device;
464 struct btrfs_fs_devices *fs_devices;
465 struct rcu_string *name;
467 u64 found_transid = btrfs_super_generation(disk_super);
469 fs_devices = find_fsid(disk_super->fsid);
471 fs_devices = alloc_fs_devices(disk_super->fsid);
472 if (IS_ERR(fs_devices))
473 return PTR_ERR(fs_devices);
475 list_add(&fs_devices->list, &fs_uuids);
476 fs_devices->latest_devid = devid;
477 fs_devices->latest_trans = found_transid;
481 device = __find_device(&fs_devices->devices, devid,
482 disk_super->dev_item.uuid);
485 if (fs_devices->opened)
488 device = btrfs_alloc_device(NULL, &devid,
489 disk_super->dev_item.uuid);
490 if (IS_ERR(device)) {
491 /* we can safely leave the fs_devices entry around */
492 return PTR_ERR(device);
495 name = rcu_string_strdup(path, GFP_NOFS);
500 rcu_assign_pointer(device->name, name);
502 mutex_lock(&fs_devices->device_list_mutex);
503 list_add_rcu(&device->dev_list, &fs_devices->devices);
504 fs_devices->num_devices++;
505 mutex_unlock(&fs_devices->device_list_mutex);
508 device->fs_devices = fs_devices;
509 } else if (!device->name || strcmp(device->name->str, path)) {
510 name = rcu_string_strdup(path, GFP_NOFS);
513 rcu_string_free(device->name);
514 rcu_assign_pointer(device->name, name);
515 if (device->missing) {
516 fs_devices->missing_devices--;
521 if (found_transid > fs_devices->latest_trans) {
522 fs_devices->latest_devid = devid;
523 fs_devices->latest_trans = found_transid;
525 *fs_devices_ret = fs_devices;
530 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
532 struct btrfs_fs_devices *fs_devices;
533 struct btrfs_device *device;
534 struct btrfs_device *orig_dev;
536 fs_devices = alloc_fs_devices(orig->fsid);
537 if (IS_ERR(fs_devices))
540 fs_devices->latest_devid = orig->latest_devid;
541 fs_devices->latest_trans = orig->latest_trans;
542 fs_devices->total_devices = orig->total_devices;
544 /* We have held the volume lock, it is safe to get the devices. */
545 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
546 struct rcu_string *name;
548 device = btrfs_alloc_device(NULL, &orig_dev->devid,
554 * This is ok to do without rcu read locked because we hold the
555 * uuid mutex so nothing we touch in here is going to disappear.
557 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
562 rcu_assign_pointer(device->name, name);
564 list_add(&device->dev_list, &fs_devices->devices);
565 device->fs_devices = fs_devices;
566 fs_devices->num_devices++;
570 free_fs_devices(fs_devices);
571 return ERR_PTR(-ENOMEM);
574 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
575 struct btrfs_fs_devices *fs_devices, int step)
577 struct btrfs_device *device, *next;
579 struct block_device *latest_bdev = NULL;
580 u64 latest_devid = 0;
581 u64 latest_transid = 0;
583 mutex_lock(&uuid_mutex);
585 /* This is the initialized path, it is safe to release the devices. */
586 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
587 if (device->in_fs_metadata) {
588 if (!device->is_tgtdev_for_dev_replace &&
590 device->generation > latest_transid)) {
591 latest_devid = device->devid;
592 latest_transid = device->generation;
593 latest_bdev = device->bdev;
598 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
600 * In the first step, keep the device which has
601 * the correct fsid and the devid that is used
602 * for the dev_replace procedure.
603 * In the second step, the dev_replace state is
604 * read from the device tree and it is known
605 * whether the procedure is really active or
606 * not, which means whether this device is
607 * used or whether it should be removed.
609 if (step == 0 || device->is_tgtdev_for_dev_replace) {
614 blkdev_put(device->bdev, device->mode);
616 fs_devices->open_devices--;
618 if (device->writeable) {
619 list_del_init(&device->dev_alloc_list);
620 device->writeable = 0;
621 if (!device->is_tgtdev_for_dev_replace)
622 fs_devices->rw_devices--;
624 list_del_init(&device->dev_list);
625 fs_devices->num_devices--;
626 rcu_string_free(device->name);
630 if (fs_devices->seed) {
631 fs_devices = fs_devices->seed;
635 fs_devices->latest_bdev = latest_bdev;
636 fs_devices->latest_devid = latest_devid;
637 fs_devices->latest_trans = latest_transid;
639 mutex_unlock(&uuid_mutex);
642 static void __free_device(struct work_struct *work)
644 struct btrfs_device *device;
646 device = container_of(work, struct btrfs_device, rcu_work);
649 blkdev_put(device->bdev, device->mode);
651 rcu_string_free(device->name);
655 static void free_device(struct rcu_head *head)
657 struct btrfs_device *device;
659 device = container_of(head, struct btrfs_device, rcu);
661 INIT_WORK(&device->rcu_work, __free_device);
662 schedule_work(&device->rcu_work);
665 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
667 struct btrfs_device *device;
669 if (--fs_devices->opened > 0)
672 mutex_lock(&fs_devices->device_list_mutex);
673 list_for_each_entry(device, &fs_devices->devices, dev_list) {
674 struct btrfs_device *new_device;
675 struct rcu_string *name;
678 fs_devices->open_devices--;
680 if (device->writeable &&
681 device->devid != BTRFS_DEV_REPLACE_DEVID) {
682 list_del_init(&device->dev_alloc_list);
683 fs_devices->rw_devices--;
686 if (device->can_discard)
687 fs_devices->num_can_discard--;
689 fs_devices->missing_devices--;
691 new_device = btrfs_alloc_device(NULL, &device->devid,
693 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
695 /* Safe because we are under uuid_mutex */
697 name = rcu_string_strdup(device->name->str, GFP_NOFS);
698 BUG_ON(!name); /* -ENOMEM */
699 rcu_assign_pointer(new_device->name, name);
702 list_replace_rcu(&device->dev_list, &new_device->dev_list);
703 new_device->fs_devices = device->fs_devices;
705 call_rcu(&device->rcu, free_device);
707 mutex_unlock(&fs_devices->device_list_mutex);
709 WARN_ON(fs_devices->open_devices);
710 WARN_ON(fs_devices->rw_devices);
711 fs_devices->opened = 0;
712 fs_devices->seeding = 0;
717 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
719 struct btrfs_fs_devices *seed_devices = NULL;
722 mutex_lock(&uuid_mutex);
723 ret = __btrfs_close_devices(fs_devices);
724 if (!fs_devices->opened) {
725 seed_devices = fs_devices->seed;
726 fs_devices->seed = NULL;
728 mutex_unlock(&uuid_mutex);
730 while (seed_devices) {
731 fs_devices = seed_devices;
732 seed_devices = fs_devices->seed;
733 __btrfs_close_devices(fs_devices);
734 free_fs_devices(fs_devices);
737 * Wait for rcu kworkers under __btrfs_close_devices
738 * to finish all blkdev_puts so device is really
739 * free when umount is done.
745 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
746 fmode_t flags, void *holder)
748 struct request_queue *q;
749 struct block_device *bdev;
750 struct list_head *head = &fs_devices->devices;
751 struct btrfs_device *device;
752 struct block_device *latest_bdev = NULL;
753 struct buffer_head *bh;
754 struct btrfs_super_block *disk_super;
755 u64 latest_devid = 0;
756 u64 latest_transid = 0;
763 list_for_each_entry(device, head, dev_list) {
769 /* Just open everything we can; ignore failures here */
770 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
774 disk_super = (struct btrfs_super_block *)bh->b_data;
775 devid = btrfs_stack_device_id(&disk_super->dev_item);
776 if (devid != device->devid)
779 if (memcmp(device->uuid, disk_super->dev_item.uuid,
783 device->generation = btrfs_super_generation(disk_super);
784 if (!latest_transid || device->generation > latest_transid) {
785 latest_devid = devid;
786 latest_transid = device->generation;
790 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
791 device->writeable = 0;
793 device->writeable = !bdev_read_only(bdev);
797 q = bdev_get_queue(bdev);
798 if (blk_queue_discard(q)) {
799 device->can_discard = 1;
800 fs_devices->num_can_discard++;
804 device->in_fs_metadata = 0;
805 device->mode = flags;
807 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
808 fs_devices->rotating = 1;
810 fs_devices->open_devices++;
811 if (device->writeable &&
812 device->devid != BTRFS_DEV_REPLACE_DEVID) {
813 fs_devices->rw_devices++;
814 list_add(&device->dev_alloc_list,
815 &fs_devices->alloc_list);
822 blkdev_put(bdev, flags);
825 if (fs_devices->open_devices == 0) {
829 fs_devices->seeding = seeding;
830 fs_devices->opened = 1;
831 fs_devices->latest_bdev = latest_bdev;
832 fs_devices->latest_devid = latest_devid;
833 fs_devices->latest_trans = latest_transid;
834 fs_devices->total_rw_bytes = 0;
839 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
840 fmode_t flags, void *holder)
844 mutex_lock(&uuid_mutex);
845 if (fs_devices->opened) {
846 fs_devices->opened++;
849 ret = __btrfs_open_devices(fs_devices, flags, holder);
851 mutex_unlock(&uuid_mutex);
856 * Look for a btrfs signature on a device. This may be called out of the mount path
857 * and we are not allowed to call set_blocksize during the scan. The superblock
858 * is read via pagecache
860 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
861 struct btrfs_fs_devices **fs_devices_ret)
863 struct btrfs_super_block *disk_super;
864 struct block_device *bdev;
875 * we would like to check all the supers, but that would make
876 * a btrfs mount succeed after a mkfs from a different FS.
877 * So, we need to add a special mount option to scan for
878 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
880 bytenr = btrfs_sb_offset(0);
882 mutex_lock(&uuid_mutex);
884 bdev = blkdev_get_by_path(path, flags, holder);
891 /* make sure our super fits in the device */
892 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
895 /* make sure our super fits in the page */
896 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
899 /* make sure our super doesn't straddle pages on disk */
900 index = bytenr >> PAGE_CACHE_SHIFT;
901 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
904 /* pull in the page with our super */
905 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
908 if (IS_ERR_OR_NULL(page))
913 /* align our pointer to the offset of the super block */
914 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
916 if (btrfs_super_bytenr(disk_super) != bytenr ||
917 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
920 devid = btrfs_stack_device_id(&disk_super->dev_item);
921 transid = btrfs_super_generation(disk_super);
922 total_devices = btrfs_super_num_devices(disk_super);
924 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
926 if (disk_super->label[0]) {
927 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
928 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
929 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
931 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
934 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
937 if (!ret && fs_devices_ret)
938 (*fs_devices_ret)->total_devices = total_devices;
942 page_cache_release(page);
945 blkdev_put(bdev, flags);
947 mutex_unlock(&uuid_mutex);
951 /* helper to account the used device space in the range */
952 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
953 u64 end, u64 *length)
955 struct btrfs_key key;
956 struct btrfs_root *root = device->dev_root;
957 struct btrfs_dev_extent *dev_extent;
958 struct btrfs_path *path;
962 struct extent_buffer *l;
966 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
969 path = btrfs_alloc_path();
974 key.objectid = device->devid;
976 key.type = BTRFS_DEV_EXTENT_KEY;
978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
982 ret = btrfs_previous_item(root, path, key.objectid, key.type);
989 slot = path->slots[0];
990 if (slot >= btrfs_header_nritems(l)) {
991 ret = btrfs_next_leaf(root, path);
999 btrfs_item_key_to_cpu(l, &key, slot);
1001 if (key.objectid < device->devid)
1004 if (key.objectid > device->devid)
1007 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1010 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1011 extent_end = key.offset + btrfs_dev_extent_length(l,
1013 if (key.offset <= start && extent_end > end) {
1014 *length = end - start + 1;
1016 } else if (key.offset <= start && extent_end > start)
1017 *length += extent_end - start;
1018 else if (key.offset > start && extent_end <= end)
1019 *length += extent_end - key.offset;
1020 else if (key.offset > start && key.offset <= end) {
1021 *length += end - key.offset + 1;
1023 } else if (key.offset > end)
1031 btrfs_free_path(path);
1035 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1036 struct btrfs_device *device,
1037 u64 *start, u64 len)
1039 struct extent_map *em;
1042 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1043 struct map_lookup *map;
1046 map = (struct map_lookup *)em->bdev;
1047 for (i = 0; i < map->num_stripes; i++) {
1048 if (map->stripes[i].dev != device)
1050 if (map->stripes[i].physical >= *start + len ||
1051 map->stripes[i].physical + em->orig_block_len <=
1054 *start = map->stripes[i].physical +
1065 * find_free_dev_extent - find free space in the specified device
1066 * @device: the device which we search the free space in
1067 * @num_bytes: the size of the free space that we need
1068 * @start: store the start of the free space.
1069 * @len: the size of the free space. that we find, or the size of the max
1070 * free space if we don't find suitable free space
1072 * this uses a pretty simple search, the expectation is that it is
1073 * called very infrequently and that a given device has a small number
1076 * @start is used to store the start of the free space if we find. But if we
1077 * don't find suitable free space, it will be used to store the start position
1078 * of the max free space.
1080 * @len is used to store the size of the free space that we find.
1081 * But if we don't find suitable free space, it is used to store the size of
1082 * the max free space.
1084 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1085 struct btrfs_device *device, u64 num_bytes,
1086 u64 *start, u64 *len)
1088 struct btrfs_key key;
1089 struct btrfs_root *root = device->dev_root;
1090 struct btrfs_dev_extent *dev_extent;
1091 struct btrfs_path *path;
1097 u64 search_end = device->total_bytes;
1100 struct extent_buffer *l;
1102 /* FIXME use last free of some kind */
1104 /* we don't want to overwrite the superblock on the drive,
1105 * so we make sure to start at an offset of at least 1MB
1107 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1109 path = btrfs_alloc_path();
1113 max_hole_start = search_start;
1117 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1123 path->search_commit_root = 1;
1124 path->skip_locking = 1;
1126 key.objectid = device->devid;
1127 key.offset = search_start;
1128 key.type = BTRFS_DEV_EXTENT_KEY;
1130 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1134 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1141 slot = path->slots[0];
1142 if (slot >= btrfs_header_nritems(l)) {
1143 ret = btrfs_next_leaf(root, path);
1151 btrfs_item_key_to_cpu(l, &key, slot);
1153 if (key.objectid < device->devid)
1156 if (key.objectid > device->devid)
1159 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1162 if (key.offset > search_start) {
1163 hole_size = key.offset - search_start;
1166 * Have to check before we set max_hole_start, otherwise
1167 * we could end up sending back this offset anyway.
1169 if (contains_pending_extent(trans, device,
1174 if (hole_size > max_hole_size) {
1175 max_hole_start = search_start;
1176 max_hole_size = hole_size;
1180 * If this free space is greater than which we need,
1181 * it must be the max free space that we have found
1182 * until now, so max_hole_start must point to the start
1183 * of this free space and the length of this free space
1184 * is stored in max_hole_size. Thus, we return
1185 * max_hole_start and max_hole_size and go back to the
1188 if (hole_size >= num_bytes) {
1194 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1195 extent_end = key.offset + btrfs_dev_extent_length(l,
1197 if (extent_end > search_start)
1198 search_start = extent_end;
1205 * At this point, search_start should be the end of
1206 * allocated dev extents, and when shrinking the device,
1207 * search_end may be smaller than search_start.
1209 if (search_end > search_start)
1210 hole_size = search_end - search_start;
1212 if (hole_size > max_hole_size) {
1213 max_hole_start = search_start;
1214 max_hole_size = hole_size;
1217 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1218 btrfs_release_path(path);
1223 if (hole_size < num_bytes)
1229 btrfs_free_path(path);
1230 *start = max_hole_start;
1232 *len = max_hole_size;
1236 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1237 struct btrfs_device *device,
1241 struct btrfs_path *path;
1242 struct btrfs_root *root = device->dev_root;
1243 struct btrfs_key key;
1244 struct btrfs_key found_key;
1245 struct extent_buffer *leaf = NULL;
1246 struct btrfs_dev_extent *extent = NULL;
1248 path = btrfs_alloc_path();
1252 key.objectid = device->devid;
1254 key.type = BTRFS_DEV_EXTENT_KEY;
1256 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1258 ret = btrfs_previous_item(root, path, key.objectid,
1259 BTRFS_DEV_EXTENT_KEY);
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1264 extent = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_dev_extent);
1266 BUG_ON(found_key.offset > start || found_key.offset +
1267 btrfs_dev_extent_length(leaf, extent) < start);
1269 btrfs_release_path(path);
1271 } else if (ret == 0) {
1272 leaf = path->nodes[0];
1273 extent = btrfs_item_ptr(leaf, path->slots[0],
1274 struct btrfs_dev_extent);
1276 btrfs_error(root->fs_info, ret, "Slot search failed");
1280 if (device->bytes_used > 0) {
1281 u64 len = btrfs_dev_extent_length(leaf, extent);
1282 device->bytes_used -= len;
1283 spin_lock(&root->fs_info->free_chunk_lock);
1284 root->fs_info->free_chunk_space += len;
1285 spin_unlock(&root->fs_info->free_chunk_lock);
1287 ret = btrfs_del_item(trans, root, path);
1289 btrfs_error(root->fs_info, ret,
1290 "Failed to remove dev extent item");
1293 btrfs_free_path(path);
1297 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1298 struct btrfs_device *device,
1299 u64 chunk_tree, u64 chunk_objectid,
1300 u64 chunk_offset, u64 start, u64 num_bytes)
1303 struct btrfs_path *path;
1304 struct btrfs_root *root = device->dev_root;
1305 struct btrfs_dev_extent *extent;
1306 struct extent_buffer *leaf;
1307 struct btrfs_key key;
1309 WARN_ON(!device->in_fs_metadata);
1310 WARN_ON(device->is_tgtdev_for_dev_replace);
1311 path = btrfs_alloc_path();
1315 key.objectid = device->devid;
1317 key.type = BTRFS_DEV_EXTENT_KEY;
1318 ret = btrfs_insert_empty_item(trans, root, path, &key,
1323 leaf = path->nodes[0];
1324 extent = btrfs_item_ptr(leaf, path->slots[0],
1325 struct btrfs_dev_extent);
1326 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1327 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1328 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1330 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1331 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1333 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1334 btrfs_mark_buffer_dirty(leaf);
1336 btrfs_free_path(path);
1340 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1342 struct extent_map_tree *em_tree;
1343 struct extent_map *em;
1347 em_tree = &fs_info->mapping_tree.map_tree;
1348 read_lock(&em_tree->lock);
1349 n = rb_last(&em_tree->map);
1351 em = rb_entry(n, struct extent_map, rb_node);
1352 ret = em->start + em->len;
1354 read_unlock(&em_tree->lock);
1359 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1363 struct btrfs_key key;
1364 struct btrfs_key found_key;
1365 struct btrfs_path *path;
1367 path = btrfs_alloc_path();
1371 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1372 key.type = BTRFS_DEV_ITEM_KEY;
1373 key.offset = (u64)-1;
1375 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1379 BUG_ON(ret == 0); /* Corruption */
1381 ret = btrfs_previous_item(fs_info->chunk_root, path,
1382 BTRFS_DEV_ITEMS_OBJECTID,
1383 BTRFS_DEV_ITEM_KEY);
1387 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1389 *devid_ret = found_key.offset + 1;
1393 btrfs_free_path(path);
1398 * the device information is stored in the chunk root
1399 * the btrfs_device struct should be fully filled in
1401 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_device *device)
1406 struct btrfs_path *path;
1407 struct btrfs_dev_item *dev_item;
1408 struct extent_buffer *leaf;
1409 struct btrfs_key key;
1412 root = root->fs_info->chunk_root;
1414 path = btrfs_alloc_path();
1418 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1419 key.type = BTRFS_DEV_ITEM_KEY;
1420 key.offset = device->devid;
1422 ret = btrfs_insert_empty_item(trans, root, path, &key,
1427 leaf = path->nodes[0];
1428 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1430 btrfs_set_device_id(leaf, dev_item, device->devid);
1431 btrfs_set_device_generation(leaf, dev_item, 0);
1432 btrfs_set_device_type(leaf, dev_item, device->type);
1433 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438 btrfs_set_device_group(leaf, dev_item, 0);
1439 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1440 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1441 btrfs_set_device_start_offset(leaf, dev_item, 0);
1443 ptr = btrfs_device_uuid(dev_item);
1444 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1445 ptr = btrfs_device_fsid(dev_item);
1446 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1447 btrfs_mark_buffer_dirty(leaf);
1451 btrfs_free_path(path);
1456 * Function to update ctime/mtime for a given device path.
1457 * Mainly used for ctime/mtime based probe like libblkid.
1459 static void update_dev_time(char *path_name)
1463 filp = filp_open(path_name, O_RDWR, 0);
1466 file_update_time(filp);
1467 filp_close(filp, NULL);
1471 static int btrfs_rm_dev_item(struct btrfs_root *root,
1472 struct btrfs_device *device)
1475 struct btrfs_path *path;
1476 struct btrfs_key key;
1477 struct btrfs_trans_handle *trans;
1479 root = root->fs_info->chunk_root;
1481 path = btrfs_alloc_path();
1485 trans = btrfs_start_transaction(root, 0);
1486 if (IS_ERR(trans)) {
1487 btrfs_free_path(path);
1488 return PTR_ERR(trans);
1490 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1491 key.type = BTRFS_DEV_ITEM_KEY;
1492 key.offset = device->devid;
1495 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1504 ret = btrfs_del_item(trans, root, path);
1508 btrfs_free_path(path);
1509 unlock_chunks(root);
1510 btrfs_commit_transaction(trans, root);
1514 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1516 struct btrfs_device *device;
1517 struct btrfs_device *next_device;
1518 struct block_device *bdev;
1519 struct buffer_head *bh = NULL;
1520 struct btrfs_super_block *disk_super;
1521 struct btrfs_fs_devices *cur_devices;
1528 bool clear_super = false;
1530 mutex_lock(&uuid_mutex);
1533 seq = read_seqbegin(&root->fs_info->profiles_lock);
1535 all_avail = root->fs_info->avail_data_alloc_bits |
1536 root->fs_info->avail_system_alloc_bits |
1537 root->fs_info->avail_metadata_alloc_bits;
1538 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1540 num_devices = root->fs_info->fs_devices->num_devices;
1541 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1542 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1543 WARN_ON(num_devices < 1);
1546 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1548 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1549 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1553 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1554 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1558 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1559 root->fs_info->fs_devices->rw_devices <= 2) {
1560 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1563 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1564 root->fs_info->fs_devices->rw_devices <= 3) {
1565 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1569 if (strcmp(device_path, "missing") == 0) {
1570 struct list_head *devices;
1571 struct btrfs_device *tmp;
1574 devices = &root->fs_info->fs_devices->devices;
1576 * It is safe to read the devices since the volume_mutex
1579 list_for_each_entry(tmp, devices, dev_list) {
1580 if (tmp->in_fs_metadata &&
1581 !tmp->is_tgtdev_for_dev_replace &&
1591 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1595 ret = btrfs_get_bdev_and_sb(device_path,
1596 FMODE_WRITE | FMODE_EXCL,
1597 root->fs_info->bdev_holder, 0,
1601 disk_super = (struct btrfs_super_block *)bh->b_data;
1602 devid = btrfs_stack_device_id(&disk_super->dev_item);
1603 dev_uuid = disk_super->dev_item.uuid;
1604 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1612 if (device->is_tgtdev_for_dev_replace) {
1613 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1617 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1618 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1622 if (device->writeable) {
1624 list_del_init(&device->dev_alloc_list);
1625 unlock_chunks(root);
1626 root->fs_info->fs_devices->rw_devices--;
1630 mutex_unlock(&uuid_mutex);
1631 ret = btrfs_shrink_device(device, 0);
1632 mutex_lock(&uuid_mutex);
1637 * TODO: the superblock still includes this device in its num_devices
1638 * counter although write_all_supers() is not locked out. This
1639 * could give a filesystem state which requires a degraded mount.
1641 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1645 spin_lock(&root->fs_info->free_chunk_lock);
1646 root->fs_info->free_chunk_space = device->total_bytes -
1648 spin_unlock(&root->fs_info->free_chunk_lock);
1650 device->in_fs_metadata = 0;
1651 btrfs_scrub_cancel_dev(root->fs_info, device);
1654 * the device list mutex makes sure that we don't change
1655 * the device list while someone else is writing out all
1656 * the device supers. Whoever is writing all supers, should
1657 * lock the device list mutex before getting the number of
1658 * devices in the super block (super_copy). Conversely,
1659 * whoever updates the number of devices in the super block
1660 * (super_copy) should hold the device list mutex.
1663 cur_devices = device->fs_devices;
1664 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1665 list_del_rcu(&device->dev_list);
1667 device->fs_devices->num_devices--;
1668 device->fs_devices->total_devices--;
1670 if (device->missing)
1671 root->fs_info->fs_devices->missing_devices--;
1673 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1674 struct btrfs_device, dev_list);
1675 if (device->bdev == root->fs_info->sb->s_bdev)
1676 root->fs_info->sb->s_bdev = next_device->bdev;
1677 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1678 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1681 device->fs_devices->open_devices--;
1683 call_rcu(&device->rcu, free_device);
1685 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1686 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1687 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1689 if (cur_devices->open_devices == 0) {
1690 struct btrfs_fs_devices *fs_devices;
1691 fs_devices = root->fs_info->fs_devices;
1692 while (fs_devices) {
1693 if (fs_devices->seed == cur_devices) {
1694 fs_devices->seed = cur_devices->seed;
1697 fs_devices = fs_devices->seed;
1699 cur_devices->seed = NULL;
1701 __btrfs_close_devices(cur_devices);
1702 unlock_chunks(root);
1703 free_fs_devices(cur_devices);
1706 root->fs_info->num_tolerated_disk_barrier_failures =
1707 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1710 * at this point, the device is zero sized. We want to
1711 * remove it from the devices list and zero out the old super
1713 if (clear_super && disk_super) {
1717 /* make sure this device isn't detected as part of
1720 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1721 set_buffer_dirty(bh);
1722 sync_dirty_buffer(bh);
1724 /* clear the mirror copies of super block on the disk
1725 * being removed, 0th copy is been taken care above and
1726 * the below would take of the rest
1728 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1729 bytenr = btrfs_sb_offset(i);
1730 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1731 i_size_read(bdev->bd_inode))
1735 bh = __bread(bdev, bytenr / 4096,
1736 BTRFS_SUPER_INFO_SIZE);
1740 disk_super = (struct btrfs_super_block *)bh->b_data;
1742 if (btrfs_super_bytenr(disk_super) != bytenr ||
1743 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1746 memset(&disk_super->magic, 0,
1747 sizeof(disk_super->magic));
1748 set_buffer_dirty(bh);
1749 sync_dirty_buffer(bh);
1756 /* Notify udev that device has changed */
1757 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1759 /* Update ctime/mtime for device path for libblkid */
1760 update_dev_time(device_path);
1766 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1768 mutex_unlock(&uuid_mutex);
1771 if (device->writeable) {
1773 list_add(&device->dev_alloc_list,
1774 &root->fs_info->fs_devices->alloc_list);
1775 unlock_chunks(root);
1776 root->fs_info->fs_devices->rw_devices++;
1781 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1782 struct btrfs_device *srcdev)
1784 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1786 list_del_rcu(&srcdev->dev_list);
1787 list_del_rcu(&srcdev->dev_alloc_list);
1788 fs_info->fs_devices->num_devices--;
1789 if (srcdev->missing) {
1790 fs_info->fs_devices->missing_devices--;
1791 fs_info->fs_devices->rw_devices++;
1793 if (srcdev->can_discard)
1794 fs_info->fs_devices->num_can_discard--;
1796 fs_info->fs_devices->open_devices--;
1798 /* zero out the old super */
1799 btrfs_scratch_superblock(srcdev);
1802 call_rcu(&srcdev->rcu, free_device);
1805 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1806 struct btrfs_device *tgtdev)
1808 struct btrfs_device *next_device;
1811 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1813 btrfs_scratch_superblock(tgtdev);
1814 fs_info->fs_devices->open_devices--;
1816 fs_info->fs_devices->num_devices--;
1817 if (tgtdev->can_discard)
1818 fs_info->fs_devices->num_can_discard++;
1820 next_device = list_entry(fs_info->fs_devices->devices.next,
1821 struct btrfs_device, dev_list);
1822 if (tgtdev->bdev == fs_info->sb->s_bdev)
1823 fs_info->sb->s_bdev = next_device->bdev;
1824 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1825 fs_info->fs_devices->latest_bdev = next_device->bdev;
1826 list_del_rcu(&tgtdev->dev_list);
1828 call_rcu(&tgtdev->rcu, free_device);
1830 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1833 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1834 struct btrfs_device **device)
1837 struct btrfs_super_block *disk_super;
1840 struct block_device *bdev;
1841 struct buffer_head *bh;
1844 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1845 root->fs_info->bdev_holder, 0, &bdev, &bh);
1848 disk_super = (struct btrfs_super_block *)bh->b_data;
1849 devid = btrfs_stack_device_id(&disk_super->dev_item);
1850 dev_uuid = disk_super->dev_item.uuid;
1851 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1856 blkdev_put(bdev, FMODE_READ);
1860 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1862 struct btrfs_device **device)
1865 if (strcmp(device_path, "missing") == 0) {
1866 struct list_head *devices;
1867 struct btrfs_device *tmp;
1869 devices = &root->fs_info->fs_devices->devices;
1871 * It is safe to read the devices since the volume_mutex
1872 * is held by the caller.
1874 list_for_each_entry(tmp, devices, dev_list) {
1875 if (tmp->in_fs_metadata && !tmp->bdev) {
1882 btrfs_err(root->fs_info, "no missing device found");
1888 return btrfs_find_device_by_path(root, device_path, device);
1893 * does all the dirty work required for changing file system's UUID.
1895 static int btrfs_prepare_sprout(struct btrfs_root *root)
1897 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1898 struct btrfs_fs_devices *old_devices;
1899 struct btrfs_fs_devices *seed_devices;
1900 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1901 struct btrfs_device *device;
1904 BUG_ON(!mutex_is_locked(&uuid_mutex));
1905 if (!fs_devices->seeding)
1908 seed_devices = __alloc_fs_devices();
1909 if (IS_ERR(seed_devices))
1910 return PTR_ERR(seed_devices);
1912 old_devices = clone_fs_devices(fs_devices);
1913 if (IS_ERR(old_devices)) {
1914 kfree(seed_devices);
1915 return PTR_ERR(old_devices);
1918 list_add(&old_devices->list, &fs_uuids);
1920 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1921 seed_devices->opened = 1;
1922 INIT_LIST_HEAD(&seed_devices->devices);
1923 INIT_LIST_HEAD(&seed_devices->alloc_list);
1924 mutex_init(&seed_devices->device_list_mutex);
1926 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1927 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1930 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1931 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1932 device->fs_devices = seed_devices;
1935 fs_devices->seeding = 0;
1936 fs_devices->num_devices = 0;
1937 fs_devices->open_devices = 0;
1938 fs_devices->seed = seed_devices;
1940 generate_random_uuid(fs_devices->fsid);
1941 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1942 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1943 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1945 super_flags = btrfs_super_flags(disk_super) &
1946 ~BTRFS_SUPER_FLAG_SEEDING;
1947 btrfs_set_super_flags(disk_super, super_flags);
1953 * strore the expected generation for seed devices in device items.
1955 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root)
1958 struct btrfs_path *path;
1959 struct extent_buffer *leaf;
1960 struct btrfs_dev_item *dev_item;
1961 struct btrfs_device *device;
1962 struct btrfs_key key;
1963 u8 fs_uuid[BTRFS_UUID_SIZE];
1964 u8 dev_uuid[BTRFS_UUID_SIZE];
1968 path = btrfs_alloc_path();
1972 root = root->fs_info->chunk_root;
1973 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1975 key.type = BTRFS_DEV_ITEM_KEY;
1978 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1982 leaf = path->nodes[0];
1984 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1985 ret = btrfs_next_leaf(root, path);
1990 leaf = path->nodes[0];
1991 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1992 btrfs_release_path(path);
1996 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1997 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1998 key.type != BTRFS_DEV_ITEM_KEY)
2001 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2002 struct btrfs_dev_item);
2003 devid = btrfs_device_id(leaf, dev_item);
2004 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2006 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2008 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2010 BUG_ON(!device); /* Logic error */
2012 if (device->fs_devices->seeding) {
2013 btrfs_set_device_generation(leaf, dev_item,
2014 device->generation);
2015 btrfs_mark_buffer_dirty(leaf);
2023 btrfs_free_path(path);
2027 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2029 struct request_queue *q;
2030 struct btrfs_trans_handle *trans;
2031 struct btrfs_device *device;
2032 struct block_device *bdev;
2033 struct list_head *devices;
2034 struct super_block *sb = root->fs_info->sb;
2035 struct rcu_string *name;
2037 int seeding_dev = 0;
2040 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2043 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2044 root->fs_info->bdev_holder);
2046 return PTR_ERR(bdev);
2048 if (root->fs_info->fs_devices->seeding) {
2050 down_write(&sb->s_umount);
2051 mutex_lock(&uuid_mutex);
2054 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2056 devices = &root->fs_info->fs_devices->devices;
2058 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2059 list_for_each_entry(device, devices, dev_list) {
2060 if (device->bdev == bdev) {
2063 &root->fs_info->fs_devices->device_list_mutex);
2067 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2069 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2070 if (IS_ERR(device)) {
2071 /* we can safely leave the fs_devices entry around */
2072 ret = PTR_ERR(device);
2076 name = rcu_string_strdup(device_path, GFP_NOFS);
2082 rcu_assign_pointer(device->name, name);
2084 trans = btrfs_start_transaction(root, 0);
2085 if (IS_ERR(trans)) {
2086 rcu_string_free(device->name);
2088 ret = PTR_ERR(trans);
2094 q = bdev_get_queue(bdev);
2095 if (blk_queue_discard(q))
2096 device->can_discard = 1;
2097 device->writeable = 1;
2098 device->generation = trans->transid;
2099 device->io_width = root->sectorsize;
2100 device->io_align = root->sectorsize;
2101 device->sector_size = root->sectorsize;
2102 device->total_bytes = i_size_read(bdev->bd_inode);
2103 device->disk_total_bytes = device->total_bytes;
2104 device->dev_root = root->fs_info->dev_root;
2105 device->bdev = bdev;
2106 device->in_fs_metadata = 1;
2107 device->is_tgtdev_for_dev_replace = 0;
2108 device->mode = FMODE_EXCL;
2109 device->dev_stats_valid = 1;
2110 set_blocksize(device->bdev, 4096);
2113 sb->s_flags &= ~MS_RDONLY;
2114 ret = btrfs_prepare_sprout(root);
2115 BUG_ON(ret); /* -ENOMEM */
2118 device->fs_devices = root->fs_info->fs_devices;
2120 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2121 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2122 list_add(&device->dev_alloc_list,
2123 &root->fs_info->fs_devices->alloc_list);
2124 root->fs_info->fs_devices->num_devices++;
2125 root->fs_info->fs_devices->open_devices++;
2126 root->fs_info->fs_devices->rw_devices++;
2127 root->fs_info->fs_devices->total_devices++;
2128 if (device->can_discard)
2129 root->fs_info->fs_devices->num_can_discard++;
2130 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2132 spin_lock(&root->fs_info->free_chunk_lock);
2133 root->fs_info->free_chunk_space += device->total_bytes;
2134 spin_unlock(&root->fs_info->free_chunk_lock);
2136 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2137 root->fs_info->fs_devices->rotating = 1;
2139 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2140 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2141 total_bytes + device->total_bytes);
2143 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2144 btrfs_set_super_num_devices(root->fs_info->super_copy,
2146 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2149 ret = init_first_rw_device(trans, root, device);
2151 btrfs_abort_transaction(trans, root, ret);
2154 ret = btrfs_finish_sprout(trans, root);
2156 btrfs_abort_transaction(trans, root, ret);
2160 ret = btrfs_add_device(trans, root, device);
2162 btrfs_abort_transaction(trans, root, ret);
2168 * we've got more storage, clear any full flags on the space
2171 btrfs_clear_space_info_full(root->fs_info);
2173 unlock_chunks(root);
2174 root->fs_info->num_tolerated_disk_barrier_failures =
2175 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2176 ret = btrfs_commit_transaction(trans, root);
2179 mutex_unlock(&uuid_mutex);
2180 up_write(&sb->s_umount);
2182 if (ret) /* transaction commit */
2185 ret = btrfs_relocate_sys_chunks(root);
2187 btrfs_error(root->fs_info, ret,
2188 "Failed to relocate sys chunks after "
2189 "device initialization. This can be fixed "
2190 "using the \"btrfs balance\" command.");
2191 trans = btrfs_attach_transaction(root);
2192 if (IS_ERR(trans)) {
2193 if (PTR_ERR(trans) == -ENOENT)
2195 return PTR_ERR(trans);
2197 ret = btrfs_commit_transaction(trans, root);
2200 /* Update ctime/mtime for libblkid */
2201 update_dev_time(device_path);
2205 unlock_chunks(root);
2206 btrfs_end_transaction(trans, root);
2207 rcu_string_free(device->name);
2210 blkdev_put(bdev, FMODE_EXCL);
2212 mutex_unlock(&uuid_mutex);
2213 up_write(&sb->s_umount);
2218 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2219 struct btrfs_device **device_out)
2221 struct request_queue *q;
2222 struct btrfs_device *device;
2223 struct block_device *bdev;
2224 struct btrfs_fs_info *fs_info = root->fs_info;
2225 struct list_head *devices;
2226 struct rcu_string *name;
2227 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2231 if (fs_info->fs_devices->seeding)
2234 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2235 fs_info->bdev_holder);
2237 return PTR_ERR(bdev);
2239 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2241 devices = &fs_info->fs_devices->devices;
2242 list_for_each_entry(device, devices, dev_list) {
2243 if (device->bdev == bdev) {
2249 device = btrfs_alloc_device(NULL, &devid, NULL);
2250 if (IS_ERR(device)) {
2251 ret = PTR_ERR(device);
2255 name = rcu_string_strdup(device_path, GFP_NOFS);
2261 rcu_assign_pointer(device->name, name);
2263 q = bdev_get_queue(bdev);
2264 if (blk_queue_discard(q))
2265 device->can_discard = 1;
2266 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2267 device->writeable = 1;
2268 device->generation = 0;
2269 device->io_width = root->sectorsize;
2270 device->io_align = root->sectorsize;
2271 device->sector_size = root->sectorsize;
2272 device->total_bytes = i_size_read(bdev->bd_inode);
2273 device->disk_total_bytes = device->total_bytes;
2274 device->dev_root = fs_info->dev_root;
2275 device->bdev = bdev;
2276 device->in_fs_metadata = 1;
2277 device->is_tgtdev_for_dev_replace = 1;
2278 device->mode = FMODE_EXCL;
2279 device->dev_stats_valid = 1;
2280 set_blocksize(device->bdev, 4096);
2281 device->fs_devices = fs_info->fs_devices;
2282 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2283 fs_info->fs_devices->num_devices++;
2284 fs_info->fs_devices->open_devices++;
2285 if (device->can_discard)
2286 fs_info->fs_devices->num_can_discard++;
2287 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2289 *device_out = device;
2293 blkdev_put(bdev, FMODE_EXCL);
2297 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2298 struct btrfs_device *tgtdev)
2300 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2301 tgtdev->io_width = fs_info->dev_root->sectorsize;
2302 tgtdev->io_align = fs_info->dev_root->sectorsize;
2303 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2304 tgtdev->dev_root = fs_info->dev_root;
2305 tgtdev->in_fs_metadata = 1;
2308 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2309 struct btrfs_device *device)
2312 struct btrfs_path *path;
2313 struct btrfs_root *root;
2314 struct btrfs_dev_item *dev_item;
2315 struct extent_buffer *leaf;
2316 struct btrfs_key key;
2318 root = device->dev_root->fs_info->chunk_root;
2320 path = btrfs_alloc_path();
2324 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2325 key.type = BTRFS_DEV_ITEM_KEY;
2326 key.offset = device->devid;
2328 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2337 leaf = path->nodes[0];
2338 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2340 btrfs_set_device_id(leaf, dev_item, device->devid);
2341 btrfs_set_device_type(leaf, dev_item, device->type);
2342 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2343 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2344 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2345 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2346 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2347 btrfs_mark_buffer_dirty(leaf);
2350 btrfs_free_path(path);
2354 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2355 struct btrfs_device *device, u64 new_size)
2357 struct btrfs_super_block *super_copy =
2358 device->dev_root->fs_info->super_copy;
2359 u64 old_total = btrfs_super_total_bytes(super_copy);
2360 u64 diff = new_size - device->total_bytes;
2362 if (!device->writeable)
2364 if (new_size <= device->total_bytes ||
2365 device->is_tgtdev_for_dev_replace)
2368 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2369 device->fs_devices->total_rw_bytes += diff;
2371 device->total_bytes = new_size;
2372 device->disk_total_bytes = new_size;
2373 btrfs_clear_space_info_full(device->dev_root->fs_info);
2375 return btrfs_update_device(trans, device);
2378 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2379 struct btrfs_device *device, u64 new_size)
2382 lock_chunks(device->dev_root);
2383 ret = __btrfs_grow_device(trans, device, new_size);
2384 unlock_chunks(device->dev_root);
2388 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2389 struct btrfs_root *root,
2390 u64 chunk_tree, u64 chunk_objectid,
2394 struct btrfs_path *path;
2395 struct btrfs_key key;
2397 root = root->fs_info->chunk_root;
2398 path = btrfs_alloc_path();
2402 key.objectid = chunk_objectid;
2403 key.offset = chunk_offset;
2404 key.type = BTRFS_CHUNK_ITEM_KEY;
2406 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2409 else if (ret > 0) { /* Logic error or corruption */
2410 btrfs_error(root->fs_info, -ENOENT,
2411 "Failed lookup while freeing chunk.");
2416 ret = btrfs_del_item(trans, root, path);
2418 btrfs_error(root->fs_info, ret,
2419 "Failed to delete chunk item.");
2421 btrfs_free_path(path);
2425 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2428 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2429 struct btrfs_disk_key *disk_key;
2430 struct btrfs_chunk *chunk;
2437 struct btrfs_key key;
2439 array_size = btrfs_super_sys_array_size(super_copy);
2441 ptr = super_copy->sys_chunk_array;
2444 while (cur < array_size) {
2445 disk_key = (struct btrfs_disk_key *)ptr;
2446 btrfs_disk_key_to_cpu(&key, disk_key);
2448 len = sizeof(*disk_key);
2450 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2451 chunk = (struct btrfs_chunk *)(ptr + len);
2452 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2453 len += btrfs_chunk_item_size(num_stripes);
2458 if (key.objectid == chunk_objectid &&
2459 key.offset == chunk_offset) {
2460 memmove(ptr, ptr + len, array_size - (cur + len));
2462 btrfs_set_super_sys_array_size(super_copy, array_size);
2471 static int btrfs_relocate_chunk(struct btrfs_root *root,
2472 u64 chunk_tree, u64 chunk_objectid,
2475 struct extent_map_tree *em_tree;
2476 struct btrfs_root *extent_root;
2477 struct btrfs_trans_handle *trans;
2478 struct extent_map *em;
2479 struct map_lookup *map;
2483 root = root->fs_info->chunk_root;
2484 extent_root = root->fs_info->extent_root;
2485 em_tree = &root->fs_info->mapping_tree.map_tree;
2487 ret = btrfs_can_relocate(extent_root, chunk_offset);
2491 /* step one, relocate all the extents inside this chunk */
2492 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2496 trans = btrfs_start_transaction(root, 0);
2497 if (IS_ERR(trans)) {
2498 ret = PTR_ERR(trans);
2499 btrfs_std_error(root->fs_info, ret);
2506 * step two, delete the device extents and the
2507 * chunk tree entries
2509 read_lock(&em_tree->lock);
2510 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2511 read_unlock(&em_tree->lock);
2513 BUG_ON(!em || em->start > chunk_offset ||
2514 em->start + em->len < chunk_offset);
2515 map = (struct map_lookup *)em->bdev;
2517 for (i = 0; i < map->num_stripes; i++) {
2518 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2519 map->stripes[i].physical);
2522 if (map->stripes[i].dev) {
2523 ret = btrfs_update_device(trans, map->stripes[i].dev);
2527 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2532 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2534 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2535 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2539 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2542 write_lock(&em_tree->lock);
2543 remove_extent_mapping(em_tree, em);
2544 write_unlock(&em_tree->lock);
2549 /* once for the tree */
2550 free_extent_map(em);
2552 free_extent_map(em);
2554 unlock_chunks(root);
2555 btrfs_end_transaction(trans, root);
2559 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2561 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2562 struct btrfs_path *path;
2563 struct extent_buffer *leaf;
2564 struct btrfs_chunk *chunk;
2565 struct btrfs_key key;
2566 struct btrfs_key found_key;
2567 u64 chunk_tree = chunk_root->root_key.objectid;
2569 bool retried = false;
2573 path = btrfs_alloc_path();
2578 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2579 key.offset = (u64)-1;
2580 key.type = BTRFS_CHUNK_ITEM_KEY;
2583 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2586 BUG_ON(ret == 0); /* Corruption */
2588 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2595 leaf = path->nodes[0];
2596 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2598 chunk = btrfs_item_ptr(leaf, path->slots[0],
2599 struct btrfs_chunk);
2600 chunk_type = btrfs_chunk_type(leaf, chunk);
2601 btrfs_release_path(path);
2603 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2604 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2613 if (found_key.offset == 0)
2615 key.offset = found_key.offset - 1;
2618 if (failed && !retried) {
2622 } else if (WARN_ON(failed && retried)) {
2626 btrfs_free_path(path);
2630 static int insert_balance_item(struct btrfs_root *root,
2631 struct btrfs_balance_control *bctl)
2633 struct btrfs_trans_handle *trans;
2634 struct btrfs_balance_item *item;
2635 struct btrfs_disk_balance_args disk_bargs;
2636 struct btrfs_path *path;
2637 struct extent_buffer *leaf;
2638 struct btrfs_key key;
2641 path = btrfs_alloc_path();
2645 trans = btrfs_start_transaction(root, 0);
2646 if (IS_ERR(trans)) {
2647 btrfs_free_path(path);
2648 return PTR_ERR(trans);
2651 key.objectid = BTRFS_BALANCE_OBJECTID;
2652 key.type = BTRFS_BALANCE_ITEM_KEY;
2655 ret = btrfs_insert_empty_item(trans, root, path, &key,
2660 leaf = path->nodes[0];
2661 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2663 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2665 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2666 btrfs_set_balance_data(leaf, item, &disk_bargs);
2667 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2668 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2669 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2670 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2672 btrfs_set_balance_flags(leaf, item, bctl->flags);
2674 btrfs_mark_buffer_dirty(leaf);
2676 btrfs_free_path(path);
2677 err = btrfs_commit_transaction(trans, root);
2683 static int del_balance_item(struct btrfs_root *root)
2685 struct btrfs_trans_handle *trans;
2686 struct btrfs_path *path;
2687 struct btrfs_key key;
2690 path = btrfs_alloc_path();
2694 trans = btrfs_start_transaction(root, 0);
2695 if (IS_ERR(trans)) {
2696 btrfs_free_path(path);
2697 return PTR_ERR(trans);
2700 key.objectid = BTRFS_BALANCE_OBJECTID;
2701 key.type = BTRFS_BALANCE_ITEM_KEY;
2704 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2712 ret = btrfs_del_item(trans, root, path);
2714 btrfs_free_path(path);
2715 err = btrfs_commit_transaction(trans, root);
2722 * This is a heuristic used to reduce the number of chunks balanced on
2723 * resume after balance was interrupted.
2725 static void update_balance_args(struct btrfs_balance_control *bctl)
2728 * Turn on soft mode for chunk types that were being converted.
2730 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2731 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2732 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2733 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2734 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2735 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2738 * Turn on usage filter if is not already used. The idea is
2739 * that chunks that we have already balanced should be
2740 * reasonably full. Don't do it for chunks that are being
2741 * converted - that will keep us from relocating unconverted
2742 * (albeit full) chunks.
2744 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2745 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2746 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2747 bctl->data.usage = 90;
2749 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2750 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2751 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2752 bctl->sys.usage = 90;
2754 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2755 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2756 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2757 bctl->meta.usage = 90;
2762 * Should be called with both balance and volume mutexes held to
2763 * serialize other volume operations (add_dev/rm_dev/resize) with
2764 * restriper. Same goes for unset_balance_control.
2766 static void set_balance_control(struct btrfs_balance_control *bctl)
2768 struct btrfs_fs_info *fs_info = bctl->fs_info;
2770 BUG_ON(fs_info->balance_ctl);
2772 spin_lock(&fs_info->balance_lock);
2773 fs_info->balance_ctl = bctl;
2774 spin_unlock(&fs_info->balance_lock);
2777 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2779 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2781 BUG_ON(!fs_info->balance_ctl);
2783 spin_lock(&fs_info->balance_lock);
2784 fs_info->balance_ctl = NULL;
2785 spin_unlock(&fs_info->balance_lock);
2791 * Balance filters. Return 1 if chunk should be filtered out
2792 * (should not be balanced).
2794 static int chunk_profiles_filter(u64 chunk_type,
2795 struct btrfs_balance_args *bargs)
2797 chunk_type = chunk_to_extended(chunk_type) &
2798 BTRFS_EXTENDED_PROFILE_MASK;
2800 if (bargs->profiles & chunk_type)
2806 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2807 struct btrfs_balance_args *bargs)
2809 struct btrfs_block_group_cache *cache;
2810 u64 chunk_used, user_thresh;
2813 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2814 chunk_used = btrfs_block_group_used(&cache->item);
2816 if (bargs->usage == 0)
2818 else if (bargs->usage > 100)
2819 user_thresh = cache->key.offset;
2821 user_thresh = div_factor_fine(cache->key.offset,
2824 if (chunk_used < user_thresh)
2827 btrfs_put_block_group(cache);
2831 static int chunk_devid_filter(struct extent_buffer *leaf,
2832 struct btrfs_chunk *chunk,
2833 struct btrfs_balance_args *bargs)
2835 struct btrfs_stripe *stripe;
2836 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2839 for (i = 0; i < num_stripes; i++) {
2840 stripe = btrfs_stripe_nr(chunk, i);
2841 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2848 /* [pstart, pend) */
2849 static int chunk_drange_filter(struct extent_buffer *leaf,
2850 struct btrfs_chunk *chunk,
2852 struct btrfs_balance_args *bargs)
2854 struct btrfs_stripe *stripe;
2855 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2861 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2864 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2865 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2866 factor = num_stripes / 2;
2867 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2868 factor = num_stripes - 1;
2869 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2870 factor = num_stripes - 2;
2872 factor = num_stripes;
2875 for (i = 0; i < num_stripes; i++) {
2876 stripe = btrfs_stripe_nr(chunk, i);
2877 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2880 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2881 stripe_length = btrfs_chunk_length(leaf, chunk);
2882 do_div(stripe_length, factor);
2884 if (stripe_offset < bargs->pend &&
2885 stripe_offset + stripe_length > bargs->pstart)
2892 /* [vstart, vend) */
2893 static int chunk_vrange_filter(struct extent_buffer *leaf,
2894 struct btrfs_chunk *chunk,
2896 struct btrfs_balance_args *bargs)
2898 if (chunk_offset < bargs->vend &&
2899 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2900 /* at least part of the chunk is inside this vrange */
2906 static int chunk_soft_convert_filter(u64 chunk_type,
2907 struct btrfs_balance_args *bargs)
2909 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2912 chunk_type = chunk_to_extended(chunk_type) &
2913 BTRFS_EXTENDED_PROFILE_MASK;
2915 if (bargs->target == chunk_type)
2921 static int should_balance_chunk(struct btrfs_root *root,
2922 struct extent_buffer *leaf,
2923 struct btrfs_chunk *chunk, u64 chunk_offset)
2925 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2926 struct btrfs_balance_args *bargs = NULL;
2927 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2930 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2931 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2935 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2936 bargs = &bctl->data;
2937 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2939 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2940 bargs = &bctl->meta;
2942 /* profiles filter */
2943 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2944 chunk_profiles_filter(chunk_type, bargs)) {
2949 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2950 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2955 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2956 chunk_devid_filter(leaf, chunk, bargs)) {
2960 /* drange filter, makes sense only with devid filter */
2961 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2962 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2967 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2968 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2972 /* soft profile changing mode */
2973 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2974 chunk_soft_convert_filter(chunk_type, bargs)) {
2979 * limited by count, must be the last filter
2981 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
2982 if (bargs->limit == 0)
2991 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2993 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2994 struct btrfs_root *chunk_root = fs_info->chunk_root;
2995 struct btrfs_root *dev_root = fs_info->dev_root;
2996 struct list_head *devices;
2997 struct btrfs_device *device;
3000 struct btrfs_chunk *chunk;
3001 struct btrfs_path *path;
3002 struct btrfs_key key;
3003 struct btrfs_key found_key;
3004 struct btrfs_trans_handle *trans;
3005 struct extent_buffer *leaf;
3008 int enospc_errors = 0;
3009 bool counting = true;
3010 u64 limit_data = bctl->data.limit;
3011 u64 limit_meta = bctl->meta.limit;
3012 u64 limit_sys = bctl->sys.limit;
3014 /* step one make some room on all the devices */
3015 devices = &fs_info->fs_devices->devices;
3016 list_for_each_entry(device, devices, dev_list) {
3017 old_size = device->total_bytes;
3018 size_to_free = div_factor(old_size, 1);
3019 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3020 if (!device->writeable ||
3021 device->total_bytes - device->bytes_used > size_to_free ||
3022 device->is_tgtdev_for_dev_replace)
3025 ret = btrfs_shrink_device(device, old_size - size_to_free);
3030 trans = btrfs_start_transaction(dev_root, 0);
3031 BUG_ON(IS_ERR(trans));
3033 ret = btrfs_grow_device(trans, device, old_size);
3036 btrfs_end_transaction(trans, dev_root);
3039 /* step two, relocate all the chunks */
3040 path = btrfs_alloc_path();
3046 /* zero out stat counters */
3047 spin_lock(&fs_info->balance_lock);
3048 memset(&bctl->stat, 0, sizeof(bctl->stat));
3049 spin_unlock(&fs_info->balance_lock);
3052 bctl->data.limit = limit_data;
3053 bctl->meta.limit = limit_meta;
3054 bctl->sys.limit = limit_sys;
3056 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3057 key.offset = (u64)-1;
3058 key.type = BTRFS_CHUNK_ITEM_KEY;
3061 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3062 atomic_read(&fs_info->balance_cancel_req)) {
3067 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3072 * this shouldn't happen, it means the last relocate
3076 BUG(); /* FIXME break ? */
3078 ret = btrfs_previous_item(chunk_root, path, 0,
3079 BTRFS_CHUNK_ITEM_KEY);
3085 leaf = path->nodes[0];
3086 slot = path->slots[0];
3087 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3089 if (found_key.objectid != key.objectid)
3092 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3095 spin_lock(&fs_info->balance_lock);
3096 bctl->stat.considered++;
3097 spin_unlock(&fs_info->balance_lock);
3100 ret = should_balance_chunk(chunk_root, leaf, chunk,
3102 btrfs_release_path(path);
3107 spin_lock(&fs_info->balance_lock);
3108 bctl->stat.expected++;
3109 spin_unlock(&fs_info->balance_lock);
3113 ret = btrfs_relocate_chunk(chunk_root,
3114 chunk_root->root_key.objectid,
3117 if (ret && ret != -ENOSPC)
3119 if (ret == -ENOSPC) {
3122 spin_lock(&fs_info->balance_lock);
3123 bctl->stat.completed++;
3124 spin_unlock(&fs_info->balance_lock);
3127 if (found_key.offset == 0)
3129 key.offset = found_key.offset - 1;
3133 btrfs_release_path(path);
3138 btrfs_free_path(path);
3139 if (enospc_errors) {
3140 btrfs_info(fs_info, "%d enospc errors during balance",
3150 * alloc_profile_is_valid - see if a given profile is valid and reduced
3151 * @flags: profile to validate
3152 * @extended: if true @flags is treated as an extended profile
3154 static int alloc_profile_is_valid(u64 flags, int extended)
3156 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3157 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3159 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3161 /* 1) check that all other bits are zeroed */
3165 /* 2) see if profile is reduced */
3167 return !extended; /* "0" is valid for usual profiles */
3169 /* true if exactly one bit set */
3170 return (flags & (flags - 1)) == 0;
3173 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3175 /* cancel requested || normal exit path */
3176 return atomic_read(&fs_info->balance_cancel_req) ||
3177 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3178 atomic_read(&fs_info->balance_cancel_req) == 0);
3181 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3185 unset_balance_control(fs_info);
3186 ret = del_balance_item(fs_info->tree_root);
3188 btrfs_std_error(fs_info, ret);
3190 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3194 * Should be called with both balance and volume mutexes held
3196 int btrfs_balance(struct btrfs_balance_control *bctl,
3197 struct btrfs_ioctl_balance_args *bargs)
3199 struct btrfs_fs_info *fs_info = bctl->fs_info;
3206 if (btrfs_fs_closing(fs_info) ||
3207 atomic_read(&fs_info->balance_pause_req) ||
3208 atomic_read(&fs_info->balance_cancel_req)) {
3213 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3214 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3218 * In case of mixed groups both data and meta should be picked,
3219 * and identical options should be given for both of them.
3221 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3222 if (mixed && (bctl->flags & allowed)) {
3223 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3224 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3225 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3226 btrfs_err(fs_info, "with mixed groups data and "
3227 "metadata balance options must be the same");
3233 num_devices = fs_info->fs_devices->num_devices;
3234 btrfs_dev_replace_lock(&fs_info->dev_replace);
3235 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3236 BUG_ON(num_devices < 1);
3239 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3240 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3241 if (num_devices == 1)
3242 allowed |= BTRFS_BLOCK_GROUP_DUP;
3243 else if (num_devices > 1)
3244 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3245 if (num_devices > 2)
3246 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3247 if (num_devices > 3)
3248 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3249 BTRFS_BLOCK_GROUP_RAID6);
3250 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3251 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3252 (bctl->data.target & ~allowed))) {
3253 btrfs_err(fs_info, "unable to start balance with target "
3254 "data profile %llu",
3259 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3260 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3261 (bctl->meta.target & ~allowed))) {
3263 "unable to start balance with target metadata profile %llu",
3268 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3269 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3270 (bctl->sys.target & ~allowed))) {
3272 "unable to start balance with target system profile %llu",
3278 /* allow dup'ed data chunks only in mixed mode */
3279 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3280 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3281 btrfs_err(fs_info, "dup for data is not allowed");
3286 /* allow to reduce meta or sys integrity only if force set */
3287 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3288 BTRFS_BLOCK_GROUP_RAID10 |
3289 BTRFS_BLOCK_GROUP_RAID5 |
3290 BTRFS_BLOCK_GROUP_RAID6;
3292 seq = read_seqbegin(&fs_info->profiles_lock);
3294 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3295 (fs_info->avail_system_alloc_bits & allowed) &&
3296 !(bctl->sys.target & allowed)) ||
3297 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3298 (fs_info->avail_metadata_alloc_bits & allowed) &&
3299 !(bctl->meta.target & allowed))) {
3300 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3301 btrfs_info(fs_info, "force reducing metadata integrity");
3303 btrfs_err(fs_info, "balance will reduce metadata "
3304 "integrity, use force if you want this");
3309 } while (read_seqretry(&fs_info->profiles_lock, seq));
3311 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3312 int num_tolerated_disk_barrier_failures;
3313 u64 target = bctl->sys.target;
3315 num_tolerated_disk_barrier_failures =
3316 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3317 if (num_tolerated_disk_barrier_failures > 0 &&
3319 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3320 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3321 num_tolerated_disk_barrier_failures = 0;
3322 else if (num_tolerated_disk_barrier_failures > 1 &&
3324 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3325 num_tolerated_disk_barrier_failures = 1;
3327 fs_info->num_tolerated_disk_barrier_failures =
3328 num_tolerated_disk_barrier_failures;
3331 ret = insert_balance_item(fs_info->tree_root, bctl);
3332 if (ret && ret != -EEXIST)
3335 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3336 BUG_ON(ret == -EEXIST);
3337 set_balance_control(bctl);
3339 BUG_ON(ret != -EEXIST);
3340 spin_lock(&fs_info->balance_lock);
3341 update_balance_args(bctl);
3342 spin_unlock(&fs_info->balance_lock);
3345 atomic_inc(&fs_info->balance_running);
3346 mutex_unlock(&fs_info->balance_mutex);
3348 ret = __btrfs_balance(fs_info);
3350 mutex_lock(&fs_info->balance_mutex);
3351 atomic_dec(&fs_info->balance_running);
3353 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3354 fs_info->num_tolerated_disk_barrier_failures =
3355 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3359 memset(bargs, 0, sizeof(*bargs));
3360 update_ioctl_balance_args(fs_info, 0, bargs);
3363 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3364 balance_need_close(fs_info)) {
3365 __cancel_balance(fs_info);
3368 wake_up(&fs_info->balance_wait_q);
3372 if (bctl->flags & BTRFS_BALANCE_RESUME)
3373 __cancel_balance(fs_info);
3376 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3381 static int balance_kthread(void *data)
3383 struct btrfs_fs_info *fs_info = data;
3386 mutex_lock(&fs_info->volume_mutex);
3387 mutex_lock(&fs_info->balance_mutex);
3389 if (fs_info->balance_ctl) {
3390 btrfs_info(fs_info, "continuing balance");
3391 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3394 mutex_unlock(&fs_info->balance_mutex);
3395 mutex_unlock(&fs_info->volume_mutex);
3400 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3402 struct task_struct *tsk;
3404 spin_lock(&fs_info->balance_lock);
3405 if (!fs_info->balance_ctl) {
3406 spin_unlock(&fs_info->balance_lock);
3409 spin_unlock(&fs_info->balance_lock);
3411 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3412 btrfs_info(fs_info, "force skipping balance");
3416 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3417 return PTR_ERR_OR_ZERO(tsk);
3420 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3422 struct btrfs_balance_control *bctl;
3423 struct btrfs_balance_item *item;
3424 struct btrfs_disk_balance_args disk_bargs;
3425 struct btrfs_path *path;
3426 struct extent_buffer *leaf;
3427 struct btrfs_key key;
3430 path = btrfs_alloc_path();
3434 key.objectid = BTRFS_BALANCE_OBJECTID;
3435 key.type = BTRFS_BALANCE_ITEM_KEY;
3438 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3441 if (ret > 0) { /* ret = -ENOENT; */
3446 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3452 leaf = path->nodes[0];
3453 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3455 bctl->fs_info = fs_info;
3456 bctl->flags = btrfs_balance_flags(leaf, item);
3457 bctl->flags |= BTRFS_BALANCE_RESUME;
3459 btrfs_balance_data(leaf, item, &disk_bargs);
3460 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3461 btrfs_balance_meta(leaf, item, &disk_bargs);
3462 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3463 btrfs_balance_sys(leaf, item, &disk_bargs);
3464 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3466 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3468 mutex_lock(&fs_info->volume_mutex);
3469 mutex_lock(&fs_info->balance_mutex);
3471 set_balance_control(bctl);
3473 mutex_unlock(&fs_info->balance_mutex);
3474 mutex_unlock(&fs_info->volume_mutex);
3476 btrfs_free_path(path);
3480 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3484 mutex_lock(&fs_info->balance_mutex);
3485 if (!fs_info->balance_ctl) {
3486 mutex_unlock(&fs_info->balance_mutex);
3490 if (atomic_read(&fs_info->balance_running)) {
3491 atomic_inc(&fs_info->balance_pause_req);
3492 mutex_unlock(&fs_info->balance_mutex);
3494 wait_event(fs_info->balance_wait_q,
3495 atomic_read(&fs_info->balance_running) == 0);
3497 mutex_lock(&fs_info->balance_mutex);
3498 /* we are good with balance_ctl ripped off from under us */
3499 BUG_ON(atomic_read(&fs_info->balance_running));
3500 atomic_dec(&fs_info->balance_pause_req);
3505 mutex_unlock(&fs_info->balance_mutex);
3509 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3511 if (fs_info->sb->s_flags & MS_RDONLY)
3514 mutex_lock(&fs_info->balance_mutex);
3515 if (!fs_info->balance_ctl) {
3516 mutex_unlock(&fs_info->balance_mutex);
3520 atomic_inc(&fs_info->balance_cancel_req);
3522 * if we are running just wait and return, balance item is
3523 * deleted in btrfs_balance in this case
3525 if (atomic_read(&fs_info->balance_running)) {
3526 mutex_unlock(&fs_info->balance_mutex);
3527 wait_event(fs_info->balance_wait_q,
3528 atomic_read(&fs_info->balance_running) == 0);
3529 mutex_lock(&fs_info->balance_mutex);
3531 /* __cancel_balance needs volume_mutex */
3532 mutex_unlock(&fs_info->balance_mutex);
3533 mutex_lock(&fs_info->volume_mutex);
3534 mutex_lock(&fs_info->balance_mutex);
3536 if (fs_info->balance_ctl)
3537 __cancel_balance(fs_info);
3539 mutex_unlock(&fs_info->volume_mutex);
3542 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3543 atomic_dec(&fs_info->balance_cancel_req);
3544 mutex_unlock(&fs_info->balance_mutex);
3548 static int btrfs_uuid_scan_kthread(void *data)
3550 struct btrfs_fs_info *fs_info = data;
3551 struct btrfs_root *root = fs_info->tree_root;
3552 struct btrfs_key key;
3553 struct btrfs_key max_key;
3554 struct btrfs_path *path = NULL;
3556 struct extent_buffer *eb;
3558 struct btrfs_root_item root_item;
3560 struct btrfs_trans_handle *trans = NULL;
3562 path = btrfs_alloc_path();
3569 key.type = BTRFS_ROOT_ITEM_KEY;
3572 max_key.objectid = (u64)-1;
3573 max_key.type = BTRFS_ROOT_ITEM_KEY;
3574 max_key.offset = (u64)-1;
3576 path->keep_locks = 1;
3579 ret = btrfs_search_forward(root, &key, path, 0);
3586 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3587 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3588 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3589 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3592 eb = path->nodes[0];
3593 slot = path->slots[0];
3594 item_size = btrfs_item_size_nr(eb, slot);
3595 if (item_size < sizeof(root_item))
3598 read_extent_buffer(eb, &root_item,
3599 btrfs_item_ptr_offset(eb, slot),
3600 (int)sizeof(root_item));
3601 if (btrfs_root_refs(&root_item) == 0)
3604 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3605 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3609 btrfs_release_path(path);
3611 * 1 - subvol uuid item
3612 * 1 - received_subvol uuid item
3614 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3615 if (IS_ERR(trans)) {
3616 ret = PTR_ERR(trans);
3624 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3625 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3627 BTRFS_UUID_KEY_SUBVOL,
3630 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3636 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3637 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3638 root_item.received_uuid,
3639 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3642 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3650 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3656 btrfs_release_path(path);
3657 if (key.offset < (u64)-1) {
3659 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3661 key.type = BTRFS_ROOT_ITEM_KEY;
3662 } else if (key.objectid < (u64)-1) {
3664 key.type = BTRFS_ROOT_ITEM_KEY;
3673 btrfs_free_path(path);
3674 if (trans && !IS_ERR(trans))
3675 btrfs_end_transaction(trans, fs_info->uuid_root);
3677 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3679 fs_info->update_uuid_tree_gen = 1;
3680 up(&fs_info->uuid_tree_rescan_sem);
3685 * Callback for btrfs_uuid_tree_iterate().
3687 * 0 check succeeded, the entry is not outdated.
3688 * < 0 if an error occured.
3689 * > 0 if the check failed, which means the caller shall remove the entry.
3691 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3692 u8 *uuid, u8 type, u64 subid)
3694 struct btrfs_key key;
3696 struct btrfs_root *subvol_root;
3698 if (type != BTRFS_UUID_KEY_SUBVOL &&
3699 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3702 key.objectid = subid;
3703 key.type = BTRFS_ROOT_ITEM_KEY;
3704 key.offset = (u64)-1;
3705 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3706 if (IS_ERR(subvol_root)) {
3707 ret = PTR_ERR(subvol_root);
3714 case BTRFS_UUID_KEY_SUBVOL:
3715 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3718 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3719 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3729 static int btrfs_uuid_rescan_kthread(void *data)
3731 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3735 * 1st step is to iterate through the existing UUID tree and
3736 * to delete all entries that contain outdated data.
3737 * 2nd step is to add all missing entries to the UUID tree.
3739 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3741 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3742 up(&fs_info->uuid_tree_rescan_sem);
3745 return btrfs_uuid_scan_kthread(data);
3748 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3750 struct btrfs_trans_handle *trans;
3751 struct btrfs_root *tree_root = fs_info->tree_root;
3752 struct btrfs_root *uuid_root;
3753 struct task_struct *task;
3760 trans = btrfs_start_transaction(tree_root, 2);
3762 return PTR_ERR(trans);
3764 uuid_root = btrfs_create_tree(trans, fs_info,
3765 BTRFS_UUID_TREE_OBJECTID);
3766 if (IS_ERR(uuid_root)) {
3767 btrfs_abort_transaction(trans, tree_root,
3768 PTR_ERR(uuid_root));
3769 return PTR_ERR(uuid_root);
3772 fs_info->uuid_root = uuid_root;
3774 ret = btrfs_commit_transaction(trans, tree_root);
3778 down(&fs_info->uuid_tree_rescan_sem);
3779 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3781 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3782 btrfs_warn(fs_info, "failed to start uuid_scan task");
3783 up(&fs_info->uuid_tree_rescan_sem);
3784 return PTR_ERR(task);
3790 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3792 struct task_struct *task;
3794 down(&fs_info->uuid_tree_rescan_sem);
3795 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3797 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3798 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3799 up(&fs_info->uuid_tree_rescan_sem);
3800 return PTR_ERR(task);
3807 * shrinking a device means finding all of the device extents past
3808 * the new size, and then following the back refs to the chunks.
3809 * The chunk relocation code actually frees the device extent
3811 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3813 struct btrfs_trans_handle *trans;
3814 struct btrfs_root *root = device->dev_root;
3815 struct btrfs_dev_extent *dev_extent = NULL;
3816 struct btrfs_path *path;
3824 bool retried = false;
3825 struct extent_buffer *l;
3826 struct btrfs_key key;
3827 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3828 u64 old_total = btrfs_super_total_bytes(super_copy);
3829 u64 old_size = device->total_bytes;
3830 u64 diff = device->total_bytes - new_size;
3832 if (device->is_tgtdev_for_dev_replace)
3835 path = btrfs_alloc_path();
3843 device->total_bytes = new_size;
3844 if (device->writeable) {
3845 device->fs_devices->total_rw_bytes -= diff;
3846 spin_lock(&root->fs_info->free_chunk_lock);
3847 root->fs_info->free_chunk_space -= diff;
3848 spin_unlock(&root->fs_info->free_chunk_lock);
3850 unlock_chunks(root);
3853 key.objectid = device->devid;
3854 key.offset = (u64)-1;
3855 key.type = BTRFS_DEV_EXTENT_KEY;
3858 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3862 ret = btrfs_previous_item(root, path, 0, key.type);
3867 btrfs_release_path(path);
3872 slot = path->slots[0];
3873 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3875 if (key.objectid != device->devid) {
3876 btrfs_release_path(path);
3880 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3881 length = btrfs_dev_extent_length(l, dev_extent);
3883 if (key.offset + length <= new_size) {
3884 btrfs_release_path(path);
3888 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3889 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3890 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3891 btrfs_release_path(path);
3893 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3895 if (ret && ret != -ENOSPC)
3899 } while (key.offset-- > 0);
3901 if (failed && !retried) {
3905 } else if (failed && retried) {
3909 device->total_bytes = old_size;
3910 if (device->writeable)
3911 device->fs_devices->total_rw_bytes += diff;
3912 spin_lock(&root->fs_info->free_chunk_lock);
3913 root->fs_info->free_chunk_space += diff;
3914 spin_unlock(&root->fs_info->free_chunk_lock);
3915 unlock_chunks(root);
3919 /* Shrinking succeeded, else we would be at "done". */
3920 trans = btrfs_start_transaction(root, 0);
3921 if (IS_ERR(trans)) {
3922 ret = PTR_ERR(trans);
3928 device->disk_total_bytes = new_size;
3929 /* Now btrfs_update_device() will change the on-disk size. */
3930 ret = btrfs_update_device(trans, device);
3932 unlock_chunks(root);
3933 btrfs_end_transaction(trans, root);
3936 WARN_ON(diff > old_total);
3937 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3938 unlock_chunks(root);
3939 btrfs_end_transaction(trans, root);
3941 btrfs_free_path(path);
3945 static int btrfs_add_system_chunk(struct btrfs_root *root,
3946 struct btrfs_key *key,
3947 struct btrfs_chunk *chunk, int item_size)
3949 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3950 struct btrfs_disk_key disk_key;
3954 array_size = btrfs_super_sys_array_size(super_copy);
3955 if (array_size + item_size + sizeof(disk_key)
3956 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3959 ptr = super_copy->sys_chunk_array + array_size;
3960 btrfs_cpu_key_to_disk(&disk_key, key);
3961 memcpy(ptr, &disk_key, sizeof(disk_key));
3962 ptr += sizeof(disk_key);
3963 memcpy(ptr, chunk, item_size);
3964 item_size += sizeof(disk_key);
3965 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3970 * sort the devices in descending order by max_avail, total_avail
3972 static int btrfs_cmp_device_info(const void *a, const void *b)
3974 const struct btrfs_device_info *di_a = a;
3975 const struct btrfs_device_info *di_b = b;
3977 if (di_a->max_avail > di_b->max_avail)
3979 if (di_a->max_avail < di_b->max_avail)
3981 if (di_a->total_avail > di_b->total_avail)
3983 if (di_a->total_avail < di_b->total_avail)
3988 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3989 [BTRFS_RAID_RAID10] = {
3992 .devs_max = 0, /* 0 == as many as possible */
3994 .devs_increment = 2,
3997 [BTRFS_RAID_RAID1] = {
4002 .devs_increment = 2,
4005 [BTRFS_RAID_DUP] = {
4010 .devs_increment = 1,
4013 [BTRFS_RAID_RAID0] = {
4018 .devs_increment = 1,
4021 [BTRFS_RAID_SINGLE] = {
4026 .devs_increment = 1,
4029 [BTRFS_RAID_RAID5] = {
4034 .devs_increment = 1,
4037 [BTRFS_RAID_RAID6] = {
4042 .devs_increment = 1,
4047 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4049 /* TODO allow them to set a preferred stripe size */
4053 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4055 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4058 btrfs_set_fs_incompat(info, RAID56);
4061 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4062 - sizeof(struct btrfs_item) \
4063 - sizeof(struct btrfs_chunk)) \
4064 / sizeof(struct btrfs_stripe) + 1)
4066 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4067 - 2 * sizeof(struct btrfs_disk_key) \
4068 - 2 * sizeof(struct btrfs_chunk)) \
4069 / sizeof(struct btrfs_stripe) + 1)
4071 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4072 struct btrfs_root *extent_root, u64 start,
4075 struct btrfs_fs_info *info = extent_root->fs_info;
4076 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4077 struct list_head *cur;
4078 struct map_lookup *map = NULL;
4079 struct extent_map_tree *em_tree;
4080 struct extent_map *em;
4081 struct btrfs_device_info *devices_info = NULL;
4083 int num_stripes; /* total number of stripes to allocate */
4084 int data_stripes; /* number of stripes that count for
4086 int sub_stripes; /* sub_stripes info for map */
4087 int dev_stripes; /* stripes per dev */
4088 int devs_max; /* max devs to use */
4089 int devs_min; /* min devs needed */
4090 int devs_increment; /* ndevs has to be a multiple of this */
4091 int ncopies; /* how many copies to data has */
4093 u64 max_stripe_size;
4097 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4103 BUG_ON(!alloc_profile_is_valid(type, 0));
4105 if (list_empty(&fs_devices->alloc_list))
4108 index = __get_raid_index(type);
4110 sub_stripes = btrfs_raid_array[index].sub_stripes;
4111 dev_stripes = btrfs_raid_array[index].dev_stripes;
4112 devs_max = btrfs_raid_array[index].devs_max;
4113 devs_min = btrfs_raid_array[index].devs_min;
4114 devs_increment = btrfs_raid_array[index].devs_increment;
4115 ncopies = btrfs_raid_array[index].ncopies;
4117 if (type & BTRFS_BLOCK_GROUP_DATA) {
4118 max_stripe_size = 1024 * 1024 * 1024;
4119 max_chunk_size = 10 * max_stripe_size;
4121 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4122 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4123 /* for larger filesystems, use larger metadata chunks */
4124 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4125 max_stripe_size = 1024 * 1024 * 1024;
4127 max_stripe_size = 256 * 1024 * 1024;
4128 max_chunk_size = max_stripe_size;
4130 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4131 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4132 max_stripe_size = 32 * 1024 * 1024;
4133 max_chunk_size = 2 * max_stripe_size;
4135 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4137 btrfs_err(info, "invalid chunk type 0x%llx requested",
4142 /* we don't want a chunk larger than 10% of writeable space */
4143 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4146 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4151 cur = fs_devices->alloc_list.next;
4154 * in the first pass through the devices list, we gather information
4155 * about the available holes on each device.
4158 while (cur != &fs_devices->alloc_list) {
4159 struct btrfs_device *device;
4163 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4167 if (!device->writeable) {
4169 "BTRFS: read-only device in alloc_list\n");
4173 if (!device->in_fs_metadata ||
4174 device->is_tgtdev_for_dev_replace)
4177 if (device->total_bytes > device->bytes_used)
4178 total_avail = device->total_bytes - device->bytes_used;
4182 /* If there is no space on this device, skip it. */
4183 if (total_avail == 0)
4186 ret = find_free_dev_extent(trans, device,
4187 max_stripe_size * dev_stripes,
4188 &dev_offset, &max_avail);
4189 if (ret && ret != -ENOSPC)
4193 max_avail = max_stripe_size * dev_stripes;
4195 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4198 if (ndevs == fs_devices->rw_devices) {
4199 WARN(1, "%s: found more than %llu devices\n",
4200 __func__, fs_devices->rw_devices);
4203 devices_info[ndevs].dev_offset = dev_offset;
4204 devices_info[ndevs].max_avail = max_avail;
4205 devices_info[ndevs].total_avail = total_avail;
4206 devices_info[ndevs].dev = device;
4211 * now sort the devices by hole size / available space
4213 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4214 btrfs_cmp_device_info, NULL);
4216 /* round down to number of usable stripes */
4217 ndevs -= ndevs % devs_increment;
4219 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4224 if (devs_max && ndevs > devs_max)
4227 * the primary goal is to maximize the number of stripes, so use as many
4228 * devices as possible, even if the stripes are not maximum sized.
4230 stripe_size = devices_info[ndevs-1].max_avail;
4231 num_stripes = ndevs * dev_stripes;
4234 * this will have to be fixed for RAID1 and RAID10 over
4237 data_stripes = num_stripes / ncopies;
4239 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4240 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4241 btrfs_super_stripesize(info->super_copy));
4242 data_stripes = num_stripes - 1;
4244 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4245 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4246 btrfs_super_stripesize(info->super_copy));
4247 data_stripes = num_stripes - 2;
4251 * Use the number of data stripes to figure out how big this chunk
4252 * is really going to be in terms of logical address space,
4253 * and compare that answer with the max chunk size
4255 if (stripe_size * data_stripes > max_chunk_size) {
4256 u64 mask = (1ULL << 24) - 1;
4257 stripe_size = max_chunk_size;
4258 do_div(stripe_size, data_stripes);
4260 /* bump the answer up to a 16MB boundary */
4261 stripe_size = (stripe_size + mask) & ~mask;
4263 /* but don't go higher than the limits we found
4264 * while searching for free extents
4266 if (stripe_size > devices_info[ndevs-1].max_avail)
4267 stripe_size = devices_info[ndevs-1].max_avail;
4270 do_div(stripe_size, dev_stripes);
4272 /* align to BTRFS_STRIPE_LEN */
4273 do_div(stripe_size, raid_stripe_len);
4274 stripe_size *= raid_stripe_len;
4276 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4281 map->num_stripes = num_stripes;
4283 for (i = 0; i < ndevs; ++i) {
4284 for (j = 0; j < dev_stripes; ++j) {
4285 int s = i * dev_stripes + j;
4286 map->stripes[s].dev = devices_info[i].dev;
4287 map->stripes[s].physical = devices_info[i].dev_offset +
4291 map->sector_size = extent_root->sectorsize;
4292 map->stripe_len = raid_stripe_len;
4293 map->io_align = raid_stripe_len;
4294 map->io_width = raid_stripe_len;
4296 map->sub_stripes = sub_stripes;
4298 num_bytes = stripe_size * data_stripes;
4300 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4302 em = alloc_extent_map();
4307 em->bdev = (struct block_device *)map;
4309 em->len = num_bytes;
4310 em->block_start = 0;
4311 em->block_len = em->len;
4312 em->orig_block_len = stripe_size;
4314 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4315 write_lock(&em_tree->lock);
4316 ret = add_extent_mapping(em_tree, em, 0);
4318 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4319 atomic_inc(&em->refs);
4321 write_unlock(&em_tree->lock);
4323 free_extent_map(em);
4327 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4328 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4331 goto error_del_extent;
4333 free_extent_map(em);
4334 check_raid56_incompat_flag(extent_root->fs_info, type);
4336 kfree(devices_info);
4340 write_lock(&em_tree->lock);
4341 remove_extent_mapping(em_tree, em);
4342 write_unlock(&em_tree->lock);
4344 /* One for our allocation */
4345 free_extent_map(em);
4346 /* One for the tree reference */
4347 free_extent_map(em);
4350 kfree(devices_info);
4354 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4355 struct btrfs_root *extent_root,
4356 u64 chunk_offset, u64 chunk_size)
4358 struct btrfs_key key;
4359 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4360 struct btrfs_device *device;
4361 struct btrfs_chunk *chunk;
4362 struct btrfs_stripe *stripe;
4363 struct extent_map_tree *em_tree;
4364 struct extent_map *em;
4365 struct map_lookup *map;
4372 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4373 read_lock(&em_tree->lock);
4374 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4375 read_unlock(&em_tree->lock);
4378 btrfs_crit(extent_root->fs_info, "unable to find logical "
4379 "%Lu len %Lu", chunk_offset, chunk_size);
4383 if (em->start != chunk_offset || em->len != chunk_size) {
4384 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4385 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4386 chunk_size, em->start, em->len);
4387 free_extent_map(em);
4391 map = (struct map_lookup *)em->bdev;
4392 item_size = btrfs_chunk_item_size(map->num_stripes);
4393 stripe_size = em->orig_block_len;
4395 chunk = kzalloc(item_size, GFP_NOFS);
4401 for (i = 0; i < map->num_stripes; i++) {
4402 device = map->stripes[i].dev;
4403 dev_offset = map->stripes[i].physical;
4405 device->bytes_used += stripe_size;
4406 ret = btrfs_update_device(trans, device);
4409 ret = btrfs_alloc_dev_extent(trans, device,
4410 chunk_root->root_key.objectid,
4411 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4412 chunk_offset, dev_offset,
4418 spin_lock(&extent_root->fs_info->free_chunk_lock);
4419 extent_root->fs_info->free_chunk_space -= (stripe_size *
4421 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4423 stripe = &chunk->stripe;
4424 for (i = 0; i < map->num_stripes; i++) {
4425 device = map->stripes[i].dev;
4426 dev_offset = map->stripes[i].physical;
4428 btrfs_set_stack_stripe_devid(stripe, device->devid);
4429 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4430 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4434 btrfs_set_stack_chunk_length(chunk, chunk_size);
4435 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4436 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4437 btrfs_set_stack_chunk_type(chunk, map->type);
4438 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4439 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4440 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4441 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4442 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4444 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4445 key.type = BTRFS_CHUNK_ITEM_KEY;
4446 key.offset = chunk_offset;
4448 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4449 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4451 * TODO: Cleanup of inserted chunk root in case of
4454 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4460 free_extent_map(em);
4465 * Chunk allocation falls into two parts. The first part does works
4466 * that make the new allocated chunk useable, but not do any operation
4467 * that modifies the chunk tree. The second part does the works that
4468 * require modifying the chunk tree. This division is important for the
4469 * bootstrap process of adding storage to a seed btrfs.
4471 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4472 struct btrfs_root *extent_root, u64 type)
4476 chunk_offset = find_next_chunk(extent_root->fs_info);
4477 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4480 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4481 struct btrfs_root *root,
4482 struct btrfs_device *device)
4485 u64 sys_chunk_offset;
4487 struct btrfs_fs_info *fs_info = root->fs_info;
4488 struct btrfs_root *extent_root = fs_info->extent_root;
4491 chunk_offset = find_next_chunk(fs_info);
4492 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4493 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4498 sys_chunk_offset = find_next_chunk(root->fs_info);
4499 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4500 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4503 btrfs_abort_transaction(trans, root, ret);
4507 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4509 btrfs_abort_transaction(trans, root, ret);
4514 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4516 struct extent_map *em;
4517 struct map_lookup *map;
4518 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4522 read_lock(&map_tree->map_tree.lock);
4523 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4524 read_unlock(&map_tree->map_tree.lock);
4528 if (btrfs_test_opt(root, DEGRADED)) {
4529 free_extent_map(em);
4533 map = (struct map_lookup *)em->bdev;
4534 for (i = 0; i < map->num_stripes; i++) {
4535 if (!map->stripes[i].dev->writeable) {
4540 free_extent_map(em);
4544 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4546 extent_map_tree_init(&tree->map_tree);
4549 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4551 struct extent_map *em;
4554 write_lock(&tree->map_tree.lock);
4555 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4557 remove_extent_mapping(&tree->map_tree, em);
4558 write_unlock(&tree->map_tree.lock);
4563 free_extent_map(em);
4564 /* once for the tree */
4565 free_extent_map(em);
4569 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4571 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4572 struct extent_map *em;
4573 struct map_lookup *map;
4574 struct extent_map_tree *em_tree = &map_tree->map_tree;
4577 read_lock(&em_tree->lock);
4578 em = lookup_extent_mapping(em_tree, logical, len);
4579 read_unlock(&em_tree->lock);
4582 * We could return errors for these cases, but that could get ugly and
4583 * we'd probably do the same thing which is just not do anything else
4584 * and exit, so return 1 so the callers don't try to use other copies.
4587 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4592 if (em->start > logical || em->start + em->len < logical) {
4593 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4594 "%Lu-%Lu", logical, logical+len, em->start,
4595 em->start + em->len);
4596 free_extent_map(em);
4600 map = (struct map_lookup *)em->bdev;
4601 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4602 ret = map->num_stripes;
4603 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4604 ret = map->sub_stripes;
4605 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4607 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4611 free_extent_map(em);
4613 btrfs_dev_replace_lock(&fs_info->dev_replace);
4614 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4616 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4621 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4622 struct btrfs_mapping_tree *map_tree,
4625 struct extent_map *em;
4626 struct map_lookup *map;
4627 struct extent_map_tree *em_tree = &map_tree->map_tree;
4628 unsigned long len = root->sectorsize;
4630 read_lock(&em_tree->lock);
4631 em = lookup_extent_mapping(em_tree, logical, len);
4632 read_unlock(&em_tree->lock);
4635 BUG_ON(em->start > logical || em->start + em->len < logical);
4636 map = (struct map_lookup *)em->bdev;
4637 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4638 BTRFS_BLOCK_GROUP_RAID6)) {
4639 len = map->stripe_len * nr_data_stripes(map);
4641 free_extent_map(em);
4645 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4646 u64 logical, u64 len, int mirror_num)
4648 struct extent_map *em;
4649 struct map_lookup *map;
4650 struct extent_map_tree *em_tree = &map_tree->map_tree;
4653 read_lock(&em_tree->lock);
4654 em = lookup_extent_mapping(em_tree, logical, len);
4655 read_unlock(&em_tree->lock);
4658 BUG_ON(em->start > logical || em->start + em->len < logical);
4659 map = (struct map_lookup *)em->bdev;
4660 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4661 BTRFS_BLOCK_GROUP_RAID6))
4663 free_extent_map(em);
4667 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4668 struct map_lookup *map, int first, int num,
4669 int optimal, int dev_replace_is_ongoing)
4673 struct btrfs_device *srcdev;
4675 if (dev_replace_is_ongoing &&
4676 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4677 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4678 srcdev = fs_info->dev_replace.srcdev;
4683 * try to avoid the drive that is the source drive for a
4684 * dev-replace procedure, only choose it if no other non-missing
4685 * mirror is available
4687 for (tolerance = 0; tolerance < 2; tolerance++) {
4688 if (map->stripes[optimal].dev->bdev &&
4689 (tolerance || map->stripes[optimal].dev != srcdev))
4691 for (i = first; i < first + num; i++) {
4692 if (map->stripes[i].dev->bdev &&
4693 (tolerance || map->stripes[i].dev != srcdev))
4698 /* we couldn't find one that doesn't fail. Just return something
4699 * and the io error handling code will clean up eventually
4704 static inline int parity_smaller(u64 a, u64 b)
4709 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4710 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4712 struct btrfs_bio_stripe s;
4719 for (i = 0; i < bbio->num_stripes - 1; i++) {
4720 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4721 s = bbio->stripes[i];
4723 bbio->stripes[i] = bbio->stripes[i+1];
4724 raid_map[i] = raid_map[i+1];
4725 bbio->stripes[i+1] = s;
4733 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4734 u64 logical, u64 *length,
4735 struct btrfs_bio **bbio_ret,
4736 int mirror_num, u64 **raid_map_ret)
4738 struct extent_map *em;
4739 struct map_lookup *map;
4740 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4741 struct extent_map_tree *em_tree = &map_tree->map_tree;
4744 u64 stripe_end_offset;
4749 u64 *raid_map = NULL;
4755 struct btrfs_bio *bbio = NULL;
4756 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4757 int dev_replace_is_ongoing = 0;
4758 int num_alloc_stripes;
4759 int patch_the_first_stripe_for_dev_replace = 0;
4760 u64 physical_to_patch_in_first_stripe = 0;
4761 u64 raid56_full_stripe_start = (u64)-1;
4763 read_lock(&em_tree->lock);
4764 em = lookup_extent_mapping(em_tree, logical, *length);
4765 read_unlock(&em_tree->lock);
4768 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4773 if (em->start > logical || em->start + em->len < logical) {
4774 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4775 "found %Lu-%Lu", logical, em->start,
4776 em->start + em->len);
4777 free_extent_map(em);
4781 map = (struct map_lookup *)em->bdev;
4782 offset = logical - em->start;
4784 stripe_len = map->stripe_len;
4787 * stripe_nr counts the total number of stripes we have to stride
4788 * to get to this block
4790 do_div(stripe_nr, stripe_len);
4792 stripe_offset = stripe_nr * stripe_len;
4793 BUG_ON(offset < stripe_offset);
4795 /* stripe_offset is the offset of this block in its stripe*/
4796 stripe_offset = offset - stripe_offset;
4798 /* if we're here for raid56, we need to know the stripe aligned start */
4799 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4800 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4801 raid56_full_stripe_start = offset;
4803 /* allow a write of a full stripe, but make sure we don't
4804 * allow straddling of stripes
4806 do_div(raid56_full_stripe_start, full_stripe_len);
4807 raid56_full_stripe_start *= full_stripe_len;
4810 if (rw & REQ_DISCARD) {
4811 /* we don't discard raid56 yet */
4813 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4817 *length = min_t(u64, em->len - offset, *length);
4818 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4820 /* For writes to RAID[56], allow a full stripeset across all disks.
4821 For other RAID types and for RAID[56] reads, just allow a single
4822 stripe (on a single disk). */
4823 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4825 max_len = stripe_len * nr_data_stripes(map) -
4826 (offset - raid56_full_stripe_start);
4828 /* we limit the length of each bio to what fits in a stripe */
4829 max_len = stripe_len - stripe_offset;
4831 *length = min_t(u64, em->len - offset, max_len);
4833 *length = em->len - offset;
4836 /* This is for when we're called from btrfs_merge_bio_hook() and all
4837 it cares about is the length */
4841 btrfs_dev_replace_lock(dev_replace);
4842 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4843 if (!dev_replace_is_ongoing)
4844 btrfs_dev_replace_unlock(dev_replace);
4846 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4847 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4848 dev_replace->tgtdev != NULL) {
4850 * in dev-replace case, for repair case (that's the only
4851 * case where the mirror is selected explicitly when
4852 * calling btrfs_map_block), blocks left of the left cursor
4853 * can also be read from the target drive.
4854 * For REQ_GET_READ_MIRRORS, the target drive is added as
4855 * the last one to the array of stripes. For READ, it also
4856 * needs to be supported using the same mirror number.
4857 * If the requested block is not left of the left cursor,
4858 * EIO is returned. This can happen because btrfs_num_copies()
4859 * returns one more in the dev-replace case.
4861 u64 tmp_length = *length;
4862 struct btrfs_bio *tmp_bbio = NULL;
4863 int tmp_num_stripes;
4864 u64 srcdev_devid = dev_replace->srcdev->devid;
4865 int index_srcdev = 0;
4867 u64 physical_of_found = 0;
4869 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4870 logical, &tmp_length, &tmp_bbio, 0, NULL);
4872 WARN_ON(tmp_bbio != NULL);
4876 tmp_num_stripes = tmp_bbio->num_stripes;
4877 if (mirror_num > tmp_num_stripes) {
4879 * REQ_GET_READ_MIRRORS does not contain this
4880 * mirror, that means that the requested area
4881 * is not left of the left cursor
4889 * process the rest of the function using the mirror_num
4890 * of the source drive. Therefore look it up first.
4891 * At the end, patch the device pointer to the one of the
4894 for (i = 0; i < tmp_num_stripes; i++) {
4895 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4897 * In case of DUP, in order to keep it
4898 * simple, only add the mirror with the
4899 * lowest physical address
4902 physical_of_found <=
4903 tmp_bbio->stripes[i].physical)
4908 tmp_bbio->stripes[i].physical;
4913 mirror_num = index_srcdev + 1;
4914 patch_the_first_stripe_for_dev_replace = 1;
4915 physical_to_patch_in_first_stripe = physical_of_found;
4924 } else if (mirror_num > map->num_stripes) {
4930 stripe_nr_orig = stripe_nr;
4931 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4932 do_div(stripe_nr_end, map->stripe_len);
4933 stripe_end_offset = stripe_nr_end * map->stripe_len -
4936 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4937 if (rw & REQ_DISCARD)
4938 num_stripes = min_t(u64, map->num_stripes,
4939 stripe_nr_end - stripe_nr_orig);
4940 stripe_index = do_div(stripe_nr, map->num_stripes);
4941 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4942 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4943 num_stripes = map->num_stripes;
4944 else if (mirror_num)
4945 stripe_index = mirror_num - 1;
4947 stripe_index = find_live_mirror(fs_info, map, 0,
4949 current->pid % map->num_stripes,
4950 dev_replace_is_ongoing);
4951 mirror_num = stripe_index + 1;
4954 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4955 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4956 num_stripes = map->num_stripes;
4957 } else if (mirror_num) {
4958 stripe_index = mirror_num - 1;
4963 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4964 int factor = map->num_stripes / map->sub_stripes;
4966 stripe_index = do_div(stripe_nr, factor);
4967 stripe_index *= map->sub_stripes;
4969 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4970 num_stripes = map->sub_stripes;
4971 else if (rw & REQ_DISCARD)
4972 num_stripes = min_t(u64, map->sub_stripes *
4973 (stripe_nr_end - stripe_nr_orig),
4975 else if (mirror_num)
4976 stripe_index += mirror_num - 1;
4978 int old_stripe_index = stripe_index;
4979 stripe_index = find_live_mirror(fs_info, map,
4981 map->sub_stripes, stripe_index +
4982 current->pid % map->sub_stripes,
4983 dev_replace_is_ongoing);
4984 mirror_num = stripe_index - old_stripe_index + 1;
4987 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4988 BTRFS_BLOCK_GROUP_RAID6)) {
4991 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4995 /* push stripe_nr back to the start of the full stripe */
4996 stripe_nr = raid56_full_stripe_start;
4997 do_div(stripe_nr, stripe_len);
4999 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5001 /* RAID[56] write or recovery. Return all stripes */
5002 num_stripes = map->num_stripes;
5003 max_errors = nr_parity_stripes(map);
5005 raid_map = kmalloc_array(num_stripes, sizeof(u64),
5012 /* Work out the disk rotation on this stripe-set */
5014 rot = do_div(tmp, num_stripes);
5016 /* Fill in the logical address of each stripe */
5017 tmp = stripe_nr * nr_data_stripes(map);
5018 for (i = 0; i < nr_data_stripes(map); i++)
5019 raid_map[(i+rot) % num_stripes] =
5020 em->start + (tmp + i) * map->stripe_len;
5022 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5023 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5024 raid_map[(i+rot+1) % num_stripes] =
5027 *length = map->stripe_len;
5032 * Mirror #0 or #1 means the original data block.
5033 * Mirror #2 is RAID5 parity block.
5034 * Mirror #3 is RAID6 Q block.
5036 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5038 stripe_index = nr_data_stripes(map) +
5041 /* We distribute the parity blocks across stripes */
5042 tmp = stripe_nr + stripe_index;
5043 stripe_index = do_div(tmp, map->num_stripes);
5047 * after this do_div call, stripe_nr is the number of stripes
5048 * on this device we have to walk to find the data, and
5049 * stripe_index is the number of our device in the stripe array
5051 stripe_index = do_div(stripe_nr, map->num_stripes);
5052 mirror_num = stripe_index + 1;
5054 BUG_ON(stripe_index >= map->num_stripes);
5056 num_alloc_stripes = num_stripes;
5057 if (dev_replace_is_ongoing) {
5058 if (rw & (REQ_WRITE | REQ_DISCARD))
5059 num_alloc_stripes <<= 1;
5060 if (rw & REQ_GET_READ_MIRRORS)
5061 num_alloc_stripes++;
5063 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5069 atomic_set(&bbio->error, 0);
5071 if (rw & REQ_DISCARD) {
5073 int sub_stripes = 0;
5074 u64 stripes_per_dev = 0;
5075 u32 remaining_stripes = 0;
5076 u32 last_stripe = 0;
5079 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5080 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5083 sub_stripes = map->sub_stripes;
5085 factor = map->num_stripes / sub_stripes;
5086 stripes_per_dev = div_u64_rem(stripe_nr_end -
5089 &remaining_stripes);
5090 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5091 last_stripe *= sub_stripes;
5094 for (i = 0; i < num_stripes; i++) {
5095 bbio->stripes[i].physical =
5096 map->stripes[stripe_index].physical +
5097 stripe_offset + stripe_nr * map->stripe_len;
5098 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5100 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5101 BTRFS_BLOCK_GROUP_RAID10)) {
5102 bbio->stripes[i].length = stripes_per_dev *
5105 if (i / sub_stripes < remaining_stripes)
5106 bbio->stripes[i].length +=
5110 * Special for the first stripe and
5113 * |-------|...|-------|
5117 if (i < sub_stripes)
5118 bbio->stripes[i].length -=
5121 if (stripe_index >= last_stripe &&
5122 stripe_index <= (last_stripe +
5124 bbio->stripes[i].length -=
5127 if (i == sub_stripes - 1)
5130 bbio->stripes[i].length = *length;
5133 if (stripe_index == map->num_stripes) {
5134 /* This could only happen for RAID0/10 */
5140 for (i = 0; i < num_stripes; i++) {
5141 bbio->stripes[i].physical =
5142 map->stripes[stripe_index].physical +
5144 stripe_nr * map->stripe_len;
5145 bbio->stripes[i].dev =
5146 map->stripes[stripe_index].dev;
5151 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5152 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5153 BTRFS_BLOCK_GROUP_RAID10 |
5154 BTRFS_BLOCK_GROUP_RAID5 |
5155 BTRFS_BLOCK_GROUP_DUP)) {
5157 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5162 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5163 dev_replace->tgtdev != NULL) {
5164 int index_where_to_add;
5165 u64 srcdev_devid = dev_replace->srcdev->devid;
5168 * duplicate the write operations while the dev replace
5169 * procedure is running. Since the copying of the old disk
5170 * to the new disk takes place at run time while the
5171 * filesystem is mounted writable, the regular write
5172 * operations to the old disk have to be duplicated to go
5173 * to the new disk as well.
5174 * Note that device->missing is handled by the caller, and
5175 * that the write to the old disk is already set up in the
5178 index_where_to_add = num_stripes;
5179 for (i = 0; i < num_stripes; i++) {
5180 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5181 /* write to new disk, too */
5182 struct btrfs_bio_stripe *new =
5183 bbio->stripes + index_where_to_add;
5184 struct btrfs_bio_stripe *old =
5187 new->physical = old->physical;
5188 new->length = old->length;
5189 new->dev = dev_replace->tgtdev;
5190 index_where_to_add++;
5194 num_stripes = index_where_to_add;
5195 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5196 dev_replace->tgtdev != NULL) {
5197 u64 srcdev_devid = dev_replace->srcdev->devid;
5198 int index_srcdev = 0;
5200 u64 physical_of_found = 0;
5203 * During the dev-replace procedure, the target drive can
5204 * also be used to read data in case it is needed to repair
5205 * a corrupt block elsewhere. This is possible if the
5206 * requested area is left of the left cursor. In this area,
5207 * the target drive is a full copy of the source drive.
5209 for (i = 0; i < num_stripes; i++) {
5210 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5212 * In case of DUP, in order to keep it
5213 * simple, only add the mirror with the
5214 * lowest physical address
5217 physical_of_found <=
5218 bbio->stripes[i].physical)
5222 physical_of_found = bbio->stripes[i].physical;
5226 u64 length = map->stripe_len;
5228 if (physical_of_found + length <=
5229 dev_replace->cursor_left) {
5230 struct btrfs_bio_stripe *tgtdev_stripe =
5231 bbio->stripes + num_stripes;
5233 tgtdev_stripe->physical = physical_of_found;
5234 tgtdev_stripe->length =
5235 bbio->stripes[index_srcdev].length;
5236 tgtdev_stripe->dev = dev_replace->tgtdev;
5244 bbio->num_stripes = num_stripes;
5245 bbio->max_errors = max_errors;
5246 bbio->mirror_num = mirror_num;
5249 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5250 * mirror_num == num_stripes + 1 && dev_replace target drive is
5251 * available as a mirror
5253 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5254 WARN_ON(num_stripes > 1);
5255 bbio->stripes[0].dev = dev_replace->tgtdev;
5256 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5257 bbio->mirror_num = map->num_stripes + 1;
5260 sort_parity_stripes(bbio, raid_map);
5261 *raid_map_ret = raid_map;
5264 if (dev_replace_is_ongoing)
5265 btrfs_dev_replace_unlock(dev_replace);
5266 free_extent_map(em);
5270 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5271 u64 logical, u64 *length,
5272 struct btrfs_bio **bbio_ret, int mirror_num)
5274 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5278 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5279 u64 chunk_start, u64 physical, u64 devid,
5280 u64 **logical, int *naddrs, int *stripe_len)
5282 struct extent_map_tree *em_tree = &map_tree->map_tree;
5283 struct extent_map *em;
5284 struct map_lookup *map;
5292 read_lock(&em_tree->lock);
5293 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5294 read_unlock(&em_tree->lock);
5297 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5302 if (em->start != chunk_start) {
5303 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5304 em->start, chunk_start);
5305 free_extent_map(em);
5308 map = (struct map_lookup *)em->bdev;
5311 rmap_len = map->stripe_len;
5313 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5314 do_div(length, map->num_stripes / map->sub_stripes);
5315 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5316 do_div(length, map->num_stripes);
5317 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5318 BTRFS_BLOCK_GROUP_RAID6)) {
5319 do_div(length, nr_data_stripes(map));
5320 rmap_len = map->stripe_len * nr_data_stripes(map);
5323 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5324 BUG_ON(!buf); /* -ENOMEM */
5326 for (i = 0; i < map->num_stripes; i++) {
5327 if (devid && map->stripes[i].dev->devid != devid)
5329 if (map->stripes[i].physical > physical ||
5330 map->stripes[i].physical + length <= physical)
5333 stripe_nr = physical - map->stripes[i].physical;
5334 do_div(stripe_nr, map->stripe_len);
5336 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5337 stripe_nr = stripe_nr * map->num_stripes + i;
5338 do_div(stripe_nr, map->sub_stripes);
5339 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5340 stripe_nr = stripe_nr * map->num_stripes + i;
5341 } /* else if RAID[56], multiply by nr_data_stripes().
5342 * Alternatively, just use rmap_len below instead of
5343 * map->stripe_len */
5345 bytenr = chunk_start + stripe_nr * rmap_len;
5346 WARN_ON(nr >= map->num_stripes);
5347 for (j = 0; j < nr; j++) {
5348 if (buf[j] == bytenr)
5352 WARN_ON(nr >= map->num_stripes);
5359 *stripe_len = rmap_len;
5361 free_extent_map(em);
5365 static void btrfs_end_bio(struct bio *bio, int err)
5367 struct btrfs_bio *bbio = bio->bi_private;
5368 struct btrfs_device *dev = bbio->stripes[0].dev;
5369 int is_orig_bio = 0;
5372 atomic_inc(&bbio->error);
5373 if (err == -EIO || err == -EREMOTEIO) {
5374 unsigned int stripe_index =
5375 btrfs_io_bio(bio)->stripe_index;
5377 BUG_ON(stripe_index >= bbio->num_stripes);
5378 dev = bbio->stripes[stripe_index].dev;
5380 if (bio->bi_rw & WRITE)
5381 btrfs_dev_stat_inc(dev,
5382 BTRFS_DEV_STAT_WRITE_ERRS);
5384 btrfs_dev_stat_inc(dev,
5385 BTRFS_DEV_STAT_READ_ERRS);
5386 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5387 btrfs_dev_stat_inc(dev,
5388 BTRFS_DEV_STAT_FLUSH_ERRS);
5389 btrfs_dev_stat_print_on_error(dev);
5394 if (bio == bbio->orig_bio)
5397 btrfs_bio_counter_dec(bbio->fs_info);
5399 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5402 bio = bbio->orig_bio;
5406 * We have original bio now. So increment bi_remaining to
5407 * account for it in endio
5409 atomic_inc(&bio->bi_remaining);
5411 bio->bi_private = bbio->private;
5412 bio->bi_end_io = bbio->end_io;
5413 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5414 /* only send an error to the higher layers if it is
5415 * beyond the tolerance of the btrfs bio
5417 if (atomic_read(&bbio->error) > bbio->max_errors) {
5421 * this bio is actually up to date, we didn't
5422 * go over the max number of errors
5424 set_bit(BIO_UPTODATE, &bio->bi_flags);
5429 bio_endio(bio, err);
5430 } else if (!is_orig_bio) {
5436 * see run_scheduled_bios for a description of why bios are collected for
5439 * This will add one bio to the pending list for a device and make sure
5440 * the work struct is scheduled.
5442 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5443 struct btrfs_device *device,
5444 int rw, struct bio *bio)
5446 int should_queue = 1;
5447 struct btrfs_pending_bios *pending_bios;
5449 if (device->missing || !device->bdev) {
5450 bio_endio(bio, -EIO);
5454 /* don't bother with additional async steps for reads, right now */
5455 if (!(rw & REQ_WRITE)) {
5457 btrfsic_submit_bio(rw, bio);
5463 * nr_async_bios allows us to reliably return congestion to the
5464 * higher layers. Otherwise, the async bio makes it appear we have
5465 * made progress against dirty pages when we've really just put it
5466 * on a queue for later
5468 atomic_inc(&root->fs_info->nr_async_bios);
5469 WARN_ON(bio->bi_next);
5470 bio->bi_next = NULL;
5473 spin_lock(&device->io_lock);
5474 if (bio->bi_rw & REQ_SYNC)
5475 pending_bios = &device->pending_sync_bios;
5477 pending_bios = &device->pending_bios;
5479 if (pending_bios->tail)
5480 pending_bios->tail->bi_next = bio;
5482 pending_bios->tail = bio;
5483 if (!pending_bios->head)
5484 pending_bios->head = bio;
5485 if (device->running_pending)
5488 spin_unlock(&device->io_lock);
5491 btrfs_queue_work(root->fs_info->submit_workers,
5495 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5498 struct bio_vec *prev;
5499 struct request_queue *q = bdev_get_queue(bdev);
5500 unsigned int max_sectors = queue_max_sectors(q);
5501 struct bvec_merge_data bvm = {
5503 .bi_sector = sector,
5504 .bi_rw = bio->bi_rw,
5507 if (WARN_ON(bio->bi_vcnt == 0))
5510 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5511 if (bio_sectors(bio) > max_sectors)
5514 if (!q->merge_bvec_fn)
5517 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5518 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5523 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5524 struct bio *bio, u64 physical, int dev_nr,
5527 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5529 bio->bi_private = bbio;
5530 btrfs_io_bio(bio)->stripe_index = dev_nr;
5531 bio->bi_end_io = btrfs_end_bio;
5532 bio->bi_iter.bi_sector = physical >> 9;
5535 struct rcu_string *name;
5538 name = rcu_dereference(dev->name);
5539 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5540 "(%s id %llu), size=%u\n", rw,
5541 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5542 name->str, dev->devid, bio->bi_size);
5546 bio->bi_bdev = dev->bdev;
5548 btrfs_bio_counter_inc_noblocked(root->fs_info);
5551 btrfs_schedule_bio(root, dev, rw, bio);
5553 btrfsic_submit_bio(rw, bio);
5556 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5557 struct bio *first_bio, struct btrfs_device *dev,
5558 int dev_nr, int rw, int async)
5560 struct bio_vec *bvec = first_bio->bi_io_vec;
5562 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5563 u64 physical = bbio->stripes[dev_nr].physical;
5566 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5570 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5571 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5572 bvec->bv_offset) < bvec->bv_len) {
5573 u64 len = bio->bi_iter.bi_size;
5575 atomic_inc(&bbio->stripes_pending);
5576 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5584 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5588 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5590 atomic_inc(&bbio->error);
5591 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5592 bio->bi_private = bbio->private;
5593 bio->bi_end_io = bbio->end_io;
5594 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5595 bio->bi_iter.bi_sector = logical >> 9;
5597 bio_endio(bio, -EIO);
5601 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5602 int mirror_num, int async_submit)
5604 struct btrfs_device *dev;
5605 struct bio *first_bio = bio;
5606 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5609 u64 *raid_map = NULL;
5613 struct btrfs_bio *bbio = NULL;
5615 length = bio->bi_iter.bi_size;
5616 map_length = length;
5618 btrfs_bio_counter_inc_blocked(root->fs_info);
5619 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5620 mirror_num, &raid_map);
5622 btrfs_bio_counter_dec(root->fs_info);
5626 total_devs = bbio->num_stripes;
5627 bbio->orig_bio = first_bio;
5628 bbio->private = first_bio->bi_private;
5629 bbio->end_io = first_bio->bi_end_io;
5630 bbio->fs_info = root->fs_info;
5631 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5634 /* In this case, map_length has been set to the length of
5635 a single stripe; not the whole write */
5637 ret = raid56_parity_write(root, bio, bbio,
5638 raid_map, map_length);
5640 ret = raid56_parity_recover(root, bio, bbio,
5641 raid_map, map_length,
5645 * FIXME, replace dosen't support raid56 yet, please fix
5648 btrfs_bio_counter_dec(root->fs_info);
5652 if (map_length < length) {
5653 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5654 logical, length, map_length);
5658 while (dev_nr < total_devs) {
5659 dev = bbio->stripes[dev_nr].dev;
5660 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5661 bbio_error(bbio, first_bio, logical);
5667 * Check and see if we're ok with this bio based on it's size
5668 * and offset with the given device.
5670 if (!bio_size_ok(dev->bdev, first_bio,
5671 bbio->stripes[dev_nr].physical >> 9)) {
5672 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5673 dev_nr, rw, async_submit);
5679 if (dev_nr < total_devs - 1) {
5680 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5681 BUG_ON(!bio); /* -ENOMEM */
5686 submit_stripe_bio(root, bbio, bio,
5687 bbio->stripes[dev_nr].physical, dev_nr, rw,
5691 btrfs_bio_counter_dec(root->fs_info);
5695 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5698 struct btrfs_device *device;
5699 struct btrfs_fs_devices *cur_devices;
5701 cur_devices = fs_info->fs_devices;
5702 while (cur_devices) {
5704 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5705 device = __find_device(&cur_devices->devices,
5710 cur_devices = cur_devices->seed;
5715 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5716 u64 devid, u8 *dev_uuid)
5718 struct btrfs_device *device;
5719 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5721 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5725 list_add(&device->dev_list, &fs_devices->devices);
5726 device->fs_devices = fs_devices;
5727 fs_devices->num_devices++;
5729 device->missing = 1;
5730 fs_devices->missing_devices++;
5736 * btrfs_alloc_device - allocate struct btrfs_device
5737 * @fs_info: used only for generating a new devid, can be NULL if
5738 * devid is provided (i.e. @devid != NULL).
5739 * @devid: a pointer to devid for this device. If NULL a new devid
5741 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5744 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5745 * on error. Returned struct is not linked onto any lists and can be
5746 * destroyed with kfree() right away.
5748 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5752 struct btrfs_device *dev;
5755 if (WARN_ON(!devid && !fs_info))
5756 return ERR_PTR(-EINVAL);
5758 dev = __alloc_device();
5767 ret = find_next_devid(fs_info, &tmp);
5770 return ERR_PTR(ret);
5776 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5778 generate_random_uuid(dev->uuid);
5780 btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5785 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5786 struct extent_buffer *leaf,
5787 struct btrfs_chunk *chunk)
5789 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5790 struct map_lookup *map;
5791 struct extent_map *em;
5795 u8 uuid[BTRFS_UUID_SIZE];
5800 logical = key->offset;
5801 length = btrfs_chunk_length(leaf, chunk);
5803 read_lock(&map_tree->map_tree.lock);
5804 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5805 read_unlock(&map_tree->map_tree.lock);
5807 /* already mapped? */
5808 if (em && em->start <= logical && em->start + em->len > logical) {
5809 free_extent_map(em);
5812 free_extent_map(em);
5815 em = alloc_extent_map();
5818 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5819 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5821 free_extent_map(em);
5825 em->bdev = (struct block_device *)map;
5826 em->start = logical;
5829 em->block_start = 0;
5830 em->block_len = em->len;
5832 map->num_stripes = num_stripes;
5833 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5834 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5835 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5836 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5837 map->type = btrfs_chunk_type(leaf, chunk);
5838 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5839 for (i = 0; i < num_stripes; i++) {
5840 map->stripes[i].physical =
5841 btrfs_stripe_offset_nr(leaf, chunk, i);
5842 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5843 read_extent_buffer(leaf, uuid, (unsigned long)
5844 btrfs_stripe_dev_uuid_nr(chunk, i),
5846 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5848 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5850 free_extent_map(em);
5853 if (!map->stripes[i].dev) {
5854 map->stripes[i].dev =
5855 add_missing_dev(root, devid, uuid);
5856 if (!map->stripes[i].dev) {
5858 free_extent_map(em);
5862 map->stripes[i].dev->in_fs_metadata = 1;
5865 write_lock(&map_tree->map_tree.lock);
5866 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5867 write_unlock(&map_tree->map_tree.lock);
5868 BUG_ON(ret); /* Tree corruption */
5869 free_extent_map(em);
5874 static void fill_device_from_item(struct extent_buffer *leaf,
5875 struct btrfs_dev_item *dev_item,
5876 struct btrfs_device *device)
5880 device->devid = btrfs_device_id(leaf, dev_item);
5881 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5882 device->total_bytes = device->disk_total_bytes;
5883 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5884 device->type = btrfs_device_type(leaf, dev_item);
5885 device->io_align = btrfs_device_io_align(leaf, dev_item);
5886 device->io_width = btrfs_device_io_width(leaf, dev_item);
5887 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5888 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5889 device->is_tgtdev_for_dev_replace = 0;
5891 ptr = btrfs_device_uuid(dev_item);
5892 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5895 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5897 struct btrfs_fs_devices *fs_devices;
5900 BUG_ON(!mutex_is_locked(&uuid_mutex));
5902 fs_devices = root->fs_info->fs_devices->seed;
5903 while (fs_devices) {
5904 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5908 fs_devices = fs_devices->seed;
5911 fs_devices = find_fsid(fsid);
5917 fs_devices = clone_fs_devices(fs_devices);
5918 if (IS_ERR(fs_devices)) {
5919 ret = PTR_ERR(fs_devices);
5923 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5924 root->fs_info->bdev_holder);
5926 free_fs_devices(fs_devices);
5930 if (!fs_devices->seeding) {
5931 __btrfs_close_devices(fs_devices);
5932 free_fs_devices(fs_devices);
5937 fs_devices->seed = root->fs_info->fs_devices->seed;
5938 root->fs_info->fs_devices->seed = fs_devices;
5943 static int read_one_dev(struct btrfs_root *root,
5944 struct extent_buffer *leaf,
5945 struct btrfs_dev_item *dev_item)
5947 struct btrfs_device *device;
5950 u8 fs_uuid[BTRFS_UUID_SIZE];
5951 u8 dev_uuid[BTRFS_UUID_SIZE];
5953 devid = btrfs_device_id(leaf, dev_item);
5954 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5956 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5959 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5960 ret = open_seed_devices(root, fs_uuid);
5961 if (ret && !btrfs_test_opt(root, DEGRADED))
5965 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5966 if (!device || !device->bdev) {
5967 if (!btrfs_test_opt(root, DEGRADED))
5971 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5972 device = add_missing_dev(root, devid, dev_uuid);
5975 } else if (!device->missing) {
5977 * this happens when a device that was properly setup
5978 * in the device info lists suddenly goes bad.
5979 * device->bdev is NULL, and so we have to set
5980 * device->missing to one here
5982 root->fs_info->fs_devices->missing_devices++;
5983 device->missing = 1;
5987 if (device->fs_devices != root->fs_info->fs_devices) {
5988 BUG_ON(device->writeable);
5989 if (device->generation !=
5990 btrfs_device_generation(leaf, dev_item))
5994 fill_device_from_item(leaf, dev_item, device);
5995 device->in_fs_metadata = 1;
5996 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5997 device->fs_devices->total_rw_bytes += device->total_bytes;
5998 spin_lock(&root->fs_info->free_chunk_lock);
5999 root->fs_info->free_chunk_space += device->total_bytes -
6001 spin_unlock(&root->fs_info->free_chunk_lock);
6007 int btrfs_read_sys_array(struct btrfs_root *root)
6009 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6010 struct extent_buffer *sb;
6011 struct btrfs_disk_key *disk_key;
6012 struct btrfs_chunk *chunk;
6014 unsigned long sb_ptr;
6020 struct btrfs_key key;
6022 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6023 BTRFS_SUPER_INFO_SIZE);
6026 btrfs_set_buffer_uptodate(sb);
6027 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6029 * The sb extent buffer is artifical and just used to read the system array.
6030 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6031 * pages up-to-date when the page is larger: extent does not cover the
6032 * whole page and consequently check_page_uptodate does not find all
6033 * the page's extents up-to-date (the hole beyond sb),
6034 * write_extent_buffer then triggers a WARN_ON.
6036 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6037 * but sb spans only this function. Add an explicit SetPageUptodate call
6038 * to silence the warning eg. on PowerPC 64.
6040 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6041 SetPageUptodate(sb->pages[0]);
6043 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6044 array_size = btrfs_super_sys_array_size(super_copy);
6046 ptr = super_copy->sys_chunk_array;
6047 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6050 while (cur < array_size) {
6051 disk_key = (struct btrfs_disk_key *)ptr;
6052 btrfs_disk_key_to_cpu(&key, disk_key);
6054 len = sizeof(*disk_key); ptr += len;
6058 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6059 chunk = (struct btrfs_chunk *)sb_ptr;
6060 ret = read_one_chunk(root, &key, sb, chunk);
6063 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6064 len = btrfs_chunk_item_size(num_stripes);
6073 free_extent_buffer(sb);
6077 int btrfs_read_chunk_tree(struct btrfs_root *root)
6079 struct btrfs_path *path;
6080 struct extent_buffer *leaf;
6081 struct btrfs_key key;
6082 struct btrfs_key found_key;
6086 root = root->fs_info->chunk_root;
6088 path = btrfs_alloc_path();
6092 mutex_lock(&uuid_mutex);
6096 * Read all device items, and then all the chunk items. All
6097 * device items are found before any chunk item (their object id
6098 * is smaller than the lowest possible object id for a chunk
6099 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6101 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6104 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6108 leaf = path->nodes[0];
6109 slot = path->slots[0];
6110 if (slot >= btrfs_header_nritems(leaf)) {
6111 ret = btrfs_next_leaf(root, path);
6118 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6119 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6120 struct btrfs_dev_item *dev_item;
6121 dev_item = btrfs_item_ptr(leaf, slot,
6122 struct btrfs_dev_item);
6123 ret = read_one_dev(root, leaf, dev_item);
6126 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6127 struct btrfs_chunk *chunk;
6128 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6129 ret = read_one_chunk(root, &found_key, leaf, chunk);
6137 unlock_chunks(root);
6138 mutex_unlock(&uuid_mutex);
6140 btrfs_free_path(path);
6144 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6146 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6147 struct btrfs_device *device;
6149 while (fs_devices) {
6150 mutex_lock(&fs_devices->device_list_mutex);
6151 list_for_each_entry(device, &fs_devices->devices, dev_list)
6152 device->dev_root = fs_info->dev_root;
6153 mutex_unlock(&fs_devices->device_list_mutex);
6155 fs_devices = fs_devices->seed;
6159 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6163 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6164 btrfs_dev_stat_reset(dev, i);
6167 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6169 struct btrfs_key key;
6170 struct btrfs_key found_key;
6171 struct btrfs_root *dev_root = fs_info->dev_root;
6172 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6173 struct extent_buffer *eb;
6176 struct btrfs_device *device;
6177 struct btrfs_path *path = NULL;
6180 path = btrfs_alloc_path();
6186 mutex_lock(&fs_devices->device_list_mutex);
6187 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6189 struct btrfs_dev_stats_item *ptr;
6192 key.type = BTRFS_DEV_STATS_KEY;
6193 key.offset = device->devid;
6194 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6196 __btrfs_reset_dev_stats(device);
6197 device->dev_stats_valid = 1;
6198 btrfs_release_path(path);
6201 slot = path->slots[0];
6202 eb = path->nodes[0];
6203 btrfs_item_key_to_cpu(eb, &found_key, slot);
6204 item_size = btrfs_item_size_nr(eb, slot);
6206 ptr = btrfs_item_ptr(eb, slot,
6207 struct btrfs_dev_stats_item);
6209 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6210 if (item_size >= (1 + i) * sizeof(__le64))
6211 btrfs_dev_stat_set(device, i,
6212 btrfs_dev_stats_value(eb, ptr, i));
6214 btrfs_dev_stat_reset(device, i);
6217 device->dev_stats_valid = 1;
6218 btrfs_dev_stat_print_on_load(device);
6219 btrfs_release_path(path);
6221 mutex_unlock(&fs_devices->device_list_mutex);
6224 btrfs_free_path(path);
6225 return ret < 0 ? ret : 0;
6228 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6229 struct btrfs_root *dev_root,
6230 struct btrfs_device *device)
6232 struct btrfs_path *path;
6233 struct btrfs_key key;
6234 struct extent_buffer *eb;
6235 struct btrfs_dev_stats_item *ptr;
6240 key.type = BTRFS_DEV_STATS_KEY;
6241 key.offset = device->devid;
6243 path = btrfs_alloc_path();
6245 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6247 printk_in_rcu(KERN_WARNING "BTRFS: "
6248 "error %d while searching for dev_stats item for device %s!\n",
6249 ret, rcu_str_deref(device->name));
6254 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6255 /* need to delete old one and insert a new one */
6256 ret = btrfs_del_item(trans, dev_root, path);
6258 printk_in_rcu(KERN_WARNING "BTRFS: "
6259 "delete too small dev_stats item for device %s failed %d!\n",
6260 rcu_str_deref(device->name), ret);
6267 /* need to insert a new item */
6268 btrfs_release_path(path);
6269 ret = btrfs_insert_empty_item(trans, dev_root, path,
6270 &key, sizeof(*ptr));
6272 printk_in_rcu(KERN_WARNING "BTRFS: "
6273 "insert dev_stats item for device %s failed %d!\n",
6274 rcu_str_deref(device->name), ret);
6279 eb = path->nodes[0];
6280 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6281 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6282 btrfs_set_dev_stats_value(eb, ptr, i,
6283 btrfs_dev_stat_read(device, i));
6284 btrfs_mark_buffer_dirty(eb);
6287 btrfs_free_path(path);
6292 * called from commit_transaction. Writes all changed device stats to disk.
6294 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6295 struct btrfs_fs_info *fs_info)
6297 struct btrfs_root *dev_root = fs_info->dev_root;
6298 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6299 struct btrfs_device *device;
6302 mutex_lock(&fs_devices->device_list_mutex);
6303 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6304 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6307 ret = update_dev_stat_item(trans, dev_root, device);
6309 device->dev_stats_dirty = 0;
6311 mutex_unlock(&fs_devices->device_list_mutex);
6316 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6318 btrfs_dev_stat_inc(dev, index);
6319 btrfs_dev_stat_print_on_error(dev);
6322 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6324 if (!dev->dev_stats_valid)
6326 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6327 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6328 rcu_str_deref(dev->name),
6329 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6330 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6331 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6332 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6333 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6336 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6340 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6341 if (btrfs_dev_stat_read(dev, i) != 0)
6343 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6344 return; /* all values == 0, suppress message */
6346 printk_in_rcu(KERN_INFO "BTRFS: "
6347 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6348 rcu_str_deref(dev->name),
6349 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6350 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6351 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6352 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6353 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6356 int btrfs_get_dev_stats(struct btrfs_root *root,
6357 struct btrfs_ioctl_get_dev_stats *stats)
6359 struct btrfs_device *dev;
6360 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6363 mutex_lock(&fs_devices->device_list_mutex);
6364 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6365 mutex_unlock(&fs_devices->device_list_mutex);
6368 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6370 } else if (!dev->dev_stats_valid) {
6371 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6373 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6374 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6375 if (stats->nr_items > i)
6377 btrfs_dev_stat_read_and_reset(dev, i);
6379 btrfs_dev_stat_reset(dev, i);
6382 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6383 if (stats->nr_items > i)
6384 stats->values[i] = btrfs_dev_stat_read(dev, i);
6386 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6387 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6391 int btrfs_scratch_superblock(struct btrfs_device *device)
6393 struct buffer_head *bh;
6394 struct btrfs_super_block *disk_super;
6396 bh = btrfs_read_dev_super(device->bdev);
6399 disk_super = (struct btrfs_super_block *)bh->b_data;
6401 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6402 set_buffer_dirty(bh);
6403 sync_dirty_buffer(bh);