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);
2546 /* once for the tree */
2547 free_extent_map(em);
2549 free_extent_map(em);
2551 unlock_chunks(root);
2552 btrfs_end_transaction(trans, root);
2556 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2558 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2559 struct btrfs_path *path;
2560 struct extent_buffer *leaf;
2561 struct btrfs_chunk *chunk;
2562 struct btrfs_key key;
2563 struct btrfs_key found_key;
2564 u64 chunk_tree = chunk_root->root_key.objectid;
2566 bool retried = false;
2570 path = btrfs_alloc_path();
2575 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2576 key.offset = (u64)-1;
2577 key.type = BTRFS_CHUNK_ITEM_KEY;
2580 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2583 BUG_ON(ret == 0); /* Corruption */
2585 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2592 leaf = path->nodes[0];
2593 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2595 chunk = btrfs_item_ptr(leaf, path->slots[0],
2596 struct btrfs_chunk);
2597 chunk_type = btrfs_chunk_type(leaf, chunk);
2598 btrfs_release_path(path);
2600 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2601 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2610 if (found_key.offset == 0)
2612 key.offset = found_key.offset - 1;
2615 if (failed && !retried) {
2619 } else if (WARN_ON(failed && retried)) {
2623 btrfs_free_path(path);
2627 static int insert_balance_item(struct btrfs_root *root,
2628 struct btrfs_balance_control *bctl)
2630 struct btrfs_trans_handle *trans;
2631 struct btrfs_balance_item *item;
2632 struct btrfs_disk_balance_args disk_bargs;
2633 struct btrfs_path *path;
2634 struct extent_buffer *leaf;
2635 struct btrfs_key key;
2638 path = btrfs_alloc_path();
2642 trans = btrfs_start_transaction(root, 0);
2643 if (IS_ERR(trans)) {
2644 btrfs_free_path(path);
2645 return PTR_ERR(trans);
2648 key.objectid = BTRFS_BALANCE_OBJECTID;
2649 key.type = BTRFS_BALANCE_ITEM_KEY;
2652 ret = btrfs_insert_empty_item(trans, root, path, &key,
2657 leaf = path->nodes[0];
2658 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2660 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2662 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2663 btrfs_set_balance_data(leaf, item, &disk_bargs);
2664 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2665 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2666 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2667 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2669 btrfs_set_balance_flags(leaf, item, bctl->flags);
2671 btrfs_mark_buffer_dirty(leaf);
2673 btrfs_free_path(path);
2674 err = btrfs_commit_transaction(trans, root);
2680 static int del_balance_item(struct btrfs_root *root)
2682 struct btrfs_trans_handle *trans;
2683 struct btrfs_path *path;
2684 struct btrfs_key key;
2687 path = btrfs_alloc_path();
2691 trans = btrfs_start_transaction(root, 0);
2692 if (IS_ERR(trans)) {
2693 btrfs_free_path(path);
2694 return PTR_ERR(trans);
2697 key.objectid = BTRFS_BALANCE_OBJECTID;
2698 key.type = BTRFS_BALANCE_ITEM_KEY;
2701 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2709 ret = btrfs_del_item(trans, root, path);
2711 btrfs_free_path(path);
2712 err = btrfs_commit_transaction(trans, root);
2719 * This is a heuristic used to reduce the number of chunks balanced on
2720 * resume after balance was interrupted.
2722 static void update_balance_args(struct btrfs_balance_control *bctl)
2725 * Turn on soft mode for chunk types that were being converted.
2727 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2728 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2729 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2730 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2731 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2732 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2735 * Turn on usage filter if is not already used. The idea is
2736 * that chunks that we have already balanced should be
2737 * reasonably full. Don't do it for chunks that are being
2738 * converted - that will keep us from relocating unconverted
2739 * (albeit full) chunks.
2741 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2742 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2743 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2744 bctl->data.usage = 90;
2746 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2747 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2748 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2749 bctl->sys.usage = 90;
2751 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2752 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2753 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2754 bctl->meta.usage = 90;
2759 * Should be called with both balance and volume mutexes held to
2760 * serialize other volume operations (add_dev/rm_dev/resize) with
2761 * restriper. Same goes for unset_balance_control.
2763 static void set_balance_control(struct btrfs_balance_control *bctl)
2765 struct btrfs_fs_info *fs_info = bctl->fs_info;
2767 BUG_ON(fs_info->balance_ctl);
2769 spin_lock(&fs_info->balance_lock);
2770 fs_info->balance_ctl = bctl;
2771 spin_unlock(&fs_info->balance_lock);
2774 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2776 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2778 BUG_ON(!fs_info->balance_ctl);
2780 spin_lock(&fs_info->balance_lock);
2781 fs_info->balance_ctl = NULL;
2782 spin_unlock(&fs_info->balance_lock);
2788 * Balance filters. Return 1 if chunk should be filtered out
2789 * (should not be balanced).
2791 static int chunk_profiles_filter(u64 chunk_type,
2792 struct btrfs_balance_args *bargs)
2794 chunk_type = chunk_to_extended(chunk_type) &
2795 BTRFS_EXTENDED_PROFILE_MASK;
2797 if (bargs->profiles & chunk_type)
2803 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2804 struct btrfs_balance_args *bargs)
2806 struct btrfs_block_group_cache *cache;
2807 u64 chunk_used, user_thresh;
2810 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2811 chunk_used = btrfs_block_group_used(&cache->item);
2813 if (bargs->usage == 0)
2815 else if (bargs->usage > 100)
2816 user_thresh = cache->key.offset;
2818 user_thresh = div_factor_fine(cache->key.offset,
2821 if (chunk_used < user_thresh)
2824 btrfs_put_block_group(cache);
2828 static int chunk_devid_filter(struct extent_buffer *leaf,
2829 struct btrfs_chunk *chunk,
2830 struct btrfs_balance_args *bargs)
2832 struct btrfs_stripe *stripe;
2833 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2836 for (i = 0; i < num_stripes; i++) {
2837 stripe = btrfs_stripe_nr(chunk, i);
2838 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2845 /* [pstart, pend) */
2846 static int chunk_drange_filter(struct extent_buffer *leaf,
2847 struct btrfs_chunk *chunk,
2849 struct btrfs_balance_args *bargs)
2851 struct btrfs_stripe *stripe;
2852 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2858 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2861 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2862 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2863 factor = num_stripes / 2;
2864 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2865 factor = num_stripes - 1;
2866 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2867 factor = num_stripes - 2;
2869 factor = num_stripes;
2872 for (i = 0; i < num_stripes; i++) {
2873 stripe = btrfs_stripe_nr(chunk, i);
2874 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2877 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2878 stripe_length = btrfs_chunk_length(leaf, chunk);
2879 do_div(stripe_length, factor);
2881 if (stripe_offset < bargs->pend &&
2882 stripe_offset + stripe_length > bargs->pstart)
2889 /* [vstart, vend) */
2890 static int chunk_vrange_filter(struct extent_buffer *leaf,
2891 struct btrfs_chunk *chunk,
2893 struct btrfs_balance_args *bargs)
2895 if (chunk_offset < bargs->vend &&
2896 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2897 /* at least part of the chunk is inside this vrange */
2903 static int chunk_soft_convert_filter(u64 chunk_type,
2904 struct btrfs_balance_args *bargs)
2906 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2909 chunk_type = chunk_to_extended(chunk_type) &
2910 BTRFS_EXTENDED_PROFILE_MASK;
2912 if (bargs->target == chunk_type)
2918 static int should_balance_chunk(struct btrfs_root *root,
2919 struct extent_buffer *leaf,
2920 struct btrfs_chunk *chunk, u64 chunk_offset)
2922 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2923 struct btrfs_balance_args *bargs = NULL;
2924 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2927 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2928 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2932 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2933 bargs = &bctl->data;
2934 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2936 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2937 bargs = &bctl->meta;
2939 /* profiles filter */
2940 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2941 chunk_profiles_filter(chunk_type, bargs)) {
2946 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2947 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2952 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2953 chunk_devid_filter(leaf, chunk, bargs)) {
2957 /* drange filter, makes sense only with devid filter */
2958 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2959 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2964 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2965 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2969 /* soft profile changing mode */
2970 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2971 chunk_soft_convert_filter(chunk_type, bargs)) {
2976 * limited by count, must be the last filter
2978 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
2979 if (bargs->limit == 0)
2988 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2990 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2991 struct btrfs_root *chunk_root = fs_info->chunk_root;
2992 struct btrfs_root *dev_root = fs_info->dev_root;
2993 struct list_head *devices;
2994 struct btrfs_device *device;
2997 struct btrfs_chunk *chunk;
2998 struct btrfs_path *path;
2999 struct btrfs_key key;
3000 struct btrfs_key found_key;
3001 struct btrfs_trans_handle *trans;
3002 struct extent_buffer *leaf;
3005 int enospc_errors = 0;
3006 bool counting = true;
3007 u64 limit_data = bctl->data.limit;
3008 u64 limit_meta = bctl->meta.limit;
3009 u64 limit_sys = bctl->sys.limit;
3011 /* step one make some room on all the devices */
3012 devices = &fs_info->fs_devices->devices;
3013 list_for_each_entry(device, devices, dev_list) {
3014 old_size = device->total_bytes;
3015 size_to_free = div_factor(old_size, 1);
3016 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3017 if (!device->writeable ||
3018 device->total_bytes - device->bytes_used > size_to_free ||
3019 device->is_tgtdev_for_dev_replace)
3022 ret = btrfs_shrink_device(device, old_size - size_to_free);
3027 trans = btrfs_start_transaction(dev_root, 0);
3028 BUG_ON(IS_ERR(trans));
3030 ret = btrfs_grow_device(trans, device, old_size);
3033 btrfs_end_transaction(trans, dev_root);
3036 /* step two, relocate all the chunks */
3037 path = btrfs_alloc_path();
3043 /* zero out stat counters */
3044 spin_lock(&fs_info->balance_lock);
3045 memset(&bctl->stat, 0, sizeof(bctl->stat));
3046 spin_unlock(&fs_info->balance_lock);
3049 bctl->data.limit = limit_data;
3050 bctl->meta.limit = limit_meta;
3051 bctl->sys.limit = limit_sys;
3053 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3054 key.offset = (u64)-1;
3055 key.type = BTRFS_CHUNK_ITEM_KEY;
3058 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3059 atomic_read(&fs_info->balance_cancel_req)) {
3064 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3069 * this shouldn't happen, it means the last relocate
3073 BUG(); /* FIXME break ? */
3075 ret = btrfs_previous_item(chunk_root, path, 0,
3076 BTRFS_CHUNK_ITEM_KEY);
3082 leaf = path->nodes[0];
3083 slot = path->slots[0];
3084 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3086 if (found_key.objectid != key.objectid)
3089 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3092 spin_lock(&fs_info->balance_lock);
3093 bctl->stat.considered++;
3094 spin_unlock(&fs_info->balance_lock);
3097 ret = should_balance_chunk(chunk_root, leaf, chunk,
3099 btrfs_release_path(path);
3104 spin_lock(&fs_info->balance_lock);
3105 bctl->stat.expected++;
3106 spin_unlock(&fs_info->balance_lock);
3110 ret = btrfs_relocate_chunk(chunk_root,
3111 chunk_root->root_key.objectid,
3114 if (ret && ret != -ENOSPC)
3116 if (ret == -ENOSPC) {
3119 spin_lock(&fs_info->balance_lock);
3120 bctl->stat.completed++;
3121 spin_unlock(&fs_info->balance_lock);
3124 if (found_key.offset == 0)
3126 key.offset = found_key.offset - 1;
3130 btrfs_release_path(path);
3135 btrfs_free_path(path);
3136 if (enospc_errors) {
3137 btrfs_info(fs_info, "%d enospc errors during balance",
3147 * alloc_profile_is_valid - see if a given profile is valid and reduced
3148 * @flags: profile to validate
3149 * @extended: if true @flags is treated as an extended profile
3151 static int alloc_profile_is_valid(u64 flags, int extended)
3153 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3154 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3156 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3158 /* 1) check that all other bits are zeroed */
3162 /* 2) see if profile is reduced */
3164 return !extended; /* "0" is valid for usual profiles */
3166 /* true if exactly one bit set */
3167 return (flags & (flags - 1)) == 0;
3170 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3172 /* cancel requested || normal exit path */
3173 return atomic_read(&fs_info->balance_cancel_req) ||
3174 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3175 atomic_read(&fs_info->balance_cancel_req) == 0);
3178 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3182 unset_balance_control(fs_info);
3183 ret = del_balance_item(fs_info->tree_root);
3185 btrfs_std_error(fs_info, ret);
3187 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3191 * Should be called with both balance and volume mutexes held
3193 int btrfs_balance(struct btrfs_balance_control *bctl,
3194 struct btrfs_ioctl_balance_args *bargs)
3196 struct btrfs_fs_info *fs_info = bctl->fs_info;
3203 if (btrfs_fs_closing(fs_info) ||
3204 atomic_read(&fs_info->balance_pause_req) ||
3205 atomic_read(&fs_info->balance_cancel_req)) {
3210 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3211 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3215 * In case of mixed groups both data and meta should be picked,
3216 * and identical options should be given for both of them.
3218 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3219 if (mixed && (bctl->flags & allowed)) {
3220 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3221 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3222 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3223 btrfs_err(fs_info, "with mixed groups data and "
3224 "metadata balance options must be the same");
3230 num_devices = fs_info->fs_devices->num_devices;
3231 btrfs_dev_replace_lock(&fs_info->dev_replace);
3232 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3233 BUG_ON(num_devices < 1);
3236 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3237 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3238 if (num_devices == 1)
3239 allowed |= BTRFS_BLOCK_GROUP_DUP;
3240 else if (num_devices > 1)
3241 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3242 if (num_devices > 2)
3243 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3244 if (num_devices > 3)
3245 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3246 BTRFS_BLOCK_GROUP_RAID6);
3247 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3248 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3249 (bctl->data.target & ~allowed))) {
3250 btrfs_err(fs_info, "unable to start balance with target "
3251 "data profile %llu",
3256 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3257 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3258 (bctl->meta.target & ~allowed))) {
3260 "unable to start balance with target metadata profile %llu",
3265 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3266 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3267 (bctl->sys.target & ~allowed))) {
3269 "unable to start balance with target system profile %llu",
3275 /* allow dup'ed data chunks only in mixed mode */
3276 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3277 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3278 btrfs_err(fs_info, "dup for data is not allowed");
3283 /* allow to reduce meta or sys integrity only if force set */
3284 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3285 BTRFS_BLOCK_GROUP_RAID10 |
3286 BTRFS_BLOCK_GROUP_RAID5 |
3287 BTRFS_BLOCK_GROUP_RAID6;
3289 seq = read_seqbegin(&fs_info->profiles_lock);
3291 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3292 (fs_info->avail_system_alloc_bits & allowed) &&
3293 !(bctl->sys.target & allowed)) ||
3294 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3295 (fs_info->avail_metadata_alloc_bits & allowed) &&
3296 !(bctl->meta.target & allowed))) {
3297 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3298 btrfs_info(fs_info, "force reducing metadata integrity");
3300 btrfs_err(fs_info, "balance will reduce metadata "
3301 "integrity, use force if you want this");
3306 } while (read_seqretry(&fs_info->profiles_lock, seq));
3308 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3309 int num_tolerated_disk_barrier_failures;
3310 u64 target = bctl->sys.target;
3312 num_tolerated_disk_barrier_failures =
3313 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3314 if (num_tolerated_disk_barrier_failures > 0 &&
3316 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3317 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3318 num_tolerated_disk_barrier_failures = 0;
3319 else if (num_tolerated_disk_barrier_failures > 1 &&
3321 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3322 num_tolerated_disk_barrier_failures = 1;
3324 fs_info->num_tolerated_disk_barrier_failures =
3325 num_tolerated_disk_barrier_failures;
3328 ret = insert_balance_item(fs_info->tree_root, bctl);
3329 if (ret && ret != -EEXIST)
3332 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3333 BUG_ON(ret == -EEXIST);
3334 set_balance_control(bctl);
3336 BUG_ON(ret != -EEXIST);
3337 spin_lock(&fs_info->balance_lock);
3338 update_balance_args(bctl);
3339 spin_unlock(&fs_info->balance_lock);
3342 atomic_inc(&fs_info->balance_running);
3343 mutex_unlock(&fs_info->balance_mutex);
3345 ret = __btrfs_balance(fs_info);
3347 mutex_lock(&fs_info->balance_mutex);
3348 atomic_dec(&fs_info->balance_running);
3350 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3351 fs_info->num_tolerated_disk_barrier_failures =
3352 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3356 memset(bargs, 0, sizeof(*bargs));
3357 update_ioctl_balance_args(fs_info, 0, bargs);
3360 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3361 balance_need_close(fs_info)) {
3362 __cancel_balance(fs_info);
3365 wake_up(&fs_info->balance_wait_q);
3369 if (bctl->flags & BTRFS_BALANCE_RESUME)
3370 __cancel_balance(fs_info);
3373 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3378 static int balance_kthread(void *data)
3380 struct btrfs_fs_info *fs_info = data;
3383 mutex_lock(&fs_info->volume_mutex);
3384 mutex_lock(&fs_info->balance_mutex);
3386 if (fs_info->balance_ctl) {
3387 btrfs_info(fs_info, "continuing balance");
3388 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3391 mutex_unlock(&fs_info->balance_mutex);
3392 mutex_unlock(&fs_info->volume_mutex);
3397 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3399 struct task_struct *tsk;
3401 spin_lock(&fs_info->balance_lock);
3402 if (!fs_info->balance_ctl) {
3403 spin_unlock(&fs_info->balance_lock);
3406 spin_unlock(&fs_info->balance_lock);
3408 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3409 btrfs_info(fs_info, "force skipping balance");
3413 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3414 return PTR_ERR_OR_ZERO(tsk);
3417 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3419 struct btrfs_balance_control *bctl;
3420 struct btrfs_balance_item *item;
3421 struct btrfs_disk_balance_args disk_bargs;
3422 struct btrfs_path *path;
3423 struct extent_buffer *leaf;
3424 struct btrfs_key key;
3427 path = btrfs_alloc_path();
3431 key.objectid = BTRFS_BALANCE_OBJECTID;
3432 key.type = BTRFS_BALANCE_ITEM_KEY;
3435 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3438 if (ret > 0) { /* ret = -ENOENT; */
3443 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3449 leaf = path->nodes[0];
3450 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3452 bctl->fs_info = fs_info;
3453 bctl->flags = btrfs_balance_flags(leaf, item);
3454 bctl->flags |= BTRFS_BALANCE_RESUME;
3456 btrfs_balance_data(leaf, item, &disk_bargs);
3457 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3458 btrfs_balance_meta(leaf, item, &disk_bargs);
3459 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3460 btrfs_balance_sys(leaf, item, &disk_bargs);
3461 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3463 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3465 mutex_lock(&fs_info->volume_mutex);
3466 mutex_lock(&fs_info->balance_mutex);
3468 set_balance_control(bctl);
3470 mutex_unlock(&fs_info->balance_mutex);
3471 mutex_unlock(&fs_info->volume_mutex);
3473 btrfs_free_path(path);
3477 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3481 mutex_lock(&fs_info->balance_mutex);
3482 if (!fs_info->balance_ctl) {
3483 mutex_unlock(&fs_info->balance_mutex);
3487 if (atomic_read(&fs_info->balance_running)) {
3488 atomic_inc(&fs_info->balance_pause_req);
3489 mutex_unlock(&fs_info->balance_mutex);
3491 wait_event(fs_info->balance_wait_q,
3492 atomic_read(&fs_info->balance_running) == 0);
3494 mutex_lock(&fs_info->balance_mutex);
3495 /* we are good with balance_ctl ripped off from under us */
3496 BUG_ON(atomic_read(&fs_info->balance_running));
3497 atomic_dec(&fs_info->balance_pause_req);
3502 mutex_unlock(&fs_info->balance_mutex);
3506 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3508 if (fs_info->sb->s_flags & MS_RDONLY)
3511 mutex_lock(&fs_info->balance_mutex);
3512 if (!fs_info->balance_ctl) {
3513 mutex_unlock(&fs_info->balance_mutex);
3517 atomic_inc(&fs_info->balance_cancel_req);
3519 * if we are running just wait and return, balance item is
3520 * deleted in btrfs_balance in this case
3522 if (atomic_read(&fs_info->balance_running)) {
3523 mutex_unlock(&fs_info->balance_mutex);
3524 wait_event(fs_info->balance_wait_q,
3525 atomic_read(&fs_info->balance_running) == 0);
3526 mutex_lock(&fs_info->balance_mutex);
3528 /* __cancel_balance needs volume_mutex */
3529 mutex_unlock(&fs_info->balance_mutex);
3530 mutex_lock(&fs_info->volume_mutex);
3531 mutex_lock(&fs_info->balance_mutex);
3533 if (fs_info->balance_ctl)
3534 __cancel_balance(fs_info);
3536 mutex_unlock(&fs_info->volume_mutex);
3539 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3540 atomic_dec(&fs_info->balance_cancel_req);
3541 mutex_unlock(&fs_info->balance_mutex);
3545 static int btrfs_uuid_scan_kthread(void *data)
3547 struct btrfs_fs_info *fs_info = data;
3548 struct btrfs_root *root = fs_info->tree_root;
3549 struct btrfs_key key;
3550 struct btrfs_key max_key;
3551 struct btrfs_path *path = NULL;
3553 struct extent_buffer *eb;
3555 struct btrfs_root_item root_item;
3557 struct btrfs_trans_handle *trans = NULL;
3559 path = btrfs_alloc_path();
3566 key.type = BTRFS_ROOT_ITEM_KEY;
3569 max_key.objectid = (u64)-1;
3570 max_key.type = BTRFS_ROOT_ITEM_KEY;
3571 max_key.offset = (u64)-1;
3573 path->keep_locks = 1;
3576 ret = btrfs_search_forward(root, &key, path, 0);
3583 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3584 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3585 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3586 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3589 eb = path->nodes[0];
3590 slot = path->slots[0];
3591 item_size = btrfs_item_size_nr(eb, slot);
3592 if (item_size < sizeof(root_item))
3595 read_extent_buffer(eb, &root_item,
3596 btrfs_item_ptr_offset(eb, slot),
3597 (int)sizeof(root_item));
3598 if (btrfs_root_refs(&root_item) == 0)
3601 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3602 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3606 btrfs_release_path(path);
3608 * 1 - subvol uuid item
3609 * 1 - received_subvol uuid item
3611 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3612 if (IS_ERR(trans)) {
3613 ret = PTR_ERR(trans);
3621 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3622 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3624 BTRFS_UUID_KEY_SUBVOL,
3627 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3633 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3634 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3635 root_item.received_uuid,
3636 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3639 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3647 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3653 btrfs_release_path(path);
3654 if (key.offset < (u64)-1) {
3656 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3658 key.type = BTRFS_ROOT_ITEM_KEY;
3659 } else if (key.objectid < (u64)-1) {
3661 key.type = BTRFS_ROOT_ITEM_KEY;
3670 btrfs_free_path(path);
3671 if (trans && !IS_ERR(trans))
3672 btrfs_end_transaction(trans, fs_info->uuid_root);
3674 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3676 fs_info->update_uuid_tree_gen = 1;
3677 up(&fs_info->uuid_tree_rescan_sem);
3682 * Callback for btrfs_uuid_tree_iterate().
3684 * 0 check succeeded, the entry is not outdated.
3685 * < 0 if an error occured.
3686 * > 0 if the check failed, which means the caller shall remove the entry.
3688 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3689 u8 *uuid, u8 type, u64 subid)
3691 struct btrfs_key key;
3693 struct btrfs_root *subvol_root;
3695 if (type != BTRFS_UUID_KEY_SUBVOL &&
3696 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3699 key.objectid = subid;
3700 key.type = BTRFS_ROOT_ITEM_KEY;
3701 key.offset = (u64)-1;
3702 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3703 if (IS_ERR(subvol_root)) {
3704 ret = PTR_ERR(subvol_root);
3711 case BTRFS_UUID_KEY_SUBVOL:
3712 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3715 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3716 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3726 static int btrfs_uuid_rescan_kthread(void *data)
3728 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3732 * 1st step is to iterate through the existing UUID tree and
3733 * to delete all entries that contain outdated data.
3734 * 2nd step is to add all missing entries to the UUID tree.
3736 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3738 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3739 up(&fs_info->uuid_tree_rescan_sem);
3742 return btrfs_uuid_scan_kthread(data);
3745 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3747 struct btrfs_trans_handle *trans;
3748 struct btrfs_root *tree_root = fs_info->tree_root;
3749 struct btrfs_root *uuid_root;
3750 struct task_struct *task;
3757 trans = btrfs_start_transaction(tree_root, 2);
3759 return PTR_ERR(trans);
3761 uuid_root = btrfs_create_tree(trans, fs_info,
3762 BTRFS_UUID_TREE_OBJECTID);
3763 if (IS_ERR(uuid_root)) {
3764 btrfs_abort_transaction(trans, tree_root,
3765 PTR_ERR(uuid_root));
3766 return PTR_ERR(uuid_root);
3769 fs_info->uuid_root = uuid_root;
3771 ret = btrfs_commit_transaction(trans, tree_root);
3775 down(&fs_info->uuid_tree_rescan_sem);
3776 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3778 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3779 btrfs_warn(fs_info, "failed to start uuid_scan task");
3780 up(&fs_info->uuid_tree_rescan_sem);
3781 return PTR_ERR(task);
3787 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3789 struct task_struct *task;
3791 down(&fs_info->uuid_tree_rescan_sem);
3792 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3794 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3795 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3796 up(&fs_info->uuid_tree_rescan_sem);
3797 return PTR_ERR(task);
3804 * shrinking a device means finding all of the device extents past
3805 * the new size, and then following the back refs to the chunks.
3806 * The chunk relocation code actually frees the device extent
3808 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3810 struct btrfs_trans_handle *trans;
3811 struct btrfs_root *root = device->dev_root;
3812 struct btrfs_dev_extent *dev_extent = NULL;
3813 struct btrfs_path *path;
3821 bool retried = false;
3822 struct extent_buffer *l;
3823 struct btrfs_key key;
3824 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3825 u64 old_total = btrfs_super_total_bytes(super_copy);
3826 u64 old_size = device->total_bytes;
3827 u64 diff = device->total_bytes - new_size;
3829 if (device->is_tgtdev_for_dev_replace)
3832 path = btrfs_alloc_path();
3840 device->total_bytes = new_size;
3841 if (device->writeable) {
3842 device->fs_devices->total_rw_bytes -= diff;
3843 spin_lock(&root->fs_info->free_chunk_lock);
3844 root->fs_info->free_chunk_space -= diff;
3845 spin_unlock(&root->fs_info->free_chunk_lock);
3847 unlock_chunks(root);
3850 key.objectid = device->devid;
3851 key.offset = (u64)-1;
3852 key.type = BTRFS_DEV_EXTENT_KEY;
3855 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3859 ret = btrfs_previous_item(root, path, 0, key.type);
3864 btrfs_release_path(path);
3869 slot = path->slots[0];
3870 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3872 if (key.objectid != device->devid) {
3873 btrfs_release_path(path);
3877 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3878 length = btrfs_dev_extent_length(l, dev_extent);
3880 if (key.offset + length <= new_size) {
3881 btrfs_release_path(path);
3885 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3886 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3887 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3888 btrfs_release_path(path);
3890 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3892 if (ret && ret != -ENOSPC)
3896 } while (key.offset-- > 0);
3898 if (failed && !retried) {
3902 } else if (failed && retried) {
3906 device->total_bytes = old_size;
3907 if (device->writeable)
3908 device->fs_devices->total_rw_bytes += diff;
3909 spin_lock(&root->fs_info->free_chunk_lock);
3910 root->fs_info->free_chunk_space += diff;
3911 spin_unlock(&root->fs_info->free_chunk_lock);
3912 unlock_chunks(root);
3916 /* Shrinking succeeded, else we would be at "done". */
3917 trans = btrfs_start_transaction(root, 0);
3918 if (IS_ERR(trans)) {
3919 ret = PTR_ERR(trans);
3925 device->disk_total_bytes = new_size;
3926 /* Now btrfs_update_device() will change the on-disk size. */
3927 ret = btrfs_update_device(trans, device);
3929 unlock_chunks(root);
3930 btrfs_end_transaction(trans, root);
3933 WARN_ON(diff > old_total);
3934 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3935 unlock_chunks(root);
3936 btrfs_end_transaction(trans, root);
3938 btrfs_free_path(path);
3942 static int btrfs_add_system_chunk(struct btrfs_root *root,
3943 struct btrfs_key *key,
3944 struct btrfs_chunk *chunk, int item_size)
3946 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3947 struct btrfs_disk_key disk_key;
3951 array_size = btrfs_super_sys_array_size(super_copy);
3952 if (array_size + item_size + sizeof(disk_key)
3953 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3956 ptr = super_copy->sys_chunk_array + array_size;
3957 btrfs_cpu_key_to_disk(&disk_key, key);
3958 memcpy(ptr, &disk_key, sizeof(disk_key));
3959 ptr += sizeof(disk_key);
3960 memcpy(ptr, chunk, item_size);
3961 item_size += sizeof(disk_key);
3962 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3967 * sort the devices in descending order by max_avail, total_avail
3969 static int btrfs_cmp_device_info(const void *a, const void *b)
3971 const struct btrfs_device_info *di_a = a;
3972 const struct btrfs_device_info *di_b = b;
3974 if (di_a->max_avail > di_b->max_avail)
3976 if (di_a->max_avail < di_b->max_avail)
3978 if (di_a->total_avail > di_b->total_avail)
3980 if (di_a->total_avail < di_b->total_avail)
3985 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3986 [BTRFS_RAID_RAID10] = {
3989 .devs_max = 0, /* 0 == as many as possible */
3991 .devs_increment = 2,
3994 [BTRFS_RAID_RAID1] = {
3999 .devs_increment = 2,
4002 [BTRFS_RAID_DUP] = {
4007 .devs_increment = 1,
4010 [BTRFS_RAID_RAID0] = {
4015 .devs_increment = 1,
4018 [BTRFS_RAID_SINGLE] = {
4023 .devs_increment = 1,
4026 [BTRFS_RAID_RAID5] = {
4031 .devs_increment = 1,
4034 [BTRFS_RAID_RAID6] = {
4039 .devs_increment = 1,
4044 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4046 /* TODO allow them to set a preferred stripe size */
4050 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4052 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4055 btrfs_set_fs_incompat(info, RAID56);
4058 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4059 - sizeof(struct btrfs_item) \
4060 - sizeof(struct btrfs_chunk)) \
4061 / sizeof(struct btrfs_stripe) + 1)
4063 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4064 - 2 * sizeof(struct btrfs_disk_key) \
4065 - 2 * sizeof(struct btrfs_chunk)) \
4066 / sizeof(struct btrfs_stripe) + 1)
4068 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4069 struct btrfs_root *extent_root, u64 start,
4072 struct btrfs_fs_info *info = extent_root->fs_info;
4073 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4074 struct list_head *cur;
4075 struct map_lookup *map = NULL;
4076 struct extent_map_tree *em_tree;
4077 struct extent_map *em;
4078 struct btrfs_device_info *devices_info = NULL;
4080 int num_stripes; /* total number of stripes to allocate */
4081 int data_stripes; /* number of stripes that count for
4083 int sub_stripes; /* sub_stripes info for map */
4084 int dev_stripes; /* stripes per dev */
4085 int devs_max; /* max devs to use */
4086 int devs_min; /* min devs needed */
4087 int devs_increment; /* ndevs has to be a multiple of this */
4088 int ncopies; /* how many copies to data has */
4090 u64 max_stripe_size;
4094 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4100 BUG_ON(!alloc_profile_is_valid(type, 0));
4102 if (list_empty(&fs_devices->alloc_list))
4105 index = __get_raid_index(type);
4107 sub_stripes = btrfs_raid_array[index].sub_stripes;
4108 dev_stripes = btrfs_raid_array[index].dev_stripes;
4109 devs_max = btrfs_raid_array[index].devs_max;
4110 devs_min = btrfs_raid_array[index].devs_min;
4111 devs_increment = btrfs_raid_array[index].devs_increment;
4112 ncopies = btrfs_raid_array[index].ncopies;
4114 if (type & BTRFS_BLOCK_GROUP_DATA) {
4115 max_stripe_size = 1024 * 1024 * 1024;
4116 max_chunk_size = 10 * max_stripe_size;
4118 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4119 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4120 /* for larger filesystems, use larger metadata chunks */
4121 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4122 max_stripe_size = 1024 * 1024 * 1024;
4124 max_stripe_size = 256 * 1024 * 1024;
4125 max_chunk_size = max_stripe_size;
4127 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4128 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4129 max_stripe_size = 32 * 1024 * 1024;
4130 max_chunk_size = 2 * max_stripe_size;
4132 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4134 btrfs_err(info, "invalid chunk type 0x%llx requested",
4139 /* we don't want a chunk larger than 10% of writeable space */
4140 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4143 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4148 cur = fs_devices->alloc_list.next;
4151 * in the first pass through the devices list, we gather information
4152 * about the available holes on each device.
4155 while (cur != &fs_devices->alloc_list) {
4156 struct btrfs_device *device;
4160 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4164 if (!device->writeable) {
4166 "BTRFS: read-only device in alloc_list\n");
4170 if (!device->in_fs_metadata ||
4171 device->is_tgtdev_for_dev_replace)
4174 if (device->total_bytes > device->bytes_used)
4175 total_avail = device->total_bytes - device->bytes_used;
4179 /* If there is no space on this device, skip it. */
4180 if (total_avail == 0)
4183 ret = find_free_dev_extent(trans, device,
4184 max_stripe_size * dev_stripes,
4185 &dev_offset, &max_avail);
4186 if (ret && ret != -ENOSPC)
4190 max_avail = max_stripe_size * dev_stripes;
4192 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4195 if (ndevs == fs_devices->rw_devices) {
4196 WARN(1, "%s: found more than %llu devices\n",
4197 __func__, fs_devices->rw_devices);
4200 devices_info[ndevs].dev_offset = dev_offset;
4201 devices_info[ndevs].max_avail = max_avail;
4202 devices_info[ndevs].total_avail = total_avail;
4203 devices_info[ndevs].dev = device;
4208 * now sort the devices by hole size / available space
4210 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4211 btrfs_cmp_device_info, NULL);
4213 /* round down to number of usable stripes */
4214 ndevs -= ndevs % devs_increment;
4216 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4221 if (devs_max && ndevs > devs_max)
4224 * the primary goal is to maximize the number of stripes, so use as many
4225 * devices as possible, even if the stripes are not maximum sized.
4227 stripe_size = devices_info[ndevs-1].max_avail;
4228 num_stripes = ndevs * dev_stripes;
4231 * this will have to be fixed for RAID1 and RAID10 over
4234 data_stripes = num_stripes / ncopies;
4236 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4237 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4238 btrfs_super_stripesize(info->super_copy));
4239 data_stripes = num_stripes - 1;
4241 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4242 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4243 btrfs_super_stripesize(info->super_copy));
4244 data_stripes = num_stripes - 2;
4248 * Use the number of data stripes to figure out how big this chunk
4249 * is really going to be in terms of logical address space,
4250 * and compare that answer with the max chunk size
4252 if (stripe_size * data_stripes > max_chunk_size) {
4253 u64 mask = (1ULL << 24) - 1;
4254 stripe_size = max_chunk_size;
4255 do_div(stripe_size, data_stripes);
4257 /* bump the answer up to a 16MB boundary */
4258 stripe_size = (stripe_size + mask) & ~mask;
4260 /* but don't go higher than the limits we found
4261 * while searching for free extents
4263 if (stripe_size > devices_info[ndevs-1].max_avail)
4264 stripe_size = devices_info[ndevs-1].max_avail;
4267 do_div(stripe_size, dev_stripes);
4269 /* align to BTRFS_STRIPE_LEN */
4270 do_div(stripe_size, raid_stripe_len);
4271 stripe_size *= raid_stripe_len;
4273 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4278 map->num_stripes = num_stripes;
4280 for (i = 0; i < ndevs; ++i) {
4281 for (j = 0; j < dev_stripes; ++j) {
4282 int s = i * dev_stripes + j;
4283 map->stripes[s].dev = devices_info[i].dev;
4284 map->stripes[s].physical = devices_info[i].dev_offset +
4288 map->sector_size = extent_root->sectorsize;
4289 map->stripe_len = raid_stripe_len;
4290 map->io_align = raid_stripe_len;
4291 map->io_width = raid_stripe_len;
4293 map->sub_stripes = sub_stripes;
4295 num_bytes = stripe_size * data_stripes;
4297 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4299 em = alloc_extent_map();
4305 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4306 em->bdev = (struct block_device *)map;
4308 em->len = num_bytes;
4309 em->block_start = 0;
4310 em->block_len = em->len;
4311 em->orig_block_len = stripe_size;
4313 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4314 write_lock(&em_tree->lock);
4315 ret = add_extent_mapping(em_tree, em, 0);
4317 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4318 atomic_inc(&em->refs);
4320 write_unlock(&em_tree->lock);
4322 free_extent_map(em);
4326 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4327 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4330 goto error_del_extent;
4332 free_extent_map(em);
4333 check_raid56_incompat_flag(extent_root->fs_info, type);
4335 kfree(devices_info);
4339 write_lock(&em_tree->lock);
4340 remove_extent_mapping(em_tree, em);
4341 write_unlock(&em_tree->lock);
4343 /* One for our allocation */
4344 free_extent_map(em);
4345 /* One for the tree reference */
4346 free_extent_map(em);
4348 kfree(devices_info);
4352 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4353 struct btrfs_root *extent_root,
4354 u64 chunk_offset, u64 chunk_size)
4356 struct btrfs_key key;
4357 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4358 struct btrfs_device *device;
4359 struct btrfs_chunk *chunk;
4360 struct btrfs_stripe *stripe;
4361 struct extent_map_tree *em_tree;
4362 struct extent_map *em;
4363 struct map_lookup *map;
4370 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4371 read_lock(&em_tree->lock);
4372 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4373 read_unlock(&em_tree->lock);
4376 btrfs_crit(extent_root->fs_info, "unable to find logical "
4377 "%Lu len %Lu", chunk_offset, chunk_size);
4381 if (em->start != chunk_offset || em->len != chunk_size) {
4382 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4383 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4384 chunk_size, em->start, em->len);
4385 free_extent_map(em);
4389 map = (struct map_lookup *)em->bdev;
4390 item_size = btrfs_chunk_item_size(map->num_stripes);
4391 stripe_size = em->orig_block_len;
4393 chunk = kzalloc(item_size, GFP_NOFS);
4399 for (i = 0; i < map->num_stripes; i++) {
4400 device = map->stripes[i].dev;
4401 dev_offset = map->stripes[i].physical;
4403 device->bytes_used += stripe_size;
4404 ret = btrfs_update_device(trans, device);
4407 ret = btrfs_alloc_dev_extent(trans, device,
4408 chunk_root->root_key.objectid,
4409 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4410 chunk_offset, dev_offset,
4416 spin_lock(&extent_root->fs_info->free_chunk_lock);
4417 extent_root->fs_info->free_chunk_space -= (stripe_size *
4419 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4421 stripe = &chunk->stripe;
4422 for (i = 0; i < map->num_stripes; i++) {
4423 device = map->stripes[i].dev;
4424 dev_offset = map->stripes[i].physical;
4426 btrfs_set_stack_stripe_devid(stripe, device->devid);
4427 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4428 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4432 btrfs_set_stack_chunk_length(chunk, chunk_size);
4433 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4434 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4435 btrfs_set_stack_chunk_type(chunk, map->type);
4436 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4437 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4438 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4439 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4440 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4442 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4443 key.type = BTRFS_CHUNK_ITEM_KEY;
4444 key.offset = chunk_offset;
4446 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4447 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4449 * TODO: Cleanup of inserted chunk root in case of
4452 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4458 free_extent_map(em);
4463 * Chunk allocation falls into two parts. The first part does works
4464 * that make the new allocated chunk useable, but not do any operation
4465 * that modifies the chunk tree. The second part does the works that
4466 * require modifying the chunk tree. This division is important for the
4467 * bootstrap process of adding storage to a seed btrfs.
4469 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4470 struct btrfs_root *extent_root, u64 type)
4474 chunk_offset = find_next_chunk(extent_root->fs_info);
4475 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4478 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4479 struct btrfs_root *root,
4480 struct btrfs_device *device)
4483 u64 sys_chunk_offset;
4485 struct btrfs_fs_info *fs_info = root->fs_info;
4486 struct btrfs_root *extent_root = fs_info->extent_root;
4489 chunk_offset = find_next_chunk(fs_info);
4490 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4491 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4496 sys_chunk_offset = find_next_chunk(root->fs_info);
4497 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4498 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4501 btrfs_abort_transaction(trans, root, ret);
4505 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4507 btrfs_abort_transaction(trans, root, ret);
4512 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4514 struct extent_map *em;
4515 struct map_lookup *map;
4516 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4520 read_lock(&map_tree->map_tree.lock);
4521 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4522 read_unlock(&map_tree->map_tree.lock);
4526 if (btrfs_test_opt(root, DEGRADED)) {
4527 free_extent_map(em);
4531 map = (struct map_lookup *)em->bdev;
4532 for (i = 0; i < map->num_stripes; i++) {
4533 if (!map->stripes[i].dev->writeable) {
4538 free_extent_map(em);
4542 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4544 extent_map_tree_init(&tree->map_tree);
4547 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4549 struct extent_map *em;
4552 write_lock(&tree->map_tree.lock);
4553 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4555 remove_extent_mapping(&tree->map_tree, em);
4556 write_unlock(&tree->map_tree.lock);
4560 free_extent_map(em);
4561 /* once for the tree */
4562 free_extent_map(em);
4566 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4568 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4569 struct extent_map *em;
4570 struct map_lookup *map;
4571 struct extent_map_tree *em_tree = &map_tree->map_tree;
4574 read_lock(&em_tree->lock);
4575 em = lookup_extent_mapping(em_tree, logical, len);
4576 read_unlock(&em_tree->lock);
4579 * We could return errors for these cases, but that could get ugly and
4580 * we'd probably do the same thing which is just not do anything else
4581 * and exit, so return 1 so the callers don't try to use other copies.
4584 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4589 if (em->start > logical || em->start + em->len < logical) {
4590 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4591 "%Lu-%Lu", logical, logical+len, em->start,
4592 em->start + em->len);
4593 free_extent_map(em);
4597 map = (struct map_lookup *)em->bdev;
4598 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4599 ret = map->num_stripes;
4600 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4601 ret = map->sub_stripes;
4602 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4604 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4608 free_extent_map(em);
4610 btrfs_dev_replace_lock(&fs_info->dev_replace);
4611 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4613 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4618 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4619 struct btrfs_mapping_tree *map_tree,
4622 struct extent_map *em;
4623 struct map_lookup *map;
4624 struct extent_map_tree *em_tree = &map_tree->map_tree;
4625 unsigned long len = root->sectorsize;
4627 read_lock(&em_tree->lock);
4628 em = lookup_extent_mapping(em_tree, logical, len);
4629 read_unlock(&em_tree->lock);
4632 BUG_ON(em->start > logical || em->start + em->len < logical);
4633 map = (struct map_lookup *)em->bdev;
4634 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4635 BTRFS_BLOCK_GROUP_RAID6)) {
4636 len = map->stripe_len * nr_data_stripes(map);
4638 free_extent_map(em);
4642 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4643 u64 logical, u64 len, int mirror_num)
4645 struct extent_map *em;
4646 struct map_lookup *map;
4647 struct extent_map_tree *em_tree = &map_tree->map_tree;
4650 read_lock(&em_tree->lock);
4651 em = lookup_extent_mapping(em_tree, logical, len);
4652 read_unlock(&em_tree->lock);
4655 BUG_ON(em->start > logical || em->start + em->len < logical);
4656 map = (struct map_lookup *)em->bdev;
4657 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4658 BTRFS_BLOCK_GROUP_RAID6))
4660 free_extent_map(em);
4664 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4665 struct map_lookup *map, int first, int num,
4666 int optimal, int dev_replace_is_ongoing)
4670 struct btrfs_device *srcdev;
4672 if (dev_replace_is_ongoing &&
4673 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4674 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4675 srcdev = fs_info->dev_replace.srcdev;
4680 * try to avoid the drive that is the source drive for a
4681 * dev-replace procedure, only choose it if no other non-missing
4682 * mirror is available
4684 for (tolerance = 0; tolerance < 2; tolerance++) {
4685 if (map->stripes[optimal].dev->bdev &&
4686 (tolerance || map->stripes[optimal].dev != srcdev))
4688 for (i = first; i < first + num; i++) {
4689 if (map->stripes[i].dev->bdev &&
4690 (tolerance || map->stripes[i].dev != srcdev))
4695 /* we couldn't find one that doesn't fail. Just return something
4696 * and the io error handling code will clean up eventually
4701 static inline int parity_smaller(u64 a, u64 b)
4706 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4707 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4709 struct btrfs_bio_stripe s;
4716 for (i = 0; i < bbio->num_stripes - 1; i++) {
4717 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4718 s = bbio->stripes[i];
4720 bbio->stripes[i] = bbio->stripes[i+1];
4721 raid_map[i] = raid_map[i+1];
4722 bbio->stripes[i+1] = s;
4730 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4731 u64 logical, u64 *length,
4732 struct btrfs_bio **bbio_ret,
4733 int mirror_num, u64 **raid_map_ret)
4735 struct extent_map *em;
4736 struct map_lookup *map;
4737 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4738 struct extent_map_tree *em_tree = &map_tree->map_tree;
4741 u64 stripe_end_offset;
4746 u64 *raid_map = NULL;
4752 struct btrfs_bio *bbio = NULL;
4753 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4754 int dev_replace_is_ongoing = 0;
4755 int num_alloc_stripes;
4756 int patch_the_first_stripe_for_dev_replace = 0;
4757 u64 physical_to_patch_in_first_stripe = 0;
4758 u64 raid56_full_stripe_start = (u64)-1;
4760 read_lock(&em_tree->lock);
4761 em = lookup_extent_mapping(em_tree, logical, *length);
4762 read_unlock(&em_tree->lock);
4765 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4770 if (em->start > logical || em->start + em->len < logical) {
4771 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4772 "found %Lu-%Lu", logical, em->start,
4773 em->start + em->len);
4774 free_extent_map(em);
4778 map = (struct map_lookup *)em->bdev;
4779 offset = logical - em->start;
4781 stripe_len = map->stripe_len;
4784 * stripe_nr counts the total number of stripes we have to stride
4785 * to get to this block
4787 do_div(stripe_nr, stripe_len);
4789 stripe_offset = stripe_nr * stripe_len;
4790 BUG_ON(offset < stripe_offset);
4792 /* stripe_offset is the offset of this block in its stripe*/
4793 stripe_offset = offset - stripe_offset;
4795 /* if we're here for raid56, we need to know the stripe aligned start */
4796 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4797 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4798 raid56_full_stripe_start = offset;
4800 /* allow a write of a full stripe, but make sure we don't
4801 * allow straddling of stripes
4803 do_div(raid56_full_stripe_start, full_stripe_len);
4804 raid56_full_stripe_start *= full_stripe_len;
4807 if (rw & REQ_DISCARD) {
4808 /* we don't discard raid56 yet */
4810 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4814 *length = min_t(u64, em->len - offset, *length);
4815 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4817 /* For writes to RAID[56], allow a full stripeset across all disks.
4818 For other RAID types and for RAID[56] reads, just allow a single
4819 stripe (on a single disk). */
4820 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4822 max_len = stripe_len * nr_data_stripes(map) -
4823 (offset - raid56_full_stripe_start);
4825 /* we limit the length of each bio to what fits in a stripe */
4826 max_len = stripe_len - stripe_offset;
4828 *length = min_t(u64, em->len - offset, max_len);
4830 *length = em->len - offset;
4833 /* This is for when we're called from btrfs_merge_bio_hook() and all
4834 it cares about is the length */
4838 btrfs_dev_replace_lock(dev_replace);
4839 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4840 if (!dev_replace_is_ongoing)
4841 btrfs_dev_replace_unlock(dev_replace);
4843 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4844 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4845 dev_replace->tgtdev != NULL) {
4847 * in dev-replace case, for repair case (that's the only
4848 * case where the mirror is selected explicitly when
4849 * calling btrfs_map_block), blocks left of the left cursor
4850 * can also be read from the target drive.
4851 * For REQ_GET_READ_MIRRORS, the target drive is added as
4852 * the last one to the array of stripes. For READ, it also
4853 * needs to be supported using the same mirror number.
4854 * If the requested block is not left of the left cursor,
4855 * EIO is returned. This can happen because btrfs_num_copies()
4856 * returns one more in the dev-replace case.
4858 u64 tmp_length = *length;
4859 struct btrfs_bio *tmp_bbio = NULL;
4860 int tmp_num_stripes;
4861 u64 srcdev_devid = dev_replace->srcdev->devid;
4862 int index_srcdev = 0;
4864 u64 physical_of_found = 0;
4866 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4867 logical, &tmp_length, &tmp_bbio, 0, NULL);
4869 WARN_ON(tmp_bbio != NULL);
4873 tmp_num_stripes = tmp_bbio->num_stripes;
4874 if (mirror_num > tmp_num_stripes) {
4876 * REQ_GET_READ_MIRRORS does not contain this
4877 * mirror, that means that the requested area
4878 * is not left of the left cursor
4886 * process the rest of the function using the mirror_num
4887 * of the source drive. Therefore look it up first.
4888 * At the end, patch the device pointer to the one of the
4891 for (i = 0; i < tmp_num_stripes; i++) {
4892 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4894 * In case of DUP, in order to keep it
4895 * simple, only add the mirror with the
4896 * lowest physical address
4899 physical_of_found <=
4900 tmp_bbio->stripes[i].physical)
4905 tmp_bbio->stripes[i].physical;
4910 mirror_num = index_srcdev + 1;
4911 patch_the_first_stripe_for_dev_replace = 1;
4912 physical_to_patch_in_first_stripe = physical_of_found;
4921 } else if (mirror_num > map->num_stripes) {
4927 stripe_nr_orig = stripe_nr;
4928 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4929 do_div(stripe_nr_end, map->stripe_len);
4930 stripe_end_offset = stripe_nr_end * map->stripe_len -
4933 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4934 if (rw & REQ_DISCARD)
4935 num_stripes = min_t(u64, map->num_stripes,
4936 stripe_nr_end - stripe_nr_orig);
4937 stripe_index = do_div(stripe_nr, map->num_stripes);
4938 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4939 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4940 num_stripes = map->num_stripes;
4941 else if (mirror_num)
4942 stripe_index = mirror_num - 1;
4944 stripe_index = find_live_mirror(fs_info, map, 0,
4946 current->pid % map->num_stripes,
4947 dev_replace_is_ongoing);
4948 mirror_num = stripe_index + 1;
4951 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4952 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4953 num_stripes = map->num_stripes;
4954 } else if (mirror_num) {
4955 stripe_index = mirror_num - 1;
4960 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4961 int factor = map->num_stripes / map->sub_stripes;
4963 stripe_index = do_div(stripe_nr, factor);
4964 stripe_index *= map->sub_stripes;
4966 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4967 num_stripes = map->sub_stripes;
4968 else if (rw & REQ_DISCARD)
4969 num_stripes = min_t(u64, map->sub_stripes *
4970 (stripe_nr_end - stripe_nr_orig),
4972 else if (mirror_num)
4973 stripe_index += mirror_num - 1;
4975 int old_stripe_index = stripe_index;
4976 stripe_index = find_live_mirror(fs_info, map,
4978 map->sub_stripes, stripe_index +
4979 current->pid % map->sub_stripes,
4980 dev_replace_is_ongoing);
4981 mirror_num = stripe_index - old_stripe_index + 1;
4984 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4985 BTRFS_BLOCK_GROUP_RAID6)) {
4988 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4992 /* push stripe_nr back to the start of the full stripe */
4993 stripe_nr = raid56_full_stripe_start;
4994 do_div(stripe_nr, stripe_len);
4996 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4998 /* RAID[56] write or recovery. Return all stripes */
4999 num_stripes = map->num_stripes;
5000 max_errors = nr_parity_stripes(map);
5002 raid_map = kmalloc_array(num_stripes, sizeof(u64),
5009 /* Work out the disk rotation on this stripe-set */
5011 rot = do_div(tmp, num_stripes);
5013 /* Fill in the logical address of each stripe */
5014 tmp = stripe_nr * nr_data_stripes(map);
5015 for (i = 0; i < nr_data_stripes(map); i++)
5016 raid_map[(i+rot) % num_stripes] =
5017 em->start + (tmp + i) * map->stripe_len;
5019 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5020 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5021 raid_map[(i+rot+1) % num_stripes] =
5024 *length = map->stripe_len;
5029 * Mirror #0 or #1 means the original data block.
5030 * Mirror #2 is RAID5 parity block.
5031 * Mirror #3 is RAID6 Q block.
5033 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5035 stripe_index = nr_data_stripes(map) +
5038 /* We distribute the parity blocks across stripes */
5039 tmp = stripe_nr + stripe_index;
5040 stripe_index = do_div(tmp, map->num_stripes);
5044 * after this do_div call, stripe_nr is the number of stripes
5045 * on this device we have to walk to find the data, and
5046 * stripe_index is the number of our device in the stripe array
5048 stripe_index = do_div(stripe_nr, map->num_stripes);
5049 mirror_num = stripe_index + 1;
5051 BUG_ON(stripe_index >= map->num_stripes);
5053 num_alloc_stripes = num_stripes;
5054 if (dev_replace_is_ongoing) {
5055 if (rw & (REQ_WRITE | REQ_DISCARD))
5056 num_alloc_stripes <<= 1;
5057 if (rw & REQ_GET_READ_MIRRORS)
5058 num_alloc_stripes++;
5060 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5066 atomic_set(&bbio->error, 0);
5068 if (rw & REQ_DISCARD) {
5070 int sub_stripes = 0;
5071 u64 stripes_per_dev = 0;
5072 u32 remaining_stripes = 0;
5073 u32 last_stripe = 0;
5076 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5077 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5080 sub_stripes = map->sub_stripes;
5082 factor = map->num_stripes / sub_stripes;
5083 stripes_per_dev = div_u64_rem(stripe_nr_end -
5086 &remaining_stripes);
5087 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5088 last_stripe *= sub_stripes;
5091 for (i = 0; i < num_stripes; i++) {
5092 bbio->stripes[i].physical =
5093 map->stripes[stripe_index].physical +
5094 stripe_offset + stripe_nr * map->stripe_len;
5095 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5097 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5098 BTRFS_BLOCK_GROUP_RAID10)) {
5099 bbio->stripes[i].length = stripes_per_dev *
5102 if (i / sub_stripes < remaining_stripes)
5103 bbio->stripes[i].length +=
5107 * Special for the first stripe and
5110 * |-------|...|-------|
5114 if (i < sub_stripes)
5115 bbio->stripes[i].length -=
5118 if (stripe_index >= last_stripe &&
5119 stripe_index <= (last_stripe +
5121 bbio->stripes[i].length -=
5124 if (i == sub_stripes - 1)
5127 bbio->stripes[i].length = *length;
5130 if (stripe_index == map->num_stripes) {
5131 /* This could only happen for RAID0/10 */
5137 for (i = 0; i < num_stripes; i++) {
5138 bbio->stripes[i].physical =
5139 map->stripes[stripe_index].physical +
5141 stripe_nr * map->stripe_len;
5142 bbio->stripes[i].dev =
5143 map->stripes[stripe_index].dev;
5148 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5149 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5150 BTRFS_BLOCK_GROUP_RAID10 |
5151 BTRFS_BLOCK_GROUP_RAID5 |
5152 BTRFS_BLOCK_GROUP_DUP)) {
5154 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5159 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5160 dev_replace->tgtdev != NULL) {
5161 int index_where_to_add;
5162 u64 srcdev_devid = dev_replace->srcdev->devid;
5165 * duplicate the write operations while the dev replace
5166 * procedure is running. Since the copying of the old disk
5167 * to the new disk takes place at run time while the
5168 * filesystem is mounted writable, the regular write
5169 * operations to the old disk have to be duplicated to go
5170 * to the new disk as well.
5171 * Note that device->missing is handled by the caller, and
5172 * that the write to the old disk is already set up in the
5175 index_where_to_add = num_stripes;
5176 for (i = 0; i < num_stripes; i++) {
5177 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5178 /* write to new disk, too */
5179 struct btrfs_bio_stripe *new =
5180 bbio->stripes + index_where_to_add;
5181 struct btrfs_bio_stripe *old =
5184 new->physical = old->physical;
5185 new->length = old->length;
5186 new->dev = dev_replace->tgtdev;
5187 index_where_to_add++;
5191 num_stripes = index_where_to_add;
5192 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5193 dev_replace->tgtdev != NULL) {
5194 u64 srcdev_devid = dev_replace->srcdev->devid;
5195 int index_srcdev = 0;
5197 u64 physical_of_found = 0;
5200 * During the dev-replace procedure, the target drive can
5201 * also be used to read data in case it is needed to repair
5202 * a corrupt block elsewhere. This is possible if the
5203 * requested area is left of the left cursor. In this area,
5204 * the target drive is a full copy of the source drive.
5206 for (i = 0; i < num_stripes; i++) {
5207 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5209 * In case of DUP, in order to keep it
5210 * simple, only add the mirror with the
5211 * lowest physical address
5214 physical_of_found <=
5215 bbio->stripes[i].physical)
5219 physical_of_found = bbio->stripes[i].physical;
5223 u64 length = map->stripe_len;
5225 if (physical_of_found + length <=
5226 dev_replace->cursor_left) {
5227 struct btrfs_bio_stripe *tgtdev_stripe =
5228 bbio->stripes + num_stripes;
5230 tgtdev_stripe->physical = physical_of_found;
5231 tgtdev_stripe->length =
5232 bbio->stripes[index_srcdev].length;
5233 tgtdev_stripe->dev = dev_replace->tgtdev;
5241 bbio->num_stripes = num_stripes;
5242 bbio->max_errors = max_errors;
5243 bbio->mirror_num = mirror_num;
5246 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5247 * mirror_num == num_stripes + 1 && dev_replace target drive is
5248 * available as a mirror
5250 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5251 WARN_ON(num_stripes > 1);
5252 bbio->stripes[0].dev = dev_replace->tgtdev;
5253 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5254 bbio->mirror_num = map->num_stripes + 1;
5257 sort_parity_stripes(bbio, raid_map);
5258 *raid_map_ret = raid_map;
5261 if (dev_replace_is_ongoing)
5262 btrfs_dev_replace_unlock(dev_replace);
5263 free_extent_map(em);
5267 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5268 u64 logical, u64 *length,
5269 struct btrfs_bio **bbio_ret, int mirror_num)
5271 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5275 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5276 u64 chunk_start, u64 physical, u64 devid,
5277 u64 **logical, int *naddrs, int *stripe_len)
5279 struct extent_map_tree *em_tree = &map_tree->map_tree;
5280 struct extent_map *em;
5281 struct map_lookup *map;
5289 read_lock(&em_tree->lock);
5290 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5291 read_unlock(&em_tree->lock);
5294 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5299 if (em->start != chunk_start) {
5300 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5301 em->start, chunk_start);
5302 free_extent_map(em);
5305 map = (struct map_lookup *)em->bdev;
5308 rmap_len = map->stripe_len;
5310 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5311 do_div(length, map->num_stripes / map->sub_stripes);
5312 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5313 do_div(length, map->num_stripes);
5314 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5315 BTRFS_BLOCK_GROUP_RAID6)) {
5316 do_div(length, nr_data_stripes(map));
5317 rmap_len = map->stripe_len * nr_data_stripes(map);
5320 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5321 BUG_ON(!buf); /* -ENOMEM */
5323 for (i = 0; i < map->num_stripes; i++) {
5324 if (devid && map->stripes[i].dev->devid != devid)
5326 if (map->stripes[i].physical > physical ||
5327 map->stripes[i].physical + length <= physical)
5330 stripe_nr = physical - map->stripes[i].physical;
5331 do_div(stripe_nr, map->stripe_len);
5333 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5334 stripe_nr = stripe_nr * map->num_stripes + i;
5335 do_div(stripe_nr, map->sub_stripes);
5336 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5337 stripe_nr = stripe_nr * map->num_stripes + i;
5338 } /* else if RAID[56], multiply by nr_data_stripes().
5339 * Alternatively, just use rmap_len below instead of
5340 * map->stripe_len */
5342 bytenr = chunk_start + stripe_nr * rmap_len;
5343 WARN_ON(nr >= map->num_stripes);
5344 for (j = 0; j < nr; j++) {
5345 if (buf[j] == bytenr)
5349 WARN_ON(nr >= map->num_stripes);
5356 *stripe_len = rmap_len;
5358 free_extent_map(em);
5362 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5364 if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5365 bio_endio_nodec(bio, err);
5367 bio_endio(bio, err);
5371 static void btrfs_end_bio(struct bio *bio, int err)
5373 struct btrfs_bio *bbio = bio->bi_private;
5374 struct btrfs_device *dev = bbio->stripes[0].dev;
5375 int is_orig_bio = 0;
5378 atomic_inc(&bbio->error);
5379 if (err == -EIO || err == -EREMOTEIO) {
5380 unsigned int stripe_index =
5381 btrfs_io_bio(bio)->stripe_index;
5383 BUG_ON(stripe_index >= bbio->num_stripes);
5384 dev = bbio->stripes[stripe_index].dev;
5386 if (bio->bi_rw & WRITE)
5387 btrfs_dev_stat_inc(dev,
5388 BTRFS_DEV_STAT_WRITE_ERRS);
5390 btrfs_dev_stat_inc(dev,
5391 BTRFS_DEV_STAT_READ_ERRS);
5392 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5393 btrfs_dev_stat_inc(dev,
5394 BTRFS_DEV_STAT_FLUSH_ERRS);
5395 btrfs_dev_stat_print_on_error(dev);
5400 if (bio == bbio->orig_bio)
5403 btrfs_bio_counter_dec(bbio->fs_info);
5405 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5408 bio = bbio->orig_bio;
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);
5428 btrfs_end_bbio(bbio, bio, err);
5429 } else if (!is_orig_bio) {
5435 * see run_scheduled_bios for a description of why bios are collected for
5438 * This will add one bio to the pending list for a device and make sure
5439 * the work struct is scheduled.
5441 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5442 struct btrfs_device *device,
5443 int rw, struct bio *bio)
5445 int should_queue = 1;
5446 struct btrfs_pending_bios *pending_bios;
5448 if (device->missing || !device->bdev) {
5449 bio_endio(bio, -EIO);
5453 /* don't bother with additional async steps for reads, right now */
5454 if (!(rw & REQ_WRITE)) {
5456 btrfsic_submit_bio(rw, bio);
5462 * nr_async_bios allows us to reliably return congestion to the
5463 * higher layers. Otherwise, the async bio makes it appear we have
5464 * made progress against dirty pages when we've really just put it
5465 * on a queue for later
5467 atomic_inc(&root->fs_info->nr_async_bios);
5468 WARN_ON(bio->bi_next);
5469 bio->bi_next = NULL;
5472 spin_lock(&device->io_lock);
5473 if (bio->bi_rw & REQ_SYNC)
5474 pending_bios = &device->pending_sync_bios;
5476 pending_bios = &device->pending_bios;
5478 if (pending_bios->tail)
5479 pending_bios->tail->bi_next = bio;
5481 pending_bios->tail = bio;
5482 if (!pending_bios->head)
5483 pending_bios->head = bio;
5484 if (device->running_pending)
5487 spin_unlock(&device->io_lock);
5490 btrfs_queue_work(root->fs_info->submit_workers,
5494 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5497 struct bio_vec *prev;
5498 struct request_queue *q = bdev_get_queue(bdev);
5499 unsigned int max_sectors = queue_max_sectors(q);
5500 struct bvec_merge_data bvm = {
5502 .bi_sector = sector,
5503 .bi_rw = bio->bi_rw,
5506 if (WARN_ON(bio->bi_vcnt == 0))
5509 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5510 if (bio_sectors(bio) > max_sectors)
5513 if (!q->merge_bvec_fn)
5516 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5517 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5522 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5523 struct bio *bio, u64 physical, int dev_nr,
5526 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5528 bio->bi_private = bbio;
5529 btrfs_io_bio(bio)->stripe_index = dev_nr;
5530 bio->bi_end_io = btrfs_end_bio;
5531 bio->bi_iter.bi_sector = physical >> 9;
5534 struct rcu_string *name;
5537 name = rcu_dereference(dev->name);
5538 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5539 "(%s id %llu), size=%u\n", rw,
5540 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5541 name->str, dev->devid, bio->bi_size);
5545 bio->bi_bdev = dev->bdev;
5547 btrfs_bio_counter_inc_noblocked(root->fs_info);
5550 btrfs_schedule_bio(root, dev, rw, bio);
5552 btrfsic_submit_bio(rw, bio);
5555 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5556 struct bio *first_bio, struct btrfs_device *dev,
5557 int dev_nr, int rw, int async)
5559 struct bio_vec *bvec = first_bio->bi_io_vec;
5561 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5562 u64 physical = bbio->stripes[dev_nr].physical;
5565 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5569 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5570 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5571 bvec->bv_offset) < bvec->bv_len) {
5572 u64 len = bio->bi_iter.bi_size;
5574 atomic_inc(&bbio->stripes_pending);
5575 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5583 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5587 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5589 atomic_inc(&bbio->error);
5590 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5591 /* Shoud be the original bio. */
5592 WARN_ON(bio != bbio->orig_bio);
5594 bio->bi_private = bbio->private;
5595 bio->bi_end_io = bbio->end_io;
5596 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5597 bio->bi_iter.bi_sector = logical >> 9;
5599 btrfs_end_bbio(bbio, bio, -EIO);
5603 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5604 int mirror_num, int async_submit)
5606 struct btrfs_device *dev;
5607 struct bio *first_bio = bio;
5608 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5611 u64 *raid_map = NULL;
5615 struct btrfs_bio *bbio = NULL;
5617 length = bio->bi_iter.bi_size;
5618 map_length = length;
5620 btrfs_bio_counter_inc_blocked(root->fs_info);
5621 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5622 mirror_num, &raid_map);
5624 btrfs_bio_counter_dec(root->fs_info);
5628 total_devs = bbio->num_stripes;
5629 bbio->orig_bio = first_bio;
5630 bbio->private = first_bio->bi_private;
5631 bbio->end_io = first_bio->bi_end_io;
5632 bbio->fs_info = root->fs_info;
5633 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5636 /* In this case, map_length has been set to the length of
5637 a single stripe; not the whole write */
5639 ret = raid56_parity_write(root, bio, bbio,
5640 raid_map, map_length);
5642 ret = raid56_parity_recover(root, bio, bbio,
5643 raid_map, map_length,
5647 * FIXME, replace dosen't support raid56 yet, please fix
5650 btrfs_bio_counter_dec(root->fs_info);
5654 if (map_length < length) {
5655 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5656 logical, length, map_length);
5660 while (dev_nr < total_devs) {
5661 dev = bbio->stripes[dev_nr].dev;
5662 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5663 bbio_error(bbio, first_bio, logical);
5669 * Check and see if we're ok with this bio based on it's size
5670 * and offset with the given device.
5672 if (!bio_size_ok(dev->bdev, first_bio,
5673 bbio->stripes[dev_nr].physical >> 9)) {
5674 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5675 dev_nr, rw, async_submit);
5681 if (dev_nr < total_devs - 1) {
5682 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5683 BUG_ON(!bio); /* -ENOMEM */
5686 bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5689 submit_stripe_bio(root, bbio, bio,
5690 bbio->stripes[dev_nr].physical, dev_nr, rw,
5694 btrfs_bio_counter_dec(root->fs_info);
5698 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5701 struct btrfs_device *device;
5702 struct btrfs_fs_devices *cur_devices;
5704 cur_devices = fs_info->fs_devices;
5705 while (cur_devices) {
5707 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5708 device = __find_device(&cur_devices->devices,
5713 cur_devices = cur_devices->seed;
5718 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5719 u64 devid, u8 *dev_uuid)
5721 struct btrfs_device *device;
5722 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5724 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5728 list_add(&device->dev_list, &fs_devices->devices);
5729 device->fs_devices = fs_devices;
5730 fs_devices->num_devices++;
5732 device->missing = 1;
5733 fs_devices->missing_devices++;
5739 * btrfs_alloc_device - allocate struct btrfs_device
5740 * @fs_info: used only for generating a new devid, can be NULL if
5741 * devid is provided (i.e. @devid != NULL).
5742 * @devid: a pointer to devid for this device. If NULL a new devid
5744 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5747 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5748 * on error. Returned struct is not linked onto any lists and can be
5749 * destroyed with kfree() right away.
5751 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5755 struct btrfs_device *dev;
5758 if (WARN_ON(!devid && !fs_info))
5759 return ERR_PTR(-EINVAL);
5761 dev = __alloc_device();
5770 ret = find_next_devid(fs_info, &tmp);
5773 return ERR_PTR(ret);
5779 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5781 generate_random_uuid(dev->uuid);
5783 btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5788 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5789 struct extent_buffer *leaf,
5790 struct btrfs_chunk *chunk)
5792 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5793 struct map_lookup *map;
5794 struct extent_map *em;
5798 u8 uuid[BTRFS_UUID_SIZE];
5803 logical = key->offset;
5804 length = btrfs_chunk_length(leaf, chunk);
5806 read_lock(&map_tree->map_tree.lock);
5807 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5808 read_unlock(&map_tree->map_tree.lock);
5810 /* already mapped? */
5811 if (em && em->start <= logical && em->start + em->len > logical) {
5812 free_extent_map(em);
5815 free_extent_map(em);
5818 em = alloc_extent_map();
5821 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5822 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5824 free_extent_map(em);
5828 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5829 em->bdev = (struct block_device *)map;
5830 em->start = logical;
5833 em->block_start = 0;
5834 em->block_len = em->len;
5836 map->num_stripes = num_stripes;
5837 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5838 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5839 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5840 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5841 map->type = btrfs_chunk_type(leaf, chunk);
5842 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5843 for (i = 0; i < num_stripes; i++) {
5844 map->stripes[i].physical =
5845 btrfs_stripe_offset_nr(leaf, chunk, i);
5846 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5847 read_extent_buffer(leaf, uuid, (unsigned long)
5848 btrfs_stripe_dev_uuid_nr(chunk, i),
5850 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5852 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5853 free_extent_map(em);
5856 if (!map->stripes[i].dev) {
5857 map->stripes[i].dev =
5858 add_missing_dev(root, devid, uuid);
5859 if (!map->stripes[i].dev) {
5860 free_extent_map(em);
5864 map->stripes[i].dev->in_fs_metadata = 1;
5867 write_lock(&map_tree->map_tree.lock);
5868 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5869 write_unlock(&map_tree->map_tree.lock);
5870 BUG_ON(ret); /* Tree corruption */
5871 free_extent_map(em);
5876 static void fill_device_from_item(struct extent_buffer *leaf,
5877 struct btrfs_dev_item *dev_item,
5878 struct btrfs_device *device)
5882 device->devid = btrfs_device_id(leaf, dev_item);
5883 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5884 device->total_bytes = device->disk_total_bytes;
5885 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5886 device->type = btrfs_device_type(leaf, dev_item);
5887 device->io_align = btrfs_device_io_align(leaf, dev_item);
5888 device->io_width = btrfs_device_io_width(leaf, dev_item);
5889 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5890 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5891 device->is_tgtdev_for_dev_replace = 0;
5893 ptr = btrfs_device_uuid(dev_item);
5894 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5897 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5899 struct btrfs_fs_devices *fs_devices;
5902 BUG_ON(!mutex_is_locked(&uuid_mutex));
5904 fs_devices = root->fs_info->fs_devices->seed;
5905 while (fs_devices) {
5906 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5910 fs_devices = fs_devices->seed;
5913 fs_devices = find_fsid(fsid);
5919 fs_devices = clone_fs_devices(fs_devices);
5920 if (IS_ERR(fs_devices)) {
5921 ret = PTR_ERR(fs_devices);
5925 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5926 root->fs_info->bdev_holder);
5928 free_fs_devices(fs_devices);
5932 if (!fs_devices->seeding) {
5933 __btrfs_close_devices(fs_devices);
5934 free_fs_devices(fs_devices);
5939 fs_devices->seed = root->fs_info->fs_devices->seed;
5940 root->fs_info->fs_devices->seed = fs_devices;
5945 static int read_one_dev(struct btrfs_root *root,
5946 struct extent_buffer *leaf,
5947 struct btrfs_dev_item *dev_item)
5949 struct btrfs_device *device;
5952 u8 fs_uuid[BTRFS_UUID_SIZE];
5953 u8 dev_uuid[BTRFS_UUID_SIZE];
5955 devid = btrfs_device_id(leaf, dev_item);
5956 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5958 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5961 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5962 ret = open_seed_devices(root, fs_uuid);
5963 if (ret && !btrfs_test_opt(root, DEGRADED))
5967 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5968 if (!device || !device->bdev) {
5969 if (!btrfs_test_opt(root, DEGRADED))
5973 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5974 device = add_missing_dev(root, devid, dev_uuid);
5977 } else if (!device->missing) {
5979 * this happens when a device that was properly setup
5980 * in the device info lists suddenly goes bad.
5981 * device->bdev is NULL, and so we have to set
5982 * device->missing to one here
5984 root->fs_info->fs_devices->missing_devices++;
5985 device->missing = 1;
5989 if (device->fs_devices != root->fs_info->fs_devices) {
5990 BUG_ON(device->writeable);
5991 if (device->generation !=
5992 btrfs_device_generation(leaf, dev_item))
5996 fill_device_from_item(leaf, dev_item, device);
5997 device->in_fs_metadata = 1;
5998 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5999 device->fs_devices->total_rw_bytes += device->total_bytes;
6000 spin_lock(&root->fs_info->free_chunk_lock);
6001 root->fs_info->free_chunk_space += device->total_bytes -
6003 spin_unlock(&root->fs_info->free_chunk_lock);
6009 int btrfs_read_sys_array(struct btrfs_root *root)
6011 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6012 struct extent_buffer *sb;
6013 struct btrfs_disk_key *disk_key;
6014 struct btrfs_chunk *chunk;
6016 unsigned long sb_ptr;
6022 struct btrfs_key key;
6024 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6025 BTRFS_SUPER_INFO_SIZE);
6028 btrfs_set_buffer_uptodate(sb);
6029 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6031 * The sb extent buffer is artifical and just used to read the system array.
6032 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6033 * pages up-to-date when the page is larger: extent does not cover the
6034 * whole page and consequently check_page_uptodate does not find all
6035 * the page's extents up-to-date (the hole beyond sb),
6036 * write_extent_buffer then triggers a WARN_ON.
6038 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6039 * but sb spans only this function. Add an explicit SetPageUptodate call
6040 * to silence the warning eg. on PowerPC 64.
6042 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6043 SetPageUptodate(sb->pages[0]);
6045 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6046 array_size = btrfs_super_sys_array_size(super_copy);
6048 ptr = super_copy->sys_chunk_array;
6049 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6052 while (cur < array_size) {
6053 disk_key = (struct btrfs_disk_key *)ptr;
6054 btrfs_disk_key_to_cpu(&key, disk_key);
6056 len = sizeof(*disk_key); ptr += len;
6060 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6061 chunk = (struct btrfs_chunk *)sb_ptr;
6062 ret = read_one_chunk(root, &key, sb, chunk);
6065 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6066 len = btrfs_chunk_item_size(num_stripes);
6075 free_extent_buffer(sb);
6079 int btrfs_read_chunk_tree(struct btrfs_root *root)
6081 struct btrfs_path *path;
6082 struct extent_buffer *leaf;
6083 struct btrfs_key key;
6084 struct btrfs_key found_key;
6088 root = root->fs_info->chunk_root;
6090 path = btrfs_alloc_path();
6094 mutex_lock(&uuid_mutex);
6098 * Read all device items, and then all the chunk items. All
6099 * device items are found before any chunk item (their object id
6100 * is smaller than the lowest possible object id for a chunk
6101 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6103 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6106 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6110 leaf = path->nodes[0];
6111 slot = path->slots[0];
6112 if (slot >= btrfs_header_nritems(leaf)) {
6113 ret = btrfs_next_leaf(root, path);
6120 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6121 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6122 struct btrfs_dev_item *dev_item;
6123 dev_item = btrfs_item_ptr(leaf, slot,
6124 struct btrfs_dev_item);
6125 ret = read_one_dev(root, leaf, dev_item);
6128 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6129 struct btrfs_chunk *chunk;
6130 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6131 ret = read_one_chunk(root, &found_key, leaf, chunk);
6139 unlock_chunks(root);
6140 mutex_unlock(&uuid_mutex);
6142 btrfs_free_path(path);
6146 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6148 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6149 struct btrfs_device *device;
6151 while (fs_devices) {
6152 mutex_lock(&fs_devices->device_list_mutex);
6153 list_for_each_entry(device, &fs_devices->devices, dev_list)
6154 device->dev_root = fs_info->dev_root;
6155 mutex_unlock(&fs_devices->device_list_mutex);
6157 fs_devices = fs_devices->seed;
6161 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6165 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6166 btrfs_dev_stat_reset(dev, i);
6169 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6171 struct btrfs_key key;
6172 struct btrfs_key found_key;
6173 struct btrfs_root *dev_root = fs_info->dev_root;
6174 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6175 struct extent_buffer *eb;
6178 struct btrfs_device *device;
6179 struct btrfs_path *path = NULL;
6182 path = btrfs_alloc_path();
6188 mutex_lock(&fs_devices->device_list_mutex);
6189 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6191 struct btrfs_dev_stats_item *ptr;
6194 key.type = BTRFS_DEV_STATS_KEY;
6195 key.offset = device->devid;
6196 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6198 __btrfs_reset_dev_stats(device);
6199 device->dev_stats_valid = 1;
6200 btrfs_release_path(path);
6203 slot = path->slots[0];
6204 eb = path->nodes[0];
6205 btrfs_item_key_to_cpu(eb, &found_key, slot);
6206 item_size = btrfs_item_size_nr(eb, slot);
6208 ptr = btrfs_item_ptr(eb, slot,
6209 struct btrfs_dev_stats_item);
6211 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6212 if (item_size >= (1 + i) * sizeof(__le64))
6213 btrfs_dev_stat_set(device, i,
6214 btrfs_dev_stats_value(eb, ptr, i));
6216 btrfs_dev_stat_reset(device, i);
6219 device->dev_stats_valid = 1;
6220 btrfs_dev_stat_print_on_load(device);
6221 btrfs_release_path(path);
6223 mutex_unlock(&fs_devices->device_list_mutex);
6226 btrfs_free_path(path);
6227 return ret < 0 ? ret : 0;
6230 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6231 struct btrfs_root *dev_root,
6232 struct btrfs_device *device)
6234 struct btrfs_path *path;
6235 struct btrfs_key key;
6236 struct extent_buffer *eb;
6237 struct btrfs_dev_stats_item *ptr;
6242 key.type = BTRFS_DEV_STATS_KEY;
6243 key.offset = device->devid;
6245 path = btrfs_alloc_path();
6247 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6249 printk_in_rcu(KERN_WARNING "BTRFS: "
6250 "error %d while searching for dev_stats item for device %s!\n",
6251 ret, rcu_str_deref(device->name));
6256 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6257 /* need to delete old one and insert a new one */
6258 ret = btrfs_del_item(trans, dev_root, path);
6260 printk_in_rcu(KERN_WARNING "BTRFS: "
6261 "delete too small dev_stats item for device %s failed %d!\n",
6262 rcu_str_deref(device->name), ret);
6269 /* need to insert a new item */
6270 btrfs_release_path(path);
6271 ret = btrfs_insert_empty_item(trans, dev_root, path,
6272 &key, sizeof(*ptr));
6274 printk_in_rcu(KERN_WARNING "BTRFS: "
6275 "insert dev_stats item for device %s failed %d!\n",
6276 rcu_str_deref(device->name), ret);
6281 eb = path->nodes[0];
6282 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6283 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6284 btrfs_set_dev_stats_value(eb, ptr, i,
6285 btrfs_dev_stat_read(device, i));
6286 btrfs_mark_buffer_dirty(eb);
6289 btrfs_free_path(path);
6294 * called from commit_transaction. Writes all changed device stats to disk.
6296 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6297 struct btrfs_fs_info *fs_info)
6299 struct btrfs_root *dev_root = fs_info->dev_root;
6300 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6301 struct btrfs_device *device;
6304 mutex_lock(&fs_devices->device_list_mutex);
6305 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6306 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6309 ret = update_dev_stat_item(trans, dev_root, device);
6311 device->dev_stats_dirty = 0;
6313 mutex_unlock(&fs_devices->device_list_mutex);
6318 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6320 btrfs_dev_stat_inc(dev, index);
6321 btrfs_dev_stat_print_on_error(dev);
6324 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6326 if (!dev->dev_stats_valid)
6328 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6329 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6330 rcu_str_deref(dev->name),
6331 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6332 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6333 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6334 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6335 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6338 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6342 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6343 if (btrfs_dev_stat_read(dev, i) != 0)
6345 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6346 return; /* all values == 0, suppress message */
6348 printk_in_rcu(KERN_INFO "BTRFS: "
6349 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6350 rcu_str_deref(dev->name),
6351 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6352 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6353 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6354 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6355 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6358 int btrfs_get_dev_stats(struct btrfs_root *root,
6359 struct btrfs_ioctl_get_dev_stats *stats)
6361 struct btrfs_device *dev;
6362 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6365 mutex_lock(&fs_devices->device_list_mutex);
6366 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6367 mutex_unlock(&fs_devices->device_list_mutex);
6370 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6372 } else if (!dev->dev_stats_valid) {
6373 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6375 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6376 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6377 if (stats->nr_items > i)
6379 btrfs_dev_stat_read_and_reset(dev, i);
6381 btrfs_dev_stat_reset(dev, i);
6384 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6385 if (stats->nr_items > i)
6386 stats->values[i] = btrfs_dev_stat_read(dev, i);
6388 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6389 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6393 int btrfs_scratch_superblock(struct btrfs_device *device)
6395 struct buffer_head *bh;
6396 struct btrfs_super_block *disk_super;
6398 bh = btrfs_read_dev_super(device->bdev);
6401 disk_super = (struct btrfs_super_block *)bh->b_data;
6403 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6404 set_buffer_dirty(bh);
6405 sync_dirty_buffer(bh);