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>
30 #include "extent_map.h"
32 #include "transaction.h"
33 #include "print-tree.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
39 #include "dev-replace.h"
41 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
42 struct btrfs_root
*root
,
43 struct btrfs_device
*device
);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
45 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
48 static DEFINE_MUTEX(uuid_mutex
);
49 static LIST_HEAD(fs_uuids
);
51 static void lock_chunks(struct btrfs_root
*root
)
53 mutex_lock(&root
->fs_info
->chunk_mutex
);
56 static void unlock_chunks(struct btrfs_root
*root
)
58 mutex_unlock(&root
->fs_info
->chunk_mutex
);
61 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
63 struct btrfs_device
*device
;
64 WARN_ON(fs_devices
->opened
);
65 while (!list_empty(&fs_devices
->devices
)) {
66 device
= list_entry(fs_devices
->devices
.next
,
67 struct btrfs_device
, dev_list
);
68 list_del(&device
->dev_list
);
69 rcu_string_free(device
->name
);
75 static void btrfs_kobject_uevent(struct block_device
*bdev
,
76 enum kobject_action action
)
80 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
82 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
84 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
85 &disk_to_dev(bdev
->bd_disk
)->kobj
);
88 void btrfs_cleanup_fs_uuids(void)
90 struct btrfs_fs_devices
*fs_devices
;
92 while (!list_empty(&fs_uuids
)) {
93 fs_devices
= list_entry(fs_uuids
.next
,
94 struct btrfs_fs_devices
, list
);
95 list_del(&fs_devices
->list
);
96 free_fs_devices(fs_devices
);
100 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
103 struct btrfs_device
*dev
;
105 list_for_each_entry(dev
, head
, dev_list
) {
106 if (dev
->devid
== devid
&&
107 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
114 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
116 struct btrfs_fs_devices
*fs_devices
;
118 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
119 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
126 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
127 int flush
, struct block_device
**bdev
,
128 struct buffer_head
**bh
)
132 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
135 ret
= PTR_ERR(*bdev
);
136 printk(KERN_INFO
"btrfs: open %s failed\n", device_path
);
141 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
142 ret
= set_blocksize(*bdev
, 4096);
144 blkdev_put(*bdev
, flags
);
147 invalidate_bdev(*bdev
);
148 *bh
= btrfs_read_dev_super(*bdev
);
151 blkdev_put(*bdev
, flags
);
163 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
164 struct bio
*head
, struct bio
*tail
)
167 struct bio
*old_head
;
169 old_head
= pending_bios
->head
;
170 pending_bios
->head
= head
;
171 if (pending_bios
->tail
)
172 tail
->bi_next
= old_head
;
174 pending_bios
->tail
= tail
;
178 * we try to collect pending bios for a device so we don't get a large
179 * number of procs sending bios down to the same device. This greatly
180 * improves the schedulers ability to collect and merge the bios.
182 * But, it also turns into a long list of bios to process and that is sure
183 * to eventually make the worker thread block. The solution here is to
184 * make some progress and then put this work struct back at the end of
185 * the list if the block device is congested. This way, multiple devices
186 * can make progress from a single worker thread.
188 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
191 struct backing_dev_info
*bdi
;
192 struct btrfs_fs_info
*fs_info
;
193 struct btrfs_pending_bios
*pending_bios
;
197 unsigned long num_run
;
198 unsigned long batch_run
= 0;
200 unsigned long last_waited
= 0;
202 int sync_pending
= 0;
203 struct blk_plug plug
;
206 * this function runs all the bios we've collected for
207 * a particular device. We don't want to wander off to
208 * another device without first sending all of these down.
209 * So, setup a plug here and finish it off before we return
211 blk_start_plug(&plug
);
213 bdi
= blk_get_backing_dev_info(device
->bdev
);
214 fs_info
= device
->dev_root
->fs_info
;
215 limit
= btrfs_async_submit_limit(fs_info
);
216 limit
= limit
* 2 / 3;
219 spin_lock(&device
->io_lock
);
224 /* take all the bios off the list at once and process them
225 * later on (without the lock held). But, remember the
226 * tail and other pointers so the bios can be properly reinserted
227 * into the list if we hit congestion
229 if (!force_reg
&& device
->pending_sync_bios
.head
) {
230 pending_bios
= &device
->pending_sync_bios
;
233 pending_bios
= &device
->pending_bios
;
237 pending
= pending_bios
->head
;
238 tail
= pending_bios
->tail
;
239 WARN_ON(pending
&& !tail
);
242 * if pending was null this time around, no bios need processing
243 * at all and we can stop. Otherwise it'll loop back up again
244 * and do an additional check so no bios are missed.
246 * device->running_pending is used to synchronize with the
249 if (device
->pending_sync_bios
.head
== NULL
&&
250 device
->pending_bios
.head
== NULL
) {
252 device
->running_pending
= 0;
255 device
->running_pending
= 1;
258 pending_bios
->head
= NULL
;
259 pending_bios
->tail
= NULL
;
261 spin_unlock(&device
->io_lock
);
266 /* we want to work on both lists, but do more bios on the
267 * sync list than the regular list
270 pending_bios
!= &device
->pending_sync_bios
&&
271 device
->pending_sync_bios
.head
) ||
272 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
273 device
->pending_bios
.head
)) {
274 spin_lock(&device
->io_lock
);
275 requeue_list(pending_bios
, pending
, tail
);
280 pending
= pending
->bi_next
;
283 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
284 waitqueue_active(&fs_info
->async_submit_wait
))
285 wake_up(&fs_info
->async_submit_wait
);
287 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
290 * if we're doing the sync list, record that our
291 * plug has some sync requests on it
293 * If we're doing the regular list and there are
294 * sync requests sitting around, unplug before
297 if (pending_bios
== &device
->pending_sync_bios
) {
299 } else if (sync_pending
) {
300 blk_finish_plug(&plug
);
301 blk_start_plug(&plug
);
305 btrfsic_submit_bio(cur
->bi_rw
, cur
);
312 * we made progress, there is more work to do and the bdi
313 * is now congested. Back off and let other work structs
316 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
317 fs_info
->fs_devices
->open_devices
> 1) {
318 struct io_context
*ioc
;
320 ioc
= current
->io_context
;
323 * the main goal here is that we don't want to
324 * block if we're going to be able to submit
325 * more requests without blocking.
327 * This code does two great things, it pokes into
328 * the elevator code from a filesystem _and_
329 * it makes assumptions about how batching works.
331 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
332 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
334 ioc
->last_waited
== last_waited
)) {
336 * we want to go through our batch of
337 * requests and stop. So, we copy out
338 * the ioc->last_waited time and test
339 * against it before looping
341 last_waited
= ioc
->last_waited
;
346 spin_lock(&device
->io_lock
);
347 requeue_list(pending_bios
, pending
, tail
);
348 device
->running_pending
= 1;
350 spin_unlock(&device
->io_lock
);
351 btrfs_requeue_work(&device
->work
);
354 /* unplug every 64 requests just for good measure */
355 if (batch_run
% 64 == 0) {
356 blk_finish_plug(&plug
);
357 blk_start_plug(&plug
);
366 spin_lock(&device
->io_lock
);
367 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
369 spin_unlock(&device
->io_lock
);
372 blk_finish_plug(&plug
);
375 static void pending_bios_fn(struct btrfs_work
*work
)
377 struct btrfs_device
*device
;
379 device
= container_of(work
, struct btrfs_device
, work
);
380 run_scheduled_bios(device
);
383 static noinline
int device_list_add(const char *path
,
384 struct btrfs_super_block
*disk_super
,
385 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
387 struct btrfs_device
*device
;
388 struct btrfs_fs_devices
*fs_devices
;
389 struct rcu_string
*name
;
390 u64 found_transid
= btrfs_super_generation(disk_super
);
392 fs_devices
= find_fsid(disk_super
->fsid
);
394 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
397 INIT_LIST_HEAD(&fs_devices
->devices
);
398 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
399 list_add(&fs_devices
->list
, &fs_uuids
);
400 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
401 fs_devices
->latest_devid
= devid
;
402 fs_devices
->latest_trans
= found_transid
;
403 mutex_init(&fs_devices
->device_list_mutex
);
406 device
= __find_device(&fs_devices
->devices
, devid
,
407 disk_super
->dev_item
.uuid
);
410 if (fs_devices
->opened
)
413 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
415 /* we can safely leave the fs_devices entry around */
418 device
->devid
= devid
;
419 device
->dev_stats_valid
= 0;
420 device
->work
.func
= pending_bios_fn
;
421 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
423 spin_lock_init(&device
->io_lock
);
425 name
= rcu_string_strdup(path
, GFP_NOFS
);
430 rcu_assign_pointer(device
->name
, name
);
431 INIT_LIST_HEAD(&device
->dev_alloc_list
);
433 /* init readahead state */
434 spin_lock_init(&device
->reada_lock
);
435 device
->reada_curr_zone
= NULL
;
436 atomic_set(&device
->reada_in_flight
, 0);
437 device
->reada_next
= 0;
438 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
439 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
441 mutex_lock(&fs_devices
->device_list_mutex
);
442 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
443 mutex_unlock(&fs_devices
->device_list_mutex
);
445 device
->fs_devices
= fs_devices
;
446 fs_devices
->num_devices
++;
447 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
448 name
= rcu_string_strdup(path
, GFP_NOFS
);
451 rcu_string_free(device
->name
);
452 rcu_assign_pointer(device
->name
, name
);
453 if (device
->missing
) {
454 fs_devices
->missing_devices
--;
459 if (found_transid
> fs_devices
->latest_trans
) {
460 fs_devices
->latest_devid
= devid
;
461 fs_devices
->latest_trans
= found_transid
;
463 *fs_devices_ret
= fs_devices
;
467 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
469 struct btrfs_fs_devices
*fs_devices
;
470 struct btrfs_device
*device
;
471 struct btrfs_device
*orig_dev
;
473 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
475 return ERR_PTR(-ENOMEM
);
477 INIT_LIST_HEAD(&fs_devices
->devices
);
478 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
479 INIT_LIST_HEAD(&fs_devices
->list
);
480 mutex_init(&fs_devices
->device_list_mutex
);
481 fs_devices
->latest_devid
= orig
->latest_devid
;
482 fs_devices
->latest_trans
= orig
->latest_trans
;
483 fs_devices
->total_devices
= orig
->total_devices
;
484 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
486 /* We have held the volume lock, it is safe to get the devices. */
487 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
488 struct rcu_string
*name
;
490 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
495 * This is ok to do without rcu read locked because we hold the
496 * uuid mutex so nothing we touch in here is going to disappear.
498 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
503 rcu_assign_pointer(device
->name
, name
);
505 device
->devid
= orig_dev
->devid
;
506 device
->work
.func
= pending_bios_fn
;
507 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
508 spin_lock_init(&device
->io_lock
);
509 INIT_LIST_HEAD(&device
->dev_list
);
510 INIT_LIST_HEAD(&device
->dev_alloc_list
);
512 list_add(&device
->dev_list
, &fs_devices
->devices
);
513 device
->fs_devices
= fs_devices
;
514 fs_devices
->num_devices
++;
518 free_fs_devices(fs_devices
);
519 return ERR_PTR(-ENOMEM
);
522 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
523 struct btrfs_fs_devices
*fs_devices
, int step
)
525 struct btrfs_device
*device
, *next
;
527 struct block_device
*latest_bdev
= NULL
;
528 u64 latest_devid
= 0;
529 u64 latest_transid
= 0;
531 mutex_lock(&uuid_mutex
);
533 /* This is the initialized path, it is safe to release the devices. */
534 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
535 if (device
->in_fs_metadata
) {
536 if (!device
->is_tgtdev_for_dev_replace
&&
538 device
->generation
> latest_transid
)) {
539 latest_devid
= device
->devid
;
540 latest_transid
= device
->generation
;
541 latest_bdev
= device
->bdev
;
546 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
548 * In the first step, keep the device which has
549 * the correct fsid and the devid that is used
550 * for the dev_replace procedure.
551 * In the second step, the dev_replace state is
552 * read from the device tree and it is known
553 * whether the procedure is really active or
554 * not, which means whether this device is
555 * used or whether it should be removed.
557 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
562 blkdev_put(device
->bdev
, device
->mode
);
564 fs_devices
->open_devices
--;
566 if (device
->writeable
) {
567 list_del_init(&device
->dev_alloc_list
);
568 device
->writeable
= 0;
569 if (!device
->is_tgtdev_for_dev_replace
)
570 fs_devices
->rw_devices
--;
572 list_del_init(&device
->dev_list
);
573 fs_devices
->num_devices
--;
574 rcu_string_free(device
->name
);
578 if (fs_devices
->seed
) {
579 fs_devices
= fs_devices
->seed
;
583 fs_devices
->latest_bdev
= latest_bdev
;
584 fs_devices
->latest_devid
= latest_devid
;
585 fs_devices
->latest_trans
= latest_transid
;
587 mutex_unlock(&uuid_mutex
);
590 static void __free_device(struct work_struct
*work
)
592 struct btrfs_device
*device
;
594 device
= container_of(work
, struct btrfs_device
, rcu_work
);
597 blkdev_put(device
->bdev
, device
->mode
);
599 rcu_string_free(device
->name
);
603 static void free_device(struct rcu_head
*head
)
605 struct btrfs_device
*device
;
607 device
= container_of(head
, struct btrfs_device
, rcu
);
609 INIT_WORK(&device
->rcu_work
, __free_device
);
610 schedule_work(&device
->rcu_work
);
613 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
615 struct btrfs_device
*device
;
617 if (--fs_devices
->opened
> 0)
620 mutex_lock(&fs_devices
->device_list_mutex
);
621 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
622 struct btrfs_device
*new_device
;
623 struct rcu_string
*name
;
626 fs_devices
->open_devices
--;
628 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
629 list_del_init(&device
->dev_alloc_list
);
630 fs_devices
->rw_devices
--;
633 if (device
->can_discard
)
634 fs_devices
->num_can_discard
--;
636 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
637 BUG_ON(!new_device
); /* -ENOMEM */
638 memcpy(new_device
, device
, sizeof(*new_device
));
640 /* Safe because we are under uuid_mutex */
642 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
643 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
644 rcu_assign_pointer(new_device
->name
, name
);
646 new_device
->bdev
= NULL
;
647 new_device
->writeable
= 0;
648 new_device
->in_fs_metadata
= 0;
649 new_device
->can_discard
= 0;
650 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
652 call_rcu(&device
->rcu
, free_device
);
654 mutex_unlock(&fs_devices
->device_list_mutex
);
656 WARN_ON(fs_devices
->open_devices
);
657 WARN_ON(fs_devices
->rw_devices
);
658 fs_devices
->opened
= 0;
659 fs_devices
->seeding
= 0;
664 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
666 struct btrfs_fs_devices
*seed_devices
= NULL
;
669 mutex_lock(&uuid_mutex
);
670 ret
= __btrfs_close_devices(fs_devices
);
671 if (!fs_devices
->opened
) {
672 seed_devices
= fs_devices
->seed
;
673 fs_devices
->seed
= NULL
;
675 mutex_unlock(&uuid_mutex
);
677 while (seed_devices
) {
678 fs_devices
= seed_devices
;
679 seed_devices
= fs_devices
->seed
;
680 __btrfs_close_devices(fs_devices
);
681 free_fs_devices(fs_devices
);
686 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
687 fmode_t flags
, void *holder
)
689 struct request_queue
*q
;
690 struct block_device
*bdev
;
691 struct list_head
*head
= &fs_devices
->devices
;
692 struct btrfs_device
*device
;
693 struct block_device
*latest_bdev
= NULL
;
694 struct buffer_head
*bh
;
695 struct btrfs_super_block
*disk_super
;
696 u64 latest_devid
= 0;
697 u64 latest_transid
= 0;
704 list_for_each_entry(device
, head
, dev_list
) {
710 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
715 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
716 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
717 if (devid
!= device
->devid
)
720 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
724 device
->generation
= btrfs_super_generation(disk_super
);
725 if (!latest_transid
|| device
->generation
> latest_transid
) {
726 latest_devid
= devid
;
727 latest_transid
= device
->generation
;
731 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
732 device
->writeable
= 0;
734 device
->writeable
= !bdev_read_only(bdev
);
738 q
= bdev_get_queue(bdev
);
739 if (blk_queue_discard(q
)) {
740 device
->can_discard
= 1;
741 fs_devices
->num_can_discard
++;
745 device
->in_fs_metadata
= 0;
746 device
->mode
= flags
;
748 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
749 fs_devices
->rotating
= 1;
751 fs_devices
->open_devices
++;
752 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
753 fs_devices
->rw_devices
++;
754 list_add(&device
->dev_alloc_list
,
755 &fs_devices
->alloc_list
);
762 blkdev_put(bdev
, flags
);
765 if (fs_devices
->open_devices
== 0) {
769 fs_devices
->seeding
= seeding
;
770 fs_devices
->opened
= 1;
771 fs_devices
->latest_bdev
= latest_bdev
;
772 fs_devices
->latest_devid
= latest_devid
;
773 fs_devices
->latest_trans
= latest_transid
;
774 fs_devices
->total_rw_bytes
= 0;
779 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
780 fmode_t flags
, void *holder
)
784 mutex_lock(&uuid_mutex
);
785 if (fs_devices
->opened
) {
786 fs_devices
->opened
++;
789 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
791 mutex_unlock(&uuid_mutex
);
795 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
796 struct btrfs_fs_devices
**fs_devices_ret
)
798 struct btrfs_super_block
*disk_super
;
799 struct block_device
*bdev
;
800 struct buffer_head
*bh
;
807 mutex_lock(&uuid_mutex
);
808 ret
= btrfs_get_bdev_and_sb(path
, flags
, holder
, 0, &bdev
, &bh
);
811 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
812 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
813 transid
= btrfs_super_generation(disk_super
);
814 total_devices
= btrfs_super_num_devices(disk_super
);
815 if (disk_super
->label
[0]) {
816 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
817 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
818 printk(KERN_INFO
"device label %s ", disk_super
->label
);
820 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
822 printk(KERN_CONT
"devid %llu transid %llu %s\n",
823 (unsigned long long)devid
, (unsigned long long)transid
, path
);
824 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
825 if (!ret
&& fs_devices_ret
)
826 (*fs_devices_ret
)->total_devices
= total_devices
;
828 blkdev_put(bdev
, flags
);
830 mutex_unlock(&uuid_mutex
);
834 /* helper to account the used device space in the range */
835 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
836 u64 end
, u64
*length
)
838 struct btrfs_key key
;
839 struct btrfs_root
*root
= device
->dev_root
;
840 struct btrfs_dev_extent
*dev_extent
;
841 struct btrfs_path
*path
;
845 struct extent_buffer
*l
;
849 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
852 path
= btrfs_alloc_path();
857 key
.objectid
= device
->devid
;
859 key
.type
= BTRFS_DEV_EXTENT_KEY
;
861 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
865 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
872 slot
= path
->slots
[0];
873 if (slot
>= btrfs_header_nritems(l
)) {
874 ret
= btrfs_next_leaf(root
, path
);
882 btrfs_item_key_to_cpu(l
, &key
, slot
);
884 if (key
.objectid
< device
->devid
)
887 if (key
.objectid
> device
->devid
)
890 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
893 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
894 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
896 if (key
.offset
<= start
&& extent_end
> end
) {
897 *length
= end
- start
+ 1;
899 } else if (key
.offset
<= start
&& extent_end
> start
)
900 *length
+= extent_end
- start
;
901 else if (key
.offset
> start
&& extent_end
<= end
)
902 *length
+= extent_end
- key
.offset
;
903 else if (key
.offset
> start
&& key
.offset
<= end
) {
904 *length
+= end
- key
.offset
+ 1;
906 } else if (key
.offset
> end
)
914 btrfs_free_path(path
);
919 * find_free_dev_extent - find free space in the specified device
920 * @device: the device which we search the free space in
921 * @num_bytes: the size of the free space that we need
922 * @start: store the start of the free space.
923 * @len: the size of the free space. that we find, or the size of the max
924 * free space if we don't find suitable free space
926 * this uses a pretty simple search, the expectation is that it is
927 * called very infrequently and that a given device has a small number
930 * @start is used to store the start of the free space if we find. But if we
931 * don't find suitable free space, it will be used to store the start position
932 * of the max free space.
934 * @len is used to store the size of the free space that we find.
935 * But if we don't find suitable free space, it is used to store the size of
936 * the max free space.
938 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
939 u64
*start
, u64
*len
)
941 struct btrfs_key key
;
942 struct btrfs_root
*root
= device
->dev_root
;
943 struct btrfs_dev_extent
*dev_extent
;
944 struct btrfs_path
*path
;
950 u64 search_end
= device
->total_bytes
;
953 struct extent_buffer
*l
;
955 /* FIXME use last free of some kind */
957 /* we don't want to overwrite the superblock on the drive,
958 * so we make sure to start at an offset of at least 1MB
960 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
962 max_hole_start
= search_start
;
966 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
971 path
= btrfs_alloc_path();
978 key
.objectid
= device
->devid
;
979 key
.offset
= search_start
;
980 key
.type
= BTRFS_DEV_EXTENT_KEY
;
982 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
986 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
993 slot
= path
->slots
[0];
994 if (slot
>= btrfs_header_nritems(l
)) {
995 ret
= btrfs_next_leaf(root
, path
);
1003 btrfs_item_key_to_cpu(l
, &key
, slot
);
1005 if (key
.objectid
< device
->devid
)
1008 if (key
.objectid
> device
->devid
)
1011 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1014 if (key
.offset
> search_start
) {
1015 hole_size
= key
.offset
- search_start
;
1017 if (hole_size
> max_hole_size
) {
1018 max_hole_start
= search_start
;
1019 max_hole_size
= hole_size
;
1023 * If this free space is greater than which we need,
1024 * it must be the max free space that we have found
1025 * until now, so max_hole_start must point to the start
1026 * of this free space and the length of this free space
1027 * is stored in max_hole_size. Thus, we return
1028 * max_hole_start and max_hole_size and go back to the
1031 if (hole_size
>= num_bytes
) {
1037 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1038 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1040 if (extent_end
> search_start
)
1041 search_start
= extent_end
;
1048 * At this point, search_start should be the end of
1049 * allocated dev extents, and when shrinking the device,
1050 * search_end may be smaller than search_start.
1052 if (search_end
> search_start
)
1053 hole_size
= search_end
- search_start
;
1055 if (hole_size
> max_hole_size
) {
1056 max_hole_start
= search_start
;
1057 max_hole_size
= hole_size
;
1061 if (hole_size
< num_bytes
)
1067 btrfs_free_path(path
);
1069 *start
= max_hole_start
;
1071 *len
= max_hole_size
;
1075 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1076 struct btrfs_device
*device
,
1080 struct btrfs_path
*path
;
1081 struct btrfs_root
*root
= device
->dev_root
;
1082 struct btrfs_key key
;
1083 struct btrfs_key found_key
;
1084 struct extent_buffer
*leaf
= NULL
;
1085 struct btrfs_dev_extent
*extent
= NULL
;
1087 path
= btrfs_alloc_path();
1091 key
.objectid
= device
->devid
;
1093 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1095 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1097 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1098 BTRFS_DEV_EXTENT_KEY
);
1101 leaf
= path
->nodes
[0];
1102 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1103 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1104 struct btrfs_dev_extent
);
1105 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1106 btrfs_dev_extent_length(leaf
, extent
) < start
);
1108 btrfs_release_path(path
);
1110 } else if (ret
== 0) {
1111 leaf
= path
->nodes
[0];
1112 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1113 struct btrfs_dev_extent
);
1115 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1119 if (device
->bytes_used
> 0) {
1120 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1121 device
->bytes_used
-= len
;
1122 spin_lock(&root
->fs_info
->free_chunk_lock
);
1123 root
->fs_info
->free_chunk_space
+= len
;
1124 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1126 ret
= btrfs_del_item(trans
, root
, path
);
1128 btrfs_error(root
->fs_info
, ret
,
1129 "Failed to remove dev extent item");
1132 btrfs_free_path(path
);
1136 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1137 struct btrfs_device
*device
,
1138 u64 chunk_tree
, u64 chunk_objectid
,
1139 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1142 struct btrfs_path
*path
;
1143 struct btrfs_root
*root
= device
->dev_root
;
1144 struct btrfs_dev_extent
*extent
;
1145 struct extent_buffer
*leaf
;
1146 struct btrfs_key key
;
1148 WARN_ON(!device
->in_fs_metadata
);
1149 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1150 path
= btrfs_alloc_path();
1154 key
.objectid
= device
->devid
;
1156 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1157 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1162 leaf
= path
->nodes
[0];
1163 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1164 struct btrfs_dev_extent
);
1165 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1166 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1167 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1169 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1170 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1173 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1174 btrfs_mark_buffer_dirty(leaf
);
1176 btrfs_free_path(path
);
1180 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1181 u64 objectid
, u64
*offset
)
1183 struct btrfs_path
*path
;
1185 struct btrfs_key key
;
1186 struct btrfs_chunk
*chunk
;
1187 struct btrfs_key found_key
;
1189 path
= btrfs_alloc_path();
1193 key
.objectid
= objectid
;
1194 key
.offset
= (u64
)-1;
1195 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1197 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1201 BUG_ON(ret
== 0); /* Corruption */
1203 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1207 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1209 if (found_key
.objectid
!= objectid
)
1212 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1213 struct btrfs_chunk
);
1214 *offset
= found_key
.offset
+
1215 btrfs_chunk_length(path
->nodes
[0], chunk
);
1220 btrfs_free_path(path
);
1224 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1227 struct btrfs_key key
;
1228 struct btrfs_key found_key
;
1229 struct btrfs_path
*path
;
1231 root
= root
->fs_info
->chunk_root
;
1233 path
= btrfs_alloc_path();
1237 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1238 key
.type
= BTRFS_DEV_ITEM_KEY
;
1239 key
.offset
= (u64
)-1;
1241 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1245 BUG_ON(ret
== 0); /* Corruption */
1247 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1248 BTRFS_DEV_ITEM_KEY
);
1252 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1254 *objectid
= found_key
.offset
+ 1;
1258 btrfs_free_path(path
);
1263 * the device information is stored in the chunk root
1264 * the btrfs_device struct should be fully filled in
1266 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1267 struct btrfs_root
*root
,
1268 struct btrfs_device
*device
)
1271 struct btrfs_path
*path
;
1272 struct btrfs_dev_item
*dev_item
;
1273 struct extent_buffer
*leaf
;
1274 struct btrfs_key key
;
1277 root
= root
->fs_info
->chunk_root
;
1279 path
= btrfs_alloc_path();
1283 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1284 key
.type
= BTRFS_DEV_ITEM_KEY
;
1285 key
.offset
= device
->devid
;
1287 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1292 leaf
= path
->nodes
[0];
1293 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1295 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1296 btrfs_set_device_generation(leaf
, dev_item
, 0);
1297 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1298 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1299 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1300 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1301 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1302 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1303 btrfs_set_device_group(leaf
, dev_item
, 0);
1304 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1305 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1306 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1308 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1309 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1310 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1311 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1312 btrfs_mark_buffer_dirty(leaf
);
1316 btrfs_free_path(path
);
1320 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1321 struct btrfs_device
*device
)
1324 struct btrfs_path
*path
;
1325 struct btrfs_key key
;
1326 struct btrfs_trans_handle
*trans
;
1328 root
= root
->fs_info
->chunk_root
;
1330 path
= btrfs_alloc_path();
1334 trans
= btrfs_start_transaction(root
, 0);
1335 if (IS_ERR(trans
)) {
1336 btrfs_free_path(path
);
1337 return PTR_ERR(trans
);
1339 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1340 key
.type
= BTRFS_DEV_ITEM_KEY
;
1341 key
.offset
= device
->devid
;
1344 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1353 ret
= btrfs_del_item(trans
, root
, path
);
1357 btrfs_free_path(path
);
1358 unlock_chunks(root
);
1359 btrfs_commit_transaction(trans
, root
);
1363 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1365 struct btrfs_device
*device
;
1366 struct btrfs_device
*next_device
;
1367 struct block_device
*bdev
;
1368 struct buffer_head
*bh
= NULL
;
1369 struct btrfs_super_block
*disk_super
;
1370 struct btrfs_fs_devices
*cur_devices
;
1376 bool clear_super
= false;
1378 mutex_lock(&uuid_mutex
);
1380 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1381 root
->fs_info
->avail_system_alloc_bits
|
1382 root
->fs_info
->avail_metadata_alloc_bits
;
1384 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1385 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1386 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1387 WARN_ON(num_devices
< 1);
1390 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1392 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1393 printk(KERN_ERR
"btrfs: unable to go below four devices "
1399 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1400 printk(KERN_ERR
"btrfs: unable to go below two "
1401 "devices on raid1\n");
1406 if (strcmp(device_path
, "missing") == 0) {
1407 struct list_head
*devices
;
1408 struct btrfs_device
*tmp
;
1411 devices
= &root
->fs_info
->fs_devices
->devices
;
1413 * It is safe to read the devices since the volume_mutex
1416 list_for_each_entry(tmp
, devices
, dev_list
) {
1417 if (tmp
->in_fs_metadata
&&
1418 !tmp
->is_tgtdev_for_dev_replace
&&
1428 printk(KERN_ERR
"btrfs: no missing devices found to "
1433 ret
= btrfs_get_bdev_and_sb(device_path
,
1434 FMODE_WRITE
| FMODE_EXCL
,
1435 root
->fs_info
->bdev_holder
, 0,
1439 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1440 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1441 dev_uuid
= disk_super
->dev_item
.uuid
;
1442 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1450 if (device
->is_tgtdev_for_dev_replace
) {
1451 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1456 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1457 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1463 if (device
->writeable
) {
1465 list_del_init(&device
->dev_alloc_list
);
1466 unlock_chunks(root
);
1467 root
->fs_info
->fs_devices
->rw_devices
--;
1471 ret
= btrfs_shrink_device(device
, 0);
1476 * TODO: the superblock still includes this device in its num_devices
1477 * counter although write_all_supers() is not locked out. This
1478 * could give a filesystem state which requires a degraded mount.
1480 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1484 spin_lock(&root
->fs_info
->free_chunk_lock
);
1485 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1487 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1489 device
->in_fs_metadata
= 0;
1490 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1493 * the device list mutex makes sure that we don't change
1494 * the device list while someone else is writing out all
1495 * the device supers.
1498 cur_devices
= device
->fs_devices
;
1499 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1500 list_del_rcu(&device
->dev_list
);
1502 device
->fs_devices
->num_devices
--;
1503 device
->fs_devices
->total_devices
--;
1505 if (device
->missing
)
1506 root
->fs_info
->fs_devices
->missing_devices
--;
1508 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1509 struct btrfs_device
, dev_list
);
1510 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1511 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1512 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1513 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1516 device
->fs_devices
->open_devices
--;
1518 call_rcu(&device
->rcu
, free_device
);
1519 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1521 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1522 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1524 if (cur_devices
->open_devices
== 0) {
1525 struct btrfs_fs_devices
*fs_devices
;
1526 fs_devices
= root
->fs_info
->fs_devices
;
1527 while (fs_devices
) {
1528 if (fs_devices
->seed
== cur_devices
)
1530 fs_devices
= fs_devices
->seed
;
1532 fs_devices
->seed
= cur_devices
->seed
;
1533 cur_devices
->seed
= NULL
;
1535 __btrfs_close_devices(cur_devices
);
1536 unlock_chunks(root
);
1537 free_fs_devices(cur_devices
);
1540 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1541 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1544 * at this point, the device is zero sized. We want to
1545 * remove it from the devices list and zero out the old super
1547 if (clear_super
&& disk_super
) {
1548 /* make sure this device isn't detected as part of
1551 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1552 set_buffer_dirty(bh
);
1553 sync_dirty_buffer(bh
);
1558 /* Notify udev that device has changed */
1559 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1564 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1566 mutex_unlock(&uuid_mutex
);
1569 if (device
->writeable
) {
1571 list_add(&device
->dev_alloc_list
,
1572 &root
->fs_info
->fs_devices
->alloc_list
);
1573 unlock_chunks(root
);
1574 root
->fs_info
->fs_devices
->rw_devices
++;
1579 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1580 struct btrfs_device
*srcdev
)
1582 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1583 list_del_rcu(&srcdev
->dev_list
);
1584 list_del_rcu(&srcdev
->dev_alloc_list
);
1585 fs_info
->fs_devices
->num_devices
--;
1586 if (srcdev
->missing
) {
1587 fs_info
->fs_devices
->missing_devices
--;
1588 fs_info
->fs_devices
->rw_devices
++;
1590 if (srcdev
->can_discard
)
1591 fs_info
->fs_devices
->num_can_discard
--;
1593 fs_info
->fs_devices
->open_devices
--;
1595 call_rcu(&srcdev
->rcu
, free_device
);
1598 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1599 struct btrfs_device
*tgtdev
)
1601 struct btrfs_device
*next_device
;
1604 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1606 btrfs_scratch_superblock(tgtdev
);
1607 fs_info
->fs_devices
->open_devices
--;
1609 fs_info
->fs_devices
->num_devices
--;
1610 if (tgtdev
->can_discard
)
1611 fs_info
->fs_devices
->num_can_discard
++;
1613 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1614 struct btrfs_device
, dev_list
);
1615 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1616 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1617 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1618 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1619 list_del_rcu(&tgtdev
->dev_list
);
1621 call_rcu(&tgtdev
->rcu
, free_device
);
1623 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1626 int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1627 struct btrfs_device
**device
)
1630 struct btrfs_super_block
*disk_super
;
1633 struct block_device
*bdev
;
1634 struct buffer_head
*bh
;
1637 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1638 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1641 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1642 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1643 dev_uuid
= disk_super
->dev_item
.uuid
;
1644 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1649 blkdev_put(bdev
, FMODE_READ
);
1653 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1655 struct btrfs_device
**device
)
1658 if (strcmp(device_path
, "missing") == 0) {
1659 struct list_head
*devices
;
1660 struct btrfs_device
*tmp
;
1662 devices
= &root
->fs_info
->fs_devices
->devices
;
1664 * It is safe to read the devices since the volume_mutex
1665 * is held by the caller.
1667 list_for_each_entry(tmp
, devices
, dev_list
) {
1668 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1675 pr_err("btrfs: no missing device found\n");
1681 return btrfs_find_device_by_path(root
, device_path
, device
);
1686 * does all the dirty work required for changing file system's UUID.
1688 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1690 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1691 struct btrfs_fs_devices
*old_devices
;
1692 struct btrfs_fs_devices
*seed_devices
;
1693 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1694 struct btrfs_device
*device
;
1697 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1698 if (!fs_devices
->seeding
)
1701 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1705 old_devices
= clone_fs_devices(fs_devices
);
1706 if (IS_ERR(old_devices
)) {
1707 kfree(seed_devices
);
1708 return PTR_ERR(old_devices
);
1711 list_add(&old_devices
->list
, &fs_uuids
);
1713 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1714 seed_devices
->opened
= 1;
1715 INIT_LIST_HEAD(&seed_devices
->devices
);
1716 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1717 mutex_init(&seed_devices
->device_list_mutex
);
1719 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1720 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1722 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1724 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1725 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1726 device
->fs_devices
= seed_devices
;
1729 fs_devices
->seeding
= 0;
1730 fs_devices
->num_devices
= 0;
1731 fs_devices
->open_devices
= 0;
1732 fs_devices
->total_devices
= 0;
1733 fs_devices
->seed
= seed_devices
;
1735 generate_random_uuid(fs_devices
->fsid
);
1736 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1737 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1738 super_flags
= btrfs_super_flags(disk_super
) &
1739 ~BTRFS_SUPER_FLAG_SEEDING
;
1740 btrfs_set_super_flags(disk_super
, super_flags
);
1746 * strore the expected generation for seed devices in device items.
1748 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1749 struct btrfs_root
*root
)
1751 struct btrfs_path
*path
;
1752 struct extent_buffer
*leaf
;
1753 struct btrfs_dev_item
*dev_item
;
1754 struct btrfs_device
*device
;
1755 struct btrfs_key key
;
1756 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1757 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1761 path
= btrfs_alloc_path();
1765 root
= root
->fs_info
->chunk_root
;
1766 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1768 key
.type
= BTRFS_DEV_ITEM_KEY
;
1771 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1775 leaf
= path
->nodes
[0];
1777 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1778 ret
= btrfs_next_leaf(root
, path
);
1783 leaf
= path
->nodes
[0];
1784 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1785 btrfs_release_path(path
);
1789 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1790 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1791 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1794 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1795 struct btrfs_dev_item
);
1796 devid
= btrfs_device_id(leaf
, dev_item
);
1797 read_extent_buffer(leaf
, dev_uuid
,
1798 (unsigned long)btrfs_device_uuid(dev_item
),
1800 read_extent_buffer(leaf
, fs_uuid
,
1801 (unsigned long)btrfs_device_fsid(dev_item
),
1803 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1805 BUG_ON(!device
); /* Logic error */
1807 if (device
->fs_devices
->seeding
) {
1808 btrfs_set_device_generation(leaf
, dev_item
,
1809 device
->generation
);
1810 btrfs_mark_buffer_dirty(leaf
);
1818 btrfs_free_path(path
);
1822 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1824 struct request_queue
*q
;
1825 struct btrfs_trans_handle
*trans
;
1826 struct btrfs_device
*device
;
1827 struct block_device
*bdev
;
1828 struct list_head
*devices
;
1829 struct super_block
*sb
= root
->fs_info
->sb
;
1830 struct rcu_string
*name
;
1832 int seeding_dev
= 0;
1835 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1838 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1839 root
->fs_info
->bdev_holder
);
1841 return PTR_ERR(bdev
);
1843 if (root
->fs_info
->fs_devices
->seeding
) {
1845 down_write(&sb
->s_umount
);
1846 mutex_lock(&uuid_mutex
);
1849 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1851 devices
= &root
->fs_info
->fs_devices
->devices
;
1853 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1854 list_for_each_entry(device
, devices
, dev_list
) {
1855 if (device
->bdev
== bdev
) {
1858 &root
->fs_info
->fs_devices
->device_list_mutex
);
1862 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1864 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1866 /* we can safely leave the fs_devices entry around */
1871 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1877 rcu_assign_pointer(device
->name
, name
);
1879 ret
= find_next_devid(root
, &device
->devid
);
1881 rcu_string_free(device
->name
);
1886 trans
= btrfs_start_transaction(root
, 0);
1887 if (IS_ERR(trans
)) {
1888 rcu_string_free(device
->name
);
1890 ret
= PTR_ERR(trans
);
1896 q
= bdev_get_queue(bdev
);
1897 if (blk_queue_discard(q
))
1898 device
->can_discard
= 1;
1899 device
->writeable
= 1;
1900 device
->work
.func
= pending_bios_fn
;
1901 generate_random_uuid(device
->uuid
);
1902 spin_lock_init(&device
->io_lock
);
1903 device
->generation
= trans
->transid
;
1904 device
->io_width
= root
->sectorsize
;
1905 device
->io_align
= root
->sectorsize
;
1906 device
->sector_size
= root
->sectorsize
;
1907 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1908 device
->disk_total_bytes
= device
->total_bytes
;
1909 device
->dev_root
= root
->fs_info
->dev_root
;
1910 device
->bdev
= bdev
;
1911 device
->in_fs_metadata
= 1;
1912 device
->is_tgtdev_for_dev_replace
= 0;
1913 device
->mode
= FMODE_EXCL
;
1914 set_blocksize(device
->bdev
, 4096);
1917 sb
->s_flags
&= ~MS_RDONLY
;
1918 ret
= btrfs_prepare_sprout(root
);
1919 BUG_ON(ret
); /* -ENOMEM */
1922 device
->fs_devices
= root
->fs_info
->fs_devices
;
1924 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1925 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1926 list_add(&device
->dev_alloc_list
,
1927 &root
->fs_info
->fs_devices
->alloc_list
);
1928 root
->fs_info
->fs_devices
->num_devices
++;
1929 root
->fs_info
->fs_devices
->open_devices
++;
1930 root
->fs_info
->fs_devices
->rw_devices
++;
1931 root
->fs_info
->fs_devices
->total_devices
++;
1932 if (device
->can_discard
)
1933 root
->fs_info
->fs_devices
->num_can_discard
++;
1934 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1936 spin_lock(&root
->fs_info
->free_chunk_lock
);
1937 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
1938 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1940 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1941 root
->fs_info
->fs_devices
->rotating
= 1;
1943 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
1944 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
1945 total_bytes
+ device
->total_bytes
);
1947 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
1948 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
1950 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1953 ret
= init_first_rw_device(trans
, root
, device
);
1955 btrfs_abort_transaction(trans
, root
, ret
);
1958 ret
= btrfs_finish_sprout(trans
, root
);
1960 btrfs_abort_transaction(trans
, root
, ret
);
1964 ret
= btrfs_add_device(trans
, root
, device
);
1966 btrfs_abort_transaction(trans
, root
, ret
);
1972 * we've got more storage, clear any full flags on the space
1975 btrfs_clear_space_info_full(root
->fs_info
);
1977 unlock_chunks(root
);
1978 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1979 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1980 ret
= btrfs_commit_transaction(trans
, root
);
1983 mutex_unlock(&uuid_mutex
);
1984 up_write(&sb
->s_umount
);
1986 if (ret
) /* transaction commit */
1989 ret
= btrfs_relocate_sys_chunks(root
);
1991 btrfs_error(root
->fs_info
, ret
,
1992 "Failed to relocate sys chunks after "
1993 "device initialization. This can be fixed "
1994 "using the \"btrfs balance\" command.");
1995 trans
= btrfs_attach_transaction(root
);
1996 if (IS_ERR(trans
)) {
1997 if (PTR_ERR(trans
) == -ENOENT
)
1999 return PTR_ERR(trans
);
2001 ret
= btrfs_commit_transaction(trans
, root
);
2007 unlock_chunks(root
);
2008 btrfs_end_transaction(trans
, root
);
2009 rcu_string_free(device
->name
);
2012 blkdev_put(bdev
, FMODE_EXCL
);
2014 mutex_unlock(&uuid_mutex
);
2015 up_write(&sb
->s_umount
);
2020 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2021 struct btrfs_device
**device_out
)
2023 struct request_queue
*q
;
2024 struct btrfs_device
*device
;
2025 struct block_device
*bdev
;
2026 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2027 struct list_head
*devices
;
2028 struct rcu_string
*name
;
2032 if (fs_info
->fs_devices
->seeding
)
2035 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2036 fs_info
->bdev_holder
);
2038 return PTR_ERR(bdev
);
2040 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2042 devices
= &fs_info
->fs_devices
->devices
;
2043 list_for_each_entry(device
, devices
, dev_list
) {
2044 if (device
->bdev
== bdev
) {
2050 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
2056 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2062 rcu_assign_pointer(device
->name
, name
);
2064 q
= bdev_get_queue(bdev
);
2065 if (blk_queue_discard(q
))
2066 device
->can_discard
= 1;
2067 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2068 device
->writeable
= 1;
2069 device
->work
.func
= pending_bios_fn
;
2070 generate_random_uuid(device
->uuid
);
2071 device
->devid
= BTRFS_DEV_REPLACE_DEVID
;
2072 spin_lock_init(&device
->io_lock
);
2073 device
->generation
= 0;
2074 device
->io_width
= root
->sectorsize
;
2075 device
->io_align
= root
->sectorsize
;
2076 device
->sector_size
= root
->sectorsize
;
2077 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2078 device
->disk_total_bytes
= device
->total_bytes
;
2079 device
->dev_root
= fs_info
->dev_root
;
2080 device
->bdev
= bdev
;
2081 device
->in_fs_metadata
= 1;
2082 device
->is_tgtdev_for_dev_replace
= 1;
2083 device
->mode
= FMODE_EXCL
;
2084 set_blocksize(device
->bdev
, 4096);
2085 device
->fs_devices
= fs_info
->fs_devices
;
2086 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2087 fs_info
->fs_devices
->num_devices
++;
2088 fs_info
->fs_devices
->open_devices
++;
2089 if (device
->can_discard
)
2090 fs_info
->fs_devices
->num_can_discard
++;
2091 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2093 *device_out
= device
;
2097 blkdev_put(bdev
, FMODE_EXCL
);
2101 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2102 struct btrfs_device
*tgtdev
)
2104 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2105 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2106 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2107 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2108 tgtdev
->dev_root
= fs_info
->dev_root
;
2109 tgtdev
->in_fs_metadata
= 1;
2112 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2113 struct btrfs_device
*device
)
2116 struct btrfs_path
*path
;
2117 struct btrfs_root
*root
;
2118 struct btrfs_dev_item
*dev_item
;
2119 struct extent_buffer
*leaf
;
2120 struct btrfs_key key
;
2122 root
= device
->dev_root
->fs_info
->chunk_root
;
2124 path
= btrfs_alloc_path();
2128 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2129 key
.type
= BTRFS_DEV_ITEM_KEY
;
2130 key
.offset
= device
->devid
;
2132 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2141 leaf
= path
->nodes
[0];
2142 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2144 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2145 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2146 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2147 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2148 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2149 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2150 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2151 btrfs_mark_buffer_dirty(leaf
);
2154 btrfs_free_path(path
);
2158 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2159 struct btrfs_device
*device
, u64 new_size
)
2161 struct btrfs_super_block
*super_copy
=
2162 device
->dev_root
->fs_info
->super_copy
;
2163 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2164 u64 diff
= new_size
- device
->total_bytes
;
2166 if (!device
->writeable
)
2168 if (new_size
<= device
->total_bytes
||
2169 device
->is_tgtdev_for_dev_replace
)
2172 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2173 device
->fs_devices
->total_rw_bytes
+= diff
;
2175 device
->total_bytes
= new_size
;
2176 device
->disk_total_bytes
= new_size
;
2177 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2179 return btrfs_update_device(trans
, device
);
2182 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2183 struct btrfs_device
*device
, u64 new_size
)
2186 lock_chunks(device
->dev_root
);
2187 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2188 unlock_chunks(device
->dev_root
);
2192 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2193 struct btrfs_root
*root
,
2194 u64 chunk_tree
, u64 chunk_objectid
,
2198 struct btrfs_path
*path
;
2199 struct btrfs_key key
;
2201 root
= root
->fs_info
->chunk_root
;
2202 path
= btrfs_alloc_path();
2206 key
.objectid
= chunk_objectid
;
2207 key
.offset
= chunk_offset
;
2208 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2210 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2213 else if (ret
> 0) { /* Logic error or corruption */
2214 btrfs_error(root
->fs_info
, -ENOENT
,
2215 "Failed lookup while freeing chunk.");
2220 ret
= btrfs_del_item(trans
, root
, path
);
2222 btrfs_error(root
->fs_info
, ret
,
2223 "Failed to delete chunk item.");
2225 btrfs_free_path(path
);
2229 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2232 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2233 struct btrfs_disk_key
*disk_key
;
2234 struct btrfs_chunk
*chunk
;
2241 struct btrfs_key key
;
2243 array_size
= btrfs_super_sys_array_size(super_copy
);
2245 ptr
= super_copy
->sys_chunk_array
;
2248 while (cur
< array_size
) {
2249 disk_key
= (struct btrfs_disk_key
*)ptr
;
2250 btrfs_disk_key_to_cpu(&key
, disk_key
);
2252 len
= sizeof(*disk_key
);
2254 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2255 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2256 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2257 len
+= btrfs_chunk_item_size(num_stripes
);
2262 if (key
.objectid
== chunk_objectid
&&
2263 key
.offset
== chunk_offset
) {
2264 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2266 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2275 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2276 u64 chunk_tree
, u64 chunk_objectid
,
2279 struct extent_map_tree
*em_tree
;
2280 struct btrfs_root
*extent_root
;
2281 struct btrfs_trans_handle
*trans
;
2282 struct extent_map
*em
;
2283 struct map_lookup
*map
;
2287 root
= root
->fs_info
->chunk_root
;
2288 extent_root
= root
->fs_info
->extent_root
;
2289 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2291 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2295 /* step one, relocate all the extents inside this chunk */
2296 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2300 trans
= btrfs_start_transaction(root
, 0);
2301 BUG_ON(IS_ERR(trans
));
2306 * step two, delete the device extents and the
2307 * chunk tree entries
2309 read_lock(&em_tree
->lock
);
2310 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2311 read_unlock(&em_tree
->lock
);
2313 BUG_ON(!em
|| em
->start
> chunk_offset
||
2314 em
->start
+ em
->len
< chunk_offset
);
2315 map
= (struct map_lookup
*)em
->bdev
;
2317 for (i
= 0; i
< map
->num_stripes
; i
++) {
2318 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2319 map
->stripes
[i
].physical
);
2322 if (map
->stripes
[i
].dev
) {
2323 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2327 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2332 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2334 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2335 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2339 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2342 write_lock(&em_tree
->lock
);
2343 remove_extent_mapping(em_tree
, em
);
2344 write_unlock(&em_tree
->lock
);
2349 /* once for the tree */
2350 free_extent_map(em
);
2352 free_extent_map(em
);
2354 unlock_chunks(root
);
2355 btrfs_end_transaction(trans
, root
);
2359 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2361 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2362 struct btrfs_path
*path
;
2363 struct extent_buffer
*leaf
;
2364 struct btrfs_chunk
*chunk
;
2365 struct btrfs_key key
;
2366 struct btrfs_key found_key
;
2367 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2369 bool retried
= false;
2373 path
= btrfs_alloc_path();
2378 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2379 key
.offset
= (u64
)-1;
2380 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2383 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2386 BUG_ON(ret
== 0); /* Corruption */
2388 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2395 leaf
= path
->nodes
[0];
2396 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2398 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2399 struct btrfs_chunk
);
2400 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2401 btrfs_release_path(path
);
2403 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2404 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2413 if (found_key
.offset
== 0)
2415 key
.offset
= found_key
.offset
- 1;
2418 if (failed
&& !retried
) {
2422 } else if (failed
&& retried
) {
2427 btrfs_free_path(path
);
2431 static int insert_balance_item(struct btrfs_root
*root
,
2432 struct btrfs_balance_control
*bctl
)
2434 struct btrfs_trans_handle
*trans
;
2435 struct btrfs_balance_item
*item
;
2436 struct btrfs_disk_balance_args disk_bargs
;
2437 struct btrfs_path
*path
;
2438 struct extent_buffer
*leaf
;
2439 struct btrfs_key key
;
2442 path
= btrfs_alloc_path();
2446 trans
= btrfs_start_transaction(root
, 0);
2447 if (IS_ERR(trans
)) {
2448 btrfs_free_path(path
);
2449 return PTR_ERR(trans
);
2452 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2453 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2456 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2461 leaf
= path
->nodes
[0];
2462 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2464 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2466 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2467 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2468 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2469 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2470 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2471 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2473 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2475 btrfs_mark_buffer_dirty(leaf
);
2477 btrfs_free_path(path
);
2478 err
= btrfs_commit_transaction(trans
, root
);
2484 static int del_balance_item(struct btrfs_root
*root
)
2486 struct btrfs_trans_handle
*trans
;
2487 struct btrfs_path
*path
;
2488 struct btrfs_key key
;
2491 path
= btrfs_alloc_path();
2495 trans
= btrfs_start_transaction(root
, 0);
2496 if (IS_ERR(trans
)) {
2497 btrfs_free_path(path
);
2498 return PTR_ERR(trans
);
2501 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2502 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2505 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2513 ret
= btrfs_del_item(trans
, root
, path
);
2515 btrfs_free_path(path
);
2516 err
= btrfs_commit_transaction(trans
, root
);
2523 * This is a heuristic used to reduce the number of chunks balanced on
2524 * resume after balance was interrupted.
2526 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2529 * Turn on soft mode for chunk types that were being converted.
2531 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2532 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2533 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2534 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2535 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2536 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2539 * Turn on usage filter if is not already used. The idea is
2540 * that chunks that we have already balanced should be
2541 * reasonably full. Don't do it for chunks that are being
2542 * converted - that will keep us from relocating unconverted
2543 * (albeit full) chunks.
2545 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2546 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2547 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2548 bctl
->data
.usage
= 90;
2550 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2551 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2552 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2553 bctl
->sys
.usage
= 90;
2555 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2556 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2557 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2558 bctl
->meta
.usage
= 90;
2563 * Should be called with both balance and volume mutexes held to
2564 * serialize other volume operations (add_dev/rm_dev/resize) with
2565 * restriper. Same goes for unset_balance_control.
2567 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2569 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2571 BUG_ON(fs_info
->balance_ctl
);
2573 spin_lock(&fs_info
->balance_lock
);
2574 fs_info
->balance_ctl
= bctl
;
2575 spin_unlock(&fs_info
->balance_lock
);
2578 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2580 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2582 BUG_ON(!fs_info
->balance_ctl
);
2584 spin_lock(&fs_info
->balance_lock
);
2585 fs_info
->balance_ctl
= NULL
;
2586 spin_unlock(&fs_info
->balance_lock
);
2592 * Balance filters. Return 1 if chunk should be filtered out
2593 * (should not be balanced).
2595 static int chunk_profiles_filter(u64 chunk_type
,
2596 struct btrfs_balance_args
*bargs
)
2598 chunk_type
= chunk_to_extended(chunk_type
) &
2599 BTRFS_EXTENDED_PROFILE_MASK
;
2601 if (bargs
->profiles
& chunk_type
)
2607 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2608 struct btrfs_balance_args
*bargs
)
2610 struct btrfs_block_group_cache
*cache
;
2611 u64 chunk_used
, user_thresh
;
2614 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2615 chunk_used
= btrfs_block_group_used(&cache
->item
);
2617 user_thresh
= div_factor_fine(cache
->key
.offset
, bargs
->usage
);
2618 if (chunk_used
< user_thresh
)
2621 btrfs_put_block_group(cache
);
2625 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2626 struct btrfs_chunk
*chunk
,
2627 struct btrfs_balance_args
*bargs
)
2629 struct btrfs_stripe
*stripe
;
2630 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2633 for (i
= 0; i
< num_stripes
; i
++) {
2634 stripe
= btrfs_stripe_nr(chunk
, i
);
2635 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2642 /* [pstart, pend) */
2643 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2644 struct btrfs_chunk
*chunk
,
2646 struct btrfs_balance_args
*bargs
)
2648 struct btrfs_stripe
*stripe
;
2649 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2655 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2658 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2659 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
))
2663 factor
= num_stripes
/ factor
;
2665 for (i
= 0; i
< num_stripes
; i
++) {
2666 stripe
= btrfs_stripe_nr(chunk
, i
);
2667 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2670 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2671 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2672 do_div(stripe_length
, factor
);
2674 if (stripe_offset
< bargs
->pend
&&
2675 stripe_offset
+ stripe_length
> bargs
->pstart
)
2682 /* [vstart, vend) */
2683 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2684 struct btrfs_chunk
*chunk
,
2686 struct btrfs_balance_args
*bargs
)
2688 if (chunk_offset
< bargs
->vend
&&
2689 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2690 /* at least part of the chunk is inside this vrange */
2696 static int chunk_soft_convert_filter(u64 chunk_type
,
2697 struct btrfs_balance_args
*bargs
)
2699 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2702 chunk_type
= chunk_to_extended(chunk_type
) &
2703 BTRFS_EXTENDED_PROFILE_MASK
;
2705 if (bargs
->target
== chunk_type
)
2711 static int should_balance_chunk(struct btrfs_root
*root
,
2712 struct extent_buffer
*leaf
,
2713 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2715 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2716 struct btrfs_balance_args
*bargs
= NULL
;
2717 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2720 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2721 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2725 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2726 bargs
= &bctl
->data
;
2727 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2729 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2730 bargs
= &bctl
->meta
;
2732 /* profiles filter */
2733 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2734 chunk_profiles_filter(chunk_type
, bargs
)) {
2739 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2740 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2745 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2746 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2750 /* drange filter, makes sense only with devid filter */
2751 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2752 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2757 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2758 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2762 /* soft profile changing mode */
2763 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2764 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2771 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2773 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2774 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2775 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2776 struct list_head
*devices
;
2777 struct btrfs_device
*device
;
2780 struct btrfs_chunk
*chunk
;
2781 struct btrfs_path
*path
;
2782 struct btrfs_key key
;
2783 struct btrfs_key found_key
;
2784 struct btrfs_trans_handle
*trans
;
2785 struct extent_buffer
*leaf
;
2788 int enospc_errors
= 0;
2789 bool counting
= true;
2791 /* step one make some room on all the devices */
2792 devices
= &fs_info
->fs_devices
->devices
;
2793 list_for_each_entry(device
, devices
, dev_list
) {
2794 old_size
= device
->total_bytes
;
2795 size_to_free
= div_factor(old_size
, 1);
2796 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2797 if (!device
->writeable
||
2798 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2799 device
->is_tgtdev_for_dev_replace
)
2802 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2807 trans
= btrfs_start_transaction(dev_root
, 0);
2808 BUG_ON(IS_ERR(trans
));
2810 ret
= btrfs_grow_device(trans
, device
, old_size
);
2813 btrfs_end_transaction(trans
, dev_root
);
2816 /* step two, relocate all the chunks */
2817 path
= btrfs_alloc_path();
2823 /* zero out stat counters */
2824 spin_lock(&fs_info
->balance_lock
);
2825 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2826 spin_unlock(&fs_info
->balance_lock
);
2828 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2829 key
.offset
= (u64
)-1;
2830 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2833 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2834 atomic_read(&fs_info
->balance_cancel_req
)) {
2839 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2844 * this shouldn't happen, it means the last relocate
2848 BUG(); /* FIXME break ? */
2850 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2851 BTRFS_CHUNK_ITEM_KEY
);
2857 leaf
= path
->nodes
[0];
2858 slot
= path
->slots
[0];
2859 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2861 if (found_key
.objectid
!= key
.objectid
)
2864 /* chunk zero is special */
2865 if (found_key
.offset
== 0)
2868 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2871 spin_lock(&fs_info
->balance_lock
);
2872 bctl
->stat
.considered
++;
2873 spin_unlock(&fs_info
->balance_lock
);
2876 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2878 btrfs_release_path(path
);
2883 spin_lock(&fs_info
->balance_lock
);
2884 bctl
->stat
.expected
++;
2885 spin_unlock(&fs_info
->balance_lock
);
2889 ret
= btrfs_relocate_chunk(chunk_root
,
2890 chunk_root
->root_key
.objectid
,
2893 if (ret
&& ret
!= -ENOSPC
)
2895 if (ret
== -ENOSPC
) {
2898 spin_lock(&fs_info
->balance_lock
);
2899 bctl
->stat
.completed
++;
2900 spin_unlock(&fs_info
->balance_lock
);
2903 key
.offset
= found_key
.offset
- 1;
2907 btrfs_release_path(path
);
2912 btrfs_free_path(path
);
2913 if (enospc_errors
) {
2914 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
2924 * alloc_profile_is_valid - see if a given profile is valid and reduced
2925 * @flags: profile to validate
2926 * @extended: if true @flags is treated as an extended profile
2928 static int alloc_profile_is_valid(u64 flags
, int extended
)
2930 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
2931 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
2933 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
2935 /* 1) check that all other bits are zeroed */
2939 /* 2) see if profile is reduced */
2941 return !extended
; /* "0" is valid for usual profiles */
2943 /* true if exactly one bit set */
2944 return (flags
& (flags
- 1)) == 0;
2947 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
2949 /* cancel requested || normal exit path */
2950 return atomic_read(&fs_info
->balance_cancel_req
) ||
2951 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
2952 atomic_read(&fs_info
->balance_cancel_req
) == 0);
2955 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
2959 unset_balance_control(fs_info
);
2960 ret
= del_balance_item(fs_info
->tree_root
);
2963 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
2966 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
2967 struct btrfs_ioctl_balance_args
*bargs
);
2970 * Should be called with both balance and volume mutexes held
2972 int btrfs_balance(struct btrfs_balance_control
*bctl
,
2973 struct btrfs_ioctl_balance_args
*bargs
)
2975 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2981 if (btrfs_fs_closing(fs_info
) ||
2982 atomic_read(&fs_info
->balance_pause_req
) ||
2983 atomic_read(&fs_info
->balance_cancel_req
)) {
2988 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
2989 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
2993 * In case of mixed groups both data and meta should be picked,
2994 * and identical options should be given for both of them.
2996 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
2997 if (mixed
&& (bctl
->flags
& allowed
)) {
2998 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
2999 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3000 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3001 printk(KERN_ERR
"btrfs: with mixed groups data and "
3002 "metadata balance options must be the same\n");
3008 num_devices
= fs_info
->fs_devices
->num_devices
;
3009 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3010 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3011 BUG_ON(num_devices
< 1);
3014 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3015 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3016 if (num_devices
== 1)
3017 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3018 else if (num_devices
< 4)
3019 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3021 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
3022 BTRFS_BLOCK_GROUP_RAID10
);
3024 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3025 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3026 (bctl
->data
.target
& ~allowed
))) {
3027 printk(KERN_ERR
"btrfs: unable to start balance with target "
3028 "data profile %llu\n",
3029 (unsigned long long)bctl
->data
.target
);
3033 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3034 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3035 (bctl
->meta
.target
& ~allowed
))) {
3036 printk(KERN_ERR
"btrfs: unable to start balance with target "
3037 "metadata profile %llu\n",
3038 (unsigned long long)bctl
->meta
.target
);
3042 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3043 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3044 (bctl
->sys
.target
& ~allowed
))) {
3045 printk(KERN_ERR
"btrfs: unable to start balance with target "
3046 "system profile %llu\n",
3047 (unsigned long long)bctl
->sys
.target
);
3052 /* allow dup'ed data chunks only in mixed mode */
3053 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3054 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3055 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
3060 /* allow to reduce meta or sys integrity only if force set */
3061 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3062 BTRFS_BLOCK_GROUP_RAID10
;
3063 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3064 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3065 !(bctl
->sys
.target
& allowed
)) ||
3066 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3067 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3068 !(bctl
->meta
.target
& allowed
))) {
3069 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3070 printk(KERN_INFO
"btrfs: force reducing metadata "
3073 printk(KERN_ERR
"btrfs: balance will reduce metadata "
3074 "integrity, use force if you want this\n");
3080 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3081 int num_tolerated_disk_barrier_failures
;
3082 u64 target
= bctl
->sys
.target
;
3084 num_tolerated_disk_barrier_failures
=
3085 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3086 if (num_tolerated_disk_barrier_failures
> 0 &&
3088 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3089 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3090 num_tolerated_disk_barrier_failures
= 0;
3091 else if (num_tolerated_disk_barrier_failures
> 1 &&
3093 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3094 num_tolerated_disk_barrier_failures
= 1;
3096 fs_info
->num_tolerated_disk_barrier_failures
=
3097 num_tolerated_disk_barrier_failures
;
3100 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3101 if (ret
&& ret
!= -EEXIST
)
3104 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3105 BUG_ON(ret
== -EEXIST
);
3106 set_balance_control(bctl
);
3108 BUG_ON(ret
!= -EEXIST
);
3109 spin_lock(&fs_info
->balance_lock
);
3110 update_balance_args(bctl
);
3111 spin_unlock(&fs_info
->balance_lock
);
3114 atomic_inc(&fs_info
->balance_running
);
3115 mutex_unlock(&fs_info
->balance_mutex
);
3117 ret
= __btrfs_balance(fs_info
);
3119 mutex_lock(&fs_info
->balance_mutex
);
3120 atomic_dec(&fs_info
->balance_running
);
3123 memset(bargs
, 0, sizeof(*bargs
));
3124 update_ioctl_balance_args(fs_info
, 0, bargs
);
3127 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3128 balance_need_close(fs_info
)) {
3129 __cancel_balance(fs_info
);
3132 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3133 fs_info
->num_tolerated_disk_barrier_failures
=
3134 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3137 wake_up(&fs_info
->balance_wait_q
);
3141 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3142 __cancel_balance(fs_info
);
3145 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3150 static int balance_kthread(void *data
)
3152 struct btrfs_fs_info
*fs_info
= data
;
3155 mutex_lock(&fs_info
->volume_mutex
);
3156 mutex_lock(&fs_info
->balance_mutex
);
3158 if (fs_info
->balance_ctl
) {
3159 printk(KERN_INFO
"btrfs: continuing balance\n");
3160 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3163 mutex_unlock(&fs_info
->balance_mutex
);
3164 mutex_unlock(&fs_info
->volume_mutex
);
3169 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3171 struct task_struct
*tsk
;
3173 spin_lock(&fs_info
->balance_lock
);
3174 if (!fs_info
->balance_ctl
) {
3175 spin_unlock(&fs_info
->balance_lock
);
3178 spin_unlock(&fs_info
->balance_lock
);
3180 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3181 printk(KERN_INFO
"btrfs: force skipping balance\n");
3185 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3187 return PTR_ERR(tsk
);
3192 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3194 struct btrfs_balance_control
*bctl
;
3195 struct btrfs_balance_item
*item
;
3196 struct btrfs_disk_balance_args disk_bargs
;
3197 struct btrfs_path
*path
;
3198 struct extent_buffer
*leaf
;
3199 struct btrfs_key key
;
3202 path
= btrfs_alloc_path();
3206 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3207 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3210 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3213 if (ret
> 0) { /* ret = -ENOENT; */
3218 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3224 leaf
= path
->nodes
[0];
3225 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3227 bctl
->fs_info
= fs_info
;
3228 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3229 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3231 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3232 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3233 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3234 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3235 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3236 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3238 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3240 mutex_lock(&fs_info
->volume_mutex
);
3241 mutex_lock(&fs_info
->balance_mutex
);
3243 set_balance_control(bctl
);
3245 mutex_unlock(&fs_info
->balance_mutex
);
3246 mutex_unlock(&fs_info
->volume_mutex
);
3248 btrfs_free_path(path
);
3252 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3256 mutex_lock(&fs_info
->balance_mutex
);
3257 if (!fs_info
->balance_ctl
) {
3258 mutex_unlock(&fs_info
->balance_mutex
);
3262 if (atomic_read(&fs_info
->balance_running
)) {
3263 atomic_inc(&fs_info
->balance_pause_req
);
3264 mutex_unlock(&fs_info
->balance_mutex
);
3266 wait_event(fs_info
->balance_wait_q
,
3267 atomic_read(&fs_info
->balance_running
) == 0);
3269 mutex_lock(&fs_info
->balance_mutex
);
3270 /* we are good with balance_ctl ripped off from under us */
3271 BUG_ON(atomic_read(&fs_info
->balance_running
));
3272 atomic_dec(&fs_info
->balance_pause_req
);
3277 mutex_unlock(&fs_info
->balance_mutex
);
3281 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3283 mutex_lock(&fs_info
->balance_mutex
);
3284 if (!fs_info
->balance_ctl
) {
3285 mutex_unlock(&fs_info
->balance_mutex
);
3289 atomic_inc(&fs_info
->balance_cancel_req
);
3291 * if we are running just wait and return, balance item is
3292 * deleted in btrfs_balance in this case
3294 if (atomic_read(&fs_info
->balance_running
)) {
3295 mutex_unlock(&fs_info
->balance_mutex
);
3296 wait_event(fs_info
->balance_wait_q
,
3297 atomic_read(&fs_info
->balance_running
) == 0);
3298 mutex_lock(&fs_info
->balance_mutex
);
3300 /* __cancel_balance needs volume_mutex */
3301 mutex_unlock(&fs_info
->balance_mutex
);
3302 mutex_lock(&fs_info
->volume_mutex
);
3303 mutex_lock(&fs_info
->balance_mutex
);
3305 if (fs_info
->balance_ctl
)
3306 __cancel_balance(fs_info
);
3308 mutex_unlock(&fs_info
->volume_mutex
);
3311 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3312 atomic_dec(&fs_info
->balance_cancel_req
);
3313 mutex_unlock(&fs_info
->balance_mutex
);
3318 * shrinking a device means finding all of the device extents past
3319 * the new size, and then following the back refs to the chunks.
3320 * The chunk relocation code actually frees the device extent
3322 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3324 struct btrfs_trans_handle
*trans
;
3325 struct btrfs_root
*root
= device
->dev_root
;
3326 struct btrfs_dev_extent
*dev_extent
= NULL
;
3327 struct btrfs_path
*path
;
3335 bool retried
= false;
3336 struct extent_buffer
*l
;
3337 struct btrfs_key key
;
3338 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3339 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3340 u64 old_size
= device
->total_bytes
;
3341 u64 diff
= device
->total_bytes
- new_size
;
3343 if (device
->is_tgtdev_for_dev_replace
)
3346 path
= btrfs_alloc_path();
3354 device
->total_bytes
= new_size
;
3355 if (device
->writeable
) {
3356 device
->fs_devices
->total_rw_bytes
-= diff
;
3357 spin_lock(&root
->fs_info
->free_chunk_lock
);
3358 root
->fs_info
->free_chunk_space
-= diff
;
3359 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3361 unlock_chunks(root
);
3364 key
.objectid
= device
->devid
;
3365 key
.offset
= (u64
)-1;
3366 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3369 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3373 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3378 btrfs_release_path(path
);
3383 slot
= path
->slots
[0];
3384 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3386 if (key
.objectid
!= device
->devid
) {
3387 btrfs_release_path(path
);
3391 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3392 length
= btrfs_dev_extent_length(l
, dev_extent
);
3394 if (key
.offset
+ length
<= new_size
) {
3395 btrfs_release_path(path
);
3399 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3400 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3401 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3402 btrfs_release_path(path
);
3404 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3406 if (ret
&& ret
!= -ENOSPC
)
3410 } while (key
.offset
-- > 0);
3412 if (failed
&& !retried
) {
3416 } else if (failed
&& retried
) {
3420 device
->total_bytes
= old_size
;
3421 if (device
->writeable
)
3422 device
->fs_devices
->total_rw_bytes
+= diff
;
3423 spin_lock(&root
->fs_info
->free_chunk_lock
);
3424 root
->fs_info
->free_chunk_space
+= diff
;
3425 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3426 unlock_chunks(root
);
3430 /* Shrinking succeeded, else we would be at "done". */
3431 trans
= btrfs_start_transaction(root
, 0);
3432 if (IS_ERR(trans
)) {
3433 ret
= PTR_ERR(trans
);
3439 device
->disk_total_bytes
= new_size
;
3440 /* Now btrfs_update_device() will change the on-disk size. */
3441 ret
= btrfs_update_device(trans
, device
);
3443 unlock_chunks(root
);
3444 btrfs_end_transaction(trans
, root
);
3447 WARN_ON(diff
> old_total
);
3448 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3449 unlock_chunks(root
);
3450 btrfs_end_transaction(trans
, root
);
3452 btrfs_free_path(path
);
3456 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3457 struct btrfs_key
*key
,
3458 struct btrfs_chunk
*chunk
, int item_size
)
3460 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3461 struct btrfs_disk_key disk_key
;
3465 array_size
= btrfs_super_sys_array_size(super_copy
);
3466 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3469 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3470 btrfs_cpu_key_to_disk(&disk_key
, key
);
3471 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3472 ptr
+= sizeof(disk_key
);
3473 memcpy(ptr
, chunk
, item_size
);
3474 item_size
+= sizeof(disk_key
);
3475 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3480 * sort the devices in descending order by max_avail, total_avail
3482 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3484 const struct btrfs_device_info
*di_a
= a
;
3485 const struct btrfs_device_info
*di_b
= b
;
3487 if (di_a
->max_avail
> di_b
->max_avail
)
3489 if (di_a
->max_avail
< di_b
->max_avail
)
3491 if (di_a
->total_avail
> di_b
->total_avail
)
3493 if (di_a
->total_avail
< di_b
->total_avail
)
3498 struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3499 { 2, 1, 0, 4, 2, 2 /* raid10 */ },
3500 { 1, 1, 2, 2, 2, 2 /* raid1 */ },
3501 { 1, 2, 1, 1, 1, 2 /* dup */ },
3502 { 1, 1, 0, 2, 1, 1 /* raid0 */ },
3503 { 1, 1, 0, 1, 1, 1 /* single */ },
3506 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3507 struct btrfs_root
*extent_root
,
3508 struct map_lookup
**map_ret
,
3509 u64
*num_bytes_out
, u64
*stripe_size_out
,
3510 u64 start
, u64 type
)
3512 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3513 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3514 struct list_head
*cur
;
3515 struct map_lookup
*map
= NULL
;
3516 struct extent_map_tree
*em_tree
;
3517 struct extent_map
*em
;
3518 struct btrfs_device_info
*devices_info
= NULL
;
3520 int num_stripes
; /* total number of stripes to allocate */
3521 int sub_stripes
; /* sub_stripes info for map */
3522 int dev_stripes
; /* stripes per dev */
3523 int devs_max
; /* max devs to use */
3524 int devs_min
; /* min devs needed */
3525 int devs_increment
; /* ndevs has to be a multiple of this */
3526 int ncopies
; /* how many copies to data has */
3528 u64 max_stripe_size
;
3537 BUG_ON(!alloc_profile_is_valid(type
, 0));
3539 if (list_empty(&fs_devices
->alloc_list
))
3542 index
= __get_raid_index(type
);
3544 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
3545 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
3546 devs_max
= btrfs_raid_array
[index
].devs_max
;
3547 devs_min
= btrfs_raid_array
[index
].devs_min
;
3548 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
3549 ncopies
= btrfs_raid_array
[index
].ncopies
;
3551 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3552 max_stripe_size
= 1024 * 1024 * 1024;
3553 max_chunk_size
= 10 * max_stripe_size
;
3554 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3555 /* for larger filesystems, use larger metadata chunks */
3556 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3557 max_stripe_size
= 1024 * 1024 * 1024;
3559 max_stripe_size
= 256 * 1024 * 1024;
3560 max_chunk_size
= max_stripe_size
;
3561 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3562 max_stripe_size
= 32 * 1024 * 1024;
3563 max_chunk_size
= 2 * max_stripe_size
;
3565 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3570 /* we don't want a chunk larger than 10% of writeable space */
3571 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3574 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3579 cur
= fs_devices
->alloc_list
.next
;
3582 * in the first pass through the devices list, we gather information
3583 * about the available holes on each device.
3586 while (cur
!= &fs_devices
->alloc_list
) {
3587 struct btrfs_device
*device
;
3591 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3595 if (!device
->writeable
) {
3597 "btrfs: read-only device in alloc_list\n");
3601 if (!device
->in_fs_metadata
||
3602 device
->is_tgtdev_for_dev_replace
)
3605 if (device
->total_bytes
> device
->bytes_used
)
3606 total_avail
= device
->total_bytes
- device
->bytes_used
;
3610 /* If there is no space on this device, skip it. */
3611 if (total_avail
== 0)
3614 ret
= find_free_dev_extent(device
,
3615 max_stripe_size
* dev_stripes
,
3616 &dev_offset
, &max_avail
);
3617 if (ret
&& ret
!= -ENOSPC
)
3621 max_avail
= max_stripe_size
* dev_stripes
;
3623 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3626 devices_info
[ndevs
].dev_offset
= dev_offset
;
3627 devices_info
[ndevs
].max_avail
= max_avail
;
3628 devices_info
[ndevs
].total_avail
= total_avail
;
3629 devices_info
[ndevs
].dev
= device
;
3631 WARN_ON(ndevs
> fs_devices
->rw_devices
);
3635 * now sort the devices by hole size / available space
3637 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3638 btrfs_cmp_device_info
, NULL
);
3640 /* round down to number of usable stripes */
3641 ndevs
-= ndevs
% devs_increment
;
3643 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3648 if (devs_max
&& ndevs
> devs_max
)
3651 * the primary goal is to maximize the number of stripes, so use as many
3652 * devices as possible, even if the stripes are not maximum sized.
3654 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3655 num_stripes
= ndevs
* dev_stripes
;
3657 if (stripe_size
* ndevs
> max_chunk_size
* ncopies
) {
3658 stripe_size
= max_chunk_size
* ncopies
;
3659 do_div(stripe_size
, ndevs
);
3662 do_div(stripe_size
, dev_stripes
);
3664 /* align to BTRFS_STRIPE_LEN */
3665 do_div(stripe_size
, BTRFS_STRIPE_LEN
);
3666 stripe_size
*= BTRFS_STRIPE_LEN
;
3668 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3673 map
->num_stripes
= num_stripes
;
3675 for (i
= 0; i
< ndevs
; ++i
) {
3676 for (j
= 0; j
< dev_stripes
; ++j
) {
3677 int s
= i
* dev_stripes
+ j
;
3678 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3679 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3683 map
->sector_size
= extent_root
->sectorsize
;
3684 map
->stripe_len
= BTRFS_STRIPE_LEN
;
3685 map
->io_align
= BTRFS_STRIPE_LEN
;
3686 map
->io_width
= BTRFS_STRIPE_LEN
;
3688 map
->sub_stripes
= sub_stripes
;
3691 num_bytes
= stripe_size
* (num_stripes
/ ncopies
);
3693 *stripe_size_out
= stripe_size
;
3694 *num_bytes_out
= num_bytes
;
3696 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3698 em
= alloc_extent_map();
3703 em
->bdev
= (struct block_device
*)map
;
3705 em
->len
= num_bytes
;
3706 em
->block_start
= 0;
3707 em
->block_len
= em
->len
;
3709 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3710 write_lock(&em_tree
->lock
);
3711 ret
= add_extent_mapping(em_tree
, em
);
3712 write_unlock(&em_tree
->lock
);
3713 free_extent_map(em
);
3717 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3718 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3723 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3724 struct btrfs_device
*device
;
3727 device
= map
->stripes
[i
].dev
;
3728 dev_offset
= map
->stripes
[i
].physical
;
3730 ret
= btrfs_alloc_dev_extent(trans
, device
,
3731 info
->chunk_root
->root_key
.objectid
,
3732 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3733 start
, dev_offset
, stripe_size
);
3735 btrfs_abort_transaction(trans
, extent_root
, ret
);
3740 kfree(devices_info
);
3745 kfree(devices_info
);
3749 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3750 struct btrfs_root
*extent_root
,
3751 struct map_lookup
*map
, u64 chunk_offset
,
3752 u64 chunk_size
, u64 stripe_size
)
3755 struct btrfs_key key
;
3756 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3757 struct btrfs_device
*device
;
3758 struct btrfs_chunk
*chunk
;
3759 struct btrfs_stripe
*stripe
;
3760 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3764 chunk
= kzalloc(item_size
, GFP_NOFS
);
3769 while (index
< map
->num_stripes
) {
3770 device
= map
->stripes
[index
].dev
;
3771 device
->bytes_used
+= stripe_size
;
3772 ret
= btrfs_update_device(trans
, device
);
3778 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
3779 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
3781 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
3784 stripe
= &chunk
->stripe
;
3785 while (index
< map
->num_stripes
) {
3786 device
= map
->stripes
[index
].dev
;
3787 dev_offset
= map
->stripes
[index
].physical
;
3789 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
3790 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
3791 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
3796 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
3797 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
3798 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
3799 btrfs_set_stack_chunk_type(chunk
, map
->type
);
3800 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
3801 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
3802 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
3803 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
3804 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
3806 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3807 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3808 key
.offset
= chunk_offset
;
3810 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
3812 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3814 * TODO: Cleanup of inserted chunk root in case of
3817 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
3827 * Chunk allocation falls into two parts. The first part does works
3828 * that make the new allocated chunk useable, but not do any operation
3829 * that modifies the chunk tree. The second part does the works that
3830 * require modifying the chunk tree. This division is important for the
3831 * bootstrap process of adding storage to a seed btrfs.
3833 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3834 struct btrfs_root
*extent_root
, u64 type
)
3839 struct map_lookup
*map
;
3840 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3843 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3848 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3849 &stripe_size
, chunk_offset
, type
);
3853 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3854 chunk_size
, stripe_size
);
3860 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
3861 struct btrfs_root
*root
,
3862 struct btrfs_device
*device
)
3865 u64 sys_chunk_offset
;
3869 u64 sys_stripe_size
;
3871 struct map_lookup
*map
;
3872 struct map_lookup
*sys_map
;
3873 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3874 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3877 ret
= find_next_chunk(fs_info
->chunk_root
,
3878 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
3882 alloc_profile
= BTRFS_BLOCK_GROUP_METADATA
|
3883 fs_info
->avail_metadata_alloc_bits
;
3884 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3886 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3887 &stripe_size
, chunk_offset
, alloc_profile
);
3891 sys_chunk_offset
= chunk_offset
+ chunk_size
;
3893 alloc_profile
= BTRFS_BLOCK_GROUP_SYSTEM
|
3894 fs_info
->avail_system_alloc_bits
;
3895 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3897 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
3898 &sys_chunk_size
, &sys_stripe_size
,
3899 sys_chunk_offset
, alloc_profile
);
3901 btrfs_abort_transaction(trans
, root
, ret
);
3905 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
3907 btrfs_abort_transaction(trans
, root
, ret
);
3912 * Modifying chunk tree needs allocating new blocks from both
3913 * system block group and metadata block group. So we only can
3914 * do operations require modifying the chunk tree after both
3915 * block groups were created.
3917 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3918 chunk_size
, stripe_size
);
3920 btrfs_abort_transaction(trans
, root
, ret
);
3924 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
3925 sys_chunk_offset
, sys_chunk_size
,
3928 btrfs_abort_transaction(trans
, root
, ret
);
3935 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
3937 struct extent_map
*em
;
3938 struct map_lookup
*map
;
3939 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
3943 read_lock(&map_tree
->map_tree
.lock
);
3944 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
3945 read_unlock(&map_tree
->map_tree
.lock
);
3949 if (btrfs_test_opt(root
, DEGRADED
)) {
3950 free_extent_map(em
);
3954 map
= (struct map_lookup
*)em
->bdev
;
3955 for (i
= 0; i
< map
->num_stripes
; i
++) {
3956 if (!map
->stripes
[i
].dev
->writeable
) {
3961 free_extent_map(em
);
3965 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
3967 extent_map_tree_init(&tree
->map_tree
);
3970 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
3972 struct extent_map
*em
;
3975 write_lock(&tree
->map_tree
.lock
);
3976 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
3978 remove_extent_mapping(&tree
->map_tree
, em
);
3979 write_unlock(&tree
->map_tree
.lock
);
3984 free_extent_map(em
);
3985 /* once for the tree */
3986 free_extent_map(em
);
3990 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
3992 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
3993 struct extent_map
*em
;
3994 struct map_lookup
*map
;
3995 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3998 read_lock(&em_tree
->lock
);
3999 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4000 read_unlock(&em_tree
->lock
);
4003 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4004 map
= (struct map_lookup
*)em
->bdev
;
4005 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4006 ret
= map
->num_stripes
;
4007 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4008 ret
= map
->sub_stripes
;
4011 free_extent_map(em
);
4013 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4014 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4016 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4021 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4022 struct map_lookup
*map
, int first
, int num
,
4023 int optimal
, int dev_replace_is_ongoing
)
4027 struct btrfs_device
*srcdev
;
4029 if (dev_replace_is_ongoing
&&
4030 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4031 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4032 srcdev
= fs_info
->dev_replace
.srcdev
;
4037 * try to avoid the drive that is the source drive for a
4038 * dev-replace procedure, only choose it if no other non-missing
4039 * mirror is available
4041 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4042 if (map
->stripes
[optimal
].dev
->bdev
&&
4043 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4045 for (i
= first
; i
< first
+ num
; i
++) {
4046 if (map
->stripes
[i
].dev
->bdev
&&
4047 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4052 /* we couldn't find one that doesn't fail. Just return something
4053 * and the io error handling code will clean up eventually
4058 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4059 u64 logical
, u64
*length
,
4060 struct btrfs_bio
**bbio_ret
,
4063 struct extent_map
*em
;
4064 struct map_lookup
*map
;
4065 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4066 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4069 u64 stripe_end_offset
;
4078 struct btrfs_bio
*bbio
= NULL
;
4079 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4080 int dev_replace_is_ongoing
= 0;
4081 int num_alloc_stripes
;
4082 int patch_the_first_stripe_for_dev_replace
= 0;
4083 u64 physical_to_patch_in_first_stripe
= 0;
4085 read_lock(&em_tree
->lock
);
4086 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4087 read_unlock(&em_tree
->lock
);
4090 printk(KERN_CRIT
"btrfs: unable to find logical %llu len %llu\n",
4091 (unsigned long long)logical
,
4092 (unsigned long long)*length
);
4096 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4097 map
= (struct map_lookup
*)em
->bdev
;
4098 offset
= logical
- em
->start
;
4102 * stripe_nr counts the total number of stripes we have to stride
4103 * to get to this block
4105 do_div(stripe_nr
, map
->stripe_len
);
4107 stripe_offset
= stripe_nr
* map
->stripe_len
;
4108 BUG_ON(offset
< stripe_offset
);
4110 /* stripe_offset is the offset of this block in its stripe*/
4111 stripe_offset
= offset
- stripe_offset
;
4113 if (rw
& REQ_DISCARD
)
4114 *length
= min_t(u64
, em
->len
- offset
, *length
);
4115 else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4116 /* we limit the length of each bio to what fits in a stripe */
4117 *length
= min_t(u64
, em
->len
- offset
,
4118 map
->stripe_len
- stripe_offset
);
4120 *length
= em
->len
- offset
;
4126 btrfs_dev_replace_lock(dev_replace
);
4127 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4128 if (!dev_replace_is_ongoing
)
4129 btrfs_dev_replace_unlock(dev_replace
);
4131 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4132 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4133 dev_replace
->tgtdev
!= NULL
) {
4135 * in dev-replace case, for repair case (that's the only
4136 * case where the mirror is selected explicitly when
4137 * calling btrfs_map_block), blocks left of the left cursor
4138 * can also be read from the target drive.
4139 * For REQ_GET_READ_MIRRORS, the target drive is added as
4140 * the last one to the array of stripes. For READ, it also
4141 * needs to be supported using the same mirror number.
4142 * If the requested block is not left of the left cursor,
4143 * EIO is returned. This can happen because btrfs_num_copies()
4144 * returns one more in the dev-replace case.
4146 u64 tmp_length
= *length
;
4147 struct btrfs_bio
*tmp_bbio
= NULL
;
4148 int tmp_num_stripes
;
4149 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4150 int index_srcdev
= 0;
4152 u64 physical_of_found
= 0;
4154 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4155 logical
, &tmp_length
, &tmp_bbio
, 0);
4157 WARN_ON(tmp_bbio
!= NULL
);
4161 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4162 if (mirror_num
> tmp_num_stripes
) {
4164 * REQ_GET_READ_MIRRORS does not contain this
4165 * mirror, that means that the requested area
4166 * is not left of the left cursor
4174 * process the rest of the function using the mirror_num
4175 * of the source drive. Therefore look it up first.
4176 * At the end, patch the device pointer to the one of the
4179 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4180 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4182 * In case of DUP, in order to keep it
4183 * simple, only add the mirror with the
4184 * lowest physical address
4187 physical_of_found
<=
4188 tmp_bbio
->stripes
[i
].physical
)
4193 tmp_bbio
->stripes
[i
].physical
;
4198 mirror_num
= index_srcdev
+ 1;
4199 patch_the_first_stripe_for_dev_replace
= 1;
4200 physical_to_patch_in_first_stripe
= physical_of_found
;
4209 } else if (mirror_num
> map
->num_stripes
) {
4215 stripe_nr_orig
= stripe_nr
;
4216 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
4217 (~(map
->stripe_len
- 1));
4218 do_div(stripe_nr_end
, map
->stripe_len
);
4219 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4221 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4222 if (rw
& REQ_DISCARD
)
4223 num_stripes
= min_t(u64
, map
->num_stripes
,
4224 stripe_nr_end
- stripe_nr_orig
);
4225 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4226 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4227 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4228 num_stripes
= map
->num_stripes
;
4229 else if (mirror_num
)
4230 stripe_index
= mirror_num
- 1;
4232 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4234 current
->pid
% map
->num_stripes
,
4235 dev_replace_is_ongoing
);
4236 mirror_num
= stripe_index
+ 1;
4239 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4240 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4241 num_stripes
= map
->num_stripes
;
4242 } else if (mirror_num
) {
4243 stripe_index
= mirror_num
- 1;
4248 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4249 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4251 stripe_index
= do_div(stripe_nr
, factor
);
4252 stripe_index
*= map
->sub_stripes
;
4254 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4255 num_stripes
= map
->sub_stripes
;
4256 else if (rw
& REQ_DISCARD
)
4257 num_stripes
= min_t(u64
, map
->sub_stripes
*
4258 (stripe_nr_end
- stripe_nr_orig
),
4260 else if (mirror_num
)
4261 stripe_index
+= mirror_num
- 1;
4263 int old_stripe_index
= stripe_index
;
4264 stripe_index
= find_live_mirror(fs_info
, map
,
4266 map
->sub_stripes
, stripe_index
+
4267 current
->pid
% map
->sub_stripes
,
4268 dev_replace_is_ongoing
);
4269 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4273 * after this do_div call, stripe_nr is the number of stripes
4274 * on this device we have to walk to find the data, and
4275 * stripe_index is the number of our device in the stripe array
4277 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4278 mirror_num
= stripe_index
+ 1;
4280 BUG_ON(stripe_index
>= map
->num_stripes
);
4282 num_alloc_stripes
= num_stripes
;
4283 if (dev_replace_is_ongoing
) {
4284 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
4285 num_alloc_stripes
<<= 1;
4286 if (rw
& REQ_GET_READ_MIRRORS
)
4287 num_alloc_stripes
++;
4289 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
4294 atomic_set(&bbio
->error
, 0);
4296 if (rw
& REQ_DISCARD
) {
4298 int sub_stripes
= 0;
4299 u64 stripes_per_dev
= 0;
4300 u32 remaining_stripes
= 0;
4301 u32 last_stripe
= 0;
4304 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
4305 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4308 sub_stripes
= map
->sub_stripes
;
4310 factor
= map
->num_stripes
/ sub_stripes
;
4311 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
4314 &remaining_stripes
);
4315 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
4316 last_stripe
*= sub_stripes
;
4319 for (i
= 0; i
< num_stripes
; i
++) {
4320 bbio
->stripes
[i
].physical
=
4321 map
->stripes
[stripe_index
].physical
+
4322 stripe_offset
+ stripe_nr
* map
->stripe_len
;
4323 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
4325 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
4326 BTRFS_BLOCK_GROUP_RAID10
)) {
4327 bbio
->stripes
[i
].length
= stripes_per_dev
*
4330 if (i
/ sub_stripes
< remaining_stripes
)
4331 bbio
->stripes
[i
].length
+=
4335 * Special for the first stripe and
4338 * |-------|...|-------|
4342 if (i
< sub_stripes
)
4343 bbio
->stripes
[i
].length
-=
4346 if (stripe_index
>= last_stripe
&&
4347 stripe_index
<= (last_stripe
+
4349 bbio
->stripes
[i
].length
-=
4352 if (i
== sub_stripes
- 1)
4355 bbio
->stripes
[i
].length
= *length
;
4358 if (stripe_index
== map
->num_stripes
) {
4359 /* This could only happen for RAID0/10 */
4365 for (i
= 0; i
< num_stripes
; i
++) {
4366 bbio
->stripes
[i
].physical
=
4367 map
->stripes
[stripe_index
].physical
+
4369 stripe_nr
* map
->stripe_len
;
4370 bbio
->stripes
[i
].dev
=
4371 map
->stripes
[stripe_index
].dev
;
4376 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
4377 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4378 BTRFS_BLOCK_GROUP_RAID10
|
4379 BTRFS_BLOCK_GROUP_DUP
)) {
4384 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
4385 dev_replace
->tgtdev
!= NULL
) {
4386 int index_where_to_add
;
4387 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4390 * duplicate the write operations while the dev replace
4391 * procedure is running. Since the copying of the old disk
4392 * to the new disk takes place at run time while the
4393 * filesystem is mounted writable, the regular write
4394 * operations to the old disk have to be duplicated to go
4395 * to the new disk as well.
4396 * Note that device->missing is handled by the caller, and
4397 * that the write to the old disk is already set up in the
4400 index_where_to_add
= num_stripes
;
4401 for (i
= 0; i
< num_stripes
; i
++) {
4402 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4403 /* write to new disk, too */
4404 struct btrfs_bio_stripe
*new =
4405 bbio
->stripes
+ index_where_to_add
;
4406 struct btrfs_bio_stripe
*old
=
4409 new->physical
= old
->physical
;
4410 new->length
= old
->length
;
4411 new->dev
= dev_replace
->tgtdev
;
4412 index_where_to_add
++;
4416 num_stripes
= index_where_to_add
;
4417 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
4418 dev_replace
->tgtdev
!= NULL
) {
4419 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4420 int index_srcdev
= 0;
4422 u64 physical_of_found
= 0;
4425 * During the dev-replace procedure, the target drive can
4426 * also be used to read data in case it is needed to repair
4427 * a corrupt block elsewhere. This is possible if the
4428 * requested area is left of the left cursor. In this area,
4429 * the target drive is a full copy of the source drive.
4431 for (i
= 0; i
< num_stripes
; i
++) {
4432 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4434 * In case of DUP, in order to keep it
4435 * simple, only add the mirror with the
4436 * lowest physical address
4439 physical_of_found
<=
4440 bbio
->stripes
[i
].physical
)
4444 physical_of_found
= bbio
->stripes
[i
].physical
;
4448 u64 length
= map
->stripe_len
;
4450 if (physical_of_found
+ length
<=
4451 dev_replace
->cursor_left
) {
4452 struct btrfs_bio_stripe
*tgtdev_stripe
=
4453 bbio
->stripes
+ num_stripes
;
4455 tgtdev_stripe
->physical
= physical_of_found
;
4456 tgtdev_stripe
->length
=
4457 bbio
->stripes
[index_srcdev
].length
;
4458 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
4466 bbio
->num_stripes
= num_stripes
;
4467 bbio
->max_errors
= max_errors
;
4468 bbio
->mirror_num
= mirror_num
;
4471 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4472 * mirror_num == num_stripes + 1 && dev_replace target drive is
4473 * available as a mirror
4475 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
4476 WARN_ON(num_stripes
> 1);
4477 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
4478 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
4479 bbio
->mirror_num
= map
->num_stripes
+ 1;
4482 if (dev_replace_is_ongoing
)
4483 btrfs_dev_replace_unlock(dev_replace
);
4484 free_extent_map(em
);
4488 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4489 u64 logical
, u64
*length
,
4490 struct btrfs_bio
**bbio_ret
, int mirror_num
)
4492 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
4496 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
4497 u64 chunk_start
, u64 physical
, u64 devid
,
4498 u64
**logical
, int *naddrs
, int *stripe_len
)
4500 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4501 struct extent_map
*em
;
4502 struct map_lookup
*map
;
4509 read_lock(&em_tree
->lock
);
4510 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
4511 read_unlock(&em_tree
->lock
);
4513 BUG_ON(!em
|| em
->start
!= chunk_start
);
4514 map
= (struct map_lookup
*)em
->bdev
;
4517 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4518 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4519 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4520 do_div(length
, map
->num_stripes
);
4522 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4523 BUG_ON(!buf
); /* -ENOMEM */
4525 for (i
= 0; i
< map
->num_stripes
; i
++) {
4526 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4528 if (map
->stripes
[i
].physical
> physical
||
4529 map
->stripes
[i
].physical
+ length
<= physical
)
4532 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4533 do_div(stripe_nr
, map
->stripe_len
);
4535 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4536 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4537 do_div(stripe_nr
, map
->sub_stripes
);
4538 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4539 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4541 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
4542 WARN_ON(nr
>= map
->num_stripes
);
4543 for (j
= 0; j
< nr
; j
++) {
4544 if (buf
[j
] == bytenr
)
4548 WARN_ON(nr
>= map
->num_stripes
);
4555 *stripe_len
= map
->stripe_len
;
4557 free_extent_map(em
);
4561 static void *merge_stripe_index_into_bio_private(void *bi_private
,
4562 unsigned int stripe_index
)
4565 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4567 * The alternative solution (instead of stealing bits from the
4568 * pointer) would be to allocate an intermediate structure
4569 * that contains the old private pointer plus the stripe_index.
4571 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
4572 BUG_ON(stripe_index
> 3);
4573 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
4576 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
4578 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
4581 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
4583 return (unsigned int)((uintptr_t)bi_private
) & 3;
4586 static void btrfs_end_bio(struct bio
*bio
, int err
)
4588 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
4589 int is_orig_bio
= 0;
4592 atomic_inc(&bbio
->error
);
4593 if (err
== -EIO
|| err
== -EREMOTEIO
) {
4594 unsigned int stripe_index
=
4595 extract_stripe_index_from_bio_private(
4597 struct btrfs_device
*dev
;
4599 BUG_ON(stripe_index
>= bbio
->num_stripes
);
4600 dev
= bbio
->stripes
[stripe_index
].dev
;
4602 if (bio
->bi_rw
& WRITE
)
4603 btrfs_dev_stat_inc(dev
,
4604 BTRFS_DEV_STAT_WRITE_ERRS
);
4606 btrfs_dev_stat_inc(dev
,
4607 BTRFS_DEV_STAT_READ_ERRS
);
4608 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
4609 btrfs_dev_stat_inc(dev
,
4610 BTRFS_DEV_STAT_FLUSH_ERRS
);
4611 btrfs_dev_stat_print_on_error(dev
);
4616 if (bio
== bbio
->orig_bio
)
4619 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4622 bio
= bbio
->orig_bio
;
4624 bio
->bi_private
= bbio
->private;
4625 bio
->bi_end_io
= bbio
->end_io
;
4626 bio
->bi_bdev
= (struct block_device
*)
4627 (unsigned long)bbio
->mirror_num
;
4628 /* only send an error to the higher layers if it is
4629 * beyond the tolerance of the multi-bio
4631 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
4635 * this bio is actually up to date, we didn't
4636 * go over the max number of errors
4638 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
4643 bio_endio(bio
, err
);
4644 } else if (!is_orig_bio
) {
4649 struct async_sched
{
4652 struct btrfs_fs_info
*info
;
4653 struct btrfs_work work
;
4657 * see run_scheduled_bios for a description of why bios are collected for
4660 * This will add one bio to the pending list for a device and make sure
4661 * the work struct is scheduled.
4663 static noinline
void schedule_bio(struct btrfs_root
*root
,
4664 struct btrfs_device
*device
,
4665 int rw
, struct bio
*bio
)
4667 int should_queue
= 1;
4668 struct btrfs_pending_bios
*pending_bios
;
4670 /* don't bother with additional async steps for reads, right now */
4671 if (!(rw
& REQ_WRITE
)) {
4673 btrfsic_submit_bio(rw
, bio
);
4679 * nr_async_bios allows us to reliably return congestion to the
4680 * higher layers. Otherwise, the async bio makes it appear we have
4681 * made progress against dirty pages when we've really just put it
4682 * on a queue for later
4684 atomic_inc(&root
->fs_info
->nr_async_bios
);
4685 WARN_ON(bio
->bi_next
);
4686 bio
->bi_next
= NULL
;
4689 spin_lock(&device
->io_lock
);
4690 if (bio
->bi_rw
& REQ_SYNC
)
4691 pending_bios
= &device
->pending_sync_bios
;
4693 pending_bios
= &device
->pending_bios
;
4695 if (pending_bios
->tail
)
4696 pending_bios
->tail
->bi_next
= bio
;
4698 pending_bios
->tail
= bio
;
4699 if (!pending_bios
->head
)
4700 pending_bios
->head
= bio
;
4701 if (device
->running_pending
)
4704 spin_unlock(&device
->io_lock
);
4707 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
4711 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
4714 struct bio_vec
*prev
;
4715 struct request_queue
*q
= bdev_get_queue(bdev
);
4716 unsigned short max_sectors
= queue_max_sectors(q
);
4717 struct bvec_merge_data bvm
= {
4719 .bi_sector
= sector
,
4720 .bi_rw
= bio
->bi_rw
,
4723 if (bio
->bi_vcnt
== 0) {
4728 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
4729 if ((bio
->bi_size
>> 9) > max_sectors
)
4732 if (!q
->merge_bvec_fn
)
4735 bvm
.bi_size
= bio
->bi_size
- prev
->bv_len
;
4736 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
4741 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4742 struct bio
*bio
, u64 physical
, int dev_nr
,
4745 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
4747 bio
->bi_private
= bbio
;
4748 bio
->bi_private
= merge_stripe_index_into_bio_private(
4749 bio
->bi_private
, (unsigned int)dev_nr
);
4750 bio
->bi_end_io
= btrfs_end_bio
;
4751 bio
->bi_sector
= physical
>> 9;
4754 struct rcu_string
*name
;
4757 name
= rcu_dereference(dev
->name
);
4758 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4759 "(%s id %llu), size=%u\n", rw
,
4760 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
4761 name
->str
, dev
->devid
, bio
->bi_size
);
4765 bio
->bi_bdev
= dev
->bdev
;
4767 schedule_bio(root
, dev
, rw
, bio
);
4769 btrfsic_submit_bio(rw
, bio
);
4772 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4773 struct bio
*first_bio
, struct btrfs_device
*dev
,
4774 int dev_nr
, int rw
, int async
)
4776 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
4778 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
4779 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
4782 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
4786 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
4787 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
4788 bvec
->bv_offset
) < bvec
->bv_len
) {
4789 u64 len
= bio
->bi_size
;
4791 atomic_inc(&bbio
->stripes_pending
);
4792 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
4800 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
4804 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
4806 atomic_inc(&bbio
->error
);
4807 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4808 bio
->bi_private
= bbio
->private;
4809 bio
->bi_end_io
= bbio
->end_io
;
4810 bio
->bi_bdev
= (struct block_device
*)
4811 (unsigned long)bbio
->mirror_num
;
4812 bio
->bi_sector
= logical
>> 9;
4814 bio_endio(bio
, -EIO
);
4818 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
4819 int mirror_num
, int async_submit
)
4821 struct btrfs_device
*dev
;
4822 struct bio
*first_bio
= bio
;
4823 u64 logical
= (u64
)bio
->bi_sector
<< 9;
4829 struct btrfs_bio
*bbio
= NULL
;
4831 length
= bio
->bi_size
;
4832 map_length
= length
;
4834 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
4839 total_devs
= bbio
->num_stripes
;
4840 if (map_length
< length
) {
4841 printk(KERN_CRIT
"btrfs: mapping failed logical %llu bio len %llu "
4842 "len %llu\n", (unsigned long long)logical
,
4843 (unsigned long long)length
,
4844 (unsigned long long)map_length
);
4848 bbio
->orig_bio
= first_bio
;
4849 bbio
->private = first_bio
->bi_private
;
4850 bbio
->end_io
= first_bio
->bi_end_io
;
4851 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
4853 while (dev_nr
< total_devs
) {
4854 dev
= bbio
->stripes
[dev_nr
].dev
;
4855 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
4856 bbio_error(bbio
, first_bio
, logical
);
4862 * Check and see if we're ok with this bio based on it's size
4863 * and offset with the given device.
4865 if (!bio_size_ok(dev
->bdev
, first_bio
,
4866 bbio
->stripes
[dev_nr
].physical
>> 9)) {
4867 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
4868 dev_nr
, rw
, async_submit
);
4874 if (dev_nr
< total_devs
- 1) {
4875 bio
= bio_clone(first_bio
, GFP_NOFS
);
4876 BUG_ON(!bio
); /* -ENOMEM */
4881 submit_stripe_bio(root
, bbio
, bio
,
4882 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
4889 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
4892 struct btrfs_device
*device
;
4893 struct btrfs_fs_devices
*cur_devices
;
4895 cur_devices
= fs_info
->fs_devices
;
4896 while (cur_devices
) {
4898 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4899 device
= __find_device(&cur_devices
->devices
,
4904 cur_devices
= cur_devices
->seed
;
4909 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
4910 u64 devid
, u8
*dev_uuid
)
4912 struct btrfs_device
*device
;
4913 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
4915 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
4918 list_add(&device
->dev_list
,
4919 &fs_devices
->devices
);
4920 device
->dev_root
= root
->fs_info
->dev_root
;
4921 device
->devid
= devid
;
4922 device
->work
.func
= pending_bios_fn
;
4923 device
->fs_devices
= fs_devices
;
4924 device
->missing
= 1;
4925 fs_devices
->num_devices
++;
4926 fs_devices
->missing_devices
++;
4927 spin_lock_init(&device
->io_lock
);
4928 INIT_LIST_HEAD(&device
->dev_alloc_list
);
4929 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
4933 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
4934 struct extent_buffer
*leaf
,
4935 struct btrfs_chunk
*chunk
)
4937 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4938 struct map_lookup
*map
;
4939 struct extent_map
*em
;
4943 u8 uuid
[BTRFS_UUID_SIZE
];
4948 logical
= key
->offset
;
4949 length
= btrfs_chunk_length(leaf
, chunk
);
4951 read_lock(&map_tree
->map_tree
.lock
);
4952 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
4953 read_unlock(&map_tree
->map_tree
.lock
);
4955 /* already mapped? */
4956 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
4957 free_extent_map(em
);
4960 free_extent_map(em
);
4963 em
= alloc_extent_map();
4966 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
4967 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4969 free_extent_map(em
);
4973 em
->bdev
= (struct block_device
*)map
;
4974 em
->start
= logical
;
4977 em
->block_start
= 0;
4978 em
->block_len
= em
->len
;
4980 map
->num_stripes
= num_stripes
;
4981 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
4982 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
4983 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
4984 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
4985 map
->type
= btrfs_chunk_type(leaf
, chunk
);
4986 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
4987 for (i
= 0; i
< num_stripes
; i
++) {
4988 map
->stripes
[i
].physical
=
4989 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
4990 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
4991 read_extent_buffer(leaf
, uuid
, (unsigned long)
4992 btrfs_stripe_dev_uuid_nr(chunk
, i
),
4994 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
4996 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
4998 free_extent_map(em
);
5001 if (!map
->stripes
[i
].dev
) {
5002 map
->stripes
[i
].dev
=
5003 add_missing_dev(root
, devid
, uuid
);
5004 if (!map
->stripes
[i
].dev
) {
5006 free_extent_map(em
);
5010 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5013 write_lock(&map_tree
->map_tree
.lock
);
5014 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
5015 write_unlock(&map_tree
->map_tree
.lock
);
5016 BUG_ON(ret
); /* Tree corruption */
5017 free_extent_map(em
);
5022 static void fill_device_from_item(struct extent_buffer
*leaf
,
5023 struct btrfs_dev_item
*dev_item
,
5024 struct btrfs_device
*device
)
5028 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5029 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5030 device
->total_bytes
= device
->disk_total_bytes
;
5031 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5032 device
->type
= btrfs_device_type(leaf
, dev_item
);
5033 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5034 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5035 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5036 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5037 device
->is_tgtdev_for_dev_replace
= 0;
5039 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
5040 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5043 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5045 struct btrfs_fs_devices
*fs_devices
;
5048 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5050 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5051 while (fs_devices
) {
5052 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5056 fs_devices
= fs_devices
->seed
;
5059 fs_devices
= find_fsid(fsid
);
5065 fs_devices
= clone_fs_devices(fs_devices
);
5066 if (IS_ERR(fs_devices
)) {
5067 ret
= PTR_ERR(fs_devices
);
5071 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5072 root
->fs_info
->bdev_holder
);
5074 free_fs_devices(fs_devices
);
5078 if (!fs_devices
->seeding
) {
5079 __btrfs_close_devices(fs_devices
);
5080 free_fs_devices(fs_devices
);
5085 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5086 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5091 static int read_one_dev(struct btrfs_root
*root
,
5092 struct extent_buffer
*leaf
,
5093 struct btrfs_dev_item
*dev_item
)
5095 struct btrfs_device
*device
;
5098 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5099 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5101 devid
= btrfs_device_id(leaf
, dev_item
);
5102 read_extent_buffer(leaf
, dev_uuid
,
5103 (unsigned long)btrfs_device_uuid(dev_item
),
5105 read_extent_buffer(leaf
, fs_uuid
,
5106 (unsigned long)btrfs_device_fsid(dev_item
),
5109 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5110 ret
= open_seed_devices(root
, fs_uuid
);
5111 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5115 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5116 if (!device
|| !device
->bdev
) {
5117 if (!btrfs_test_opt(root
, DEGRADED
))
5121 printk(KERN_WARNING
"warning devid %llu missing\n",
5122 (unsigned long long)devid
);
5123 device
= add_missing_dev(root
, devid
, dev_uuid
);
5126 } else if (!device
->missing
) {
5128 * this happens when a device that was properly setup
5129 * in the device info lists suddenly goes bad.
5130 * device->bdev is NULL, and so we have to set
5131 * device->missing to one here
5133 root
->fs_info
->fs_devices
->missing_devices
++;
5134 device
->missing
= 1;
5138 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5139 BUG_ON(device
->writeable
);
5140 if (device
->generation
!=
5141 btrfs_device_generation(leaf
, dev_item
))
5145 fill_device_from_item(leaf
, dev_item
, device
);
5146 device
->dev_root
= root
->fs_info
->dev_root
;
5147 device
->in_fs_metadata
= 1;
5148 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5149 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5150 spin_lock(&root
->fs_info
->free_chunk_lock
);
5151 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5153 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5159 int btrfs_read_sys_array(struct btrfs_root
*root
)
5161 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5162 struct extent_buffer
*sb
;
5163 struct btrfs_disk_key
*disk_key
;
5164 struct btrfs_chunk
*chunk
;
5166 unsigned long sb_ptr
;
5172 struct btrfs_key key
;
5174 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5175 BTRFS_SUPER_INFO_SIZE
);
5178 btrfs_set_buffer_uptodate(sb
);
5179 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5181 * The sb extent buffer is artifical and just used to read the system array.
5182 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5183 * pages up-to-date when the page is larger: extent does not cover the
5184 * whole page and consequently check_page_uptodate does not find all
5185 * the page's extents up-to-date (the hole beyond sb),
5186 * write_extent_buffer then triggers a WARN_ON.
5188 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5189 * but sb spans only this function. Add an explicit SetPageUptodate call
5190 * to silence the warning eg. on PowerPC 64.
5192 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5193 SetPageUptodate(sb
->pages
[0]);
5195 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5196 array_size
= btrfs_super_sys_array_size(super_copy
);
5198 ptr
= super_copy
->sys_chunk_array
;
5199 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
5202 while (cur
< array_size
) {
5203 disk_key
= (struct btrfs_disk_key
*)ptr
;
5204 btrfs_disk_key_to_cpu(&key
, disk_key
);
5206 len
= sizeof(*disk_key
); ptr
+= len
;
5210 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5211 chunk
= (struct btrfs_chunk
*)sb_ptr
;
5212 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
5215 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
5216 len
= btrfs_chunk_item_size(num_stripes
);
5225 free_extent_buffer(sb
);
5229 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
5231 struct btrfs_path
*path
;
5232 struct extent_buffer
*leaf
;
5233 struct btrfs_key key
;
5234 struct btrfs_key found_key
;
5238 root
= root
->fs_info
->chunk_root
;
5240 path
= btrfs_alloc_path();
5244 mutex_lock(&uuid_mutex
);
5247 /* first we search for all of the device items, and then we
5248 * read in all of the chunk items. This way we can create chunk
5249 * mappings that reference all of the devices that are afound
5251 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
5255 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5259 leaf
= path
->nodes
[0];
5260 slot
= path
->slots
[0];
5261 if (slot
>= btrfs_header_nritems(leaf
)) {
5262 ret
= btrfs_next_leaf(root
, path
);
5269 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5270 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5271 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
5273 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
5274 struct btrfs_dev_item
*dev_item
;
5275 dev_item
= btrfs_item_ptr(leaf
, slot
,
5276 struct btrfs_dev_item
);
5277 ret
= read_one_dev(root
, leaf
, dev_item
);
5281 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5282 struct btrfs_chunk
*chunk
;
5283 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
5284 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
5290 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5292 btrfs_release_path(path
);
5297 unlock_chunks(root
);
5298 mutex_unlock(&uuid_mutex
);
5300 btrfs_free_path(path
);
5304 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
5308 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5309 btrfs_dev_stat_reset(dev
, i
);
5312 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
5314 struct btrfs_key key
;
5315 struct btrfs_key found_key
;
5316 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5317 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5318 struct extent_buffer
*eb
;
5321 struct btrfs_device
*device
;
5322 struct btrfs_path
*path
= NULL
;
5325 path
= btrfs_alloc_path();
5331 mutex_lock(&fs_devices
->device_list_mutex
);
5332 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5334 struct btrfs_dev_stats_item
*ptr
;
5337 key
.type
= BTRFS_DEV_STATS_KEY
;
5338 key
.offset
= device
->devid
;
5339 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
5341 __btrfs_reset_dev_stats(device
);
5342 device
->dev_stats_valid
= 1;
5343 btrfs_release_path(path
);
5346 slot
= path
->slots
[0];
5347 eb
= path
->nodes
[0];
5348 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5349 item_size
= btrfs_item_size_nr(eb
, slot
);
5351 ptr
= btrfs_item_ptr(eb
, slot
,
5352 struct btrfs_dev_stats_item
);
5354 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5355 if (item_size
>= (1 + i
) * sizeof(__le64
))
5356 btrfs_dev_stat_set(device
, i
,
5357 btrfs_dev_stats_value(eb
, ptr
, i
));
5359 btrfs_dev_stat_reset(device
, i
);
5362 device
->dev_stats_valid
= 1;
5363 btrfs_dev_stat_print_on_load(device
);
5364 btrfs_release_path(path
);
5366 mutex_unlock(&fs_devices
->device_list_mutex
);
5369 btrfs_free_path(path
);
5370 return ret
< 0 ? ret
: 0;
5373 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
5374 struct btrfs_root
*dev_root
,
5375 struct btrfs_device
*device
)
5377 struct btrfs_path
*path
;
5378 struct btrfs_key key
;
5379 struct extent_buffer
*eb
;
5380 struct btrfs_dev_stats_item
*ptr
;
5385 key
.type
= BTRFS_DEV_STATS_KEY
;
5386 key
.offset
= device
->devid
;
5388 path
= btrfs_alloc_path();
5390 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
5392 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
5393 ret
, rcu_str_deref(device
->name
));
5398 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
5399 /* need to delete old one and insert a new one */
5400 ret
= btrfs_del_item(trans
, dev_root
, path
);
5402 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
5403 rcu_str_deref(device
->name
), ret
);
5410 /* need to insert a new item */
5411 btrfs_release_path(path
);
5412 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
5413 &key
, sizeof(*ptr
));
5415 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
5416 rcu_str_deref(device
->name
), ret
);
5421 eb
= path
->nodes
[0];
5422 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
5423 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5424 btrfs_set_dev_stats_value(eb
, ptr
, i
,
5425 btrfs_dev_stat_read(device
, i
));
5426 btrfs_mark_buffer_dirty(eb
);
5429 btrfs_free_path(path
);
5434 * called from commit_transaction. Writes all changed device stats to disk.
5436 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
5437 struct btrfs_fs_info
*fs_info
)
5439 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5440 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5441 struct btrfs_device
*device
;
5444 mutex_lock(&fs_devices
->device_list_mutex
);
5445 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5446 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
5449 ret
= update_dev_stat_item(trans
, dev_root
, device
);
5451 device
->dev_stats_dirty
= 0;
5453 mutex_unlock(&fs_devices
->device_list_mutex
);
5458 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
5460 btrfs_dev_stat_inc(dev
, index
);
5461 btrfs_dev_stat_print_on_error(dev
);
5464 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
5466 if (!dev
->dev_stats_valid
)
5468 printk_ratelimited_in_rcu(KERN_ERR
5469 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5470 rcu_str_deref(dev
->name
),
5471 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5472 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5473 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5474 btrfs_dev_stat_read(dev
,
5475 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5476 btrfs_dev_stat_read(dev
,
5477 BTRFS_DEV_STAT_GENERATION_ERRS
));
5480 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
5484 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5485 if (btrfs_dev_stat_read(dev
, i
) != 0)
5487 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
5488 return; /* all values == 0, suppress message */
5490 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5491 rcu_str_deref(dev
->name
),
5492 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5493 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5494 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5495 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5496 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
5499 int btrfs_get_dev_stats(struct btrfs_root
*root
,
5500 struct btrfs_ioctl_get_dev_stats
*stats
)
5502 struct btrfs_device
*dev
;
5503 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5506 mutex_lock(&fs_devices
->device_list_mutex
);
5507 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
5508 mutex_unlock(&fs_devices
->device_list_mutex
);
5512 "btrfs: get dev_stats failed, device not found\n");
5514 } else if (!dev
->dev_stats_valid
) {
5516 "btrfs: get dev_stats failed, not yet valid\n");
5518 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
5519 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5520 if (stats
->nr_items
> i
)
5522 btrfs_dev_stat_read_and_reset(dev
, i
);
5524 btrfs_dev_stat_reset(dev
, i
);
5527 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5528 if (stats
->nr_items
> i
)
5529 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
5531 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
5532 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
5536 int btrfs_scratch_superblock(struct btrfs_device
*device
)
5538 struct buffer_head
*bh
;
5539 struct btrfs_super_block
*disk_super
;
5541 bh
= btrfs_read_dev_super(device
->bdev
);
5544 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
5546 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
5547 set_buffer_dirty(bh
);
5548 sync_dirty_buffer(bh
);