/fs/btrfs/volumes.c
C | 3718 lines | 2949 code | 513 blank | 256 comment | 540 complexity | 58b4fc98f53631a2fc9f626b94464381 MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.0, AGPL-1.0
Large files files are truncated, but you can click here to view the full file
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
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 <asm/div64.h>
27#include "compat.h"
28#include "ctree.h"
29#include "extent_map.h"
30#include "disk-io.h"
31#include "transaction.h"
32#include "print-tree.h"
33#include "volumes.h"
34#include "async-thread.h"
35
36static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
41static DEFINE_MUTEX(uuid_mutex);
42static LIST_HEAD(fs_uuids);
43
44static void lock_chunks(struct btrfs_root *root)
45{
46 mutex_lock(&root->fs_info->chunk_mutex);
47}
48
49static void unlock_chunks(struct btrfs_root *root)
50{
51 mutex_unlock(&root->fs_info->chunk_mutex);
52}
53
54static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
55{
56 struct btrfs_device *device;
57 WARN_ON(fs_devices->opened);
58 while (!list_empty(&fs_devices->devices)) {
59 device = list_entry(fs_devices->devices.next,
60 struct btrfs_device, dev_list);
61 list_del(&device->dev_list);
62 kfree(device->name);
63 kfree(device);
64 }
65 kfree(fs_devices);
66}
67
68int btrfs_cleanup_fs_uuids(void)
69{
70 struct btrfs_fs_devices *fs_devices;
71
72 while (!list_empty(&fs_uuids)) {
73 fs_devices = list_entry(fs_uuids.next,
74 struct btrfs_fs_devices, list);
75 list_del(&fs_devices->list);
76 free_fs_devices(fs_devices);
77 }
78 return 0;
79}
80
81static noinline struct btrfs_device *__find_device(struct list_head *head,
82 u64 devid, u8 *uuid)
83{
84 struct btrfs_device *dev;
85
86 list_for_each_entry(dev, head, dev_list) {
87 if (dev->devid == devid &&
88 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
89 return dev;
90 }
91 }
92 return NULL;
93}
94
95static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
96{
97 struct btrfs_fs_devices *fs_devices;
98
99 list_for_each_entry(fs_devices, &fs_uuids, list) {
100 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
101 return fs_devices;
102 }
103 return NULL;
104}
105
106static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 struct bio *head, struct bio *tail)
108{
109
110 struct bio *old_head;
111
112 old_head = pending_bios->head;
113 pending_bios->head = head;
114 if (pending_bios->tail)
115 tail->bi_next = old_head;
116 else
117 pending_bios->tail = tail;
118}
119
120/*
121 * we try to collect pending bios for a device so we don't get a large
122 * number of procs sending bios down to the same device. This greatly
123 * improves the schedulers ability to collect and merge the bios.
124 *
125 * But, it also turns into a long list of bios to process and that is sure
126 * to eventually make the worker thread block. The solution here is to
127 * make some progress and then put this work struct back at the end of
128 * the list if the block device is congested. This way, multiple devices
129 * can make progress from a single worker thread.
130 */
131static noinline int run_scheduled_bios(struct btrfs_device *device)
132{
133 struct bio *pending;
134 struct backing_dev_info *bdi;
135 struct btrfs_fs_info *fs_info;
136 struct btrfs_pending_bios *pending_bios;
137 struct bio *tail;
138 struct bio *cur;
139 int again = 0;
140 unsigned long num_run;
141 unsigned long batch_run = 0;
142 unsigned long limit;
143 unsigned long last_waited = 0;
144 int force_reg = 0;
145 struct blk_plug plug;
146
147 /*
148 * this function runs all the bios we've collected for
149 * a particular device. We don't want to wander off to
150 * another device without first sending all of these down.
151 * So, setup a plug here and finish it off before we return
152 */
153 blk_start_plug(&plug);
154
155 bdi = blk_get_backing_dev_info(device->bdev);
156 fs_info = device->dev_root->fs_info;
157 limit = btrfs_async_submit_limit(fs_info);
158 limit = limit * 2 / 3;
159
160loop:
161 spin_lock(&device->io_lock);
162
163loop_lock:
164 num_run = 0;
165
166 /* take all the bios off the list at once and process them
167 * later on (without the lock held). But, remember the
168 * tail and other pointers so the bios can be properly reinserted
169 * into the list if we hit congestion
170 */
171 if (!force_reg && device->pending_sync_bios.head) {
172 pending_bios = &device->pending_sync_bios;
173 force_reg = 1;
174 } else {
175 pending_bios = &device->pending_bios;
176 force_reg = 0;
177 }
178
179 pending = pending_bios->head;
180 tail = pending_bios->tail;
181 WARN_ON(pending && !tail);
182
183 /*
184 * if pending was null this time around, no bios need processing
185 * at all and we can stop. Otherwise it'll loop back up again
186 * and do an additional check so no bios are missed.
187 *
188 * device->running_pending is used to synchronize with the
189 * schedule_bio code.
190 */
191 if (device->pending_sync_bios.head == NULL &&
192 device->pending_bios.head == NULL) {
193 again = 0;
194 device->running_pending = 0;
195 } else {
196 again = 1;
197 device->running_pending = 1;
198 }
199
200 pending_bios->head = NULL;
201 pending_bios->tail = NULL;
202
203 spin_unlock(&device->io_lock);
204
205 while (pending) {
206
207 rmb();
208 /* we want to work on both lists, but do more bios on the
209 * sync list than the regular list
210 */
211 if ((num_run > 32 &&
212 pending_bios != &device->pending_sync_bios &&
213 device->pending_sync_bios.head) ||
214 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
215 device->pending_bios.head)) {
216 spin_lock(&device->io_lock);
217 requeue_list(pending_bios, pending, tail);
218 goto loop_lock;
219 }
220
221 cur = pending;
222 pending = pending->bi_next;
223 cur->bi_next = NULL;
224 atomic_dec(&fs_info->nr_async_bios);
225
226 if (atomic_read(&fs_info->nr_async_bios) < limit &&
227 waitqueue_active(&fs_info->async_submit_wait))
228 wake_up(&fs_info->async_submit_wait);
229
230 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
231
232 submit_bio(cur->bi_rw, cur);
233 num_run++;
234 batch_run++;
235 if (need_resched())
236 cond_resched();
237
238 /*
239 * we made progress, there is more work to do and the bdi
240 * is now congested. Back off and let other work structs
241 * run instead
242 */
243 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
244 fs_info->fs_devices->open_devices > 1) {
245 struct io_context *ioc;
246
247 ioc = current->io_context;
248
249 /*
250 * the main goal here is that we don't want to
251 * block if we're going to be able to submit
252 * more requests without blocking.
253 *
254 * This code does two great things, it pokes into
255 * the elevator code from a filesystem _and_
256 * it makes assumptions about how batching works.
257 */
258 if (ioc && ioc->nr_batch_requests > 0 &&
259 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
260 (last_waited == 0 ||
261 ioc->last_waited == last_waited)) {
262 /*
263 * we want to go through our batch of
264 * requests and stop. So, we copy out
265 * the ioc->last_waited time and test
266 * against it before looping
267 */
268 last_waited = ioc->last_waited;
269 if (need_resched())
270 cond_resched();
271 continue;
272 }
273 spin_lock(&device->io_lock);
274 requeue_list(pending_bios, pending, tail);
275 device->running_pending = 1;
276
277 spin_unlock(&device->io_lock);
278 btrfs_requeue_work(&device->work);
279 goto done;
280 }
281 }
282
283 cond_resched();
284 if (again)
285 goto loop;
286
287 spin_lock(&device->io_lock);
288 if (device->pending_bios.head || device->pending_sync_bios.head)
289 goto loop_lock;
290 spin_unlock(&device->io_lock);
291
292done:
293 blk_finish_plug(&plug);
294 return 0;
295}
296
297static void pending_bios_fn(struct btrfs_work *work)
298{
299 struct btrfs_device *device;
300
301 device = container_of(work, struct btrfs_device, work);
302 run_scheduled_bios(device);
303}
304
305static noinline int device_list_add(const char *path,
306 struct btrfs_super_block *disk_super,
307 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
308{
309 struct btrfs_device *device;
310 struct btrfs_fs_devices *fs_devices;
311 u64 found_transid = btrfs_super_generation(disk_super);
312 char *name;
313
314 fs_devices = find_fsid(disk_super->fsid);
315 if (!fs_devices) {
316 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
317 if (!fs_devices)
318 return -ENOMEM;
319 INIT_LIST_HEAD(&fs_devices->devices);
320 INIT_LIST_HEAD(&fs_devices->alloc_list);
321 list_add(&fs_devices->list, &fs_uuids);
322 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
323 fs_devices->latest_devid = devid;
324 fs_devices->latest_trans = found_transid;
325 mutex_init(&fs_devices->device_list_mutex);
326 device = NULL;
327 } else {
328 device = __find_device(&fs_devices->devices, devid,
329 disk_super->dev_item.uuid);
330 }
331 if (!device) {
332 if (fs_devices->opened)
333 return -EBUSY;
334
335 device = kzalloc(sizeof(*device), GFP_NOFS);
336 if (!device) {
337 /* we can safely leave the fs_devices entry around */
338 return -ENOMEM;
339 }
340 device->devid = devid;
341 device->work.func = pending_bios_fn;
342 memcpy(device->uuid, disk_super->dev_item.uuid,
343 BTRFS_UUID_SIZE);
344 spin_lock_init(&device->io_lock);
345 device->name = kstrdup(path, GFP_NOFS);
346 if (!device->name) {
347 kfree(device);
348 return -ENOMEM;
349 }
350 INIT_LIST_HEAD(&device->dev_alloc_list);
351
352 mutex_lock(&fs_devices->device_list_mutex);
353 list_add_rcu(&device->dev_list, &fs_devices->devices);
354 mutex_unlock(&fs_devices->device_list_mutex);
355
356 device->fs_devices = fs_devices;
357 fs_devices->num_devices++;
358 } else if (!device->name || strcmp(device->name, path)) {
359 name = kstrdup(path, GFP_NOFS);
360 if (!name)
361 return -ENOMEM;
362 kfree(device->name);
363 device->name = name;
364 if (device->missing) {
365 fs_devices->missing_devices--;
366 device->missing = 0;
367 }
368 }
369
370 if (found_transid > fs_devices->latest_trans) {
371 fs_devices->latest_devid = devid;
372 fs_devices->latest_trans = found_transid;
373 }
374 *fs_devices_ret = fs_devices;
375 return 0;
376}
377
378static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
379{
380 struct btrfs_fs_devices *fs_devices;
381 struct btrfs_device *device;
382 struct btrfs_device *orig_dev;
383
384 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
385 if (!fs_devices)
386 return ERR_PTR(-ENOMEM);
387
388 INIT_LIST_HEAD(&fs_devices->devices);
389 INIT_LIST_HEAD(&fs_devices->alloc_list);
390 INIT_LIST_HEAD(&fs_devices->list);
391 mutex_init(&fs_devices->device_list_mutex);
392 fs_devices->latest_devid = orig->latest_devid;
393 fs_devices->latest_trans = orig->latest_trans;
394 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
395
396 /* We have held the volume lock, it is safe to get the devices. */
397 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
398 device = kzalloc(sizeof(*device), GFP_NOFS);
399 if (!device)
400 goto error;
401
402 device->name = kstrdup(orig_dev->name, GFP_NOFS);
403 if (!device->name) {
404 kfree(device);
405 goto error;
406 }
407
408 device->devid = orig_dev->devid;
409 device->work.func = pending_bios_fn;
410 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
411 spin_lock_init(&device->io_lock);
412 INIT_LIST_HEAD(&device->dev_list);
413 INIT_LIST_HEAD(&device->dev_alloc_list);
414
415 list_add(&device->dev_list, &fs_devices->devices);
416 device->fs_devices = fs_devices;
417 fs_devices->num_devices++;
418 }
419 return fs_devices;
420error:
421 free_fs_devices(fs_devices);
422 return ERR_PTR(-ENOMEM);
423}
424
425int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
426{
427 struct btrfs_device *device, *next;
428
429 mutex_lock(&uuid_mutex);
430again:
431 /* This is the initialized path, it is safe to release the devices. */
432 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
433 if (device->in_fs_metadata)
434 continue;
435
436 if (device->bdev) {
437 blkdev_put(device->bdev, device->mode);
438 device->bdev = NULL;
439 fs_devices->open_devices--;
440 }
441 if (device->writeable) {
442 list_del_init(&device->dev_alloc_list);
443 device->writeable = 0;
444 fs_devices->rw_devices--;
445 }
446 list_del_init(&device->dev_list);
447 fs_devices->num_devices--;
448 kfree(device->name);
449 kfree(device);
450 }
451
452 if (fs_devices->seed) {
453 fs_devices = fs_devices->seed;
454 goto again;
455 }
456
457 mutex_unlock(&uuid_mutex);
458 return 0;
459}
460
461static void __free_device(struct work_struct *work)
462{
463 struct btrfs_device *device;
464
465 device = container_of(work, struct btrfs_device, rcu_work);
466
467 if (device->bdev)
468 blkdev_put(device->bdev, device->mode);
469
470 kfree(device->name);
471 kfree(device);
472}
473
474static void free_device(struct rcu_head *head)
475{
476 struct btrfs_device *device;
477
478 device = container_of(head, struct btrfs_device, rcu);
479
480 INIT_WORK(&device->rcu_work, __free_device);
481 schedule_work(&device->rcu_work);
482}
483
484static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
485{
486 struct btrfs_device *device;
487
488 if (--fs_devices->opened > 0)
489 return 0;
490
491 mutex_lock(&fs_devices->device_list_mutex);
492 list_for_each_entry(device, &fs_devices->devices, dev_list) {
493 struct btrfs_device *new_device;
494
495 if (device->bdev)
496 fs_devices->open_devices--;
497
498 if (device->writeable) {
499 list_del_init(&device->dev_alloc_list);
500 fs_devices->rw_devices--;
501 }
502
503 if (device->can_discard)
504 fs_devices->num_can_discard--;
505
506 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
507 BUG_ON(!new_device);
508 memcpy(new_device, device, sizeof(*new_device));
509 new_device->name = kstrdup(device->name, GFP_NOFS);
510 BUG_ON(device->name && !new_device->name);
511 new_device->bdev = NULL;
512 new_device->writeable = 0;
513 new_device->in_fs_metadata = 0;
514 new_device->can_discard = 0;
515 list_replace_rcu(&device->dev_list, &new_device->dev_list);
516
517 call_rcu(&device->rcu, free_device);
518 }
519 mutex_unlock(&fs_devices->device_list_mutex);
520
521 WARN_ON(fs_devices->open_devices);
522 WARN_ON(fs_devices->rw_devices);
523 fs_devices->opened = 0;
524 fs_devices->seeding = 0;
525
526 return 0;
527}
528
529int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
530{
531 struct btrfs_fs_devices *seed_devices = NULL;
532 int ret;
533
534 mutex_lock(&uuid_mutex);
535 ret = __btrfs_close_devices(fs_devices);
536 if (!fs_devices->opened) {
537 seed_devices = fs_devices->seed;
538 fs_devices->seed = NULL;
539 }
540 mutex_unlock(&uuid_mutex);
541
542 while (seed_devices) {
543 fs_devices = seed_devices;
544 seed_devices = fs_devices->seed;
545 __btrfs_close_devices(fs_devices);
546 free_fs_devices(fs_devices);
547 }
548 return ret;
549}
550
551static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
552 fmode_t flags, void *holder)
553{
554 struct request_queue *q;
555 struct block_device *bdev;
556 struct list_head *head = &fs_devices->devices;
557 struct btrfs_device *device;
558 struct block_device *latest_bdev = NULL;
559 struct buffer_head *bh;
560 struct btrfs_super_block *disk_super;
561 u64 latest_devid = 0;
562 u64 latest_transid = 0;
563 u64 devid;
564 int seeding = 1;
565 int ret = 0;
566
567 flags |= FMODE_EXCL;
568
569 list_for_each_entry(device, head, dev_list) {
570 if (device->bdev)
571 continue;
572 if (!device->name)
573 continue;
574
575 bdev = blkdev_get_by_path(device->name, flags, holder);
576 if (IS_ERR(bdev)) {
577 printk(KERN_INFO "open %s failed\n", device->name);
578 goto error;
579 }
580 set_blocksize(bdev, 4096);
581
582 bh = btrfs_read_dev_super(bdev);
583 if (!bh) {
584 ret = -EINVAL;
585 goto error_close;
586 }
587
588 disk_super = (struct btrfs_super_block *)bh->b_data;
589 devid = btrfs_stack_device_id(&disk_super->dev_item);
590 if (devid != device->devid)
591 goto error_brelse;
592
593 if (memcmp(device->uuid, disk_super->dev_item.uuid,
594 BTRFS_UUID_SIZE))
595 goto error_brelse;
596
597 device->generation = btrfs_super_generation(disk_super);
598 if (!latest_transid || device->generation > latest_transid) {
599 latest_devid = devid;
600 latest_transid = device->generation;
601 latest_bdev = bdev;
602 }
603
604 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
605 device->writeable = 0;
606 } else {
607 device->writeable = !bdev_read_only(bdev);
608 seeding = 0;
609 }
610
611 q = bdev_get_queue(bdev);
612 if (blk_queue_discard(q)) {
613 device->can_discard = 1;
614 fs_devices->num_can_discard++;
615 }
616
617 device->bdev = bdev;
618 device->in_fs_metadata = 0;
619 device->mode = flags;
620
621 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
622 fs_devices->rotating = 1;
623
624 fs_devices->open_devices++;
625 if (device->writeable) {
626 fs_devices->rw_devices++;
627 list_add(&device->dev_alloc_list,
628 &fs_devices->alloc_list);
629 }
630 brelse(bh);
631 continue;
632
633error_brelse:
634 brelse(bh);
635error_close:
636 blkdev_put(bdev, flags);
637error:
638 continue;
639 }
640 if (fs_devices->open_devices == 0) {
641 ret = -EIO;
642 goto out;
643 }
644 fs_devices->seeding = seeding;
645 fs_devices->opened = 1;
646 fs_devices->latest_bdev = latest_bdev;
647 fs_devices->latest_devid = latest_devid;
648 fs_devices->latest_trans = latest_transid;
649 fs_devices->total_rw_bytes = 0;
650out:
651 return ret;
652}
653
654int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
655 fmode_t flags, void *holder)
656{
657 int ret;
658
659 mutex_lock(&uuid_mutex);
660 if (fs_devices->opened) {
661 fs_devices->opened++;
662 ret = 0;
663 } else {
664 ret = __btrfs_open_devices(fs_devices, flags, holder);
665 }
666 mutex_unlock(&uuid_mutex);
667 return ret;
668}
669
670int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
671 struct btrfs_fs_devices **fs_devices_ret)
672{
673 struct btrfs_super_block *disk_super;
674 struct block_device *bdev;
675 struct buffer_head *bh;
676 int ret;
677 u64 devid;
678 u64 transid;
679
680 mutex_lock(&uuid_mutex);
681
682 flags |= FMODE_EXCL;
683 bdev = blkdev_get_by_path(path, flags, holder);
684
685 if (IS_ERR(bdev)) {
686 ret = PTR_ERR(bdev);
687 goto error;
688 }
689
690 ret = set_blocksize(bdev, 4096);
691 if (ret)
692 goto error_close;
693 bh = btrfs_read_dev_super(bdev);
694 if (!bh) {
695 ret = -EINVAL;
696 goto error_close;
697 }
698 disk_super = (struct btrfs_super_block *)bh->b_data;
699 devid = btrfs_stack_device_id(&disk_super->dev_item);
700 transid = btrfs_super_generation(disk_super);
701 if (disk_super->label[0])
702 printk(KERN_INFO "device label %s ", disk_super->label);
703 else
704 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
705 printk(KERN_CONT "devid %llu transid %llu %s\n",
706 (unsigned long long)devid, (unsigned long long)transid, path);
707 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
708
709 brelse(bh);
710error_close:
711 blkdev_put(bdev, flags);
712error:
713 mutex_unlock(&uuid_mutex);
714 return ret;
715}
716
717/* helper to account the used device space in the range */
718int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
719 u64 end, u64 *length)
720{
721 struct btrfs_key key;
722 struct btrfs_root *root = device->dev_root;
723 struct btrfs_dev_extent *dev_extent;
724 struct btrfs_path *path;
725 u64 extent_end;
726 int ret;
727 int slot;
728 struct extent_buffer *l;
729
730 *length = 0;
731
732 if (start >= device->total_bytes)
733 return 0;
734
735 path = btrfs_alloc_path();
736 if (!path)
737 return -ENOMEM;
738 path->reada = 2;
739
740 key.objectid = device->devid;
741 key.offset = start;
742 key.type = BTRFS_DEV_EXTENT_KEY;
743
744 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
745 if (ret < 0)
746 goto out;
747 if (ret > 0) {
748 ret = btrfs_previous_item(root, path, key.objectid, key.type);
749 if (ret < 0)
750 goto out;
751 }
752
753 while (1) {
754 l = path->nodes[0];
755 slot = path->slots[0];
756 if (slot >= btrfs_header_nritems(l)) {
757 ret = btrfs_next_leaf(root, path);
758 if (ret == 0)
759 continue;
760 if (ret < 0)
761 goto out;
762
763 break;
764 }
765 btrfs_item_key_to_cpu(l, &key, slot);
766
767 if (key.objectid < device->devid)
768 goto next;
769
770 if (key.objectid > device->devid)
771 break;
772
773 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
774 goto next;
775
776 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
777 extent_end = key.offset + btrfs_dev_extent_length(l,
778 dev_extent);
779 if (key.offset <= start && extent_end > end) {
780 *length = end - start + 1;
781 break;
782 } else if (key.offset <= start && extent_end > start)
783 *length += extent_end - start;
784 else if (key.offset > start && extent_end <= end)
785 *length += extent_end - key.offset;
786 else if (key.offset > start && key.offset <= end) {
787 *length += end - key.offset + 1;
788 break;
789 } else if (key.offset > end)
790 break;
791
792next:
793 path->slots[0]++;
794 }
795 ret = 0;
796out:
797 btrfs_free_path(path);
798 return ret;
799}
800
801/*
802 * find_free_dev_extent - find free space in the specified device
803 * @trans: transaction handler
804 * @device: the device which we search the free space in
805 * @num_bytes: the size of the free space that we need
806 * @start: store the start of the free space.
807 * @len: the size of the free space. that we find, or the size of the max
808 * free space if we don't find suitable free space
809 *
810 * this uses a pretty simple search, the expectation is that it is
811 * called very infrequently and that a given device has a small number
812 * of extents
813 *
814 * @start is used to store the start of the free space if we find. But if we
815 * don't find suitable free space, it will be used to store the start position
816 * of the max free space.
817 *
818 * @len is used to store the size of the free space that we find.
819 * But if we don't find suitable free space, it is used to store the size of
820 * the max free space.
821 */
822int find_free_dev_extent(struct btrfs_trans_handle *trans,
823 struct btrfs_device *device, u64 num_bytes,
824 u64 *start, u64 *len)
825{
826 struct btrfs_key key;
827 struct btrfs_root *root = device->dev_root;
828 struct btrfs_dev_extent *dev_extent;
829 struct btrfs_path *path;
830 u64 hole_size;
831 u64 max_hole_start;
832 u64 max_hole_size;
833 u64 extent_end;
834 u64 search_start;
835 u64 search_end = device->total_bytes;
836 int ret;
837 int slot;
838 struct extent_buffer *l;
839
840 /* FIXME use last free of some kind */
841
842 /* we don't want to overwrite the superblock on the drive,
843 * so we make sure to start at an offset of at least 1MB
844 */
845 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
846
847 max_hole_start = search_start;
848 max_hole_size = 0;
849
850 if (search_start >= search_end) {
851 ret = -ENOSPC;
852 goto error;
853 }
854
855 path = btrfs_alloc_path();
856 if (!path) {
857 ret = -ENOMEM;
858 goto error;
859 }
860 path->reada = 2;
861
862 key.objectid = device->devid;
863 key.offset = search_start;
864 key.type = BTRFS_DEV_EXTENT_KEY;
865
866 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
867 if (ret < 0)
868 goto out;
869 if (ret > 0) {
870 ret = btrfs_previous_item(root, path, key.objectid, key.type);
871 if (ret < 0)
872 goto out;
873 }
874
875 while (1) {
876 l = path->nodes[0];
877 slot = path->slots[0];
878 if (slot >= btrfs_header_nritems(l)) {
879 ret = btrfs_next_leaf(root, path);
880 if (ret == 0)
881 continue;
882 if (ret < 0)
883 goto out;
884
885 break;
886 }
887 btrfs_item_key_to_cpu(l, &key, slot);
888
889 if (key.objectid < device->devid)
890 goto next;
891
892 if (key.objectid > device->devid)
893 break;
894
895 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
896 goto next;
897
898 if (key.offset > search_start) {
899 hole_size = key.offset - search_start;
900
901 if (hole_size > max_hole_size) {
902 max_hole_start = search_start;
903 max_hole_size = hole_size;
904 }
905
906 /*
907 * If this free space is greater than which we need,
908 * it must be the max free space that we have found
909 * until now, so max_hole_start must point to the start
910 * of this free space and the length of this free space
911 * is stored in max_hole_size. Thus, we return
912 * max_hole_start and max_hole_size and go back to the
913 * caller.
914 */
915 if (hole_size >= num_bytes) {
916 ret = 0;
917 goto out;
918 }
919 }
920
921 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
922 extent_end = key.offset + btrfs_dev_extent_length(l,
923 dev_extent);
924 if (extent_end > search_start)
925 search_start = extent_end;
926next:
927 path->slots[0]++;
928 cond_resched();
929 }
930
931 hole_size = search_end- search_start;
932 if (hole_size > max_hole_size) {
933 max_hole_start = search_start;
934 max_hole_size = hole_size;
935 }
936
937 /* See above. */
938 if (hole_size < num_bytes)
939 ret = -ENOSPC;
940 else
941 ret = 0;
942
943out:
944 btrfs_free_path(path);
945error:
946 *start = max_hole_start;
947 if (len)
948 *len = max_hole_size;
949 return ret;
950}
951
952static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
953 struct btrfs_device *device,
954 u64 start)
955{
956 int ret;
957 struct btrfs_path *path;
958 struct btrfs_root *root = device->dev_root;
959 struct btrfs_key key;
960 struct btrfs_key found_key;
961 struct extent_buffer *leaf = NULL;
962 struct btrfs_dev_extent *extent = NULL;
963
964 path = btrfs_alloc_path();
965 if (!path)
966 return -ENOMEM;
967
968 key.objectid = device->devid;
969 key.offset = start;
970 key.type = BTRFS_DEV_EXTENT_KEY;
971
972 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
973 if (ret > 0) {
974 ret = btrfs_previous_item(root, path, key.objectid,
975 BTRFS_DEV_EXTENT_KEY);
976 if (ret)
977 goto out;
978 leaf = path->nodes[0];
979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
980 extent = btrfs_item_ptr(leaf, path->slots[0],
981 struct btrfs_dev_extent);
982 BUG_ON(found_key.offset > start || found_key.offset +
983 btrfs_dev_extent_length(leaf, extent) < start);
984 } else if (ret == 0) {
985 leaf = path->nodes[0];
986 extent = btrfs_item_ptr(leaf, path->slots[0],
987 struct btrfs_dev_extent);
988 }
989 BUG_ON(ret);
990
991 if (device->bytes_used > 0)
992 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
993 ret = btrfs_del_item(trans, root, path);
994
995out:
996 btrfs_free_path(path);
997 return ret;
998}
999
1000int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1001 struct btrfs_device *device,
1002 u64 chunk_tree, u64 chunk_objectid,
1003 u64 chunk_offset, u64 start, u64 num_bytes)
1004{
1005 int ret;
1006 struct btrfs_path *path;
1007 struct btrfs_root *root = device->dev_root;
1008 struct btrfs_dev_extent *extent;
1009 struct extent_buffer *leaf;
1010 struct btrfs_key key;
1011
1012 WARN_ON(!device->in_fs_metadata);
1013 path = btrfs_alloc_path();
1014 if (!path)
1015 return -ENOMEM;
1016
1017 key.objectid = device->devid;
1018 key.offset = start;
1019 key.type = BTRFS_DEV_EXTENT_KEY;
1020 ret = btrfs_insert_empty_item(trans, root, path, &key,
1021 sizeof(*extent));
1022 BUG_ON(ret);
1023
1024 leaf = path->nodes[0];
1025 extent = btrfs_item_ptr(leaf, path->slots[0],
1026 struct btrfs_dev_extent);
1027 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1028 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1029 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1030
1031 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1032 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1033 BTRFS_UUID_SIZE);
1034
1035 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1036 btrfs_mark_buffer_dirty(leaf);
1037 btrfs_free_path(path);
1038 return ret;
1039}
1040
1041static noinline int find_next_chunk(struct btrfs_root *root,
1042 u64 objectid, u64 *offset)
1043{
1044 struct btrfs_path *path;
1045 int ret;
1046 struct btrfs_key key;
1047 struct btrfs_chunk *chunk;
1048 struct btrfs_key found_key;
1049
1050 path = btrfs_alloc_path();
1051 BUG_ON(!path);
1052
1053 key.objectid = objectid;
1054 key.offset = (u64)-1;
1055 key.type = BTRFS_CHUNK_ITEM_KEY;
1056
1057 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1058 if (ret < 0)
1059 goto error;
1060
1061 BUG_ON(ret == 0);
1062
1063 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1064 if (ret) {
1065 *offset = 0;
1066 } else {
1067 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1068 path->slots[0]);
1069 if (found_key.objectid != objectid)
1070 *offset = 0;
1071 else {
1072 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1073 struct btrfs_chunk);
1074 *offset = found_key.offset +
1075 btrfs_chunk_length(path->nodes[0], chunk);
1076 }
1077 }
1078 ret = 0;
1079error:
1080 btrfs_free_path(path);
1081 return ret;
1082}
1083
1084static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1085{
1086 int ret;
1087 struct btrfs_key key;
1088 struct btrfs_key found_key;
1089 struct btrfs_path *path;
1090
1091 root = root->fs_info->chunk_root;
1092
1093 path = btrfs_alloc_path();
1094 if (!path)
1095 return -ENOMEM;
1096
1097 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1098 key.type = BTRFS_DEV_ITEM_KEY;
1099 key.offset = (u64)-1;
1100
1101 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1102 if (ret < 0)
1103 goto error;
1104
1105 BUG_ON(ret == 0);
1106
1107 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1108 BTRFS_DEV_ITEM_KEY);
1109 if (ret) {
1110 *objectid = 1;
1111 } else {
1112 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1113 path->slots[0]);
1114 *objectid = found_key.offset + 1;
1115 }
1116 ret = 0;
1117error:
1118 btrfs_free_path(path);
1119 return ret;
1120}
1121
1122/*
1123 * the device information is stored in the chunk root
1124 * the btrfs_device struct should be fully filled in
1125 */
1126int btrfs_add_device(struct btrfs_trans_handle *trans,
1127 struct btrfs_root *root,
1128 struct btrfs_device *device)
1129{
1130 int ret;
1131 struct btrfs_path *path;
1132 struct btrfs_dev_item *dev_item;
1133 struct extent_buffer *leaf;
1134 struct btrfs_key key;
1135 unsigned long ptr;
1136
1137 root = root->fs_info->chunk_root;
1138
1139 path = btrfs_alloc_path();
1140 if (!path)
1141 return -ENOMEM;
1142
1143 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1144 key.type = BTRFS_DEV_ITEM_KEY;
1145 key.offset = device->devid;
1146
1147 ret = btrfs_insert_empty_item(trans, root, path, &key,
1148 sizeof(*dev_item));
1149 if (ret)
1150 goto out;
1151
1152 leaf = path->nodes[0];
1153 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1154
1155 btrfs_set_device_id(leaf, dev_item, device->devid);
1156 btrfs_set_device_generation(leaf, dev_item, 0);
1157 btrfs_set_device_type(leaf, dev_item, device->type);
1158 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1159 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1160 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1161 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1162 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1163 btrfs_set_device_group(leaf, dev_item, 0);
1164 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1165 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1166 btrfs_set_device_start_offset(leaf, dev_item, 0);
1167
1168 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1169 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1170 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1171 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1172 btrfs_mark_buffer_dirty(leaf);
1173
1174 ret = 0;
1175out:
1176 btrfs_free_path(path);
1177 return ret;
1178}
1179
1180static int btrfs_rm_dev_item(struct btrfs_root *root,
1181 struct btrfs_device *device)
1182{
1183 int ret;
1184 struct btrfs_path *path;
1185 struct btrfs_key key;
1186 struct btrfs_trans_handle *trans;
1187
1188 root = root->fs_info->chunk_root;
1189
1190 path = btrfs_alloc_path();
1191 if (!path)
1192 return -ENOMEM;
1193
1194 trans = btrfs_start_transaction(root, 0);
1195 if (IS_ERR(trans)) {
1196 btrfs_free_path(path);
1197 return PTR_ERR(trans);
1198 }
1199 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1200 key.type = BTRFS_DEV_ITEM_KEY;
1201 key.offset = device->devid;
1202 lock_chunks(root);
1203
1204 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1205 if (ret < 0)
1206 goto out;
1207
1208 if (ret > 0) {
1209 ret = -ENOENT;
1210 goto out;
1211 }
1212
1213 ret = btrfs_del_item(trans, root, path);
1214 if (ret)
1215 goto out;
1216out:
1217 btrfs_free_path(path);
1218 unlock_chunks(root);
1219 btrfs_commit_transaction(trans, root);
1220 return ret;
1221}
1222
1223int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1224{
1225 struct btrfs_device *device;
1226 struct btrfs_device *next_device;
1227 struct block_device *bdev;
1228 struct buffer_head *bh = NULL;
1229 struct btrfs_super_block *disk_super;
1230 struct btrfs_fs_devices *cur_devices;
1231 u64 all_avail;
1232 u64 devid;
1233 u64 num_devices;
1234 u8 *dev_uuid;
1235 int ret = 0;
1236 bool clear_super = false;
1237
1238 mutex_lock(&uuid_mutex);
1239 mutex_lock(&root->fs_info->volume_mutex);
1240
1241 all_avail = root->fs_info->avail_data_alloc_bits |
1242 root->fs_info->avail_system_alloc_bits |
1243 root->fs_info->avail_metadata_alloc_bits;
1244
1245 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1246 root->fs_info->fs_devices->num_devices <= 4) {
1247 printk(KERN_ERR "btrfs: unable to go below four devices "
1248 "on raid10\n");
1249 ret = -EINVAL;
1250 goto out;
1251 }
1252
1253 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1254 root->fs_info->fs_devices->num_devices <= 2) {
1255 printk(KERN_ERR "btrfs: unable to go below two "
1256 "devices on raid1\n");
1257 ret = -EINVAL;
1258 goto out;
1259 }
1260
1261 if (strcmp(device_path, "missing") == 0) {
1262 struct list_head *devices;
1263 struct btrfs_device *tmp;
1264
1265 device = NULL;
1266 devices = &root->fs_info->fs_devices->devices;
1267 /*
1268 * It is safe to read the devices since the volume_mutex
1269 * is held.
1270 */
1271 list_for_each_entry(tmp, devices, dev_list) {
1272 if (tmp->in_fs_metadata && !tmp->bdev) {
1273 device = tmp;
1274 break;
1275 }
1276 }
1277 bdev = NULL;
1278 bh = NULL;
1279 disk_super = NULL;
1280 if (!device) {
1281 printk(KERN_ERR "btrfs: no missing devices found to "
1282 "remove\n");
1283 goto out;
1284 }
1285 } else {
1286 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1287 root->fs_info->bdev_holder);
1288 if (IS_ERR(bdev)) {
1289 ret = PTR_ERR(bdev);
1290 goto out;
1291 }
1292
1293 set_blocksize(bdev, 4096);
1294 bh = btrfs_read_dev_super(bdev);
1295 if (!bh) {
1296 ret = -EINVAL;
1297 goto error_close;
1298 }
1299 disk_super = (struct btrfs_super_block *)bh->b_data;
1300 devid = btrfs_stack_device_id(&disk_super->dev_item);
1301 dev_uuid = disk_super->dev_item.uuid;
1302 device = btrfs_find_device(root, devid, dev_uuid,
1303 disk_super->fsid);
1304 if (!device) {
1305 ret = -ENOENT;
1306 goto error_brelse;
1307 }
1308 }
1309
1310 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1311 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1312 "device\n");
1313 ret = -EINVAL;
1314 goto error_brelse;
1315 }
1316
1317 if (device->writeable) {
1318 lock_chunks(root);
1319 list_del_init(&device->dev_alloc_list);
1320 unlock_chunks(root);
1321 root->fs_info->fs_devices->rw_devices--;
1322 clear_super = true;
1323 }
1324
1325 ret = btrfs_shrink_device(device, 0);
1326 if (ret)
1327 goto error_undo;
1328
1329 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1330 if (ret)
1331 goto error_undo;
1332
1333 device->in_fs_metadata = 0;
1334 btrfs_scrub_cancel_dev(root, device);
1335
1336 /*
1337 * the device list mutex makes sure that we don't change
1338 * the device list while someone else is writing out all
1339 * the device supers.
1340 */
1341
1342 cur_devices = device->fs_devices;
1343 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1344 list_del_rcu(&device->dev_list);
1345
1346 device->fs_devices->num_devices--;
1347
1348 if (device->missing)
1349 root->fs_info->fs_devices->missing_devices--;
1350
1351 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1352 struct btrfs_device, dev_list);
1353 if (device->bdev == root->fs_info->sb->s_bdev)
1354 root->fs_info->sb->s_bdev = next_device->bdev;
1355 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1356 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1357
1358 if (device->bdev)
1359 device->fs_devices->open_devices--;
1360
1361 call_rcu(&device->rcu, free_device);
1362 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1363
1364 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1365 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1366
1367 if (cur_devices->open_devices == 0) {
1368 struct btrfs_fs_devices *fs_devices;
1369 fs_devices = root->fs_info->fs_devices;
1370 while (fs_devices) {
1371 if (fs_devices->seed == cur_devices)
1372 break;
1373 fs_devices = fs_devices->seed;
1374 }
1375 fs_devices->seed = cur_devices->seed;
1376 cur_devices->seed = NULL;
1377 lock_chunks(root);
1378 __btrfs_close_devices(cur_devices);
1379 unlock_chunks(root);
1380 free_fs_devices(cur_devices);
1381 }
1382
1383 /*
1384 * at this point, the device is zero sized. We want to
1385 * remove it from the devices list and zero out the old super
1386 */
1387 if (clear_super) {
1388 /* make sure this device isn't detected as part of
1389 * the FS anymore
1390 */
1391 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1392 set_buffer_dirty(bh);
1393 sync_dirty_buffer(bh);
1394 }
1395
1396 ret = 0;
1397
1398error_brelse:
1399 brelse(bh);
1400error_close:
1401 if (bdev)
1402 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1403out:
1404 mutex_unlock(&root->fs_info->volume_mutex);
1405 mutex_unlock(&uuid_mutex);
1406 return ret;
1407error_undo:
1408 if (device->writeable) {
1409 lock_chunks(root);
1410 list_add(&device->dev_alloc_list,
1411 &root->fs_info->fs_devices->alloc_list);
1412 unlock_chunks(root);
1413 root->fs_info->fs_devices->rw_devices++;
1414 }
1415 goto error_brelse;
1416}
1417
1418/*
1419 * does all the dirty work required for changing file system's UUID.
1420 */
1421static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root)
1423{
1424 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1425 struct btrfs_fs_devices *old_devices;
1426 struct btrfs_fs_devices *seed_devices;
1427 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1428 struct btrfs_device *device;
1429 u64 super_flags;
1430
1431 BUG_ON(!mutex_is_locked(&uuid_mutex));
1432 if (!fs_devices->seeding)
1433 return -EINVAL;
1434
1435 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1436 if (!seed_devices)
1437 return -ENOMEM;
1438
1439 old_devices = clone_fs_devices(fs_devices);
1440 if (IS_ERR(old_devices)) {
1441 kfree(seed_devices);
1442 return PTR_ERR(old_devices);
1443 }
1444
1445 list_add(&old_devices->list, &fs_uuids);
1446
1447 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1448 seed_devices->opened = 1;
1449 INIT_LIST_HEAD(&seed_devices->devices);
1450 INIT_LIST_HEAD(&seed_devices->alloc_list);
1451 mutex_init(&seed_devices->device_list_mutex);
1452
1453 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1454 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1455 synchronize_rcu);
1456 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1457
1458 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1459 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1460 device->fs_devices = seed_devices;
1461 }
1462
1463 fs_devices->seeding = 0;
1464 fs_devices->num_devices = 0;
1465 fs_devices->open_devices = 0;
1466 fs_devices->seed = seed_devices;
1467
1468 generate_random_uuid(fs_devices->fsid);
1469 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1470 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1471 super_flags = btrfs_super_flags(disk_super) &
1472 ~BTRFS_SUPER_FLAG_SEEDING;
1473 btrfs_set_super_flags(disk_super, super_flags);
1474
1475 return 0;
1476}
1477
1478/*
1479 * strore the expected generation for seed devices in device items.
1480 */
1481static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root)
1483{
1484 struct btrfs_path *path;
1485 struct extent_buffer *leaf;
1486 struct btrfs_dev_item *dev_item;
1487 struct btrfs_device *device;
1488 struct btrfs_key key;
1489 u8 fs_uuid[BTRFS_UUID_SIZE];
1490 u8 dev_uuid[BTRFS_UUID_SIZE];
1491 u64 devid;
1492 int ret;
1493
1494 path = btrfs_alloc_path();
1495 if (!path)
1496 return -ENOMEM;
1497
1498 root = root->fs_info->chunk_root;
1499 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1500 key.offset = 0;
1501 key.type = BTRFS_DEV_ITEM_KEY;
1502
1503 while (1) {
1504 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1505 if (ret < 0)
1506 goto error;
1507
1508 leaf = path->nodes[0];
1509next_slot:
1510 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1511 ret = btrfs_next_leaf(root, path);
1512 if (ret > 0)
1513 break;
1514 if (ret < 0)
1515 goto error;
1516 leaf = path->nodes[0];
1517 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1518 btrfs_release_path(path);
1519 continue;
1520 }
1521
1522 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1523 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1524 key.type != BTRFS_DEV_ITEM_KEY)
1525 break;
1526
1527 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1528 struct btrfs_dev_item);
1529 devid = btrfs_device_id(leaf, dev_item);
1530 read_extent_buffer(leaf, dev_uuid,
1531 (unsigned long)btrfs_device_uuid(dev_item),
1532 BTRFS_UUID_SIZE);
1533 read_extent_buffer(leaf, fs_uuid,
1534 (unsigned long)btrfs_device_fsid(dev_item),
1535 BTRFS_UUID_SIZE);
1536 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1537 BUG_ON(!device);
1538
1539 if (device->fs_devices->seeding) {
1540 btrfs_set_device_generation(leaf, dev_item,
1541 device->generation);
1542 btrfs_mark_buffer_dirty(leaf);
1543 }
1544
1545 path->slots[0]++;
1546 goto next_slot;
1547 }
1548 ret = 0;
1549error:
1550 btrfs_free_path(path);
1551 return ret;
1552}
1553
1554int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1555{
1556 struct request_queue *q;
1557 struct btrfs_trans_handle *trans;
1558 struct btrfs_device *device;
1559 struct block_device *bdev;
1560 struct list_head *devices;
1561 struct super_block *sb = root->fs_info->sb;
1562 u64 total_bytes;
1563 int seeding_dev = 0;
1564 int ret = 0;
1565
1566 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1567 return -EINVAL;
1568
1569 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1570 root->fs_info->bdev_holder);
1571 if (IS_ERR(bdev))
1572 return PTR_ERR(bdev);
1573
1574 if (root->fs_info->fs_devices->seeding) {
1575 seeding_dev = 1;
1576 down_write(&sb->s_umount);
1577 mutex_lock(&uuid_mutex);
1578 }
1579
1580 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1581 mutex_lock(&root->fs_info->volume_mutex);
1582
1583 devices = &root->fs_info->fs_devices->devices;
1584 /*
1585 * we have the volume lock, so we don't need the extra
1586 * device list mutex while reading the list here.
1587 */
1588 list_for_each_entry(device, devices, dev_list) {
1589 if (device->bdev == bdev) {
1590 ret = -EEXIST;
1591 goto error;
1592 }
1593 }
1594
1595 device = kzalloc(sizeof(*device), GFP_NOFS);
1596 if (!device) {
1597 /* we can safely leave the fs_devices entry around */
1598 ret = -ENOMEM;
1599 goto error;
1600 }
1601
1602 device->name = kstrdup(device_path, GFP_NOFS);
1603 if (!device->name) {
1604 kfree(device);
1605 ret = -ENOMEM;
1606 goto error;
1607 }
1608
1609 ret = find_next_devid(root, &device->devid);
1610 if (ret) {
1611 kfree(device->name);
1612 kfree(device);
1613 goto error;
1614 }
1615
1616 trans = btrfs_start_transaction(root, 0);
1617 if (IS_ERR(trans)) {
1618 kfree(device->name);
1619 kfree(device);
1620 ret = PTR_ERR(trans);
1621 goto error;
1622 }
1623
1624 lock_chunks(root);
1625
1626 q = bdev_get_queue(bdev);
1627 if (blk_queue_discard(q))
1628 device->can_discard = 1;
1629 device->writeable = 1;
1630 device->work.func = pending_bios_fn;
1631 generate_random_uuid(device->uuid);
1632 spin_lock_init(&device->io_lock);
1633 device->generation = trans->transid;
1634 device->io_width = root->sectorsize;
1635 device->io_align = root->sectorsize;
1636 device->sector_size = root->sectorsize;
1637 device->total_bytes = i_size_read(bdev->bd_inode);
1638 device->disk_total_bytes = device->total_bytes;
1639 device->dev_root = root->fs_info->dev_root;
1640 device->bdev = bdev;
1641 device->in_fs_metadata = 1;
1642 device->mode = FMODE_EXCL;
1643 set_blocksize(device->bdev, 4096);
1644
1645 if (seeding_dev) {
1646 sb->s_flags &= ~MS_RDONLY;
1647 ret = btrfs_prepare_sprout(trans, root);
1648 BUG_ON(ret);
1649 }
1650
1651 device->fs_devices = root->fs_info->fs_devices;
1652
1653 /*
1654 * we don't want write_supers to jump in here with our device
1655 * half setup
1656 */
1657 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1658 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1659 list_add(&device->dev_alloc_list,
1660 &root->fs_info->fs_devices->alloc_list);
1661 root->fs_info->fs_devices->num_devices++;
1662 root->fs_info->fs_devices->open_devices++;
1663 root->fs_info->fs_devices->rw_devices++;
1664 if (device->can_discard)
1665 root->fs_info->fs_devices->num_can_discard++;
1666 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1667
1668 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1669 root->fs_info->fs_devices->rotating = 1;
1670
1671 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1672 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1673 total_bytes + device->total_bytes);
1674
1675 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1676 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1677 total_bytes + 1);
1678 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1679
1680 if (seeding_dev) {
1681 ret = init_first_rw_device(trans, root, device);
1682 BUG_ON(ret);
1683 ret = btrfs_finish_sprout(trans, root);
1684 BUG_ON(ret);
1685 } else {
1686 ret = btrfs_add_device(trans, root, device);
1687 }
1688
1689 /*
1690 * we've got more storage, clear any full flags on the space
1691 * infos
1692 */
1693 btrfs_clear_space_info_full(root->fs_info);
1694
1695 unlock_chunks(root);
1696 btrfs_commit_transaction(trans, root);
1697
1698 if (seeding_dev) {
1699 mutex_unlock(&uuid_mutex);
1700 up_write(&sb->s_umount);
1701
1702 ret = btrfs_relocate_sys_chunks(root);
1703 BUG_ON(ret);
1704 }
1705out:
1706 mutex_unlock(&root->fs_info->volume_mutex);
1707 return ret;
1708error:
1709 blkdev_put(bdev, FMODE_EXCL);
1710 if (seeding_dev) {
1711 mutex_unlock(&uuid_mutex);
1712 up_write(&sb->s_umount);
1713 }
1714 goto out;
1715}
1716
1717static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1718 struct btrfs_device *device)
1719{
1720 int ret;
1721 struct btrfs_path *path;
1722 struct btrfs_root *root;
1723 struct btrfs_dev_item *dev_item;
1724 struct extent_buffer *leaf;
1725 struct btrfs_key key;
1726
1727 root = device->dev_root->fs_info->chunk_root;
1728
1729 path = btrfs_alloc_path();
1730 if (!path)
1731 return -ENOMEM;
1732
1733 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1734 key.type = BTRFS_DEV_ITEM_KEY;
1735 key.offset = device->devid;
1736
1737 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1738 if (ret < 0)
1739 goto out;
1740
1741 if (ret > 0) {
1742 ret = -ENOENT;
1743 goto out;
1744 }
1745
1746 leaf = path->nodes[0];
1747 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1748
1749 btrfs_set_device_id(leaf, dev_item, device->devid);
1750 btrfs_set_device_type(leaf, dev_item, device->type);
1751 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1752 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1753 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1754 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1755 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1756 btrfs_mark_buffer_dirty(leaf);
1757
1758out:
1759 btrfs_free_path(path);
1760 return ret;
1761}
1762
1763static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1764 struct btrfs_device *device, u64 new_size)
1765{
1766 struct btrfs_super_block *super_copy =
1767 &device->dev_root->fs_info->super_copy;
1768 u64 old_total = btrfs_super_total_bytes(super_copy);
1769 u64 diff = new_size - device->total_bytes;
1770
1771 if (!device->writeable)
1772 return -EACCES;
1773 if (new_size <= device->total_bytes)
1774 return -EINVAL;
1775
1776 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1777 device->fs_devices->total_rw_bytes += diff;
1778
1779 device->total_bytes = new_size;
1780 device->disk_total_bytes = new_size;
1781 btrfs_clear_space_info_full(device->dev_root->fs_info);
1782
1783 return btrfs_update_device(trans, device);
1784}
1785
1786int btrfs_grow_device(struct btrfs_trans_handle *trans,
1787 struct btrfs_device *device, u64 new_size)
1788{
1789 int ret;
1790 lock_chunks(device->dev_root);
1791 ret = __btrfs_grow_device(trans, device, new_size);
1792 unlock_chunks(device->dev_root);
1793 return ret;
1794}
1795
1796static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1797 struct btrfs_root *root,
1798 u64 chunk_tree, u64 chunk_objectid,
1799 u64 chunk_offset)
1800{
1801 int ret;
1802 struct btrfs_path *path;
1803 struct btrfs_key key;
1804
1805 root = root->fs_info->chunk_root;
1806 path = btrfs_alloc_path();
1807 if (!path)
1808 return -ENOMEM;
1809
1810 key.objectid = chunk_objectid;
1811 key.offset = chunk_offset;
1812 key.type = BTRFS_CHUNK_ITEM_KEY;
1813
1814 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1815 BUG_ON(ret);
1816
1817 ret = btrfs_del_item(trans, root, path);
1818
1819 btrfs_free_path(path);
1820 return ret;
1821}
1822
1823static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1824 chunk_offset)
1825{
1826 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1827 struct btrfs_disk_key *disk_key;
1828 struct btrfs_chunk *chunk;
1829 u8 *ptr;
1830 int ret = 0;
1831 u32 num_stripes;
1832 u32 array_size;
1833 u32 len = 0;
1834 u32 cur;
1835 struct btrfs_key key;
1836
1837 array_size = btrfs_super_sys_array_size(super_copy);
1838
1839 ptr = super_copy->sys_chunk_array;
1840 cur = 0;
1841
1842 while (cur < array_size) {
1843 disk_key = (struct btrfs_disk_key *)ptr;
1844 btrfs_disk_key_to_cpu(&key, disk_key);
1845
1846 len = sizeof(*disk_key);
1847
1848 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1849 chunk = (struct btrfs_chunk *)(ptr + len);
1850 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1851 len += btrfs_chunk_item_size(num_stripes);
1852 } else {
1853 ret = -EIO;
1854 break;
1855 }
1856 if (key.objectid == chunk_objectid &&
1857 key.offset == chunk_offset) {
1858 memmove(ptr, ptr + len, array_size - (cur + len));
1859 array_size -= len;
1860 btrfs_set_super_sys_array_size(super_copy, array_size);
1861 } else {
1862 ptr += len;
1863 cur += len;
1864 }
1865 }
1866 return ret;
1867}
1868
1869static int btrfs_relocate_chunk(struct btrfs_root *root,
1870 u64 chunk_tree, u64 chunk_objectid,
1871 u64 chunk_offset)
1872{
1873 struct extent_map_tree *em_tree;
1874 struct btrfs_root *extent_root;
1875 struct btrfs_trans_handle *trans;
1876 struct extent_map *em;
1877 struct map_lookup *map;
1878 int ret;
1879 int i;
1880
1881 root = root->fs_info->chunk_root;
1882 extent_root = root->fs_info->extent_root;
1883 em_tree = &root->fs_info->mapping_tree.map_tree;
1884
1885 ret = btrfs_can_relocate(extent_root, chunk_offset);
1886 if (ret)
1887 return -ENOSPC;
1888
1889 /* step one, relocate all the extents inside this chunk */
1890 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1891 if (ret)
1892 return ret;
1893
1894 trans = btrfs_start_transaction(root, 0);
1895 BUG_ON(IS_ERR(trans));
1896
1897 lock_chunks(root);
1898
1899 /*
1900 * step two, delete the device extents and the
1901 * chunk tree entries
1902 */
1903 read_lock(&em_tree->lock);
1904 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1905 read_unlock(&em_tree->lock);
1906
1907 BUG_ON(em->start > chunk_offset ||
1908 em->start + em->len < chunk_offset);
1909 map = (struct map_lookup *)em->bdev;
1910
1911 for (i = 0; i < map->num_stripes; i++) {
1912 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1913 map->stripes[i].physical);
1914 BUG_ON(ret);
1915
1916 if (map->stripes[i].dev) {
1917 ret = btrfs_update_device(trans, map->stripes[i].dev);
1918 BUG_ON(ret);
1919 }
1920 }
1921 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1922 chunk_offset);
1923
1924 BUG_ON(ret);
1925
1926 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1927
1928 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1929 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1930 BUG_ON(ret);
1931 }
1932
1933 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1934 BUG_ON(ret);
1935
1936 write_lock(&em_tree->lock);
1937 remove_extent_mapping(em_tree, em);
1938 write_unlock(&em_tree->lock);
1939
1940 kfree(map);
1941 em->bdev = NULL;
1942
1943 /* once for the tree */
1944 free_extent_map(em);
1945 /* once for us */
1946 free_extent_map(em);
1947
1948 unlock_chunks(root);
1949 btrfs_end_transaction(trans, root);
1950 return 0;
1951}
1952
1953static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1954{
1955 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1956 struct btrfs_path *path;
1957 struct extent_buffer *leaf;
1958 struct btrfs_chunk *chunk;
1959 struct btrfs_key key;
1960 struct btrfs_key found_key;
1961 u64 chunk_tree = chunk_root->root_key.objectid;
1962 u64 chunk_type;
1963 bool retried = false;
1964 int failed = 0;
1965 int ret;
1966
1967 path = btrfs_alloc_path();
1968 if (!path)
1969 return -ENOMEM;
1970
1971again:
1972 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1973 key.offset = (u64)-1;
1974 key.type = BTRFS_CHUNK_ITEM_KEY;
1975
1976 while (1) {
1977 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1978 if (ret < 0)
1979 goto error;
1980 BUG_ON(ret == 0);
1981
1982 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1983 key.type);
1984 if (ret …Large files files are truncated, but you can click here to view the full file