projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: return void in functions without error conditions
[deliverable/linux.git] / fs / btrfs / volumes.c
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 <linux/kthread.h>
27 #include <asm/div64.h>
28 #include "compat.h"
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37
38 static int init_first_rw_device(struct btrfs_trans_handle *trans,
39 struct btrfs_root *root,
40 struct btrfs_device *device);
41 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
42
43 static DEFINE_MUTEX(uuid_mutex);
44 static LIST_HEAD(fs_uuids);
45
46 static void lock_chunks(struct btrfs_root *root)
47 {
48 mutex_lock(&root->fs_info->chunk_mutex);
49 }
50
51 static void unlock_chunks(struct btrfs_root *root)
52 {
53 mutex_unlock(&root->fs_info->chunk_mutex);
54 }
55
56 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
57 {
58 struct btrfs_device *device;
59 WARN_ON(fs_devices->opened);
60 while (!list_empty(&fs_devices->devices)) {
61 device = list_entry(fs_devices->devices.next,
62 struct btrfs_device, dev_list);
63 list_del(&device->dev_list);
64 kfree(device->name);
65 kfree(device);
66 }
67 kfree(fs_devices);
68 }
69
70 void btrfs_cleanup_fs_uuids(void)
71 {
72 struct btrfs_fs_devices *fs_devices;
73
74 while (!list_empty(&fs_uuids)) {
75 fs_devices = list_entry(fs_uuids.next,
76 struct btrfs_fs_devices, list);
77 list_del(&fs_devices->list);
78 free_fs_devices(fs_devices);
79 }
80 }
81
82 static noinline struct btrfs_device *__find_device(struct list_head *head,
83 u64 devid, u8 *uuid)
84 {
85 struct btrfs_device *dev;
86
87 list_for_each_entry(dev, head, dev_list) {
88 if (dev->devid == devid &&
89 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
90 return dev;
91 }
92 }
93 return NULL;
94 }
95
96 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
97 {
98 struct btrfs_fs_devices *fs_devices;
99
100 list_for_each_entry(fs_devices, &fs_uuids, list) {
101 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
102 return fs_devices;
103 }
104 return NULL;
105 }
106
107 static void requeue_list(struct btrfs_pending_bios *pending_bios,
108 struct bio *head, struct bio *tail)
109 {
110
111 struct bio *old_head;
112
113 old_head = pending_bios->head;
114 pending_bios->head = head;
115 if (pending_bios->tail)
116 tail->bi_next = old_head;
117 else
118 pending_bios->tail = tail;
119 }
120
121 /*
122 * we try to collect pending bios for a device so we don't get a large
123 * number of procs sending bios down to the same device. This greatly
124 * improves the schedulers ability to collect and merge the bios.
125 *
126 * But, it also turns into a long list of bios to process and that is sure
127 * to eventually make the worker thread block. The solution here is to
128 * make some progress and then put this work struct back at the end of
129 * the list if the block device is congested. This way, multiple devices
130 * can make progress from a single worker thread.
131 */
132 static noinline void run_scheduled_bios(struct btrfs_device *device)
133 {
134 struct bio *pending;
135 struct backing_dev_info *bdi;
136 struct btrfs_fs_info *fs_info;
137 struct btrfs_pending_bios *pending_bios;
138 struct bio *tail;
139 struct bio *cur;
140 int again = 0;
141 unsigned long num_run;
142 unsigned long batch_run = 0;
143 unsigned long limit;
144 unsigned long last_waited = 0;
145 int force_reg = 0;
146 int sync_pending = 0;
147 struct blk_plug plug;
148
149 /*
150 * this function runs all the bios we've collected for
151 * a particular device. We don't want to wander off to
152 * another device without first sending all of these down.
153 * So, setup a plug here and finish it off before we return
154 */
155 blk_start_plug(&plug);
156
157 bdi = blk_get_backing_dev_info(device->bdev);
158 fs_info = device->dev_root->fs_info;
159 limit = btrfs_async_submit_limit(fs_info);
160 limit = limit * 2 / 3;
161
162 loop:
163 spin_lock(&device->io_lock);
164
165 loop_lock:
166 num_run = 0;
167
168 /* take all the bios off the list at once and process them
169 * later on (without the lock held). But, remember the
170 * tail and other pointers so the bios can be properly reinserted
171 * into the list if we hit congestion
172 */
173 if (!force_reg && device->pending_sync_bios.head) {
174 pending_bios = &device->pending_sync_bios;
175 force_reg = 1;
176 } else {
177 pending_bios = &device->pending_bios;
178 force_reg = 0;
179 }
180
181 pending = pending_bios->head;
182 tail = pending_bios->tail;
183 WARN_ON(pending && !tail);
184
185 /*
186 * if pending was null this time around, no bios need processing
187 * at all and we can stop. Otherwise it'll loop back up again
188 * and do an additional check so no bios are missed.
189 *
190 * device->running_pending is used to synchronize with the
191 * schedule_bio code.
192 */
193 if (device->pending_sync_bios.head == NULL &&
194 device->pending_bios.head == NULL) {
195 again = 0;
196 device->running_pending = 0;
197 } else {
198 again = 1;
199 device->running_pending = 1;
200 }
201
202 pending_bios->head = NULL;
203 pending_bios->tail = NULL;
204
205 spin_unlock(&device->io_lock);
206
207 while (pending) {
208
209 rmb();
210 /* we want to work on both lists, but do more bios on the
211 * sync list than the regular list
212 */
213 if ((num_run > 32 &&
214 pending_bios != &device->pending_sync_bios &&
215 device->pending_sync_bios.head) ||
216 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
217 device->pending_bios.head)) {
218 spin_lock(&device->io_lock);
219 requeue_list(pending_bios, pending, tail);
220 goto loop_lock;
221 }
222
223 cur = pending;
224 pending = pending->bi_next;
225 cur->bi_next = NULL;
226 atomic_dec(&fs_info->nr_async_bios);
227
228 if (atomic_read(&fs_info->nr_async_bios) < limit &&
229 waitqueue_active(&fs_info->async_submit_wait))
230 wake_up(&fs_info->async_submit_wait);
231
232 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
233
234 /*
235 * if we're doing the sync list, record that our
236 * plug has some sync requests on it
237 *
238 * If we're doing the regular list and there are
239 * sync requests sitting around, unplug before
240 * we add more
241 */
242 if (pending_bios == &device->pending_sync_bios) {
243 sync_pending = 1;
244 } else if (sync_pending) {
245 blk_finish_plug(&plug);
246 blk_start_plug(&plug);
247 sync_pending = 0;
248 }
249
250 btrfsic_submit_bio(cur->bi_rw, cur);
251 num_run++;
252 batch_run++;
253 if (need_resched())
254 cond_resched();
255
256 /*
257 * we made progress, there is more work to do and the bdi
258 * is now congested. Back off and let other work structs
259 * run instead
260 */
261 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
262 fs_info->fs_devices->open_devices > 1) {
263 struct io_context *ioc;
264
265 ioc = current->io_context;
266
267 /*
268 * the main goal here is that we don't want to
269 * block if we're going to be able to submit
270 * more requests without blocking.
271 *
272 * This code does two great things, it pokes into
273 * the elevator code from a filesystem _and_
274 * it makes assumptions about how batching works.
275 */
276 if (ioc && ioc->nr_batch_requests > 0 &&
277 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
278 (last_waited == 0 ||
279 ioc->last_waited == last_waited)) {
280 /*
281 * we want to go through our batch of
282 * requests and stop. So, we copy out
283 * the ioc->last_waited time and test
284 * against it before looping
285 */
286 last_waited = ioc->last_waited;
287 if (need_resched())
288 cond_resched();
289 continue;
290 }
291 spin_lock(&device->io_lock);
292 requeue_list(pending_bios, pending, tail);
293 device->running_pending = 1;
294
295 spin_unlock(&device->io_lock);
296 btrfs_requeue_work(&device->work);
297 goto done;
298 }
299 /* unplug every 64 requests just for good measure */
300 if (batch_run % 64 == 0) {
301 blk_finish_plug(&plug);
302 blk_start_plug(&plug);
303 sync_pending = 0;
304 }
305 }
306
307 cond_resched();
308 if (again)
309 goto loop;
310
311 spin_lock(&device->io_lock);
312 if (device->pending_bios.head || device->pending_sync_bios.head)
313 goto loop_lock;
314 spin_unlock(&device->io_lock);
315
316 done:
317 blk_finish_plug(&plug);
318 }
319
320 static void pending_bios_fn(struct btrfs_work *work)
321 {
322 struct btrfs_device *device;
323
324 device = container_of(work, struct btrfs_device, work);
325 run_scheduled_bios(device);
326 }
327
328 static noinline int device_list_add(const char *path,
329 struct btrfs_super_block *disk_super,
330 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
331 {
332 struct btrfs_device *device;
333 struct btrfs_fs_devices *fs_devices;
334 u64 found_transid = btrfs_super_generation(disk_super);
335 char *name;
336
337 fs_devices = find_fsid(disk_super->fsid);
338 if (!fs_devices) {
339 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
340 if (!fs_devices)
341 return -ENOMEM;
342 INIT_LIST_HEAD(&fs_devices->devices);
343 INIT_LIST_HEAD(&fs_devices->alloc_list);
344 list_add(&fs_devices->list, &fs_uuids);
345 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
346 fs_devices->latest_devid = devid;
347 fs_devices->latest_trans = found_transid;
348 mutex_init(&fs_devices->device_list_mutex);
349 device = NULL;
350 } else {
351 device = __find_device(&fs_devices->devices, devid,
352 disk_super->dev_item.uuid);
353 }
354 if (!device) {
355 if (fs_devices->opened)
356 return -EBUSY;
357
358 device = kzalloc(sizeof(*device), GFP_NOFS);
359 if (!device) {
360 /* we can safely leave the fs_devices entry around */
361 return -ENOMEM;
362 }
363 device->devid = devid;
364 device->work.func = pending_bios_fn;
365 memcpy(device->uuid, disk_super->dev_item.uuid,
366 BTRFS_UUID_SIZE);
367 spin_lock_init(&device->io_lock);
368 device->name = kstrdup(path, GFP_NOFS);
369 if (!device->name) {
370 kfree(device);
371 return -ENOMEM;
372 }
373 INIT_LIST_HEAD(&device->dev_alloc_list);
374
375 /* init readahead state */
376 spin_lock_init(&device->reada_lock);
377 device->reada_curr_zone = NULL;
378 atomic_set(&device->reada_in_flight, 0);
379 device->reada_next = 0;
380 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
381 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
382
383 mutex_lock(&fs_devices->device_list_mutex);
384 list_add_rcu(&device->dev_list, &fs_devices->devices);
385 mutex_unlock(&fs_devices->device_list_mutex);
386
387 device->fs_devices = fs_devices;
388 fs_devices->num_devices++;
389 } else if (!device->name || strcmp(device->name, path)) {
390 name = kstrdup(path, GFP_NOFS);
391 if (!name)
392 return -ENOMEM;
393 kfree(device->name);
394 device->name = name;
395 if (device->missing) {
396 fs_devices->missing_devices--;
397 device->missing = 0;
398 }
399 }
400
401 if (found_transid > fs_devices->latest_trans) {
402 fs_devices->latest_devid = devid;
403 fs_devices->latest_trans = found_transid;
404 }
405 *fs_devices_ret = fs_devices;
406 return 0;
407 }
408
409 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
410 {
411 struct btrfs_fs_devices *fs_devices;
412 struct btrfs_device *device;
413 struct btrfs_device *orig_dev;
414
415 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
416 if (!fs_devices)
417 return ERR_PTR(-ENOMEM);
418
419 INIT_LIST_HEAD(&fs_devices->devices);
420 INIT_LIST_HEAD(&fs_devices->alloc_list);
421 INIT_LIST_HEAD(&fs_devices->list);
422 mutex_init(&fs_devices->device_list_mutex);
423 fs_devices->latest_devid = orig->latest_devid;
424 fs_devices->latest_trans = orig->latest_trans;
425 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
426
427 /* We have held the volume lock, it is safe to get the devices. */
428 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
429 device = kzalloc(sizeof(*device), GFP_NOFS);
430 if (!device)
431 goto error;
432
433 device->name = kstrdup(orig_dev->name, GFP_NOFS);
434 if (!device->name) {
435 kfree(device);
436 goto error;
437 }
438
439 device->devid = orig_dev->devid;
440 device->work.func = pending_bios_fn;
441 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
442 spin_lock_init(&device->io_lock);
443 INIT_LIST_HEAD(&device->dev_list);
444 INIT_LIST_HEAD(&device->dev_alloc_list);
445
446 list_add(&device->dev_list, &fs_devices->devices);
447 device->fs_devices = fs_devices;
448 fs_devices->num_devices++;
449 }
450 return fs_devices;
451 error:
452 free_fs_devices(fs_devices);
453 return ERR_PTR(-ENOMEM);
454 }
455
456 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
457 {
458 struct btrfs_device *device, *next;
459
460 struct block_device *latest_bdev = NULL;
461 u64 latest_devid = 0;
462 u64 latest_transid = 0;
463
464 mutex_lock(&uuid_mutex);
465 again:
466 /* This is the initialized path, it is safe to release the devices. */
467 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
468 if (device->in_fs_metadata) {
469 if (!latest_transid ||
470 device->generation > latest_transid) {
471 latest_devid = device->devid;
472 latest_transid = device->generation;
473 latest_bdev = device->bdev;
474 }
475 continue;
476 }
477
478 if (device->bdev) {
479 blkdev_put(device->bdev, device->mode);
480 device->bdev = NULL;
481 fs_devices->open_devices--;
482 }
483 if (device->writeable) {
484 list_del_init(&device->dev_alloc_list);
485 device->writeable = 0;
486 fs_devices->rw_devices--;
487 }
488 list_del_init(&device->dev_list);
489 fs_devices->num_devices--;
490 kfree(device->name);
491 kfree(device);
492 }
493
494 if (fs_devices->seed) {
495 fs_devices = fs_devices->seed;
496 goto again;
497 }
498
499 fs_devices->latest_bdev = latest_bdev;
500 fs_devices->latest_devid = latest_devid;
501 fs_devices->latest_trans = latest_transid;
502
503 mutex_unlock(&uuid_mutex);
504 }
505
506 static void __free_device(struct work_struct *work)
507 {
508 struct btrfs_device *device;
509
510 device = container_of(work, struct btrfs_device, rcu_work);
511
512 if (device->bdev)
513 blkdev_put(device->bdev, device->mode);
514
515 kfree(device->name);
516 kfree(device);
517 }
518
519 static void free_device(struct rcu_head *head)
520 {
521 struct btrfs_device *device;
522
523 device = container_of(head, struct btrfs_device, rcu);
524
525 INIT_WORK(&device->rcu_work, __free_device);
526 schedule_work(&device->rcu_work);
527 }
528
529 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
530 {
531 struct btrfs_device *device;
532
533 if (--fs_devices->opened > 0)
534 return 0;
535
536 mutex_lock(&fs_devices->device_list_mutex);
537 list_for_each_entry(device, &fs_devices->devices, dev_list) {
538 struct btrfs_device *new_device;
539
540 if (device->bdev)
541 fs_devices->open_devices--;
542
543 if (device->writeable) {
544 list_del_init(&device->dev_alloc_list);
545 fs_devices->rw_devices--;
546 }
547
548 if (device->can_discard)
549 fs_devices->num_can_discard--;
550
551 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
552 BUG_ON(!new_device);
553 memcpy(new_device, device, sizeof(*new_device));
554 new_device->name = kstrdup(device->name, GFP_NOFS);
555 BUG_ON(device->name && !new_device->name);
556 new_device->bdev = NULL;
557 new_device->writeable = 0;
558 new_device->in_fs_metadata = 0;
559 new_device->can_discard = 0;
560 list_replace_rcu(&device->dev_list, &new_device->dev_list);
561
562 call_rcu(&device->rcu, free_device);
563 }
564 mutex_unlock(&fs_devices->device_list_mutex);
565
566 WARN_ON(fs_devices->open_devices);
567 WARN_ON(fs_devices->rw_devices);
568 fs_devices->opened = 0;
569 fs_devices->seeding = 0;
570
571 return 0;
572 }
573
574 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
575 {
576 struct btrfs_fs_devices *seed_devices = NULL;
577 int ret;
578
579 mutex_lock(&uuid_mutex);
580 ret = __btrfs_close_devices(fs_devices);
581 if (!fs_devices->opened) {
582 seed_devices = fs_devices->seed;
583 fs_devices->seed = NULL;
584 }
585 mutex_unlock(&uuid_mutex);
586
587 while (seed_devices) {
588 fs_devices = seed_devices;
589 seed_devices = fs_devices->seed;
590 __btrfs_close_devices(fs_devices);
591 free_fs_devices(fs_devices);
592 }
593 return ret;
594 }
595
596 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
597 fmode_t flags, void *holder)
598 {
599 struct request_queue *q;
600 struct block_device *bdev;
601 struct list_head *head = &fs_devices->devices;
602 struct btrfs_device *device;
603 struct block_device *latest_bdev = NULL;
604 struct buffer_head *bh;
605 struct btrfs_super_block *disk_super;
606 u64 latest_devid = 0;
607 u64 latest_transid = 0;
608 u64 devid;
609 int seeding = 1;
610 int ret = 0;
611
612 flags |= FMODE_EXCL;
613
614 list_for_each_entry(device, head, dev_list) {
615 if (device->bdev)
616 continue;
617 if (!device->name)
618 continue;
619
620 bdev = blkdev_get_by_path(device->name, flags, holder);
621 if (IS_ERR(bdev)) {
622 printk(KERN_INFO "open %s failed\n", device->name);
623 goto error;
624 }
625 set_blocksize(bdev, 4096);
626
627 bh = btrfs_read_dev_super(bdev);
628 if (!bh)
629 goto error_close;
630
631 disk_super = (struct btrfs_super_block *)bh->b_data;
632 devid = btrfs_stack_device_id(&disk_super->dev_item);
633 if (devid != device->devid)
634 goto error_brelse;
635
636 if (memcmp(device->uuid, disk_super->dev_item.uuid,
637 BTRFS_UUID_SIZE))
638 goto error_brelse;
639
640 device->generation = btrfs_super_generation(disk_super);
641 if (!latest_transid || device->generation > latest_transid) {
642 latest_devid = devid;
643 latest_transid = device->generation;
644 latest_bdev = bdev;
645 }
646
647 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
648 device->writeable = 0;
649 } else {
650 device->writeable = !bdev_read_only(bdev);
651 seeding = 0;
652 }
653
654 q = bdev_get_queue(bdev);
655 if (blk_queue_discard(q)) {
656 device->can_discard = 1;
657 fs_devices->num_can_discard++;
658 }
659
660 device->bdev = bdev;
661 device->in_fs_metadata = 0;
662 device->mode = flags;
663
664 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
665 fs_devices->rotating = 1;
666
667 fs_devices->open_devices++;
668 if (device->writeable) {
669 fs_devices->rw_devices++;
670 list_add(&device->dev_alloc_list,
671 &fs_devices->alloc_list);
672 }
673 brelse(bh);
674 continue;
675
676 error_brelse:
677 brelse(bh);
678 error_close:
679 blkdev_put(bdev, flags);
680 error:
681 continue;
682 }
683 if (fs_devices->open_devices == 0) {
684 ret = -EINVAL;
685 goto out;
686 }
687 fs_devices->seeding = seeding;
688 fs_devices->opened = 1;
689 fs_devices->latest_bdev = latest_bdev;
690 fs_devices->latest_devid = latest_devid;
691 fs_devices->latest_trans = latest_transid;
692 fs_devices->total_rw_bytes = 0;
693 out:
694 return ret;
695 }
696
697 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
698 fmode_t flags, void *holder)
699 {
700 int ret;
701
702 mutex_lock(&uuid_mutex);
703 if (fs_devices->opened) {
704 fs_devices->opened++;
705 ret = 0;
706 } else {
707 ret = __btrfs_open_devices(fs_devices, flags, holder);
708 }
709 mutex_unlock(&uuid_mutex);
710 return ret;
711 }
712
713 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
714 struct btrfs_fs_devices **fs_devices_ret)
715 {
716 struct btrfs_super_block *disk_super;
717 struct block_device *bdev;
718 struct buffer_head *bh;
719 int ret;
720 u64 devid;
721 u64 transid;
722
723 flags |= FMODE_EXCL;
724 bdev = blkdev_get_by_path(path, flags, holder);
725
726 if (IS_ERR(bdev)) {
727 ret = PTR_ERR(bdev);
728 goto error;
729 }
730
731 mutex_lock(&uuid_mutex);
732 ret = set_blocksize(bdev, 4096);
733 if (ret)
734 goto error_close;
735 bh = btrfs_read_dev_super(bdev);
736 if (!bh) {
737 ret = -EINVAL;
738 goto error_close;
739 }
740 disk_super = (struct btrfs_super_block *)bh->b_data;
741 devid = btrfs_stack_device_id(&disk_super->dev_item);
742 transid = btrfs_super_generation(disk_super);
743 if (disk_super->label[0])
744 printk(KERN_INFO "device label %s ", disk_super->label);
745 else
746 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
747 printk(KERN_CONT "devid %llu transid %llu %s\n",
748 (unsigned long long)devid, (unsigned long long)transid, path);
749 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
750
751 brelse(bh);
752 error_close:
753 mutex_unlock(&uuid_mutex);
754 blkdev_put(bdev, flags);
755 error:
756 return ret;
757 }
758
759 /* helper to account the used device space in the range */
760 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
761 u64 end, u64 *length)
762 {
763 struct btrfs_key key;
764 struct btrfs_root *root = device->dev_root;
765 struct btrfs_dev_extent *dev_extent;
766 struct btrfs_path *path;
767 u64 extent_end;
768 int ret;
769 int slot;
770 struct extent_buffer *l;
771
772 *length = 0;
773
774 if (start >= device->total_bytes)
775 return 0;
776
777 path = btrfs_alloc_path();
778 if (!path)
779 return -ENOMEM;
780 path->reada = 2;
781
782 key.objectid = device->devid;
783 key.offset = start;
784 key.type = BTRFS_DEV_EXTENT_KEY;
785
786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
787 if (ret < 0)
788 goto out;
789 if (ret > 0) {
790 ret = btrfs_previous_item(root, path, key.objectid, key.type);
791 if (ret < 0)
792 goto out;
793 }
794
795 while (1) {
796 l = path->nodes[0];
797 slot = path->slots[0];
798 if (slot >= btrfs_header_nritems(l)) {
799 ret = btrfs_next_leaf(root, path);
800 if (ret == 0)
801 continue;
802 if (ret < 0)
803 goto out;
804
805 break;
806 }
807 btrfs_item_key_to_cpu(l, &key, slot);
808
809 if (key.objectid < device->devid)
810 goto next;
811
812 if (key.objectid > device->devid)
813 break;
814
815 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
816 goto next;
817
818 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
819 extent_end = key.offset + btrfs_dev_extent_length(l,
820 dev_extent);
821 if (key.offset <= start && extent_end > end) {
822 *length = end - start + 1;
823 break;
824 } else if (key.offset <= start && extent_end > start)
825 *length += extent_end - start;
826 else if (key.offset > start && extent_end <= end)
827 *length += extent_end - key.offset;
828 else if (key.offset > start && key.offset <= end) {
829 *length += end - key.offset + 1;
830 break;
831 } else if (key.offset > end)
832 break;
833
834 next:
835 path->slots[0]++;
836 }
837 ret = 0;
838 out:
839 btrfs_free_path(path);
840 return ret;
841 }
842
843 /*
844 * find_free_dev_extent - find free space in the specified device
845 * @device: the device which we search the free space in
846 * @num_bytes: the size of the free space that we need
847 * @start: store the start of the free space.
848 * @len: the size of the free space. that we find, or the size of the max
849 * free space if we don't find suitable free space
850 *
851 * this uses a pretty simple search, the expectation is that it is
852 * called very infrequently and that a given device has a small number
853 * of extents
854 *
855 * @start is used to store the start of the free space if we find. But if we
856 * don't find suitable free space, it will be used to store the start position
857 * of the max free space.
858 *
859 * @len is used to store the size of the free space that we find.
860 * But if we don't find suitable free space, it is used to store the size of
861 * the max free space.
862 */
863 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
864 u64 *start, u64 *len)
865 {
866 struct btrfs_key key;
867 struct btrfs_root *root = device->dev_root;
868 struct btrfs_dev_extent *dev_extent;
869 struct btrfs_path *path;
870 u64 hole_size;
871 u64 max_hole_start;
872 u64 max_hole_size;
873 u64 extent_end;
874 u64 search_start;
875 u64 search_end = device->total_bytes;
876 int ret;
877 int slot;
878 struct extent_buffer *l;
879
880 /* FIXME use last free of some kind */
881
882 /* we don't want to overwrite the superblock on the drive,
883 * so we make sure to start at an offset of at least 1MB
884 */
885 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
886
887 max_hole_start = search_start;
888 max_hole_size = 0;
889 hole_size = 0;
890
891 if (search_start >= search_end) {
892 ret = -ENOSPC;
893 goto error;
894 }
895
896 path = btrfs_alloc_path();
897 if (!path) {
898 ret = -ENOMEM;
899 goto error;
900 }
901 path->reada = 2;
902
903 key.objectid = device->devid;
904 key.offset = search_start;
905 key.type = BTRFS_DEV_EXTENT_KEY;
906
907 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
908 if (ret < 0)
909 goto out;
910 if (ret > 0) {
911 ret = btrfs_previous_item(root, path, key.objectid, key.type);
912 if (ret < 0)
913 goto out;
914 }
915
916 while (1) {
917 l = path->nodes[0];
918 slot = path->slots[0];
919 if (slot >= btrfs_header_nritems(l)) {
920 ret = btrfs_next_leaf(root, path);
921 if (ret == 0)
922 continue;
923 if (ret < 0)
924 goto out;
925
926 break;
927 }
928 btrfs_item_key_to_cpu(l, &key, slot);
929
930 if (key.objectid < device->devid)
931 goto next;
932
933 if (key.objectid > device->devid)
934 break;
935
936 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
937 goto next;
938
939 if (key.offset > search_start) {
940 hole_size = key.offset - search_start;
941
942 if (hole_size > max_hole_size) {
943 max_hole_start = search_start;
944 max_hole_size = hole_size;
945 }
946
947 /*
948 * If this free space is greater than which we need,
949 * it must be the max free space that we have found
950 * until now, so max_hole_start must point to the start
951 * of this free space and the length of this free space
952 * is stored in max_hole_size. Thus, we return
953 * max_hole_start and max_hole_size and go back to the
954 * caller.
955 */
956 if (hole_size >= num_bytes) {
957 ret = 0;
958 goto out;
959 }
960 }
961
962 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
963 extent_end = key.offset + btrfs_dev_extent_length(l,
964 dev_extent);
965 if (extent_end > search_start)
966 search_start = extent_end;
967 next:
968 path->slots[0]++;
969 cond_resched();
970 }
971
972 /*
973 * At this point, search_start should be the end of
974 * allocated dev extents, and when shrinking the device,
975 * search_end may be smaller than search_start.
976 */
977 if (search_end > search_start)
978 hole_size = search_end - search_start;
979
980 if (hole_size > max_hole_size) {
981 max_hole_start = search_start;
982 max_hole_size = hole_size;
983 }
984
985 /* See above. */
986 if (hole_size < num_bytes)
987 ret = -ENOSPC;
988 else
989 ret = 0;
990
991 out:
992 btrfs_free_path(path);
993 error:
994 *start = max_hole_start;
995 if (len)
996 *len = max_hole_size;
997 return ret;
998 }
999
1000 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1001 struct btrfs_device *device,
1002 u64 start)
1003 {
1004 int ret;
1005 struct btrfs_path *path;
1006 struct btrfs_root *root = device->dev_root;
1007 struct btrfs_key key;
1008 struct btrfs_key found_key;
1009 struct extent_buffer *leaf = NULL;
1010 struct btrfs_dev_extent *extent = NULL;
1011
1012 path = btrfs_alloc_path();
1013 if (!path)
1014 return -ENOMEM;
1015
1016 key.objectid = device->devid;
1017 key.offset = start;
1018 key.type = BTRFS_DEV_EXTENT_KEY;
1019 again:
1020 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1021 if (ret > 0) {
1022 ret = btrfs_previous_item(root, path, key.objectid,
1023 BTRFS_DEV_EXTENT_KEY);
1024 if (ret)
1025 goto out;
1026 leaf = path->nodes[0];
1027 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1028 extent = btrfs_item_ptr(leaf, path->slots[0],
1029 struct btrfs_dev_extent);
1030 BUG_ON(found_key.offset > start || found_key.offset +
1031 btrfs_dev_extent_length(leaf, extent) < start);
1032 key = found_key;
1033 btrfs_release_path(path);
1034 goto again;
1035 } else if (ret == 0) {
1036 leaf = path->nodes[0];
1037 extent = btrfs_item_ptr(leaf, path->slots[0],
1038 struct btrfs_dev_extent);
1039 }
1040 BUG_ON(ret);
1041
1042 if (device->bytes_used > 0) {
1043 u64 len = btrfs_dev_extent_length(leaf, extent);
1044 device->bytes_used -= len;
1045 spin_lock(&root->fs_info->free_chunk_lock);
1046 root->fs_info->free_chunk_space += len;
1047 spin_unlock(&root->fs_info->free_chunk_lock);
1048 }
1049 ret = btrfs_del_item(trans, root, path);
1050
1051 out:
1052 btrfs_free_path(path);
1053 return ret;
1054 }
1055
1056 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1057 struct btrfs_device *device,
1058 u64 chunk_tree, u64 chunk_objectid,
1059 u64 chunk_offset, u64 start, u64 num_bytes)
1060 {
1061 int ret;
1062 struct btrfs_path *path;
1063 struct btrfs_root *root = device->dev_root;
1064 struct btrfs_dev_extent *extent;
1065 struct extent_buffer *leaf;
1066 struct btrfs_key key;
1067
1068 WARN_ON(!device->in_fs_metadata);
1069 path = btrfs_alloc_path();
1070 if (!path)
1071 return -ENOMEM;
1072
1073 key.objectid = device->devid;
1074 key.offset = start;
1075 key.type = BTRFS_DEV_EXTENT_KEY;
1076 ret = btrfs_insert_empty_item(trans, root, path, &key,
1077 sizeof(*extent));
1078 BUG_ON(ret);
1079
1080 leaf = path->nodes[0];
1081 extent = btrfs_item_ptr(leaf, path->slots[0],
1082 struct btrfs_dev_extent);
1083 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1084 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1085 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1086
1087 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1088 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1089 BTRFS_UUID_SIZE);
1090
1091 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1092 btrfs_mark_buffer_dirty(leaf);
1093 btrfs_free_path(path);
1094 return ret;
1095 }
1096
1097 static noinline int find_next_chunk(struct btrfs_root *root,
1098 u64 objectid, u64 *offset)
1099 {
1100 struct btrfs_path *path;
1101 int ret;
1102 struct btrfs_key key;
1103 struct btrfs_chunk *chunk;
1104 struct btrfs_key found_key;
1105
1106 path = btrfs_alloc_path();
1107 if (!path)
1108 return -ENOMEM;
1109
1110 key.objectid = objectid;
1111 key.offset = (u64)-1;
1112 key.type = BTRFS_CHUNK_ITEM_KEY;
1113
1114 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1115 if (ret < 0)
1116 goto error;
1117
1118 BUG_ON(ret == 0);
1119
1120 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1121 if (ret) {
1122 *offset = 0;
1123 } else {
1124 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1125 path->slots[0]);
1126 if (found_key.objectid != objectid)
1127 *offset = 0;
1128 else {
1129 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1130 struct btrfs_chunk);
1131 *offset = found_key.offset +
1132 btrfs_chunk_length(path->nodes[0], chunk);
1133 }
1134 }
1135 ret = 0;
1136 error:
1137 btrfs_free_path(path);
1138 return ret;
1139 }
1140
1141 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1142 {
1143 int ret;
1144 struct btrfs_key key;
1145 struct btrfs_key found_key;
1146 struct btrfs_path *path;
1147
1148 root = root->fs_info->chunk_root;
1149
1150 path = btrfs_alloc_path();
1151 if (!path)
1152 return -ENOMEM;
1153
1154 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1155 key.type = BTRFS_DEV_ITEM_KEY;
1156 key.offset = (u64)-1;
1157
1158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1159 if (ret < 0)
1160 goto error;
1161
1162 BUG_ON(ret == 0);
1163
1164 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1165 BTRFS_DEV_ITEM_KEY);
1166 if (ret) {
1167 *objectid = 1;
1168 } else {
1169 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1170 path->slots[0]);
1171 *objectid = found_key.offset + 1;
1172 }
1173 ret = 0;
1174 error:
1175 btrfs_free_path(path);
1176 return ret;
1177 }
1178
1179 /*
1180 * the device information is stored in the chunk root
1181 * the btrfs_device struct should be fully filled in
1182 */
1183 int btrfs_add_device(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_device *device)
1186 {
1187 int ret;
1188 struct btrfs_path *path;
1189 struct btrfs_dev_item *dev_item;
1190 struct extent_buffer *leaf;
1191 struct btrfs_key key;
1192 unsigned long ptr;
1193
1194 root = root->fs_info->chunk_root;
1195
1196 path = btrfs_alloc_path();
1197 if (!path)
1198 return -ENOMEM;
1199
1200 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1201 key.type = BTRFS_DEV_ITEM_KEY;
1202 key.offset = device->devid;
1203
1204 ret = btrfs_insert_empty_item(trans, root, path, &key,
1205 sizeof(*dev_item));
1206 if (ret)
1207 goto out;
1208
1209 leaf = path->nodes[0];
1210 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1211
1212 btrfs_set_device_id(leaf, dev_item, device->devid);
1213 btrfs_set_device_generation(leaf, dev_item, 0);
1214 btrfs_set_device_type(leaf, dev_item, device->type);
1215 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1216 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1217 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1218 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1219 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1220 btrfs_set_device_group(leaf, dev_item, 0);
1221 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1222 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1223 btrfs_set_device_start_offset(leaf, dev_item, 0);
1224
1225 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1226 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1227 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1228 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1229 btrfs_mark_buffer_dirty(leaf);
1230
1231 ret = 0;
1232 out:
1233 btrfs_free_path(path);
1234 return ret;
1235 }
1236
1237 static int btrfs_rm_dev_item(struct btrfs_root *root,
1238 struct btrfs_device *device)
1239 {
1240 int ret;
1241 struct btrfs_path *path;
1242 struct btrfs_key key;
1243 struct btrfs_trans_handle *trans;
1244
1245 root = root->fs_info->chunk_root;
1246
1247 path = btrfs_alloc_path();
1248 if (!path)
1249 return -ENOMEM;
1250
1251 trans = btrfs_start_transaction(root, 0);
1252 if (IS_ERR(trans)) {
1253 btrfs_free_path(path);
1254 return PTR_ERR(trans);
1255 }
1256 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1257 key.type = BTRFS_DEV_ITEM_KEY;
1258 key.offset = device->devid;
1259 lock_chunks(root);
1260
1261 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1262 if (ret < 0)
1263 goto out;
1264
1265 if (ret > 0) {
1266 ret = -ENOENT;
1267 goto out;
1268 }
1269
1270 ret = btrfs_del_item(trans, root, path);
1271 if (ret)
1272 goto out;
1273 out:
1274 btrfs_free_path(path);
1275 unlock_chunks(root);
1276 btrfs_commit_transaction(trans, root);
1277 return ret;
1278 }
1279
1280 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1281 {
1282 struct btrfs_device *device;
1283 struct btrfs_device *next_device;
1284 struct block_device *bdev;
1285 struct buffer_head *bh = NULL;
1286 struct btrfs_super_block *disk_super;
1287 struct btrfs_fs_devices *cur_devices;
1288 u64 all_avail;
1289 u64 devid;
1290 u64 num_devices;
1291 u8 *dev_uuid;
1292 int ret = 0;
1293 bool clear_super = false;
1294
1295 mutex_lock(&uuid_mutex);
1296
1297 all_avail = root->fs_info->avail_data_alloc_bits |
1298 root->fs_info->avail_system_alloc_bits |
1299 root->fs_info->avail_metadata_alloc_bits;
1300
1301 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1302 root->fs_info->fs_devices->num_devices <= 4) {
1303 printk(KERN_ERR "btrfs: unable to go below four devices "
1304 "on raid10\n");
1305 ret = -EINVAL;
1306 goto out;
1307 }
1308
1309 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1310 root->fs_info->fs_devices->num_devices <= 2) {
1311 printk(KERN_ERR "btrfs: unable to go below two "
1312 "devices on raid1\n");
1313 ret = -EINVAL;
1314 goto out;
1315 }
1316
1317 if (strcmp(device_path, "missing") == 0) {
1318 struct list_head *devices;
1319 struct btrfs_device *tmp;
1320
1321 device = NULL;
1322 devices = &root->fs_info->fs_devices->devices;
1323 /*
1324 * It is safe to read the devices since the volume_mutex
1325 * is held.
1326 */
1327 list_for_each_entry(tmp, devices, dev_list) {
1328 if (tmp->in_fs_metadata && !tmp->bdev) {
1329 device = tmp;
1330 break;
1331 }
1332 }
1333 bdev = NULL;
1334 bh = NULL;
1335 disk_super = NULL;
1336 if (!device) {
1337 printk(KERN_ERR "btrfs: no missing devices found to "
1338 "remove\n");
1339 goto out;
1340 }
1341 } else {
1342 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1343 root->fs_info->bdev_holder);
1344 if (IS_ERR(bdev)) {
1345 ret = PTR_ERR(bdev);
1346 goto out;
1347 }
1348
1349 set_blocksize(bdev, 4096);
1350 bh = btrfs_read_dev_super(bdev);
1351 if (!bh) {
1352 ret = -EINVAL;
1353 goto error_close;
1354 }
1355 disk_super = (struct btrfs_super_block *)bh->b_data;
1356 devid = btrfs_stack_device_id(&disk_super->dev_item);
1357 dev_uuid = disk_super->dev_item.uuid;
1358 device = btrfs_find_device(root, devid, dev_uuid,
1359 disk_super->fsid);
1360 if (!device) {
1361 ret = -ENOENT;
1362 goto error_brelse;
1363 }
1364 }
1365
1366 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1367 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1368 "device\n");
1369 ret = -EINVAL;
1370 goto error_brelse;
1371 }
1372
1373 if (device->writeable) {
1374 lock_chunks(root);
1375 list_del_init(&device->dev_alloc_list);
1376 unlock_chunks(root);
1377 root->fs_info->fs_devices->rw_devices--;
1378 clear_super = true;
1379 }
1380
1381 ret = btrfs_shrink_device(device, 0);
1382 if (ret)
1383 goto error_undo;
1384
1385 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1386 if (ret)
1387 goto error_undo;
1388
1389 spin_lock(&root->fs_info->free_chunk_lock);
1390 root->fs_info->free_chunk_space = device->total_bytes -
1391 device->bytes_used;
1392 spin_unlock(&root->fs_info->free_chunk_lock);
1393
1394 device->in_fs_metadata = 0;
1395 btrfs_scrub_cancel_dev(root, device);
1396
1397 /*
1398 * the device list mutex makes sure that we don't change
1399 * the device list while someone else is writing out all
1400 * the device supers.
1401 */
1402
1403 cur_devices = device->fs_devices;
1404 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1405 list_del_rcu(&device->dev_list);
1406
1407 device->fs_devices->num_devices--;
1408
1409 if (device->missing)
1410 root->fs_info->fs_devices->missing_devices--;
1411
1412 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1413 struct btrfs_device, dev_list);
1414 if (device->bdev == root->fs_info->sb->s_bdev)
1415 root->fs_info->sb->s_bdev = next_device->bdev;
1416 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1417 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1418
1419 if (device->bdev)
1420 device->fs_devices->open_devices--;
1421
1422 call_rcu(&device->rcu, free_device);
1423 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1424
1425 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1426 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1427
1428 if (cur_devices->open_devices == 0) {
1429 struct btrfs_fs_devices *fs_devices;
1430 fs_devices = root->fs_info->fs_devices;
1431 while (fs_devices) {
1432 if (fs_devices->seed == cur_devices)
1433 break;
1434 fs_devices = fs_devices->seed;
1435 }
1436 fs_devices->seed = cur_devices->seed;
1437 cur_devices->seed = NULL;
1438 lock_chunks(root);
1439 __btrfs_close_devices(cur_devices);
1440 unlock_chunks(root);
1441 free_fs_devices(cur_devices);
1442 }
1443
1444 /*
1445 * at this point, the device is zero sized. We want to
1446 * remove it from the devices list and zero out the old super
1447 */
1448 if (clear_super) {
1449 /* make sure this device isn't detected as part of
1450 * the FS anymore
1451 */
1452 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1453 set_buffer_dirty(bh);
1454 sync_dirty_buffer(bh);
1455 }
1456
1457 ret = 0;
1458
1459 error_brelse:
1460 brelse(bh);
1461 error_close:
1462 if (bdev)
1463 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1464 out:
1465 mutex_unlock(&uuid_mutex);
1466 return ret;
1467 error_undo:
1468 if (device->writeable) {
1469 lock_chunks(root);
1470 list_add(&device->dev_alloc_list,
1471 &root->fs_info->fs_devices->alloc_list);
1472 unlock_chunks(root);
1473 root->fs_info->fs_devices->rw_devices++;
1474 }
1475 goto error_brelse;
1476 }
1477
1478 /*
1479 * does all the dirty work required for changing file system's UUID.
1480 */
1481 static int btrfs_prepare_sprout(struct btrfs_root *root)
1482 {
1483 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1484 struct btrfs_fs_devices *old_devices;
1485 struct btrfs_fs_devices *seed_devices;
1486 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1487 struct btrfs_device *device;
1488 u64 super_flags;
1489
1490 BUG_ON(!mutex_is_locked(&uuid_mutex));
1491 if (!fs_devices->seeding)
1492 return -EINVAL;
1493
1494 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1495 if (!seed_devices)
1496 return -ENOMEM;
1497
1498 old_devices = clone_fs_devices(fs_devices);
1499 if (IS_ERR(old_devices)) {
1500 kfree(seed_devices);
1501 return PTR_ERR(old_devices);
1502 }
1503
1504 list_add(&old_devices->list, &fs_uuids);
1505
1506 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1507 seed_devices->opened = 1;
1508 INIT_LIST_HEAD(&seed_devices->devices);
1509 INIT_LIST_HEAD(&seed_devices->alloc_list);
1510 mutex_init(&seed_devices->device_list_mutex);
1511
1512 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1513 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1514 synchronize_rcu);
1515 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1516
1517 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1518 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1519 device->fs_devices = seed_devices;
1520 }
1521
1522 fs_devices->seeding = 0;
1523 fs_devices->num_devices = 0;
1524 fs_devices->open_devices = 0;
1525 fs_devices->seed = seed_devices;
1526
1527 generate_random_uuid(fs_devices->fsid);
1528 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1529 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1530 super_flags = btrfs_super_flags(disk_super) &
1531 ~BTRFS_SUPER_FLAG_SEEDING;
1532 btrfs_set_super_flags(disk_super, super_flags);
1533
1534 return 0;
1535 }
1536
1537 /*
1538 * strore the expected generation for seed devices in device items.
1539 */
1540 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1541 struct btrfs_root *root)
1542 {
1543 struct btrfs_path *path;
1544 struct extent_buffer *leaf;
1545 struct btrfs_dev_item *dev_item;
1546 struct btrfs_device *device;
1547 struct btrfs_key key;
1548 u8 fs_uuid[BTRFS_UUID_SIZE];
1549 u8 dev_uuid[BTRFS_UUID_SIZE];
1550 u64 devid;
1551 int ret;
1552
1553 path = btrfs_alloc_path();
1554 if (!path)
1555 return -ENOMEM;
1556
1557 root = root->fs_info->chunk_root;
1558 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1559 key.offset = 0;
1560 key.type = BTRFS_DEV_ITEM_KEY;
1561
1562 while (1) {
1563 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1564 if (ret < 0)
1565 goto error;
1566
1567 leaf = path->nodes[0];
1568 next_slot:
1569 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1570 ret = btrfs_next_leaf(root, path);
1571 if (ret > 0)
1572 break;
1573 if (ret < 0)
1574 goto error;
1575 leaf = path->nodes[0];
1576 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1577 btrfs_release_path(path);
1578 continue;
1579 }
1580
1581 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1582 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1583 key.type != BTRFS_DEV_ITEM_KEY)
1584 break;
1585
1586 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1587 struct btrfs_dev_item);
1588 devid = btrfs_device_id(leaf, dev_item);
1589 read_extent_buffer(leaf, dev_uuid,
1590 (unsigned long)btrfs_device_uuid(dev_item),
1591 BTRFS_UUID_SIZE);
1592 read_extent_buffer(leaf, fs_uuid,
1593 (unsigned long)btrfs_device_fsid(dev_item),
1594 BTRFS_UUID_SIZE);
1595 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1596 BUG_ON(!device);
1597
1598 if (device->fs_devices->seeding) {
1599 btrfs_set_device_generation(leaf, dev_item,
1600 device->generation);
1601 btrfs_mark_buffer_dirty(leaf);
1602 }
1603
1604 path->slots[0]++;
1605 goto next_slot;
1606 }
1607 ret = 0;
1608 error:
1609 btrfs_free_path(path);
1610 return ret;
1611 }
1612
1613 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1614 {
1615 struct request_queue *q;
1616 struct btrfs_trans_handle *trans;
1617 struct btrfs_device *device;
1618 struct block_device *bdev;
1619 struct list_head *devices;
1620 struct super_block *sb = root->fs_info->sb;
1621 u64 total_bytes;
1622 int seeding_dev = 0;
1623 int ret = 0;
1624
1625 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1626 return -EINVAL;
1627
1628 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1629 root->fs_info->bdev_holder);
1630 if (IS_ERR(bdev))
1631 return PTR_ERR(bdev);
1632
1633 if (root->fs_info->fs_devices->seeding) {
1634 seeding_dev = 1;
1635 down_write(&sb->s_umount);
1636 mutex_lock(&uuid_mutex);
1637 }
1638
1639 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1640
1641 devices = &root->fs_info->fs_devices->devices;
1642 /*
1643 * we have the volume lock, so we don't need the extra
1644 * device list mutex while reading the list here.
1645 */
1646 list_for_each_entry(device, devices, dev_list) {
1647 if (device->bdev == bdev) {
1648 ret = -EEXIST;
1649 goto error;
1650 }
1651 }
1652
1653 device = kzalloc(sizeof(*device), GFP_NOFS);
1654 if (!device) {
1655 /* we can safely leave the fs_devices entry around */
1656 ret = -ENOMEM;
1657 goto error;
1658 }
1659
1660 device->name = kstrdup(device_path, GFP_NOFS);
1661 if (!device->name) {
1662 kfree(device);
1663 ret = -ENOMEM;
1664 goto error;
1665 }
1666
1667 ret = find_next_devid(root, &device->devid);
1668 if (ret) {
1669 kfree(device->name);
1670 kfree(device);
1671 goto error;
1672 }
1673
1674 trans = btrfs_start_transaction(root, 0);
1675 if (IS_ERR(trans)) {
1676 kfree(device->name);
1677 kfree(device);
1678 ret = PTR_ERR(trans);
1679 goto error;
1680 }
1681
1682 lock_chunks(root);
1683
1684 q = bdev_get_queue(bdev);
1685 if (blk_queue_discard(q))
1686 device->can_discard = 1;
1687 device->writeable = 1;
1688 device->work.func = pending_bios_fn;
1689 generate_random_uuid(device->uuid);
1690 spin_lock_init(&device->io_lock);
1691 device->generation = trans->transid;
1692 device->io_width = root->sectorsize;
1693 device->io_align = root->sectorsize;
1694 device->sector_size = root->sectorsize;
1695 device->total_bytes = i_size_read(bdev->bd_inode);
1696 device->disk_total_bytes = device->total_bytes;
1697 device->dev_root = root->fs_info->dev_root;
1698 device->bdev = bdev;
1699 device->in_fs_metadata = 1;
1700 device->mode = FMODE_EXCL;
1701 set_blocksize(device->bdev, 4096);
1702
1703 if (seeding_dev) {
1704 sb->s_flags &= ~MS_RDONLY;
1705 ret = btrfs_prepare_sprout(root);
1706 BUG_ON(ret);
1707 }
1708
1709 device->fs_devices = root->fs_info->fs_devices;
1710
1711 /*
1712 * we don't want write_supers to jump in here with our device
1713 * half setup
1714 */
1715 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1716 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1717 list_add(&device->dev_alloc_list,
1718 &root->fs_info->fs_devices->alloc_list);
1719 root->fs_info->fs_devices->num_devices++;
1720 root->fs_info->fs_devices->open_devices++;
1721 root->fs_info->fs_devices->rw_devices++;
1722 if (device->can_discard)
1723 root->fs_info->fs_devices->num_can_discard++;
1724 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1725
1726 spin_lock(&root->fs_info->free_chunk_lock);
1727 root->fs_info->free_chunk_space += device->total_bytes;
1728 spin_unlock(&root->fs_info->free_chunk_lock);
1729
1730 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1731 root->fs_info->fs_devices->rotating = 1;
1732
1733 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1734 btrfs_set_super_total_bytes(root->fs_info->super_copy,
1735 total_bytes + device->total_bytes);
1736
1737 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1738 btrfs_set_super_num_devices(root->fs_info->super_copy,
1739 total_bytes + 1);
1740 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1741
1742 if (seeding_dev) {
1743 ret = init_first_rw_device(trans, root, device);
1744 BUG_ON(ret);
1745 ret = btrfs_finish_sprout(trans, root);
1746 BUG_ON(ret);
1747 } else {
1748 ret = btrfs_add_device(trans, root, device);
1749 }
1750
1751 /*
1752 * we've got more storage, clear any full flags on the space
1753 * infos
1754 */
1755 btrfs_clear_space_info_full(root->fs_info);
1756
1757 unlock_chunks(root);
1758 btrfs_commit_transaction(trans, root);
1759
1760 if (seeding_dev) {
1761 mutex_unlock(&uuid_mutex);
1762 up_write(&sb->s_umount);
1763
1764 ret = btrfs_relocate_sys_chunks(root);
1765 BUG_ON(ret);
1766 }
1767
1768 return ret;
1769 error:
1770 blkdev_put(bdev, FMODE_EXCL);
1771 if (seeding_dev) {
1772 mutex_unlock(&uuid_mutex);
1773 up_write(&sb->s_umount);
1774 }
1775 return ret;
1776 }
1777
1778 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1779 struct btrfs_device *device)
1780 {
1781 int ret;
1782 struct btrfs_path *path;
1783 struct btrfs_root *root;
1784 struct btrfs_dev_item *dev_item;
1785 struct extent_buffer *leaf;
1786 struct btrfs_key key;
1787
1788 root = device->dev_root->fs_info->chunk_root;
1789
1790 path = btrfs_alloc_path();
1791 if (!path)
1792 return -ENOMEM;
1793
1794 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1795 key.type = BTRFS_DEV_ITEM_KEY;
1796 key.offset = device->devid;
1797
1798 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1799 if (ret < 0)
1800 goto out;
1801
1802 if (ret > 0) {
1803 ret = -ENOENT;
1804 goto out;
1805 }
1806
1807 leaf = path->nodes[0];
1808 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1809
1810 btrfs_set_device_id(leaf, dev_item, device->devid);
1811 btrfs_set_device_type(leaf, dev_item, device->type);
1812 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1813 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1814 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1815 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1816 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1817 btrfs_mark_buffer_dirty(leaf);
1818
1819 out:
1820 btrfs_free_path(path);
1821 return ret;
1822 }
1823
1824 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1825 struct btrfs_device *device, u64 new_size)
1826 {
1827 struct btrfs_super_block *super_copy =
1828 device->dev_root->fs_info->super_copy;
1829 u64 old_total = btrfs_super_total_bytes(super_copy);
1830 u64 diff = new_size - device->total_bytes;
1831
1832 if (!device->writeable)
1833 return -EACCES;
1834 if (new_size <= device->total_bytes)
1835 return -EINVAL;
1836
1837 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1838 device->fs_devices->total_rw_bytes += diff;
1839
1840 device->total_bytes = new_size;
1841 device->disk_total_bytes = new_size;
1842 btrfs_clear_space_info_full(device->dev_root->fs_info);
1843
1844 return btrfs_update_device(trans, device);
1845 }
1846
1847 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1848 struct btrfs_device *device, u64 new_size)
1849 {
1850 int ret;
1851 lock_chunks(device->dev_root);
1852 ret = __btrfs_grow_device(trans, device, new_size);
1853 unlock_chunks(device->dev_root);
1854 return ret;
1855 }
1856
1857 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1858 struct btrfs_root *root,
1859 u64 chunk_tree, u64 chunk_objectid,
1860 u64 chunk_offset)
1861 {
1862 int ret;
1863 struct btrfs_path *path;
1864 struct btrfs_key key;
1865
1866 root = root->fs_info->chunk_root;
1867 path = btrfs_alloc_path();
1868 if (!path)
1869 return -ENOMEM;
1870
1871 key.objectid = chunk_objectid;
1872 key.offset = chunk_offset;
1873 key.type = BTRFS_CHUNK_ITEM_KEY;
1874
1875 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1876 BUG_ON(ret);
1877
1878 ret = btrfs_del_item(trans, root, path);
1879
1880 btrfs_free_path(path);
1881 return ret;
1882 }
1883
1884 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1885 chunk_offset)
1886 {
1887 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1888 struct btrfs_disk_key *disk_key;
1889 struct btrfs_chunk *chunk;
1890 u8 *ptr;
1891 int ret = 0;
1892 u32 num_stripes;
1893 u32 array_size;
1894 u32 len = 0;
1895 u32 cur;
1896 struct btrfs_key key;
1897
1898 array_size = btrfs_super_sys_array_size(super_copy);
1899
1900 ptr = super_copy->sys_chunk_array;
1901 cur = 0;
1902
1903 while (cur < array_size) {
1904 disk_key = (struct btrfs_disk_key *)ptr;
1905 btrfs_disk_key_to_cpu(&key, disk_key);
1906
1907 len = sizeof(*disk_key);
1908
1909 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1910 chunk = (struct btrfs_chunk *)(ptr + len);
1911 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1912 len += btrfs_chunk_item_size(num_stripes);
1913 } else {
1914 ret = -EIO;
1915 break;
1916 }
1917 if (key.objectid == chunk_objectid &&
1918 key.offset == chunk_offset) {
1919 memmove(ptr, ptr + len, array_size - (cur + len));
1920 array_size -= len;
1921 btrfs_set_super_sys_array_size(super_copy, array_size);
1922 } else {
1923 ptr += len;
1924 cur += len;
1925 }
1926 }
1927 return ret;
1928 }
1929
1930 static int btrfs_relocate_chunk(struct btrfs_root *root,
1931 u64 chunk_tree, u64 chunk_objectid,
1932 u64 chunk_offset)
1933 {
1934 struct extent_map_tree *em_tree;
1935 struct btrfs_root *extent_root;
1936 struct btrfs_trans_handle *trans;
1937 struct extent_map *em;
1938 struct map_lookup *map;
1939 int ret;
1940 int i;
1941
1942 root = root->fs_info->chunk_root;
1943 extent_root = root->fs_info->extent_root;
1944 em_tree = &root->fs_info->mapping_tree.map_tree;
1945
1946 ret = btrfs_can_relocate(extent_root, chunk_offset);
1947 if (ret)
1948 return -ENOSPC;
1949
1950 /* step one, relocate all the extents inside this chunk */
1951 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1952 if (ret)
1953 return ret;
1954
1955 trans = btrfs_start_transaction(root, 0);
1956 BUG_ON(IS_ERR(trans));
1957
1958 lock_chunks(root);
1959
1960 /*
1961 * step two, delete the device extents and the
1962 * chunk tree entries
1963 */
1964 read_lock(&em_tree->lock);
1965 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1966 read_unlock(&em_tree->lock);
1967
1968 BUG_ON(!em || em->start > chunk_offset ||
1969 em->start + em->len < chunk_offset);
1970 map = (struct map_lookup *)em->bdev;
1971
1972 for (i = 0; i < map->num_stripes; i++) {
1973 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1974 map->stripes[i].physical);
1975 BUG_ON(ret);
1976
1977 if (map->stripes[i].dev) {
1978 ret = btrfs_update_device(trans, map->stripes[i].dev);
1979 BUG_ON(ret);
1980 }
1981 }
1982 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1983 chunk_offset);
1984
1985 BUG_ON(ret);
1986
1987 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1988
1989 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1990 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1991 BUG_ON(ret);
1992 }
1993
1994 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1995 BUG_ON(ret);
1996
1997 write_lock(&em_tree->lock);
1998 remove_extent_mapping(em_tree, em);
1999 write_unlock(&em_tree->lock);
2000
2001 kfree(map);
2002 em->bdev = NULL;
2003
2004 /* once for the tree */
2005 free_extent_map(em);
2006 /* once for us */
2007 free_extent_map(em);
2008
2009 unlock_chunks(root);
2010 btrfs_end_transaction(trans, root);
2011 return 0;
2012 }
2013
2014 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2015 {
2016 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2017 struct btrfs_path *path;
2018 struct extent_buffer *leaf;
2019 struct btrfs_chunk *chunk;
2020 struct btrfs_key key;
2021 struct btrfs_key found_key;
2022 u64 chunk_tree = chunk_root->root_key.objectid;
2023 u64 chunk_type;
2024 bool retried = false;
2025 int failed = 0;
2026 int ret;
2027
2028 path = btrfs_alloc_path();
2029 if (!path)
2030 return -ENOMEM;
2031
2032 again:
2033 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2034 key.offset = (u64)-1;
2035 key.type = BTRFS_CHUNK_ITEM_KEY;
2036
2037 while (1) {
2038 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2039 if (ret < 0)
2040 goto error;
2041 BUG_ON(ret == 0);
2042
2043 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2044 key.type);
2045 if (ret < 0)
2046 goto error;
2047 if (ret > 0)
2048 break;
2049
2050 leaf = path->nodes[0];
2051 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2052
2053 chunk = btrfs_item_ptr(leaf, path->slots[0],
2054 struct btrfs_chunk);
2055 chunk_type = btrfs_chunk_type(leaf, chunk);
2056 btrfs_release_path(path);
2057
2058 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2059 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2060 found_key.objectid,
2061 found_key.offset);
2062 if (ret == -ENOSPC)
2063 failed++;
2064 else if (ret)
2065 BUG();
2066 }
2067
2068 if (found_key.offset == 0)
2069 break;
2070 key.offset = found_key.offset - 1;
2071 }
2072 ret = 0;
2073 if (failed && !retried) {
2074 failed = 0;
2075 retried = true;
2076 goto again;
2077 } else if (failed && retried) {
2078 WARN_ON(1);
2079 ret = -ENOSPC;
2080 }
2081 error:
2082 btrfs_free_path(path);
2083 return ret;
2084 }
2085
2086 static int insert_balance_item(struct btrfs_root *root,
2087 struct btrfs_balance_control *bctl)
2088 {
2089 struct btrfs_trans_handle *trans;
2090 struct btrfs_balance_item *item;
2091 struct btrfs_disk_balance_args disk_bargs;
2092 struct btrfs_path *path;
2093 struct extent_buffer *leaf;
2094 struct btrfs_key key;
2095 int ret, err;
2096
2097 path = btrfs_alloc_path();
2098 if (!path)
2099 return -ENOMEM;
2100
2101 trans = btrfs_start_transaction(root, 0);
2102 if (IS_ERR(trans)) {
2103 btrfs_free_path(path);
2104 return PTR_ERR(trans);
2105 }
2106
2107 key.objectid = BTRFS_BALANCE_OBJECTID;
2108 key.type = BTRFS_BALANCE_ITEM_KEY;
2109 key.offset = 0;
2110
2111 ret = btrfs_insert_empty_item(trans, root, path, &key,
2112 sizeof(*item));
2113 if (ret)
2114 goto out;
2115
2116 leaf = path->nodes[0];
2117 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2118
2119 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2120
2121 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2122 btrfs_set_balance_data(leaf, item, &disk_bargs);
2123 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2124 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2125 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2126 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2127
2128 btrfs_set_balance_flags(leaf, item, bctl->flags);
2129
2130 btrfs_mark_buffer_dirty(leaf);
2131 out:
2132 btrfs_free_path(path);
2133 err = btrfs_commit_transaction(trans, root);
2134 if (err && !ret)
2135 ret = err;
2136 return ret;
2137 }
2138
2139 static int del_balance_item(struct btrfs_root *root)
2140 {
2141 struct btrfs_trans_handle *trans;
2142 struct btrfs_path *path;
2143 struct btrfs_key key;
2144 int ret, err;
2145
2146 path = btrfs_alloc_path();
2147 if (!path)
2148 return -ENOMEM;
2149
2150 trans = btrfs_start_transaction(root, 0);
2151 if (IS_ERR(trans)) {
2152 btrfs_free_path(path);
2153 return PTR_ERR(trans);
2154 }
2155
2156 key.objectid = BTRFS_BALANCE_OBJECTID;
2157 key.type = BTRFS_BALANCE_ITEM_KEY;
2158 key.offset = 0;
2159
2160 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2161 if (ret < 0)
2162 goto out;
2163 if (ret > 0) {
2164 ret = -ENOENT;
2165 goto out;
2166 }
2167
2168 ret = btrfs_del_item(trans, root, path);
2169 out:
2170 btrfs_free_path(path);
2171 err = btrfs_commit_transaction(trans, root);
2172 if (err && !ret)
2173 ret = err;
2174 return ret;
2175 }
2176
2177 /*
2178 * This is a heuristic used to reduce the number of chunks balanced on
2179 * resume after balance was interrupted.
2180 */
2181 static void update_balance_args(struct btrfs_balance_control *bctl)
2182 {
2183 /*
2184 * Turn on soft mode for chunk types that were being converted.
2185 */
2186 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2187 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2188 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2189 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2190 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2191 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2192
2193 /*
2194 * Turn on usage filter if is not already used. The idea is
2195 * that chunks that we have already balanced should be
2196 * reasonably full. Don't do it for chunks that are being
2197 * converted - that will keep us from relocating unconverted
2198 * (albeit full) chunks.
2199 */
2200 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2201 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2202 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2203 bctl->data.usage = 90;
2204 }
2205 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2206 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2207 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2208 bctl->sys.usage = 90;
2209 }
2210 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2211 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2212 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2213 bctl->meta.usage = 90;
2214 }
2215 }
2216
2217 /*
2218 * Should be called with both balance and volume mutexes held to
2219 * serialize other volume operations (add_dev/rm_dev/resize) with
2220 * restriper. Same goes for unset_balance_control.
2221 */
2222 static void set_balance_control(struct btrfs_balance_control *bctl)
2223 {
2224 struct btrfs_fs_info *fs_info = bctl->fs_info;
2225
2226 BUG_ON(fs_info->balance_ctl);
2227
2228 spin_lock(&fs_info->balance_lock);
2229 fs_info->balance_ctl = bctl;
2230 spin_unlock(&fs_info->balance_lock);
2231 }
2232
2233 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2234 {
2235 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2236
2237 BUG_ON(!fs_info->balance_ctl);
2238
2239 spin_lock(&fs_info->balance_lock);
2240 fs_info->balance_ctl = NULL;
2241 spin_unlock(&fs_info->balance_lock);
2242
2243 kfree(bctl);
2244 }
2245
2246 /*
2247 * Balance filters. Return 1 if chunk should be filtered out
2248 * (should not be balanced).
2249 */
2250 static int chunk_profiles_filter(u64 chunk_profile,
2251 struct btrfs_balance_args *bargs)
2252 {
2253 chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
2254
2255 if (chunk_profile == 0)
2256 chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2257
2258 if (bargs->profiles & chunk_profile)
2259 return 0;
2260
2261 return 1;
2262 }
2263
2264 static u64 div_factor_fine(u64 num, int factor)
2265 {
2266 if (factor <= 0)
2267 return 0;
2268 if (factor >= 100)
2269 return num;
2270
2271 num *= factor;
2272 do_div(num, 100);
2273 return num;
2274 }
2275
2276 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2277 struct btrfs_balance_args *bargs)
2278 {
2279 struct btrfs_block_group_cache *cache;
2280 u64 chunk_used, user_thresh;
2281 int ret = 1;
2282
2283 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2284 chunk_used = btrfs_block_group_used(&cache->item);
2285
2286 user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2287 if (chunk_used < user_thresh)
2288 ret = 0;
2289
2290 btrfs_put_block_group(cache);
2291 return ret;
2292 }
2293
2294 static int chunk_devid_filter(struct extent_buffer *leaf,
2295 struct btrfs_chunk *chunk,
2296 struct btrfs_balance_args *bargs)
2297 {
2298 struct btrfs_stripe *stripe;
2299 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2300 int i;
2301
2302 for (i = 0; i < num_stripes; i++) {
2303 stripe = btrfs_stripe_nr(chunk, i);
2304 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2305 return 0;
2306 }
2307
2308 return 1;
2309 }
2310
2311 /* [pstart, pend) */
2312 static int chunk_drange_filter(struct extent_buffer *leaf,
2313 struct btrfs_chunk *chunk,
2314 u64 chunk_offset,
2315 struct btrfs_balance_args *bargs)
2316 {
2317 struct btrfs_stripe *stripe;
2318 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2319 u64 stripe_offset;
2320 u64 stripe_length;
2321 int factor;
2322 int i;
2323
2324 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2325 return 0;
2326
2327 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2328 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2329 factor = 2;
2330 else
2331 factor = 1;
2332 factor = num_stripes / factor;
2333
2334 for (i = 0; i < num_stripes; i++) {
2335 stripe = btrfs_stripe_nr(chunk, i);
2336 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2337 continue;
2338
2339 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2340 stripe_length = btrfs_chunk_length(leaf, chunk);
2341 do_div(stripe_length, factor);
2342
2343 if (stripe_offset < bargs->pend &&
2344 stripe_offset + stripe_length > bargs->pstart)
2345 return 0;
2346 }
2347
2348 return 1;
2349 }
2350
2351 /* [vstart, vend) */
2352 static int chunk_vrange_filter(struct extent_buffer *leaf,
2353 struct btrfs_chunk *chunk,
2354 u64 chunk_offset,
2355 struct btrfs_balance_args *bargs)
2356 {
2357 if (chunk_offset < bargs->vend &&
2358 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2359 /* at least part of the chunk is inside this vrange */
2360 return 0;
2361
2362 return 1;
2363 }
2364
2365 static int chunk_soft_convert_filter(u64 chunk_profile,
2366 struct btrfs_balance_args *bargs)
2367 {
2368 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2369 return 0;
2370
2371 chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
2372
2373 if (chunk_profile == 0)
2374 chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2375
2376 if (bargs->target & chunk_profile)
2377 return 1;
2378
2379 return 0;
2380 }
2381
2382 static int should_balance_chunk(struct btrfs_root *root,
2383 struct extent_buffer *leaf,
2384 struct btrfs_chunk *chunk, u64 chunk_offset)
2385 {
2386 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2387 struct btrfs_balance_args *bargs = NULL;
2388 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2389
2390 /* type filter */
2391 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2392 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2393 return 0;
2394 }
2395
2396 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2397 bargs = &bctl->data;
2398 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2399 bargs = &bctl->sys;
2400 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2401 bargs = &bctl->meta;
2402
2403 /* profiles filter */
2404 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2405 chunk_profiles_filter(chunk_type, bargs)) {
2406 return 0;
2407 }
2408
2409 /* usage filter */
2410 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2411 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2412 return 0;
2413 }
2414
2415 /* devid filter */
2416 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2417 chunk_devid_filter(leaf, chunk, bargs)) {
2418 return 0;
2419 }
2420
2421 /* drange filter, makes sense only with devid filter */
2422 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2423 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2424 return 0;
2425 }
2426
2427 /* vrange filter */
2428 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2429 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2430 return 0;
2431 }
2432
2433 /* soft profile changing mode */
2434 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2435 chunk_soft_convert_filter(chunk_type, bargs)) {
2436 return 0;
2437 }
2438
2439 return 1;
2440 }
2441
2442 static u64 div_factor(u64 num, int factor)
2443 {
2444 if (factor == 10)
2445 return num;
2446 num *= factor;
2447 do_div(num, 10);
2448 return num;
2449 }
2450
2451 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2452 {
2453 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2454 struct btrfs_root *chunk_root = fs_info->chunk_root;
2455 struct btrfs_root *dev_root = fs_info->dev_root;
2456 struct list_head *devices;
2457 struct btrfs_device *device;
2458 u64 old_size;
2459 u64 size_to_free;
2460 struct btrfs_chunk *chunk;
2461 struct btrfs_path *path;
2462 struct btrfs_key key;
2463 struct btrfs_key found_key;
2464 struct btrfs_trans_handle *trans;
2465 struct extent_buffer *leaf;
2466 int slot;
2467 int ret;
2468 int enospc_errors = 0;
2469 bool counting = true;
2470
2471 /* step one make some room on all the devices */
2472 devices = &fs_info->fs_devices->devices;
2473 list_for_each_entry(device, devices, dev_list) {
2474 old_size = device->total_bytes;
2475 size_to_free = div_factor(old_size, 1);
2476 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2477 if (!device->writeable ||
2478 device->total_bytes - device->bytes_used > size_to_free)
2479 continue;
2480
2481 ret = btrfs_shrink_device(device, old_size - size_to_free);
2482 if (ret == -ENOSPC)
2483 break;
2484 BUG_ON(ret);
2485
2486 trans = btrfs_start_transaction(dev_root, 0);
2487 BUG_ON(IS_ERR(trans));
2488
2489 ret = btrfs_grow_device(trans, device, old_size);
2490 BUG_ON(ret);
2491
2492 btrfs_end_transaction(trans, dev_root);
2493 }
2494
2495 /* step two, relocate all the chunks */
2496 path = btrfs_alloc_path();
2497 if (!path) {
2498 ret = -ENOMEM;
2499 goto error;
2500 }
2501
2502 /* zero out stat counters */
2503 spin_lock(&fs_info->balance_lock);
2504 memset(&bctl->stat, 0, sizeof(bctl->stat));
2505 spin_unlock(&fs_info->balance_lock);
2506 again:
2507 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2508 key.offset = (u64)-1;
2509 key.type = BTRFS_CHUNK_ITEM_KEY;
2510
2511 while (1) {
2512 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2513 atomic_read(&fs_info->balance_cancel_req)) {
2514 ret = -ECANCELED;
2515 goto error;
2516 }
2517
2518 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2519 if (ret < 0)
2520 goto error;
2521
2522 /*
2523 * this shouldn't happen, it means the last relocate
2524 * failed
2525 */
2526 if (ret == 0)
2527 BUG(); /* FIXME break ? */
2528
2529 ret = btrfs_previous_item(chunk_root, path, 0,
2530 BTRFS_CHUNK_ITEM_KEY);
2531 if (ret) {
2532 ret = 0;
2533 break;
2534 }
2535
2536 leaf = path->nodes[0];
2537 slot = path->slots[0];
2538 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2539
2540 if (found_key.objectid != key.objectid)
2541 break;
2542
2543 /* chunk zero is special */
2544 if (found_key.offset == 0)
2545 break;
2546
2547 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2548
2549 if (!counting) {
2550 spin_lock(&fs_info->balance_lock);
2551 bctl->stat.considered++;
2552 spin_unlock(&fs_info->balance_lock);
2553 }
2554
2555 ret = should_balance_chunk(chunk_root, leaf, chunk,
2556 found_key.offset);
2557 btrfs_release_path(path);
2558 if (!ret)
2559 goto loop;
2560
2561 if (counting) {
2562 spin_lock(&fs_info->balance_lock);
2563 bctl->stat.expected++;
2564 spin_unlock(&fs_info->balance_lock);
2565 goto loop;
2566 }
2567
2568 ret = btrfs_relocate_chunk(chunk_root,
2569 chunk_root->root_key.objectid,
2570 found_key.objectid,
2571 found_key.offset);
2572 if (ret && ret != -ENOSPC)
2573 goto error;
2574 if (ret == -ENOSPC) {
2575 enospc_errors++;
2576 } else {
2577 spin_lock(&fs_info->balance_lock);
2578 bctl->stat.completed++;
2579 spin_unlock(&fs_info->balance_lock);
2580 }
2581 loop:
2582 key.offset = found_key.offset - 1;
2583 }
2584
2585 if (counting) {
2586 btrfs_release_path(path);
2587 counting = false;
2588 goto again;
2589 }
2590 error:
2591 btrfs_free_path(path);
2592 if (enospc_errors) {
2593 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2594 enospc_errors);
2595 if (!ret)
2596 ret = -ENOSPC;
2597 }
2598
2599 return ret;
2600 }
2601
2602 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2603 {
2604 /* cancel requested || normal exit path */
2605 return atomic_read(&fs_info->balance_cancel_req) ||
2606 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2607 atomic_read(&fs_info->balance_cancel_req) == 0);
2608 }
2609
2610 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2611 {
2612 int ret;
2613
2614 unset_balance_control(fs_info);
2615 ret = del_balance_item(fs_info->tree_root);
2616 BUG_ON(ret);
2617 }
2618
2619 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2620 struct btrfs_ioctl_balance_args *bargs);
2621
2622 /*
2623 * Should be called with both balance and volume mutexes held
2624 */
2625 int btrfs_balance(struct btrfs_balance_control *bctl,
2626 struct btrfs_ioctl_balance_args *bargs)
2627 {
2628 struct btrfs_fs_info *fs_info = bctl->fs_info;
2629 u64 allowed;
2630 int ret;
2631
2632 if (btrfs_fs_closing(fs_info) ||
2633 atomic_read(&fs_info->balance_pause_req) ||
2634 atomic_read(&fs_info->balance_cancel_req)) {
2635 ret = -EINVAL;
2636 goto out;
2637 }
2638
2639 /*
2640 * In case of mixed groups both data and meta should be picked,
2641 * and identical options should be given for both of them.
2642 */
2643 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2644 if ((allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2645 (bctl->flags & (BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA))) {
2646 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2647 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2648 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2649 printk(KERN_ERR "btrfs: with mixed groups data and "
2650 "metadata balance options must be the same\n");
2651 ret = -EINVAL;
2652 goto out;
2653 }
2654 }
2655
2656 /*
2657 * Profile changing sanity checks. Skip them if a simple
2658 * balance is requested.
2659 */
2660 if (!((bctl->data.flags | bctl->sys.flags | bctl->meta.flags) &
2661 BTRFS_BALANCE_ARGS_CONVERT))
2662 goto do_balance;
2663
2664 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2665 if (fs_info->fs_devices->num_devices == 1)
2666 allowed |= BTRFS_BLOCK_GROUP_DUP;
2667 else if (fs_info->fs_devices->num_devices < 4)
2668 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2669 else
2670 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2671 BTRFS_BLOCK_GROUP_RAID10);
2672
2673 if (!profile_is_valid(bctl->data.target, 1) ||
2674 bctl->data.target & ~allowed) {
2675 printk(KERN_ERR "btrfs: unable to start balance with target "
2676 "data profile %llu\n",
2677 (unsigned long long)bctl->data.target);
2678 ret = -EINVAL;
2679 goto out;
2680 }
2681 if (!profile_is_valid(bctl->meta.target, 1) ||
2682 bctl->meta.target & ~allowed) {
2683 printk(KERN_ERR "btrfs: unable to start balance with target "
2684 "metadata profile %llu\n",
2685 (unsigned long long)bctl->meta.target);
2686 ret = -EINVAL;
2687 goto out;
2688 }
2689 if (!profile_is_valid(bctl->sys.target, 1) ||
2690 bctl->sys.target & ~allowed) {
2691 printk(KERN_ERR "btrfs: unable to start balance with target "
2692 "system profile %llu\n",
2693 (unsigned long long)bctl->sys.target);
2694 ret = -EINVAL;
2695 goto out;
2696 }
2697
2698 if (bctl->data.target & BTRFS_BLOCK_GROUP_DUP) {
2699 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2700 ret = -EINVAL;
2701 goto out;
2702 }
2703
2704 /* allow to reduce meta or sys integrity only if force set */
2705 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2706 BTRFS_BLOCK_GROUP_RAID10;
2707 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2708 (fs_info->avail_system_alloc_bits & allowed) &&
2709 !(bctl->sys.target & allowed)) ||
2710 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2711 (fs_info->avail_metadata_alloc_bits & allowed) &&
2712 !(bctl->meta.target & allowed))) {
2713 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2714 printk(KERN_INFO "btrfs: force reducing metadata "
2715 "integrity\n");
2716 } else {
2717 printk(KERN_ERR "btrfs: balance will reduce metadata "
2718 "integrity, use force if you want this\n");
2719 ret = -EINVAL;
2720 goto out;
2721 }
2722 }
2723
2724 do_balance:
2725 ret = insert_balance_item(fs_info->tree_root, bctl);
2726 if (ret && ret != -EEXIST)
2727 goto out;
2728
2729 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2730 BUG_ON(ret == -EEXIST);
2731 set_balance_control(bctl);
2732 } else {
2733 BUG_ON(ret != -EEXIST);
2734 spin_lock(&fs_info->balance_lock);
2735 update_balance_args(bctl);
2736 spin_unlock(&fs_info->balance_lock);
2737 }
2738
2739 atomic_inc(&fs_info->balance_running);
2740 mutex_unlock(&fs_info->balance_mutex);
2741
2742 ret = __btrfs_balance(fs_info);
2743
2744 mutex_lock(&fs_info->balance_mutex);
2745 atomic_dec(&fs_info->balance_running);
2746
2747 if (bargs) {
2748 memset(bargs, 0, sizeof(*bargs));
2749 update_ioctl_balance_args(fs_info, 0, bargs);
2750 }
2751
2752 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2753 balance_need_close(fs_info)) {
2754 __cancel_balance(fs_info);
2755 }
2756
2757 wake_up(&fs_info->balance_wait_q);
2758
2759 return ret;
2760 out:
2761 if (bctl->flags & BTRFS_BALANCE_RESUME)
2762 __cancel_balance(fs_info);
2763 else
2764 kfree(bctl);
2765 return ret;
2766 }
2767
2768 static int balance_kthread(void *data)
2769 {
2770 struct btrfs_balance_control *bctl =
2771 (struct btrfs_balance_control *)data;
2772 struct btrfs_fs_info *fs_info = bctl->fs_info;
2773 int ret = 0;
2774
2775 mutex_lock(&fs_info->volume_mutex);
2776 mutex_lock(&fs_info->balance_mutex);
2777
2778 set_balance_control(bctl);
2779
2780 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2781 printk(KERN_INFO "btrfs: force skipping balance\n");
2782 } else {
2783 printk(KERN_INFO "btrfs: continuing balance\n");
2784 ret = btrfs_balance(bctl, NULL);
2785 }
2786
2787 mutex_unlock(&fs_info->balance_mutex);
2788 mutex_unlock(&fs_info->volume_mutex);
2789 return ret;
2790 }
2791
2792 int btrfs_recover_balance(struct btrfs_root *tree_root)
2793 {
2794 struct task_struct *tsk;
2795 struct btrfs_balance_control *bctl;
2796 struct btrfs_balance_item *item;
2797 struct btrfs_disk_balance_args disk_bargs;
2798 struct btrfs_path *path;
2799 struct extent_buffer *leaf;
2800 struct btrfs_key key;
2801 int ret;
2802
2803 path = btrfs_alloc_path();
2804 if (!path)
2805 return -ENOMEM;
2806
2807 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2808 if (!bctl) {
2809 ret = -ENOMEM;
2810 goto out;
2811 }
2812
2813 key.objectid = BTRFS_BALANCE_OBJECTID;
2814 key.type = BTRFS_BALANCE_ITEM_KEY;
2815 key.offset = 0;
2816
2817 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2818 if (ret < 0)
2819 goto out_bctl;
2820 if (ret > 0) { /* ret = -ENOENT; */
2821 ret = 0;
2822 goto out_bctl;
2823 }
2824
2825 leaf = path->nodes[0];
2826 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2827
2828 bctl->fs_info = tree_root->fs_info;
2829 bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME;
2830
2831 btrfs_balance_data(leaf, item, &disk_bargs);
2832 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2833 btrfs_balance_meta(leaf, item, &disk_bargs);
2834 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2835 btrfs_balance_sys(leaf, item, &disk_bargs);
2836 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2837
2838 tsk = kthread_run(balance_kthread, bctl, "btrfs-balance");
2839 if (IS_ERR(tsk))
2840 ret = PTR_ERR(tsk);
2841 else
2842 goto out;
2843
2844 out_bctl:
2845 kfree(bctl);
2846 out:
2847 btrfs_free_path(path);
2848 return ret;
2849 }
2850
2851 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2852 {
2853 int ret = 0;
2854
2855 mutex_lock(&fs_info->balance_mutex);
2856 if (!fs_info->balance_ctl) {
2857 mutex_unlock(&fs_info->balance_mutex);
2858 return -ENOTCONN;
2859 }
2860
2861 if (atomic_read(&fs_info->balance_running)) {
2862 atomic_inc(&fs_info->balance_pause_req);
2863 mutex_unlock(&fs_info->balance_mutex);
2864
2865 wait_event(fs_info->balance_wait_q,
2866 atomic_read(&fs_info->balance_running) == 0);
2867
2868 mutex_lock(&fs_info->balance_mutex);
2869 /* we are good with balance_ctl ripped off from under us */
2870 BUG_ON(atomic_read(&fs_info->balance_running));
2871 atomic_dec(&fs_info->balance_pause_req);
2872 } else {
2873 ret = -ENOTCONN;
2874 }
2875
2876 mutex_unlock(&fs_info->balance_mutex);
2877 return ret;
2878 }
2879
2880 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2881 {
2882 mutex_lock(&fs_info->balance_mutex);
2883 if (!fs_info->balance_ctl) {
2884 mutex_unlock(&fs_info->balance_mutex);
2885 return -ENOTCONN;
2886 }
2887
2888 atomic_inc(&fs_info->balance_cancel_req);
2889 /*
2890 * if we are running just wait and return, balance item is
2891 * deleted in btrfs_balance in this case
2892 */
2893 if (atomic_read(&fs_info->balance_running)) {
2894 mutex_unlock(&fs_info->balance_mutex);
2895 wait_event(fs_info->balance_wait_q,
2896 atomic_read(&fs_info->balance_running) == 0);
2897 mutex_lock(&fs_info->balance_mutex);
2898 } else {
2899 /* __cancel_balance needs volume_mutex */
2900 mutex_unlock(&fs_info->balance_mutex);
2901 mutex_lock(&fs_info->volume_mutex);
2902 mutex_lock(&fs_info->balance_mutex);
2903
2904 if (fs_info->balance_ctl)
2905 __cancel_balance(fs_info);
2906
2907 mutex_unlock(&fs_info->volume_mutex);
2908 }
2909
2910 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
2911 atomic_dec(&fs_info->balance_cancel_req);
2912 mutex_unlock(&fs_info->balance_mutex);
2913 return 0;
2914 }
2915
2916 /*
2917 * shrinking a device means finding all of the device extents past
2918 * the new size, and then following the back refs to the chunks.
2919 * The chunk relocation code actually frees the device extent
2920 */
2921 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2922 {
2923 struct btrfs_trans_handle *trans;
2924 struct btrfs_root *root = device->dev_root;
2925 struct btrfs_dev_extent *dev_extent = NULL;
2926 struct btrfs_path *path;
2927 u64 length;
2928 u64 chunk_tree;
2929 u64 chunk_objectid;
2930 u64 chunk_offset;
2931 int ret;
2932 int slot;
2933 int failed = 0;
2934 bool retried = false;
2935 struct extent_buffer *l;
2936 struct btrfs_key key;
2937 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2938 u64 old_total = btrfs_super_total_bytes(super_copy);
2939 u64 old_size = device->total_bytes;
2940 u64 diff = device->total_bytes - new_size;
2941
2942 if (new_size >= device->total_bytes)
2943 return -EINVAL;
2944
2945 path = btrfs_alloc_path();
2946 if (!path)
2947 return -ENOMEM;
2948
2949 path->reada = 2;
2950
2951 lock_chunks(root);
2952
2953 device->total_bytes = new_size;
2954 if (device->writeable) {
2955 device->fs_devices->total_rw_bytes -= diff;
2956 spin_lock(&root->fs_info->free_chunk_lock);
2957 root->fs_info->free_chunk_space -= diff;
2958 spin_unlock(&root->fs_info->free_chunk_lock);
2959 }
2960 unlock_chunks(root);
2961
2962 again:
2963 key.objectid = device->devid;
2964 key.offset = (u64)-1;
2965 key.type = BTRFS_DEV_EXTENT_KEY;
2966
2967 while (1) {
2968 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2969 if (ret < 0)
2970 goto done;
2971
2972 ret = btrfs_previous_item(root, path, 0, key.type);
2973 if (ret < 0)
2974 goto done;
2975 if (ret) {
2976 ret = 0;
2977 btrfs_release_path(path);
2978 break;
2979 }
2980
2981 l = path->nodes[0];
2982 slot = path->slots[0];
2983 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2984
2985 if (key.objectid != device->devid) {
2986 btrfs_release_path(path);
2987 break;
2988 }
2989
2990 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2991 length = btrfs_dev_extent_length(l, dev_extent);
2992
2993 if (key.offset + length <= new_size) {
2994 btrfs_release_path(path);
2995 break;
2996 }
2997
2998 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2999 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3000 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3001 btrfs_release_path(path);
3002
3003 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3004 chunk_offset);
3005 if (ret && ret != -ENOSPC)
3006 goto done;
3007 if (ret == -ENOSPC)
3008 failed++;
3009 key.offset -= 1;
3010 }
3011
3012 if (failed && !retried) {
3013 failed = 0;
3014 retried = true;
3015 goto again;
3016 } else if (failed && retried) {
3017 ret = -ENOSPC;
3018 lock_chunks(root);
3019
3020 device->total_bytes = old_size;
3021 if (device->writeable)
3022 device->fs_devices->total_rw_bytes += diff;
3023 spin_lock(&root->fs_info->free_chunk_lock);
3024 root->fs_info->free_chunk_space += diff;
3025 spin_unlock(&root->fs_info->free_chunk_lock);
3026 unlock_chunks(root);
3027 goto done;
3028 }
3029
3030 /* Shrinking succeeded, else we would be at "done". */
3031 trans = btrfs_start_transaction(root, 0);
3032 if (IS_ERR(trans)) {
3033 ret = PTR_ERR(trans);
3034 goto done;
3035 }
3036
3037 lock_chunks(root);
3038
3039 device->disk_total_bytes = new_size;
3040 /* Now btrfs_update_device() will change the on-disk size. */
3041 ret = btrfs_update_device(trans, device);
3042 if (ret) {
3043 unlock_chunks(root);
3044 btrfs_end_transaction(trans, root);
3045 goto done;
3046 }
3047 WARN_ON(diff > old_total);
3048 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3049 unlock_chunks(root);
3050 btrfs_end_transaction(trans, root);
3051 done:
3052 btrfs_free_path(path);
3053 return ret;
3054 }
3055
3056 static int btrfs_add_system_chunk(struct btrfs_root *root,
3057 struct btrfs_key *key,
3058 struct btrfs_chunk *chunk, int item_size)
3059 {
3060 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3061 struct btrfs_disk_key disk_key;
3062 u32 array_size;
3063 u8 *ptr;
3064
3065 array_size = btrfs_super_sys_array_size(super_copy);
3066 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3067 return -EFBIG;
3068
3069 ptr = super_copy->sys_chunk_array + array_size;
3070 btrfs_cpu_key_to_disk(&disk_key, key);
3071 memcpy(ptr, &disk_key, sizeof(disk_key));
3072 ptr += sizeof(disk_key);
3073 memcpy(ptr, chunk, item_size);
3074 item_size += sizeof(disk_key);
3075 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3076 return 0;
3077 }
3078
3079 /*
3080 * sort the devices in descending order by max_avail, total_avail
3081 */
3082 static int btrfs_cmp_device_info(const void *a, const void *b)
3083 {
3084 const struct btrfs_device_info *di_a = a;
3085 const struct btrfs_device_info *di_b = b;
3086
3087 if (di_a->max_avail > di_b->max_avail)
3088 return -1;
3089 if (di_a->max_avail < di_b->max_avail)
3090 return 1;
3091 if (di_a->total_avail > di_b->total_avail)
3092 return -1;
3093 if (di_a->total_avail < di_b->total_avail)
3094 return 1;
3095 return 0;
3096 }
3097
3098 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3099 struct btrfs_root *extent_root,
3100 struct map_lookup **map_ret,
3101 u64 *num_bytes_out, u64 *stripe_size_out,
3102 u64 start, u64 type)
3103 {
3104 struct btrfs_fs_info *info = extent_root->fs_info;
3105 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3106 struct list_head *cur;
3107 struct map_lookup *map = NULL;
3108 struct extent_map_tree *em_tree;
3109 struct extent_map *em;
3110 struct btrfs_device_info *devices_info = NULL;
3111 u64 total_avail;
3112 int num_stripes; /* total number of stripes to allocate */
3113 int sub_stripes; /* sub_stripes info for map */
3114 int dev_stripes; /* stripes per dev */
3115 int devs_max; /* max devs to use */
3116 int devs_min; /* min devs needed */
3117 int devs_increment; /* ndevs has to be a multiple of this */
3118 int ncopies; /* how many copies to data has */
3119 int ret;
3120 u64 max_stripe_size;
3121 u64 max_chunk_size;
3122 u64 stripe_size;
3123 u64 num_bytes;
3124 int ndevs;
3125 int i;
3126 int j;
3127
3128 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
3129 (type & BTRFS_BLOCK_GROUP_DUP)) {
3130 WARN_ON(1);
3131 type &= ~BTRFS_BLOCK_GROUP_DUP;
3132 }
3133
3134 if (list_empty(&fs_devices->alloc_list))
3135 return -ENOSPC;
3136
3137 sub_stripes = 1;
3138 dev_stripes = 1;
3139 devs_increment = 1;
3140 ncopies = 1;
3141 devs_max = 0; /* 0 == as many as possible */
3142 devs_min = 1;
3143
3144 /*
3145 * define the properties of each RAID type.
3146 * FIXME: move this to a global table and use it in all RAID
3147 * calculation code
3148 */
3149 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3150 dev_stripes = 2;
3151 ncopies = 2;
3152 devs_max = 1;
3153 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3154 devs_min = 2;
3155 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3156 devs_increment = 2;
3157 ncopies = 2;
3158 devs_max = 2;
3159 devs_min = 2;
3160 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3161 sub_stripes = 2;
3162 devs_increment = 2;
3163 ncopies = 2;
3164 devs_min = 4;
3165 } else {
3166 devs_max = 1;
3167 }
3168
3169 if (type & BTRFS_BLOCK_GROUP_DATA) {
3170 max_stripe_size = 1024 * 1024 * 1024;
3171 max_chunk_size = 10 * max_stripe_size;
3172 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3173 /* for larger filesystems, use larger metadata chunks */
3174 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3175 max_stripe_size = 1024 * 1024 * 1024;
3176 else
3177 max_stripe_size = 256 * 1024 * 1024;
3178 max_chunk_size = max_stripe_size;
3179 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3180 max_stripe_size = 32 * 1024 * 1024;
3181 max_chunk_size = 2 * max_stripe_size;
3182 } else {
3183 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3184 type);
3185 BUG_ON(1);
3186 }
3187
3188 /* we don't want a chunk larger than 10% of writeable space */
3189 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3190 max_chunk_size);
3191
3192 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3193 GFP_NOFS);
3194 if (!devices_info)
3195 return -ENOMEM;
3196
3197 cur = fs_devices->alloc_list.next;
3198
3199 /*
3200 * in the first pass through the devices list, we gather information
3201 * about the available holes on each device.
3202 */
3203 ndevs = 0;
3204 while (cur != &fs_devices->alloc_list) {
3205 struct btrfs_device *device;
3206 u64 max_avail;
3207 u64 dev_offset;
3208
3209 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3210
3211 cur = cur->next;
3212
3213 if (!device->writeable) {
3214 printk(KERN_ERR
3215 "btrfs: read-only device in alloc_list\n");
3216 WARN_ON(1);
3217 continue;
3218 }
3219
3220 if (!device->in_fs_metadata)
3221 continue;
3222
3223 if (device->total_bytes > device->bytes_used)
3224 total_avail = device->total_bytes - device->bytes_used;
3225 else
3226 total_avail = 0;
3227
3228 /* If there is no space on this device, skip it. */
3229 if (total_avail == 0)
3230 continue;
3231
3232 ret = find_free_dev_extent(device,
3233 max_stripe_size * dev_stripes,
3234 &dev_offset, &max_avail);
3235 if (ret && ret != -ENOSPC)
3236 goto error;
3237
3238 if (ret == 0)
3239 max_avail = max_stripe_size * dev_stripes;
3240
3241 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3242 continue;
3243
3244 devices_info[ndevs].dev_offset = dev_offset;
3245 devices_info[ndevs].max_avail = max_avail;
3246 devices_info[ndevs].total_avail = total_avail;
3247 devices_info[ndevs].dev = device;
3248 ++ndevs;
3249 }
3250
3251 /*
3252 * now sort the devices by hole size / available space
3253 */
3254 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3255 btrfs_cmp_device_info, NULL);
3256
3257 /* round down to number of usable stripes */
3258 ndevs -= ndevs % devs_increment;
3259
3260 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3261 ret = -ENOSPC;
3262 goto error;
3263 }
3264
3265 if (devs_max && ndevs > devs_max)
3266 ndevs = devs_max;
3267 /*
3268 * the primary goal is to maximize the number of stripes, so use as many
3269 * devices as possible, even if the stripes are not maximum sized.
3270 */
3271 stripe_size = devices_info[ndevs-1].max_avail;
3272 num_stripes = ndevs * dev_stripes;
3273
3274 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
3275 stripe_size = max_chunk_size * ncopies;
3276 do_div(stripe_size, num_stripes);
3277 }
3278
3279 do_div(stripe_size, dev_stripes);
3280 do_div(stripe_size, BTRFS_STRIPE_LEN);
3281 stripe_size *= BTRFS_STRIPE_LEN;
3282
3283 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3284 if (!map) {
3285 ret = -ENOMEM;
3286 goto error;
3287 }
3288 map->num_stripes = num_stripes;
3289
3290 for (i = 0; i < ndevs; ++i) {
3291 for (j = 0; j < dev_stripes; ++j) {
3292 int s = i * dev_stripes + j;
3293 map->stripes[s].dev = devices_info[i].dev;
3294 map->stripes[s].physical = devices_info[i].dev_offset +
3295 j * stripe_size;
3296 }
3297 }
3298 map->sector_size = extent_root->sectorsize;
3299 map->stripe_len = BTRFS_STRIPE_LEN;
3300 map->io_align = BTRFS_STRIPE_LEN;
3301 map->io_width = BTRFS_STRIPE_LEN;
3302 map->type = type;
3303 map->sub_stripes = sub_stripes;
3304
3305 *map_ret = map;
3306 num_bytes = stripe_size * (num_stripes / ncopies);
3307
3308 *stripe_size_out = stripe_size;
3309 *num_bytes_out = num_bytes;
3310
3311 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3312
3313 em = alloc_extent_map();
3314 if (!em) {
3315 ret = -ENOMEM;
3316 goto error;
3317 }
3318 em->bdev = (struct block_device *)map;
3319 em->start = start;
3320 em->len = num_bytes;
3321 em->block_start = 0;
3322 em->block_len = em->len;
3323
3324 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3325 write_lock(&em_tree->lock);
3326 ret = add_extent_mapping(em_tree, em);
3327 write_unlock(&em_tree->lock);
3328 BUG_ON(ret);
3329 free_extent_map(em);
3330
3331 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3332 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3333 start, num_bytes);
3334 BUG_ON(ret);
3335
3336 for (i = 0; i < map->num_stripes; ++i) {
3337 struct btrfs_device *device;
3338 u64 dev_offset;
3339
3340 device = map->stripes[i].dev;
3341 dev_offset = map->stripes[i].physical;
3342
3343 ret = btrfs_alloc_dev_extent(trans, device,
3344 info->chunk_root->root_key.objectid,
3345 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3346 start, dev_offset, stripe_size);
3347 BUG_ON(ret);
3348 }
3349
3350 kfree(devices_info);
3351 return 0;
3352
3353 error:
3354 kfree(map);
3355 kfree(devices_info);
3356 return ret;
3357 }
3358
3359 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3360 struct btrfs_root *extent_root,
3361 struct map_lookup *map, u64 chunk_offset,
3362 u64 chunk_size, u64 stripe_size)
3363 {
3364 u64 dev_offset;
3365 struct btrfs_key key;
3366 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3367 struct btrfs_device *device;
3368 struct btrfs_chunk *chunk;
3369 struct btrfs_stripe *stripe;
3370 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3371 int index = 0;
3372 int ret;
3373
3374 chunk = kzalloc(item_size, GFP_NOFS);
3375 if (!chunk)
3376 return -ENOMEM;
3377
3378 index = 0;
3379 while (index < map->num_stripes) {
3380 device = map->stripes[index].dev;
3381 device->bytes_used += stripe_size;
3382 ret = btrfs_update_device(trans, device);
3383 BUG_ON(ret);
3384 index++;
3385 }
3386
3387 spin_lock(&extent_root->fs_info->free_chunk_lock);
3388 extent_root->fs_info->free_chunk_space -= (stripe_size *
3389 map->num_stripes);
3390 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3391
3392 index = 0;
3393 stripe = &chunk->stripe;
3394 while (index < map->num_stripes) {
3395 device = map->stripes[index].dev;
3396 dev_offset = map->stripes[index].physical;
3397
3398 btrfs_set_stack_stripe_devid(stripe, device->devid);
3399 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3400 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3401 stripe++;
3402 index++;
3403 }
3404
3405 btrfs_set_stack_chunk_length(chunk, chunk_size);
3406 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3407 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3408 btrfs_set_stack_chunk_type(chunk, map->type);
3409 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3410 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3411 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3412 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3413 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3414
3415 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3416 key.type = BTRFS_CHUNK_ITEM_KEY;
3417 key.offset = chunk_offset;
3418
3419 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3420 BUG_ON(ret);
3421
3422 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3423 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3424 item_size);
3425 BUG_ON(ret);
3426 }
3427
3428 kfree(chunk);
3429 return 0;
3430 }
3431
3432 /*
3433 * Chunk allocation falls into two parts. The first part does works
3434 * that make the new allocated chunk useable, but not do any operation
3435 * that modifies the chunk tree. The second part does the works that
3436 * require modifying the chunk tree. This division is important for the
3437 * bootstrap process of adding storage to a seed btrfs.
3438 */
3439 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3440 struct btrfs_root *extent_root, u64 type)
3441 {
3442 u64 chunk_offset;
3443 u64 chunk_size;
3444 u64 stripe_size;
3445 struct map_lookup *map;
3446 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3447 int ret;
3448
3449 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3450 &chunk_offset);
3451 if (ret)
3452 return ret;
3453
3454 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3455 &stripe_size, chunk_offset, type);
3456 if (ret)
3457 return ret;
3458
3459 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3460 chunk_size, stripe_size);
3461 BUG_ON(ret);
3462 return 0;
3463 }
3464
3465 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3466 struct btrfs_root *root,
3467 struct btrfs_device *device)
3468 {
3469 u64 chunk_offset;
3470 u64 sys_chunk_offset;
3471 u64 chunk_size;
3472 u64 sys_chunk_size;
3473 u64 stripe_size;
3474 u64 sys_stripe_size;
3475 u64 alloc_profile;
3476 struct map_lookup *map;
3477 struct map_lookup *sys_map;
3478 struct btrfs_fs_info *fs_info = root->fs_info;
3479 struct btrfs_root *extent_root = fs_info->extent_root;
3480 int ret;
3481
3482 ret = find_next_chunk(fs_info->chunk_root,
3483 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3484 if (ret)
3485 return ret;
3486
3487 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3488 fs_info->avail_metadata_alloc_bits;
3489 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3490
3491 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3492 &stripe_size, chunk_offset, alloc_profile);
3493 BUG_ON(ret);
3494
3495 sys_chunk_offset = chunk_offset + chunk_size;
3496
3497 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3498 fs_info->avail_system_alloc_bits;
3499 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3500
3501 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3502 &sys_chunk_size, &sys_stripe_size,
3503 sys_chunk_offset, alloc_profile);
3504 BUG_ON(ret);
3505
3506 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3507 BUG_ON(ret);
3508
3509 /*
3510 * Modifying chunk tree needs allocating new blocks from both
3511 * system block group and metadata block group. So we only can
3512 * do operations require modifying the chunk tree after both
3513 * block groups were created.
3514 */
3515 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3516 chunk_size, stripe_size);
3517 BUG_ON(ret);
3518
3519 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3520 sys_chunk_offset, sys_chunk_size,
3521 sys_stripe_size);
3522 BUG_ON(ret);
3523 return 0;
3524 }
3525
3526 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3527 {
3528 struct extent_map *em;
3529 struct map_lookup *map;
3530 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3531 int readonly = 0;
3532 int i;
3533
3534 read_lock(&map_tree->map_tree.lock);
3535 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3536 read_unlock(&map_tree->map_tree.lock);
3537 if (!em)
3538 return 1;
3539
3540 if (btrfs_test_opt(root, DEGRADED)) {
3541 free_extent_map(em);
3542 return 0;
3543 }
3544
3545 map = (struct map_lookup *)em->bdev;
3546 for (i = 0; i < map->num_stripes; i++) {
3547 if (!map->stripes[i].dev->writeable) {
3548 readonly = 1;
3549 break;
3550 }
3551 }
3552 free_extent_map(em);
3553 return readonly;
3554 }
3555
3556 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3557 {
3558 extent_map_tree_init(&tree->map_tree);
3559 }
3560
3561 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3562 {
3563 struct extent_map *em;
3564
3565 while (1) {
3566 write_lock(&tree->map_tree.lock);
3567 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3568 if (em)
3569 remove_extent_mapping(&tree->map_tree, em);
3570 write_unlock(&tree->map_tree.lock);
3571 if (!em)
3572 break;
3573 kfree(em->bdev);
3574 /* once for us */
3575 free_extent_map(em);
3576 /* once for the tree */
3577 free_extent_map(em);
3578 }
3579 }
3580
3581 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3582 {
3583 struct extent_map *em;
3584 struct map_lookup *map;
3585 struct extent_map_tree *em_tree = &map_tree->map_tree;
3586 int ret;
3587
3588 read_lock(&em_tree->lock);
3589 em = lookup_extent_mapping(em_tree, logical, len);
3590 read_unlock(&em_tree->lock);
3591 BUG_ON(!em);
3592
3593 BUG_ON(em->start > logical || em->start + em->len < logical);
3594 map = (struct map_lookup *)em->bdev;
3595 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3596 ret = map->num_stripes;
3597 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3598 ret = map->sub_stripes;
3599 else
3600 ret = 1;
3601 free_extent_map(em);
3602 return ret;
3603 }
3604
3605 static int find_live_mirror(struct map_lookup *map, int first, int num,
3606 int optimal)
3607 {
3608 int i;
3609 if (map->stripes[optimal].dev->bdev)
3610 return optimal;
3611 for (i = first; i < first + num; i++) {
3612 if (map->stripes[i].dev->bdev)
3613 return i;
3614 }
3615 /* we couldn't find one that doesn't fail. Just return something
3616 * and the io error handling code will clean up eventually
3617 */
3618 return optimal;
3619 }
3620
3621 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3622 u64 logical, u64 *length,
3623 struct btrfs_bio **bbio_ret,
3624 int mirror_num)
3625 {
3626 struct extent_map *em;
3627 struct map_lookup *map;
3628 struct extent_map_tree *em_tree = &map_tree->map_tree;
3629 u64 offset;
3630 u64 stripe_offset;
3631 u64 stripe_end_offset;
3632 u64 stripe_nr;
3633 u64 stripe_nr_orig;
3634 u64 stripe_nr_end;
3635 int stripe_index;
3636 int i;
3637 int ret = 0;
3638 int num_stripes;
3639 int max_errors = 0;
3640 struct btrfs_bio *bbio = NULL;
3641
3642 read_lock(&em_tree->lock);
3643 em = lookup_extent_mapping(em_tree, logical, *length);
3644 read_unlock(&em_tree->lock);
3645
3646 if (!em) {
3647 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3648 (unsigned long long)logical,
3649 (unsigned long long)*length);
3650 BUG();
3651 }
3652
3653 BUG_ON(em->start > logical || em->start + em->len < logical);
3654 map = (struct map_lookup *)em->bdev;
3655 offset = logical - em->start;
3656
3657 if (mirror_num > map->num_stripes)
3658 mirror_num = 0;
3659
3660 stripe_nr = offset;
3661 /*
3662 * stripe_nr counts the total number of stripes we have to stride
3663 * to get to this block
3664 */
3665 do_div(stripe_nr, map->stripe_len);
3666
3667 stripe_offset = stripe_nr * map->stripe_len;
3668 BUG_ON(offset < stripe_offset);
3669
3670 /* stripe_offset is the offset of this block in its stripe*/
3671 stripe_offset = offset - stripe_offset;
3672
3673 if (rw & REQ_DISCARD)
3674 *length = min_t(u64, em->len - offset, *length);
3675 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3676 /* we limit the length of each bio to what fits in a stripe */
3677 *length = min_t(u64, em->len - offset,
3678 map->stripe_len - stripe_offset);
3679 } else {
3680 *length = em->len - offset;
3681 }
3682
3683 if (!bbio_ret)
3684 goto out;
3685
3686 num_stripes = 1;
3687 stripe_index = 0;
3688 stripe_nr_orig = stripe_nr;
3689 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3690 (~(map->stripe_len - 1));
3691 do_div(stripe_nr_end, map->stripe_len);
3692 stripe_end_offset = stripe_nr_end * map->stripe_len -
3693 (offset + *length);
3694 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3695 if (rw & REQ_DISCARD)
3696 num_stripes = min_t(u64, map->num_stripes,
3697 stripe_nr_end - stripe_nr_orig);
3698 stripe_index = do_div(stripe_nr, map->num_stripes);
3699 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3700 if (rw & (REQ_WRITE | REQ_DISCARD))
3701 num_stripes = map->num_stripes;
3702 else if (mirror_num)
3703 stripe_index = mirror_num - 1;
3704 else {
3705 stripe_index = find_live_mirror(map, 0,
3706 map->num_stripes,
3707 current->pid % map->num_stripes);
3708 mirror_num = stripe_index + 1;
3709 }
3710
3711 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3712 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3713 num_stripes = map->num_stripes;
3714 } else if (mirror_num) {
3715 stripe_index = mirror_num - 1;
3716 } else {
3717 mirror_num = 1;
3718 }
3719
3720 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3721 int factor = map->num_stripes / map->sub_stripes;
3722
3723 stripe_index = do_div(stripe_nr, factor);
3724 stripe_index *= map->sub_stripes;
3725
3726 if (rw & REQ_WRITE)
3727 num_stripes = map->sub_stripes;
3728 else if (rw & REQ_DISCARD)
3729 num_stripes = min_t(u64, map->sub_stripes *
3730 (stripe_nr_end - stripe_nr_orig),
3731 map->num_stripes);
3732 else if (mirror_num)
3733 stripe_index += mirror_num - 1;
3734 else {
3735 stripe_index = find_live_mirror(map, stripe_index,
3736 map->sub_stripes, stripe_index +
3737 current->pid % map->sub_stripes);
3738 mirror_num = stripe_index + 1;
3739 }
3740 } else {
3741 /*
3742 * after this do_div call, stripe_nr is the number of stripes
3743 * on this device we have to walk to find the data, and
3744 * stripe_index is the number of our device in the stripe array
3745 */
3746 stripe_index = do_div(stripe_nr, map->num_stripes);
3747 mirror_num = stripe_index + 1;
3748 }
3749 BUG_ON(stripe_index >= map->num_stripes);
3750
3751 bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3752 if (!bbio) {
3753 ret = -ENOMEM;
3754 goto out;
3755 }
3756 atomic_set(&bbio->error, 0);
3757
3758 if (rw & REQ_DISCARD) {
3759 int factor = 0;
3760 int sub_stripes = 0;
3761 u64 stripes_per_dev = 0;
3762 u32 remaining_stripes = 0;
3763
3764 if (map->type &
3765 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3766 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3767 sub_stripes = 1;
3768 else
3769 sub_stripes = map->sub_stripes;
3770
3771 factor = map->num_stripes / sub_stripes;
3772 stripes_per_dev = div_u64_rem(stripe_nr_end -
3773 stripe_nr_orig,
3774 factor,
3775 &remaining_stripes);
3776 }
3777
3778 for (i = 0; i < num_stripes; i++) {
3779 bbio->stripes[i].physical =
3780 map->stripes[stripe_index].physical +
3781 stripe_offset + stripe_nr * map->stripe_len;
3782 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3783
3784 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3785 BTRFS_BLOCK_GROUP_RAID10)) {
3786 bbio->stripes[i].length = stripes_per_dev *
3787 map->stripe_len;
3788 if (i / sub_stripes < remaining_stripes)
3789 bbio->stripes[i].length +=
3790 map->stripe_len;
3791 if (i < sub_stripes)
3792 bbio->stripes[i].length -=
3793 stripe_offset;
3794 if ((i / sub_stripes + 1) %
3795 sub_stripes == remaining_stripes)
3796 bbio->stripes[i].length -=
3797 stripe_end_offset;
3798 if (i == sub_stripes - 1)
3799 stripe_offset = 0;
3800 } else
3801 bbio->stripes[i].length = *length;
3802
3803 stripe_index++;
3804 if (stripe_index == map->num_stripes) {
3805 /* This could only happen for RAID0/10 */
3806 stripe_index = 0;
3807 stripe_nr++;
3808 }
3809 }
3810 } else {
3811 for (i = 0; i < num_stripes; i++) {
3812 bbio->stripes[i].physical =
3813 map->stripes[stripe_index].physical +
3814 stripe_offset +
3815 stripe_nr * map->stripe_len;
3816 bbio->stripes[i].dev =
3817 map->stripes[stripe_index].dev;
3818 stripe_index++;
3819 }
3820 }
3821
3822 if (rw & REQ_WRITE) {
3823 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3824 BTRFS_BLOCK_GROUP_RAID10 |
3825 BTRFS_BLOCK_GROUP_DUP)) {
3826 max_errors = 1;
3827 }
3828 }
3829
3830 *bbio_ret = bbio;
3831 bbio->num_stripes = num_stripes;
3832 bbio->max_errors = max_errors;
3833 bbio->mirror_num = mirror_num;
3834 out:
3835 free_extent_map(em);
3836 return ret;
3837 }
3838
3839 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3840 u64 logical, u64 *length,
3841 struct btrfs_bio **bbio_ret, int mirror_num)
3842 {
3843 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3844 mirror_num);
3845 }
3846
3847 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3848 u64 chunk_start, u64 physical, u64 devid,
3849 u64 **logical, int *naddrs, int *stripe_len)
3850 {
3851 struct extent_map_tree *em_tree = &map_tree->map_tree;
3852 struct extent_map *em;
3853 struct map_lookup *map;
3854 u64 *buf;
3855 u64 bytenr;
3856 u64 length;
3857 u64 stripe_nr;
3858 int i, j, nr = 0;
3859
3860 read_lock(&em_tree->lock);
3861 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3862 read_unlock(&em_tree->lock);
3863
3864 BUG_ON(!em || em->start != chunk_start);
3865 map = (struct map_lookup *)em->bdev;
3866
3867 length = em->len;
3868 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3869 do_div(length, map->num_stripes / map->sub_stripes);
3870 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3871 do_div(length, map->num_stripes);
3872
3873 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3874 BUG_ON(!buf);
3875
3876 for (i = 0; i < map->num_stripes; i++) {
3877 if (devid && map->stripes[i].dev->devid != devid)
3878 continue;
3879 if (map->stripes[i].physical > physical ||
3880 map->stripes[i].physical + length <= physical)
3881 continue;
3882
3883 stripe_nr = physical - map->stripes[i].physical;
3884 do_div(stripe_nr, map->stripe_len);
3885
3886 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3887 stripe_nr = stripe_nr * map->num_stripes + i;
3888 do_div(stripe_nr, map->sub_stripes);
3889 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3890 stripe_nr = stripe_nr * map->num_stripes + i;
3891 }
3892 bytenr = chunk_start + stripe_nr * map->stripe_len;
3893 WARN_ON(nr >= map->num_stripes);
3894 for (j = 0; j < nr; j++) {
3895 if (buf[j] == bytenr)
3896 break;
3897 }
3898 if (j == nr) {
3899 WARN_ON(nr >= map->num_stripes);
3900 buf[nr++] = bytenr;
3901 }
3902 }
3903
3904 *logical = buf;
3905 *naddrs = nr;
3906 *stripe_len = map->stripe_len;
3907
3908 free_extent_map(em);
3909 return 0;
3910 }
3911
3912 static void btrfs_end_bio(struct bio *bio, int err)
3913 {
3914 struct btrfs_bio *bbio = bio->bi_private;
3915 int is_orig_bio = 0;
3916
3917 if (err)
3918 atomic_inc(&bbio->error);
3919
3920 if (bio == bbio->orig_bio)
3921 is_orig_bio = 1;
3922
3923 if (atomic_dec_and_test(&bbio->stripes_pending)) {
3924 if (!is_orig_bio) {
3925 bio_put(bio);
3926 bio = bbio->orig_bio;
3927 }
3928 bio->bi_private = bbio->private;
3929 bio->bi_end_io = bbio->end_io;
3930 bio->bi_bdev = (struct block_device *)
3931 (unsigned long)bbio->mirror_num;
3932 /* only send an error to the higher layers if it is
3933 * beyond the tolerance of the multi-bio
3934 */
3935 if (atomic_read(&bbio->error) > bbio->max_errors) {
3936 err = -EIO;
3937 } else {
3938 /*
3939 * this bio is actually up to date, we didn't
3940 * go over the max number of errors
3941 */
3942 set_bit(BIO_UPTODATE, &bio->bi_flags);
3943 err = 0;
3944 }
3945 kfree(bbio);
3946
3947 bio_endio(bio, err);
3948 } else if (!is_orig_bio) {
3949 bio_put(bio);
3950 }
3951 }
3952
3953 struct async_sched {
3954 struct bio *bio;
3955 int rw;
3956 struct btrfs_fs_info *info;
3957 struct btrfs_work work;
3958 };
3959
3960 /*
3961 * see run_scheduled_bios for a description of why bios are collected for
3962 * async submit.
3963 *
3964 * This will add one bio to the pending list for a device and make sure
3965 * the work struct is scheduled.
3966 */
3967 static noinline void schedule_bio(struct btrfs_root *root,
3968 struct btrfs_device *device,
3969 int rw, struct bio *bio)
3970 {
3971 int should_queue = 1;
3972 struct btrfs_pending_bios *pending_bios;
3973
3974 /* don't bother with additional async steps for reads, right now */
3975 if (!(rw & REQ_WRITE)) {
3976 bio_get(bio);
3977 btrfsic_submit_bio(rw, bio);
3978 bio_put(bio);
3979 return;
3980 }
3981
3982 /*
3983 * nr_async_bios allows us to reliably return congestion to the
3984 * higher layers. Otherwise, the async bio makes it appear we have
3985 * made progress against dirty pages when we've really just put it
3986 * on a queue for later
3987 */
3988 atomic_inc(&root->fs_info->nr_async_bios);
3989 WARN_ON(bio->bi_next);
3990 bio->bi_next = NULL;
3991 bio->bi_rw |= rw;
3992
3993 spin_lock(&device->io_lock);
3994 if (bio->bi_rw & REQ_SYNC)
3995 pending_bios = &device->pending_sync_bios;
3996 else
3997 pending_bios = &device->pending_bios;
3998
3999 if (pending_bios->tail)
4000 pending_bios->tail->bi_next = bio;
4001
4002 pending_bios->tail = bio;
4003 if (!pending_bios->head)
4004 pending_bios->head = bio;
4005 if (device->running_pending)
4006 should_queue = 0;
4007
4008 spin_unlock(&device->io_lock);
4009
4010 if (should_queue)
4011 btrfs_queue_worker(&root->fs_info->submit_workers,
4012 &device->work);
4013 }
4014
4015 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4016 int mirror_num, int async_submit)
4017 {
4018 struct btrfs_mapping_tree *map_tree;
4019 struct btrfs_device *dev;
4020 struct bio *first_bio = bio;
4021 u64 logical = (u64)bio->bi_sector << 9;
4022 u64 length = 0;
4023 u64 map_length;
4024 int ret;
4025 int dev_nr = 0;
4026 int total_devs = 1;
4027 struct btrfs_bio *bbio = NULL;
4028
4029 length = bio->bi_size;
4030 map_tree = &root->fs_info->mapping_tree;
4031 map_length = length;
4032
4033 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4034 mirror_num);
4035 BUG_ON(ret);
4036
4037 total_devs = bbio->num_stripes;
4038 if (map_length < length) {
4039 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4040 "len %llu\n", (unsigned long long)logical,
4041 (unsigned long long)length,
4042 (unsigned long long)map_length);
4043 BUG();
4044 }
4045
4046 bbio->orig_bio = first_bio;
4047 bbio->private = first_bio->bi_private;
4048 bbio->end_io = first_bio->bi_end_io;
4049 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4050
4051 while (dev_nr < total_devs) {
4052 if (dev_nr < total_devs - 1) {
4053 bio = bio_clone(first_bio, GFP_NOFS);
4054 BUG_ON(!bio);
4055 } else {
4056 bio = first_bio;
4057 }
4058 bio->bi_private = bbio;
4059 bio->bi_end_io = btrfs_end_bio;
4060 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4061 dev = bbio->stripes[dev_nr].dev;
4062 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4063 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4064 "(%s id %llu), size=%u\n", rw,
4065 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4066 dev->name, dev->devid, bio->bi_size);
4067 bio->bi_bdev = dev->bdev;
4068 if (async_submit)
4069 schedule_bio(root, dev, rw, bio);
4070 else
4071 btrfsic_submit_bio(rw, bio);
4072 } else {
4073 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4074 bio->bi_sector = logical >> 9;
4075 bio_endio(bio, -EIO);
4076 }
4077 dev_nr++;
4078 }
4079 return 0;
4080 }
4081
4082 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4083 u8 *uuid, u8 *fsid)
4084 {
4085 struct btrfs_device *device;
4086 struct btrfs_fs_devices *cur_devices;
4087
4088 cur_devices = root->fs_info->fs_devices;
4089 while (cur_devices) {
4090 if (!fsid ||
4091 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4092 device = __find_device(&cur_devices->devices,
4093 devid, uuid);
4094 if (device)
4095 return device;
4096 }
4097 cur_devices = cur_devices->seed;
4098 }
4099 return NULL;
4100 }
4101
4102 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4103 u64 devid, u8 *dev_uuid)
4104 {
4105 struct btrfs_device *device;
4106 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4107
4108 device = kzalloc(sizeof(*device), GFP_NOFS);
4109 if (!device)
4110 return NULL;
4111 list_add(&device->dev_list,
4112 &fs_devices->devices);
4113 device->dev_root = root->fs_info->dev_root;
4114 device->devid = devid;
4115 device->work.func = pending_bios_fn;
4116 device->fs_devices = fs_devices;
4117 device->missing = 1;
4118 fs_devices->num_devices++;
4119 fs_devices->missing_devices++;
4120 spin_lock_init(&device->io_lock);
4121 INIT_LIST_HEAD(&device->dev_alloc_list);
4122 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4123 return device;
4124 }
4125
4126 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4127 struct extent_buffer *leaf,
4128 struct btrfs_chunk *chunk)
4129 {
4130 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4131 struct map_lookup *map;
4132 struct extent_map *em;
4133 u64 logical;
4134 u64 length;
4135 u64 devid;
4136 u8 uuid[BTRFS_UUID_SIZE];
4137 int num_stripes;
4138 int ret;
4139 int i;
4140
4141 logical = key->offset;
4142 length = btrfs_chunk_length(leaf, chunk);
4143
4144 read_lock(&map_tree->map_tree.lock);
4145 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4146 read_unlock(&map_tree->map_tree.lock);
4147
4148 /* already mapped? */
4149 if (em && em->start <= logical && em->start + em->len > logical) {
4150 free_extent_map(em);
4151 return 0;
4152 } else if (em) {
4153 free_extent_map(em);
4154 }
4155
4156 em = alloc_extent_map();
4157 if (!em)
4158 return -ENOMEM;
4159 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4160 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4161 if (!map) {
4162 free_extent_map(em);
4163 return -ENOMEM;
4164 }
4165
4166 em->bdev = (struct block_device *)map;
4167 em->start = logical;
4168 em->len = length;
4169 em->block_start = 0;
4170 em->block_len = em->len;
4171
4172 map->num_stripes = num_stripes;
4173 map->io_width = btrfs_chunk_io_width(leaf, chunk);
4174 map->io_align = btrfs_chunk_io_align(leaf, chunk);
4175 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4176 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4177 map->type = btrfs_chunk_type(leaf, chunk);
4178 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4179 for (i = 0; i < num_stripes; i++) {
4180 map->stripes[i].physical =
4181 btrfs_stripe_offset_nr(leaf, chunk, i);
4182 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4183 read_extent_buffer(leaf, uuid, (unsigned long)
4184 btrfs_stripe_dev_uuid_nr(chunk, i),
4185 BTRFS_UUID_SIZE);
4186 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4187 NULL);
4188 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4189 kfree(map);
4190 free_extent_map(em);
4191 return -EIO;
4192 }
4193 if (!map->stripes[i].dev) {
4194 map->stripes[i].dev =
4195 add_missing_dev(root, devid, uuid);
4196 if (!map->stripes[i].dev) {
4197 kfree(map);
4198 free_extent_map(em);
4199 return -EIO;
4200 }
4201 }
4202 map->stripes[i].dev->in_fs_metadata = 1;
4203 }
4204
4205 write_lock(&map_tree->map_tree.lock);
4206 ret = add_extent_mapping(&map_tree->map_tree, em);
4207 write_unlock(&map_tree->map_tree.lock);
4208 BUG_ON(ret);
4209 free_extent_map(em);
4210
4211 return 0;
4212 }
4213
4214 static void fill_device_from_item(struct extent_buffer *leaf,
4215 struct btrfs_dev_item *dev_item,
4216 struct btrfs_device *device)
4217 {
4218 unsigned long ptr;
4219
4220 device->devid = btrfs_device_id(leaf, dev_item);
4221 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4222 device->total_bytes = device->disk_total_bytes;
4223 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4224 device->type = btrfs_device_type(leaf, dev_item);
4225 device->io_align = btrfs_device_io_align(leaf, dev_item);
4226 device->io_width = btrfs_device_io_width(leaf, dev_item);
4227 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4228
4229 ptr = (unsigned long)btrfs_device_uuid(dev_item);
4230 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4231 }
4232
4233 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4234 {
4235 struct btrfs_fs_devices *fs_devices;
4236 int ret;
4237
4238 BUG_ON(!mutex_is_locked(&uuid_mutex));
4239
4240 fs_devices = root->fs_info->fs_devices->seed;
4241 while (fs_devices) {
4242 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4243 ret = 0;
4244 goto out;
4245 }
4246 fs_devices = fs_devices->seed;
4247 }
4248
4249 fs_devices = find_fsid(fsid);
4250 if (!fs_devices) {
4251 ret = -ENOENT;
4252 goto out;
4253 }
4254
4255 fs_devices = clone_fs_devices(fs_devices);
4256 if (IS_ERR(fs_devices)) {
4257 ret = PTR_ERR(fs_devices);
4258 goto out;
4259 }
4260
4261 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4262 root->fs_info->bdev_holder);
4263 if (ret)
4264 goto out;
4265
4266 if (!fs_devices->seeding) {
4267 __btrfs_close_devices(fs_devices);
4268 free_fs_devices(fs_devices);
4269 ret = -EINVAL;
4270 goto out;
4271 }
4272
4273 fs_devices->seed = root->fs_info->fs_devices->seed;
4274 root->fs_info->fs_devices->seed = fs_devices;
4275 out:
4276 return ret;
4277 }
4278
4279 static int read_one_dev(struct btrfs_root *root,
4280 struct extent_buffer *leaf,
4281 struct btrfs_dev_item *dev_item)
4282 {
4283 struct btrfs_device *device;
4284 u64 devid;
4285 int ret;
4286 u8 fs_uuid[BTRFS_UUID_SIZE];
4287 u8 dev_uuid[BTRFS_UUID_SIZE];
4288
4289 devid = btrfs_device_id(leaf, dev_item);
4290 read_extent_buffer(leaf, dev_uuid,
4291 (unsigned long)btrfs_device_uuid(dev_item),
4292 BTRFS_UUID_SIZE);
4293 read_extent_buffer(leaf, fs_uuid,
4294 (unsigned long)btrfs_device_fsid(dev_item),
4295 BTRFS_UUID_SIZE);
4296
4297 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4298 ret = open_seed_devices(root, fs_uuid);
4299 if (ret && !btrfs_test_opt(root, DEGRADED))
4300 return ret;
4301 }
4302
4303 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4304 if (!device || !device->bdev) {
4305 if (!btrfs_test_opt(root, DEGRADED))
4306 return -EIO;
4307
4308 if (!device) {
4309 printk(KERN_WARNING "warning devid %llu missing\n",
4310 (unsigned long long)devid);
4311 device = add_missing_dev(root, devid, dev_uuid);
4312 if (!device)
4313 return -ENOMEM;
4314 } else if (!device->missing) {
4315 /*
4316 * this happens when a device that was properly setup
4317 * in the device info lists suddenly goes bad.
4318 * device->bdev is NULL, and so we have to set
4319 * device->missing to one here
4320 */
4321 root->fs_info->fs_devices->missing_devices++;
4322 device->missing = 1;
4323 }
4324 }
4325
4326 if (device->fs_devices != root->fs_info->fs_devices) {
4327 BUG_ON(device->writeable);
4328 if (device->generation !=
4329 btrfs_device_generation(leaf, dev_item))
4330 return -EINVAL;
4331 }
4332
4333 fill_device_from_item(leaf, dev_item, device);
4334 device->dev_root = root->fs_info->dev_root;
4335 device->in_fs_metadata = 1;
4336 if (device->writeable) {
4337 device->fs_devices->total_rw_bytes += device->total_bytes;
4338 spin_lock(&root->fs_info->free_chunk_lock);
4339 root->fs_info->free_chunk_space += device->total_bytes -
4340 device->bytes_used;
4341 spin_unlock(&root->fs_info->free_chunk_lock);
4342 }
4343 ret = 0;
4344 return ret;
4345 }
4346
4347 int btrfs_read_sys_array(struct btrfs_root *root)
4348 {
4349 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4350 struct extent_buffer *sb;
4351 struct btrfs_disk_key *disk_key;
4352 struct btrfs_chunk *chunk;
4353 u8 *ptr;
4354 unsigned long sb_ptr;
4355 int ret = 0;
4356 u32 num_stripes;
4357 u32 array_size;
4358 u32 len = 0;
4359 u32 cur;
4360 struct btrfs_key key;
4361
4362 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4363 BTRFS_SUPER_INFO_SIZE);
4364 if (!sb)
4365 return -ENOMEM;
4366 btrfs_set_buffer_uptodate(sb);
4367 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4368 /*
4369 * The sb extent buffer is artifical and just used to read the system array.
4370 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4371 * pages up-to-date when the page is larger: extent does not cover the
4372 * whole page and consequently check_page_uptodate does not find all
4373 * the page's extents up-to-date (the hole beyond sb),
4374 * write_extent_buffer then triggers a WARN_ON.
4375 *
4376 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4377 * but sb spans only this function. Add an explicit SetPageUptodate call
4378 * to silence the warning eg. on PowerPC 64.
4379 */
4380 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4381 SetPageUptodate(sb->first_page);
4382
4383 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4384 array_size = btrfs_super_sys_array_size(super_copy);
4385
4386 ptr = super_copy->sys_chunk_array;
4387 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4388 cur = 0;
4389
4390 while (cur < array_size) {
4391 disk_key = (struct btrfs_disk_key *)ptr;
4392 btrfs_disk_key_to_cpu(&key, disk_key);
4393
4394 len = sizeof(*disk_key); ptr += len;
4395 sb_ptr += len;
4396 cur += len;
4397
4398 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4399 chunk = (struct btrfs_chunk *)sb_ptr;
4400 ret = read_one_chunk(root, &key, sb, chunk);
4401 if (ret)
4402 break;
4403 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4404 len = btrfs_chunk_item_size(num_stripes);
4405 } else {
4406 ret = -EIO;
4407 break;
4408 }
4409 ptr += len;
4410 sb_ptr += len;
4411 cur += len;
4412 }
4413 free_extent_buffer(sb);
4414 return ret;
4415 }
4416
4417 int btrfs_read_chunk_tree(struct btrfs_root *root)
4418 {
4419 struct btrfs_path *path;
4420 struct extent_buffer *leaf;
4421 struct btrfs_key key;
4422 struct btrfs_key found_key;
4423 int ret;
4424 int slot;
4425
4426 root = root->fs_info->chunk_root;
4427
4428 path = btrfs_alloc_path();
4429 if (!path)
4430 return -ENOMEM;
4431
4432 mutex_lock(&uuid_mutex);
4433 lock_chunks(root);
4434
4435 /* first we search for all of the device items, and then we
4436 * read in all of the chunk items. This way we can create chunk
4437 * mappings that reference all of the devices that are afound
4438 */
4439 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4440 key.offset = 0;
4441 key.type = 0;
4442 again:
4443 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4444 if (ret < 0)
4445 goto error;
4446 while (1) {
4447 leaf = path->nodes[0];
4448 slot = path->slots[0];
4449 if (slot >= btrfs_header_nritems(leaf)) {
4450 ret = btrfs_next_leaf(root, path);
4451 if (ret == 0)
4452 continue;
4453 if (ret < 0)
4454 goto error;
4455 break;
4456 }
4457 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4458 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4459 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4460 break;
4461 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4462 struct btrfs_dev_item *dev_item;
4463 dev_item = btrfs_item_ptr(leaf, slot,
4464 struct btrfs_dev_item);
4465 ret = read_one_dev(root, leaf, dev_item);
4466 if (ret)
4467 goto error;
4468 }
4469 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4470 struct btrfs_chunk *chunk;
4471 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4472 ret = read_one_chunk(root, &found_key, leaf, chunk);
4473 if (ret)
4474 goto error;
4475 }
4476 path->slots[0]++;
4477 }
4478 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4479 key.objectid = 0;
4480 btrfs_release_path(path);
4481 goto again;
4482 }
4483 ret = 0;
4484 error:
4485 unlock_chunks(root);
4486 mutex_unlock(&uuid_mutex);
4487
4488 btrfs_free_path(path);
4489 return ret;
4490 }
Linux kernel - Deliverables
This page took 0.136143 seconds and 5 git commands to generate.