projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: fix chunk allocation error handling
[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/ratelimit.h>
27 #include <linux/kthread.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 #include "rcu-string.h"
38 #include "math.h"
39 #include "dev-replace.h"
40
41 static int init_first_rw_device(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 struct btrfs_device *device);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
45 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
47
48 static DEFINE_MUTEX(uuid_mutex);
49 static LIST_HEAD(fs_uuids);
50
51 static void lock_chunks(struct btrfs_root *root)
52 {
53 mutex_lock(&root->fs_info->chunk_mutex);
54 }
55
56 static void unlock_chunks(struct btrfs_root *root)
57 {
58 mutex_unlock(&root->fs_info->chunk_mutex);
59 }
60
61 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62 {
63 struct btrfs_device *device;
64 WARN_ON(fs_devices->opened);
65 while (!list_empty(&fs_devices->devices)) {
66 device = list_entry(fs_devices->devices.next,
67 struct btrfs_device, dev_list);
68 list_del(&device->dev_list);
69 rcu_string_free(device->name);
70 kfree(device);
71 }
72 kfree(fs_devices);
73 }
74
75 static void btrfs_kobject_uevent(struct block_device *bdev,
76 enum kobject_action action)
77 {
78 int ret;
79
80 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
81 if (ret)
82 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
83 action,
84 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
85 &disk_to_dev(bdev->bd_disk)->kobj);
86 }
87
88 void btrfs_cleanup_fs_uuids(void)
89 {
90 struct btrfs_fs_devices *fs_devices;
91
92 while (!list_empty(&fs_uuids)) {
93 fs_devices = list_entry(fs_uuids.next,
94 struct btrfs_fs_devices, list);
95 list_del(&fs_devices->list);
96 free_fs_devices(fs_devices);
97 }
98 }
99
100 static noinline struct btrfs_device *__find_device(struct list_head *head,
101 u64 devid, u8 *uuid)
102 {
103 struct btrfs_device *dev;
104
105 list_for_each_entry(dev, head, dev_list) {
106 if (dev->devid == devid &&
107 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
108 return dev;
109 }
110 }
111 return NULL;
112 }
113
114 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 {
116 struct btrfs_fs_devices *fs_devices;
117
118 list_for_each_entry(fs_devices, &fs_uuids, list) {
119 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
120 return fs_devices;
121 }
122 return NULL;
123 }
124
125 static int
126 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
127 int flush, struct block_device **bdev,
128 struct buffer_head **bh)
129 {
130 int ret;
131
132 *bdev = blkdev_get_by_path(device_path, flags, holder);
133
134 if (IS_ERR(*bdev)) {
135 ret = PTR_ERR(*bdev);
136 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
137 goto error;
138 }
139
140 if (flush)
141 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
142 ret = set_blocksize(*bdev, 4096);
143 if (ret) {
144 blkdev_put(*bdev, flags);
145 goto error;
146 }
147 invalidate_bdev(*bdev);
148 *bh = btrfs_read_dev_super(*bdev);
149 if (!*bh) {
150 ret = -EINVAL;
151 blkdev_put(*bdev, flags);
152 goto error;
153 }
154
155 return 0;
156
157 error:
158 *bdev = NULL;
159 *bh = NULL;
160 return ret;
161 }
162
163 static void requeue_list(struct btrfs_pending_bios *pending_bios,
164 struct bio *head, struct bio *tail)
165 {
166
167 struct bio *old_head;
168
169 old_head = pending_bios->head;
170 pending_bios->head = head;
171 if (pending_bios->tail)
172 tail->bi_next = old_head;
173 else
174 pending_bios->tail = tail;
175 }
176
177 /*
178 * we try to collect pending bios for a device so we don't get a large
179 * number of procs sending bios down to the same device. This greatly
180 * improves the schedulers ability to collect and merge the bios.
181 *
182 * But, it also turns into a long list of bios to process and that is sure
183 * to eventually make the worker thread block. The solution here is to
184 * make some progress and then put this work struct back at the end of
185 * the list if the block device is congested. This way, multiple devices
186 * can make progress from a single worker thread.
187 */
188 static noinline void run_scheduled_bios(struct btrfs_device *device)
189 {
190 struct bio *pending;
191 struct backing_dev_info *bdi;
192 struct btrfs_fs_info *fs_info;
193 struct btrfs_pending_bios *pending_bios;
194 struct bio *tail;
195 struct bio *cur;
196 int again = 0;
197 unsigned long num_run;
198 unsigned long batch_run = 0;
199 unsigned long limit;
200 unsigned long last_waited = 0;
201 int force_reg = 0;
202 int sync_pending = 0;
203 struct blk_plug plug;
204
205 /*
206 * this function runs all the bios we've collected for
207 * a particular device. We don't want to wander off to
208 * another device without first sending all of these down.
209 * So, setup a plug here and finish it off before we return
210 */
211 blk_start_plug(&plug);
212
213 bdi = blk_get_backing_dev_info(device->bdev);
214 fs_info = device->dev_root->fs_info;
215 limit = btrfs_async_submit_limit(fs_info);
216 limit = limit * 2 / 3;
217
218 loop:
219 spin_lock(&device->io_lock);
220
221 loop_lock:
222 num_run = 0;
223
224 /* take all the bios off the list at once and process them
225 * later on (without the lock held). But, remember the
226 * tail and other pointers so the bios can be properly reinserted
227 * into the list if we hit congestion
228 */
229 if (!force_reg && device->pending_sync_bios.head) {
230 pending_bios = &device->pending_sync_bios;
231 force_reg = 1;
232 } else {
233 pending_bios = &device->pending_bios;
234 force_reg = 0;
235 }
236
237 pending = pending_bios->head;
238 tail = pending_bios->tail;
239 WARN_ON(pending && !tail);
240
241 /*
242 * if pending was null this time around, no bios need processing
243 * at all and we can stop. Otherwise it'll loop back up again
244 * and do an additional check so no bios are missed.
245 *
246 * device->running_pending is used to synchronize with the
247 * schedule_bio code.
248 */
249 if (device->pending_sync_bios.head == NULL &&
250 device->pending_bios.head == NULL) {
251 again = 0;
252 device->running_pending = 0;
253 } else {
254 again = 1;
255 device->running_pending = 1;
256 }
257
258 pending_bios->head = NULL;
259 pending_bios->tail = NULL;
260
261 spin_unlock(&device->io_lock);
262
263 while (pending) {
264
265 rmb();
266 /* we want to work on both lists, but do more bios on the
267 * sync list than the regular list
268 */
269 if ((num_run > 32 &&
270 pending_bios != &device->pending_sync_bios &&
271 device->pending_sync_bios.head) ||
272 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
273 device->pending_bios.head)) {
274 spin_lock(&device->io_lock);
275 requeue_list(pending_bios, pending, tail);
276 goto loop_lock;
277 }
278
279 cur = pending;
280 pending = pending->bi_next;
281 cur->bi_next = NULL;
282
283 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
284 waitqueue_active(&fs_info->async_submit_wait))
285 wake_up(&fs_info->async_submit_wait);
286
287 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
288
289 /*
290 * if we're doing the sync list, record that our
291 * plug has some sync requests on it
292 *
293 * If we're doing the regular list and there are
294 * sync requests sitting around, unplug before
295 * we add more
296 */
297 if (pending_bios == &device->pending_sync_bios) {
298 sync_pending = 1;
299 } else if (sync_pending) {
300 blk_finish_plug(&plug);
301 blk_start_plug(&plug);
302 sync_pending = 0;
303 }
304
305 btrfsic_submit_bio(cur->bi_rw, cur);
306 num_run++;
307 batch_run++;
308 if (need_resched())
309 cond_resched();
310
311 /*
312 * we made progress, there is more work to do and the bdi
313 * is now congested. Back off and let other work structs
314 * run instead
315 */
316 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
317 fs_info->fs_devices->open_devices > 1) {
318 struct io_context *ioc;
319
320 ioc = current->io_context;
321
322 /*
323 * the main goal here is that we don't want to
324 * block if we're going to be able to submit
325 * more requests without blocking.
326 *
327 * This code does two great things, it pokes into
328 * the elevator code from a filesystem _and_
329 * it makes assumptions about how batching works.
330 */
331 if (ioc && ioc->nr_batch_requests > 0 &&
332 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
333 (last_waited == 0 ||
334 ioc->last_waited == last_waited)) {
335 /*
336 * we want to go through our batch of
337 * requests and stop. So, we copy out
338 * the ioc->last_waited time and test
339 * against it before looping
340 */
341 last_waited = ioc->last_waited;
342 if (need_resched())
343 cond_resched();
344 continue;
345 }
346 spin_lock(&device->io_lock);
347 requeue_list(pending_bios, pending, tail);
348 device->running_pending = 1;
349
350 spin_unlock(&device->io_lock);
351 btrfs_requeue_work(&device->work);
352 goto done;
353 }
354 /* unplug every 64 requests just for good measure */
355 if (batch_run % 64 == 0) {
356 blk_finish_plug(&plug);
357 blk_start_plug(&plug);
358 sync_pending = 0;
359 }
360 }
361
362 cond_resched();
363 if (again)
364 goto loop;
365
366 spin_lock(&device->io_lock);
367 if (device->pending_bios.head || device->pending_sync_bios.head)
368 goto loop_lock;
369 spin_unlock(&device->io_lock);
370
371 done:
372 blk_finish_plug(&plug);
373 }
374
375 static void pending_bios_fn(struct btrfs_work *work)
376 {
377 struct btrfs_device *device;
378
379 device = container_of(work, struct btrfs_device, work);
380 run_scheduled_bios(device);
381 }
382
383 static noinline int device_list_add(const char *path,
384 struct btrfs_super_block *disk_super,
385 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
386 {
387 struct btrfs_device *device;
388 struct btrfs_fs_devices *fs_devices;
389 struct rcu_string *name;
390 u64 found_transid = btrfs_super_generation(disk_super);
391
392 fs_devices = find_fsid(disk_super->fsid);
393 if (!fs_devices) {
394 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
395 if (!fs_devices)
396 return -ENOMEM;
397 INIT_LIST_HEAD(&fs_devices->devices);
398 INIT_LIST_HEAD(&fs_devices->alloc_list);
399 list_add(&fs_devices->list, &fs_uuids);
400 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
401 fs_devices->latest_devid = devid;
402 fs_devices->latest_trans = found_transid;
403 mutex_init(&fs_devices->device_list_mutex);
404 device = NULL;
405 } else {
406 device = __find_device(&fs_devices->devices, devid,
407 disk_super->dev_item.uuid);
408 }
409 if (!device) {
410 if (fs_devices->opened)
411 return -EBUSY;
412
413 device = kzalloc(sizeof(*device), GFP_NOFS);
414 if (!device) {
415 /* we can safely leave the fs_devices entry around */
416 return -ENOMEM;
417 }
418 device->devid = devid;
419 device->dev_stats_valid = 0;
420 device->work.func = pending_bios_fn;
421 memcpy(device->uuid, disk_super->dev_item.uuid,
422 BTRFS_UUID_SIZE);
423 spin_lock_init(&device->io_lock);
424
425 name = rcu_string_strdup(path, GFP_NOFS);
426 if (!name) {
427 kfree(device);
428 return -ENOMEM;
429 }
430 rcu_assign_pointer(device->name, name);
431 INIT_LIST_HEAD(&device->dev_alloc_list);
432
433 /* init readahead state */
434 spin_lock_init(&device->reada_lock);
435 device->reada_curr_zone = NULL;
436 atomic_set(&device->reada_in_flight, 0);
437 device->reada_next = 0;
438 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
439 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
440
441 mutex_lock(&fs_devices->device_list_mutex);
442 list_add_rcu(&device->dev_list, &fs_devices->devices);
443 mutex_unlock(&fs_devices->device_list_mutex);
444
445 device->fs_devices = fs_devices;
446 fs_devices->num_devices++;
447 } else if (!device->name || strcmp(device->name->str, path)) {
448 name = rcu_string_strdup(path, GFP_NOFS);
449 if (!name)
450 return -ENOMEM;
451 rcu_string_free(device->name);
452 rcu_assign_pointer(device->name, name);
453 if (device->missing) {
454 fs_devices->missing_devices--;
455 device->missing = 0;
456 }
457 }
458
459 if (found_transid > fs_devices->latest_trans) {
460 fs_devices->latest_devid = devid;
461 fs_devices->latest_trans = found_transid;
462 }
463 *fs_devices_ret = fs_devices;
464 return 0;
465 }
466
467 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
468 {
469 struct btrfs_fs_devices *fs_devices;
470 struct btrfs_device *device;
471 struct btrfs_device *orig_dev;
472
473 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
474 if (!fs_devices)
475 return ERR_PTR(-ENOMEM);
476
477 INIT_LIST_HEAD(&fs_devices->devices);
478 INIT_LIST_HEAD(&fs_devices->alloc_list);
479 INIT_LIST_HEAD(&fs_devices->list);
480 mutex_init(&fs_devices->device_list_mutex);
481 fs_devices->latest_devid = orig->latest_devid;
482 fs_devices->latest_trans = orig->latest_trans;
483 fs_devices->total_devices = orig->total_devices;
484 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
485
486 /* We have held the volume lock, it is safe to get the devices. */
487 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
488 struct rcu_string *name;
489
490 device = kzalloc(sizeof(*device), GFP_NOFS);
491 if (!device)
492 goto error;
493
494 /*
495 * This is ok to do without rcu read locked because we hold the
496 * uuid mutex so nothing we touch in here is going to disappear.
497 */
498 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
499 if (!name) {
500 kfree(device);
501 goto error;
502 }
503 rcu_assign_pointer(device->name, name);
504
505 device->devid = orig_dev->devid;
506 device->work.func = pending_bios_fn;
507 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
508 spin_lock_init(&device->io_lock);
509 INIT_LIST_HEAD(&device->dev_list);
510 INIT_LIST_HEAD(&device->dev_alloc_list);
511
512 list_add(&device->dev_list, &fs_devices->devices);
513 device->fs_devices = fs_devices;
514 fs_devices->num_devices++;
515 }
516 return fs_devices;
517 error:
518 free_fs_devices(fs_devices);
519 return ERR_PTR(-ENOMEM);
520 }
521
522 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
523 struct btrfs_fs_devices *fs_devices, int step)
524 {
525 struct btrfs_device *device, *next;
526
527 struct block_device *latest_bdev = NULL;
528 u64 latest_devid = 0;
529 u64 latest_transid = 0;
530
531 mutex_lock(&uuid_mutex);
532 again:
533 /* This is the initialized path, it is safe to release the devices. */
534 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
535 if (device->in_fs_metadata) {
536 if (!device->is_tgtdev_for_dev_replace &&
537 (!latest_transid ||
538 device->generation > latest_transid)) {
539 latest_devid = device->devid;
540 latest_transid = device->generation;
541 latest_bdev = device->bdev;
542 }
543 continue;
544 }
545
546 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
547 /*
548 * In the first step, keep the device which has
549 * the correct fsid and the devid that is used
550 * for the dev_replace procedure.
551 * In the second step, the dev_replace state is
552 * read from the device tree and it is known
553 * whether the procedure is really active or
554 * not, which means whether this device is
555 * used or whether it should be removed.
556 */
557 if (step == 0 || device->is_tgtdev_for_dev_replace) {
558 continue;
559 }
560 }
561 if (device->bdev) {
562 blkdev_put(device->bdev, device->mode);
563 device->bdev = NULL;
564 fs_devices->open_devices--;
565 }
566 if (device->writeable) {
567 list_del_init(&device->dev_alloc_list);
568 device->writeable = 0;
569 if (!device->is_tgtdev_for_dev_replace)
570 fs_devices->rw_devices--;
571 }
572 list_del_init(&device->dev_list);
573 fs_devices->num_devices--;
574 rcu_string_free(device->name);
575 kfree(device);
576 }
577
578 if (fs_devices->seed) {
579 fs_devices = fs_devices->seed;
580 goto again;
581 }
582
583 fs_devices->latest_bdev = latest_bdev;
584 fs_devices->latest_devid = latest_devid;
585 fs_devices->latest_trans = latest_transid;
586
587 mutex_unlock(&uuid_mutex);
588 }
589
590 static void __free_device(struct work_struct *work)
591 {
592 struct btrfs_device *device;
593
594 device = container_of(work, struct btrfs_device, rcu_work);
595
596 if (device->bdev)
597 blkdev_put(device->bdev, device->mode);
598
599 rcu_string_free(device->name);
600 kfree(device);
601 }
602
603 static void free_device(struct rcu_head *head)
604 {
605 struct btrfs_device *device;
606
607 device = container_of(head, struct btrfs_device, rcu);
608
609 INIT_WORK(&device->rcu_work, __free_device);
610 schedule_work(&device->rcu_work);
611 }
612
613 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
614 {
615 struct btrfs_device *device;
616
617 if (--fs_devices->opened > 0)
618 return 0;
619
620 mutex_lock(&fs_devices->device_list_mutex);
621 list_for_each_entry(device, &fs_devices->devices, dev_list) {
622 struct btrfs_device *new_device;
623 struct rcu_string *name;
624
625 if (device->bdev)
626 fs_devices->open_devices--;
627
628 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
629 list_del_init(&device->dev_alloc_list);
630 fs_devices->rw_devices--;
631 }
632
633 if (device->can_discard)
634 fs_devices->num_can_discard--;
635
636 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
637 BUG_ON(!new_device); /* -ENOMEM */
638 memcpy(new_device, device, sizeof(*new_device));
639
640 /* Safe because we are under uuid_mutex */
641 if (device->name) {
642 name = rcu_string_strdup(device->name->str, GFP_NOFS);
643 BUG_ON(device->name && !name); /* -ENOMEM */
644 rcu_assign_pointer(new_device->name, name);
645 }
646 new_device->bdev = NULL;
647 new_device->writeable = 0;
648 new_device->in_fs_metadata = 0;
649 new_device->can_discard = 0;
650 list_replace_rcu(&device->dev_list, &new_device->dev_list);
651
652 call_rcu(&device->rcu, free_device);
653 }
654 mutex_unlock(&fs_devices->device_list_mutex);
655
656 WARN_ON(fs_devices->open_devices);
657 WARN_ON(fs_devices->rw_devices);
658 fs_devices->opened = 0;
659 fs_devices->seeding = 0;
660
661 return 0;
662 }
663
664 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
665 {
666 struct btrfs_fs_devices *seed_devices = NULL;
667 int ret;
668
669 mutex_lock(&uuid_mutex);
670 ret = __btrfs_close_devices(fs_devices);
671 if (!fs_devices->opened) {
672 seed_devices = fs_devices->seed;
673 fs_devices->seed = NULL;
674 }
675 mutex_unlock(&uuid_mutex);
676
677 while (seed_devices) {
678 fs_devices = seed_devices;
679 seed_devices = fs_devices->seed;
680 __btrfs_close_devices(fs_devices);
681 free_fs_devices(fs_devices);
682 }
683 return ret;
684 }
685
686 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
687 fmode_t flags, void *holder)
688 {
689 struct request_queue *q;
690 struct block_device *bdev;
691 struct list_head *head = &fs_devices->devices;
692 struct btrfs_device *device;
693 struct block_device *latest_bdev = NULL;
694 struct buffer_head *bh;
695 struct btrfs_super_block *disk_super;
696 u64 latest_devid = 0;
697 u64 latest_transid = 0;
698 u64 devid;
699 int seeding = 1;
700 int ret = 0;
701
702 flags |= FMODE_EXCL;
703
704 list_for_each_entry(device, head, dev_list) {
705 if (device->bdev)
706 continue;
707 if (!device->name)
708 continue;
709
710 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
711 &bdev, &bh);
712 if (ret)
713 continue;
714
715 disk_super = (struct btrfs_super_block *)bh->b_data;
716 devid = btrfs_stack_device_id(&disk_super->dev_item);
717 if (devid != device->devid)
718 goto error_brelse;
719
720 if (memcmp(device->uuid, disk_super->dev_item.uuid,
721 BTRFS_UUID_SIZE))
722 goto error_brelse;
723
724 device->generation = btrfs_super_generation(disk_super);
725 if (!latest_transid || device->generation > latest_transid) {
726 latest_devid = devid;
727 latest_transid = device->generation;
728 latest_bdev = bdev;
729 }
730
731 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
732 device->writeable = 0;
733 } else {
734 device->writeable = !bdev_read_only(bdev);
735 seeding = 0;
736 }
737
738 q = bdev_get_queue(bdev);
739 if (blk_queue_discard(q)) {
740 device->can_discard = 1;
741 fs_devices->num_can_discard++;
742 }
743
744 device->bdev = bdev;
745 device->in_fs_metadata = 0;
746 device->mode = flags;
747
748 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
749 fs_devices->rotating = 1;
750
751 fs_devices->open_devices++;
752 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
753 fs_devices->rw_devices++;
754 list_add(&device->dev_alloc_list,
755 &fs_devices->alloc_list);
756 }
757 brelse(bh);
758 continue;
759
760 error_brelse:
761 brelse(bh);
762 blkdev_put(bdev, flags);
763 continue;
764 }
765 if (fs_devices->open_devices == 0) {
766 ret = -EINVAL;
767 goto out;
768 }
769 fs_devices->seeding = seeding;
770 fs_devices->opened = 1;
771 fs_devices->latest_bdev = latest_bdev;
772 fs_devices->latest_devid = latest_devid;
773 fs_devices->latest_trans = latest_transid;
774 fs_devices->total_rw_bytes = 0;
775 out:
776 return ret;
777 }
778
779 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
780 fmode_t flags, void *holder)
781 {
782 int ret;
783
784 mutex_lock(&uuid_mutex);
785 if (fs_devices->opened) {
786 fs_devices->opened++;
787 ret = 0;
788 } else {
789 ret = __btrfs_open_devices(fs_devices, flags, holder);
790 }
791 mutex_unlock(&uuid_mutex);
792 return ret;
793 }
794
795 /*
796 * Look for a btrfs signature on a device. This may be called out of the mount path
797 * and we are not allowed to call set_blocksize during the scan. The superblock
798 * is read via pagecache
799 */
800 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
801 struct btrfs_fs_devices **fs_devices_ret)
802 {
803 struct btrfs_super_block *disk_super;
804 struct block_device *bdev;
805 struct page *page;
806 void *p;
807 int ret = -EINVAL;
808 u64 devid;
809 u64 transid;
810 u64 total_devices;
811 u64 bytenr;
812 pgoff_t index;
813
814 /*
815 * we would like to check all the supers, but that would make
816 * a btrfs mount succeed after a mkfs from a different FS.
817 * So, we need to add a special mount option to scan for
818 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
819 */
820 bytenr = btrfs_sb_offset(0);
821 flags |= FMODE_EXCL;
822 mutex_lock(&uuid_mutex);
823
824 bdev = blkdev_get_by_path(path, flags, holder);
825
826 if (IS_ERR(bdev)) {
827 ret = PTR_ERR(bdev);
828 printk(KERN_INFO "btrfs: open %s failed\n", path);
829 goto error;
830 }
831
832 /* make sure our super fits in the device */
833 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
834 goto error_bdev_put;
835
836 /* make sure our super fits in the page */
837 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
838 goto error_bdev_put;
839
840 /* make sure our super doesn't straddle pages on disk */
841 index = bytenr >> PAGE_CACHE_SHIFT;
842 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
843 goto error_bdev_put;
844
845 /* pull in the page with our super */
846 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
847 index, GFP_NOFS);
848
849 if (IS_ERR_OR_NULL(page))
850 goto error_bdev_put;
851
852 p = kmap(page);
853
854 /* align our pointer to the offset of the super block */
855 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
856
857 if (btrfs_super_bytenr(disk_super) != bytenr ||
858 strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
859 sizeof(disk_super->magic)))
860 goto error_unmap;
861
862 devid = btrfs_stack_device_id(&disk_super->dev_item);
863 transid = btrfs_super_generation(disk_super);
864 total_devices = btrfs_super_num_devices(disk_super);
865
866 if (disk_super->label[0]) {
867 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
868 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
869 printk(KERN_INFO "device label %s ", disk_super->label);
870 } else {
871 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
872 }
873
874 printk(KERN_CONT "devid %llu transid %llu %s\n",
875 (unsigned long long)devid, (unsigned long long)transid, path);
876
877 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
878 if (!ret && fs_devices_ret)
879 (*fs_devices_ret)->total_devices = total_devices;
880
881 error_unmap:
882 kunmap(page);
883 page_cache_release(page);
884
885 error_bdev_put:
886 blkdev_put(bdev, flags);
887 error:
888 mutex_unlock(&uuid_mutex);
889 return ret;
890 }
891
892 /* helper to account the used device space in the range */
893 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
894 u64 end, u64 *length)
895 {
896 struct btrfs_key key;
897 struct btrfs_root *root = device->dev_root;
898 struct btrfs_dev_extent *dev_extent;
899 struct btrfs_path *path;
900 u64 extent_end;
901 int ret;
902 int slot;
903 struct extent_buffer *l;
904
905 *length = 0;
906
907 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
908 return 0;
909
910 path = btrfs_alloc_path();
911 if (!path)
912 return -ENOMEM;
913 path->reada = 2;
914
915 key.objectid = device->devid;
916 key.offset = start;
917 key.type = BTRFS_DEV_EXTENT_KEY;
918
919 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
920 if (ret < 0)
921 goto out;
922 if (ret > 0) {
923 ret = btrfs_previous_item(root, path, key.objectid, key.type);
924 if (ret < 0)
925 goto out;
926 }
927
928 while (1) {
929 l = path->nodes[0];
930 slot = path->slots[0];
931 if (slot >= btrfs_header_nritems(l)) {
932 ret = btrfs_next_leaf(root, path);
933 if (ret == 0)
934 continue;
935 if (ret < 0)
936 goto out;
937
938 break;
939 }
940 btrfs_item_key_to_cpu(l, &key, slot);
941
942 if (key.objectid < device->devid)
943 goto next;
944
945 if (key.objectid > device->devid)
946 break;
947
948 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
949 goto next;
950
951 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
952 extent_end = key.offset + btrfs_dev_extent_length(l,
953 dev_extent);
954 if (key.offset <= start && extent_end > end) {
955 *length = end - start + 1;
956 break;
957 } else if (key.offset <= start && extent_end > start)
958 *length += extent_end - start;
959 else if (key.offset > start && extent_end <= end)
960 *length += extent_end - key.offset;
961 else if (key.offset > start && key.offset <= end) {
962 *length += end - key.offset + 1;
963 break;
964 } else if (key.offset > end)
965 break;
966
967 next:
968 path->slots[0]++;
969 }
970 ret = 0;
971 out:
972 btrfs_free_path(path);
973 return ret;
974 }
975
976 /*
977 * find_free_dev_extent - find free space in the specified device
978 * @device: the device which we search the free space in
979 * @num_bytes: the size of the free space that we need
980 * @start: store the start of the free space.
981 * @len: the size of the free space. that we find, or the size of the max
982 * free space if we don't find suitable free space
983 *
984 * this uses a pretty simple search, the expectation is that it is
985 * called very infrequently and that a given device has a small number
986 * of extents
987 *
988 * @start is used to store the start of the free space if we find. But if we
989 * don't find suitable free space, it will be used to store the start position
990 * of the max free space.
991 *
992 * @len is used to store the size of the free space that we find.
993 * But if we don't find suitable free space, it is used to store the size of
994 * the max free space.
995 */
996 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
997 u64 *start, u64 *len)
998 {
999 struct btrfs_key key;
1000 struct btrfs_root *root = device->dev_root;
1001 struct btrfs_dev_extent *dev_extent;
1002 struct btrfs_path *path;
1003 u64 hole_size;
1004 u64 max_hole_start;
1005 u64 max_hole_size;
1006 u64 extent_end;
1007 u64 search_start;
1008 u64 search_end = device->total_bytes;
1009 int ret;
1010 int slot;
1011 struct extent_buffer *l;
1012
1013 /* FIXME use last free of some kind */
1014
1015 /* we don't want to overwrite the superblock on the drive,
1016 * so we make sure to start at an offset of at least 1MB
1017 */
1018 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1019
1020 max_hole_start = search_start;
1021 max_hole_size = 0;
1022 hole_size = 0;
1023
1024 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1025 ret = -ENOSPC;
1026 goto error;
1027 }
1028
1029 path = btrfs_alloc_path();
1030 if (!path) {
1031 ret = -ENOMEM;
1032 goto error;
1033 }
1034 path->reada = 2;
1035
1036 key.objectid = device->devid;
1037 key.offset = search_start;
1038 key.type = BTRFS_DEV_EXTENT_KEY;
1039
1040 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1041 if (ret < 0)
1042 goto out;
1043 if (ret > 0) {
1044 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1045 if (ret < 0)
1046 goto out;
1047 }
1048
1049 while (1) {
1050 l = path->nodes[0];
1051 slot = path->slots[0];
1052 if (slot >= btrfs_header_nritems(l)) {
1053 ret = btrfs_next_leaf(root, path);
1054 if (ret == 0)
1055 continue;
1056 if (ret < 0)
1057 goto out;
1058
1059 break;
1060 }
1061 btrfs_item_key_to_cpu(l, &key, slot);
1062
1063 if (key.objectid < device->devid)
1064 goto next;
1065
1066 if (key.objectid > device->devid)
1067 break;
1068
1069 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1070 goto next;
1071
1072 if (key.offset > search_start) {
1073 hole_size = key.offset - search_start;
1074
1075 if (hole_size > max_hole_size) {
1076 max_hole_start = search_start;
1077 max_hole_size = hole_size;
1078 }
1079
1080 /*
1081 * If this free space is greater than which we need,
1082 * it must be the max free space that we have found
1083 * until now, so max_hole_start must point to the start
1084 * of this free space and the length of this free space
1085 * is stored in max_hole_size. Thus, we return
1086 * max_hole_start and max_hole_size and go back to the
1087 * caller.
1088 */
1089 if (hole_size >= num_bytes) {
1090 ret = 0;
1091 goto out;
1092 }
1093 }
1094
1095 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1096 extent_end = key.offset + btrfs_dev_extent_length(l,
1097 dev_extent);
1098 if (extent_end > search_start)
1099 search_start = extent_end;
1100 next:
1101 path->slots[0]++;
1102 cond_resched();
1103 }
1104
1105 /*
1106 * At this point, search_start should be the end of
1107 * allocated dev extents, and when shrinking the device,
1108 * search_end may be smaller than search_start.
1109 */
1110 if (search_end > search_start)
1111 hole_size = search_end - search_start;
1112
1113 if (hole_size > max_hole_size) {
1114 max_hole_start = search_start;
1115 max_hole_size = hole_size;
1116 }
1117
1118 /* See above. */
1119 if (hole_size < num_bytes)
1120 ret = -ENOSPC;
1121 else
1122 ret = 0;
1123
1124 out:
1125 btrfs_free_path(path);
1126 error:
1127 *start = max_hole_start;
1128 if (len)
1129 *len = max_hole_size;
1130 return ret;
1131 }
1132
1133 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1134 struct btrfs_device *device,
1135 u64 start)
1136 {
1137 int ret;
1138 struct btrfs_path *path;
1139 struct btrfs_root *root = device->dev_root;
1140 struct btrfs_key key;
1141 struct btrfs_key found_key;
1142 struct extent_buffer *leaf = NULL;
1143 struct btrfs_dev_extent *extent = NULL;
1144
1145 path = btrfs_alloc_path();
1146 if (!path)
1147 return -ENOMEM;
1148
1149 key.objectid = device->devid;
1150 key.offset = start;
1151 key.type = BTRFS_DEV_EXTENT_KEY;
1152 again:
1153 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1154 if (ret > 0) {
1155 ret = btrfs_previous_item(root, path, key.objectid,
1156 BTRFS_DEV_EXTENT_KEY);
1157 if (ret)
1158 goto out;
1159 leaf = path->nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1161 extent = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_dev_extent);
1163 BUG_ON(found_key.offset > start || found_key.offset +
1164 btrfs_dev_extent_length(leaf, extent) < start);
1165 key = found_key;
1166 btrfs_release_path(path);
1167 goto again;
1168 } else if (ret == 0) {
1169 leaf = path->nodes[0];
1170 extent = btrfs_item_ptr(leaf, path->slots[0],
1171 struct btrfs_dev_extent);
1172 } else {
1173 btrfs_error(root->fs_info, ret, "Slot search failed");
1174 goto out;
1175 }
1176
1177 if (device->bytes_used > 0) {
1178 u64 len = btrfs_dev_extent_length(leaf, extent);
1179 device->bytes_used -= len;
1180 spin_lock(&root->fs_info->free_chunk_lock);
1181 root->fs_info->free_chunk_space += len;
1182 spin_unlock(&root->fs_info->free_chunk_lock);
1183 }
1184 ret = btrfs_del_item(trans, root, path);
1185 if (ret) {
1186 btrfs_error(root->fs_info, ret,
1187 "Failed to remove dev extent item");
1188 }
1189 out:
1190 btrfs_free_path(path);
1191 return ret;
1192 }
1193
1194 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1195 struct btrfs_device *device,
1196 u64 chunk_tree, u64 chunk_objectid,
1197 u64 chunk_offset, u64 start, u64 num_bytes)
1198 {
1199 int ret;
1200 struct btrfs_path *path;
1201 struct btrfs_root *root = device->dev_root;
1202 struct btrfs_dev_extent *extent;
1203 struct extent_buffer *leaf;
1204 struct btrfs_key key;
1205
1206 WARN_ON(!device->in_fs_metadata);
1207 WARN_ON(device->is_tgtdev_for_dev_replace);
1208 path = btrfs_alloc_path();
1209 if (!path)
1210 return -ENOMEM;
1211
1212 key.objectid = device->devid;
1213 key.offset = start;
1214 key.type = BTRFS_DEV_EXTENT_KEY;
1215 ret = btrfs_insert_empty_item(trans, root, path, &key,
1216 sizeof(*extent));
1217 if (ret)
1218 goto out;
1219
1220 leaf = path->nodes[0];
1221 extent = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_dev_extent);
1223 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1224 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1225 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1226
1227 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1228 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1229 BTRFS_UUID_SIZE);
1230
1231 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1232 btrfs_mark_buffer_dirty(leaf);
1233 out:
1234 btrfs_free_path(path);
1235 return ret;
1236 }
1237
1238 static noinline int find_next_chunk(struct btrfs_root *root,
1239 u64 objectid, u64 *offset)
1240 {
1241 struct btrfs_path *path;
1242 int ret;
1243 struct btrfs_key key;
1244 struct btrfs_chunk *chunk;
1245 struct btrfs_key found_key;
1246
1247 path = btrfs_alloc_path();
1248 if (!path)
1249 return -ENOMEM;
1250
1251 key.objectid = objectid;
1252 key.offset = (u64)-1;
1253 key.type = BTRFS_CHUNK_ITEM_KEY;
1254
1255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1256 if (ret < 0)
1257 goto error;
1258
1259 BUG_ON(ret == 0); /* Corruption */
1260
1261 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1262 if (ret) {
1263 *offset = 0;
1264 } else {
1265 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1266 path->slots[0]);
1267 if (found_key.objectid != objectid)
1268 *offset = 0;
1269 else {
1270 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1271 struct btrfs_chunk);
1272 *offset = found_key.offset +
1273 btrfs_chunk_length(path->nodes[0], chunk);
1274 }
1275 }
1276 ret = 0;
1277 error:
1278 btrfs_free_path(path);
1279 return ret;
1280 }
1281
1282 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1283 {
1284 int ret;
1285 struct btrfs_key key;
1286 struct btrfs_key found_key;
1287 struct btrfs_path *path;
1288
1289 root = root->fs_info->chunk_root;
1290
1291 path = btrfs_alloc_path();
1292 if (!path)
1293 return -ENOMEM;
1294
1295 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1296 key.type = BTRFS_DEV_ITEM_KEY;
1297 key.offset = (u64)-1;
1298
1299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1300 if (ret < 0)
1301 goto error;
1302
1303 BUG_ON(ret == 0); /* Corruption */
1304
1305 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1306 BTRFS_DEV_ITEM_KEY);
1307 if (ret) {
1308 *objectid = 1;
1309 } else {
1310 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1311 path->slots[0]);
1312 *objectid = found_key.offset + 1;
1313 }
1314 ret = 0;
1315 error:
1316 btrfs_free_path(path);
1317 return ret;
1318 }
1319
1320 /*
1321 * the device information is stored in the chunk root
1322 * the btrfs_device struct should be fully filled in
1323 */
1324 int btrfs_add_device(struct btrfs_trans_handle *trans,
1325 struct btrfs_root *root,
1326 struct btrfs_device *device)
1327 {
1328 int ret;
1329 struct btrfs_path *path;
1330 struct btrfs_dev_item *dev_item;
1331 struct extent_buffer *leaf;
1332 struct btrfs_key key;
1333 unsigned long ptr;
1334
1335 root = root->fs_info->chunk_root;
1336
1337 path = btrfs_alloc_path();
1338 if (!path)
1339 return -ENOMEM;
1340
1341 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1342 key.type = BTRFS_DEV_ITEM_KEY;
1343 key.offset = device->devid;
1344
1345 ret = btrfs_insert_empty_item(trans, root, path, &key,
1346 sizeof(*dev_item));
1347 if (ret)
1348 goto out;
1349
1350 leaf = path->nodes[0];
1351 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1352
1353 btrfs_set_device_id(leaf, dev_item, device->devid);
1354 btrfs_set_device_generation(leaf, dev_item, 0);
1355 btrfs_set_device_type(leaf, dev_item, device->type);
1356 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1357 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1358 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1359 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1360 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1361 btrfs_set_device_group(leaf, dev_item, 0);
1362 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1363 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1364 btrfs_set_device_start_offset(leaf, dev_item, 0);
1365
1366 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1367 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1368 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1369 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1370 btrfs_mark_buffer_dirty(leaf);
1371
1372 ret = 0;
1373 out:
1374 btrfs_free_path(path);
1375 return ret;
1376 }
1377
1378 static int btrfs_rm_dev_item(struct btrfs_root *root,
1379 struct btrfs_device *device)
1380 {
1381 int ret;
1382 struct btrfs_path *path;
1383 struct btrfs_key key;
1384 struct btrfs_trans_handle *trans;
1385
1386 root = root->fs_info->chunk_root;
1387
1388 path = btrfs_alloc_path();
1389 if (!path)
1390 return -ENOMEM;
1391
1392 trans = btrfs_start_transaction(root, 0);
1393 if (IS_ERR(trans)) {
1394 btrfs_free_path(path);
1395 return PTR_ERR(trans);
1396 }
1397 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1398 key.type = BTRFS_DEV_ITEM_KEY;
1399 key.offset = device->devid;
1400 lock_chunks(root);
1401
1402 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 if (ret < 0)
1404 goto out;
1405
1406 if (ret > 0) {
1407 ret = -ENOENT;
1408 goto out;
1409 }
1410
1411 ret = btrfs_del_item(trans, root, path);
1412 if (ret)
1413 goto out;
1414 out:
1415 btrfs_free_path(path);
1416 unlock_chunks(root);
1417 btrfs_commit_transaction(trans, root);
1418 return ret;
1419 }
1420
1421 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1422 {
1423 struct btrfs_device *device;
1424 struct btrfs_device *next_device;
1425 struct block_device *bdev;
1426 struct buffer_head *bh = NULL;
1427 struct btrfs_super_block *disk_super;
1428 struct btrfs_fs_devices *cur_devices;
1429 u64 all_avail;
1430 u64 devid;
1431 u64 num_devices;
1432 u8 *dev_uuid;
1433 unsigned seq;
1434 int ret = 0;
1435 bool clear_super = false;
1436
1437 mutex_lock(&uuid_mutex);
1438
1439 do {
1440 seq = read_seqbegin(&root->fs_info->profiles_lock);
1441
1442 all_avail = root->fs_info->avail_data_alloc_bits |
1443 root->fs_info->avail_system_alloc_bits |
1444 root->fs_info->avail_metadata_alloc_bits;
1445 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1446
1447 num_devices = root->fs_info->fs_devices->num_devices;
1448 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1449 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1450 WARN_ON(num_devices < 1);
1451 num_devices--;
1452 }
1453 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1454
1455 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1456 printk(KERN_ERR "btrfs: unable to go below four devices "
1457 "on raid10\n");
1458 ret = -EINVAL;
1459 goto out;
1460 }
1461
1462 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1463 printk(KERN_ERR "btrfs: unable to go below two "
1464 "devices on raid1\n");
1465 ret = -EINVAL;
1466 goto out;
1467 }
1468
1469 if (strcmp(device_path, "missing") == 0) {
1470 struct list_head *devices;
1471 struct btrfs_device *tmp;
1472
1473 device = NULL;
1474 devices = &root->fs_info->fs_devices->devices;
1475 /*
1476 * It is safe to read the devices since the volume_mutex
1477 * is held.
1478 */
1479 list_for_each_entry(tmp, devices, dev_list) {
1480 if (tmp->in_fs_metadata &&
1481 !tmp->is_tgtdev_for_dev_replace &&
1482 !tmp->bdev) {
1483 device = tmp;
1484 break;
1485 }
1486 }
1487 bdev = NULL;
1488 bh = NULL;
1489 disk_super = NULL;
1490 if (!device) {
1491 printk(KERN_ERR "btrfs: no missing devices found to "
1492 "remove\n");
1493 goto out;
1494 }
1495 } else {
1496 ret = btrfs_get_bdev_and_sb(device_path,
1497 FMODE_WRITE | FMODE_EXCL,
1498 root->fs_info->bdev_holder, 0,
1499 &bdev, &bh);
1500 if (ret)
1501 goto out;
1502 disk_super = (struct btrfs_super_block *)bh->b_data;
1503 devid = btrfs_stack_device_id(&disk_super->dev_item);
1504 dev_uuid = disk_super->dev_item.uuid;
1505 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1506 disk_super->fsid);
1507 if (!device) {
1508 ret = -ENOENT;
1509 goto error_brelse;
1510 }
1511 }
1512
1513 if (device->is_tgtdev_for_dev_replace) {
1514 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1515 ret = -EINVAL;
1516 goto error_brelse;
1517 }
1518
1519 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1520 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1521 "device\n");
1522 ret = -EINVAL;
1523 goto error_brelse;
1524 }
1525
1526 if (device->writeable) {
1527 lock_chunks(root);
1528 list_del_init(&device->dev_alloc_list);
1529 unlock_chunks(root);
1530 root->fs_info->fs_devices->rw_devices--;
1531 clear_super = true;
1532 }
1533
1534 ret = btrfs_shrink_device(device, 0);
1535 if (ret)
1536 goto error_undo;
1537
1538 /*
1539 * TODO: the superblock still includes this device in its num_devices
1540 * counter although write_all_supers() is not locked out. This
1541 * could give a filesystem state which requires a degraded mount.
1542 */
1543 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1544 if (ret)
1545 goto error_undo;
1546
1547 spin_lock(&root->fs_info->free_chunk_lock);
1548 root->fs_info->free_chunk_space = device->total_bytes -
1549 device->bytes_used;
1550 spin_unlock(&root->fs_info->free_chunk_lock);
1551
1552 device->in_fs_metadata = 0;
1553 btrfs_scrub_cancel_dev(root->fs_info, device);
1554
1555 /*
1556 * the device list mutex makes sure that we don't change
1557 * the device list while someone else is writing out all
1558 * the device supers.
1559 */
1560
1561 cur_devices = device->fs_devices;
1562 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1563 list_del_rcu(&device->dev_list);
1564
1565 device->fs_devices->num_devices--;
1566 device->fs_devices->total_devices--;
1567
1568 if (device->missing)
1569 root->fs_info->fs_devices->missing_devices--;
1570
1571 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1572 struct btrfs_device, dev_list);
1573 if (device->bdev == root->fs_info->sb->s_bdev)
1574 root->fs_info->sb->s_bdev = next_device->bdev;
1575 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1576 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1577
1578 if (device->bdev)
1579 device->fs_devices->open_devices--;
1580
1581 call_rcu(&device->rcu, free_device);
1582 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1583
1584 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1585 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1586
1587 if (cur_devices->open_devices == 0) {
1588 struct btrfs_fs_devices *fs_devices;
1589 fs_devices = root->fs_info->fs_devices;
1590 while (fs_devices) {
1591 if (fs_devices->seed == cur_devices)
1592 break;
1593 fs_devices = fs_devices->seed;
1594 }
1595 fs_devices->seed = cur_devices->seed;
1596 cur_devices->seed = NULL;
1597 lock_chunks(root);
1598 __btrfs_close_devices(cur_devices);
1599 unlock_chunks(root);
1600 free_fs_devices(cur_devices);
1601 }
1602
1603 root->fs_info->num_tolerated_disk_barrier_failures =
1604 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1605
1606 /*
1607 * at this point, the device is zero sized. We want to
1608 * remove it from the devices list and zero out the old super
1609 */
1610 if (clear_super && disk_super) {
1611 /* make sure this device isn't detected as part of
1612 * the FS anymore
1613 */
1614 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1615 set_buffer_dirty(bh);
1616 sync_dirty_buffer(bh);
1617 }
1618
1619 ret = 0;
1620
1621 /* Notify udev that device has changed */
1622 if (bdev)
1623 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1624
1625 error_brelse:
1626 brelse(bh);
1627 if (bdev)
1628 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1629 out:
1630 mutex_unlock(&uuid_mutex);
1631 return ret;
1632 error_undo:
1633 if (device->writeable) {
1634 lock_chunks(root);
1635 list_add(&device->dev_alloc_list,
1636 &root->fs_info->fs_devices->alloc_list);
1637 unlock_chunks(root);
1638 root->fs_info->fs_devices->rw_devices++;
1639 }
1640 goto error_brelse;
1641 }
1642
1643 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1644 struct btrfs_device *srcdev)
1645 {
1646 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1647 list_del_rcu(&srcdev->dev_list);
1648 list_del_rcu(&srcdev->dev_alloc_list);
1649 fs_info->fs_devices->num_devices--;
1650 if (srcdev->missing) {
1651 fs_info->fs_devices->missing_devices--;
1652 fs_info->fs_devices->rw_devices++;
1653 }
1654 if (srcdev->can_discard)
1655 fs_info->fs_devices->num_can_discard--;
1656 if (srcdev->bdev)
1657 fs_info->fs_devices->open_devices--;
1658
1659 call_rcu(&srcdev->rcu, free_device);
1660 }
1661
1662 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1663 struct btrfs_device *tgtdev)
1664 {
1665 struct btrfs_device *next_device;
1666
1667 WARN_ON(!tgtdev);
1668 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1669 if (tgtdev->bdev) {
1670 btrfs_scratch_superblock(tgtdev);
1671 fs_info->fs_devices->open_devices--;
1672 }
1673 fs_info->fs_devices->num_devices--;
1674 if (tgtdev->can_discard)
1675 fs_info->fs_devices->num_can_discard++;
1676
1677 next_device = list_entry(fs_info->fs_devices->devices.next,
1678 struct btrfs_device, dev_list);
1679 if (tgtdev->bdev == fs_info->sb->s_bdev)
1680 fs_info->sb->s_bdev = next_device->bdev;
1681 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1682 fs_info->fs_devices->latest_bdev = next_device->bdev;
1683 list_del_rcu(&tgtdev->dev_list);
1684
1685 call_rcu(&tgtdev->rcu, free_device);
1686
1687 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1688 }
1689
1690 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1691 struct btrfs_device **device)
1692 {
1693 int ret = 0;
1694 struct btrfs_super_block *disk_super;
1695 u64 devid;
1696 u8 *dev_uuid;
1697 struct block_device *bdev;
1698 struct buffer_head *bh;
1699
1700 *device = NULL;
1701 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1702 root->fs_info->bdev_holder, 0, &bdev, &bh);
1703 if (ret)
1704 return ret;
1705 disk_super = (struct btrfs_super_block *)bh->b_data;
1706 devid = btrfs_stack_device_id(&disk_super->dev_item);
1707 dev_uuid = disk_super->dev_item.uuid;
1708 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1709 disk_super->fsid);
1710 brelse(bh);
1711 if (!*device)
1712 ret = -ENOENT;
1713 blkdev_put(bdev, FMODE_READ);
1714 return ret;
1715 }
1716
1717 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1718 char *device_path,
1719 struct btrfs_device **device)
1720 {
1721 *device = NULL;
1722 if (strcmp(device_path, "missing") == 0) {
1723 struct list_head *devices;
1724 struct btrfs_device *tmp;
1725
1726 devices = &root->fs_info->fs_devices->devices;
1727 /*
1728 * It is safe to read the devices since the volume_mutex
1729 * is held by the caller.
1730 */
1731 list_for_each_entry(tmp, devices, dev_list) {
1732 if (tmp->in_fs_metadata && !tmp->bdev) {
1733 *device = tmp;
1734 break;
1735 }
1736 }
1737
1738 if (!*device) {
1739 pr_err("btrfs: no missing device found\n");
1740 return -ENOENT;
1741 }
1742
1743 return 0;
1744 } else {
1745 return btrfs_find_device_by_path(root, device_path, device);
1746 }
1747 }
1748
1749 /*
1750 * does all the dirty work required for changing file system's UUID.
1751 */
1752 static int btrfs_prepare_sprout(struct btrfs_root *root)
1753 {
1754 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1755 struct btrfs_fs_devices *old_devices;
1756 struct btrfs_fs_devices *seed_devices;
1757 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1758 struct btrfs_device *device;
1759 u64 super_flags;
1760
1761 BUG_ON(!mutex_is_locked(&uuid_mutex));
1762 if (!fs_devices->seeding)
1763 return -EINVAL;
1764
1765 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1766 if (!seed_devices)
1767 return -ENOMEM;
1768
1769 old_devices = clone_fs_devices(fs_devices);
1770 if (IS_ERR(old_devices)) {
1771 kfree(seed_devices);
1772 return PTR_ERR(old_devices);
1773 }
1774
1775 list_add(&old_devices->list, &fs_uuids);
1776
1777 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1778 seed_devices->opened = 1;
1779 INIT_LIST_HEAD(&seed_devices->devices);
1780 INIT_LIST_HEAD(&seed_devices->alloc_list);
1781 mutex_init(&seed_devices->device_list_mutex);
1782
1783 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1784 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1785 synchronize_rcu);
1786 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1787
1788 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1789 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1790 device->fs_devices = seed_devices;
1791 }
1792
1793 fs_devices->seeding = 0;
1794 fs_devices->num_devices = 0;
1795 fs_devices->open_devices = 0;
1796 fs_devices->total_devices = 0;
1797 fs_devices->seed = seed_devices;
1798
1799 generate_random_uuid(fs_devices->fsid);
1800 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1801 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1802 super_flags = btrfs_super_flags(disk_super) &
1803 ~BTRFS_SUPER_FLAG_SEEDING;
1804 btrfs_set_super_flags(disk_super, super_flags);
1805
1806 return 0;
1807 }
1808
1809 /*
1810 * strore the expected generation for seed devices in device items.
1811 */
1812 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1813 struct btrfs_root *root)
1814 {
1815 struct btrfs_path *path;
1816 struct extent_buffer *leaf;
1817 struct btrfs_dev_item *dev_item;
1818 struct btrfs_device *device;
1819 struct btrfs_key key;
1820 u8 fs_uuid[BTRFS_UUID_SIZE];
1821 u8 dev_uuid[BTRFS_UUID_SIZE];
1822 u64 devid;
1823 int ret;
1824
1825 path = btrfs_alloc_path();
1826 if (!path)
1827 return -ENOMEM;
1828
1829 root = root->fs_info->chunk_root;
1830 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1831 key.offset = 0;
1832 key.type = BTRFS_DEV_ITEM_KEY;
1833
1834 while (1) {
1835 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1836 if (ret < 0)
1837 goto error;
1838
1839 leaf = path->nodes[0];
1840 next_slot:
1841 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1842 ret = btrfs_next_leaf(root, path);
1843 if (ret > 0)
1844 break;
1845 if (ret < 0)
1846 goto error;
1847 leaf = path->nodes[0];
1848 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1849 btrfs_release_path(path);
1850 continue;
1851 }
1852
1853 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1854 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1855 key.type != BTRFS_DEV_ITEM_KEY)
1856 break;
1857
1858 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1859 struct btrfs_dev_item);
1860 devid = btrfs_device_id(leaf, dev_item);
1861 read_extent_buffer(leaf, dev_uuid,
1862 (unsigned long)btrfs_device_uuid(dev_item),
1863 BTRFS_UUID_SIZE);
1864 read_extent_buffer(leaf, fs_uuid,
1865 (unsigned long)btrfs_device_fsid(dev_item),
1866 BTRFS_UUID_SIZE);
1867 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1868 fs_uuid);
1869 BUG_ON(!device); /* Logic error */
1870
1871 if (device->fs_devices->seeding) {
1872 btrfs_set_device_generation(leaf, dev_item,
1873 device->generation);
1874 btrfs_mark_buffer_dirty(leaf);
1875 }
1876
1877 path->slots[0]++;
1878 goto next_slot;
1879 }
1880 ret = 0;
1881 error:
1882 btrfs_free_path(path);
1883 return ret;
1884 }
1885
1886 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1887 {
1888 struct request_queue *q;
1889 struct btrfs_trans_handle *trans;
1890 struct btrfs_device *device;
1891 struct block_device *bdev;
1892 struct list_head *devices;
1893 struct super_block *sb = root->fs_info->sb;
1894 struct rcu_string *name;
1895 u64 total_bytes;
1896 int seeding_dev = 0;
1897 int ret = 0;
1898
1899 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1900 return -EROFS;
1901
1902 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1903 root->fs_info->bdev_holder);
1904 if (IS_ERR(bdev))
1905 return PTR_ERR(bdev);
1906
1907 if (root->fs_info->fs_devices->seeding) {
1908 seeding_dev = 1;
1909 down_write(&sb->s_umount);
1910 mutex_lock(&uuid_mutex);
1911 }
1912
1913 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1914
1915 devices = &root->fs_info->fs_devices->devices;
1916
1917 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1918 list_for_each_entry(device, devices, dev_list) {
1919 if (device->bdev == bdev) {
1920 ret = -EEXIST;
1921 mutex_unlock(
1922 &root->fs_info->fs_devices->device_list_mutex);
1923 goto error;
1924 }
1925 }
1926 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1927
1928 device = kzalloc(sizeof(*device), GFP_NOFS);
1929 if (!device) {
1930 /* we can safely leave the fs_devices entry around */
1931 ret = -ENOMEM;
1932 goto error;
1933 }
1934
1935 name = rcu_string_strdup(device_path, GFP_NOFS);
1936 if (!name) {
1937 kfree(device);
1938 ret = -ENOMEM;
1939 goto error;
1940 }
1941 rcu_assign_pointer(device->name, name);
1942
1943 ret = find_next_devid(root, &device->devid);
1944 if (ret) {
1945 rcu_string_free(device->name);
1946 kfree(device);
1947 goto error;
1948 }
1949
1950 trans = btrfs_start_transaction(root, 0);
1951 if (IS_ERR(trans)) {
1952 rcu_string_free(device->name);
1953 kfree(device);
1954 ret = PTR_ERR(trans);
1955 goto error;
1956 }
1957
1958 lock_chunks(root);
1959
1960 q = bdev_get_queue(bdev);
1961 if (blk_queue_discard(q))
1962 device->can_discard = 1;
1963 device->writeable = 1;
1964 device->work.func = pending_bios_fn;
1965 generate_random_uuid(device->uuid);
1966 spin_lock_init(&device->io_lock);
1967 device->generation = trans->transid;
1968 device->io_width = root->sectorsize;
1969 device->io_align = root->sectorsize;
1970 device->sector_size = root->sectorsize;
1971 device->total_bytes = i_size_read(bdev->bd_inode);
1972 device->disk_total_bytes = device->total_bytes;
1973 device->dev_root = root->fs_info->dev_root;
1974 device->bdev = bdev;
1975 device->in_fs_metadata = 1;
1976 device->is_tgtdev_for_dev_replace = 0;
1977 device->mode = FMODE_EXCL;
1978 set_blocksize(device->bdev, 4096);
1979
1980 if (seeding_dev) {
1981 sb->s_flags &= ~MS_RDONLY;
1982 ret = btrfs_prepare_sprout(root);
1983 BUG_ON(ret); /* -ENOMEM */
1984 }
1985
1986 device->fs_devices = root->fs_info->fs_devices;
1987
1988 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1989 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1990 list_add(&device->dev_alloc_list,
1991 &root->fs_info->fs_devices->alloc_list);
1992 root->fs_info->fs_devices->num_devices++;
1993 root->fs_info->fs_devices->open_devices++;
1994 root->fs_info->fs_devices->rw_devices++;
1995 root->fs_info->fs_devices->total_devices++;
1996 if (device->can_discard)
1997 root->fs_info->fs_devices->num_can_discard++;
1998 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1999
2000 spin_lock(&root->fs_info->free_chunk_lock);
2001 root->fs_info->free_chunk_space += device->total_bytes;
2002 spin_unlock(&root->fs_info->free_chunk_lock);
2003
2004 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2005 root->fs_info->fs_devices->rotating = 1;
2006
2007 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2008 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2009 total_bytes + device->total_bytes);
2010
2011 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2012 btrfs_set_super_num_devices(root->fs_info->super_copy,
2013 total_bytes + 1);
2014 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2015
2016 if (seeding_dev) {
2017 ret = init_first_rw_device(trans, root, device);
2018 if (ret) {
2019 btrfs_abort_transaction(trans, root, ret);
2020 goto error_trans;
2021 }
2022 ret = btrfs_finish_sprout(trans, root);
2023 if (ret) {
2024 btrfs_abort_transaction(trans, root, ret);
2025 goto error_trans;
2026 }
2027 } else {
2028 ret = btrfs_add_device(trans, root, device);
2029 if (ret) {
2030 btrfs_abort_transaction(trans, root, ret);
2031 goto error_trans;
2032 }
2033 }
2034
2035 /*
2036 * we've got more storage, clear any full flags on the space
2037 * infos
2038 */
2039 btrfs_clear_space_info_full(root->fs_info);
2040
2041 unlock_chunks(root);
2042 root->fs_info->num_tolerated_disk_barrier_failures =
2043 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2044 ret = btrfs_commit_transaction(trans, root);
2045
2046 if (seeding_dev) {
2047 mutex_unlock(&uuid_mutex);
2048 up_write(&sb->s_umount);
2049
2050 if (ret) /* transaction commit */
2051 return ret;
2052
2053 ret = btrfs_relocate_sys_chunks(root);
2054 if (ret < 0)
2055 btrfs_error(root->fs_info, ret,
2056 "Failed to relocate sys chunks after "
2057 "device initialization. This can be fixed "
2058 "using the \"btrfs balance\" command.");
2059 trans = btrfs_attach_transaction(root);
2060 if (IS_ERR(trans)) {
2061 if (PTR_ERR(trans) == -ENOENT)
2062 return 0;
2063 return PTR_ERR(trans);
2064 }
2065 ret = btrfs_commit_transaction(trans, root);
2066 }
2067
2068 return ret;
2069
2070 error_trans:
2071 unlock_chunks(root);
2072 btrfs_end_transaction(trans, root);
2073 rcu_string_free(device->name);
2074 kfree(device);
2075 error:
2076 blkdev_put(bdev, FMODE_EXCL);
2077 if (seeding_dev) {
2078 mutex_unlock(&uuid_mutex);
2079 up_write(&sb->s_umount);
2080 }
2081 return ret;
2082 }
2083
2084 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2085 struct btrfs_device **device_out)
2086 {
2087 struct request_queue *q;
2088 struct btrfs_device *device;
2089 struct block_device *bdev;
2090 struct btrfs_fs_info *fs_info = root->fs_info;
2091 struct list_head *devices;
2092 struct rcu_string *name;
2093 int ret = 0;
2094
2095 *device_out = NULL;
2096 if (fs_info->fs_devices->seeding)
2097 return -EINVAL;
2098
2099 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2100 fs_info->bdev_holder);
2101 if (IS_ERR(bdev))
2102 return PTR_ERR(bdev);
2103
2104 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2105
2106 devices = &fs_info->fs_devices->devices;
2107 list_for_each_entry(device, devices, dev_list) {
2108 if (device->bdev == bdev) {
2109 ret = -EEXIST;
2110 goto error;
2111 }
2112 }
2113
2114 device = kzalloc(sizeof(*device), GFP_NOFS);
2115 if (!device) {
2116 ret = -ENOMEM;
2117 goto error;
2118 }
2119
2120 name = rcu_string_strdup(device_path, GFP_NOFS);
2121 if (!name) {
2122 kfree(device);
2123 ret = -ENOMEM;
2124 goto error;
2125 }
2126 rcu_assign_pointer(device->name, name);
2127
2128 q = bdev_get_queue(bdev);
2129 if (blk_queue_discard(q))
2130 device->can_discard = 1;
2131 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2132 device->writeable = 1;
2133 device->work.func = pending_bios_fn;
2134 generate_random_uuid(device->uuid);
2135 device->devid = BTRFS_DEV_REPLACE_DEVID;
2136 spin_lock_init(&device->io_lock);
2137 device->generation = 0;
2138 device->io_width = root->sectorsize;
2139 device->io_align = root->sectorsize;
2140 device->sector_size = root->sectorsize;
2141 device->total_bytes = i_size_read(bdev->bd_inode);
2142 device->disk_total_bytes = device->total_bytes;
2143 device->dev_root = fs_info->dev_root;
2144 device->bdev = bdev;
2145 device->in_fs_metadata = 1;
2146 device->is_tgtdev_for_dev_replace = 1;
2147 device->mode = FMODE_EXCL;
2148 set_blocksize(device->bdev, 4096);
2149 device->fs_devices = fs_info->fs_devices;
2150 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2151 fs_info->fs_devices->num_devices++;
2152 fs_info->fs_devices->open_devices++;
2153 if (device->can_discard)
2154 fs_info->fs_devices->num_can_discard++;
2155 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2156
2157 *device_out = device;
2158 return ret;
2159
2160 error:
2161 blkdev_put(bdev, FMODE_EXCL);
2162 return ret;
2163 }
2164
2165 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2166 struct btrfs_device *tgtdev)
2167 {
2168 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2169 tgtdev->io_width = fs_info->dev_root->sectorsize;
2170 tgtdev->io_align = fs_info->dev_root->sectorsize;
2171 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2172 tgtdev->dev_root = fs_info->dev_root;
2173 tgtdev->in_fs_metadata = 1;
2174 }
2175
2176 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2177 struct btrfs_device *device)
2178 {
2179 int ret;
2180 struct btrfs_path *path;
2181 struct btrfs_root *root;
2182 struct btrfs_dev_item *dev_item;
2183 struct extent_buffer *leaf;
2184 struct btrfs_key key;
2185
2186 root = device->dev_root->fs_info->chunk_root;
2187
2188 path = btrfs_alloc_path();
2189 if (!path)
2190 return -ENOMEM;
2191
2192 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2193 key.type = BTRFS_DEV_ITEM_KEY;
2194 key.offset = device->devid;
2195
2196 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2197 if (ret < 0)
2198 goto out;
2199
2200 if (ret > 0) {
2201 ret = -ENOENT;
2202 goto out;
2203 }
2204
2205 leaf = path->nodes[0];
2206 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2207
2208 btrfs_set_device_id(leaf, dev_item, device->devid);
2209 btrfs_set_device_type(leaf, dev_item, device->type);
2210 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2211 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2212 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2213 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2214 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2215 btrfs_mark_buffer_dirty(leaf);
2216
2217 out:
2218 btrfs_free_path(path);
2219 return ret;
2220 }
2221
2222 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2223 struct btrfs_device *device, u64 new_size)
2224 {
2225 struct btrfs_super_block *super_copy =
2226 device->dev_root->fs_info->super_copy;
2227 u64 old_total = btrfs_super_total_bytes(super_copy);
2228 u64 diff = new_size - device->total_bytes;
2229
2230 if (!device->writeable)
2231 return -EACCES;
2232 if (new_size <= device->total_bytes ||
2233 device->is_tgtdev_for_dev_replace)
2234 return -EINVAL;
2235
2236 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2237 device->fs_devices->total_rw_bytes += diff;
2238
2239 device->total_bytes = new_size;
2240 device->disk_total_bytes = new_size;
2241 btrfs_clear_space_info_full(device->dev_root->fs_info);
2242
2243 return btrfs_update_device(trans, device);
2244 }
2245
2246 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2247 struct btrfs_device *device, u64 new_size)
2248 {
2249 int ret;
2250 lock_chunks(device->dev_root);
2251 ret = __btrfs_grow_device(trans, device, new_size);
2252 unlock_chunks(device->dev_root);
2253 return ret;
2254 }
2255
2256 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2257 struct btrfs_root *root,
2258 u64 chunk_tree, u64 chunk_objectid,
2259 u64 chunk_offset)
2260 {
2261 int ret;
2262 struct btrfs_path *path;
2263 struct btrfs_key key;
2264
2265 root = root->fs_info->chunk_root;
2266 path = btrfs_alloc_path();
2267 if (!path)
2268 return -ENOMEM;
2269
2270 key.objectid = chunk_objectid;
2271 key.offset = chunk_offset;
2272 key.type = BTRFS_CHUNK_ITEM_KEY;
2273
2274 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2275 if (ret < 0)
2276 goto out;
2277 else if (ret > 0) { /* Logic error or corruption */
2278 btrfs_error(root->fs_info, -ENOENT,
2279 "Failed lookup while freeing chunk.");
2280 ret = -ENOENT;
2281 goto out;
2282 }
2283
2284 ret = btrfs_del_item(trans, root, path);
2285 if (ret < 0)
2286 btrfs_error(root->fs_info, ret,
2287 "Failed to delete chunk item.");
2288 out:
2289 btrfs_free_path(path);
2290 return ret;
2291 }
2292
2293 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2294 chunk_offset)
2295 {
2296 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2297 struct btrfs_disk_key *disk_key;
2298 struct btrfs_chunk *chunk;
2299 u8 *ptr;
2300 int ret = 0;
2301 u32 num_stripes;
2302 u32 array_size;
2303 u32 len = 0;
2304 u32 cur;
2305 struct btrfs_key key;
2306
2307 array_size = btrfs_super_sys_array_size(super_copy);
2308
2309 ptr = super_copy->sys_chunk_array;
2310 cur = 0;
2311
2312 while (cur < array_size) {
2313 disk_key = (struct btrfs_disk_key *)ptr;
2314 btrfs_disk_key_to_cpu(&key, disk_key);
2315
2316 len = sizeof(*disk_key);
2317
2318 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2319 chunk = (struct btrfs_chunk *)(ptr + len);
2320 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2321 len += btrfs_chunk_item_size(num_stripes);
2322 } else {
2323 ret = -EIO;
2324 break;
2325 }
2326 if (key.objectid == chunk_objectid &&
2327 key.offset == chunk_offset) {
2328 memmove(ptr, ptr + len, array_size - (cur + len));
2329 array_size -= len;
2330 btrfs_set_super_sys_array_size(super_copy, array_size);
2331 } else {
2332 ptr += len;
2333 cur += len;
2334 }
2335 }
2336 return ret;
2337 }
2338
2339 static int btrfs_relocate_chunk(struct btrfs_root *root,
2340 u64 chunk_tree, u64 chunk_objectid,
2341 u64 chunk_offset)
2342 {
2343 struct extent_map_tree *em_tree;
2344 struct btrfs_root *extent_root;
2345 struct btrfs_trans_handle *trans;
2346 struct extent_map *em;
2347 struct map_lookup *map;
2348 int ret;
2349 int i;
2350
2351 root = root->fs_info->chunk_root;
2352 extent_root = root->fs_info->extent_root;
2353 em_tree = &root->fs_info->mapping_tree.map_tree;
2354
2355 ret = btrfs_can_relocate(extent_root, chunk_offset);
2356 if (ret)
2357 return -ENOSPC;
2358
2359 /* step one, relocate all the extents inside this chunk */
2360 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2361 if (ret)
2362 return ret;
2363
2364 trans = btrfs_start_transaction(root, 0);
2365 BUG_ON(IS_ERR(trans));
2366
2367 lock_chunks(root);
2368
2369 /*
2370 * step two, delete the device extents and the
2371 * chunk tree entries
2372 */
2373 read_lock(&em_tree->lock);
2374 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2375 read_unlock(&em_tree->lock);
2376
2377 BUG_ON(!em || em->start > chunk_offset ||
2378 em->start + em->len < chunk_offset);
2379 map = (struct map_lookup *)em->bdev;
2380
2381 for (i = 0; i < map->num_stripes; i++) {
2382 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2383 map->stripes[i].physical);
2384 BUG_ON(ret);
2385
2386 if (map->stripes[i].dev) {
2387 ret = btrfs_update_device(trans, map->stripes[i].dev);
2388 BUG_ON(ret);
2389 }
2390 }
2391 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2392 chunk_offset);
2393
2394 BUG_ON(ret);
2395
2396 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2397
2398 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2399 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2400 BUG_ON(ret);
2401 }
2402
2403 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2404 BUG_ON(ret);
2405
2406 write_lock(&em_tree->lock);
2407 remove_extent_mapping(em_tree, em);
2408 write_unlock(&em_tree->lock);
2409
2410 kfree(map);
2411 em->bdev = NULL;
2412
2413 /* once for the tree */
2414 free_extent_map(em);
2415 /* once for us */
2416 free_extent_map(em);
2417
2418 unlock_chunks(root);
2419 btrfs_end_transaction(trans, root);
2420 return 0;
2421 }
2422
2423 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2424 {
2425 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2426 struct btrfs_path *path;
2427 struct extent_buffer *leaf;
2428 struct btrfs_chunk *chunk;
2429 struct btrfs_key key;
2430 struct btrfs_key found_key;
2431 u64 chunk_tree = chunk_root->root_key.objectid;
2432 u64 chunk_type;
2433 bool retried = false;
2434 int failed = 0;
2435 int ret;
2436
2437 path = btrfs_alloc_path();
2438 if (!path)
2439 return -ENOMEM;
2440
2441 again:
2442 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2443 key.offset = (u64)-1;
2444 key.type = BTRFS_CHUNK_ITEM_KEY;
2445
2446 while (1) {
2447 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2448 if (ret < 0)
2449 goto error;
2450 BUG_ON(ret == 0); /* Corruption */
2451
2452 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2453 key.type);
2454 if (ret < 0)
2455 goto error;
2456 if (ret > 0)
2457 break;
2458
2459 leaf = path->nodes[0];
2460 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2461
2462 chunk = btrfs_item_ptr(leaf, path->slots[0],
2463 struct btrfs_chunk);
2464 chunk_type = btrfs_chunk_type(leaf, chunk);
2465 btrfs_release_path(path);
2466
2467 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2468 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2469 found_key.objectid,
2470 found_key.offset);
2471 if (ret == -ENOSPC)
2472 failed++;
2473 else if (ret)
2474 BUG();
2475 }
2476
2477 if (found_key.offset == 0)
2478 break;
2479 key.offset = found_key.offset - 1;
2480 }
2481 ret = 0;
2482 if (failed && !retried) {
2483 failed = 0;
2484 retried = true;
2485 goto again;
2486 } else if (failed && retried) {
2487 WARN_ON(1);
2488 ret = -ENOSPC;
2489 }
2490 error:
2491 btrfs_free_path(path);
2492 return ret;
2493 }
2494
2495 static int insert_balance_item(struct btrfs_root *root,
2496 struct btrfs_balance_control *bctl)
2497 {
2498 struct btrfs_trans_handle *trans;
2499 struct btrfs_balance_item *item;
2500 struct btrfs_disk_balance_args disk_bargs;
2501 struct btrfs_path *path;
2502 struct extent_buffer *leaf;
2503 struct btrfs_key key;
2504 int ret, err;
2505
2506 path = btrfs_alloc_path();
2507 if (!path)
2508 return -ENOMEM;
2509
2510 trans = btrfs_start_transaction(root, 0);
2511 if (IS_ERR(trans)) {
2512 btrfs_free_path(path);
2513 return PTR_ERR(trans);
2514 }
2515
2516 key.objectid = BTRFS_BALANCE_OBJECTID;
2517 key.type = BTRFS_BALANCE_ITEM_KEY;
2518 key.offset = 0;
2519
2520 ret = btrfs_insert_empty_item(trans, root, path, &key,
2521 sizeof(*item));
2522 if (ret)
2523 goto out;
2524
2525 leaf = path->nodes[0];
2526 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2527
2528 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2529
2530 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2531 btrfs_set_balance_data(leaf, item, &disk_bargs);
2532 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2533 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2534 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2535 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2536
2537 btrfs_set_balance_flags(leaf, item, bctl->flags);
2538
2539 btrfs_mark_buffer_dirty(leaf);
2540 out:
2541 btrfs_free_path(path);
2542 err = btrfs_commit_transaction(trans, root);
2543 if (err && !ret)
2544 ret = err;
2545 return ret;
2546 }
2547
2548 static int del_balance_item(struct btrfs_root *root)
2549 {
2550 struct btrfs_trans_handle *trans;
2551 struct btrfs_path *path;
2552 struct btrfs_key key;
2553 int ret, err;
2554
2555 path = btrfs_alloc_path();
2556 if (!path)
2557 return -ENOMEM;
2558
2559 trans = btrfs_start_transaction(root, 0);
2560 if (IS_ERR(trans)) {
2561 btrfs_free_path(path);
2562 return PTR_ERR(trans);
2563 }
2564
2565 key.objectid = BTRFS_BALANCE_OBJECTID;
2566 key.type = BTRFS_BALANCE_ITEM_KEY;
2567 key.offset = 0;
2568
2569 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2570 if (ret < 0)
2571 goto out;
2572 if (ret > 0) {
2573 ret = -ENOENT;
2574 goto out;
2575 }
2576
2577 ret = btrfs_del_item(trans, root, path);
2578 out:
2579 btrfs_free_path(path);
2580 err = btrfs_commit_transaction(trans, root);
2581 if (err && !ret)
2582 ret = err;
2583 return ret;
2584 }
2585
2586 /*
2587 * This is a heuristic used to reduce the number of chunks balanced on
2588 * resume after balance was interrupted.
2589 */
2590 static void update_balance_args(struct btrfs_balance_control *bctl)
2591 {
2592 /*
2593 * Turn on soft mode for chunk types that were being converted.
2594 */
2595 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2596 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2597 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2598 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2599 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2600 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2601
2602 /*
2603 * Turn on usage filter if is not already used. The idea is
2604 * that chunks that we have already balanced should be
2605 * reasonably full. Don't do it for chunks that are being
2606 * converted - that will keep us from relocating unconverted
2607 * (albeit full) chunks.
2608 */
2609 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2610 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2611 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2612 bctl->data.usage = 90;
2613 }
2614 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2615 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2616 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2617 bctl->sys.usage = 90;
2618 }
2619 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2620 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2621 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2622 bctl->meta.usage = 90;
2623 }
2624 }
2625
2626 /*
2627 * Should be called with both balance and volume mutexes held to
2628 * serialize other volume operations (add_dev/rm_dev/resize) with
2629 * restriper. Same goes for unset_balance_control.
2630 */
2631 static void set_balance_control(struct btrfs_balance_control *bctl)
2632 {
2633 struct btrfs_fs_info *fs_info = bctl->fs_info;
2634
2635 BUG_ON(fs_info->balance_ctl);
2636
2637 spin_lock(&fs_info->balance_lock);
2638 fs_info->balance_ctl = bctl;
2639 spin_unlock(&fs_info->balance_lock);
2640 }
2641
2642 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2643 {
2644 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2645
2646 BUG_ON(!fs_info->balance_ctl);
2647
2648 spin_lock(&fs_info->balance_lock);
2649 fs_info->balance_ctl = NULL;
2650 spin_unlock(&fs_info->balance_lock);
2651
2652 kfree(bctl);
2653 }
2654
2655 /*
2656 * Balance filters. Return 1 if chunk should be filtered out
2657 * (should not be balanced).
2658 */
2659 static int chunk_profiles_filter(u64 chunk_type,
2660 struct btrfs_balance_args *bargs)
2661 {
2662 chunk_type = chunk_to_extended(chunk_type) &
2663 BTRFS_EXTENDED_PROFILE_MASK;
2664
2665 if (bargs->profiles & chunk_type)
2666 return 0;
2667
2668 return 1;
2669 }
2670
2671 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2672 struct btrfs_balance_args *bargs)
2673 {
2674 struct btrfs_block_group_cache *cache;
2675 u64 chunk_used, user_thresh;
2676 int ret = 1;
2677
2678 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2679 chunk_used = btrfs_block_group_used(&cache->item);
2680
2681 if (bargs->usage == 0)
2682 user_thresh = 0;
2683 else if (bargs->usage > 100)
2684 user_thresh = cache->key.offset;
2685 else
2686 user_thresh = div_factor_fine(cache->key.offset,
2687 bargs->usage);
2688
2689 if (chunk_used < user_thresh)
2690 ret = 0;
2691
2692 btrfs_put_block_group(cache);
2693 return ret;
2694 }
2695
2696 static int chunk_devid_filter(struct extent_buffer *leaf,
2697 struct btrfs_chunk *chunk,
2698 struct btrfs_balance_args *bargs)
2699 {
2700 struct btrfs_stripe *stripe;
2701 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2702 int i;
2703
2704 for (i = 0; i < num_stripes; i++) {
2705 stripe = btrfs_stripe_nr(chunk, i);
2706 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2707 return 0;
2708 }
2709
2710 return 1;
2711 }
2712
2713 /* [pstart, pend) */
2714 static int chunk_drange_filter(struct extent_buffer *leaf,
2715 struct btrfs_chunk *chunk,
2716 u64 chunk_offset,
2717 struct btrfs_balance_args *bargs)
2718 {
2719 struct btrfs_stripe *stripe;
2720 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2721 u64 stripe_offset;
2722 u64 stripe_length;
2723 int factor;
2724 int i;
2725
2726 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2727 return 0;
2728
2729 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2730 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2731 factor = 2;
2732 else
2733 factor = 1;
2734 factor = num_stripes / factor;
2735
2736 for (i = 0; i < num_stripes; i++) {
2737 stripe = btrfs_stripe_nr(chunk, i);
2738 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2739 continue;
2740
2741 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2742 stripe_length = btrfs_chunk_length(leaf, chunk);
2743 do_div(stripe_length, factor);
2744
2745 if (stripe_offset < bargs->pend &&
2746 stripe_offset + stripe_length > bargs->pstart)
2747 return 0;
2748 }
2749
2750 return 1;
2751 }
2752
2753 /* [vstart, vend) */
2754 static int chunk_vrange_filter(struct extent_buffer *leaf,
2755 struct btrfs_chunk *chunk,
2756 u64 chunk_offset,
2757 struct btrfs_balance_args *bargs)
2758 {
2759 if (chunk_offset < bargs->vend &&
2760 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2761 /* at least part of the chunk is inside this vrange */
2762 return 0;
2763
2764 return 1;
2765 }
2766
2767 static int chunk_soft_convert_filter(u64 chunk_type,
2768 struct btrfs_balance_args *bargs)
2769 {
2770 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2771 return 0;
2772
2773 chunk_type = chunk_to_extended(chunk_type) &
2774 BTRFS_EXTENDED_PROFILE_MASK;
2775
2776 if (bargs->target == chunk_type)
2777 return 1;
2778
2779 return 0;
2780 }
2781
2782 static int should_balance_chunk(struct btrfs_root *root,
2783 struct extent_buffer *leaf,
2784 struct btrfs_chunk *chunk, u64 chunk_offset)
2785 {
2786 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2787 struct btrfs_balance_args *bargs = NULL;
2788 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2789
2790 /* type filter */
2791 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2792 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2793 return 0;
2794 }
2795
2796 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2797 bargs = &bctl->data;
2798 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2799 bargs = &bctl->sys;
2800 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2801 bargs = &bctl->meta;
2802
2803 /* profiles filter */
2804 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2805 chunk_profiles_filter(chunk_type, bargs)) {
2806 return 0;
2807 }
2808
2809 /* usage filter */
2810 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2811 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2812 return 0;
2813 }
2814
2815 /* devid filter */
2816 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2817 chunk_devid_filter(leaf, chunk, bargs)) {
2818 return 0;
2819 }
2820
2821 /* drange filter, makes sense only with devid filter */
2822 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2823 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2824 return 0;
2825 }
2826
2827 /* vrange filter */
2828 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2829 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2830 return 0;
2831 }
2832
2833 /* soft profile changing mode */
2834 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2835 chunk_soft_convert_filter(chunk_type, bargs)) {
2836 return 0;
2837 }
2838
2839 return 1;
2840 }
2841
2842 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2843 {
2844 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2845 struct btrfs_root *chunk_root = fs_info->chunk_root;
2846 struct btrfs_root *dev_root = fs_info->dev_root;
2847 struct list_head *devices;
2848 struct btrfs_device *device;
2849 u64 old_size;
2850 u64 size_to_free;
2851 struct btrfs_chunk *chunk;
2852 struct btrfs_path *path;
2853 struct btrfs_key key;
2854 struct btrfs_key found_key;
2855 struct btrfs_trans_handle *trans;
2856 struct extent_buffer *leaf;
2857 int slot;
2858 int ret;
2859 int enospc_errors = 0;
2860 bool counting = true;
2861
2862 /* step one make some room on all the devices */
2863 devices = &fs_info->fs_devices->devices;
2864 list_for_each_entry(device, devices, dev_list) {
2865 old_size = device->total_bytes;
2866 size_to_free = div_factor(old_size, 1);
2867 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2868 if (!device->writeable ||
2869 device->total_bytes - device->bytes_used > size_to_free ||
2870 device->is_tgtdev_for_dev_replace)
2871 continue;
2872
2873 ret = btrfs_shrink_device(device, old_size - size_to_free);
2874 if (ret == -ENOSPC)
2875 break;
2876 BUG_ON(ret);
2877
2878 trans = btrfs_start_transaction(dev_root, 0);
2879 BUG_ON(IS_ERR(trans));
2880
2881 ret = btrfs_grow_device(trans, device, old_size);
2882 BUG_ON(ret);
2883
2884 btrfs_end_transaction(trans, dev_root);
2885 }
2886
2887 /* step two, relocate all the chunks */
2888 path = btrfs_alloc_path();
2889 if (!path) {
2890 ret = -ENOMEM;
2891 goto error;
2892 }
2893
2894 /* zero out stat counters */
2895 spin_lock(&fs_info->balance_lock);
2896 memset(&bctl->stat, 0, sizeof(bctl->stat));
2897 spin_unlock(&fs_info->balance_lock);
2898 again:
2899 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2900 key.offset = (u64)-1;
2901 key.type = BTRFS_CHUNK_ITEM_KEY;
2902
2903 while (1) {
2904 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2905 atomic_read(&fs_info->balance_cancel_req)) {
2906 ret = -ECANCELED;
2907 goto error;
2908 }
2909
2910 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2911 if (ret < 0)
2912 goto error;
2913
2914 /*
2915 * this shouldn't happen, it means the last relocate
2916 * failed
2917 */
2918 if (ret == 0)
2919 BUG(); /* FIXME break ? */
2920
2921 ret = btrfs_previous_item(chunk_root, path, 0,
2922 BTRFS_CHUNK_ITEM_KEY);
2923 if (ret) {
2924 ret = 0;
2925 break;
2926 }
2927
2928 leaf = path->nodes[0];
2929 slot = path->slots[0];
2930 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2931
2932 if (found_key.objectid != key.objectid)
2933 break;
2934
2935 /* chunk zero is special */
2936 if (found_key.offset == 0)
2937 break;
2938
2939 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2940
2941 if (!counting) {
2942 spin_lock(&fs_info->balance_lock);
2943 bctl->stat.considered++;
2944 spin_unlock(&fs_info->balance_lock);
2945 }
2946
2947 ret = should_balance_chunk(chunk_root, leaf, chunk,
2948 found_key.offset);
2949 btrfs_release_path(path);
2950 if (!ret)
2951 goto loop;
2952
2953 if (counting) {
2954 spin_lock(&fs_info->balance_lock);
2955 bctl->stat.expected++;
2956 spin_unlock(&fs_info->balance_lock);
2957 goto loop;
2958 }
2959
2960 ret = btrfs_relocate_chunk(chunk_root,
2961 chunk_root->root_key.objectid,
2962 found_key.objectid,
2963 found_key.offset);
2964 if (ret && ret != -ENOSPC)
2965 goto error;
2966 if (ret == -ENOSPC) {
2967 enospc_errors++;
2968 } else {
2969 spin_lock(&fs_info->balance_lock);
2970 bctl->stat.completed++;
2971 spin_unlock(&fs_info->balance_lock);
2972 }
2973 loop:
2974 key.offset = found_key.offset - 1;
2975 }
2976
2977 if (counting) {
2978 btrfs_release_path(path);
2979 counting = false;
2980 goto again;
2981 }
2982 error:
2983 btrfs_free_path(path);
2984 if (enospc_errors) {
2985 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2986 enospc_errors);
2987 if (!ret)
2988 ret = -ENOSPC;
2989 }
2990
2991 return ret;
2992 }
2993
2994 /**
2995 * alloc_profile_is_valid - see if a given profile is valid and reduced
2996 * @flags: profile to validate
2997 * @extended: if true @flags is treated as an extended profile
2998 */
2999 static int alloc_profile_is_valid(u64 flags, int extended)
3000 {
3001 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3002 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3003
3004 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3005
3006 /* 1) check that all other bits are zeroed */
3007 if (flags & ~mask)
3008 return 0;
3009
3010 /* 2) see if profile is reduced */
3011 if (flags == 0)
3012 return !extended; /* "0" is valid for usual profiles */
3013
3014 /* true if exactly one bit set */
3015 return (flags & (flags - 1)) == 0;
3016 }
3017
3018 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3019 {
3020 /* cancel requested || normal exit path */
3021 return atomic_read(&fs_info->balance_cancel_req) ||
3022 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3023 atomic_read(&fs_info->balance_cancel_req) == 0);
3024 }
3025
3026 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3027 {
3028 int ret;
3029
3030 unset_balance_control(fs_info);
3031 ret = del_balance_item(fs_info->tree_root);
3032 BUG_ON(ret);
3033
3034 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3035 }
3036
3037 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3038 struct btrfs_ioctl_balance_args *bargs);
3039
3040 /*
3041 * Should be called with both balance and volume mutexes held
3042 */
3043 int btrfs_balance(struct btrfs_balance_control *bctl,
3044 struct btrfs_ioctl_balance_args *bargs)
3045 {
3046 struct btrfs_fs_info *fs_info = bctl->fs_info;
3047 u64 allowed;
3048 int mixed = 0;
3049 int ret;
3050 u64 num_devices;
3051 unsigned seq;
3052
3053 if (btrfs_fs_closing(fs_info) ||
3054 atomic_read(&fs_info->balance_pause_req) ||
3055 atomic_read(&fs_info->balance_cancel_req)) {
3056 ret = -EINVAL;
3057 goto out;
3058 }
3059
3060 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3061 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3062 mixed = 1;
3063
3064 /*
3065 * In case of mixed groups both data and meta should be picked,
3066 * and identical options should be given for both of them.
3067 */
3068 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3069 if (mixed && (bctl->flags & allowed)) {
3070 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3071 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3072 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3073 printk(KERN_ERR "btrfs: with mixed groups data and "
3074 "metadata balance options must be the same\n");
3075 ret = -EINVAL;
3076 goto out;
3077 }
3078 }
3079
3080 num_devices = fs_info->fs_devices->num_devices;
3081 btrfs_dev_replace_lock(&fs_info->dev_replace);
3082 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3083 BUG_ON(num_devices < 1);
3084 num_devices--;
3085 }
3086 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3087 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3088 if (num_devices == 1)
3089 allowed |= BTRFS_BLOCK_GROUP_DUP;
3090 else if (num_devices < 4)
3091 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3092 else
3093 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3094 BTRFS_BLOCK_GROUP_RAID10);
3095
3096 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3097 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3098 (bctl->data.target & ~allowed))) {
3099 printk(KERN_ERR "btrfs: unable to start balance with target "
3100 "data profile %llu\n",
3101 (unsigned long long)bctl->data.target);
3102 ret = -EINVAL;
3103 goto out;
3104 }
3105 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3106 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3107 (bctl->meta.target & ~allowed))) {
3108 printk(KERN_ERR "btrfs: unable to start balance with target "
3109 "metadata profile %llu\n",
3110 (unsigned long long)bctl->meta.target);
3111 ret = -EINVAL;
3112 goto out;
3113 }
3114 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3115 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3116 (bctl->sys.target & ~allowed))) {
3117 printk(KERN_ERR "btrfs: unable to start balance with target "
3118 "system profile %llu\n",
3119 (unsigned long long)bctl->sys.target);
3120 ret = -EINVAL;
3121 goto out;
3122 }
3123
3124 /* allow dup'ed data chunks only in mixed mode */
3125 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3126 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3127 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3128 ret = -EINVAL;
3129 goto out;
3130 }
3131
3132 /* allow to reduce meta or sys integrity only if force set */
3133 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3134 BTRFS_BLOCK_GROUP_RAID10;
3135 do {
3136 seq = read_seqbegin(&fs_info->profiles_lock);
3137
3138 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139 (fs_info->avail_system_alloc_bits & allowed) &&
3140 !(bctl->sys.target & allowed)) ||
3141 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3142 (fs_info->avail_metadata_alloc_bits & allowed) &&
3143 !(bctl->meta.target & allowed))) {
3144 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3145 printk(KERN_INFO "btrfs: force reducing metadata "
3146 "integrity\n");
3147 } else {
3148 printk(KERN_ERR "btrfs: balance will reduce metadata "
3149 "integrity, use force if you want this\n");
3150 ret = -EINVAL;
3151 goto out;
3152 }
3153 }
3154 } while (read_seqretry(&fs_info->profiles_lock, seq));
3155
3156 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3157 int num_tolerated_disk_barrier_failures;
3158 u64 target = bctl->sys.target;
3159
3160 num_tolerated_disk_barrier_failures =
3161 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3162 if (num_tolerated_disk_barrier_failures > 0 &&
3163 (target &
3164 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3165 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3166 num_tolerated_disk_barrier_failures = 0;
3167 else if (num_tolerated_disk_barrier_failures > 1 &&
3168 (target &
3169 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3170 num_tolerated_disk_barrier_failures = 1;
3171
3172 fs_info->num_tolerated_disk_barrier_failures =
3173 num_tolerated_disk_barrier_failures;
3174 }
3175
3176 ret = insert_balance_item(fs_info->tree_root, bctl);
3177 if (ret && ret != -EEXIST)
3178 goto out;
3179
3180 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3181 BUG_ON(ret == -EEXIST);
3182 set_balance_control(bctl);
3183 } else {
3184 BUG_ON(ret != -EEXIST);
3185 spin_lock(&fs_info->balance_lock);
3186 update_balance_args(bctl);
3187 spin_unlock(&fs_info->balance_lock);
3188 }
3189
3190 atomic_inc(&fs_info->balance_running);
3191 mutex_unlock(&fs_info->balance_mutex);
3192
3193 ret = __btrfs_balance(fs_info);
3194
3195 mutex_lock(&fs_info->balance_mutex);
3196 atomic_dec(&fs_info->balance_running);
3197
3198 if (bargs) {
3199 memset(bargs, 0, sizeof(*bargs));
3200 update_ioctl_balance_args(fs_info, 0, bargs);
3201 }
3202
3203 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3204 balance_need_close(fs_info)) {
3205 __cancel_balance(fs_info);
3206 }
3207
3208 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3209 fs_info->num_tolerated_disk_barrier_failures =
3210 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3211 }
3212
3213 wake_up(&fs_info->balance_wait_q);
3214
3215 return ret;
3216 out:
3217 if (bctl->flags & BTRFS_BALANCE_RESUME)
3218 __cancel_balance(fs_info);
3219 else {
3220 kfree(bctl);
3221 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3222 }
3223 return ret;
3224 }
3225
3226 static int balance_kthread(void *data)
3227 {
3228 struct btrfs_fs_info *fs_info = data;
3229 int ret = 0;
3230
3231 mutex_lock(&fs_info->volume_mutex);
3232 mutex_lock(&fs_info->balance_mutex);
3233
3234 if (fs_info->balance_ctl) {
3235 printk(KERN_INFO "btrfs: continuing balance\n");
3236 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3237 }
3238
3239 mutex_unlock(&fs_info->balance_mutex);
3240 mutex_unlock(&fs_info->volume_mutex);
3241
3242 return ret;
3243 }
3244
3245 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3246 {
3247 struct task_struct *tsk;
3248
3249 spin_lock(&fs_info->balance_lock);
3250 if (!fs_info->balance_ctl) {
3251 spin_unlock(&fs_info->balance_lock);
3252 return 0;
3253 }
3254 spin_unlock(&fs_info->balance_lock);
3255
3256 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3257 printk(KERN_INFO "btrfs: force skipping balance\n");
3258 return 0;
3259 }
3260
3261 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3262 if (IS_ERR(tsk))
3263 return PTR_ERR(tsk);
3264
3265 return 0;
3266 }
3267
3268 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3269 {
3270 struct btrfs_balance_control *bctl;
3271 struct btrfs_balance_item *item;
3272 struct btrfs_disk_balance_args disk_bargs;
3273 struct btrfs_path *path;
3274 struct extent_buffer *leaf;
3275 struct btrfs_key key;
3276 int ret;
3277
3278 path = btrfs_alloc_path();
3279 if (!path)
3280 return -ENOMEM;
3281
3282 key.objectid = BTRFS_BALANCE_OBJECTID;
3283 key.type = BTRFS_BALANCE_ITEM_KEY;
3284 key.offset = 0;
3285
3286 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3287 if (ret < 0)
3288 goto out;
3289 if (ret > 0) { /* ret = -ENOENT; */
3290 ret = 0;
3291 goto out;
3292 }
3293
3294 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3295 if (!bctl) {
3296 ret = -ENOMEM;
3297 goto out;
3298 }
3299
3300 leaf = path->nodes[0];
3301 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3302
3303 bctl->fs_info = fs_info;
3304 bctl->flags = btrfs_balance_flags(leaf, item);
3305 bctl->flags |= BTRFS_BALANCE_RESUME;
3306
3307 btrfs_balance_data(leaf, item, &disk_bargs);
3308 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3309 btrfs_balance_meta(leaf, item, &disk_bargs);
3310 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3311 btrfs_balance_sys(leaf, item, &disk_bargs);
3312 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3313
3314 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3315
3316 mutex_lock(&fs_info->volume_mutex);
3317 mutex_lock(&fs_info->balance_mutex);
3318
3319 set_balance_control(bctl);
3320
3321 mutex_unlock(&fs_info->balance_mutex);
3322 mutex_unlock(&fs_info->volume_mutex);
3323 out:
3324 btrfs_free_path(path);
3325 return ret;
3326 }
3327
3328 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3329 {
3330 int ret = 0;
3331
3332 mutex_lock(&fs_info->balance_mutex);
3333 if (!fs_info->balance_ctl) {
3334 mutex_unlock(&fs_info->balance_mutex);
3335 return -ENOTCONN;
3336 }
3337
3338 if (atomic_read(&fs_info->balance_running)) {
3339 atomic_inc(&fs_info->balance_pause_req);
3340 mutex_unlock(&fs_info->balance_mutex);
3341
3342 wait_event(fs_info->balance_wait_q,
3343 atomic_read(&fs_info->balance_running) == 0);
3344
3345 mutex_lock(&fs_info->balance_mutex);
3346 /* we are good with balance_ctl ripped off from under us */
3347 BUG_ON(atomic_read(&fs_info->balance_running));
3348 atomic_dec(&fs_info->balance_pause_req);
3349 } else {
3350 ret = -ENOTCONN;
3351 }
3352
3353 mutex_unlock(&fs_info->balance_mutex);
3354 return ret;
3355 }
3356
3357 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3358 {
3359 mutex_lock(&fs_info->balance_mutex);
3360 if (!fs_info->balance_ctl) {
3361 mutex_unlock(&fs_info->balance_mutex);
3362 return -ENOTCONN;
3363 }
3364
3365 atomic_inc(&fs_info->balance_cancel_req);
3366 /*
3367 * if we are running just wait and return, balance item is
3368 * deleted in btrfs_balance in this case
3369 */
3370 if (atomic_read(&fs_info->balance_running)) {
3371 mutex_unlock(&fs_info->balance_mutex);
3372 wait_event(fs_info->balance_wait_q,
3373 atomic_read(&fs_info->balance_running) == 0);
3374 mutex_lock(&fs_info->balance_mutex);
3375 } else {
3376 /* __cancel_balance needs volume_mutex */
3377 mutex_unlock(&fs_info->balance_mutex);
3378 mutex_lock(&fs_info->volume_mutex);
3379 mutex_lock(&fs_info->balance_mutex);
3380
3381 if (fs_info->balance_ctl)
3382 __cancel_balance(fs_info);
3383
3384 mutex_unlock(&fs_info->volume_mutex);
3385 }
3386
3387 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3388 atomic_dec(&fs_info->balance_cancel_req);
3389 mutex_unlock(&fs_info->balance_mutex);
3390 return 0;
3391 }
3392
3393 /*
3394 * shrinking a device means finding all of the device extents past
3395 * the new size, and then following the back refs to the chunks.
3396 * The chunk relocation code actually frees the device extent
3397 */
3398 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3399 {
3400 struct btrfs_trans_handle *trans;
3401 struct btrfs_root *root = device->dev_root;
3402 struct btrfs_dev_extent *dev_extent = NULL;
3403 struct btrfs_path *path;
3404 u64 length;
3405 u64 chunk_tree;
3406 u64 chunk_objectid;
3407 u64 chunk_offset;
3408 int ret;
3409 int slot;
3410 int failed = 0;
3411 bool retried = false;
3412 struct extent_buffer *l;
3413 struct btrfs_key key;
3414 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3415 u64 old_total = btrfs_super_total_bytes(super_copy);
3416 u64 old_size = device->total_bytes;
3417 u64 diff = device->total_bytes - new_size;
3418
3419 if (device->is_tgtdev_for_dev_replace)
3420 return -EINVAL;
3421
3422 path = btrfs_alloc_path();
3423 if (!path)
3424 return -ENOMEM;
3425
3426 path->reada = 2;
3427
3428 lock_chunks(root);
3429
3430 device->total_bytes = new_size;
3431 if (device->writeable) {
3432 device->fs_devices->total_rw_bytes -= diff;
3433 spin_lock(&root->fs_info->free_chunk_lock);
3434 root->fs_info->free_chunk_space -= diff;
3435 spin_unlock(&root->fs_info->free_chunk_lock);
3436 }
3437 unlock_chunks(root);
3438
3439 again:
3440 key.objectid = device->devid;
3441 key.offset = (u64)-1;
3442 key.type = BTRFS_DEV_EXTENT_KEY;
3443
3444 do {
3445 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3446 if (ret < 0)
3447 goto done;
3448
3449 ret = btrfs_previous_item(root, path, 0, key.type);
3450 if (ret < 0)
3451 goto done;
3452 if (ret) {
3453 ret = 0;
3454 btrfs_release_path(path);
3455 break;
3456 }
3457
3458 l = path->nodes[0];
3459 slot = path->slots[0];
3460 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3461
3462 if (key.objectid != device->devid) {
3463 btrfs_release_path(path);
3464 break;
3465 }
3466
3467 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3468 length = btrfs_dev_extent_length(l, dev_extent);
3469
3470 if (key.offset + length <= new_size) {
3471 btrfs_release_path(path);
3472 break;
3473 }
3474
3475 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3476 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3477 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3478 btrfs_release_path(path);
3479
3480 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3481 chunk_offset);
3482 if (ret && ret != -ENOSPC)
3483 goto done;
3484 if (ret == -ENOSPC)
3485 failed++;
3486 } while (key.offset-- > 0);
3487
3488 if (failed && !retried) {
3489 failed = 0;
3490 retried = true;
3491 goto again;
3492 } else if (failed && retried) {
3493 ret = -ENOSPC;
3494 lock_chunks(root);
3495
3496 device->total_bytes = old_size;
3497 if (device->writeable)
3498 device->fs_devices->total_rw_bytes += diff;
3499 spin_lock(&root->fs_info->free_chunk_lock);
3500 root->fs_info->free_chunk_space += diff;
3501 spin_unlock(&root->fs_info->free_chunk_lock);
3502 unlock_chunks(root);
3503 goto done;
3504 }
3505
3506 /* Shrinking succeeded, else we would be at "done". */
3507 trans = btrfs_start_transaction(root, 0);
3508 if (IS_ERR(trans)) {
3509 ret = PTR_ERR(trans);
3510 goto done;
3511 }
3512
3513 lock_chunks(root);
3514
3515 device->disk_total_bytes = new_size;
3516 /* Now btrfs_update_device() will change the on-disk size. */
3517 ret = btrfs_update_device(trans, device);
3518 if (ret) {
3519 unlock_chunks(root);
3520 btrfs_end_transaction(trans, root);
3521 goto done;
3522 }
3523 WARN_ON(diff > old_total);
3524 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3525 unlock_chunks(root);
3526 btrfs_end_transaction(trans, root);
3527 done:
3528 btrfs_free_path(path);
3529 return ret;
3530 }
3531
3532 static int btrfs_add_system_chunk(struct btrfs_root *root,
3533 struct btrfs_key *key,
3534 struct btrfs_chunk *chunk, int item_size)
3535 {
3536 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3537 struct btrfs_disk_key disk_key;
3538 u32 array_size;
3539 u8 *ptr;
3540
3541 array_size = btrfs_super_sys_array_size(super_copy);
3542 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3543 return -EFBIG;
3544
3545 ptr = super_copy->sys_chunk_array + array_size;
3546 btrfs_cpu_key_to_disk(&disk_key, key);
3547 memcpy(ptr, &disk_key, sizeof(disk_key));
3548 ptr += sizeof(disk_key);
3549 memcpy(ptr, chunk, item_size);
3550 item_size += sizeof(disk_key);
3551 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3552 return 0;
3553 }
3554
3555 /*
3556 * sort the devices in descending order by max_avail, total_avail
3557 */
3558 static int btrfs_cmp_device_info(const void *a, const void *b)
3559 {
3560 const struct btrfs_device_info *di_a = a;
3561 const struct btrfs_device_info *di_b = b;
3562
3563 if (di_a->max_avail > di_b->max_avail)
3564 return -1;
3565 if (di_a->max_avail < di_b->max_avail)
3566 return 1;
3567 if (di_a->total_avail > di_b->total_avail)
3568 return -1;
3569 if (di_a->total_avail < di_b->total_avail)
3570 return 1;
3571 return 0;
3572 }
3573
3574 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3575 [BTRFS_RAID_RAID10] = {
3576 .sub_stripes = 2,
3577 .dev_stripes = 1,
3578 .devs_max = 0, /* 0 == as many as possible */
3579 .devs_min = 4,
3580 .devs_increment = 2,
3581 .ncopies = 2,
3582 },
3583 [BTRFS_RAID_RAID1] = {
3584 .sub_stripes = 1,
3585 .dev_stripes = 1,
3586 .devs_max = 2,
3587 .devs_min = 2,
3588 .devs_increment = 2,
3589 .ncopies = 2,
3590 },
3591 [BTRFS_RAID_DUP] = {
3592 .sub_stripes = 1,
3593 .dev_stripes = 2,
3594 .devs_max = 1,
3595 .devs_min = 1,
3596 .devs_increment = 1,
3597 .ncopies = 2,
3598 },
3599 [BTRFS_RAID_RAID0] = {
3600 .sub_stripes = 1,
3601 .dev_stripes = 1,
3602 .devs_max = 0,
3603 .devs_min = 2,
3604 .devs_increment = 1,
3605 .ncopies = 1,
3606 },
3607 [BTRFS_RAID_SINGLE] = {
3608 .sub_stripes = 1,
3609 .dev_stripes = 1,
3610 .devs_max = 1,
3611 .devs_min = 1,
3612 .devs_increment = 1,
3613 .ncopies = 1,
3614 },
3615 };
3616
3617 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3618 struct btrfs_root *extent_root,
3619 struct map_lookup **map_ret,
3620 u64 *num_bytes_out, u64 *stripe_size_out,
3621 u64 start, u64 type)
3622 {
3623 struct btrfs_fs_info *info = extent_root->fs_info;
3624 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3625 struct list_head *cur;
3626 struct map_lookup *map = NULL;
3627 struct extent_map_tree *em_tree;
3628 struct extent_map *em;
3629 struct btrfs_device_info *devices_info = NULL;
3630 u64 total_avail;
3631 int num_stripes; /* total number of stripes to allocate */
3632 int sub_stripes; /* sub_stripes info for map */
3633 int dev_stripes; /* stripes per dev */
3634 int devs_max; /* max devs to use */
3635 int devs_min; /* min devs needed */
3636 int devs_increment; /* ndevs has to be a multiple of this */
3637 int ncopies; /* how many copies to data has */
3638 int ret;
3639 u64 max_stripe_size;
3640 u64 max_chunk_size;
3641 u64 stripe_size;
3642 u64 num_bytes;
3643 int ndevs;
3644 int i;
3645 int j;
3646 int index;
3647
3648 BUG_ON(!alloc_profile_is_valid(type, 0));
3649
3650 if (list_empty(&fs_devices->alloc_list))
3651 return -ENOSPC;
3652
3653 index = __get_raid_index(type);
3654
3655 sub_stripes = btrfs_raid_array[index].sub_stripes;
3656 dev_stripes = btrfs_raid_array[index].dev_stripes;
3657 devs_max = btrfs_raid_array[index].devs_max;
3658 devs_min = btrfs_raid_array[index].devs_min;
3659 devs_increment = btrfs_raid_array[index].devs_increment;
3660 ncopies = btrfs_raid_array[index].ncopies;
3661
3662 if (type & BTRFS_BLOCK_GROUP_DATA) {
3663 max_stripe_size = 1024 * 1024 * 1024;
3664 max_chunk_size = 10 * max_stripe_size;
3665 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3666 /* for larger filesystems, use larger metadata chunks */
3667 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3668 max_stripe_size = 1024 * 1024 * 1024;
3669 else
3670 max_stripe_size = 256 * 1024 * 1024;
3671 max_chunk_size = max_stripe_size;
3672 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3673 max_stripe_size = 32 * 1024 * 1024;
3674 max_chunk_size = 2 * max_stripe_size;
3675 } else {
3676 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3677 type);
3678 BUG_ON(1);
3679 }
3680
3681 /* we don't want a chunk larger than 10% of writeable space */
3682 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3683 max_chunk_size);
3684
3685 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3686 GFP_NOFS);
3687 if (!devices_info)
3688 return -ENOMEM;
3689
3690 cur = fs_devices->alloc_list.next;
3691
3692 /*
3693 * in the first pass through the devices list, we gather information
3694 * about the available holes on each device.
3695 */
3696 ndevs = 0;
3697 while (cur != &fs_devices->alloc_list) {
3698 struct btrfs_device *device;
3699 u64 max_avail;
3700 u64 dev_offset;
3701
3702 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3703
3704 cur = cur->next;
3705
3706 if (!device->writeable) {
3707 WARN(1, KERN_ERR
3708 "btrfs: read-only device in alloc_list\n");
3709 continue;
3710 }
3711
3712 if (!device->in_fs_metadata ||
3713 device->is_tgtdev_for_dev_replace)
3714 continue;
3715
3716 if (device->total_bytes > device->bytes_used)
3717 total_avail = device->total_bytes - device->bytes_used;
3718 else
3719 total_avail = 0;
3720
3721 /* If there is no space on this device, skip it. */
3722 if (total_avail == 0)
3723 continue;
3724
3725 ret = find_free_dev_extent(device,
3726 max_stripe_size * dev_stripes,
3727 &dev_offset, &max_avail);
3728 if (ret && ret != -ENOSPC)
3729 goto error;
3730
3731 if (ret == 0)
3732 max_avail = max_stripe_size * dev_stripes;
3733
3734 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3735 continue;
3736
3737 devices_info[ndevs].dev_offset = dev_offset;
3738 devices_info[ndevs].max_avail = max_avail;
3739 devices_info[ndevs].total_avail = total_avail;
3740 devices_info[ndevs].dev = device;
3741 ++ndevs;
3742 WARN_ON(ndevs > fs_devices->rw_devices);
3743 }
3744
3745 /*
3746 * now sort the devices by hole size / available space
3747 */
3748 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3749 btrfs_cmp_device_info, NULL);
3750
3751 /* round down to number of usable stripes */
3752 ndevs -= ndevs % devs_increment;
3753
3754 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3755 ret = -ENOSPC;
3756 goto error;
3757 }
3758
3759 if (devs_max && ndevs > devs_max)
3760 ndevs = devs_max;
3761 /*
3762 * the primary goal is to maximize the number of stripes, so use as many
3763 * devices as possible, even if the stripes are not maximum sized.
3764 */
3765 stripe_size = devices_info[ndevs-1].max_avail;
3766 num_stripes = ndevs * dev_stripes;
3767
3768 if (stripe_size * ndevs > max_chunk_size * ncopies) {
3769 stripe_size = max_chunk_size * ncopies;
3770 do_div(stripe_size, ndevs);
3771 }
3772
3773 do_div(stripe_size, dev_stripes);
3774
3775 /* align to BTRFS_STRIPE_LEN */
3776 do_div(stripe_size, BTRFS_STRIPE_LEN);
3777 stripe_size *= BTRFS_STRIPE_LEN;
3778
3779 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3780 if (!map) {
3781 ret = -ENOMEM;
3782 goto error;
3783 }
3784 map->num_stripes = num_stripes;
3785
3786 for (i = 0; i < ndevs; ++i) {
3787 for (j = 0; j < dev_stripes; ++j) {
3788 int s = i * dev_stripes + j;
3789 map->stripes[s].dev = devices_info[i].dev;
3790 map->stripes[s].physical = devices_info[i].dev_offset +
3791 j * stripe_size;
3792 }
3793 }
3794 map->sector_size = extent_root->sectorsize;
3795 map->stripe_len = BTRFS_STRIPE_LEN;
3796 map->io_align = BTRFS_STRIPE_LEN;
3797 map->io_width = BTRFS_STRIPE_LEN;
3798 map->type = type;
3799 map->sub_stripes = sub_stripes;
3800
3801 *map_ret = map;
3802 num_bytes = stripe_size * (num_stripes / ncopies);
3803
3804 *stripe_size_out = stripe_size;
3805 *num_bytes_out = num_bytes;
3806
3807 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3808
3809 em = alloc_extent_map();
3810 if (!em) {
3811 ret = -ENOMEM;
3812 goto error;
3813 }
3814 em->bdev = (struct block_device *)map;
3815 em->start = start;
3816 em->len = num_bytes;
3817 em->block_start = 0;
3818 em->block_len = em->len;
3819
3820 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3821 write_lock(&em_tree->lock);
3822 ret = add_extent_mapping(em_tree, em);
3823 write_unlock(&em_tree->lock);
3824 free_extent_map(em);
3825 if (ret)
3826 goto error;
3827
3828 for (i = 0; i < map->num_stripes; ++i) {
3829 struct btrfs_device *device;
3830 u64 dev_offset;
3831
3832 device = map->stripes[i].dev;
3833 dev_offset = map->stripes[i].physical;
3834
3835 ret = btrfs_alloc_dev_extent(trans, device,
3836 info->chunk_root->root_key.objectid,
3837 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3838 start, dev_offset, stripe_size);
3839 if (ret)
3840 goto error_dev_extent;
3841 }
3842
3843 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3844 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3845 start, num_bytes);
3846 if (ret) {
3847 i = map->num_stripes - 1;
3848 goto error_dev_extent;
3849 }
3850
3851 kfree(devices_info);
3852 return 0;
3853
3854 error_dev_extent:
3855 for (; i >= 0; i--) {
3856 struct btrfs_device *device;
3857 int err;
3858
3859 device = map->stripes[i].dev;
3860 err = btrfs_free_dev_extent(trans, device, start);
3861 if (err) {
3862 btrfs_abort_transaction(trans, extent_root, err);
3863 break;
3864 }
3865 }
3866 error:
3867 kfree(map);
3868 kfree(devices_info);
3869 return ret;
3870 }
3871
3872 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3873 struct btrfs_root *extent_root,
3874 struct map_lookup *map, u64 chunk_offset,
3875 u64 chunk_size, u64 stripe_size)
3876 {
3877 u64 dev_offset;
3878 struct btrfs_key key;
3879 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3880 struct btrfs_device *device;
3881 struct btrfs_chunk *chunk;
3882 struct btrfs_stripe *stripe;
3883 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3884 int index = 0;
3885 int ret;
3886
3887 chunk = kzalloc(item_size, GFP_NOFS);
3888 if (!chunk)
3889 return -ENOMEM;
3890
3891 index = 0;
3892 while (index < map->num_stripes) {
3893 device = map->stripes[index].dev;
3894 device->bytes_used += stripe_size;
3895 ret = btrfs_update_device(trans, device);
3896 if (ret)
3897 goto out_free;
3898 index++;
3899 }
3900
3901 spin_lock(&extent_root->fs_info->free_chunk_lock);
3902 extent_root->fs_info->free_chunk_space -= (stripe_size *
3903 map->num_stripes);
3904 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3905
3906 index = 0;
3907 stripe = &chunk->stripe;
3908 while (index < map->num_stripes) {
3909 device = map->stripes[index].dev;
3910 dev_offset = map->stripes[index].physical;
3911
3912 btrfs_set_stack_stripe_devid(stripe, device->devid);
3913 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3914 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3915 stripe++;
3916 index++;
3917 }
3918
3919 btrfs_set_stack_chunk_length(chunk, chunk_size);
3920 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3921 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3922 btrfs_set_stack_chunk_type(chunk, map->type);
3923 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3924 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3925 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3926 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3927 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3928
3929 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3930 key.type = BTRFS_CHUNK_ITEM_KEY;
3931 key.offset = chunk_offset;
3932
3933 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3934
3935 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3936 /*
3937 * TODO: Cleanup of inserted chunk root in case of
3938 * failure.
3939 */
3940 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3941 item_size);
3942 }
3943
3944 out_free:
3945 kfree(chunk);
3946 return ret;
3947 }
3948
3949 /*
3950 * Chunk allocation falls into two parts. The first part does works
3951 * that make the new allocated chunk useable, but not do any operation
3952 * that modifies the chunk tree. The second part does the works that
3953 * require modifying the chunk tree. This division is important for the
3954 * bootstrap process of adding storage to a seed btrfs.
3955 */
3956 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3957 struct btrfs_root *extent_root, u64 type)
3958 {
3959 u64 chunk_offset;
3960 u64 chunk_size;
3961 u64 stripe_size;
3962 struct map_lookup *map;
3963 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3964 int ret;
3965
3966 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3967 &chunk_offset);
3968 if (ret)
3969 return ret;
3970
3971 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3972 &stripe_size, chunk_offset, type);
3973 if (ret)
3974 return ret;
3975
3976 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3977 chunk_size, stripe_size);
3978 if (ret)
3979 return ret;
3980 return 0;
3981 }
3982
3983 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3984 struct btrfs_root *root,
3985 struct btrfs_device *device)
3986 {
3987 u64 chunk_offset;
3988 u64 sys_chunk_offset;
3989 u64 chunk_size;
3990 u64 sys_chunk_size;
3991 u64 stripe_size;
3992 u64 sys_stripe_size;
3993 u64 alloc_profile;
3994 struct map_lookup *map;
3995 struct map_lookup *sys_map;
3996 struct btrfs_fs_info *fs_info = root->fs_info;
3997 struct btrfs_root *extent_root = fs_info->extent_root;
3998 int ret;
3999
4000 ret = find_next_chunk(fs_info->chunk_root,
4001 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4002 if (ret)
4003 return ret;
4004
4005 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4006 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4007 &stripe_size, chunk_offset, alloc_profile);
4008 if (ret)
4009 return ret;
4010
4011 sys_chunk_offset = chunk_offset + chunk_size;
4012
4013 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4014 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4015 &sys_chunk_size, &sys_stripe_size,
4016 sys_chunk_offset, alloc_profile);
4017 if (ret) {
4018 btrfs_abort_transaction(trans, root, ret);
4019 goto out;
4020 }
4021
4022 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4023 if (ret) {
4024 btrfs_abort_transaction(trans, root, ret);
4025 goto out;
4026 }
4027
4028 /*
4029 * Modifying chunk tree needs allocating new blocks from both
4030 * system block group and metadata block group. So we only can
4031 * do operations require modifying the chunk tree after both
4032 * block groups were created.
4033 */
4034 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4035 chunk_size, stripe_size);
4036 if (ret) {
4037 btrfs_abort_transaction(trans, root, ret);
4038 goto out;
4039 }
4040
4041 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4042 sys_chunk_offset, sys_chunk_size,
4043 sys_stripe_size);
4044 if (ret)
4045 btrfs_abort_transaction(trans, root, ret);
4046
4047 out:
4048
4049 return ret;
4050 }
4051
4052 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4053 {
4054 struct extent_map *em;
4055 struct map_lookup *map;
4056 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4057 int readonly = 0;
4058 int i;
4059
4060 read_lock(&map_tree->map_tree.lock);
4061 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4062 read_unlock(&map_tree->map_tree.lock);
4063 if (!em)
4064 return 1;
4065
4066 if (btrfs_test_opt(root, DEGRADED)) {
4067 free_extent_map(em);
4068 return 0;
4069 }
4070
4071 map = (struct map_lookup *)em->bdev;
4072 for (i = 0; i < map->num_stripes; i++) {
4073 if (!map->stripes[i].dev->writeable) {
4074 readonly = 1;
4075 break;
4076 }
4077 }
4078 free_extent_map(em);
4079 return readonly;
4080 }
4081
4082 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4083 {
4084 extent_map_tree_init(&tree->map_tree);
4085 }
4086
4087 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4088 {
4089 struct extent_map *em;
4090
4091 while (1) {
4092 write_lock(&tree->map_tree.lock);
4093 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4094 if (em)
4095 remove_extent_mapping(&tree->map_tree, em);
4096 write_unlock(&tree->map_tree.lock);
4097 if (!em)
4098 break;
4099 kfree(em->bdev);
4100 /* once for us */
4101 free_extent_map(em);
4102 /* once for the tree */
4103 free_extent_map(em);
4104 }
4105 }
4106
4107 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4108 {
4109 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4110 struct extent_map *em;
4111 struct map_lookup *map;
4112 struct extent_map_tree *em_tree = &map_tree->map_tree;
4113 int ret;
4114
4115 read_lock(&em_tree->lock);
4116 em = lookup_extent_mapping(em_tree, logical, len);
4117 read_unlock(&em_tree->lock);
4118 BUG_ON(!em);
4119
4120 BUG_ON(em->start > logical || em->start + em->len < logical);
4121 map = (struct map_lookup *)em->bdev;
4122 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4123 ret = map->num_stripes;
4124 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4125 ret = map->sub_stripes;
4126 else
4127 ret = 1;
4128 free_extent_map(em);
4129
4130 btrfs_dev_replace_lock(&fs_info->dev_replace);
4131 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4132 ret++;
4133 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4134
4135 return ret;
4136 }
4137
4138 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4139 struct map_lookup *map, int first, int num,
4140 int optimal, int dev_replace_is_ongoing)
4141 {
4142 int i;
4143 int tolerance;
4144 struct btrfs_device *srcdev;
4145
4146 if (dev_replace_is_ongoing &&
4147 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4148 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4149 srcdev = fs_info->dev_replace.srcdev;
4150 else
4151 srcdev = NULL;
4152
4153 /*
4154 * try to avoid the drive that is the source drive for a
4155 * dev-replace procedure, only choose it if no other non-missing
4156 * mirror is available
4157 */
4158 for (tolerance = 0; tolerance < 2; tolerance++) {
4159 if (map->stripes[optimal].dev->bdev &&
4160 (tolerance || map->stripes[optimal].dev != srcdev))
4161 return optimal;
4162 for (i = first; i < first + num; i++) {
4163 if (map->stripes[i].dev->bdev &&
4164 (tolerance || map->stripes[i].dev != srcdev))
4165 return i;
4166 }
4167 }
4168
4169 /* we couldn't find one that doesn't fail. Just return something
4170 * and the io error handling code will clean up eventually
4171 */
4172 return optimal;
4173 }
4174
4175 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4176 u64 logical, u64 *length,
4177 struct btrfs_bio **bbio_ret,
4178 int mirror_num)
4179 {
4180 struct extent_map *em;
4181 struct map_lookup *map;
4182 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4183 struct extent_map_tree *em_tree = &map_tree->map_tree;
4184 u64 offset;
4185 u64 stripe_offset;
4186 u64 stripe_end_offset;
4187 u64 stripe_nr;
4188 u64 stripe_nr_orig;
4189 u64 stripe_nr_end;
4190 int stripe_index;
4191 int i;
4192 int ret = 0;
4193 int num_stripes;
4194 int max_errors = 0;
4195 struct btrfs_bio *bbio = NULL;
4196 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4197 int dev_replace_is_ongoing = 0;
4198 int num_alloc_stripes;
4199 int patch_the_first_stripe_for_dev_replace = 0;
4200 u64 physical_to_patch_in_first_stripe = 0;
4201
4202 read_lock(&em_tree->lock);
4203 em = lookup_extent_mapping(em_tree, logical, *length);
4204 read_unlock(&em_tree->lock);
4205
4206 if (!em) {
4207 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4208 (unsigned long long)logical,
4209 (unsigned long long)*length);
4210 BUG();
4211 }
4212
4213 BUG_ON(em->start > logical || em->start + em->len < logical);
4214 map = (struct map_lookup *)em->bdev;
4215 offset = logical - em->start;
4216
4217 stripe_nr = offset;
4218 /*
4219 * stripe_nr counts the total number of stripes we have to stride
4220 * to get to this block
4221 */
4222 do_div(stripe_nr, map->stripe_len);
4223
4224 stripe_offset = stripe_nr * map->stripe_len;
4225 BUG_ON(offset < stripe_offset);
4226
4227 /* stripe_offset is the offset of this block in its stripe*/
4228 stripe_offset = offset - stripe_offset;
4229
4230 if (rw & REQ_DISCARD)
4231 *length = min_t(u64, em->len - offset, *length);
4232 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4233 /* we limit the length of each bio to what fits in a stripe */
4234 *length = min_t(u64, em->len - offset,
4235 map->stripe_len - stripe_offset);
4236 } else {
4237 *length = em->len - offset;
4238 }
4239
4240 if (!bbio_ret)
4241 goto out;
4242
4243 btrfs_dev_replace_lock(dev_replace);
4244 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4245 if (!dev_replace_is_ongoing)
4246 btrfs_dev_replace_unlock(dev_replace);
4247
4248 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4249 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4250 dev_replace->tgtdev != NULL) {
4251 /*
4252 * in dev-replace case, for repair case (that's the only
4253 * case where the mirror is selected explicitly when
4254 * calling btrfs_map_block), blocks left of the left cursor
4255 * can also be read from the target drive.
4256 * For REQ_GET_READ_MIRRORS, the target drive is added as
4257 * the last one to the array of stripes. For READ, it also
4258 * needs to be supported using the same mirror number.
4259 * If the requested block is not left of the left cursor,
4260 * EIO is returned. This can happen because btrfs_num_copies()
4261 * returns one more in the dev-replace case.
4262 */
4263 u64 tmp_length = *length;
4264 struct btrfs_bio *tmp_bbio = NULL;
4265 int tmp_num_stripes;
4266 u64 srcdev_devid = dev_replace->srcdev->devid;
4267 int index_srcdev = 0;
4268 int found = 0;
4269 u64 physical_of_found = 0;
4270
4271 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4272 logical, &tmp_length, &tmp_bbio, 0);
4273 if (ret) {
4274 WARN_ON(tmp_bbio != NULL);
4275 goto out;
4276 }
4277
4278 tmp_num_stripes = tmp_bbio->num_stripes;
4279 if (mirror_num > tmp_num_stripes) {
4280 /*
4281 * REQ_GET_READ_MIRRORS does not contain this
4282 * mirror, that means that the requested area
4283 * is not left of the left cursor
4284 */
4285 ret = -EIO;
4286 kfree(tmp_bbio);
4287 goto out;
4288 }
4289
4290 /*
4291 * process the rest of the function using the mirror_num
4292 * of the source drive. Therefore look it up first.
4293 * At the end, patch the device pointer to the one of the
4294 * target drive.
4295 */
4296 for (i = 0; i < tmp_num_stripes; i++) {
4297 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4298 /*
4299 * In case of DUP, in order to keep it
4300 * simple, only add the mirror with the
4301 * lowest physical address
4302 */
4303 if (found &&
4304 physical_of_found <=
4305 tmp_bbio->stripes[i].physical)
4306 continue;
4307 index_srcdev = i;
4308 found = 1;
4309 physical_of_found =
4310 tmp_bbio->stripes[i].physical;
4311 }
4312 }
4313
4314 if (found) {
4315 mirror_num = index_srcdev + 1;
4316 patch_the_first_stripe_for_dev_replace = 1;
4317 physical_to_patch_in_first_stripe = physical_of_found;
4318 } else {
4319 WARN_ON(1);
4320 ret = -EIO;
4321 kfree(tmp_bbio);
4322 goto out;
4323 }
4324
4325 kfree(tmp_bbio);
4326 } else if (mirror_num > map->num_stripes) {
4327 mirror_num = 0;
4328 }
4329
4330 num_stripes = 1;
4331 stripe_index = 0;
4332 stripe_nr_orig = stripe_nr;
4333 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4334 (~(map->stripe_len - 1));
4335 do_div(stripe_nr_end, map->stripe_len);
4336 stripe_end_offset = stripe_nr_end * map->stripe_len -
4337 (offset + *length);
4338 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4339 if (rw & REQ_DISCARD)
4340 num_stripes = min_t(u64, map->num_stripes,
4341 stripe_nr_end - stripe_nr_orig);
4342 stripe_index = do_div(stripe_nr, map->num_stripes);
4343 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4344 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4345 num_stripes = map->num_stripes;
4346 else if (mirror_num)
4347 stripe_index = mirror_num - 1;
4348 else {
4349 stripe_index = find_live_mirror(fs_info, map, 0,
4350 map->num_stripes,
4351 current->pid % map->num_stripes,
4352 dev_replace_is_ongoing);
4353 mirror_num = stripe_index + 1;
4354 }
4355
4356 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4357 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4358 num_stripes = map->num_stripes;
4359 } else if (mirror_num) {
4360 stripe_index = mirror_num - 1;
4361 } else {
4362 mirror_num = 1;
4363 }
4364
4365 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4366 int factor = map->num_stripes / map->sub_stripes;
4367
4368 stripe_index = do_div(stripe_nr, factor);
4369 stripe_index *= map->sub_stripes;
4370
4371 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4372 num_stripes = map->sub_stripes;
4373 else if (rw & REQ_DISCARD)
4374 num_stripes = min_t(u64, map->sub_stripes *
4375 (stripe_nr_end - stripe_nr_orig),
4376 map->num_stripes);
4377 else if (mirror_num)
4378 stripe_index += mirror_num - 1;
4379 else {
4380 int old_stripe_index = stripe_index;
4381 stripe_index = find_live_mirror(fs_info, map,
4382 stripe_index,
4383 map->sub_stripes, stripe_index +
4384 current->pid % map->sub_stripes,
4385 dev_replace_is_ongoing);
4386 mirror_num = stripe_index - old_stripe_index + 1;
4387 }
4388 } else {
4389 /*
4390 * after this do_div call, stripe_nr is the number of stripes
4391 * on this device we have to walk to find the data, and
4392 * stripe_index is the number of our device in the stripe array
4393 */
4394 stripe_index = do_div(stripe_nr, map->num_stripes);
4395 mirror_num = stripe_index + 1;
4396 }
4397 BUG_ON(stripe_index >= map->num_stripes);
4398
4399 num_alloc_stripes = num_stripes;
4400 if (dev_replace_is_ongoing) {
4401 if (rw & (REQ_WRITE | REQ_DISCARD))
4402 num_alloc_stripes <<= 1;
4403 if (rw & REQ_GET_READ_MIRRORS)
4404 num_alloc_stripes++;
4405 }
4406 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4407 if (!bbio) {
4408 ret = -ENOMEM;
4409 goto out;
4410 }
4411 atomic_set(&bbio->error, 0);
4412
4413 if (rw & REQ_DISCARD) {
4414 int factor = 0;
4415 int sub_stripes = 0;
4416 u64 stripes_per_dev = 0;
4417 u32 remaining_stripes = 0;
4418 u32 last_stripe = 0;
4419
4420 if (map->type &
4421 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4422 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4423 sub_stripes = 1;
4424 else
4425 sub_stripes = map->sub_stripes;
4426
4427 factor = map->num_stripes / sub_stripes;
4428 stripes_per_dev = div_u64_rem(stripe_nr_end -
4429 stripe_nr_orig,
4430 factor,
4431 &remaining_stripes);
4432 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4433 last_stripe *= sub_stripes;
4434 }
4435
4436 for (i = 0; i < num_stripes; i++) {
4437 bbio->stripes[i].physical =
4438 map->stripes[stripe_index].physical +
4439 stripe_offset + stripe_nr * map->stripe_len;
4440 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4441
4442 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4443 BTRFS_BLOCK_GROUP_RAID10)) {
4444 bbio->stripes[i].length = stripes_per_dev *
4445 map->stripe_len;
4446
4447 if (i / sub_stripes < remaining_stripes)
4448 bbio->stripes[i].length +=
4449 map->stripe_len;
4450
4451 /*
4452 * Special for the first stripe and
4453 * the last stripe:
4454 *
4455 * |-------|...|-------|
4456 * |----------|
4457 * off end_off
4458 */
4459 if (i < sub_stripes)
4460 bbio->stripes[i].length -=
4461 stripe_offset;
4462
4463 if (stripe_index >= last_stripe &&
4464 stripe_index <= (last_stripe +
4465 sub_stripes - 1))
4466 bbio->stripes[i].length -=
4467 stripe_end_offset;
4468
4469 if (i == sub_stripes - 1)
4470 stripe_offset = 0;
4471 } else
4472 bbio->stripes[i].length = *length;
4473
4474 stripe_index++;
4475 if (stripe_index == map->num_stripes) {
4476 /* This could only happen for RAID0/10 */
4477 stripe_index = 0;
4478 stripe_nr++;
4479 }
4480 }
4481 } else {
4482 for (i = 0; i < num_stripes; i++) {
4483 bbio->stripes[i].physical =
4484 map->stripes[stripe_index].physical +
4485 stripe_offset +
4486 stripe_nr * map->stripe_len;
4487 bbio->stripes[i].dev =
4488 map->stripes[stripe_index].dev;
4489 stripe_index++;
4490 }
4491 }
4492
4493 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4494 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4495 BTRFS_BLOCK_GROUP_RAID10 |
4496 BTRFS_BLOCK_GROUP_DUP)) {
4497 max_errors = 1;
4498 }
4499 }
4500
4501 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4502 dev_replace->tgtdev != NULL) {
4503 int index_where_to_add;
4504 u64 srcdev_devid = dev_replace->srcdev->devid;
4505
4506 /*
4507 * duplicate the write operations while the dev replace
4508 * procedure is running. Since the copying of the old disk
4509 * to the new disk takes place at run time while the
4510 * filesystem is mounted writable, the regular write
4511 * operations to the old disk have to be duplicated to go
4512 * to the new disk as well.
4513 * Note that device->missing is handled by the caller, and
4514 * that the write to the old disk is already set up in the
4515 * stripes array.
4516 */
4517 index_where_to_add = num_stripes;
4518 for (i = 0; i < num_stripes; i++) {
4519 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4520 /* write to new disk, too */
4521 struct btrfs_bio_stripe *new =
4522 bbio->stripes + index_where_to_add;
4523 struct btrfs_bio_stripe *old =
4524 bbio->stripes + i;
4525
4526 new->physical = old->physical;
4527 new->length = old->length;
4528 new->dev = dev_replace->tgtdev;
4529 index_where_to_add++;
4530 max_errors++;
4531 }
4532 }
4533 num_stripes = index_where_to_add;
4534 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4535 dev_replace->tgtdev != NULL) {
4536 u64 srcdev_devid = dev_replace->srcdev->devid;
4537 int index_srcdev = 0;
4538 int found = 0;
4539 u64 physical_of_found = 0;
4540
4541 /*
4542 * During the dev-replace procedure, the target drive can
4543 * also be used to read data in case it is needed to repair
4544 * a corrupt block elsewhere. This is possible if the
4545 * requested area is left of the left cursor. In this area,
4546 * the target drive is a full copy of the source drive.
4547 */
4548 for (i = 0; i < num_stripes; i++) {
4549 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4550 /*
4551 * In case of DUP, in order to keep it
4552 * simple, only add the mirror with the
4553 * lowest physical address
4554 */
4555 if (found &&
4556 physical_of_found <=
4557 bbio->stripes[i].physical)
4558 continue;
4559 index_srcdev = i;
4560 found = 1;
4561 physical_of_found = bbio->stripes[i].physical;
4562 }
4563 }
4564 if (found) {
4565 u64 length = map->stripe_len;
4566
4567 if (physical_of_found + length <=
4568 dev_replace->cursor_left) {
4569 struct btrfs_bio_stripe *tgtdev_stripe =
4570 bbio->stripes + num_stripes;
4571
4572 tgtdev_stripe->physical = physical_of_found;
4573 tgtdev_stripe->length =
4574 bbio->stripes[index_srcdev].length;
4575 tgtdev_stripe->dev = dev_replace->tgtdev;
4576
4577 num_stripes++;
4578 }
4579 }
4580 }
4581
4582 *bbio_ret = bbio;
4583 bbio->num_stripes = num_stripes;
4584 bbio->max_errors = max_errors;
4585 bbio->mirror_num = mirror_num;
4586
4587 /*
4588 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4589 * mirror_num == num_stripes + 1 && dev_replace target drive is
4590 * available as a mirror
4591 */
4592 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4593 WARN_ON(num_stripes > 1);
4594 bbio->stripes[0].dev = dev_replace->tgtdev;
4595 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4596 bbio->mirror_num = map->num_stripes + 1;
4597 }
4598 out:
4599 if (dev_replace_is_ongoing)
4600 btrfs_dev_replace_unlock(dev_replace);
4601 free_extent_map(em);
4602 return ret;
4603 }
4604
4605 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4606 u64 logical, u64 *length,
4607 struct btrfs_bio **bbio_ret, int mirror_num)
4608 {
4609 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4610 mirror_num);
4611 }
4612
4613 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4614 u64 chunk_start, u64 physical, u64 devid,
4615 u64 **logical, int *naddrs, int *stripe_len)
4616 {
4617 struct extent_map_tree *em_tree = &map_tree->map_tree;
4618 struct extent_map *em;
4619 struct map_lookup *map;
4620 u64 *buf;
4621 u64 bytenr;
4622 u64 length;
4623 u64 stripe_nr;
4624 int i, j, nr = 0;
4625
4626 read_lock(&em_tree->lock);
4627 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4628 read_unlock(&em_tree->lock);
4629
4630 BUG_ON(!em || em->start != chunk_start);
4631 map = (struct map_lookup *)em->bdev;
4632
4633 length = em->len;
4634 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4635 do_div(length, map->num_stripes / map->sub_stripes);
4636 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4637 do_div(length, map->num_stripes);
4638
4639 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4640 BUG_ON(!buf); /* -ENOMEM */
4641
4642 for (i = 0; i < map->num_stripes; i++) {
4643 if (devid && map->stripes[i].dev->devid != devid)
4644 continue;
4645 if (map->stripes[i].physical > physical ||
4646 map->stripes[i].physical + length <= physical)
4647 continue;
4648
4649 stripe_nr = physical - map->stripes[i].physical;
4650 do_div(stripe_nr, map->stripe_len);
4651
4652 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4653 stripe_nr = stripe_nr * map->num_stripes + i;
4654 do_div(stripe_nr, map->sub_stripes);
4655 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4656 stripe_nr = stripe_nr * map->num_stripes + i;
4657 }
4658 bytenr = chunk_start + stripe_nr * map->stripe_len;
4659 WARN_ON(nr >= map->num_stripes);
4660 for (j = 0; j < nr; j++) {
4661 if (buf[j] == bytenr)
4662 break;
4663 }
4664 if (j == nr) {
4665 WARN_ON(nr >= map->num_stripes);
4666 buf[nr++] = bytenr;
4667 }
4668 }
4669
4670 *logical = buf;
4671 *naddrs = nr;
4672 *stripe_len = map->stripe_len;
4673
4674 free_extent_map(em);
4675 return 0;
4676 }
4677
4678 static void *merge_stripe_index_into_bio_private(void *bi_private,
4679 unsigned int stripe_index)
4680 {
4681 /*
4682 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4683 * at most 1.
4684 * The alternative solution (instead of stealing bits from the
4685 * pointer) would be to allocate an intermediate structure
4686 * that contains the old private pointer plus the stripe_index.
4687 */
4688 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4689 BUG_ON(stripe_index > 3);
4690 return (void *)(((uintptr_t)bi_private) | stripe_index);
4691 }
4692
4693 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4694 {
4695 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4696 }
4697
4698 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4699 {
4700 return (unsigned int)((uintptr_t)bi_private) & 3;
4701 }
4702
4703 static void btrfs_end_bio(struct bio *bio, int err)
4704 {
4705 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4706 int is_orig_bio = 0;
4707
4708 if (err) {
4709 atomic_inc(&bbio->error);
4710 if (err == -EIO || err == -EREMOTEIO) {
4711 unsigned int stripe_index =
4712 extract_stripe_index_from_bio_private(
4713 bio->bi_private);
4714 struct btrfs_device *dev;
4715
4716 BUG_ON(stripe_index >= bbio->num_stripes);
4717 dev = bbio->stripes[stripe_index].dev;
4718 if (dev->bdev) {
4719 if (bio->bi_rw & WRITE)
4720 btrfs_dev_stat_inc(dev,
4721 BTRFS_DEV_STAT_WRITE_ERRS);
4722 else
4723 btrfs_dev_stat_inc(dev,
4724 BTRFS_DEV_STAT_READ_ERRS);
4725 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4726 btrfs_dev_stat_inc(dev,
4727 BTRFS_DEV_STAT_FLUSH_ERRS);
4728 btrfs_dev_stat_print_on_error(dev);
4729 }
4730 }
4731 }
4732
4733 if (bio == bbio->orig_bio)
4734 is_orig_bio = 1;
4735
4736 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4737 if (!is_orig_bio) {
4738 bio_put(bio);
4739 bio = bbio->orig_bio;
4740 }
4741 bio->bi_private = bbio->private;
4742 bio->bi_end_io = bbio->end_io;
4743 bio->bi_bdev = (struct block_device *)
4744 (unsigned long)bbio->mirror_num;
4745 /* only send an error to the higher layers if it is
4746 * beyond the tolerance of the multi-bio
4747 */
4748 if (atomic_read(&bbio->error) > bbio->max_errors) {
4749 err = -EIO;
4750 } else {
4751 /*
4752 * this bio is actually up to date, we didn't
4753 * go over the max number of errors
4754 */
4755 set_bit(BIO_UPTODATE, &bio->bi_flags);
4756 err = 0;
4757 }
4758 kfree(bbio);
4759
4760 bio_endio(bio, err);
4761 } else if (!is_orig_bio) {
4762 bio_put(bio);
4763 }
4764 }
4765
4766 struct async_sched {
4767 struct bio *bio;
4768 int rw;
4769 struct btrfs_fs_info *info;
4770 struct btrfs_work work;
4771 };
4772
4773 /*
4774 * see run_scheduled_bios for a description of why bios are collected for
4775 * async submit.
4776 *
4777 * This will add one bio to the pending list for a device and make sure
4778 * the work struct is scheduled.
4779 */
4780 static noinline void schedule_bio(struct btrfs_root *root,
4781 struct btrfs_device *device,
4782 int rw, struct bio *bio)
4783 {
4784 int should_queue = 1;
4785 struct btrfs_pending_bios *pending_bios;
4786
4787 /* don't bother with additional async steps for reads, right now */
4788 if (!(rw & REQ_WRITE)) {
4789 bio_get(bio);
4790 btrfsic_submit_bio(rw, bio);
4791 bio_put(bio);
4792 return;
4793 }
4794
4795 /*
4796 * nr_async_bios allows us to reliably return congestion to the
4797 * higher layers. Otherwise, the async bio makes it appear we have
4798 * made progress against dirty pages when we've really just put it
4799 * on a queue for later
4800 */
4801 atomic_inc(&root->fs_info->nr_async_bios);
4802 WARN_ON(bio->bi_next);
4803 bio->bi_next = NULL;
4804 bio->bi_rw |= rw;
4805
4806 spin_lock(&device->io_lock);
4807 if (bio->bi_rw & REQ_SYNC)
4808 pending_bios = &device->pending_sync_bios;
4809 else
4810 pending_bios = &device->pending_bios;
4811
4812 if (pending_bios->tail)
4813 pending_bios->tail->bi_next = bio;
4814
4815 pending_bios->tail = bio;
4816 if (!pending_bios->head)
4817 pending_bios->head = bio;
4818 if (device->running_pending)
4819 should_queue = 0;
4820
4821 spin_unlock(&device->io_lock);
4822
4823 if (should_queue)
4824 btrfs_queue_worker(&root->fs_info->submit_workers,
4825 &device->work);
4826 }
4827
4828 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4829 sector_t sector)
4830 {
4831 struct bio_vec *prev;
4832 struct request_queue *q = bdev_get_queue(bdev);
4833 unsigned short max_sectors = queue_max_sectors(q);
4834 struct bvec_merge_data bvm = {
4835 .bi_bdev = bdev,
4836 .bi_sector = sector,
4837 .bi_rw = bio->bi_rw,
4838 };
4839
4840 if (bio->bi_vcnt == 0) {
4841 WARN_ON(1);
4842 return 1;
4843 }
4844
4845 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4846 if ((bio->bi_size >> 9) > max_sectors)
4847 return 0;
4848
4849 if (!q->merge_bvec_fn)
4850 return 1;
4851
4852 bvm.bi_size = bio->bi_size - prev->bv_len;
4853 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4854 return 0;
4855 return 1;
4856 }
4857
4858 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4859 struct bio *bio, u64 physical, int dev_nr,
4860 int rw, int async)
4861 {
4862 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4863
4864 bio->bi_private = bbio;
4865 bio->bi_private = merge_stripe_index_into_bio_private(
4866 bio->bi_private, (unsigned int)dev_nr);
4867 bio->bi_end_io = btrfs_end_bio;
4868 bio->bi_sector = physical >> 9;
4869 #ifdef DEBUG
4870 {
4871 struct rcu_string *name;
4872
4873 rcu_read_lock();
4874 name = rcu_dereference(dev->name);
4875 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4876 "(%s id %llu), size=%u\n", rw,
4877 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4878 name->str, dev->devid, bio->bi_size);
4879 rcu_read_unlock();
4880 }
4881 #endif
4882 bio->bi_bdev = dev->bdev;
4883 if (async)
4884 schedule_bio(root, dev, rw, bio);
4885 else
4886 btrfsic_submit_bio(rw, bio);
4887 }
4888
4889 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4890 struct bio *first_bio, struct btrfs_device *dev,
4891 int dev_nr, int rw, int async)
4892 {
4893 struct bio_vec *bvec = first_bio->bi_io_vec;
4894 struct bio *bio;
4895 int nr_vecs = bio_get_nr_vecs(dev->bdev);
4896 u64 physical = bbio->stripes[dev_nr].physical;
4897
4898 again:
4899 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4900 if (!bio)
4901 return -ENOMEM;
4902
4903 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4904 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4905 bvec->bv_offset) < bvec->bv_len) {
4906 u64 len = bio->bi_size;
4907
4908 atomic_inc(&bbio->stripes_pending);
4909 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4910 rw, async);
4911 physical += len;
4912 goto again;
4913 }
4914 bvec++;
4915 }
4916
4917 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4918 return 0;
4919 }
4920
4921 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4922 {
4923 atomic_inc(&bbio->error);
4924 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4925 bio->bi_private = bbio->private;
4926 bio->bi_end_io = bbio->end_io;
4927 bio->bi_bdev = (struct block_device *)
4928 (unsigned long)bbio->mirror_num;
4929 bio->bi_sector = logical >> 9;
4930 kfree(bbio);
4931 bio_endio(bio, -EIO);
4932 }
4933 }
4934
4935 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4936 int mirror_num, int async_submit)
4937 {
4938 struct btrfs_device *dev;
4939 struct bio *first_bio = bio;
4940 u64 logical = (u64)bio->bi_sector << 9;
4941 u64 length = 0;
4942 u64 map_length;
4943 int ret;
4944 int dev_nr = 0;
4945 int total_devs = 1;
4946 struct btrfs_bio *bbio = NULL;
4947
4948 length = bio->bi_size;
4949 map_length = length;
4950
4951 ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4952 mirror_num);
4953 if (ret)
4954 return ret;
4955
4956 total_devs = bbio->num_stripes;
4957 if (map_length < length) {
4958 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4959 "len %llu\n", (unsigned long long)logical,
4960 (unsigned long long)length,
4961 (unsigned long long)map_length);
4962 BUG();
4963 }
4964
4965 bbio->orig_bio = first_bio;
4966 bbio->private = first_bio->bi_private;
4967 bbio->end_io = first_bio->bi_end_io;
4968 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4969
4970 while (dev_nr < total_devs) {
4971 dev = bbio->stripes[dev_nr].dev;
4972 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4973 bbio_error(bbio, first_bio, logical);
4974 dev_nr++;
4975 continue;
4976 }
4977
4978 /*
4979 * Check and see if we're ok with this bio based on it's size
4980 * and offset with the given device.
4981 */
4982 if (!bio_size_ok(dev->bdev, first_bio,
4983 bbio->stripes[dev_nr].physical >> 9)) {
4984 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4985 dev_nr, rw, async_submit);
4986 BUG_ON(ret);
4987 dev_nr++;
4988 continue;
4989 }
4990
4991 if (dev_nr < total_devs - 1) {
4992 bio = bio_clone(first_bio, GFP_NOFS);
4993 BUG_ON(!bio); /* -ENOMEM */
4994 } else {
4995 bio = first_bio;
4996 }
4997
4998 submit_stripe_bio(root, bbio, bio,
4999 bbio->stripes[dev_nr].physical, dev_nr, rw,
5000 async_submit);
5001 dev_nr++;
5002 }
5003 return 0;
5004 }
5005
5006 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5007 u8 *uuid, u8 *fsid)
5008 {
5009 struct btrfs_device *device;
5010 struct btrfs_fs_devices *cur_devices;
5011
5012 cur_devices = fs_info->fs_devices;
5013 while (cur_devices) {
5014 if (!fsid ||
5015 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5016 device = __find_device(&cur_devices->devices,
5017 devid, uuid);
5018 if (device)
5019 return device;
5020 }
5021 cur_devices = cur_devices->seed;
5022 }
5023 return NULL;
5024 }
5025
5026 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5027 u64 devid, u8 *dev_uuid)
5028 {
5029 struct btrfs_device *device;
5030 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5031
5032 device = kzalloc(sizeof(*device), GFP_NOFS);
5033 if (!device)
5034 return NULL;
5035 list_add(&device->dev_list,
5036 &fs_devices->devices);
5037 device->dev_root = root->fs_info->dev_root;
5038 device->devid = devid;
5039 device->work.func = pending_bios_fn;
5040 device->fs_devices = fs_devices;
5041 device->missing = 1;
5042 fs_devices->num_devices++;
5043 fs_devices->missing_devices++;
5044 spin_lock_init(&device->io_lock);
5045 INIT_LIST_HEAD(&device->dev_alloc_list);
5046 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5047 return device;
5048 }
5049
5050 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5051 struct extent_buffer *leaf,
5052 struct btrfs_chunk *chunk)
5053 {
5054 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5055 struct map_lookup *map;
5056 struct extent_map *em;
5057 u64 logical;
5058 u64 length;
5059 u64 devid;
5060 u8 uuid[BTRFS_UUID_SIZE];
5061 int num_stripes;
5062 int ret;
5063 int i;
5064
5065 logical = key->offset;
5066 length = btrfs_chunk_length(leaf, chunk);
5067
5068 read_lock(&map_tree->map_tree.lock);
5069 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5070 read_unlock(&map_tree->map_tree.lock);
5071
5072 /* already mapped? */
5073 if (em && em->start <= logical && em->start + em->len > logical) {
5074 free_extent_map(em);
5075 return 0;
5076 } else if (em) {
5077 free_extent_map(em);
5078 }
5079
5080 em = alloc_extent_map();
5081 if (!em)
5082 return -ENOMEM;
5083 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5084 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5085 if (!map) {
5086 free_extent_map(em);
5087 return -ENOMEM;
5088 }
5089
5090 em->bdev = (struct block_device *)map;
5091 em->start = logical;
5092 em->len = length;
5093 em->orig_start = 0;
5094 em->block_start = 0;
5095 em->block_len = em->len;
5096
5097 map->num_stripes = num_stripes;
5098 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5099 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5100 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5101 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5102 map->type = btrfs_chunk_type(leaf, chunk);
5103 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5104 for (i = 0; i < num_stripes; i++) {
5105 map->stripes[i].physical =
5106 btrfs_stripe_offset_nr(leaf, chunk, i);
5107 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5108 read_extent_buffer(leaf, uuid, (unsigned long)
5109 btrfs_stripe_dev_uuid_nr(chunk, i),
5110 BTRFS_UUID_SIZE);
5111 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5112 uuid, NULL);
5113 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5114 kfree(map);
5115 free_extent_map(em);
5116 return -EIO;
5117 }
5118 if (!map->stripes[i].dev) {
5119 map->stripes[i].dev =
5120 add_missing_dev(root, devid, uuid);
5121 if (!map->stripes[i].dev) {
5122 kfree(map);
5123 free_extent_map(em);
5124 return -EIO;
5125 }
5126 }
5127 map->stripes[i].dev->in_fs_metadata = 1;
5128 }
5129
5130 write_lock(&map_tree->map_tree.lock);
5131 ret = add_extent_mapping(&map_tree->map_tree, em);
5132 write_unlock(&map_tree->map_tree.lock);
5133 BUG_ON(ret); /* Tree corruption */
5134 free_extent_map(em);
5135
5136 return 0;
5137 }
5138
5139 static void fill_device_from_item(struct extent_buffer *leaf,
5140 struct btrfs_dev_item *dev_item,
5141 struct btrfs_device *device)
5142 {
5143 unsigned long ptr;
5144
5145 device->devid = btrfs_device_id(leaf, dev_item);
5146 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5147 device->total_bytes = device->disk_total_bytes;
5148 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5149 device->type = btrfs_device_type(leaf, dev_item);
5150 device->io_align = btrfs_device_io_align(leaf, dev_item);
5151 device->io_width = btrfs_device_io_width(leaf, dev_item);
5152 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5153 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5154 device->is_tgtdev_for_dev_replace = 0;
5155
5156 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5157 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5158 }
5159
5160 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5161 {
5162 struct btrfs_fs_devices *fs_devices;
5163 int ret;
5164
5165 BUG_ON(!mutex_is_locked(&uuid_mutex));
5166
5167 fs_devices = root->fs_info->fs_devices->seed;
5168 while (fs_devices) {
5169 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5170 ret = 0;
5171 goto out;
5172 }
5173 fs_devices = fs_devices->seed;
5174 }
5175
5176 fs_devices = find_fsid(fsid);
5177 if (!fs_devices) {
5178 ret = -ENOENT;
5179 goto out;
5180 }
5181
5182 fs_devices = clone_fs_devices(fs_devices);
5183 if (IS_ERR(fs_devices)) {
5184 ret = PTR_ERR(fs_devices);
5185 goto out;
5186 }
5187
5188 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5189 root->fs_info->bdev_holder);
5190 if (ret) {
5191 free_fs_devices(fs_devices);
5192 goto out;
5193 }
5194
5195 if (!fs_devices->seeding) {
5196 __btrfs_close_devices(fs_devices);
5197 free_fs_devices(fs_devices);
5198 ret = -EINVAL;
5199 goto out;
5200 }
5201
5202 fs_devices->seed = root->fs_info->fs_devices->seed;
5203 root->fs_info->fs_devices->seed = fs_devices;
5204 out:
5205 return ret;
5206 }
5207
5208 static int read_one_dev(struct btrfs_root *root,
5209 struct extent_buffer *leaf,
5210 struct btrfs_dev_item *dev_item)
5211 {
5212 struct btrfs_device *device;
5213 u64 devid;
5214 int ret;
5215 u8 fs_uuid[BTRFS_UUID_SIZE];
5216 u8 dev_uuid[BTRFS_UUID_SIZE];
5217
5218 devid = btrfs_device_id(leaf, dev_item);
5219 read_extent_buffer(leaf, dev_uuid,
5220 (unsigned long)btrfs_device_uuid(dev_item),
5221 BTRFS_UUID_SIZE);
5222 read_extent_buffer(leaf, fs_uuid,
5223 (unsigned long)btrfs_device_fsid(dev_item),
5224 BTRFS_UUID_SIZE);
5225
5226 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5227 ret = open_seed_devices(root, fs_uuid);
5228 if (ret && !btrfs_test_opt(root, DEGRADED))
5229 return ret;
5230 }
5231
5232 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5233 if (!device || !device->bdev) {
5234 if (!btrfs_test_opt(root, DEGRADED))
5235 return -EIO;
5236
5237 if (!device) {
5238 printk(KERN_WARNING "warning devid %llu missing\n",
5239 (unsigned long long)devid);
5240 device = add_missing_dev(root, devid, dev_uuid);
5241 if (!device)
5242 return -ENOMEM;
5243 } else if (!device->missing) {
5244 /*
5245 * this happens when a device that was properly setup
5246 * in the device info lists suddenly goes bad.
5247 * device->bdev is NULL, and so we have to set
5248 * device->missing to one here
5249 */
5250 root->fs_info->fs_devices->missing_devices++;
5251 device->missing = 1;
5252 }
5253 }
5254
5255 if (device->fs_devices != root->fs_info->fs_devices) {
5256 BUG_ON(device->writeable);
5257 if (device->generation !=
5258 btrfs_device_generation(leaf, dev_item))
5259 return -EINVAL;
5260 }
5261
5262 fill_device_from_item(leaf, dev_item, device);
5263 device->dev_root = root->fs_info->dev_root;
5264 device->in_fs_metadata = 1;
5265 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5266 device->fs_devices->total_rw_bytes += device->total_bytes;
5267 spin_lock(&root->fs_info->free_chunk_lock);
5268 root->fs_info->free_chunk_space += device->total_bytes -
5269 device->bytes_used;
5270 spin_unlock(&root->fs_info->free_chunk_lock);
5271 }
5272 ret = 0;
5273 return ret;
5274 }
5275
5276 int btrfs_read_sys_array(struct btrfs_root *root)
5277 {
5278 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5279 struct extent_buffer *sb;
5280 struct btrfs_disk_key *disk_key;
5281 struct btrfs_chunk *chunk;
5282 u8 *ptr;
5283 unsigned long sb_ptr;
5284 int ret = 0;
5285 u32 num_stripes;
5286 u32 array_size;
5287 u32 len = 0;
5288 u32 cur;
5289 struct btrfs_key key;
5290
5291 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5292 BTRFS_SUPER_INFO_SIZE);
5293 if (!sb)
5294 return -ENOMEM;
5295 btrfs_set_buffer_uptodate(sb);
5296 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5297 /*
5298 * The sb extent buffer is artifical and just used to read the system array.
5299 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5300 * pages up-to-date when the page is larger: extent does not cover the
5301 * whole page and consequently check_page_uptodate does not find all
5302 * the page's extents up-to-date (the hole beyond sb),
5303 * write_extent_buffer then triggers a WARN_ON.
5304 *
5305 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5306 * but sb spans only this function. Add an explicit SetPageUptodate call
5307 * to silence the warning eg. on PowerPC 64.
5308 */
5309 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5310 SetPageUptodate(sb->pages[0]);
5311
5312 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5313 array_size = btrfs_super_sys_array_size(super_copy);
5314
5315 ptr = super_copy->sys_chunk_array;
5316 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5317 cur = 0;
5318
5319 while (cur < array_size) {
5320 disk_key = (struct btrfs_disk_key *)ptr;
5321 btrfs_disk_key_to_cpu(&key, disk_key);
5322
5323 len = sizeof(*disk_key); ptr += len;
5324 sb_ptr += len;
5325 cur += len;
5326
5327 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5328 chunk = (struct btrfs_chunk *)sb_ptr;
5329 ret = read_one_chunk(root, &key, sb, chunk);
5330 if (ret)
5331 break;
5332 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5333 len = btrfs_chunk_item_size(num_stripes);
5334 } else {
5335 ret = -EIO;
5336 break;
5337 }
5338 ptr += len;
5339 sb_ptr += len;
5340 cur += len;
5341 }
5342 free_extent_buffer(sb);
5343 return ret;
5344 }
5345
5346 int btrfs_read_chunk_tree(struct btrfs_root *root)
5347 {
5348 struct btrfs_path *path;
5349 struct extent_buffer *leaf;
5350 struct btrfs_key key;
5351 struct btrfs_key found_key;
5352 int ret;
5353 int slot;
5354
5355 root = root->fs_info->chunk_root;
5356
5357 path = btrfs_alloc_path();
5358 if (!path)
5359 return -ENOMEM;
5360
5361 mutex_lock(&uuid_mutex);
5362 lock_chunks(root);
5363
5364 /* first we search for all of the device items, and then we
5365 * read in all of the chunk items. This way we can create chunk
5366 * mappings that reference all of the devices that are afound
5367 */
5368 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5369 key.offset = 0;
5370 key.type = 0;
5371 again:
5372 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5373 if (ret < 0)
5374 goto error;
5375 while (1) {
5376 leaf = path->nodes[0];
5377 slot = path->slots[0];
5378 if (slot >= btrfs_header_nritems(leaf)) {
5379 ret = btrfs_next_leaf(root, path);
5380 if (ret == 0)
5381 continue;
5382 if (ret < 0)
5383 goto error;
5384 break;
5385 }
5386 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5387 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5388 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5389 break;
5390 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5391 struct btrfs_dev_item *dev_item;
5392 dev_item = btrfs_item_ptr(leaf, slot,
5393 struct btrfs_dev_item);
5394 ret = read_one_dev(root, leaf, dev_item);
5395 if (ret)
5396 goto error;
5397 }
5398 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5399 struct btrfs_chunk *chunk;
5400 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5401 ret = read_one_chunk(root, &found_key, leaf, chunk);
5402 if (ret)
5403 goto error;
5404 }
5405 path->slots[0]++;
5406 }
5407 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5408 key.objectid = 0;
5409 btrfs_release_path(path);
5410 goto again;
5411 }
5412 ret = 0;
5413 error:
5414 unlock_chunks(root);
5415 mutex_unlock(&uuid_mutex);
5416
5417 btrfs_free_path(path);
5418 return ret;
5419 }
5420
5421 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5422 {
5423 int i;
5424
5425 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5426 btrfs_dev_stat_reset(dev, i);
5427 }
5428
5429 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5430 {
5431 struct btrfs_key key;
5432 struct btrfs_key found_key;
5433 struct btrfs_root *dev_root = fs_info->dev_root;
5434 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5435 struct extent_buffer *eb;
5436 int slot;
5437 int ret = 0;
5438 struct btrfs_device *device;
5439 struct btrfs_path *path = NULL;
5440 int i;
5441
5442 path = btrfs_alloc_path();
5443 if (!path) {
5444 ret = -ENOMEM;
5445 goto out;
5446 }
5447
5448 mutex_lock(&fs_devices->device_list_mutex);
5449 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5450 int item_size;
5451 struct btrfs_dev_stats_item *ptr;
5452
5453 key.objectid = 0;
5454 key.type = BTRFS_DEV_STATS_KEY;
5455 key.offset = device->devid;
5456 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5457 if (ret) {
5458 __btrfs_reset_dev_stats(device);
5459 device->dev_stats_valid = 1;
5460 btrfs_release_path(path);
5461 continue;
5462 }
5463 slot = path->slots[0];
5464 eb = path->nodes[0];
5465 btrfs_item_key_to_cpu(eb, &found_key, slot);
5466 item_size = btrfs_item_size_nr(eb, slot);
5467
5468 ptr = btrfs_item_ptr(eb, slot,
5469 struct btrfs_dev_stats_item);
5470
5471 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5472 if (item_size >= (1 + i) * sizeof(__le64))
5473 btrfs_dev_stat_set(device, i,
5474 btrfs_dev_stats_value(eb, ptr, i));
5475 else
5476 btrfs_dev_stat_reset(device, i);
5477 }
5478
5479 device->dev_stats_valid = 1;
5480 btrfs_dev_stat_print_on_load(device);
5481 btrfs_release_path(path);
5482 }
5483 mutex_unlock(&fs_devices->device_list_mutex);
5484
5485 out:
5486 btrfs_free_path(path);
5487 return ret < 0 ? ret : 0;
5488 }
5489
5490 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5491 struct btrfs_root *dev_root,
5492 struct btrfs_device *device)
5493 {
5494 struct btrfs_path *path;
5495 struct btrfs_key key;
5496 struct extent_buffer *eb;
5497 struct btrfs_dev_stats_item *ptr;
5498 int ret;
5499 int i;
5500
5501 key.objectid = 0;
5502 key.type = BTRFS_DEV_STATS_KEY;
5503 key.offset = device->devid;
5504
5505 path = btrfs_alloc_path();
5506 BUG_ON(!path);
5507 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5508 if (ret < 0) {
5509 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5510 ret, rcu_str_deref(device->name));
5511 goto out;
5512 }
5513
5514 if (ret == 0 &&
5515 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5516 /* need to delete old one and insert a new one */
5517 ret = btrfs_del_item(trans, dev_root, path);
5518 if (ret != 0) {
5519 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5520 rcu_str_deref(device->name), ret);
5521 goto out;
5522 }
5523 ret = 1;
5524 }
5525
5526 if (ret == 1) {
5527 /* need to insert a new item */
5528 btrfs_release_path(path);
5529 ret = btrfs_insert_empty_item(trans, dev_root, path,
5530 &key, sizeof(*ptr));
5531 if (ret < 0) {
5532 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5533 rcu_str_deref(device->name), ret);
5534 goto out;
5535 }
5536 }
5537
5538 eb = path->nodes[0];
5539 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5540 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5541 btrfs_set_dev_stats_value(eb, ptr, i,
5542 btrfs_dev_stat_read(device, i));
5543 btrfs_mark_buffer_dirty(eb);
5544
5545 out:
5546 btrfs_free_path(path);
5547 return ret;
5548 }
5549
5550 /*
5551 * called from commit_transaction. Writes all changed device stats to disk.
5552 */
5553 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5554 struct btrfs_fs_info *fs_info)
5555 {
5556 struct btrfs_root *dev_root = fs_info->dev_root;
5557 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5558 struct btrfs_device *device;
5559 int ret = 0;
5560
5561 mutex_lock(&fs_devices->device_list_mutex);
5562 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5563 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5564 continue;
5565
5566 ret = update_dev_stat_item(trans, dev_root, device);
5567 if (!ret)
5568 device->dev_stats_dirty = 0;
5569 }
5570 mutex_unlock(&fs_devices->device_list_mutex);
5571
5572 return ret;
5573 }
5574
5575 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5576 {
5577 btrfs_dev_stat_inc(dev, index);
5578 btrfs_dev_stat_print_on_error(dev);
5579 }
5580
5581 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5582 {
5583 if (!dev->dev_stats_valid)
5584 return;
5585 printk_ratelimited_in_rcu(KERN_ERR
5586 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5587 rcu_str_deref(dev->name),
5588 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5589 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5590 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5591 btrfs_dev_stat_read(dev,
5592 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5593 btrfs_dev_stat_read(dev,
5594 BTRFS_DEV_STAT_GENERATION_ERRS));
5595 }
5596
5597 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5598 {
5599 int i;
5600
5601 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5602 if (btrfs_dev_stat_read(dev, i) != 0)
5603 break;
5604 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5605 return; /* all values == 0, suppress message */
5606
5607 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5608 rcu_str_deref(dev->name),
5609 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5610 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5611 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5612 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5613 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5614 }
5615
5616 int btrfs_get_dev_stats(struct btrfs_root *root,
5617 struct btrfs_ioctl_get_dev_stats *stats)
5618 {
5619 struct btrfs_device *dev;
5620 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5621 int i;
5622
5623 mutex_lock(&fs_devices->device_list_mutex);
5624 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5625 mutex_unlock(&fs_devices->device_list_mutex);
5626
5627 if (!dev) {
5628 printk(KERN_WARNING
5629 "btrfs: get dev_stats failed, device not found\n");
5630 return -ENODEV;
5631 } else if (!dev->dev_stats_valid) {
5632 printk(KERN_WARNING
5633 "btrfs: get dev_stats failed, not yet valid\n");
5634 return -ENODEV;
5635 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5636 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5637 if (stats->nr_items > i)
5638 stats->values[i] =
5639 btrfs_dev_stat_read_and_reset(dev, i);
5640 else
5641 btrfs_dev_stat_reset(dev, i);
5642 }
5643 } else {
5644 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5645 if (stats->nr_items > i)
5646 stats->values[i] = btrfs_dev_stat_read(dev, i);
5647 }
5648 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5649 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5650 return 0;
5651 }
5652
5653 int btrfs_scratch_superblock(struct btrfs_device *device)
5654 {
5655 struct buffer_head *bh;
5656 struct btrfs_super_block *disk_super;
5657
5658 bh = btrfs_read_dev_super(device->bdev);
5659 if (!bh)
5660 return -EINVAL;
5661 disk_super = (struct btrfs_super_block *)bh->b_data;
5662
5663 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5664 set_buffer_dirty(bh);
5665 sync_dirty_buffer(bh);
5666 brelse(bh);
5667
5668 return 0;
5669 }
Linux kernel - Deliverables
This page took 0.152439 seconds and 6 git commands to generate.