?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.2.1
[linux/fpc-iii.git] / fs / btrfs / volumes.c
blobfbe7c104531c9e5eaf8b347f46d0b8a783f88322
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 <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43 #include "sysfs.h"
44
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
52
53 DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
55 struct list_head *btrfs_get_fs_uuids(void)
56 {
57 return &fs_uuids;
58 }
59
60 static struct btrfs_fs_devices *__alloc_fs_devices(void)
61 {
62 struct btrfs_fs_devices *fs_devs;
63
64 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
65 if (!fs_devs)
66 return ERR_PTR(-ENOMEM);
67
68 mutex_init(&fs_devs->device_list_mutex);
69
70 INIT_LIST_HEAD(&fs_devs->devices);
71 INIT_LIST_HEAD(&fs_devs->resized_devices);
72 INIT_LIST_HEAD(&fs_devs->alloc_list);
73 INIT_LIST_HEAD(&fs_devs->list);
74
75 return fs_devs;
76 }
77
78 /**
79 * alloc_fs_devices - allocate struct btrfs_fs_devices
80 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
81 * generated.
82 *
83 * Return: a pointer to a new &struct btrfs_fs_devices on success;
84 * ERR_PTR() on error. Returned struct is not linked onto any lists and
85 * can be destroyed with kfree() right away.
86 */
87 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
88 {
89 struct btrfs_fs_devices *fs_devs;
90
91 fs_devs = __alloc_fs_devices();
92 if (IS_ERR(fs_devs))
93 return fs_devs;
94
95 if (fsid)
96 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
97 else
98 generate_random_uuid(fs_devs->fsid);
99
100 return fs_devs;
101 }
102
103 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
104 {
105 struct btrfs_device *device;
106 WARN_ON(fs_devices->opened);
107 while (!list_empty(&fs_devices->devices)) {
108 device = list_entry(fs_devices->devices.next,
109 struct btrfs_device, dev_list);
110 list_del(&device->dev_list);
111 rcu_string_free(device->name);
112 kfree(device);
113 }
114 kfree(fs_devices);
115 }
116
117 static void btrfs_kobject_uevent(struct block_device *bdev,
118 enum kobject_action action)
119 {
120 int ret;
121
122 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
123 if (ret)
124 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
125 action,
126 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
127 &disk_to_dev(bdev->bd_disk)->kobj);
128 }
129
130 void btrfs_cleanup_fs_uuids(void)
131 {
132 struct btrfs_fs_devices *fs_devices;
133
134 while (!list_empty(&fs_uuids)) {
135 fs_devices = list_entry(fs_uuids.next,
136 struct btrfs_fs_devices, list);
137 list_del(&fs_devices->list);
138 free_fs_devices(fs_devices);
139 }
140 }
141
142 static struct btrfs_device *__alloc_device(void)
143 {
144 struct btrfs_device *dev;
145
146 dev = kzalloc(sizeof(*dev), GFP_NOFS);
147 if (!dev)
148 return ERR_PTR(-ENOMEM);
149
150 INIT_LIST_HEAD(&dev->dev_list);
151 INIT_LIST_HEAD(&dev->dev_alloc_list);
152 INIT_LIST_HEAD(&dev->resized_list);
153
154 spin_lock_init(&dev->io_lock);
155
156 spin_lock_init(&dev->reada_lock);
157 atomic_set(&dev->reada_in_flight, 0);
158 atomic_set(&dev->dev_stats_ccnt, 0);
159 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
160 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
161
162 return dev;
163 }
164
165 static noinline struct btrfs_device *__find_device(struct list_head *head,
166 u64 devid, u8 *uuid)
167 {
168 struct btrfs_device *dev;
169
170 list_for_each_entry(dev, head, dev_list) {
171 if (dev->devid == devid &&
172 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
173 return dev;
174 }
175 }
176 return NULL;
177 }
178
179 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
180 {
181 struct btrfs_fs_devices *fs_devices;
182
183 list_for_each_entry(fs_devices, &fs_uuids, list) {
184 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
185 return fs_devices;
186 }
187 return NULL;
188 }
189
190 static int
191 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
192 int flush, struct block_device **bdev,
193 struct buffer_head **bh)
194 {
195 int ret;
196
197 *bdev = blkdev_get_by_path(device_path, flags, holder);
198
199 if (IS_ERR(*bdev)) {
200 ret = PTR_ERR(*bdev);
201 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
202 goto error;
203 }
204
205 if (flush)
206 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
207 ret = set_blocksize(*bdev, 4096);
208 if (ret) {
209 blkdev_put(*bdev, flags);
210 goto error;
211 }
212 invalidate_bdev(*bdev);
213 *bh = btrfs_read_dev_super(*bdev);
214 if (!*bh) {
215 ret = -EINVAL;
216 blkdev_put(*bdev, flags);
217 goto error;
218 }
219
220 return 0;
221
222 error:
223 *bdev = NULL;
224 *bh = NULL;
225 return ret;
226 }
227
228 static void requeue_list(struct btrfs_pending_bios *pending_bios,
229 struct bio *head, struct bio *tail)
230 {
231
232 struct bio *old_head;
233
234 old_head = pending_bios->head;
235 pending_bios->head = head;
236 if (pending_bios->tail)
237 tail->bi_next = old_head;
238 else
239 pending_bios->tail = tail;
240 }
241
242 /*
243 * we try to collect pending bios for a device so we don't get a large
244 * number of procs sending bios down to the same device. This greatly
245 * improves the schedulers ability to collect and merge the bios.
246 *
247 * But, it also turns into a long list of bios to process and that is sure
248 * to eventually make the worker thread block. The solution here is to
249 * make some progress and then put this work struct back at the end of
250 * the list if the block device is congested. This way, multiple devices
251 * can make progress from a single worker thread.
252 */
253 static noinline void run_scheduled_bios(struct btrfs_device *device)
254 {
255 struct bio *pending;
256 struct backing_dev_info *bdi;
257 struct btrfs_fs_info *fs_info;
258 struct btrfs_pending_bios *pending_bios;
259 struct bio *tail;
260 struct bio *cur;
261 int again = 0;
262 unsigned long num_run;
263 unsigned long batch_run = 0;
264 unsigned long limit;
265 unsigned long last_waited = 0;
266 int force_reg = 0;
267 int sync_pending = 0;
268 struct blk_plug plug;
269
270 /*
271 * this function runs all the bios we've collected for
272 * a particular device. We don't want to wander off to
273 * another device without first sending all of these down.
274 * So, setup a plug here and finish it off before we return
275 */
276 blk_start_plug(&plug);
277
278 bdi = blk_get_backing_dev_info(device->bdev);
279 fs_info = device->dev_root->fs_info;
280 limit = btrfs_async_submit_limit(fs_info);
281 limit = limit * 2 / 3;
282
283 loop:
284 spin_lock(&device->io_lock);
285
286 loop_lock:
287 num_run = 0;
288
289 /* take all the bios off the list at once and process them
290 * later on (without the lock held). But, remember the
291 * tail and other pointers so the bios can be properly reinserted
292 * into the list if we hit congestion
293 */
294 if (!force_reg && device->pending_sync_bios.head) {
295 pending_bios = &device->pending_sync_bios;
296 force_reg = 1;
297 } else {
298 pending_bios = &device->pending_bios;
299 force_reg = 0;
300 }
301
302 pending = pending_bios->head;
303 tail = pending_bios->tail;
304 WARN_ON(pending && !tail);
305
306 /*
307 * if pending was null this time around, no bios need processing
308 * at all and we can stop. Otherwise it'll loop back up again
309 * and do an additional check so no bios are missed.
310 *
311 * device->running_pending is used to synchronize with the
312 * schedule_bio code.
313 */
314 if (device->pending_sync_bios.head == NULL &&
315 device->pending_bios.head == NULL) {
316 again = 0;
317 device->running_pending = 0;
318 } else {
319 again = 1;
320 device->running_pending = 1;
321 }
322
323 pending_bios->head = NULL;
324 pending_bios->tail = NULL;
325
326 spin_unlock(&device->io_lock);
327
328 while (pending) {
329
330 rmb();
331 /* we want to work on both lists, but do more bios on the
332 * sync list than the regular list
333 */
334 if ((num_run > 32 &&
335 pending_bios != &device->pending_sync_bios &&
336 device->pending_sync_bios.head) ||
337 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
338 device->pending_bios.head)) {
339 spin_lock(&device->io_lock);
340 requeue_list(pending_bios, pending, tail);
341 goto loop_lock;
342 }
343
344 cur = pending;
345 pending = pending->bi_next;
346 cur->bi_next = NULL;
347
348 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
349 waitqueue_active(&fs_info->async_submit_wait))
350 wake_up(&fs_info->async_submit_wait);
351
352 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
353
354 /*
355 * if we're doing the sync list, record that our
356 * plug has some sync requests on it
357 *
358 * If we're doing the regular list and there are
359 * sync requests sitting around, unplug before
360 * we add more
361 */
362 if (pending_bios == &device->pending_sync_bios) {
363 sync_pending = 1;
364 } else if (sync_pending) {
365 blk_finish_plug(&plug);
366 blk_start_plug(&plug);
367 sync_pending = 0;
368 }
369
370 btrfsic_submit_bio(cur->bi_rw, cur);
371 num_run++;
372 batch_run++;
373
374 cond_resched();
375
376 /*
377 * we made progress, there is more work to do and the bdi
378 * is now congested. Back off and let other work structs
379 * run instead
380 */
381 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
382 fs_info->fs_devices->open_devices > 1) {
383 struct io_context *ioc;
384
385 ioc = current->io_context;
386
387 /*
388 * the main goal here is that we don't want to
389 * block if we're going to be able to submit
390 * more requests without blocking.
391 *
392 * This code does two great things, it pokes into
393 * the elevator code from a filesystem _and_
394 * it makes assumptions about how batching works.
395 */
396 if (ioc && ioc->nr_batch_requests > 0 &&
397 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
398 (last_waited == 0 ||
399 ioc->last_waited == last_waited)) {
400 /*
401 * we want to go through our batch of
402 * requests and stop. So, we copy out
403 * the ioc->last_waited time and test
404 * against it before looping
405 */
406 last_waited = ioc->last_waited;
407 cond_resched();
408 continue;
409 }
410 spin_lock(&device->io_lock);
411 requeue_list(pending_bios, pending, tail);
412 device->running_pending = 1;
413
414 spin_unlock(&device->io_lock);
415 btrfs_queue_work(fs_info->submit_workers,
416 &device->work);
417 goto done;
418 }
419 /* unplug every 64 requests just for good measure */
420 if (batch_run % 64 == 0) {
421 blk_finish_plug(&plug);
422 blk_start_plug(&plug);
423 sync_pending = 0;
424 }
425 }
426
427 cond_resched();
428 if (again)
429 goto loop;
430
431 spin_lock(&device->io_lock);
432 if (device->pending_bios.head || device->pending_sync_bios.head)
433 goto loop_lock;
434 spin_unlock(&device->io_lock);
435
436 done:
437 blk_finish_plug(&plug);
438 }
439
440 static void pending_bios_fn(struct btrfs_work *work)
441 {
442 struct btrfs_device *device;
443
444 device = container_of(work, struct btrfs_device, work);
445 run_scheduled_bios(device);
446 }
447
448
449 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
450 {
451 struct btrfs_fs_devices *fs_devs;
452 struct btrfs_device *dev;
453
454 if (!cur_dev->name)
455 return;
456
457 list_for_each_entry(fs_devs, &fs_uuids, list) {
458 int del = 1;
459
460 if (fs_devs->opened)
461 continue;
462 if (fs_devs->seeding)
463 continue;
464
465 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
466
467 if (dev == cur_dev)
468 continue;
469 if (!dev->name)
470 continue;
471
472 /*
473 * Todo: This won't be enough. What if the same device
474 * comes back (with new uuid and) with its mapper path?
475 * But for now, this does help as mostly an admin will
476 * either use mapper or non mapper path throughout.
477 */
478 rcu_read_lock();
479 del = strcmp(rcu_str_deref(dev->name),
480 rcu_str_deref(cur_dev->name));
481 rcu_read_unlock();
482 if (!del)
483 break;
484 }
485
486 if (!del) {
487 /* delete the stale device */
488 if (fs_devs->num_devices == 1) {
489 btrfs_sysfs_remove_fsid(fs_devs);
490 list_del(&fs_devs->list);
491 free_fs_devices(fs_devs);
492 } else {
493 fs_devs->num_devices--;
494 list_del(&dev->dev_list);
495 rcu_string_free(dev->name);
496 kfree(dev);
497 }
498 break;
499 }
500 }
501 }
502
503 /*
504 * Add new device to list of registered devices
505 *
506 * Returns:
507 * 1 - first time device is seen
508 * 0 - device already known
509 * < 0 - error
510 */
511 static noinline int device_list_add(const char *path,
512 struct btrfs_super_block *disk_super,
513 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
514 {
515 struct btrfs_device *device;
516 struct btrfs_fs_devices *fs_devices;
517 struct rcu_string *name;
518 int ret = 0;
519 u64 found_transid = btrfs_super_generation(disk_super);
520
521 fs_devices = find_fsid(disk_super->fsid);
522 if (!fs_devices) {
523 fs_devices = alloc_fs_devices(disk_super->fsid);
524 if (IS_ERR(fs_devices))
525 return PTR_ERR(fs_devices);
526
527 list_add(&fs_devices->list, &fs_uuids);
528
529 device = NULL;
530 } else {
531 device = __find_device(&fs_devices->devices, devid,
532 disk_super->dev_item.uuid);
533 }
534
535 if (!device) {
536 if (fs_devices->opened)
537 return -EBUSY;
538
539 device = btrfs_alloc_device(NULL, &devid,
540 disk_super->dev_item.uuid);
541 if (IS_ERR(device)) {
542 /* we can safely leave the fs_devices entry around */
543 return PTR_ERR(device);
544 }
545
546 name = rcu_string_strdup(path, GFP_NOFS);
547 if (!name) {
548 kfree(device);
549 return -ENOMEM;
550 }
551 rcu_assign_pointer(device->name, name);
552
553 mutex_lock(&fs_devices->device_list_mutex);
554 list_add_rcu(&device->dev_list, &fs_devices->devices);
555 fs_devices->num_devices++;
556 mutex_unlock(&fs_devices->device_list_mutex);
557
558 ret = 1;
559 device->fs_devices = fs_devices;
560 } else if (!device->name || strcmp(device->name->str, path)) {
561 /*
562 * When FS is already mounted.
563 * 1. If you are here and if the device->name is NULL that
564 * means this device was missing at time of FS mount.
565 * 2. If you are here and if the device->name is different
566 * from 'path' that means either
567 * a. The same device disappeared and reappeared with
568 * different name. or
569 * b. The missing-disk-which-was-replaced, has
570 * reappeared now.
571 *
572 * We must allow 1 and 2a above. But 2b would be a spurious
573 * and unintentional.
574 *
575 * Further in case of 1 and 2a above, the disk at 'path'
576 * would have missed some transaction when it was away and
577 * in case of 2a the stale bdev has to be updated as well.
578 * 2b must not be allowed at all time.
579 */
580
581 /*
582 * For now, we do allow update to btrfs_fs_device through the
583 * btrfs dev scan cli after FS has been mounted. We're still
584 * tracking a problem where systems fail mount by subvolume id
585 * when we reject replacement on a mounted FS.
586 */
587 if (!fs_devices->opened && found_transid < device->generation) {
588 /*
589 * That is if the FS is _not_ mounted and if you
590 * are here, that means there is more than one
591 * disk with same uuid and devid.We keep the one
592 * with larger generation number or the last-in if
593 * generation are equal.
594 */
595 return -EEXIST;
596 }
597
598 name = rcu_string_strdup(path, GFP_NOFS);
599 if (!name)
600 return -ENOMEM;
601 rcu_string_free(device->name);
602 rcu_assign_pointer(device->name, name);
603 if (device->missing) {
604 fs_devices->missing_devices--;
605 device->missing = 0;
606 }
607 }
608
609 /*
610 * Unmount does not free the btrfs_device struct but would zero
611 * generation along with most of the other members. So just update
612 * it back. We need it to pick the disk with largest generation
613 * (as above).
614 */
615 if (!fs_devices->opened)
616 device->generation = found_transid;
617
618 /*
619 * if there is new btrfs on an already registered device,
620 * then remove the stale device entry.
621 */
622 btrfs_free_stale_device(device);
623
624 *fs_devices_ret = fs_devices;
625
626 return ret;
627 }
628
629 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
630 {
631 struct btrfs_fs_devices *fs_devices;
632 struct btrfs_device *device;
633 struct btrfs_device *orig_dev;
634
635 fs_devices = alloc_fs_devices(orig->fsid);
636 if (IS_ERR(fs_devices))
637 return fs_devices;
638
639 mutex_lock(&orig->device_list_mutex);
640 fs_devices->total_devices = orig->total_devices;
641
642 /* We have held the volume lock, it is safe to get the devices. */
643 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
644 struct rcu_string *name;
645
646 device = btrfs_alloc_device(NULL, &orig_dev->devid,
647 orig_dev->uuid);
648 if (IS_ERR(device))
649 goto error;
650
651 /*
652 * This is ok to do without rcu read locked because we hold the
653 * uuid mutex so nothing we touch in here is going to disappear.
654 */
655 if (orig_dev->name) {
656 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
657 if (!name) {
658 kfree(device);
659 goto error;
660 }
661 rcu_assign_pointer(device->name, name);
662 }
663
664 list_add(&device->dev_list, &fs_devices->devices);
665 device->fs_devices = fs_devices;
666 fs_devices->num_devices++;
667 }
668 mutex_unlock(&orig->device_list_mutex);
669 return fs_devices;
670 error:
671 mutex_unlock(&orig->device_list_mutex);
672 free_fs_devices(fs_devices);
673 return ERR_PTR(-ENOMEM);
674 }
675
676 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
677 {
678 struct btrfs_device *device, *next;
679 struct btrfs_device *latest_dev = NULL;
680
681 mutex_lock(&uuid_mutex);
682 again:
683 /* This is the initialized path, it is safe to release the devices. */
684 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
685 if (device->in_fs_metadata) {
686 if (!device->is_tgtdev_for_dev_replace &&
687 (!latest_dev ||
688 device->generation > latest_dev->generation)) {
689 latest_dev = device;
690 }
691 continue;
692 }
693
694 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
695 /*
696 * In the first step, keep the device which has
697 * the correct fsid and the devid that is used
698 * for the dev_replace procedure.
699 * In the second step, the dev_replace state is
700 * read from the device tree and it is known
701 * whether the procedure is really active or
702 * not, which means whether this device is
703 * used or whether it should be removed.
704 */
705 if (step == 0 || device->is_tgtdev_for_dev_replace) {
706 continue;
707 }
708 }
709 if (device->bdev) {
710 blkdev_put(device->bdev, device->mode);
711 device->bdev = NULL;
712 fs_devices->open_devices--;
713 }
714 if (device->writeable) {
715 list_del_init(&device->dev_alloc_list);
716 device->writeable = 0;
717 if (!device->is_tgtdev_for_dev_replace)
718 fs_devices->rw_devices--;
719 }
720 list_del_init(&device->dev_list);
721 fs_devices->num_devices--;
722 rcu_string_free(device->name);
723 kfree(device);
724 }
725
726 if (fs_devices->seed) {
727 fs_devices = fs_devices->seed;
728 goto again;
729 }
730
731 fs_devices->latest_bdev = latest_dev->bdev;
732
733 mutex_unlock(&uuid_mutex);
734 }
735
736 static void __free_device(struct work_struct *work)
737 {
738 struct btrfs_device *device;
739
740 device = container_of(work, struct btrfs_device, rcu_work);
741
742 if (device->bdev)
743 blkdev_put(device->bdev, device->mode);
744
745 rcu_string_free(device->name);
746 kfree(device);
747 }
748
749 static void free_device(struct rcu_head *head)
750 {
751 struct btrfs_device *device;
752
753 device = container_of(head, struct btrfs_device, rcu);
754
755 INIT_WORK(&device->rcu_work, __free_device);
756 schedule_work(&device->rcu_work);
757 }
758
759 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
760 {
761 struct btrfs_device *device, *tmp;
762
763 if (--fs_devices->opened > 0)
764 return 0;
765
766 mutex_lock(&fs_devices->device_list_mutex);
767 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
768 struct btrfs_device *new_device;
769 struct rcu_string *name;
770
771 if (device->bdev)
772 fs_devices->open_devices--;
773
774 if (device->writeable &&
775 device->devid != BTRFS_DEV_REPLACE_DEVID) {
776 list_del_init(&device->dev_alloc_list);
777 fs_devices->rw_devices--;
778 }
779
780 if (device->missing)
781 fs_devices->missing_devices--;
782
783 new_device = btrfs_alloc_device(NULL, &device->devid,
784 device->uuid);
785 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
786
787 /* Safe because we are under uuid_mutex */
788 if (device->name) {
789 name = rcu_string_strdup(device->name->str, GFP_NOFS);
790 BUG_ON(!name); /* -ENOMEM */
791 rcu_assign_pointer(new_device->name, name);
792 }
793
794 list_replace_rcu(&device->dev_list, &new_device->dev_list);
795 new_device->fs_devices = device->fs_devices;
796
797 call_rcu(&device->rcu, free_device);
798 }
799 mutex_unlock(&fs_devices->device_list_mutex);
800
801 WARN_ON(fs_devices->open_devices);
802 WARN_ON(fs_devices->rw_devices);
803 fs_devices->opened = 0;
804 fs_devices->seeding = 0;
805
806 return 0;
807 }
808
809 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
810 {
811 struct btrfs_fs_devices *seed_devices = NULL;
812 int ret;
813
814 mutex_lock(&uuid_mutex);
815 ret = __btrfs_close_devices(fs_devices);
816 if (!fs_devices->opened) {
817 seed_devices = fs_devices->seed;
818 fs_devices->seed = NULL;
819 }
820 mutex_unlock(&uuid_mutex);
821
822 while (seed_devices) {
823 fs_devices = seed_devices;
824 seed_devices = fs_devices->seed;
825 __btrfs_close_devices(fs_devices);
826 free_fs_devices(fs_devices);
827 }
828 /*
829 * Wait for rcu kworkers under __btrfs_close_devices
830 * to finish all blkdev_puts so device is really
831 * free when umount is done.
832 */
833 rcu_barrier();
834 return ret;
835 }
836
837 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
838 fmode_t flags, void *holder)
839 {
840 struct request_queue *q;
841 struct block_device *bdev;
842 struct list_head *head = &fs_devices->devices;
843 struct btrfs_device *device;
844 struct btrfs_device *latest_dev = NULL;
845 struct buffer_head *bh;
846 struct btrfs_super_block *disk_super;
847 u64 devid;
848 int seeding = 1;
849 int ret = 0;
850
851 flags |= FMODE_EXCL;
852
853 list_for_each_entry(device, head, dev_list) {
854 if (device->bdev)
855 continue;
856 if (!device->name)
857 continue;
858
859 /* Just open everything we can; ignore failures here */
860 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
861 &bdev, &bh))
862 continue;
863
864 disk_super = (struct btrfs_super_block *)bh->b_data;
865 devid = btrfs_stack_device_id(&disk_super->dev_item);
866 if (devid != device->devid)
867 goto error_brelse;
868
869 if (memcmp(device->uuid, disk_super->dev_item.uuid,
870 BTRFS_UUID_SIZE))
871 goto error_brelse;
872
873 device->generation = btrfs_super_generation(disk_super);
874 if (!latest_dev ||
875 device->generation > latest_dev->generation)
876 latest_dev = device;
877
878 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
879 device->writeable = 0;
880 } else {
881 device->writeable = !bdev_read_only(bdev);
882 seeding = 0;
883 }
884
885 q = bdev_get_queue(bdev);
886 if (blk_queue_discard(q))
887 device->can_discard = 1;
888
889 device->bdev = bdev;
890 device->in_fs_metadata = 0;
891 device->mode = flags;
892
893 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
894 fs_devices->rotating = 1;
895
896 fs_devices->open_devices++;
897 if (device->writeable &&
898 device->devid != BTRFS_DEV_REPLACE_DEVID) {
899 fs_devices->rw_devices++;
900 list_add(&device->dev_alloc_list,
901 &fs_devices->alloc_list);
902 }
903 brelse(bh);
904 continue;
905
906 error_brelse:
907 brelse(bh);
908 blkdev_put(bdev, flags);
909 continue;
910 }
911 if (fs_devices->open_devices == 0) {
912 ret = -EINVAL;
913 goto out;
914 }
915 fs_devices->seeding = seeding;
916 fs_devices->opened = 1;
917 fs_devices->latest_bdev = latest_dev->bdev;
918 fs_devices->total_rw_bytes = 0;
919 out:
920 return ret;
921 }
922
923 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
924 fmode_t flags, void *holder)
925 {
926 int ret;
927
928 mutex_lock(&uuid_mutex);
929 if (fs_devices->opened) {
930 fs_devices->opened++;
931 ret = 0;
932 } else {
933 ret = __btrfs_open_devices(fs_devices, flags, holder);
934 }
935 mutex_unlock(&uuid_mutex);
936 return ret;
937 }
938
939 /*
940 * Look for a btrfs signature on a device. This may be called out of the mount path
941 * and we are not allowed to call set_blocksize during the scan. The superblock
942 * is read via pagecache
943 */
944 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
945 struct btrfs_fs_devices **fs_devices_ret)
946 {
947 struct btrfs_super_block *disk_super;
948 struct block_device *bdev;
949 struct page *page;
950 void *p;
951 int ret = -EINVAL;
952 u64 devid;
953 u64 transid;
954 u64 total_devices;
955 u64 bytenr;
956 pgoff_t index;
957
958 /*
959 * we would like to check all the supers, but that would make
960 * a btrfs mount succeed after a mkfs from a different FS.
961 * So, we need to add a special mount option to scan for
962 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
963 */
964 bytenr = btrfs_sb_offset(0);
965 flags |= FMODE_EXCL;
966 mutex_lock(&uuid_mutex);
967
968 bdev = blkdev_get_by_path(path, flags, holder);
969
970 if (IS_ERR(bdev)) {
971 ret = PTR_ERR(bdev);
972 goto error;
973 }
974
975 /* make sure our super fits in the device */
976 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
977 goto error_bdev_put;
978
979 /* make sure our super fits in the page */
980 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
981 goto error_bdev_put;
982
983 /* make sure our super doesn't straddle pages on disk */
984 index = bytenr >> PAGE_CACHE_SHIFT;
985 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
986 goto error_bdev_put;
987
988 /* pull in the page with our super */
989 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
990 index, GFP_NOFS);
991
992 if (IS_ERR_OR_NULL(page))
993 goto error_bdev_put;
994
995 p = kmap(page);
996
997 /* align our pointer to the offset of the super block */
998 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
999
1000 if (btrfs_super_bytenr(disk_super) != bytenr ||
1001 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1002 goto error_unmap;
1003
1004 devid = btrfs_stack_device_id(&disk_super->dev_item);
1005 transid = btrfs_super_generation(disk_super);
1006 total_devices = btrfs_super_num_devices(disk_super);
1007
1008 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1009 if (ret > 0) {
1010 if (disk_super->label[0]) {
1011 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1012 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1013 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1014 } else {
1015 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1016 }
1017
1018 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1019 ret = 0;
1020 }
1021 if (!ret && fs_devices_ret)
1022 (*fs_devices_ret)->total_devices = total_devices;
1023
1024 error_unmap:
1025 kunmap(page);
1026 page_cache_release(page);
1027
1028 error_bdev_put:
1029 blkdev_put(bdev, flags);
1030 error:
1031 mutex_unlock(&uuid_mutex);
1032 return ret;
1033 }
1034
1035 /* helper to account the used device space in the range */
1036 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1037 u64 end, u64 *length)
1038 {
1039 struct btrfs_key key;
1040 struct btrfs_root *root = device->dev_root;
1041 struct btrfs_dev_extent *dev_extent;
1042 struct btrfs_path *path;
1043 u64 extent_end;
1044 int ret;
1045 int slot;
1046 struct extent_buffer *l;
1047
1048 *length = 0;
1049
1050 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1051 return 0;
1052
1053 path = btrfs_alloc_path();
1054 if (!path)
1055 return -ENOMEM;
1056 path->reada = 2;
1057
1058 key.objectid = device->devid;
1059 key.offset = start;
1060 key.type = BTRFS_DEV_EXTENT_KEY;
1061
1062 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1063 if (ret < 0)
1064 goto out;
1065 if (ret > 0) {
1066 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1067 if (ret < 0)
1068 goto out;
1069 }
1070
1071 while (1) {
1072 l = path->nodes[0];
1073 slot = path->slots[0];
1074 if (slot >= btrfs_header_nritems(l)) {
1075 ret = btrfs_next_leaf(root, path);
1076 if (ret == 0)
1077 continue;
1078 if (ret < 0)
1079 goto out;
1080
1081 break;
1082 }
1083 btrfs_item_key_to_cpu(l, &key, slot);
1084
1085 if (key.objectid < device->devid)
1086 goto next;
1087
1088 if (key.objectid > device->devid)
1089 break;
1090
1091 if (key.type != BTRFS_DEV_EXTENT_KEY)
1092 goto next;
1093
1094 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1095 extent_end = key.offset + btrfs_dev_extent_length(l,
1096 dev_extent);
1097 if (key.offset <= start && extent_end > end) {
1098 *length = end - start + 1;
1099 break;
1100 } else if (key.offset <= start && extent_end > start)
1101 *length += extent_end - start;
1102 else if (key.offset > start && extent_end <= end)
1103 *length += extent_end - key.offset;
1104 else if (key.offset > start && key.offset <= end) {
1105 *length += end - key.offset + 1;
1106 break;
1107 } else if (key.offset > end)
1108 break;
1109
1110 next:
1111 path->slots[0]++;
1112 }
1113 ret = 0;
1114 out:
1115 btrfs_free_path(path);
1116 return ret;
1117 }
1118
1119 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1120 struct btrfs_device *device,
1121 u64 *start, u64 len)
1122 {
1123 struct extent_map *em;
1124 struct list_head *search_list = &trans->transaction->pending_chunks;
1125 int ret = 0;
1126 u64 physical_start = *start;
1127
1128 again:
1129 list_for_each_entry(em, search_list, list) {
1130 struct map_lookup *map;
1131 int i;
1132
1133 map = (struct map_lookup *)em->bdev;
1134 for (i = 0; i < map->num_stripes; i++) {
1135 u64 end;
1136
1137 if (map->stripes[i].dev != device)
1138 continue;
1139 if (map->stripes[i].physical >= physical_start + len ||
1140 map->stripes[i].physical + em->orig_block_len <=
1141 physical_start)
1142 continue;
1143 /*
1144 * Make sure that while processing the pinned list we do
1145 * not override our *start with a lower value, because
1146 * we can have pinned chunks that fall within this
1147 * device hole and that have lower physical addresses
1148 * than the pending chunks we processed before. If we
1149 * do not take this special care we can end up getting
1150 * 2 pending chunks that start at the same physical
1151 * device offsets because the end offset of a pinned
1152 * chunk can be equal to the start offset of some
1153 * pending chunk.
1154 */
1155 end = map->stripes[i].physical + em->orig_block_len;
1156 if (end > *start) {
1157 *start = end;
1158 ret = 1;
1159 }
1160 }
1161 }
1162 if (search_list == &trans->transaction->pending_chunks) {
1163 search_list = &trans->root->fs_info->pinned_chunks;
1164 goto again;
1165 }
1166
1167 return ret;
1168 }
1169
1170
1171 /*
1172 * find_free_dev_extent - find free space in the specified device
1173 * @device: the device which we search the free space in
1174 * @num_bytes: the size of the free space that we need
1175 * @start: store the start of the free space.
1176 * @len: the size of the free space. that we find, or the size of the max
1177 * free space if we don't find suitable free space
1178 *
1179 * this uses a pretty simple search, the expectation is that it is
1180 * called very infrequently and that a given device has a small number
1181 * of extents
1182 *
1183 * @start is used to store the start of the free space if we find. But if we
1184 * don't find suitable free space, it will be used to store the start position
1185 * of the max free space.
1186 *
1187 * @len is used to store the size of the free space that we find.
1188 * But if we don't find suitable free space, it is used to store the size of
1189 * the max free space.
1190 */
1191 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1192 struct btrfs_device *device, u64 num_bytes,
1193 u64 *start, u64 *len)
1194 {
1195 struct btrfs_key key;
1196 struct btrfs_root *root = device->dev_root;
1197 struct btrfs_dev_extent *dev_extent;
1198 struct btrfs_path *path;
1199 u64 hole_size;
1200 u64 max_hole_start;
1201 u64 max_hole_size;
1202 u64 extent_end;
1203 u64 search_start;
1204 u64 search_end = device->total_bytes;
1205 int ret;
1206 int slot;
1207 struct extent_buffer *l;
1208
1209 /* FIXME use last free of some kind */
1210
1211 /* we don't want to overwrite the superblock on the drive,
1212 * so we make sure to start at an offset of at least 1MB
1213 */
1214 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1215
1216 path = btrfs_alloc_path();
1217 if (!path)
1218 return -ENOMEM;
1219
1220 max_hole_start = search_start;
1221 max_hole_size = 0;
1222
1223 again:
1224 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1225 ret = -ENOSPC;
1226 goto out;
1227 }
1228
1229 path->reada = 2;
1230 path->search_commit_root = 1;
1231 path->skip_locking = 1;
1232
1233 key.objectid = device->devid;
1234 key.offset = search_start;
1235 key.type = BTRFS_DEV_EXTENT_KEY;
1236
1237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1238 if (ret < 0)
1239 goto out;
1240 if (ret > 0) {
1241 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1242 if (ret < 0)
1243 goto out;
1244 }
1245
1246 while (1) {
1247 l = path->nodes[0];
1248 slot = path->slots[0];
1249 if (slot >= btrfs_header_nritems(l)) {
1250 ret = btrfs_next_leaf(root, path);
1251 if (ret == 0)
1252 continue;
1253 if (ret < 0)
1254 goto out;
1255
1256 break;
1257 }
1258 btrfs_item_key_to_cpu(l, &key, slot);
1259
1260 if (key.objectid < device->devid)
1261 goto next;
1262
1263 if (key.objectid > device->devid)
1264 break;
1265
1266 if (key.type != BTRFS_DEV_EXTENT_KEY)
1267 goto next;
1268
1269 if (key.offset > search_start) {
1270 hole_size = key.offset - search_start;
1271
1272 /*
1273 * Have to check before we set max_hole_start, otherwise
1274 * we could end up sending back this offset anyway.
1275 */
1276 if (contains_pending_extent(trans, device,
1277 &search_start,
1278 hole_size)) {
1279 if (key.offset >= search_start) {
1280 hole_size = key.offset - search_start;
1281 } else {
1282 WARN_ON_ONCE(1);
1283 hole_size = 0;
1284 }
1285 }
1286
1287 if (hole_size > max_hole_size) {
1288 max_hole_start = search_start;
1289 max_hole_size = hole_size;
1290 }
1291
1292 /*
1293 * If this free space is greater than which we need,
1294 * it must be the max free space that we have found
1295 * until now, so max_hole_start must point to the start
1296 * of this free space and the length of this free space
1297 * is stored in max_hole_size. Thus, we return
1298 * max_hole_start and max_hole_size and go back to the
1299 * caller.
1300 */
1301 if (hole_size >= num_bytes) {
1302 ret = 0;
1303 goto out;
1304 }
1305 }
1306
1307 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1308 extent_end = key.offset + btrfs_dev_extent_length(l,
1309 dev_extent);
1310 if (extent_end > search_start)
1311 search_start = extent_end;
1312 next:
1313 path->slots[0]++;
1314 cond_resched();
1315 }
1316
1317 /*
1318 * At this point, search_start should be the end of
1319 * allocated dev extents, and when shrinking the device,
1320 * search_end may be smaller than search_start.
1321 */
1322 if (search_end > search_start) {
1323 hole_size = search_end - search_start;
1324
1325 if (contains_pending_extent(trans, device, &search_start,
1326 hole_size)) {
1327 btrfs_release_path(path);
1328 goto again;
1329 }
1330
1331 if (hole_size > max_hole_size) {
1332 max_hole_start = search_start;
1333 max_hole_size = hole_size;
1334 }
1335 }
1336
1337 /* See above. */
1338 if (max_hole_size < num_bytes)
1339 ret = -ENOSPC;
1340 else
1341 ret = 0;
1342
1343 out:
1344 btrfs_free_path(path);
1345 *start = max_hole_start;
1346 if (len)
1347 *len = max_hole_size;
1348 return ret;
1349 }
1350
1351 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1352 struct btrfs_device *device,
1353 u64 start, u64 *dev_extent_len)
1354 {
1355 int ret;
1356 struct btrfs_path *path;
1357 struct btrfs_root *root = device->dev_root;
1358 struct btrfs_key key;
1359 struct btrfs_key found_key;
1360 struct extent_buffer *leaf = NULL;
1361 struct btrfs_dev_extent *extent = NULL;
1362
1363 path = btrfs_alloc_path();
1364 if (!path)
1365 return -ENOMEM;
1366
1367 key.objectid = device->devid;
1368 key.offset = start;
1369 key.type = BTRFS_DEV_EXTENT_KEY;
1370 again:
1371 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1372 if (ret > 0) {
1373 ret = btrfs_previous_item(root, path, key.objectid,
1374 BTRFS_DEV_EXTENT_KEY);
1375 if (ret)
1376 goto out;
1377 leaf = path->nodes[0];
1378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1379 extent = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_dev_extent);
1381 BUG_ON(found_key.offset > start || found_key.offset +
1382 btrfs_dev_extent_length(leaf, extent) < start);
1383 key = found_key;
1384 btrfs_release_path(path);
1385 goto again;
1386 } else if (ret == 0) {
1387 leaf = path->nodes[0];
1388 extent = btrfs_item_ptr(leaf, path->slots[0],
1389 struct btrfs_dev_extent);
1390 } else {
1391 btrfs_error(root->fs_info, ret, "Slot search failed");
1392 goto out;
1393 }
1394
1395 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1396
1397 ret = btrfs_del_item(trans, root, path);
1398 if (ret) {
1399 btrfs_error(root->fs_info, ret,
1400 "Failed to remove dev extent item");
1401 } else {
1402 trans->transaction->have_free_bgs = 1;
1403 }
1404 out:
1405 btrfs_free_path(path);
1406 return ret;
1407 }
1408
1409 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1410 struct btrfs_device *device,
1411 u64 chunk_tree, u64 chunk_objectid,
1412 u64 chunk_offset, u64 start, u64 num_bytes)
1413 {
1414 int ret;
1415 struct btrfs_path *path;
1416 struct btrfs_root *root = device->dev_root;
1417 struct btrfs_dev_extent *extent;
1418 struct extent_buffer *leaf;
1419 struct btrfs_key key;
1420
1421 WARN_ON(!device->in_fs_metadata);
1422 WARN_ON(device->is_tgtdev_for_dev_replace);
1423 path = btrfs_alloc_path();
1424 if (!path)
1425 return -ENOMEM;
1426
1427 key.objectid = device->devid;
1428 key.offset = start;
1429 key.type = BTRFS_DEV_EXTENT_KEY;
1430 ret = btrfs_insert_empty_item(trans, root, path, &key,
1431 sizeof(*extent));
1432 if (ret)
1433 goto out;
1434
1435 leaf = path->nodes[0];
1436 extent = btrfs_item_ptr(leaf, path->slots[0],
1437 struct btrfs_dev_extent);
1438 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1439 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1440 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1441
1442 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1443 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1444
1445 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1446 btrfs_mark_buffer_dirty(leaf);
1447 out:
1448 btrfs_free_path(path);
1449 return ret;
1450 }
1451
1452 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1453 {
1454 struct extent_map_tree *em_tree;
1455 struct extent_map *em;
1456 struct rb_node *n;
1457 u64 ret = 0;
1458
1459 em_tree = &fs_info->mapping_tree.map_tree;
1460 read_lock(&em_tree->lock);
1461 n = rb_last(&em_tree->map);
1462 if (n) {
1463 em = rb_entry(n, struct extent_map, rb_node);
1464 ret = em->start + em->len;
1465 }
1466 read_unlock(&em_tree->lock);
1467
1468 return ret;
1469 }
1470
1471 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1472 u64 *devid_ret)
1473 {
1474 int ret;
1475 struct btrfs_key key;
1476 struct btrfs_key found_key;
1477 struct btrfs_path *path;
1478
1479 path = btrfs_alloc_path();
1480 if (!path)
1481 return -ENOMEM;
1482
1483 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1484 key.type = BTRFS_DEV_ITEM_KEY;
1485 key.offset = (u64)-1;
1486
1487 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1488 if (ret < 0)
1489 goto error;
1490
1491 BUG_ON(ret == 0); /* Corruption */
1492
1493 ret = btrfs_previous_item(fs_info->chunk_root, path,
1494 BTRFS_DEV_ITEMS_OBJECTID,
1495 BTRFS_DEV_ITEM_KEY);
1496 if (ret) {
1497 *devid_ret = 1;
1498 } else {
1499 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1500 path->slots[0]);
1501 *devid_ret = found_key.offset + 1;
1502 }
1503 ret = 0;
1504 error:
1505 btrfs_free_path(path);
1506 return ret;
1507 }
1508
1509 /*
1510 * the device information is stored in the chunk root
1511 * the btrfs_device struct should be fully filled in
1512 */
1513 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1514 struct btrfs_root *root,
1515 struct btrfs_device *device)
1516 {
1517 int ret;
1518 struct btrfs_path *path;
1519 struct btrfs_dev_item *dev_item;
1520 struct extent_buffer *leaf;
1521 struct btrfs_key key;
1522 unsigned long ptr;
1523
1524 root = root->fs_info->chunk_root;
1525
1526 path = btrfs_alloc_path();
1527 if (!path)
1528 return -ENOMEM;
1529
1530 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1531 key.type = BTRFS_DEV_ITEM_KEY;
1532 key.offset = device->devid;
1533
1534 ret = btrfs_insert_empty_item(trans, root, path, &key,
1535 sizeof(*dev_item));
1536 if (ret)
1537 goto out;
1538
1539 leaf = path->nodes[0];
1540 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1541
1542 btrfs_set_device_id(leaf, dev_item, device->devid);
1543 btrfs_set_device_generation(leaf, dev_item, 0);
1544 btrfs_set_device_type(leaf, dev_item, device->type);
1545 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1546 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1547 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1548 btrfs_set_device_total_bytes(leaf, dev_item,
1549 btrfs_device_get_disk_total_bytes(device));
1550 btrfs_set_device_bytes_used(leaf, dev_item,
1551 btrfs_device_get_bytes_used(device));
1552 btrfs_set_device_group(leaf, dev_item, 0);
1553 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1554 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1555 btrfs_set_device_start_offset(leaf, dev_item, 0);
1556
1557 ptr = btrfs_device_uuid(dev_item);
1558 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1559 ptr = btrfs_device_fsid(dev_item);
1560 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1561 btrfs_mark_buffer_dirty(leaf);
1562
1563 ret = 0;
1564 out:
1565 btrfs_free_path(path);
1566 return ret;
1567 }
1568
1569 /*
1570 * Function to update ctime/mtime for a given device path.
1571 * Mainly used for ctime/mtime based probe like libblkid.
1572 */
1573 static void update_dev_time(char *path_name)
1574 {
1575 struct file *filp;
1576
1577 filp = filp_open(path_name, O_RDWR, 0);
1578 if (IS_ERR(filp))
1579 return;
1580 file_update_time(filp);
1581 filp_close(filp, NULL);
1582 return;
1583 }
1584
1585 static int btrfs_rm_dev_item(struct btrfs_root *root,
1586 struct btrfs_device *device)
1587 {
1588 int ret;
1589 struct btrfs_path *path;
1590 struct btrfs_key key;
1591 struct btrfs_trans_handle *trans;
1592
1593 root = root->fs_info->chunk_root;
1594
1595 path = btrfs_alloc_path();
1596 if (!path)
1597 return -ENOMEM;
1598
1599 trans = btrfs_start_transaction(root, 0);
1600 if (IS_ERR(trans)) {
1601 btrfs_free_path(path);
1602 return PTR_ERR(trans);
1603 }
1604 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1605 key.type = BTRFS_DEV_ITEM_KEY;
1606 key.offset = device->devid;
1607
1608 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1609 if (ret < 0)
1610 goto out;
1611
1612 if (ret > 0) {
1613 ret = -ENOENT;
1614 goto out;
1615 }
1616
1617 ret = btrfs_del_item(trans, root, path);
1618 if (ret)
1619 goto out;
1620 out:
1621 btrfs_free_path(path);
1622 btrfs_commit_transaction(trans, root);
1623 return ret;
1624 }
1625
1626 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1627 {
1628 struct btrfs_device *device;
1629 struct btrfs_device *next_device;
1630 struct block_device *bdev;
1631 struct buffer_head *bh = NULL;
1632 struct btrfs_super_block *disk_super;
1633 struct btrfs_fs_devices *cur_devices;
1634 u64 all_avail;
1635 u64 devid;
1636 u64 num_devices;
1637 u8 *dev_uuid;
1638 unsigned seq;
1639 int ret = 0;
1640 bool clear_super = false;
1641
1642 mutex_lock(&uuid_mutex);
1643
1644 do {
1645 seq = read_seqbegin(&root->fs_info->profiles_lock);
1646
1647 all_avail = root->fs_info->avail_data_alloc_bits |
1648 root->fs_info->avail_system_alloc_bits |
1649 root->fs_info->avail_metadata_alloc_bits;
1650 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1651
1652 num_devices = root->fs_info->fs_devices->num_devices;
1653 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1654 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1655 WARN_ON(num_devices < 1);
1656 num_devices--;
1657 }
1658 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1659
1660 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1661 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1662 goto out;
1663 }
1664
1665 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1666 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1667 goto out;
1668 }
1669
1670 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1671 root->fs_info->fs_devices->rw_devices <= 2) {
1672 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1673 goto out;
1674 }
1675 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1676 root->fs_info->fs_devices->rw_devices <= 3) {
1677 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1678 goto out;
1679 }
1680
1681 if (strcmp(device_path, "missing") == 0) {
1682 struct list_head *devices;
1683 struct btrfs_device *tmp;
1684
1685 device = NULL;
1686 devices = &root->fs_info->fs_devices->devices;
1687 /*
1688 * It is safe to read the devices since the volume_mutex
1689 * is held.
1690 */
1691 list_for_each_entry(tmp, devices, dev_list) {
1692 if (tmp->in_fs_metadata &&
1693 !tmp->is_tgtdev_for_dev_replace &&
1694 !tmp->bdev) {
1695 device = tmp;
1696 break;
1697 }
1698 }
1699 bdev = NULL;
1700 bh = NULL;
1701 disk_super = NULL;
1702 if (!device) {
1703 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1704 goto out;
1705 }
1706 } else {
1707 ret = btrfs_get_bdev_and_sb(device_path,
1708 FMODE_WRITE | FMODE_EXCL,
1709 root->fs_info->bdev_holder, 0,
1710 &bdev, &bh);
1711 if (ret)
1712 goto out;
1713 disk_super = (struct btrfs_super_block *)bh->b_data;
1714 devid = btrfs_stack_device_id(&disk_super->dev_item);
1715 dev_uuid = disk_super->dev_item.uuid;
1716 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1717 disk_super->fsid);
1718 if (!device) {
1719 ret = -ENOENT;
1720 goto error_brelse;
1721 }
1722 }
1723
1724 if (device->is_tgtdev_for_dev_replace) {
1725 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1726 goto error_brelse;
1727 }
1728
1729 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1730 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1731 goto error_brelse;
1732 }
1733
1734 if (device->writeable) {
1735 lock_chunks(root);
1736 list_del_init(&device->dev_alloc_list);
1737 device->fs_devices->rw_devices--;
1738 unlock_chunks(root);
1739 clear_super = true;
1740 }
1741
1742 mutex_unlock(&uuid_mutex);
1743 ret = btrfs_shrink_device(device, 0);
1744 mutex_lock(&uuid_mutex);
1745 if (ret)
1746 goto error_undo;
1747
1748 /*
1749 * TODO: the superblock still includes this device in its num_devices
1750 * counter although write_all_supers() is not locked out. This
1751 * could give a filesystem state which requires a degraded mount.
1752 */
1753 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1754 if (ret)
1755 goto error_undo;
1756
1757 device->in_fs_metadata = 0;
1758 btrfs_scrub_cancel_dev(root->fs_info, device);
1759
1760 /*
1761 * the device list mutex makes sure that we don't change
1762 * the device list while someone else is writing out all
1763 * the device supers. Whoever is writing all supers, should
1764 * lock the device list mutex before getting the number of
1765 * devices in the super block (super_copy). Conversely,
1766 * whoever updates the number of devices in the super block
1767 * (super_copy) should hold the device list mutex.
1768 */
1769
1770 cur_devices = device->fs_devices;
1771 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1772 list_del_rcu(&device->dev_list);
1773
1774 device->fs_devices->num_devices--;
1775 device->fs_devices->total_devices--;
1776
1777 if (device->missing)
1778 device->fs_devices->missing_devices--;
1779
1780 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1781 struct btrfs_device, dev_list);
1782 if (device->bdev == root->fs_info->sb->s_bdev)
1783 root->fs_info->sb->s_bdev = next_device->bdev;
1784 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1785 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1786
1787 if (device->bdev) {
1788 device->fs_devices->open_devices--;
1789 /* remove sysfs entry */
1790 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
1791 }
1792
1793 call_rcu(&device->rcu, free_device);
1794
1795 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1796 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1797 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1798
1799 if (cur_devices->open_devices == 0) {
1800 struct btrfs_fs_devices *fs_devices;
1801 fs_devices = root->fs_info->fs_devices;
1802 while (fs_devices) {
1803 if (fs_devices->seed == cur_devices) {
1804 fs_devices->seed = cur_devices->seed;
1805 break;
1806 }
1807 fs_devices = fs_devices->seed;
1808 }
1809 cur_devices->seed = NULL;
1810 __btrfs_close_devices(cur_devices);
1811 free_fs_devices(cur_devices);
1812 }
1813
1814 root->fs_info->num_tolerated_disk_barrier_failures =
1815 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1816
1817 /*
1818 * at this point, the device is zero sized. We want to
1819 * remove it from the devices list and zero out the old super
1820 */
1821 if (clear_super && disk_super) {
1822 u64 bytenr;
1823 int i;
1824
1825 /* make sure this device isn't detected as part of
1826 * the FS anymore
1827 */
1828 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1829 set_buffer_dirty(bh);
1830 sync_dirty_buffer(bh);
1831
1832 /* clear the mirror copies of super block on the disk
1833 * being removed, 0th copy is been taken care above and
1834 * the below would take of the rest
1835 */
1836 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1837 bytenr = btrfs_sb_offset(i);
1838 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1839 i_size_read(bdev->bd_inode))
1840 break;
1841
1842 brelse(bh);
1843 bh = __bread(bdev, bytenr / 4096,
1844 BTRFS_SUPER_INFO_SIZE);
1845 if (!bh)
1846 continue;
1847
1848 disk_super = (struct btrfs_super_block *)bh->b_data;
1849
1850 if (btrfs_super_bytenr(disk_super) != bytenr ||
1851 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1852 continue;
1853 }
1854 memset(&disk_super->magic, 0,
1855 sizeof(disk_super->magic));
1856 set_buffer_dirty(bh);
1857 sync_dirty_buffer(bh);
1858 }
1859 }
1860
1861 ret = 0;
1862
1863 if (bdev) {
1864 /* Notify udev that device has changed */
1865 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1866
1867 /* Update ctime/mtime for device path for libblkid */
1868 update_dev_time(device_path);
1869 }
1870
1871 error_brelse:
1872 brelse(bh);
1873 if (bdev)
1874 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1875 out:
1876 mutex_unlock(&uuid_mutex);
1877 return ret;
1878 error_undo:
1879 if (device->writeable) {
1880 lock_chunks(root);
1881 list_add(&device->dev_alloc_list,
1882 &root->fs_info->fs_devices->alloc_list);
1883 device->fs_devices->rw_devices++;
1884 unlock_chunks(root);
1885 }
1886 goto error_brelse;
1887 }
1888
1889 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1890 struct btrfs_device *srcdev)
1891 {
1892 struct btrfs_fs_devices *fs_devices;
1893
1894 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1895
1896 /*
1897 * in case of fs with no seed, srcdev->fs_devices will point
1898 * to fs_devices of fs_info. However when the dev being replaced is
1899 * a seed dev it will point to the seed's local fs_devices. In short
1900 * srcdev will have its correct fs_devices in both the cases.
1901 */
1902 fs_devices = srcdev->fs_devices;
1903
1904 list_del_rcu(&srcdev->dev_list);
1905 list_del_rcu(&srcdev->dev_alloc_list);
1906 fs_devices->num_devices--;
1907 if (srcdev->missing)
1908 fs_devices->missing_devices--;
1909
1910 if (srcdev->writeable) {
1911 fs_devices->rw_devices--;
1912 /* zero out the old super if it is writable */
1913 btrfs_scratch_superblock(srcdev);
1914 }
1915
1916 if (srcdev->bdev)
1917 fs_devices->open_devices--;
1918 }
1919
1920 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1921 struct btrfs_device *srcdev)
1922 {
1923 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1924
1925 call_rcu(&srcdev->rcu, free_device);
1926
1927 /*
1928 * unless fs_devices is seed fs, num_devices shouldn't go
1929 * zero
1930 */
1931 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1932
1933 /* if this is no devs we rather delete the fs_devices */
1934 if (!fs_devices->num_devices) {
1935 struct btrfs_fs_devices *tmp_fs_devices;
1936
1937 tmp_fs_devices = fs_info->fs_devices;
1938 while (tmp_fs_devices) {
1939 if (tmp_fs_devices->seed == fs_devices) {
1940 tmp_fs_devices->seed = fs_devices->seed;
1941 break;
1942 }
1943 tmp_fs_devices = tmp_fs_devices->seed;
1944 }
1945 fs_devices->seed = NULL;
1946 __btrfs_close_devices(fs_devices);
1947 free_fs_devices(fs_devices);
1948 }
1949 }
1950
1951 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1952 struct btrfs_device *tgtdev)
1953 {
1954 struct btrfs_device *next_device;
1955
1956 mutex_lock(&uuid_mutex);
1957 WARN_ON(!tgtdev);
1958 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1959
1960 btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev);
1961
1962 if (tgtdev->bdev) {
1963 btrfs_scratch_superblock(tgtdev);
1964 fs_info->fs_devices->open_devices--;
1965 }
1966 fs_info->fs_devices->num_devices--;
1967
1968 next_device = list_entry(fs_info->fs_devices->devices.next,
1969 struct btrfs_device, dev_list);
1970 if (tgtdev->bdev == fs_info->sb->s_bdev)
1971 fs_info->sb->s_bdev = next_device->bdev;
1972 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1973 fs_info->fs_devices->latest_bdev = next_device->bdev;
1974 list_del_rcu(&tgtdev->dev_list);
1975
1976 call_rcu(&tgtdev->rcu, free_device);
1977
1978 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1979 mutex_unlock(&uuid_mutex);
1980 }
1981
1982 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1983 struct btrfs_device **device)
1984 {
1985 int ret = 0;
1986 struct btrfs_super_block *disk_super;
1987 u64 devid;
1988 u8 *dev_uuid;
1989 struct block_device *bdev;
1990 struct buffer_head *bh;
1991
1992 *device = NULL;
1993 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1994 root->fs_info->bdev_holder, 0, &bdev, &bh);
1995 if (ret)
1996 return ret;
1997 disk_super = (struct btrfs_super_block *)bh->b_data;
1998 devid = btrfs_stack_device_id(&disk_super->dev_item);
1999 dev_uuid = disk_super->dev_item.uuid;
2000 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2001 disk_super->fsid);
2002 brelse(bh);
2003 if (!*device)
2004 ret = -ENOENT;
2005 blkdev_put(bdev, FMODE_READ);
2006 return ret;
2007 }
2008
2009 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2010 char *device_path,
2011 struct btrfs_device **device)
2012 {
2013 *device = NULL;
2014 if (strcmp(device_path, "missing") == 0) {
2015 struct list_head *devices;
2016 struct btrfs_device *tmp;
2017
2018 devices = &root->fs_info->fs_devices->devices;
2019 /*
2020 * It is safe to read the devices since the volume_mutex
2021 * is held by the caller.
2022 */
2023 list_for_each_entry(tmp, devices, dev_list) {
2024 if (tmp->in_fs_metadata && !tmp->bdev) {
2025 *device = tmp;
2026 break;
2027 }
2028 }
2029
2030 if (!*device) {
2031 btrfs_err(root->fs_info, "no missing device found");
2032 return -ENOENT;
2033 }
2034
2035 return 0;
2036 } else {
2037 return btrfs_find_device_by_path(root, device_path, device);
2038 }
2039 }
2040
2041 /*
2042 * does all the dirty work required for changing file system's UUID.
2043 */
2044 static int btrfs_prepare_sprout(struct btrfs_root *root)
2045 {
2046 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2047 struct btrfs_fs_devices *old_devices;
2048 struct btrfs_fs_devices *seed_devices;
2049 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2050 struct btrfs_device *device;
2051 u64 super_flags;
2052
2053 BUG_ON(!mutex_is_locked(&uuid_mutex));
2054 if (!fs_devices->seeding)
2055 return -EINVAL;
2056
2057 seed_devices = __alloc_fs_devices();
2058 if (IS_ERR(seed_devices))
2059 return PTR_ERR(seed_devices);
2060
2061 old_devices = clone_fs_devices(fs_devices);
2062 if (IS_ERR(old_devices)) {
2063 kfree(seed_devices);
2064 return PTR_ERR(old_devices);
2065 }
2066
2067 list_add(&old_devices->list, &fs_uuids);
2068
2069 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2070 seed_devices->opened = 1;
2071 INIT_LIST_HEAD(&seed_devices->devices);
2072 INIT_LIST_HEAD(&seed_devices->alloc_list);
2073 mutex_init(&seed_devices->device_list_mutex);
2074
2075 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2076 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2077 synchronize_rcu);
2078 list_for_each_entry(device, &seed_devices->devices, dev_list)
2079 device->fs_devices = seed_devices;
2080
2081 lock_chunks(root);
2082 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2083 unlock_chunks(root);
2084
2085 fs_devices->seeding = 0;
2086 fs_devices->num_devices = 0;
2087 fs_devices->open_devices = 0;
2088 fs_devices->missing_devices = 0;
2089 fs_devices->rotating = 0;
2090 fs_devices->seed = seed_devices;
2091
2092 generate_random_uuid(fs_devices->fsid);
2093 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2094 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2095 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2096
2097 super_flags = btrfs_super_flags(disk_super) &
2098 ~BTRFS_SUPER_FLAG_SEEDING;
2099 btrfs_set_super_flags(disk_super, super_flags);
2100
2101 return 0;
2102 }
2103
2104 /*
2105 * strore the expected generation for seed devices in device items.
2106 */
2107 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *root)
2109 {
2110 struct btrfs_path *path;
2111 struct extent_buffer *leaf;
2112 struct btrfs_dev_item *dev_item;
2113 struct btrfs_device *device;
2114 struct btrfs_key key;
2115 u8 fs_uuid[BTRFS_UUID_SIZE];
2116 u8 dev_uuid[BTRFS_UUID_SIZE];
2117 u64 devid;
2118 int ret;
2119
2120 path = btrfs_alloc_path();
2121 if (!path)
2122 return -ENOMEM;
2123
2124 root = root->fs_info->chunk_root;
2125 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2126 key.offset = 0;
2127 key.type = BTRFS_DEV_ITEM_KEY;
2128
2129 while (1) {
2130 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2131 if (ret < 0)
2132 goto error;
2133
2134 leaf = path->nodes[0];
2135 next_slot:
2136 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2137 ret = btrfs_next_leaf(root, path);
2138 if (ret > 0)
2139 break;
2140 if (ret < 0)
2141 goto error;
2142 leaf = path->nodes[0];
2143 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2144 btrfs_release_path(path);
2145 continue;
2146 }
2147
2148 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2149 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2150 key.type != BTRFS_DEV_ITEM_KEY)
2151 break;
2152
2153 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2154 struct btrfs_dev_item);
2155 devid = btrfs_device_id(leaf, dev_item);
2156 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2157 BTRFS_UUID_SIZE);
2158 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2159 BTRFS_UUID_SIZE);
2160 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2161 fs_uuid);
2162 BUG_ON(!device); /* Logic error */
2163
2164 if (device->fs_devices->seeding) {
2165 btrfs_set_device_generation(leaf, dev_item,
2166 device->generation);
2167 btrfs_mark_buffer_dirty(leaf);
2168 }
2169
2170 path->slots[0]++;
2171 goto next_slot;
2172 }
2173 ret = 0;
2174 error:
2175 btrfs_free_path(path);
2176 return ret;
2177 }
2178
2179 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2180 {
2181 struct request_queue *q;
2182 struct btrfs_trans_handle *trans;
2183 struct btrfs_device *device;
2184 struct block_device *bdev;
2185 struct list_head *devices;
2186 struct super_block *sb = root->fs_info->sb;
2187 struct rcu_string *name;
2188 u64 tmp;
2189 int seeding_dev = 0;
2190 int ret = 0;
2191
2192 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2193 return -EROFS;
2194
2195 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2196 root->fs_info->bdev_holder);
2197 if (IS_ERR(bdev))
2198 return PTR_ERR(bdev);
2199
2200 if (root->fs_info->fs_devices->seeding) {
2201 seeding_dev = 1;
2202 down_write(&sb->s_umount);
2203 mutex_lock(&uuid_mutex);
2204 }
2205
2206 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2207
2208 devices = &root->fs_info->fs_devices->devices;
2209
2210 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2211 list_for_each_entry(device, devices, dev_list) {
2212 if (device->bdev == bdev) {
2213 ret = -EEXIST;
2214 mutex_unlock(
2215 &root->fs_info->fs_devices->device_list_mutex);
2216 goto error;
2217 }
2218 }
2219 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2220
2221 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2222 if (IS_ERR(device)) {
2223 /* we can safely leave the fs_devices entry around */
2224 ret = PTR_ERR(device);
2225 goto error;
2226 }
2227
2228 name = rcu_string_strdup(device_path, GFP_NOFS);
2229 if (!name) {
2230 kfree(device);
2231 ret = -ENOMEM;
2232 goto error;
2233 }
2234 rcu_assign_pointer(device->name, name);
2235
2236 trans = btrfs_start_transaction(root, 0);
2237 if (IS_ERR(trans)) {
2238 rcu_string_free(device->name);
2239 kfree(device);
2240 ret = PTR_ERR(trans);
2241 goto error;
2242 }
2243
2244 q = bdev_get_queue(bdev);
2245 if (blk_queue_discard(q))
2246 device->can_discard = 1;
2247 device->writeable = 1;
2248 device->generation = trans->transid;
2249 device->io_width = root->sectorsize;
2250 device->io_align = root->sectorsize;
2251 device->sector_size = root->sectorsize;
2252 device->total_bytes = i_size_read(bdev->bd_inode);
2253 device->disk_total_bytes = device->total_bytes;
2254 device->commit_total_bytes = device->total_bytes;
2255 device->dev_root = root->fs_info->dev_root;
2256 device->bdev = bdev;
2257 device->in_fs_metadata = 1;
2258 device->is_tgtdev_for_dev_replace = 0;
2259 device->mode = FMODE_EXCL;
2260 device->dev_stats_valid = 1;
2261 set_blocksize(device->bdev, 4096);
2262
2263 if (seeding_dev) {
2264 sb->s_flags &= ~MS_RDONLY;
2265 ret = btrfs_prepare_sprout(root);
2266 BUG_ON(ret); /* -ENOMEM */
2267 }
2268
2269 device->fs_devices = root->fs_info->fs_devices;
2270
2271 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2272 lock_chunks(root);
2273 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2274 list_add(&device->dev_alloc_list,
2275 &root->fs_info->fs_devices->alloc_list);
2276 root->fs_info->fs_devices->num_devices++;
2277 root->fs_info->fs_devices->open_devices++;
2278 root->fs_info->fs_devices->rw_devices++;
2279 root->fs_info->fs_devices->total_devices++;
2280 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2281
2282 spin_lock(&root->fs_info->free_chunk_lock);
2283 root->fs_info->free_chunk_space += device->total_bytes;
2284 spin_unlock(&root->fs_info->free_chunk_lock);
2285
2286 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2287 root->fs_info->fs_devices->rotating = 1;
2288
2289 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2290 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2291 tmp + device->total_bytes);
2292
2293 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2294 btrfs_set_super_num_devices(root->fs_info->super_copy,
2295 tmp + 1);
2296
2297 /* add sysfs device entry */
2298 btrfs_kobj_add_device(root->fs_info->fs_devices, device);
2299
2300 /*
2301 * we've got more storage, clear any full flags on the space
2302 * infos
2303 */
2304 btrfs_clear_space_info_full(root->fs_info);
2305
2306 unlock_chunks(root);
2307 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308
2309 if (seeding_dev) {
2310 lock_chunks(root);
2311 ret = init_first_rw_device(trans, root, device);
2312 unlock_chunks(root);
2313 if (ret) {
2314 btrfs_abort_transaction(trans, root, ret);
2315 goto error_trans;
2316 }
2317 }
2318
2319 ret = btrfs_add_device(trans, root, device);
2320 if (ret) {
2321 btrfs_abort_transaction(trans, root, ret);
2322 goto error_trans;
2323 }
2324
2325 if (seeding_dev) {
2326 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2327
2328 ret = btrfs_finish_sprout(trans, root);
2329 if (ret) {
2330 btrfs_abort_transaction(trans, root, ret);
2331 goto error_trans;
2332 }
2333
2334 /* Sprouting would change fsid of the mounted root,
2335 * so rename the fsid on the sysfs
2336 */
2337 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2338 root->fs_info->fsid);
2339 if (kobject_rename(&root->fs_info->fs_devices->super_kobj,
2340 fsid_buf))
2341 pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n");
2342 }
2343
2344 root->fs_info->num_tolerated_disk_barrier_failures =
2345 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2346 ret = btrfs_commit_transaction(trans, root);
2347
2348 if (seeding_dev) {
2349 mutex_unlock(&uuid_mutex);
2350 up_write(&sb->s_umount);
2351
2352 if (ret) /* transaction commit */
2353 return ret;
2354
2355 ret = btrfs_relocate_sys_chunks(root);
2356 if (ret < 0)
2357 btrfs_error(root->fs_info, ret,
2358 "Failed to relocate sys chunks after "
2359 "device initialization. This can be fixed "
2360 "using the \"btrfs balance\" command.");
2361 trans = btrfs_attach_transaction(root);
2362 if (IS_ERR(trans)) {
2363 if (PTR_ERR(trans) == -ENOENT)
2364 return 0;
2365 return PTR_ERR(trans);
2366 }
2367 ret = btrfs_commit_transaction(trans, root);
2368 }
2369
2370 /* Update ctime/mtime for libblkid */
2371 update_dev_time(device_path);
2372 return ret;
2373
2374 error_trans:
2375 btrfs_end_transaction(trans, root);
2376 rcu_string_free(device->name);
2377 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
2378 kfree(device);
2379 error:
2380 blkdev_put(bdev, FMODE_EXCL);
2381 if (seeding_dev) {
2382 mutex_unlock(&uuid_mutex);
2383 up_write(&sb->s_umount);
2384 }
2385 return ret;
2386 }
2387
2388 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2389 struct btrfs_device *srcdev,
2390 struct btrfs_device **device_out)
2391 {
2392 struct request_queue *q;
2393 struct btrfs_device *device;
2394 struct block_device *bdev;
2395 struct btrfs_fs_info *fs_info = root->fs_info;
2396 struct list_head *devices;
2397 struct rcu_string *name;
2398 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2399 int ret = 0;
2400
2401 *device_out = NULL;
2402 if (fs_info->fs_devices->seeding) {
2403 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2404 return -EINVAL;
2405 }
2406
2407 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2408 fs_info->bdev_holder);
2409 if (IS_ERR(bdev)) {
2410 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2411 return PTR_ERR(bdev);
2412 }
2413
2414 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2415
2416 devices = &fs_info->fs_devices->devices;
2417 list_for_each_entry(device, devices, dev_list) {
2418 if (device->bdev == bdev) {
2419 btrfs_err(fs_info, "target device is in the filesystem!");
2420 ret = -EEXIST;
2421 goto error;
2422 }
2423 }
2424
2425
2426 if (i_size_read(bdev->bd_inode) <
2427 btrfs_device_get_total_bytes(srcdev)) {
2428 btrfs_err(fs_info, "target device is smaller than source device!");
2429 ret = -EINVAL;
2430 goto error;
2431 }
2432
2433
2434 device = btrfs_alloc_device(NULL, &devid, NULL);
2435 if (IS_ERR(device)) {
2436 ret = PTR_ERR(device);
2437 goto error;
2438 }
2439
2440 name = rcu_string_strdup(device_path, GFP_NOFS);
2441 if (!name) {
2442 kfree(device);
2443 ret = -ENOMEM;
2444 goto error;
2445 }
2446 rcu_assign_pointer(device->name, name);
2447
2448 q = bdev_get_queue(bdev);
2449 if (blk_queue_discard(q))
2450 device->can_discard = 1;
2451 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2452 device->writeable = 1;
2453 device->generation = 0;
2454 device->io_width = root->sectorsize;
2455 device->io_align = root->sectorsize;
2456 device->sector_size = root->sectorsize;
2457 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2458 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2459 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2460 ASSERT(list_empty(&srcdev->resized_list));
2461 device->commit_total_bytes = srcdev->commit_total_bytes;
2462 device->commit_bytes_used = device->bytes_used;
2463 device->dev_root = fs_info->dev_root;
2464 device->bdev = bdev;
2465 device->in_fs_metadata = 1;
2466 device->is_tgtdev_for_dev_replace = 1;
2467 device->mode = FMODE_EXCL;
2468 device->dev_stats_valid = 1;
2469 set_blocksize(device->bdev, 4096);
2470 device->fs_devices = fs_info->fs_devices;
2471 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2472 fs_info->fs_devices->num_devices++;
2473 fs_info->fs_devices->open_devices++;
2474 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2475
2476 *device_out = device;
2477 return ret;
2478
2479 error:
2480 blkdev_put(bdev, FMODE_EXCL);
2481 return ret;
2482 }
2483
2484 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2485 struct btrfs_device *tgtdev)
2486 {
2487 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2488 tgtdev->io_width = fs_info->dev_root->sectorsize;
2489 tgtdev->io_align = fs_info->dev_root->sectorsize;
2490 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2491 tgtdev->dev_root = fs_info->dev_root;
2492 tgtdev->in_fs_metadata = 1;
2493 }
2494
2495 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2496 struct btrfs_device *device)
2497 {
2498 int ret;
2499 struct btrfs_path *path;
2500 struct btrfs_root *root;
2501 struct btrfs_dev_item *dev_item;
2502 struct extent_buffer *leaf;
2503 struct btrfs_key key;
2504
2505 root = device->dev_root->fs_info->chunk_root;
2506
2507 path = btrfs_alloc_path();
2508 if (!path)
2509 return -ENOMEM;
2510
2511 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2512 key.type = BTRFS_DEV_ITEM_KEY;
2513 key.offset = device->devid;
2514
2515 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2516 if (ret < 0)
2517 goto out;
2518
2519 if (ret > 0) {
2520 ret = -ENOENT;
2521 goto out;
2522 }
2523
2524 leaf = path->nodes[0];
2525 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2526
2527 btrfs_set_device_id(leaf, dev_item, device->devid);
2528 btrfs_set_device_type(leaf, dev_item, device->type);
2529 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2530 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2531 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2532 btrfs_set_device_total_bytes(leaf, dev_item,
2533 btrfs_device_get_disk_total_bytes(device));
2534 btrfs_set_device_bytes_used(leaf, dev_item,
2535 btrfs_device_get_bytes_used(device));
2536 btrfs_mark_buffer_dirty(leaf);
2537
2538 out:
2539 btrfs_free_path(path);
2540 return ret;
2541 }
2542
2543 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2544 struct btrfs_device *device, u64 new_size)
2545 {
2546 struct btrfs_super_block *super_copy =
2547 device->dev_root->fs_info->super_copy;
2548 struct btrfs_fs_devices *fs_devices;
2549 u64 old_total;
2550 u64 diff;
2551
2552 if (!device->writeable)
2553 return -EACCES;
2554
2555 lock_chunks(device->dev_root);
2556 old_total = btrfs_super_total_bytes(super_copy);
2557 diff = new_size - device->total_bytes;
2558
2559 if (new_size <= device->total_bytes ||
2560 device->is_tgtdev_for_dev_replace) {
2561 unlock_chunks(device->dev_root);
2562 return -EINVAL;
2563 }
2564
2565 fs_devices = device->dev_root->fs_info->fs_devices;
2566
2567 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2568 device->fs_devices->total_rw_bytes += diff;
2569
2570 btrfs_device_set_total_bytes(device, new_size);
2571 btrfs_device_set_disk_total_bytes(device, new_size);
2572 btrfs_clear_space_info_full(device->dev_root->fs_info);
2573 if (list_empty(&device->resized_list))
2574 list_add_tail(&device->resized_list,
2575 &fs_devices->resized_devices);
2576 unlock_chunks(device->dev_root);
2577
2578 return btrfs_update_device(trans, device);
2579 }
2580
2581 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2582 struct btrfs_root *root, u64 chunk_objectid,
2583 u64 chunk_offset)
2584 {
2585 int ret;
2586 struct btrfs_path *path;
2587 struct btrfs_key key;
2588
2589 root = root->fs_info->chunk_root;
2590 path = btrfs_alloc_path();
2591 if (!path)
2592 return -ENOMEM;
2593
2594 key.objectid = chunk_objectid;
2595 key.offset = chunk_offset;
2596 key.type = BTRFS_CHUNK_ITEM_KEY;
2597
2598 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2599 if (ret < 0)
2600 goto out;
2601 else if (ret > 0) { /* Logic error or corruption */
2602 btrfs_error(root->fs_info, -ENOENT,
2603 "Failed lookup while freeing chunk.");
2604 ret = -ENOENT;
2605 goto out;
2606 }
2607
2608 ret = btrfs_del_item(trans, root, path);
2609 if (ret < 0)
2610 btrfs_error(root->fs_info, ret,
2611 "Failed to delete chunk item.");
2612 out:
2613 btrfs_free_path(path);
2614 return ret;
2615 }
2616
2617 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2618 chunk_offset)
2619 {
2620 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2621 struct btrfs_disk_key *disk_key;
2622 struct btrfs_chunk *chunk;
2623 u8 *ptr;
2624 int ret = 0;
2625 u32 num_stripes;
2626 u32 array_size;
2627 u32 len = 0;
2628 u32 cur;
2629 struct btrfs_key key;
2630
2631 lock_chunks(root);
2632 array_size = btrfs_super_sys_array_size(super_copy);
2633
2634 ptr = super_copy->sys_chunk_array;
2635 cur = 0;
2636
2637 while (cur < array_size) {
2638 disk_key = (struct btrfs_disk_key *)ptr;
2639 btrfs_disk_key_to_cpu(&key, disk_key);
2640
2641 len = sizeof(*disk_key);
2642
2643 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2644 chunk = (struct btrfs_chunk *)(ptr + len);
2645 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2646 len += btrfs_chunk_item_size(num_stripes);
2647 } else {
2648 ret = -EIO;
2649 break;
2650 }
2651 if (key.objectid == chunk_objectid &&
2652 key.offset == chunk_offset) {
2653 memmove(ptr, ptr + len, array_size - (cur + len));
2654 array_size -= len;
2655 btrfs_set_super_sys_array_size(super_copy, array_size);
2656 } else {
2657 ptr += len;
2658 cur += len;
2659 }
2660 }
2661 unlock_chunks(root);
2662 return ret;
2663 }
2664
2665 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *root, u64 chunk_offset)
2667 {
2668 struct extent_map_tree *em_tree;
2669 struct extent_map *em;
2670 struct btrfs_root *extent_root = root->fs_info->extent_root;
2671 struct map_lookup *map;
2672 u64 dev_extent_len = 0;
2673 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2674 int i, ret = 0;
2675
2676 /* Just in case */
2677 root = root->fs_info->chunk_root;
2678 em_tree = &root->fs_info->mapping_tree.map_tree;
2679
2680 read_lock(&em_tree->lock);
2681 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2682 read_unlock(&em_tree->lock);
2683
2684 if (!em || em->start > chunk_offset ||
2685 em->start + em->len < chunk_offset) {
2686 /*
2687 * This is a logic error, but we don't want to just rely on the
2688 * user having built with ASSERT enabled, so if ASSERT doens't
2689 * do anything we still error out.
2690 */
2691 ASSERT(0);
2692 if (em)
2693 free_extent_map(em);
2694 return -EINVAL;
2695 }
2696 map = (struct map_lookup *)em->bdev;
2697 lock_chunks(root->fs_info->chunk_root);
2698 check_system_chunk(trans, extent_root, map->type);
2699 unlock_chunks(root->fs_info->chunk_root);
2700
2701 for (i = 0; i < map->num_stripes; i++) {
2702 struct btrfs_device *device = map->stripes[i].dev;
2703 ret = btrfs_free_dev_extent(trans, device,
2704 map->stripes[i].physical,
2705 &dev_extent_len);
2706 if (ret) {
2707 btrfs_abort_transaction(trans, root, ret);
2708 goto out;
2709 }
2710
2711 if (device->bytes_used > 0) {
2712 lock_chunks(root);
2713 btrfs_device_set_bytes_used(device,
2714 device->bytes_used - dev_extent_len);
2715 spin_lock(&root->fs_info->free_chunk_lock);
2716 root->fs_info->free_chunk_space += dev_extent_len;
2717 spin_unlock(&root->fs_info->free_chunk_lock);
2718 btrfs_clear_space_info_full(root->fs_info);
2719 unlock_chunks(root);
2720 }
2721
2722 if (map->stripes[i].dev) {
2723 ret = btrfs_update_device(trans, map->stripes[i].dev);
2724 if (ret) {
2725 btrfs_abort_transaction(trans, root, ret);
2726 goto out;
2727 }
2728 }
2729 }
2730 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2731 if (ret) {
2732 btrfs_abort_transaction(trans, root, ret);
2733 goto out;
2734 }
2735
2736 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2737
2738 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2739 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2740 if (ret) {
2741 btrfs_abort_transaction(trans, root, ret);
2742 goto out;
2743 }
2744 }
2745
2746 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2747 if (ret) {
2748 btrfs_abort_transaction(trans, extent_root, ret);
2749 goto out;
2750 }
2751
2752 out:
2753 /* once for us */
2754 free_extent_map(em);
2755 return ret;
2756 }
2757
2758 static int btrfs_relocate_chunk(struct btrfs_root *root,
2759 u64 chunk_objectid,
2760 u64 chunk_offset)
2761 {
2762 struct btrfs_root *extent_root;
2763 struct btrfs_trans_handle *trans;
2764 int ret;
2765
2766 root = root->fs_info->chunk_root;
2767 extent_root = root->fs_info->extent_root;
2768
2769 /*
2770 * Prevent races with automatic removal of unused block groups.
2771 * After we relocate and before we remove the chunk with offset
2772 * chunk_offset, automatic removal of the block group can kick in,
2773 * resulting in a failure when calling btrfs_remove_chunk() below.
2774 *
2775 * Make sure to acquire this mutex before doing a tree search (dev
2776 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2777 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2778 * we release the path used to search the chunk/dev tree and before
2779 * the current task acquires this mutex and calls us.
2780 */
2781 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2782
2783 ret = btrfs_can_relocate(extent_root, chunk_offset);
2784 if (ret)
2785 return -ENOSPC;
2786
2787 /* step one, relocate all the extents inside this chunk */
2788 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2789 if (ret)
2790 return ret;
2791
2792 trans = btrfs_start_transaction(root, 0);
2793 if (IS_ERR(trans)) {
2794 ret = PTR_ERR(trans);
2795 btrfs_std_error(root->fs_info, ret);
2796 return ret;
2797 }
2798
2799 /*
2800 * step two, delete the device extents and the
2801 * chunk tree entries
2802 */
2803 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2804 btrfs_end_transaction(trans, root);
2805 return ret;
2806 }
2807
2808 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2809 {
2810 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2811 struct btrfs_path *path;
2812 struct extent_buffer *leaf;
2813 struct btrfs_chunk *chunk;
2814 struct btrfs_key key;
2815 struct btrfs_key found_key;
2816 u64 chunk_type;
2817 bool retried = false;
2818 int failed = 0;
2819 int ret;
2820
2821 path = btrfs_alloc_path();
2822 if (!path)
2823 return -ENOMEM;
2824
2825 again:
2826 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2827 key.offset = (u64)-1;
2828 key.type = BTRFS_CHUNK_ITEM_KEY;
2829
2830 while (1) {
2831 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2832 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2833 if (ret < 0) {
2834 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2835 goto error;
2836 }
2837 BUG_ON(ret == 0); /* Corruption */
2838
2839 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2840 key.type);
2841 if (ret)
2842 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2843 if (ret < 0)
2844 goto error;
2845 if (ret > 0)
2846 break;
2847
2848 leaf = path->nodes[0];
2849 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2850
2851 chunk = btrfs_item_ptr(leaf, path->slots[0],
2852 struct btrfs_chunk);
2853 chunk_type = btrfs_chunk_type(leaf, chunk);
2854 btrfs_release_path(path);
2855
2856 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2857 ret = btrfs_relocate_chunk(chunk_root,
2858 found_key.objectid,
2859 found_key.offset);
2860 if (ret == -ENOSPC)
2861 failed++;
2862 else
2863 BUG_ON(ret);
2864 }
2865 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2866
2867 if (found_key.offset == 0)
2868 break;
2869 key.offset = found_key.offset - 1;
2870 }
2871 ret = 0;
2872 if (failed && !retried) {
2873 failed = 0;
2874 retried = true;
2875 goto again;
2876 } else if (WARN_ON(failed && retried)) {
2877 ret = -ENOSPC;
2878 }
2879 error:
2880 btrfs_free_path(path);
2881 return ret;
2882 }
2883
2884 static int insert_balance_item(struct btrfs_root *root,
2885 struct btrfs_balance_control *bctl)
2886 {
2887 struct btrfs_trans_handle *trans;
2888 struct btrfs_balance_item *item;
2889 struct btrfs_disk_balance_args disk_bargs;
2890 struct btrfs_path *path;
2891 struct extent_buffer *leaf;
2892 struct btrfs_key key;
2893 int ret, err;
2894
2895 path = btrfs_alloc_path();
2896 if (!path)
2897 return -ENOMEM;
2898
2899 trans = btrfs_start_transaction(root, 0);
2900 if (IS_ERR(trans)) {
2901 btrfs_free_path(path);
2902 return PTR_ERR(trans);
2903 }
2904
2905 key.objectid = BTRFS_BALANCE_OBJECTID;
2906 key.type = BTRFS_BALANCE_ITEM_KEY;
2907 key.offset = 0;
2908
2909 ret = btrfs_insert_empty_item(trans, root, path, &key,
2910 sizeof(*item));
2911 if (ret)
2912 goto out;
2913
2914 leaf = path->nodes[0];
2915 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2916
2917 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2918
2919 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2920 btrfs_set_balance_data(leaf, item, &disk_bargs);
2921 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2922 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2923 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2924 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2925
2926 btrfs_set_balance_flags(leaf, item, bctl->flags);
2927
2928 btrfs_mark_buffer_dirty(leaf);
2929 out:
2930 btrfs_free_path(path);
2931 err = btrfs_commit_transaction(trans, root);
2932 if (err && !ret)
2933 ret = err;
2934 return ret;
2935 }
2936
2937 static int del_balance_item(struct btrfs_root *root)
2938 {
2939 struct btrfs_trans_handle *trans;
2940 struct btrfs_path *path;
2941 struct btrfs_key key;
2942 int ret, err;
2943
2944 path = btrfs_alloc_path();
2945 if (!path)
2946 return -ENOMEM;
2947
2948 trans = btrfs_start_transaction(root, 0);
2949 if (IS_ERR(trans)) {
2950 btrfs_free_path(path);
2951 return PTR_ERR(trans);
2952 }
2953
2954 key.objectid = BTRFS_BALANCE_OBJECTID;
2955 key.type = BTRFS_BALANCE_ITEM_KEY;
2956 key.offset = 0;
2957
2958 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2959 if (ret < 0)
2960 goto out;
2961 if (ret > 0) {
2962 ret = -ENOENT;
2963 goto out;
2964 }
2965
2966 ret = btrfs_del_item(trans, root, path);
2967 out:
2968 btrfs_free_path(path);
2969 err = btrfs_commit_transaction(trans, root);
2970 if (err && !ret)
2971 ret = err;
2972 return ret;
2973 }
2974
2975 /*
2976 * This is a heuristic used to reduce the number of chunks balanced on
2977 * resume after balance was interrupted.
2978 */
2979 static void update_balance_args(struct btrfs_balance_control *bctl)
2980 {
2981 /*
2982 * Turn on soft mode for chunk types that were being converted.
2983 */
2984 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2985 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2986 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2987 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2988 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2989 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2990
2991 /*
2992 * Turn on usage filter if is not already used. The idea is
2993 * that chunks that we have already balanced should be
2994 * reasonably full. Don't do it for chunks that are being
2995 * converted - that will keep us from relocating unconverted
2996 * (albeit full) chunks.
2997 */
2998 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2999 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3000 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3001 bctl->data.usage = 90;
3002 }
3003 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3004 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3005 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3006 bctl->sys.usage = 90;
3007 }
3008 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3009 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3010 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3011 bctl->meta.usage = 90;
3012 }
3013 }
3014
3015 /*
3016 * Should be called with both balance and volume mutexes held to
3017 * serialize other volume operations (add_dev/rm_dev/resize) with
3018 * restriper. Same goes for unset_balance_control.
3019 */
3020 static void set_balance_control(struct btrfs_balance_control *bctl)
3021 {
3022 struct btrfs_fs_info *fs_info = bctl->fs_info;
3023
3024 BUG_ON(fs_info->balance_ctl);
3025
3026 spin_lock(&fs_info->balance_lock);
3027 fs_info->balance_ctl = bctl;
3028 spin_unlock(&fs_info->balance_lock);
3029 }
3030
3031 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3032 {
3033 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3034
3035 BUG_ON(!fs_info->balance_ctl);
3036
3037 spin_lock(&fs_info->balance_lock);
3038 fs_info->balance_ctl = NULL;
3039 spin_unlock(&fs_info->balance_lock);
3040
3041 kfree(bctl);
3042 }
3043
3044 /*
3045 * Balance filters. Return 1 if chunk should be filtered out
3046 * (should not be balanced).
3047 */
3048 static int chunk_profiles_filter(u64 chunk_type,
3049 struct btrfs_balance_args *bargs)
3050 {
3051 chunk_type = chunk_to_extended(chunk_type) &
3052 BTRFS_EXTENDED_PROFILE_MASK;
3053
3054 if (bargs->profiles & chunk_type)
3055 return 0;
3056
3057 return 1;
3058 }
3059
3060 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3061 struct btrfs_balance_args *bargs)
3062 {
3063 struct btrfs_block_group_cache *cache;
3064 u64 chunk_used, user_thresh;
3065 int ret = 1;
3066
3067 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3068 chunk_used = btrfs_block_group_used(&cache->item);
3069
3070 if (bargs->usage == 0)
3071 user_thresh = 1;
3072 else if (bargs->usage > 100)
3073 user_thresh = cache->key.offset;
3074 else
3075 user_thresh = div_factor_fine(cache->key.offset,
3076 bargs->usage);
3077
3078 if (chunk_used < user_thresh)
3079 ret = 0;
3080
3081 btrfs_put_block_group(cache);
3082 return ret;
3083 }
3084
3085 static int chunk_devid_filter(struct extent_buffer *leaf,
3086 struct btrfs_chunk *chunk,
3087 struct btrfs_balance_args *bargs)
3088 {
3089 struct btrfs_stripe *stripe;
3090 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3091 int i;
3092
3093 for (i = 0; i < num_stripes; i++) {
3094 stripe = btrfs_stripe_nr(chunk, i);
3095 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3096 return 0;
3097 }
3098
3099 return 1;
3100 }
3101
3102 /* [pstart, pend) */
3103 static int chunk_drange_filter(struct extent_buffer *leaf,
3104 struct btrfs_chunk *chunk,
3105 u64 chunk_offset,
3106 struct btrfs_balance_args *bargs)
3107 {
3108 struct btrfs_stripe *stripe;
3109 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3110 u64 stripe_offset;
3111 u64 stripe_length;
3112 int factor;
3113 int i;
3114
3115 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3116 return 0;
3117
3118 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3119 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3120 factor = num_stripes / 2;
3121 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3122 factor = num_stripes - 1;
3123 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3124 factor = num_stripes - 2;
3125 } else {
3126 factor = num_stripes;
3127 }
3128
3129 for (i = 0; i < num_stripes; i++) {
3130 stripe = btrfs_stripe_nr(chunk, i);
3131 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3132 continue;
3133
3134 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3135 stripe_length = btrfs_chunk_length(leaf, chunk);
3136 stripe_length = div_u64(stripe_length, factor);
3137
3138 if (stripe_offset < bargs->pend &&
3139 stripe_offset + stripe_length > bargs->pstart)
3140 return 0;
3141 }
3142
3143 return 1;
3144 }
3145
3146 /* [vstart, vend) */
3147 static int chunk_vrange_filter(struct extent_buffer *leaf,
3148 struct btrfs_chunk *chunk,
3149 u64 chunk_offset,
3150 struct btrfs_balance_args *bargs)
3151 {
3152 if (chunk_offset < bargs->vend &&
3153 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3154 /* at least part of the chunk is inside this vrange */
3155 return 0;
3156
3157 return 1;
3158 }
3159
3160 static int chunk_soft_convert_filter(u64 chunk_type,
3161 struct btrfs_balance_args *bargs)
3162 {
3163 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3164 return 0;
3165
3166 chunk_type = chunk_to_extended(chunk_type) &
3167 BTRFS_EXTENDED_PROFILE_MASK;
3168
3169 if (bargs->target == chunk_type)
3170 return 1;
3171
3172 return 0;
3173 }
3174
3175 static int should_balance_chunk(struct btrfs_root *root,
3176 struct extent_buffer *leaf,
3177 struct btrfs_chunk *chunk, u64 chunk_offset)
3178 {
3179 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3180 struct btrfs_balance_args *bargs = NULL;
3181 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3182
3183 /* type filter */
3184 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3185 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3186 return 0;
3187 }
3188
3189 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3190 bargs = &bctl->data;
3191 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3192 bargs = &bctl->sys;
3193 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3194 bargs = &bctl->meta;
3195
3196 /* profiles filter */
3197 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3198 chunk_profiles_filter(chunk_type, bargs)) {
3199 return 0;
3200 }
3201
3202 /* usage filter */
3203 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3204 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3205 return 0;
3206 }
3207
3208 /* devid filter */
3209 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3210 chunk_devid_filter(leaf, chunk, bargs)) {
3211 return 0;
3212 }
3213
3214 /* drange filter, makes sense only with devid filter */
3215 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3216 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3217 return 0;
3218 }
3219
3220 /* vrange filter */
3221 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3222 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3223 return 0;
3224 }
3225
3226 /* soft profile changing mode */
3227 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3228 chunk_soft_convert_filter(chunk_type, bargs)) {
3229 return 0;
3230 }
3231
3232 /*
3233 * limited by count, must be the last filter
3234 */
3235 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3236 if (bargs->limit == 0)
3237 return 0;
3238 else
3239 bargs->limit--;
3240 }
3241
3242 return 1;
3243 }
3244
3245 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3246 {
3247 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3248 struct btrfs_root *chunk_root = fs_info->chunk_root;
3249 struct btrfs_root *dev_root = fs_info->dev_root;
3250 struct list_head *devices;
3251 struct btrfs_device *device;
3252 u64 old_size;
3253 u64 size_to_free;
3254 struct btrfs_chunk *chunk;
3255 struct btrfs_path *path;
3256 struct btrfs_key key;
3257 struct btrfs_key found_key;
3258 struct btrfs_trans_handle *trans;
3259 struct extent_buffer *leaf;
3260 int slot;
3261 int ret;
3262 int enospc_errors = 0;
3263 bool counting = true;
3264 u64 limit_data = bctl->data.limit;
3265 u64 limit_meta = bctl->meta.limit;
3266 u64 limit_sys = bctl->sys.limit;
3267
3268 /* step one make some room on all the devices */
3269 devices = &fs_info->fs_devices->devices;
3270 list_for_each_entry(device, devices, dev_list) {
3271 old_size = btrfs_device_get_total_bytes(device);
3272 size_to_free = div_factor(old_size, 1);
3273 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3274 if (!device->writeable ||
3275 btrfs_device_get_total_bytes(device) -
3276 btrfs_device_get_bytes_used(device) > size_to_free ||
3277 device->is_tgtdev_for_dev_replace)
3278 continue;
3279
3280 ret = btrfs_shrink_device(device, old_size - size_to_free);
3281 if (ret == -ENOSPC)
3282 break;
3283 BUG_ON(ret);
3284
3285 trans = btrfs_start_transaction(dev_root, 0);
3286 BUG_ON(IS_ERR(trans));
3287
3288 ret = btrfs_grow_device(trans, device, old_size);
3289 BUG_ON(ret);
3290
3291 btrfs_end_transaction(trans, dev_root);
3292 }
3293
3294 /* step two, relocate all the chunks */
3295 path = btrfs_alloc_path();
3296 if (!path) {
3297 ret = -ENOMEM;
3298 goto error;
3299 }
3300
3301 /* zero out stat counters */
3302 spin_lock(&fs_info->balance_lock);
3303 memset(&bctl->stat, 0, sizeof(bctl->stat));
3304 spin_unlock(&fs_info->balance_lock);
3305 again:
3306 if (!counting) {
3307 bctl->data.limit = limit_data;
3308 bctl->meta.limit = limit_meta;
3309 bctl->sys.limit = limit_sys;
3310 }
3311 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3312 key.offset = (u64)-1;
3313 key.type = BTRFS_CHUNK_ITEM_KEY;
3314
3315 while (1) {
3316 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3317 atomic_read(&fs_info->balance_cancel_req)) {
3318 ret = -ECANCELED;
3319 goto error;
3320 }
3321
3322 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3323 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3324 if (ret < 0) {
3325 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3326 goto error;
3327 }
3328
3329 /*
3330 * this shouldn't happen, it means the last relocate
3331 * failed
3332 */
3333 if (ret == 0)
3334 BUG(); /* FIXME break ? */
3335
3336 ret = btrfs_previous_item(chunk_root, path, 0,
3337 BTRFS_CHUNK_ITEM_KEY);
3338 if (ret) {
3339 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3340 ret = 0;
3341 break;
3342 }
3343
3344 leaf = path->nodes[0];
3345 slot = path->slots[0];
3346 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3347
3348 if (found_key.objectid != key.objectid) {
3349 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3350 break;
3351 }
3352
3353 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3354
3355 if (!counting) {
3356 spin_lock(&fs_info->balance_lock);
3357 bctl->stat.considered++;
3358 spin_unlock(&fs_info->balance_lock);
3359 }
3360
3361 ret = should_balance_chunk(chunk_root, leaf, chunk,
3362 found_key.offset);
3363 btrfs_release_path(path);
3364 if (!ret) {
3365 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3366 goto loop;
3367 }
3368
3369 if (counting) {
3370 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3371 spin_lock(&fs_info->balance_lock);
3372 bctl->stat.expected++;
3373 spin_unlock(&fs_info->balance_lock);
3374 goto loop;
3375 }
3376
3377 ret = btrfs_relocate_chunk(chunk_root,
3378 found_key.objectid,
3379 found_key.offset);
3380 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3381 if (ret && ret != -ENOSPC)
3382 goto error;
3383 if (ret == -ENOSPC) {
3384 enospc_errors++;
3385 } else {
3386 spin_lock(&fs_info->balance_lock);
3387 bctl->stat.completed++;
3388 spin_unlock(&fs_info->balance_lock);
3389 }
3390 loop:
3391 if (found_key.offset == 0)
3392 break;
3393 key.offset = found_key.offset - 1;
3394 }
3395
3396 if (counting) {
3397 btrfs_release_path(path);
3398 counting = false;
3399 goto again;
3400 }
3401 error:
3402 btrfs_free_path(path);
3403 if (enospc_errors) {
3404 btrfs_info(fs_info, "%d enospc errors during balance",
3405 enospc_errors);
3406 if (!ret)
3407 ret = -ENOSPC;
3408 }
3409
3410 return ret;
3411 }
3412
3413 /**
3414 * alloc_profile_is_valid - see if a given profile is valid and reduced
3415 * @flags: profile to validate
3416 * @extended: if true @flags is treated as an extended profile
3417 */
3418 static int alloc_profile_is_valid(u64 flags, int extended)
3419 {
3420 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3421 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3422
3423 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3424
3425 /* 1) check that all other bits are zeroed */
3426 if (flags & ~mask)
3427 return 0;
3428
3429 /* 2) see if profile is reduced */
3430 if (flags == 0)
3431 return !extended; /* "0" is valid for usual profiles */
3432
3433 /* true if exactly one bit set */
3434 return (flags & (flags - 1)) == 0;
3435 }
3436
3437 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3438 {
3439 /* cancel requested || normal exit path */
3440 return atomic_read(&fs_info->balance_cancel_req) ||
3441 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3442 atomic_read(&fs_info->balance_cancel_req) == 0);
3443 }
3444
3445 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3446 {
3447 int ret;
3448
3449 unset_balance_control(fs_info);
3450 ret = del_balance_item(fs_info->tree_root);
3451 if (ret)
3452 btrfs_std_error(fs_info, ret);
3453
3454 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3455 }
3456
3457 /*
3458 * Should be called with both balance and volume mutexes held
3459 */
3460 int btrfs_balance(struct btrfs_balance_control *bctl,
3461 struct btrfs_ioctl_balance_args *bargs)
3462 {
3463 struct btrfs_fs_info *fs_info = bctl->fs_info;
3464 u64 allowed;
3465 int mixed = 0;
3466 int ret;
3467 u64 num_devices;
3468 unsigned seq;
3469
3470 if (btrfs_fs_closing(fs_info) ||
3471 atomic_read(&fs_info->balance_pause_req) ||
3472 atomic_read(&fs_info->balance_cancel_req)) {
3473 ret = -EINVAL;
3474 goto out;
3475 }
3476
3477 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3478 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3479 mixed = 1;
3480
3481 /*
3482 * In case of mixed groups both data and meta should be picked,
3483 * and identical options should be given for both of them.
3484 */
3485 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3486 if (mixed && (bctl->flags & allowed)) {
3487 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3488 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3489 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3490 btrfs_err(fs_info, "with mixed groups data and "
3491 "metadata balance options must be the same");
3492 ret = -EINVAL;
3493 goto out;
3494 }
3495 }
3496
3497 num_devices = fs_info->fs_devices->num_devices;
3498 btrfs_dev_replace_lock(&fs_info->dev_replace);
3499 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3500 BUG_ON(num_devices < 1);
3501 num_devices--;
3502 }
3503 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3504 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3505 if (num_devices == 1)
3506 allowed |= BTRFS_BLOCK_GROUP_DUP;
3507 else if (num_devices > 1)
3508 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3509 if (num_devices > 2)
3510 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3511 if (num_devices > 3)
3512 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3513 BTRFS_BLOCK_GROUP_RAID6);
3514 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3515 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3516 (bctl->data.target & ~allowed))) {
3517 btrfs_err(fs_info, "unable to start balance with target "
3518 "data profile %llu",
3519 bctl->data.target);
3520 ret = -EINVAL;
3521 goto out;
3522 }
3523 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3524 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3525 (bctl->meta.target & ~allowed))) {
3526 btrfs_err(fs_info,
3527 "unable to start balance with target metadata profile %llu",
3528 bctl->meta.target);
3529 ret = -EINVAL;
3530 goto out;
3531 }
3532 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3533 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3534 (bctl->sys.target & ~allowed))) {
3535 btrfs_err(fs_info,
3536 "unable to start balance with target system profile %llu",
3537 bctl->sys.target);
3538 ret = -EINVAL;
3539 goto out;
3540 }
3541
3542 /* allow dup'ed data chunks only in mixed mode */
3543 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3544 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3545 btrfs_err(fs_info, "dup for data is not allowed");
3546 ret = -EINVAL;
3547 goto out;
3548 }
3549
3550 /* allow to reduce meta or sys integrity only if force set */
3551 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3552 BTRFS_BLOCK_GROUP_RAID10 |
3553 BTRFS_BLOCK_GROUP_RAID5 |
3554 BTRFS_BLOCK_GROUP_RAID6;
3555 do {
3556 seq = read_seqbegin(&fs_info->profiles_lock);
3557
3558 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3559 (fs_info->avail_system_alloc_bits & allowed) &&
3560 !(bctl->sys.target & allowed)) ||
3561 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3562 (fs_info->avail_metadata_alloc_bits & allowed) &&
3563 !(bctl->meta.target & allowed))) {
3564 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3565 btrfs_info(fs_info, "force reducing metadata integrity");
3566 } else {
3567 btrfs_err(fs_info, "balance will reduce metadata "
3568 "integrity, use force if you want this");
3569 ret = -EINVAL;
3570 goto out;
3571 }
3572 }
3573 } while (read_seqretry(&fs_info->profiles_lock, seq));
3574
3575 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3576 int num_tolerated_disk_barrier_failures;
3577 u64 target = bctl->sys.target;
3578
3579 num_tolerated_disk_barrier_failures =
3580 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3581 if (num_tolerated_disk_barrier_failures > 0 &&
3582 (target &
3583 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3584 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3585 num_tolerated_disk_barrier_failures = 0;
3586 else if (num_tolerated_disk_barrier_failures > 1 &&
3587 (target &
3588 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3589 num_tolerated_disk_barrier_failures = 1;
3590
3591 fs_info->num_tolerated_disk_barrier_failures =
3592 num_tolerated_disk_barrier_failures;
3593 }
3594
3595 ret = insert_balance_item(fs_info->tree_root, bctl);
3596 if (ret && ret != -EEXIST)
3597 goto out;
3598
3599 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3600 BUG_ON(ret == -EEXIST);
3601 set_balance_control(bctl);
3602 } else {
3603 BUG_ON(ret != -EEXIST);
3604 spin_lock(&fs_info->balance_lock);
3605 update_balance_args(bctl);
3606 spin_unlock(&fs_info->balance_lock);
3607 }
3608
3609 atomic_inc(&fs_info->balance_running);
3610 mutex_unlock(&fs_info->balance_mutex);
3611
3612 ret = __btrfs_balance(fs_info);
3613
3614 mutex_lock(&fs_info->balance_mutex);
3615 atomic_dec(&fs_info->balance_running);
3616
3617 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3618 fs_info->num_tolerated_disk_barrier_failures =
3619 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3620 }
3621
3622 if (bargs) {
3623 memset(bargs, 0, sizeof(*bargs));
3624 update_ioctl_balance_args(fs_info, 0, bargs);
3625 }
3626
3627 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3628 balance_need_close(fs_info)) {
3629 __cancel_balance(fs_info);
3630 }
3631
3632 wake_up(&fs_info->balance_wait_q);
3633
3634 return ret;
3635 out:
3636 if (bctl->flags & BTRFS_BALANCE_RESUME)
3637 __cancel_balance(fs_info);
3638 else {
3639 kfree(bctl);
3640 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3641 }
3642 return ret;
3643 }
3644
3645 static int balance_kthread(void *data)
3646 {
3647 struct btrfs_fs_info *fs_info = data;
3648 int ret = 0;
3649
3650 mutex_lock(&fs_info->volume_mutex);
3651 mutex_lock(&fs_info->balance_mutex);
3652
3653 if (fs_info->balance_ctl) {
3654 btrfs_info(fs_info, "continuing balance");
3655 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3656 }
3657
3658 mutex_unlock(&fs_info->balance_mutex);
3659 mutex_unlock(&fs_info->volume_mutex);
3660
3661 return ret;
3662 }
3663
3664 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3665 {
3666 struct task_struct *tsk;
3667
3668 spin_lock(&fs_info->balance_lock);
3669 if (!fs_info->balance_ctl) {
3670 spin_unlock(&fs_info->balance_lock);
3671 return 0;
3672 }
3673 spin_unlock(&fs_info->balance_lock);
3674
3675 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3676 btrfs_info(fs_info, "force skipping balance");
3677 return 0;
3678 }
3679
3680 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3681 return PTR_ERR_OR_ZERO(tsk);
3682 }
3683
3684 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3685 {
3686 struct btrfs_balance_control *bctl;
3687 struct btrfs_balance_item *item;
3688 struct btrfs_disk_balance_args disk_bargs;
3689 struct btrfs_path *path;
3690 struct extent_buffer *leaf;
3691 struct btrfs_key key;
3692 int ret;
3693
3694 path = btrfs_alloc_path();
3695 if (!path)
3696 return -ENOMEM;
3697
3698 key.objectid = BTRFS_BALANCE_OBJECTID;
3699 key.type = BTRFS_BALANCE_ITEM_KEY;
3700 key.offset = 0;
3701
3702 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3703 if (ret < 0)
3704 goto out;
3705 if (ret > 0) { /* ret = -ENOENT; */
3706 ret = 0;
3707 goto out;
3708 }
3709
3710 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3711 if (!bctl) {
3712 ret = -ENOMEM;
3713 goto out;
3714 }
3715
3716 leaf = path->nodes[0];
3717 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3718
3719 bctl->fs_info = fs_info;
3720 bctl->flags = btrfs_balance_flags(leaf, item);
3721 bctl->flags |= BTRFS_BALANCE_RESUME;
3722
3723 btrfs_balance_data(leaf, item, &disk_bargs);
3724 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3725 btrfs_balance_meta(leaf, item, &disk_bargs);
3726 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3727 btrfs_balance_sys(leaf, item, &disk_bargs);
3728 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3729
3730 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3731
3732 mutex_lock(&fs_info->volume_mutex);
3733 mutex_lock(&fs_info->balance_mutex);
3734
3735 set_balance_control(bctl);
3736
3737 mutex_unlock(&fs_info->balance_mutex);
3738 mutex_unlock(&fs_info->volume_mutex);
3739 out:
3740 btrfs_free_path(path);
3741 return ret;
3742 }
3743
3744 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3745 {
3746 int ret = 0;
3747
3748 mutex_lock(&fs_info->balance_mutex);
3749 if (!fs_info->balance_ctl) {
3750 mutex_unlock(&fs_info->balance_mutex);
3751 return -ENOTCONN;
3752 }
3753
3754 if (atomic_read(&fs_info->balance_running)) {
3755 atomic_inc(&fs_info->balance_pause_req);
3756 mutex_unlock(&fs_info->balance_mutex);
3757
3758 wait_event(fs_info->balance_wait_q,
3759 atomic_read(&fs_info->balance_running) == 0);
3760
3761 mutex_lock(&fs_info->balance_mutex);
3762 /* we are good with balance_ctl ripped off from under us */
3763 BUG_ON(atomic_read(&fs_info->balance_running));
3764 atomic_dec(&fs_info->balance_pause_req);
3765 } else {
3766 ret = -ENOTCONN;
3767 }
3768
3769 mutex_unlock(&fs_info->balance_mutex);
3770 return ret;
3771 }
3772
3773 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3774 {
3775 if (fs_info->sb->s_flags & MS_RDONLY)
3776 return -EROFS;
3777
3778 mutex_lock(&fs_info->balance_mutex);
3779 if (!fs_info->balance_ctl) {
3780 mutex_unlock(&fs_info->balance_mutex);
3781 return -ENOTCONN;
3782 }
3783
3784 atomic_inc(&fs_info->balance_cancel_req);
3785 /*
3786 * if we are running just wait and return, balance item is
3787 * deleted in btrfs_balance in this case
3788 */
3789 if (atomic_read(&fs_info->balance_running)) {
3790 mutex_unlock(&fs_info->balance_mutex);
3791 wait_event(fs_info->balance_wait_q,
3792 atomic_read(&fs_info->balance_running) == 0);
3793 mutex_lock(&fs_info->balance_mutex);
3794 } else {
3795 /* __cancel_balance needs volume_mutex */
3796 mutex_unlock(&fs_info->balance_mutex);
3797 mutex_lock(&fs_info->volume_mutex);
3798 mutex_lock(&fs_info->balance_mutex);
3799
3800 if (fs_info->balance_ctl)
3801 __cancel_balance(fs_info);
3802
3803 mutex_unlock(&fs_info->volume_mutex);
3804 }
3805
3806 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3807 atomic_dec(&fs_info->balance_cancel_req);
3808 mutex_unlock(&fs_info->balance_mutex);
3809 return 0;
3810 }
3811
3812 static int btrfs_uuid_scan_kthread(void *data)
3813 {
3814 struct btrfs_fs_info *fs_info = data;
3815 struct btrfs_root *root = fs_info->tree_root;
3816 struct btrfs_key key;
3817 struct btrfs_key max_key;
3818 struct btrfs_path *path = NULL;
3819 int ret = 0;
3820 struct extent_buffer *eb;
3821 int slot;
3822 struct btrfs_root_item root_item;
3823 u32 item_size;
3824 struct btrfs_trans_handle *trans = NULL;
3825
3826 path = btrfs_alloc_path();
3827 if (!path) {
3828 ret = -ENOMEM;
3829 goto out;
3830 }
3831
3832 key.objectid = 0;
3833 key.type = BTRFS_ROOT_ITEM_KEY;
3834 key.offset = 0;
3835
3836 max_key.objectid = (u64)-1;
3837 max_key.type = BTRFS_ROOT_ITEM_KEY;
3838 max_key.offset = (u64)-1;
3839
3840 while (1) {
3841 ret = btrfs_search_forward(root, &key, path, 0);
3842 if (ret) {
3843 if (ret > 0)
3844 ret = 0;
3845 break;
3846 }
3847
3848 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3849 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3850 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3851 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3852 goto skip;
3853
3854 eb = path->nodes[0];
3855 slot = path->slots[0];
3856 item_size = btrfs_item_size_nr(eb, slot);
3857 if (item_size < sizeof(root_item))
3858 goto skip;
3859
3860 read_extent_buffer(eb, &root_item,
3861 btrfs_item_ptr_offset(eb, slot),
3862 (int)sizeof(root_item));
3863 if (btrfs_root_refs(&root_item) == 0)
3864 goto skip;
3865
3866 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3867 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3868 if (trans)
3869 goto update_tree;
3870
3871 btrfs_release_path(path);
3872 /*
3873 * 1 - subvol uuid item
3874 * 1 - received_subvol uuid item
3875 */
3876 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3877 if (IS_ERR(trans)) {
3878 ret = PTR_ERR(trans);
3879 break;
3880 }
3881 continue;
3882 } else {
3883 goto skip;
3884 }
3885 update_tree:
3886 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3887 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3888 root_item.uuid,
3889 BTRFS_UUID_KEY_SUBVOL,
3890 key.objectid);
3891 if (ret < 0) {
3892 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3893 ret);
3894 break;
3895 }
3896 }
3897
3898 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3899 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3900 root_item.received_uuid,
3901 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3902 key.objectid);
3903 if (ret < 0) {
3904 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3905 ret);
3906 break;
3907 }
3908 }
3909
3910 skip:
3911 if (trans) {
3912 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3913 trans = NULL;
3914 if (ret)
3915 break;
3916 }
3917
3918 btrfs_release_path(path);
3919 if (key.offset < (u64)-1) {
3920 key.offset++;
3921 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3922 key.offset = 0;
3923 key.type = BTRFS_ROOT_ITEM_KEY;
3924 } else if (key.objectid < (u64)-1) {
3925 key.offset = 0;
3926 key.type = BTRFS_ROOT_ITEM_KEY;
3927 key.objectid++;
3928 } else {
3929 break;
3930 }
3931 cond_resched();
3932 }
3933
3934 out:
3935 btrfs_free_path(path);
3936 if (trans && !IS_ERR(trans))
3937 btrfs_end_transaction(trans, fs_info->uuid_root);
3938 if (ret)
3939 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3940 else
3941 fs_info->update_uuid_tree_gen = 1;
3942 up(&fs_info->uuid_tree_rescan_sem);
3943 return 0;
3944 }
3945
3946 /*
3947 * Callback for btrfs_uuid_tree_iterate().
3948 * returns:
3949 * 0 check succeeded, the entry is not outdated.
3950 * < 0 if an error occured.
3951 * > 0 if the check failed, which means the caller shall remove the entry.
3952 */
3953 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3954 u8 *uuid, u8 type, u64 subid)
3955 {
3956 struct btrfs_key key;
3957 int ret = 0;
3958 struct btrfs_root *subvol_root;
3959
3960 if (type != BTRFS_UUID_KEY_SUBVOL &&
3961 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3962 goto out;
3963
3964 key.objectid = subid;
3965 key.type = BTRFS_ROOT_ITEM_KEY;
3966 key.offset = (u64)-1;
3967 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3968 if (IS_ERR(subvol_root)) {
3969 ret = PTR_ERR(subvol_root);
3970 if (ret == -ENOENT)
3971 ret = 1;
3972 goto out;
3973 }
3974
3975 switch (type) {
3976 case BTRFS_UUID_KEY_SUBVOL:
3977 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3978 ret = 1;
3979 break;
3980 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3981 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3982 BTRFS_UUID_SIZE))
3983 ret = 1;
3984 break;
3985 }
3986
3987 out:
3988 return ret;
3989 }
3990
3991 static int btrfs_uuid_rescan_kthread(void *data)
3992 {
3993 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3994 int ret;
3995
3996 /*
3997 * 1st step is to iterate through the existing UUID tree and
3998 * to delete all entries that contain outdated data.
3999 * 2nd step is to add all missing entries to the UUID tree.
4000 */
4001 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4002 if (ret < 0) {
4003 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4004 up(&fs_info->uuid_tree_rescan_sem);
4005 return ret;
4006 }
4007 return btrfs_uuid_scan_kthread(data);
4008 }
4009
4010 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4011 {
4012 struct btrfs_trans_handle *trans;
4013 struct btrfs_root *tree_root = fs_info->tree_root;
4014 struct btrfs_root *uuid_root;
4015 struct task_struct *task;
4016 int ret;
4017
4018 /*
4019 * 1 - root node
4020 * 1 - root item
4021 */
4022 trans = btrfs_start_transaction(tree_root, 2);
4023 if (IS_ERR(trans))
4024 return PTR_ERR(trans);
4025
4026 uuid_root = btrfs_create_tree(trans, fs_info,
4027 BTRFS_UUID_TREE_OBJECTID);
4028 if (IS_ERR(uuid_root)) {
4029 ret = PTR_ERR(uuid_root);
4030 btrfs_abort_transaction(trans, tree_root, ret);
4031 return ret;
4032 }
4033
4034 fs_info->uuid_root = uuid_root;
4035
4036 ret = btrfs_commit_transaction(trans, tree_root);
4037 if (ret)
4038 return ret;
4039
4040 down(&fs_info->uuid_tree_rescan_sem);
4041 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4042 if (IS_ERR(task)) {
4043 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4044 btrfs_warn(fs_info, "failed to start uuid_scan task");
4045 up(&fs_info->uuid_tree_rescan_sem);
4046 return PTR_ERR(task);
4047 }
4048
4049 return 0;
4050 }
4051
4052 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4053 {
4054 struct task_struct *task;
4055
4056 down(&fs_info->uuid_tree_rescan_sem);
4057 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4058 if (IS_ERR(task)) {
4059 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4060 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4061 up(&fs_info->uuid_tree_rescan_sem);
4062 return PTR_ERR(task);
4063 }
4064
4065 return 0;
4066 }
4067
4068 /*
4069 * shrinking a device means finding all of the device extents past
4070 * the new size, and then following the back refs to the chunks.
4071 * The chunk relocation code actually frees the device extent
4072 */
4073 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4074 {
4075 struct btrfs_trans_handle *trans;
4076 struct btrfs_root *root = device->dev_root;
4077 struct btrfs_dev_extent *dev_extent = NULL;
4078 struct btrfs_path *path;
4079 u64 length;
4080 u64 chunk_objectid;
4081 u64 chunk_offset;
4082 int ret;
4083 int slot;
4084 int failed = 0;
4085 bool retried = false;
4086 bool checked_pending_chunks = false;
4087 struct extent_buffer *l;
4088 struct btrfs_key key;
4089 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4090 u64 old_total = btrfs_super_total_bytes(super_copy);
4091 u64 old_size = btrfs_device_get_total_bytes(device);
4092 u64 diff = old_size - new_size;
4093
4094 if (device->is_tgtdev_for_dev_replace)
4095 return -EINVAL;
4096
4097 path = btrfs_alloc_path();
4098 if (!path)
4099 return -ENOMEM;
4100
4101 path->reada = 2;
4102
4103 lock_chunks(root);
4104
4105 btrfs_device_set_total_bytes(device, new_size);
4106 if (device->writeable) {
4107 device->fs_devices->total_rw_bytes -= diff;
4108 spin_lock(&root->fs_info->free_chunk_lock);
4109 root->fs_info->free_chunk_space -= diff;
4110 spin_unlock(&root->fs_info->free_chunk_lock);
4111 }
4112 unlock_chunks(root);
4113
4114 again:
4115 key.objectid = device->devid;
4116 key.offset = (u64)-1;
4117 key.type = BTRFS_DEV_EXTENT_KEY;
4118
4119 do {
4120 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4122 if (ret < 0) {
4123 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4124 goto done;
4125 }
4126
4127 ret = btrfs_previous_item(root, path, 0, key.type);
4128 if (ret)
4129 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4130 if (ret < 0)
4131 goto done;
4132 if (ret) {
4133 ret = 0;
4134 btrfs_release_path(path);
4135 break;
4136 }
4137
4138 l = path->nodes[0];
4139 slot = path->slots[0];
4140 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4141
4142 if (key.objectid != device->devid) {
4143 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4144 btrfs_release_path(path);
4145 break;
4146 }
4147
4148 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4149 length = btrfs_dev_extent_length(l, dev_extent);
4150
4151 if (key.offset + length <= new_size) {
4152 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4153 btrfs_release_path(path);
4154 break;
4155 }
4156
4157 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4158 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4159 btrfs_release_path(path);
4160
4161 ret = btrfs_relocate_chunk(root, chunk_objectid, chunk_offset);
4162 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4163 if (ret && ret != -ENOSPC)
4164 goto done;
4165 if (ret == -ENOSPC)
4166 failed++;
4167 } while (key.offset-- > 0);
4168
4169 if (failed && !retried) {
4170 failed = 0;
4171 retried = true;
4172 goto again;
4173 } else if (failed && retried) {
4174 ret = -ENOSPC;
4175 goto done;
4176 }
4177
4178 /* Shrinking succeeded, else we would be at "done". */
4179 trans = btrfs_start_transaction(root, 0);
4180 if (IS_ERR(trans)) {
4181 ret = PTR_ERR(trans);
4182 goto done;
4183 }
4184
4185 lock_chunks(root);
4186
4187 /*
4188 * We checked in the above loop all device extents that were already in
4189 * the device tree. However before we have updated the device's
4190 * total_bytes to the new size, we might have had chunk allocations that
4191 * have not complete yet (new block groups attached to transaction
4192 * handles), and therefore their device extents were not yet in the
4193 * device tree and we missed them in the loop above. So if we have any
4194 * pending chunk using a device extent that overlaps the device range
4195 * that we can not use anymore, commit the current transaction and
4196 * repeat the search on the device tree - this way we guarantee we will
4197 * not have chunks using device extents that end beyond 'new_size'.
4198 */
4199 if (!checked_pending_chunks) {
4200 u64 start = new_size;
4201 u64 len = old_size - new_size;
4202
4203 if (contains_pending_extent(trans, device, &start, len)) {
4204 unlock_chunks(root);
4205 checked_pending_chunks = true;
4206 failed = 0;
4207 retried = false;
4208 ret = btrfs_commit_transaction(trans, root);
4209 if (ret)
4210 goto done;
4211 goto again;
4212 }
4213 }
4214
4215 btrfs_device_set_disk_total_bytes(device, new_size);
4216 if (list_empty(&device->resized_list))
4217 list_add_tail(&device->resized_list,
4218 &root->fs_info->fs_devices->resized_devices);
4219
4220 WARN_ON(diff > old_total);
4221 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4222 unlock_chunks(root);
4223
4224 /* Now btrfs_update_device() will change the on-disk size. */
4225 ret = btrfs_update_device(trans, device);
4226 btrfs_end_transaction(trans, root);
4227 done:
4228 btrfs_free_path(path);
4229 if (ret) {
4230 lock_chunks(root);
4231 btrfs_device_set_total_bytes(device, old_size);
4232 if (device->writeable)
4233 device->fs_devices->total_rw_bytes += diff;
4234 spin_lock(&root->fs_info->free_chunk_lock);
4235 root->fs_info->free_chunk_space += diff;
4236 spin_unlock(&root->fs_info->free_chunk_lock);
4237 unlock_chunks(root);
4238 }
4239 return ret;
4240 }
4241
4242 static int btrfs_add_system_chunk(struct btrfs_root *root,
4243 struct btrfs_key *key,
4244 struct btrfs_chunk *chunk, int item_size)
4245 {
4246 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4247 struct btrfs_disk_key disk_key;
4248 u32 array_size;
4249 u8 *ptr;
4250
4251 lock_chunks(root);
4252 array_size = btrfs_super_sys_array_size(super_copy);
4253 if (array_size + item_size + sizeof(disk_key)
4254 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4255 unlock_chunks(root);
4256 return -EFBIG;
4257 }
4258
4259 ptr = super_copy->sys_chunk_array + array_size;
4260 btrfs_cpu_key_to_disk(&disk_key, key);
4261 memcpy(ptr, &disk_key, sizeof(disk_key));
4262 ptr += sizeof(disk_key);
4263 memcpy(ptr, chunk, item_size);
4264 item_size += sizeof(disk_key);
4265 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4266 unlock_chunks(root);
4267
4268 return 0;
4269 }
4270
4271 /*
4272 * sort the devices in descending order by max_avail, total_avail
4273 */
4274 static int btrfs_cmp_device_info(const void *a, const void *b)
4275 {
4276 const struct btrfs_device_info *di_a = a;
4277 const struct btrfs_device_info *di_b = b;
4278
4279 if (di_a->max_avail > di_b->max_avail)
4280 return -1;
4281 if (di_a->max_avail < di_b->max_avail)
4282 return 1;
4283 if (di_a->total_avail > di_b->total_avail)
4284 return -1;
4285 if (di_a->total_avail < di_b->total_avail)
4286 return 1;
4287 return 0;
4288 }
4289
4290 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4291 [BTRFS_RAID_RAID10] = {
4292 .sub_stripes = 2,
4293 .dev_stripes = 1,
4294 .devs_max = 0, /* 0 == as many as possible */
4295 .devs_min = 4,
4296 .devs_increment = 2,
4297 .ncopies = 2,
4298 },
4299 [BTRFS_RAID_RAID1] = {
4300 .sub_stripes = 1,
4301 .dev_stripes = 1,
4302 .devs_max = 2,
4303 .devs_min = 2,
4304 .devs_increment = 2,
4305 .ncopies = 2,
4306 },
4307 [BTRFS_RAID_DUP] = {
4308 .sub_stripes = 1,
4309 .dev_stripes = 2,
4310 .devs_max = 1,
4311 .devs_min = 1,
4312 .devs_increment = 1,
4313 .ncopies = 2,
4314 },
4315 [BTRFS_RAID_RAID0] = {
4316 .sub_stripes = 1,
4317 .dev_stripes = 1,
4318 .devs_max = 0,
4319 .devs_min = 2,
4320 .devs_increment = 1,
4321 .ncopies = 1,
4322 },
4323 [BTRFS_RAID_SINGLE] = {
4324 .sub_stripes = 1,
4325 .dev_stripes = 1,
4326 .devs_max = 1,
4327 .devs_min = 1,
4328 .devs_increment = 1,
4329 .ncopies = 1,
4330 },
4331 [BTRFS_RAID_RAID5] = {
4332 .sub_stripes = 1,
4333 .dev_stripes = 1,
4334 .devs_max = 0,
4335 .devs_min = 2,
4336 .devs_increment = 1,
4337 .ncopies = 2,
4338 },
4339 [BTRFS_RAID_RAID6] = {
4340 .sub_stripes = 1,
4341 .dev_stripes = 1,
4342 .devs_max = 0,
4343 .devs_min = 3,
4344 .devs_increment = 1,
4345 .ncopies = 3,
4346 },
4347 };
4348
4349 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4350 {
4351 /* TODO allow them to set a preferred stripe size */
4352 return 64 * 1024;
4353 }
4354
4355 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4356 {
4357 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4358 return;
4359
4360 btrfs_set_fs_incompat(info, RAID56);
4361 }
4362
4363 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4364 - sizeof(struct btrfs_item) \
4365 - sizeof(struct btrfs_chunk)) \
4366 / sizeof(struct btrfs_stripe) + 1)
4367
4368 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4369 - 2 * sizeof(struct btrfs_disk_key) \
4370 - 2 * sizeof(struct btrfs_chunk)) \
4371 / sizeof(struct btrfs_stripe) + 1)
4372
4373 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4374 struct btrfs_root *extent_root, u64 start,
4375 u64 type)
4376 {
4377 struct btrfs_fs_info *info = extent_root->fs_info;
4378 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4379 struct list_head *cur;
4380 struct map_lookup *map = NULL;
4381 struct extent_map_tree *em_tree;
4382 struct extent_map *em;
4383 struct btrfs_device_info *devices_info = NULL;
4384 u64 total_avail;
4385 int num_stripes; /* total number of stripes to allocate */
4386 int data_stripes; /* number of stripes that count for
4387 block group size */
4388 int sub_stripes; /* sub_stripes info for map */
4389 int dev_stripes; /* stripes per dev */
4390 int devs_max; /* max devs to use */
4391 int devs_min; /* min devs needed */
4392 int devs_increment; /* ndevs has to be a multiple of this */
4393 int ncopies; /* how many copies to data has */
4394 int ret;
4395 u64 max_stripe_size;
4396 u64 max_chunk_size;
4397 u64 stripe_size;
4398 u64 num_bytes;
4399 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4400 int ndevs;
4401 int i;
4402 int j;
4403 int index;
4404
4405 BUG_ON(!alloc_profile_is_valid(type, 0));
4406
4407 if (list_empty(&fs_devices->alloc_list))
4408 return -ENOSPC;
4409
4410 index = __get_raid_index(type);
4411
4412 sub_stripes = btrfs_raid_array[index].sub_stripes;
4413 dev_stripes = btrfs_raid_array[index].dev_stripes;
4414 devs_max = btrfs_raid_array[index].devs_max;
4415 devs_min = btrfs_raid_array[index].devs_min;
4416 devs_increment = btrfs_raid_array[index].devs_increment;
4417 ncopies = btrfs_raid_array[index].ncopies;
4418
4419 if (type & BTRFS_BLOCK_GROUP_DATA) {
4420 max_stripe_size = 1024 * 1024 * 1024;
4421 max_chunk_size = 10 * max_stripe_size;
4422 if (!devs_max)
4423 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4424 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4425 /* for larger filesystems, use larger metadata chunks */
4426 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4427 max_stripe_size = 1024 * 1024 * 1024;
4428 else
4429 max_stripe_size = 256 * 1024 * 1024;
4430 max_chunk_size = max_stripe_size;
4431 if (!devs_max)
4432 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4433 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4434 max_stripe_size = 32 * 1024 * 1024;
4435 max_chunk_size = 2 * max_stripe_size;
4436 if (!devs_max)
4437 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4438 } else {
4439 btrfs_err(info, "invalid chunk type 0x%llx requested",
4440 type);
4441 BUG_ON(1);
4442 }
4443
4444 /* we don't want a chunk larger than 10% of writeable space */
4445 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4446 max_chunk_size);
4447
4448 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4449 GFP_NOFS);
4450 if (!devices_info)
4451 return -ENOMEM;
4452
4453 cur = fs_devices->alloc_list.next;
4454
4455 /*
4456 * in the first pass through the devices list, we gather information
4457 * about the available holes on each device.
4458 */
4459 ndevs = 0;
4460 while (cur != &fs_devices->alloc_list) {
4461 struct btrfs_device *device;
4462 u64 max_avail;
4463 u64 dev_offset;
4464
4465 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4466
4467 cur = cur->next;
4468
4469 if (!device->writeable) {
4470 WARN(1, KERN_ERR
4471 "BTRFS: read-only device in alloc_list\n");
4472 continue;
4473 }
4474
4475 if (!device->in_fs_metadata ||
4476 device->is_tgtdev_for_dev_replace)
4477 continue;
4478
4479 if (device->total_bytes > device->bytes_used)
4480 total_avail = device->total_bytes - device->bytes_used;
4481 else
4482 total_avail = 0;
4483
4484 /* If there is no space on this device, skip it. */
4485 if (total_avail == 0)
4486 continue;
4487
4488 ret = find_free_dev_extent(trans, device,
4489 max_stripe_size * dev_stripes,
4490 &dev_offset, &max_avail);
4491 if (ret && ret != -ENOSPC)
4492 goto error;
4493
4494 if (ret == 0)
4495 max_avail = max_stripe_size * dev_stripes;
4496
4497 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4498 continue;
4499
4500 if (ndevs == fs_devices->rw_devices) {
4501 WARN(1, "%s: found more than %llu devices\n",
4502 __func__, fs_devices->rw_devices);
4503 break;
4504 }
4505 devices_info[ndevs].dev_offset = dev_offset;
4506 devices_info[ndevs].max_avail = max_avail;
4507 devices_info[ndevs].total_avail = total_avail;
4508 devices_info[ndevs].dev = device;
4509 ++ndevs;
4510 }
4511
4512 /*
4513 * now sort the devices by hole size / available space
4514 */
4515 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4516 btrfs_cmp_device_info, NULL);
4517
4518 /* round down to number of usable stripes */
4519 ndevs -= ndevs % devs_increment;
4520
4521 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4522 ret = -ENOSPC;
4523 goto error;
4524 }
4525
4526 if (devs_max && ndevs > devs_max)
4527 ndevs = devs_max;
4528 /*
4529 * the primary goal is to maximize the number of stripes, so use as many
4530 * devices as possible, even if the stripes are not maximum sized.
4531 */
4532 stripe_size = devices_info[ndevs-1].max_avail;
4533 num_stripes = ndevs * dev_stripes;
4534
4535 /*
4536 * this will have to be fixed for RAID1 and RAID10 over
4537 * more drives
4538 */
4539 data_stripes = num_stripes / ncopies;
4540
4541 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4542 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4543 btrfs_super_stripesize(info->super_copy));
4544 data_stripes = num_stripes - 1;
4545 }
4546 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4547 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4548 btrfs_super_stripesize(info->super_copy));
4549 data_stripes = num_stripes - 2;
4550 }
4551
4552 /*
4553 * Use the number of data stripes to figure out how big this chunk
4554 * is really going to be in terms of logical address space,
4555 * and compare that answer with the max chunk size
4556 */
4557 if (stripe_size * data_stripes > max_chunk_size) {
4558 u64 mask = (1ULL << 24) - 1;
4559
4560 stripe_size = div_u64(max_chunk_size, data_stripes);
4561
4562 /* bump the answer up to a 16MB boundary */
4563 stripe_size = (stripe_size + mask) & ~mask;
4564
4565 /* but don't go higher than the limits we found
4566 * while searching for free extents
4567 */
4568 if (stripe_size > devices_info[ndevs-1].max_avail)
4569 stripe_size = devices_info[ndevs-1].max_avail;
4570 }
4571
4572 stripe_size = div_u64(stripe_size, dev_stripes);
4573
4574 /* align to BTRFS_STRIPE_LEN */
4575 stripe_size = div_u64(stripe_size, raid_stripe_len);
4576 stripe_size *= raid_stripe_len;
4577
4578 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4579 if (!map) {
4580 ret = -ENOMEM;
4581 goto error;
4582 }
4583 map->num_stripes = num_stripes;
4584
4585 for (i = 0; i < ndevs; ++i) {
4586 for (j = 0; j < dev_stripes; ++j) {
4587 int s = i * dev_stripes + j;
4588 map->stripes[s].dev = devices_info[i].dev;
4589 map->stripes[s].physical = devices_info[i].dev_offset +
4590 j * stripe_size;
4591 }
4592 }
4593 map->sector_size = extent_root->sectorsize;
4594 map->stripe_len = raid_stripe_len;
4595 map->io_align = raid_stripe_len;
4596 map->io_width = raid_stripe_len;
4597 map->type = type;
4598 map->sub_stripes = sub_stripes;
4599
4600 num_bytes = stripe_size * data_stripes;
4601
4602 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4603
4604 em = alloc_extent_map();
4605 if (!em) {
4606 kfree(map);
4607 ret = -ENOMEM;
4608 goto error;
4609 }
4610 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4611 em->bdev = (struct block_device *)map;
4612 em->start = start;
4613 em->len = num_bytes;
4614 em->block_start = 0;
4615 em->block_len = em->len;
4616 em->orig_block_len = stripe_size;
4617
4618 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4619 write_lock(&em_tree->lock);
4620 ret = add_extent_mapping(em_tree, em, 0);
4621 if (!ret) {
4622 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4623 atomic_inc(&em->refs);
4624 }
4625 write_unlock(&em_tree->lock);
4626 if (ret) {
4627 free_extent_map(em);
4628 goto error;
4629 }
4630
4631 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4632 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4633 start, num_bytes);
4634 if (ret)
4635 goto error_del_extent;
4636
4637 for (i = 0; i < map->num_stripes; i++) {
4638 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4639 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4640 }
4641
4642 spin_lock(&extent_root->fs_info->free_chunk_lock);
4643 extent_root->fs_info->free_chunk_space -= (stripe_size *
4644 map->num_stripes);
4645 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4646
4647 free_extent_map(em);
4648 check_raid56_incompat_flag(extent_root->fs_info, type);
4649
4650 kfree(devices_info);
4651 return 0;
4652
4653 error_del_extent:
4654 write_lock(&em_tree->lock);
4655 remove_extent_mapping(em_tree, em);
4656 write_unlock(&em_tree->lock);
4657
4658 /* One for our allocation */
4659 free_extent_map(em);
4660 /* One for the tree reference */
4661 free_extent_map(em);
4662 /* One for the pending_chunks list reference */
4663 free_extent_map(em);
4664 error:
4665 kfree(devices_info);
4666 return ret;
4667 }
4668
4669 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4670 struct btrfs_root *extent_root,
4671 u64 chunk_offset, u64 chunk_size)
4672 {
4673 struct btrfs_key key;
4674 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4675 struct btrfs_device *device;
4676 struct btrfs_chunk *chunk;
4677 struct btrfs_stripe *stripe;
4678 struct extent_map_tree *em_tree;
4679 struct extent_map *em;
4680 struct map_lookup *map;
4681 size_t item_size;
4682 u64 dev_offset;
4683 u64 stripe_size;
4684 int i = 0;
4685 int ret;
4686
4687 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4688 read_lock(&em_tree->lock);
4689 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4690 read_unlock(&em_tree->lock);
4691
4692 if (!em) {
4693 btrfs_crit(extent_root->fs_info, "unable to find logical "
4694 "%Lu len %Lu", chunk_offset, chunk_size);
4695 return -EINVAL;
4696 }
4697
4698 if (em->start != chunk_offset || em->len != chunk_size) {
4699 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4700 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4701 chunk_size, em->start, em->len);
4702 free_extent_map(em);
4703 return -EINVAL;
4704 }
4705
4706 map = (struct map_lookup *)em->bdev;
4707 item_size = btrfs_chunk_item_size(map->num_stripes);
4708 stripe_size = em->orig_block_len;
4709
4710 chunk = kzalloc(item_size, GFP_NOFS);
4711 if (!chunk) {
4712 ret = -ENOMEM;
4713 goto out;
4714 }
4715
4716 for (i = 0; i < map->num_stripes; i++) {
4717 device = map->stripes[i].dev;
4718 dev_offset = map->stripes[i].physical;
4719
4720 ret = btrfs_update_device(trans, device);
4721 if (ret)
4722 goto out;
4723 ret = btrfs_alloc_dev_extent(trans, device,
4724 chunk_root->root_key.objectid,
4725 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4726 chunk_offset, dev_offset,
4727 stripe_size);
4728 if (ret)
4729 goto out;
4730 }
4731
4732 stripe = &chunk->stripe;
4733 for (i = 0; i < map->num_stripes; i++) {
4734 device = map->stripes[i].dev;
4735 dev_offset = map->stripes[i].physical;
4736
4737 btrfs_set_stack_stripe_devid(stripe, device->devid);
4738 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4739 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4740 stripe++;
4741 }
4742
4743 btrfs_set_stack_chunk_length(chunk, chunk_size);
4744 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4745 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4746 btrfs_set_stack_chunk_type(chunk, map->type);
4747 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4748 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4749 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4750 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4751 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4752
4753 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4754 key.type = BTRFS_CHUNK_ITEM_KEY;
4755 key.offset = chunk_offset;
4756
4757 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4758 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4759 /*
4760 * TODO: Cleanup of inserted chunk root in case of
4761 * failure.
4762 */
4763 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4764 item_size);
4765 }
4766
4767 out:
4768 kfree(chunk);
4769 free_extent_map(em);
4770 return ret;
4771 }
4772
4773 /*
4774 * Chunk allocation falls into two parts. The first part does works
4775 * that make the new allocated chunk useable, but not do any operation
4776 * that modifies the chunk tree. The second part does the works that
4777 * require modifying the chunk tree. This division is important for the
4778 * bootstrap process of adding storage to a seed btrfs.
4779 */
4780 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4781 struct btrfs_root *extent_root, u64 type)
4782 {
4783 u64 chunk_offset;
4784
4785 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4786 chunk_offset = find_next_chunk(extent_root->fs_info);
4787 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4788 }
4789
4790 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4791 struct btrfs_root *root,
4792 struct btrfs_device *device)
4793 {
4794 u64 chunk_offset;
4795 u64 sys_chunk_offset;
4796 u64 alloc_profile;
4797 struct btrfs_fs_info *fs_info = root->fs_info;
4798 struct btrfs_root *extent_root = fs_info->extent_root;
4799 int ret;
4800
4801 chunk_offset = find_next_chunk(fs_info);
4802 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4803 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4804 alloc_profile);
4805 if (ret)
4806 return ret;
4807
4808 sys_chunk_offset = find_next_chunk(root->fs_info);
4809 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4810 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4811 alloc_profile);
4812 return ret;
4813 }
4814
4815 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4816 {
4817 int max_errors;
4818
4819 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4820 BTRFS_BLOCK_GROUP_RAID10 |
4821 BTRFS_BLOCK_GROUP_RAID5 |
4822 BTRFS_BLOCK_GROUP_DUP)) {
4823 max_errors = 1;
4824 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4825 max_errors = 2;
4826 } else {
4827 max_errors = 0;
4828 }
4829
4830 return max_errors;
4831 }
4832
4833 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4834 {
4835 struct extent_map *em;
4836 struct map_lookup *map;
4837 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4838 int readonly = 0;
4839 int miss_ndevs = 0;
4840 int i;
4841
4842 read_lock(&map_tree->map_tree.lock);
4843 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4844 read_unlock(&map_tree->map_tree.lock);
4845 if (!em)
4846 return 1;
4847
4848 map = (struct map_lookup *)em->bdev;
4849 for (i = 0; i < map->num_stripes; i++) {
4850 if (map->stripes[i].dev->missing) {
4851 miss_ndevs++;
4852 continue;
4853 }
4854
4855 if (!map->stripes[i].dev->writeable) {
4856 readonly = 1;
4857 goto end;
4858 }
4859 }
4860
4861 /*
4862 * If the number of missing devices is larger than max errors,
4863 * we can not write the data into that chunk successfully, so
4864 * set it readonly.
4865 */
4866 if (miss_ndevs > btrfs_chunk_max_errors(map))
4867 readonly = 1;
4868 end:
4869 free_extent_map(em);
4870 return readonly;
4871 }
4872
4873 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4874 {
4875 extent_map_tree_init(&tree->map_tree);
4876 }
4877
4878 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4879 {
4880 struct extent_map *em;
4881
4882 while (1) {
4883 write_lock(&tree->map_tree.lock);
4884 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4885 if (em)
4886 remove_extent_mapping(&tree->map_tree, em);
4887 write_unlock(&tree->map_tree.lock);
4888 if (!em)
4889 break;
4890 /* once for us */
4891 free_extent_map(em);
4892 /* once for the tree */
4893 free_extent_map(em);
4894 }
4895 }
4896
4897 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4898 {
4899 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4900 struct extent_map *em;
4901 struct map_lookup *map;
4902 struct extent_map_tree *em_tree = &map_tree->map_tree;
4903 int ret;
4904
4905 read_lock(&em_tree->lock);
4906 em = lookup_extent_mapping(em_tree, logical, len);
4907 read_unlock(&em_tree->lock);
4908
4909 /*
4910 * We could return errors for these cases, but that could get ugly and
4911 * we'd probably do the same thing which is just not do anything else
4912 * and exit, so return 1 so the callers don't try to use other copies.
4913 */
4914 if (!em) {
4915 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4916 logical+len);
4917 return 1;
4918 }
4919
4920 if (em->start > logical || em->start + em->len < logical) {
4921 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4922 "%Lu-%Lu", logical, logical+len, em->start,
4923 em->start + em->len);
4924 free_extent_map(em);
4925 return 1;
4926 }
4927
4928 map = (struct map_lookup *)em->bdev;
4929 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4930 ret = map->num_stripes;
4931 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4932 ret = map->sub_stripes;
4933 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4934 ret = 2;
4935 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4936 ret = 3;
4937 else
4938 ret = 1;
4939 free_extent_map(em);
4940
4941 btrfs_dev_replace_lock(&fs_info->dev_replace);
4942 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4943 ret++;
4944 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4945
4946 return ret;
4947 }
4948
4949 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4950 struct btrfs_mapping_tree *map_tree,
4951 u64 logical)
4952 {
4953 struct extent_map *em;
4954 struct map_lookup *map;
4955 struct extent_map_tree *em_tree = &map_tree->map_tree;
4956 unsigned long len = root->sectorsize;
4957
4958 read_lock(&em_tree->lock);
4959 em = lookup_extent_mapping(em_tree, logical, len);
4960 read_unlock(&em_tree->lock);
4961 BUG_ON(!em);
4962
4963 BUG_ON(em->start > logical || em->start + em->len < logical);
4964 map = (struct map_lookup *)em->bdev;
4965 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4966 len = map->stripe_len * nr_data_stripes(map);
4967 free_extent_map(em);
4968 return len;
4969 }
4970
4971 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4972 u64 logical, u64 len, int mirror_num)
4973 {
4974 struct extent_map *em;
4975 struct map_lookup *map;
4976 struct extent_map_tree *em_tree = &map_tree->map_tree;
4977 int ret = 0;
4978
4979 read_lock(&em_tree->lock);
4980 em = lookup_extent_mapping(em_tree, logical, len);
4981 read_unlock(&em_tree->lock);
4982 BUG_ON(!em);
4983
4984 BUG_ON(em->start > logical || em->start + em->len < logical);
4985 map = (struct map_lookup *)em->bdev;
4986 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4987 ret = 1;
4988 free_extent_map(em);
4989 return ret;
4990 }
4991
4992 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4993 struct map_lookup *map, int first, int num,
4994 int optimal, int dev_replace_is_ongoing)
4995 {
4996 int i;
4997 int tolerance;
4998 struct btrfs_device *srcdev;
4999
5000 if (dev_replace_is_ongoing &&
5001 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5002 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5003 srcdev = fs_info->dev_replace.srcdev;
5004 else
5005 srcdev = NULL;
5006
5007 /*
5008 * try to avoid the drive that is the source drive for a
5009 * dev-replace procedure, only choose it if no other non-missing
5010 * mirror is available
5011 */
5012 for (tolerance = 0; tolerance < 2; tolerance++) {
5013 if (map->stripes[optimal].dev->bdev &&
5014 (tolerance || map->stripes[optimal].dev != srcdev))
5015 return optimal;
5016 for (i = first; i < first + num; i++) {
5017 if (map->stripes[i].dev->bdev &&
5018 (tolerance || map->stripes[i].dev != srcdev))
5019 return i;
5020 }
5021 }
5022
5023 /* we couldn't find one that doesn't fail. Just return something
5024 * and the io error handling code will clean up eventually
5025 */
5026 return optimal;
5027 }
5028
5029 static inline int parity_smaller(u64 a, u64 b)
5030 {
5031 return a > b;
5032 }
5033
5034 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5035 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5036 {
5037 struct btrfs_bio_stripe s;
5038 int i;
5039 u64 l;
5040 int again = 1;
5041
5042 while (again) {
5043 again = 0;
5044 for (i = 0; i < num_stripes - 1; i++) {
5045 if (parity_smaller(bbio->raid_map[i],
5046 bbio->raid_map[i+1])) {
5047 s = bbio->stripes[i];
5048 l = bbio->raid_map[i];
5049 bbio->stripes[i] = bbio->stripes[i+1];
5050 bbio->raid_map[i] = bbio->raid_map[i+1];
5051 bbio->stripes[i+1] = s;
5052 bbio->raid_map[i+1] = l;
5053
5054 again = 1;
5055 }
5056 }
5057 }
5058 }
5059
5060 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5061 {
5062 struct btrfs_bio *bbio = kzalloc(
5063 /* the size of the btrfs_bio */
5064 sizeof(struct btrfs_bio) +
5065 /* plus the variable array for the stripes */
5066 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5067 /* plus the variable array for the tgt dev */
5068 sizeof(int) * (real_stripes) +
5069 /*
5070 * plus the raid_map, which includes both the tgt dev
5071 * and the stripes
5072 */
5073 sizeof(u64) * (total_stripes),
5074 GFP_NOFS);
5075 if (!bbio)
5076 return NULL;
5077
5078 atomic_set(&bbio->error, 0);
5079 atomic_set(&bbio->refs, 1);
5080
5081 return bbio;
5082 }
5083
5084 void btrfs_get_bbio(struct btrfs_bio *bbio)
5085 {
5086 WARN_ON(!atomic_read(&bbio->refs));
5087 atomic_inc(&bbio->refs);
5088 }
5089
5090 void btrfs_put_bbio(struct btrfs_bio *bbio)
5091 {
5092 if (!bbio)
5093 return;
5094 if (atomic_dec_and_test(&bbio->refs))
5095 kfree(bbio);
5096 }
5097
5098 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5099 u64 logical, u64 *length,
5100 struct btrfs_bio **bbio_ret,
5101 int mirror_num, int need_raid_map)
5102 {
5103 struct extent_map *em;
5104 struct map_lookup *map;
5105 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5106 struct extent_map_tree *em_tree = &map_tree->map_tree;
5107 u64 offset;
5108 u64 stripe_offset;
5109 u64 stripe_end_offset;
5110 u64 stripe_nr;
5111 u64 stripe_nr_orig;
5112 u64 stripe_nr_end;
5113 u64 stripe_len;
5114 u32 stripe_index;
5115 int i;
5116 int ret = 0;
5117 int num_stripes;
5118 int max_errors = 0;
5119 int tgtdev_indexes = 0;
5120 struct btrfs_bio *bbio = NULL;
5121 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5122 int dev_replace_is_ongoing = 0;
5123 int num_alloc_stripes;
5124 int patch_the_first_stripe_for_dev_replace = 0;
5125 u64 physical_to_patch_in_first_stripe = 0;
5126 u64 raid56_full_stripe_start = (u64)-1;
5127
5128 read_lock(&em_tree->lock);
5129 em = lookup_extent_mapping(em_tree, logical, *length);
5130 read_unlock(&em_tree->lock);
5131
5132 if (!em) {
5133 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5134 logical, *length);
5135 return -EINVAL;
5136 }
5137
5138 if (em->start > logical || em->start + em->len < logical) {
5139 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5140 "found %Lu-%Lu", logical, em->start,
5141 em->start + em->len);
5142 free_extent_map(em);
5143 return -EINVAL;
5144 }
5145
5146 map = (struct map_lookup *)em->bdev;
5147 offset = logical - em->start;
5148
5149 stripe_len = map->stripe_len;
5150 stripe_nr = offset;
5151 /*
5152 * stripe_nr counts the total number of stripes we have to stride
5153 * to get to this block
5154 */
5155 stripe_nr = div64_u64(stripe_nr, stripe_len);
5156
5157 stripe_offset = stripe_nr * stripe_len;
5158 BUG_ON(offset < stripe_offset);
5159
5160 /* stripe_offset is the offset of this block in its stripe*/
5161 stripe_offset = offset - stripe_offset;
5162
5163 /* if we're here for raid56, we need to know the stripe aligned start */
5164 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5165 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5166 raid56_full_stripe_start = offset;
5167
5168 /* allow a write of a full stripe, but make sure we don't
5169 * allow straddling of stripes
5170 */
5171 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5172 full_stripe_len);
5173 raid56_full_stripe_start *= full_stripe_len;
5174 }
5175
5176 if (rw & REQ_DISCARD) {
5177 /* we don't discard raid56 yet */
5178 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5179 ret = -EOPNOTSUPP;
5180 goto out;
5181 }
5182 *length = min_t(u64, em->len - offset, *length);
5183 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5184 u64 max_len;
5185 /* For writes to RAID[56], allow a full stripeset across all disks.
5186 For other RAID types and for RAID[56] reads, just allow a single
5187 stripe (on a single disk). */
5188 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5189 (rw & REQ_WRITE)) {
5190 max_len = stripe_len * nr_data_stripes(map) -
5191 (offset - raid56_full_stripe_start);
5192 } else {
5193 /* we limit the length of each bio to what fits in a stripe */
5194 max_len = stripe_len - stripe_offset;
5195 }
5196 *length = min_t(u64, em->len - offset, max_len);
5197 } else {
5198 *length = em->len - offset;
5199 }
5200
5201 /* This is for when we're called from btrfs_merge_bio_hook() and all
5202 it cares about is the length */
5203 if (!bbio_ret)
5204 goto out;
5205
5206 btrfs_dev_replace_lock(dev_replace);
5207 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5208 if (!dev_replace_is_ongoing)
5209 btrfs_dev_replace_unlock(dev_replace);
5210
5211 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5212 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5213 dev_replace->tgtdev != NULL) {
5214 /*
5215 * in dev-replace case, for repair case (that's the only
5216 * case where the mirror is selected explicitly when
5217 * calling btrfs_map_block), blocks left of the left cursor
5218 * can also be read from the target drive.
5219 * For REQ_GET_READ_MIRRORS, the target drive is added as
5220 * the last one to the array of stripes. For READ, it also
5221 * needs to be supported using the same mirror number.
5222 * If the requested block is not left of the left cursor,
5223 * EIO is returned. This can happen because btrfs_num_copies()
5224 * returns one more in the dev-replace case.
5225 */
5226 u64 tmp_length = *length;
5227 struct btrfs_bio *tmp_bbio = NULL;
5228 int tmp_num_stripes;
5229 u64 srcdev_devid = dev_replace->srcdev->devid;
5230 int index_srcdev = 0;
5231 int found = 0;
5232 u64 physical_of_found = 0;
5233
5234 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5235 logical, &tmp_length, &tmp_bbio, 0, 0);
5236 if (ret) {
5237 WARN_ON(tmp_bbio != NULL);
5238 goto out;
5239 }
5240
5241 tmp_num_stripes = tmp_bbio->num_stripes;
5242 if (mirror_num > tmp_num_stripes) {
5243 /*
5244 * REQ_GET_READ_MIRRORS does not contain this
5245 * mirror, that means that the requested area
5246 * is not left of the left cursor
5247 */
5248 ret = -EIO;
5249 btrfs_put_bbio(tmp_bbio);
5250 goto out;
5251 }
5252
5253 /*
5254 * process the rest of the function using the mirror_num
5255 * of the source drive. Therefore look it up first.
5256 * At the end, patch the device pointer to the one of the
5257 * target drive.
5258 */
5259 for (i = 0; i < tmp_num_stripes; i++) {
5260 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5261 /*
5262 * In case of DUP, in order to keep it
5263 * simple, only add the mirror with the
5264 * lowest physical address
5265 */
5266 if (found &&
5267 physical_of_found <=
5268 tmp_bbio->stripes[i].physical)
5269 continue;
5270 index_srcdev = i;
5271 found = 1;
5272 physical_of_found =
5273 tmp_bbio->stripes[i].physical;
5274 }
5275 }
5276
5277 if (found) {
5278 mirror_num = index_srcdev + 1;
5279 patch_the_first_stripe_for_dev_replace = 1;
5280 physical_to_patch_in_first_stripe = physical_of_found;
5281 } else {
5282 WARN_ON(1);
5283 ret = -EIO;
5284 btrfs_put_bbio(tmp_bbio);
5285 goto out;
5286 }
5287
5288 btrfs_put_bbio(tmp_bbio);
5289 } else if (mirror_num > map->num_stripes) {
5290 mirror_num = 0;
5291 }
5292
5293 num_stripes = 1;
5294 stripe_index = 0;
5295 stripe_nr_orig = stripe_nr;
5296 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5297 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5298 stripe_end_offset = stripe_nr_end * map->stripe_len -
5299 (offset + *length);
5300
5301 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5302 if (rw & REQ_DISCARD)
5303 num_stripes = min_t(u64, map->num_stripes,
5304 stripe_nr_end - stripe_nr_orig);
5305 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5306 &stripe_index);
5307 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5308 mirror_num = 1;
5309 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5310 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5311 num_stripes = map->num_stripes;
5312 else if (mirror_num)
5313 stripe_index = mirror_num - 1;
5314 else {
5315 stripe_index = find_live_mirror(fs_info, map, 0,
5316 map->num_stripes,
5317 current->pid % map->num_stripes,
5318 dev_replace_is_ongoing);
5319 mirror_num = stripe_index + 1;
5320 }
5321
5322 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5323 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5324 num_stripes = map->num_stripes;
5325 } else if (mirror_num) {
5326 stripe_index = mirror_num - 1;
5327 } else {
5328 mirror_num = 1;
5329 }
5330
5331 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5332 u32 factor = map->num_stripes / map->sub_stripes;
5333
5334 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5335 stripe_index *= map->sub_stripes;
5336
5337 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5338 num_stripes = map->sub_stripes;
5339 else if (rw & REQ_DISCARD)
5340 num_stripes = min_t(u64, map->sub_stripes *
5341 (stripe_nr_end - stripe_nr_orig),
5342 map->num_stripes);
5343 else if (mirror_num)
5344 stripe_index += mirror_num - 1;
5345 else {
5346 int old_stripe_index = stripe_index;
5347 stripe_index = find_live_mirror(fs_info, map,
5348 stripe_index,
5349 map->sub_stripes, stripe_index +
5350 current->pid % map->sub_stripes,
5351 dev_replace_is_ongoing);
5352 mirror_num = stripe_index - old_stripe_index + 1;
5353 }
5354
5355 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5356 if (need_raid_map &&
5357 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5358 mirror_num > 1)) {
5359 /* push stripe_nr back to the start of the full stripe */
5360 stripe_nr = div_u64(raid56_full_stripe_start,
5361 stripe_len * nr_data_stripes(map));
5362
5363 /* RAID[56] write or recovery. Return all stripes */
5364 num_stripes = map->num_stripes;
5365 max_errors = nr_parity_stripes(map);
5366
5367 *length = map->stripe_len;
5368 stripe_index = 0;
5369 stripe_offset = 0;
5370 } else {
5371 /*
5372 * Mirror #0 or #1 means the original data block.
5373 * Mirror #2 is RAID5 parity block.
5374 * Mirror #3 is RAID6 Q block.
5375 */
5376 stripe_nr = div_u64_rem(stripe_nr,
5377 nr_data_stripes(map), &stripe_index);
5378 if (mirror_num > 1)
5379 stripe_index = nr_data_stripes(map) +
5380 mirror_num - 2;
5381
5382 /* We distribute the parity blocks across stripes */
5383 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5384 &stripe_index);
5385 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5386 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5387 mirror_num = 1;
5388 }
5389 } else {
5390 /*
5391 * after this, stripe_nr is the number of stripes on this
5392 * device we have to walk to find the data, and stripe_index is
5393 * the number of our device in the stripe array
5394 */
5395 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5396 &stripe_index);
5397 mirror_num = stripe_index + 1;
5398 }
5399 BUG_ON(stripe_index >= map->num_stripes);
5400
5401 num_alloc_stripes = num_stripes;
5402 if (dev_replace_is_ongoing) {
5403 if (rw & (REQ_WRITE | REQ_DISCARD))
5404 num_alloc_stripes <<= 1;
5405 if (rw & REQ_GET_READ_MIRRORS)
5406 num_alloc_stripes++;
5407 tgtdev_indexes = num_stripes;
5408 }
5409
5410 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5411 if (!bbio) {
5412 ret = -ENOMEM;
5413 goto out;
5414 }
5415 if (dev_replace_is_ongoing)
5416 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5417
5418 /* build raid_map */
5419 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5420 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5421 mirror_num > 1)) {
5422 u64 tmp;
5423 unsigned rot;
5424
5425 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5426 sizeof(struct btrfs_bio_stripe) *
5427 num_alloc_stripes +
5428 sizeof(int) * tgtdev_indexes);
5429
5430 /* Work out the disk rotation on this stripe-set */
5431 div_u64_rem(stripe_nr, num_stripes, &rot);
5432
5433 /* Fill in the logical address of each stripe */
5434 tmp = stripe_nr * nr_data_stripes(map);
5435 for (i = 0; i < nr_data_stripes(map); i++)
5436 bbio->raid_map[(i+rot) % num_stripes] =
5437 em->start + (tmp + i) * map->stripe_len;
5438
5439 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5440 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5441 bbio->raid_map[(i+rot+1) % num_stripes] =
5442 RAID6_Q_STRIPE;
5443 }
5444
5445 if (rw & REQ_DISCARD) {
5446 u32 factor = 0;
5447 u32 sub_stripes = 0;
5448 u64 stripes_per_dev = 0;
5449 u32 remaining_stripes = 0;
5450 u32 last_stripe = 0;
5451
5452 if (map->type &
5453 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5454 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5455 sub_stripes = 1;
5456 else
5457 sub_stripes = map->sub_stripes;
5458
5459 factor = map->num_stripes / sub_stripes;
5460 stripes_per_dev = div_u64_rem(stripe_nr_end -
5461 stripe_nr_orig,
5462 factor,
5463 &remaining_stripes);
5464 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5465 last_stripe *= sub_stripes;
5466 }
5467
5468 for (i = 0; i < num_stripes; i++) {
5469 bbio->stripes[i].physical =
5470 map->stripes[stripe_index].physical +
5471 stripe_offset + stripe_nr * map->stripe_len;
5472 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5473
5474 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5475 BTRFS_BLOCK_GROUP_RAID10)) {
5476 bbio->stripes[i].length = stripes_per_dev *
5477 map->stripe_len;
5478
5479 if (i / sub_stripes < remaining_stripes)
5480 bbio->stripes[i].length +=
5481 map->stripe_len;
5482
5483 /*
5484 * Special for the first stripe and
5485 * the last stripe:
5486 *
5487 * |-------|...|-------|
5488 * |----------|
5489 * off end_off
5490 */
5491 if (i < sub_stripes)
5492 bbio->stripes[i].length -=
5493 stripe_offset;
5494
5495 if (stripe_index >= last_stripe &&
5496 stripe_index <= (last_stripe +
5497 sub_stripes - 1))
5498 bbio->stripes[i].length -=
5499 stripe_end_offset;
5500
5501 if (i == sub_stripes - 1)
5502 stripe_offset = 0;
5503 } else
5504 bbio->stripes[i].length = *length;
5505
5506 stripe_index++;
5507 if (stripe_index == map->num_stripes) {
5508 /* This could only happen for RAID0/10 */
5509 stripe_index = 0;
5510 stripe_nr++;
5511 }
5512 }
5513 } else {
5514 for (i = 0; i < num_stripes; i++) {
5515 bbio->stripes[i].physical =
5516 map->stripes[stripe_index].physical +
5517 stripe_offset +
5518 stripe_nr * map->stripe_len;
5519 bbio->stripes[i].dev =
5520 map->stripes[stripe_index].dev;
5521 stripe_index++;
5522 }
5523 }
5524
5525 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5526 max_errors = btrfs_chunk_max_errors(map);
5527
5528 if (bbio->raid_map)
5529 sort_parity_stripes(bbio, num_stripes);
5530
5531 tgtdev_indexes = 0;
5532 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5533 dev_replace->tgtdev != NULL) {
5534 int index_where_to_add;
5535 u64 srcdev_devid = dev_replace->srcdev->devid;
5536
5537 /*
5538 * duplicate the write operations while the dev replace
5539 * procedure is running. Since the copying of the old disk
5540 * to the new disk takes place at run time while the
5541 * filesystem is mounted writable, the regular write
5542 * operations to the old disk have to be duplicated to go
5543 * to the new disk as well.
5544 * Note that device->missing is handled by the caller, and
5545 * that the write to the old disk is already set up in the
5546 * stripes array.
5547 */
5548 index_where_to_add = num_stripes;
5549 for (i = 0; i < num_stripes; i++) {
5550 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5551 /* write to new disk, too */
5552 struct btrfs_bio_stripe *new =
5553 bbio->stripes + index_where_to_add;
5554 struct btrfs_bio_stripe *old =
5555 bbio->stripes + i;
5556
5557 new->physical = old->physical;
5558 new->length = old->length;
5559 new->dev = dev_replace->tgtdev;
5560 bbio->tgtdev_map[i] = index_where_to_add;
5561 index_where_to_add++;
5562 max_errors++;
5563 tgtdev_indexes++;
5564 }
5565 }
5566 num_stripes = index_where_to_add;
5567 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5568 dev_replace->tgtdev != NULL) {
5569 u64 srcdev_devid = dev_replace->srcdev->devid;
5570 int index_srcdev = 0;
5571 int found = 0;
5572 u64 physical_of_found = 0;
5573
5574 /*
5575 * During the dev-replace procedure, the target drive can
5576 * also be used to read data in case it is needed to repair
5577 * a corrupt block elsewhere. This is possible if the
5578 * requested area is left of the left cursor. In this area,
5579 * the target drive is a full copy of the source drive.
5580 */
5581 for (i = 0; i < num_stripes; i++) {
5582 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5583 /*
5584 * In case of DUP, in order to keep it
5585 * simple, only add the mirror with the
5586 * lowest physical address
5587 */
5588 if (found &&
5589 physical_of_found <=
5590 bbio->stripes[i].physical)
5591 continue;
5592 index_srcdev = i;
5593 found = 1;
5594 physical_of_found = bbio->stripes[i].physical;
5595 }
5596 }
5597 if (found) {
5598 if (physical_of_found + map->stripe_len <=
5599 dev_replace->cursor_left) {
5600 struct btrfs_bio_stripe *tgtdev_stripe =
5601 bbio->stripes + num_stripes;
5602
5603 tgtdev_stripe->physical = physical_of_found;
5604 tgtdev_stripe->length =
5605 bbio->stripes[index_srcdev].length;
5606 tgtdev_stripe->dev = dev_replace->tgtdev;
5607 bbio->tgtdev_map[index_srcdev] = num_stripes;
5608
5609 tgtdev_indexes++;
5610 num_stripes++;
5611 }
5612 }
5613 }
5614
5615 *bbio_ret = bbio;
5616 bbio->map_type = map->type;
5617 bbio->num_stripes = num_stripes;
5618 bbio->max_errors = max_errors;
5619 bbio->mirror_num = mirror_num;
5620 bbio->num_tgtdevs = tgtdev_indexes;
5621
5622 /*
5623 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5624 * mirror_num == num_stripes + 1 && dev_replace target drive is
5625 * available as a mirror
5626 */
5627 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5628 WARN_ON(num_stripes > 1);
5629 bbio->stripes[0].dev = dev_replace->tgtdev;
5630 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5631 bbio->mirror_num = map->num_stripes + 1;
5632 }
5633 out:
5634 if (dev_replace_is_ongoing)
5635 btrfs_dev_replace_unlock(dev_replace);
5636 free_extent_map(em);
5637 return ret;
5638 }
5639
5640 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5641 u64 logical, u64 *length,
5642 struct btrfs_bio **bbio_ret, int mirror_num)
5643 {
5644 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5645 mirror_num, 0);
5646 }
5647
5648 /* For Scrub/replace */
5649 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5650 u64 logical, u64 *length,
5651 struct btrfs_bio **bbio_ret, int mirror_num,
5652 int need_raid_map)
5653 {
5654 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5655 mirror_num, need_raid_map);
5656 }
5657
5658 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5659 u64 chunk_start, u64 physical, u64 devid,
5660 u64 **logical, int *naddrs, int *stripe_len)
5661 {
5662 struct extent_map_tree *em_tree = &map_tree->map_tree;
5663 struct extent_map *em;
5664 struct map_lookup *map;
5665 u64 *buf;
5666 u64 bytenr;
5667 u64 length;
5668 u64 stripe_nr;
5669 u64 rmap_len;
5670 int i, j, nr = 0;
5671
5672 read_lock(&em_tree->lock);
5673 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5674 read_unlock(&em_tree->lock);
5675
5676 if (!em) {
5677 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5678 chunk_start);
5679 return -EIO;
5680 }
5681
5682 if (em->start != chunk_start) {
5683 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5684 em->start, chunk_start);
5685 free_extent_map(em);
5686 return -EIO;
5687 }
5688 map = (struct map_lookup *)em->bdev;
5689
5690 length = em->len;
5691 rmap_len = map->stripe_len;
5692
5693 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5694 length = div_u64(length, map->num_stripes / map->sub_stripes);
5695 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5696 length = div_u64(length, map->num_stripes);
5697 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5698 length = div_u64(length, nr_data_stripes(map));
5699 rmap_len = map->stripe_len * nr_data_stripes(map);
5700 }
5701
5702 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5703 BUG_ON(!buf); /* -ENOMEM */
5704
5705 for (i = 0; i < map->num_stripes; i++) {
5706 if (devid && map->stripes[i].dev->devid != devid)
5707 continue;
5708 if (map->stripes[i].physical > physical ||
5709 map->stripes[i].physical + length <= physical)
5710 continue;
5711
5712 stripe_nr = physical - map->stripes[i].physical;
5713 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5714
5715 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5716 stripe_nr = stripe_nr * map->num_stripes + i;
5717 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5718 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5719 stripe_nr = stripe_nr * map->num_stripes + i;
5720 } /* else if RAID[56], multiply by nr_data_stripes().
5721 * Alternatively, just use rmap_len below instead of
5722 * map->stripe_len */
5723
5724 bytenr = chunk_start + stripe_nr * rmap_len;
5725 WARN_ON(nr >= map->num_stripes);
5726 for (j = 0; j < nr; j++) {
5727 if (buf[j] == bytenr)
5728 break;
5729 }
5730 if (j == nr) {
5731 WARN_ON(nr >= map->num_stripes);
5732 buf[nr++] = bytenr;
5733 }
5734 }
5735
5736 *logical = buf;
5737 *naddrs = nr;
5738 *stripe_len = rmap_len;
5739
5740 free_extent_map(em);
5741 return 0;
5742 }
5743
5744 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5745 {
5746 bio->bi_private = bbio->private;
5747 bio->bi_end_io = bbio->end_io;
5748 bio_endio(bio, err);
5749
5750 btrfs_put_bbio(bbio);
5751 }
5752
5753 static void btrfs_end_bio(struct bio *bio, int err)
5754 {
5755 struct btrfs_bio *bbio = bio->bi_private;
5756 int is_orig_bio = 0;
5757
5758 if (err) {
5759 atomic_inc(&bbio->error);
5760 if (err == -EIO || err == -EREMOTEIO) {
5761 unsigned int stripe_index =
5762 btrfs_io_bio(bio)->stripe_index;
5763 struct btrfs_device *dev;
5764
5765 BUG_ON(stripe_index >= bbio->num_stripes);
5766 dev = bbio->stripes[stripe_index].dev;
5767 if (dev->bdev) {
5768 if (bio->bi_rw & WRITE)
5769 btrfs_dev_stat_inc(dev,
5770 BTRFS_DEV_STAT_WRITE_ERRS);
5771 else
5772 btrfs_dev_stat_inc(dev,
5773 BTRFS_DEV_STAT_READ_ERRS);
5774 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5775 btrfs_dev_stat_inc(dev,
5776 BTRFS_DEV_STAT_FLUSH_ERRS);
5777 btrfs_dev_stat_print_on_error(dev);
5778 }
5779 }
5780 }
5781
5782 if (bio == bbio->orig_bio)
5783 is_orig_bio = 1;
5784
5785 btrfs_bio_counter_dec(bbio->fs_info);
5786
5787 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5788 if (!is_orig_bio) {
5789 bio_put(bio);
5790 bio = bbio->orig_bio;
5791 }
5792
5793 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5794 /* only send an error to the higher layers if it is
5795 * beyond the tolerance of the btrfs bio
5796 */
5797 if (atomic_read(&bbio->error) > bbio->max_errors) {
5798 err = -EIO;
5799 } else {
5800 /*
5801 * this bio is actually up to date, we didn't
5802 * go over the max number of errors
5803 */
5804 set_bit(BIO_UPTODATE, &bio->bi_flags);
5805 err = 0;
5806 }
5807
5808 btrfs_end_bbio(bbio, bio, err);
5809 } else if (!is_orig_bio) {
5810 bio_put(bio);
5811 }
5812 }
5813
5814 /*
5815 * see run_scheduled_bios for a description of why bios are collected for
5816 * async submit.
5817 *
5818 * This will add one bio to the pending list for a device and make sure
5819 * the work struct is scheduled.
5820 */
5821 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5822 struct btrfs_device *device,
5823 int rw, struct bio *bio)
5824 {
5825 int should_queue = 1;
5826 struct btrfs_pending_bios *pending_bios;
5827
5828 if (device->missing || !device->bdev) {
5829 bio_endio(bio, -EIO);
5830 return;
5831 }
5832
5833 /* don't bother with additional async steps for reads, right now */
5834 if (!(rw & REQ_WRITE)) {
5835 bio_get(bio);
5836 btrfsic_submit_bio(rw, bio);
5837 bio_put(bio);
5838 return;
5839 }
5840
5841 /*
5842 * nr_async_bios allows us to reliably return congestion to the
5843 * higher layers. Otherwise, the async bio makes it appear we have
5844 * made progress against dirty pages when we've really just put it
5845 * on a queue for later
5846 */
5847 atomic_inc(&root->fs_info->nr_async_bios);
5848 WARN_ON(bio->bi_next);
5849 bio->bi_next = NULL;
5850 bio->bi_rw |= rw;
5851
5852 spin_lock(&device->io_lock);
5853 if (bio->bi_rw & REQ_SYNC)
5854 pending_bios = &device->pending_sync_bios;
5855 else
5856 pending_bios = &device->pending_bios;
5857
5858 if (pending_bios->tail)
5859 pending_bios->tail->bi_next = bio;
5860
5861 pending_bios->tail = bio;
5862 if (!pending_bios->head)
5863 pending_bios->head = bio;
5864 if (device->running_pending)
5865 should_queue = 0;
5866
5867 spin_unlock(&device->io_lock);
5868
5869 if (should_queue)
5870 btrfs_queue_work(root->fs_info->submit_workers,
5871 &device->work);
5872 }
5873
5874 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5875 sector_t sector)
5876 {
5877 struct bio_vec *prev;
5878 struct request_queue *q = bdev_get_queue(bdev);
5879 unsigned int max_sectors = queue_max_sectors(q);
5880 struct bvec_merge_data bvm = {
5881 .bi_bdev = bdev,
5882 .bi_sector = sector,
5883 .bi_rw = bio->bi_rw,
5884 };
5885
5886 if (WARN_ON(bio->bi_vcnt == 0))
5887 return 1;
5888
5889 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5890 if (bio_sectors(bio) > max_sectors)
5891 return 0;
5892
5893 if (!q->merge_bvec_fn)
5894 return 1;
5895
5896 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5897 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5898 return 0;
5899 return 1;
5900 }
5901
5902 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5903 struct bio *bio, u64 physical, int dev_nr,
5904 int rw, int async)
5905 {
5906 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5907
5908 bio->bi_private = bbio;
5909 btrfs_io_bio(bio)->stripe_index = dev_nr;
5910 bio->bi_end_io = btrfs_end_bio;
5911 bio->bi_iter.bi_sector = physical >> 9;
5912 #ifdef DEBUG
5913 {
5914 struct rcu_string *name;
5915
5916 rcu_read_lock();
5917 name = rcu_dereference(dev->name);
5918 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5919 "(%s id %llu), size=%u\n", rw,
5920 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5921 name->str, dev->devid, bio->bi_iter.bi_size);
5922 rcu_read_unlock();
5923 }
5924 #endif
5925 bio->bi_bdev = dev->bdev;
5926
5927 btrfs_bio_counter_inc_noblocked(root->fs_info);
5928
5929 if (async)
5930 btrfs_schedule_bio(root, dev, rw, bio);
5931 else
5932 btrfsic_submit_bio(rw, bio);
5933 }
5934
5935 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5936 struct bio *first_bio, struct btrfs_device *dev,
5937 int dev_nr, int rw, int async)
5938 {
5939 struct bio_vec *bvec = first_bio->bi_io_vec;
5940 struct bio *bio;
5941 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5942 u64 physical = bbio->stripes[dev_nr].physical;
5943
5944 again:
5945 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5946 if (!bio)
5947 return -ENOMEM;
5948
5949 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5950 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5951 bvec->bv_offset) < bvec->bv_len) {
5952 u64 len = bio->bi_iter.bi_size;
5953
5954 atomic_inc(&bbio->stripes_pending);
5955 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5956 rw, async);
5957 physical += len;
5958 goto again;
5959 }
5960 bvec++;
5961 }
5962
5963 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5964 return 0;
5965 }
5966
5967 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5968 {
5969 atomic_inc(&bbio->error);
5970 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5971 /* Shoud be the original bio. */
5972 WARN_ON(bio != bbio->orig_bio);
5973
5974 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5975 bio->bi_iter.bi_sector = logical >> 9;
5976
5977 btrfs_end_bbio(bbio, bio, -EIO);
5978 }
5979 }
5980
5981 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5982 int mirror_num, int async_submit)
5983 {
5984 struct btrfs_device *dev;
5985 struct bio *first_bio = bio;
5986 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5987 u64 length = 0;
5988 u64 map_length;
5989 int ret;
5990 int dev_nr;
5991 int total_devs;
5992 struct btrfs_bio *bbio = NULL;
5993
5994 length = bio->bi_iter.bi_size;
5995 map_length = length;
5996
5997 btrfs_bio_counter_inc_blocked(root->fs_info);
5998 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5999 mirror_num, 1);
6000 if (ret) {
6001 btrfs_bio_counter_dec(root->fs_info);
6002 return ret;
6003 }
6004
6005 total_devs = bbio->num_stripes;
6006 bbio->orig_bio = first_bio;
6007 bbio->private = first_bio->bi_private;
6008 bbio->end_io = first_bio->bi_end_io;
6009 bbio->fs_info = root->fs_info;
6010 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6011
6012 if (bbio->raid_map) {
6013 /* In this case, map_length has been set to the length of
6014 a single stripe; not the whole write */
6015 if (rw & WRITE) {
6016 ret = raid56_parity_write(root, bio, bbio, map_length);
6017 } else {
6018 ret = raid56_parity_recover(root, bio, bbio, map_length,
6019 mirror_num, 1);
6020 }
6021
6022 btrfs_bio_counter_dec(root->fs_info);
6023 return ret;
6024 }
6025
6026 if (map_length < length) {
6027 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6028 logical, length, map_length);
6029 BUG();
6030 }
6031
6032 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6033 dev = bbio->stripes[dev_nr].dev;
6034 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6035 bbio_error(bbio, first_bio, logical);
6036 continue;
6037 }
6038
6039 /*
6040 * Check and see if we're ok with this bio based on it's size
6041 * and offset with the given device.
6042 */
6043 if (!bio_size_ok(dev->bdev, first_bio,
6044 bbio->stripes[dev_nr].physical >> 9)) {
6045 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
6046 dev_nr, rw, async_submit);
6047 BUG_ON(ret);
6048 continue;
6049 }
6050
6051 if (dev_nr < total_devs - 1) {
6052 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6053 BUG_ON(!bio); /* -ENOMEM */
6054 } else
6055 bio = first_bio;
6056
6057 submit_stripe_bio(root, bbio, bio,
6058 bbio->stripes[dev_nr].physical, dev_nr, rw,
6059 async_submit);
6060 }
6061 btrfs_bio_counter_dec(root->fs_info);
6062 return 0;
6063 }
6064
6065 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6066 u8 *uuid, u8 *fsid)
6067 {
6068 struct btrfs_device *device;
6069 struct btrfs_fs_devices *cur_devices;
6070
6071 cur_devices = fs_info->fs_devices;
6072 while (cur_devices) {
6073 if (!fsid ||
6074 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6075 device = __find_device(&cur_devices->devices,
6076 devid, uuid);
6077 if (device)
6078 return device;
6079 }
6080 cur_devices = cur_devices->seed;
6081 }
6082 return NULL;
6083 }
6084
6085 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6086 struct btrfs_fs_devices *fs_devices,
6087 u64 devid, u8 *dev_uuid)
6088 {
6089 struct btrfs_device *device;
6090
6091 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6092 if (IS_ERR(device))
6093 return NULL;
6094
6095 list_add(&device->dev_list, &fs_devices->devices);
6096 device->fs_devices = fs_devices;
6097 fs_devices->num_devices++;
6098
6099 device->missing = 1;
6100 fs_devices->missing_devices++;
6101
6102 return device;
6103 }
6104
6105 /**
6106 * btrfs_alloc_device - allocate struct btrfs_device
6107 * @fs_info: used only for generating a new devid, can be NULL if
6108 * devid is provided (i.e. @devid != NULL).
6109 * @devid: a pointer to devid for this device. If NULL a new devid
6110 * is generated.
6111 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6112 * is generated.
6113 *
6114 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6115 * on error. Returned struct is not linked onto any lists and can be
6116 * destroyed with kfree() right away.
6117 */
6118 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6119 const u64 *devid,
6120 const u8 *uuid)
6121 {
6122 struct btrfs_device *dev;
6123 u64 tmp;
6124
6125 if (WARN_ON(!devid && !fs_info))
6126 return ERR_PTR(-EINVAL);
6127
6128 dev = __alloc_device();
6129 if (IS_ERR(dev))
6130 return dev;
6131
6132 if (devid)
6133 tmp = *devid;
6134 else {
6135 int ret;
6136
6137 ret = find_next_devid(fs_info, &tmp);
6138 if (ret) {
6139 kfree(dev);
6140 return ERR_PTR(ret);
6141 }
6142 }
6143 dev->devid = tmp;
6144
6145 if (uuid)
6146 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6147 else
6148 generate_random_uuid(dev->uuid);
6149
6150 btrfs_init_work(&dev->work, btrfs_submit_helper,
6151 pending_bios_fn, NULL, NULL);
6152
6153 return dev;
6154 }
6155
6156 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6157 struct extent_buffer *leaf,
6158 struct btrfs_chunk *chunk)
6159 {
6160 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6161 struct map_lookup *map;
6162 struct extent_map *em;
6163 u64 logical;
6164 u64 length;
6165 u64 devid;
6166 u8 uuid[BTRFS_UUID_SIZE];
6167 int num_stripes;
6168 int ret;
6169 int i;
6170
6171 logical = key->offset;
6172 length = btrfs_chunk_length(leaf, chunk);
6173
6174 read_lock(&map_tree->map_tree.lock);
6175 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6176 read_unlock(&map_tree->map_tree.lock);
6177
6178 /* already mapped? */
6179 if (em && em->start <= logical && em->start + em->len > logical) {
6180 free_extent_map(em);
6181 return 0;
6182 } else if (em) {
6183 free_extent_map(em);
6184 }
6185
6186 em = alloc_extent_map();
6187 if (!em)
6188 return -ENOMEM;
6189 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6190 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6191 if (!map) {
6192 free_extent_map(em);
6193 return -ENOMEM;
6194 }
6195
6196 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6197 em->bdev = (struct block_device *)map;
6198 em->start = logical;
6199 em->len = length;
6200 em->orig_start = 0;
6201 em->block_start = 0;
6202 em->block_len = em->len;
6203
6204 map->num_stripes = num_stripes;
6205 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6206 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6207 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6208 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6209 map->type = btrfs_chunk_type(leaf, chunk);
6210 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6211 for (i = 0; i < num_stripes; i++) {
6212 map->stripes[i].physical =
6213 btrfs_stripe_offset_nr(leaf, chunk, i);
6214 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6215 read_extent_buffer(leaf, uuid, (unsigned long)
6216 btrfs_stripe_dev_uuid_nr(chunk, i),
6217 BTRFS_UUID_SIZE);
6218 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6219 uuid, NULL);
6220 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6221 free_extent_map(em);
6222 return -EIO;
6223 }
6224 if (!map->stripes[i].dev) {
6225 map->stripes[i].dev =
6226 add_missing_dev(root, root->fs_info->fs_devices,
6227 devid, uuid);
6228 if (!map->stripes[i].dev) {
6229 free_extent_map(em);
6230 return -EIO;
6231 }
6232 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6233 devid, uuid);
6234 }
6235 map->stripes[i].dev->in_fs_metadata = 1;
6236 }
6237
6238 write_lock(&map_tree->map_tree.lock);
6239 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6240 write_unlock(&map_tree->map_tree.lock);
6241 BUG_ON(ret); /* Tree corruption */
6242 free_extent_map(em);
6243
6244 return 0;
6245 }
6246
6247 static void fill_device_from_item(struct extent_buffer *leaf,
6248 struct btrfs_dev_item *dev_item,
6249 struct btrfs_device *device)
6250 {
6251 unsigned long ptr;
6252
6253 device->devid = btrfs_device_id(leaf, dev_item);
6254 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6255 device->total_bytes = device->disk_total_bytes;
6256 device->commit_total_bytes = device->disk_total_bytes;
6257 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6258 device->commit_bytes_used = device->bytes_used;
6259 device->type = btrfs_device_type(leaf, dev_item);
6260 device->io_align = btrfs_device_io_align(leaf, dev_item);
6261 device->io_width = btrfs_device_io_width(leaf, dev_item);
6262 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6263 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6264 device->is_tgtdev_for_dev_replace = 0;
6265
6266 ptr = btrfs_device_uuid(dev_item);
6267 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6268 }
6269
6270 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6271 u8 *fsid)
6272 {
6273 struct btrfs_fs_devices *fs_devices;
6274 int ret;
6275
6276 BUG_ON(!mutex_is_locked(&uuid_mutex));
6277
6278 fs_devices = root->fs_info->fs_devices->seed;
6279 while (fs_devices) {
6280 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6281 return fs_devices;
6282
6283 fs_devices = fs_devices->seed;
6284 }
6285
6286 fs_devices = find_fsid(fsid);
6287 if (!fs_devices) {
6288 if (!btrfs_test_opt(root, DEGRADED))
6289 return ERR_PTR(-ENOENT);
6290
6291 fs_devices = alloc_fs_devices(fsid);
6292 if (IS_ERR(fs_devices))
6293 return fs_devices;
6294
6295 fs_devices->seeding = 1;
6296 fs_devices->opened = 1;
6297 return fs_devices;
6298 }
6299
6300 fs_devices = clone_fs_devices(fs_devices);
6301 if (IS_ERR(fs_devices))
6302 return fs_devices;
6303
6304 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6305 root->fs_info->bdev_holder);
6306 if (ret) {
6307 free_fs_devices(fs_devices);
6308 fs_devices = ERR_PTR(ret);
6309 goto out;
6310 }
6311
6312 if (!fs_devices->seeding) {
6313 __btrfs_close_devices(fs_devices);
6314 free_fs_devices(fs_devices);
6315 fs_devices = ERR_PTR(-EINVAL);
6316 goto out;
6317 }
6318
6319 fs_devices->seed = root->fs_info->fs_devices->seed;
6320 root->fs_info->fs_devices->seed = fs_devices;
6321 out:
6322 return fs_devices;
6323 }
6324
6325 static int read_one_dev(struct btrfs_root *root,
6326 struct extent_buffer *leaf,
6327 struct btrfs_dev_item *dev_item)
6328 {
6329 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6330 struct btrfs_device *device;
6331 u64 devid;
6332 int ret;
6333 u8 fs_uuid[BTRFS_UUID_SIZE];
6334 u8 dev_uuid[BTRFS_UUID_SIZE];
6335
6336 devid = btrfs_device_id(leaf, dev_item);
6337 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6338 BTRFS_UUID_SIZE);
6339 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6340 BTRFS_UUID_SIZE);
6341
6342 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6343 fs_devices = open_seed_devices(root, fs_uuid);
6344 if (IS_ERR(fs_devices))
6345 return PTR_ERR(fs_devices);
6346 }
6347
6348 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6349 if (!device) {
6350 if (!btrfs_test_opt(root, DEGRADED))
6351 return -EIO;
6352
6353 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6354 if (!device)
6355 return -ENOMEM;
6356 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6357 devid, dev_uuid);
6358 } else {
6359 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6360 return -EIO;
6361
6362 if(!device->bdev && !device->missing) {
6363 /*
6364 * this happens when a device that was properly setup
6365 * in the device info lists suddenly goes bad.
6366 * device->bdev is NULL, and so we have to set
6367 * device->missing to one here
6368 */
6369 device->fs_devices->missing_devices++;
6370 device->missing = 1;
6371 }
6372
6373 /* Move the device to its own fs_devices */
6374 if (device->fs_devices != fs_devices) {
6375 ASSERT(device->missing);
6376
6377 list_move(&device->dev_list, &fs_devices->devices);
6378 device->fs_devices->num_devices--;
6379 fs_devices->num_devices++;
6380
6381 device->fs_devices->missing_devices--;
6382 fs_devices->missing_devices++;
6383
6384 device->fs_devices = fs_devices;
6385 }
6386 }
6387
6388 if (device->fs_devices != root->fs_info->fs_devices) {
6389 BUG_ON(device->writeable);
6390 if (device->generation !=
6391 btrfs_device_generation(leaf, dev_item))
6392 return -EINVAL;
6393 }
6394
6395 fill_device_from_item(leaf, dev_item, device);
6396 device->in_fs_metadata = 1;
6397 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6398 device->fs_devices->total_rw_bytes += device->total_bytes;
6399 spin_lock(&root->fs_info->free_chunk_lock);
6400 root->fs_info->free_chunk_space += device->total_bytes -
6401 device->bytes_used;
6402 spin_unlock(&root->fs_info->free_chunk_lock);
6403 }
6404 ret = 0;
6405 return ret;
6406 }
6407
6408 int btrfs_read_sys_array(struct btrfs_root *root)
6409 {
6410 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6411 struct extent_buffer *sb;
6412 struct btrfs_disk_key *disk_key;
6413 struct btrfs_chunk *chunk;
6414 u8 *array_ptr;
6415 unsigned long sb_array_offset;
6416 int ret = 0;
6417 u32 num_stripes;
6418 u32 array_size;
6419 u32 len = 0;
6420 u32 cur_offset;
6421 struct btrfs_key key;
6422
6423 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6424 /*
6425 * This will create extent buffer of nodesize, superblock size is
6426 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6427 * overallocate but we can keep it as-is, only the first page is used.
6428 */
6429 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6430 if (!sb)
6431 return -ENOMEM;
6432 btrfs_set_buffer_uptodate(sb);
6433 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6434 /*
6435 * The sb extent buffer is artifical and just used to read the system array.
6436 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6437 * pages up-to-date when the page is larger: extent does not cover the
6438 * whole page and consequently check_page_uptodate does not find all
6439 * the page's extents up-to-date (the hole beyond sb),
6440 * write_extent_buffer then triggers a WARN_ON.
6441 *
6442 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6443 * but sb spans only this function. Add an explicit SetPageUptodate call
6444 * to silence the warning eg. on PowerPC 64.
6445 */
6446 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6447 SetPageUptodate(sb->pages[0]);
6448
6449 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6450 array_size = btrfs_super_sys_array_size(super_copy);
6451
6452 array_ptr = super_copy->sys_chunk_array;
6453 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6454 cur_offset = 0;
6455
6456 while (cur_offset < array_size) {
6457 disk_key = (struct btrfs_disk_key *)array_ptr;
6458 len = sizeof(*disk_key);
6459 if (cur_offset + len > array_size)
6460 goto out_short_read;
6461
6462 btrfs_disk_key_to_cpu(&key, disk_key);
6463
6464 array_ptr += len;
6465 sb_array_offset += len;
6466 cur_offset += len;
6467
6468 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6469 chunk = (struct btrfs_chunk *)sb_array_offset;
6470 /*
6471 * At least one btrfs_chunk with one stripe must be
6472 * present, exact stripe count check comes afterwards
6473 */
6474 len = btrfs_chunk_item_size(1);
6475 if (cur_offset + len > array_size)
6476 goto out_short_read;
6477
6478 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6479 len = btrfs_chunk_item_size(num_stripes);
6480 if (cur_offset + len > array_size)
6481 goto out_short_read;
6482
6483 ret = read_one_chunk(root, &key, sb, chunk);
6484 if (ret)
6485 break;
6486 } else {
6487 ret = -EIO;
6488 break;
6489 }
6490 array_ptr += len;
6491 sb_array_offset += len;
6492 cur_offset += len;
6493 }
6494 free_extent_buffer(sb);
6495 return ret;
6496
6497 out_short_read:
6498 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6499 len, cur_offset);
6500 free_extent_buffer(sb);
6501 return -EIO;
6502 }
6503
6504 int btrfs_read_chunk_tree(struct btrfs_root *root)
6505 {
6506 struct btrfs_path *path;
6507 struct extent_buffer *leaf;
6508 struct btrfs_key key;
6509 struct btrfs_key found_key;
6510 int ret;
6511 int slot;
6512
6513 root = root->fs_info->chunk_root;
6514
6515 path = btrfs_alloc_path();
6516 if (!path)
6517 return -ENOMEM;
6518
6519 mutex_lock(&uuid_mutex);
6520 lock_chunks(root);
6521
6522 /*
6523 * Read all device items, and then all the chunk items. All
6524 * device items are found before any chunk item (their object id
6525 * is smaller than the lowest possible object id for a chunk
6526 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6527 */
6528 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6529 key.offset = 0;
6530 key.type = 0;
6531 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6532 if (ret < 0)
6533 goto error;
6534 while (1) {
6535 leaf = path->nodes[0];
6536 slot = path->slots[0];
6537 if (slot >= btrfs_header_nritems(leaf)) {
6538 ret = btrfs_next_leaf(root, path);
6539 if (ret == 0)
6540 continue;
6541 if (ret < 0)
6542 goto error;
6543 break;
6544 }
6545 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6546 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6547 struct btrfs_dev_item *dev_item;
6548 dev_item = btrfs_item_ptr(leaf, slot,
6549 struct btrfs_dev_item);
6550 ret = read_one_dev(root, leaf, dev_item);
6551 if (ret)
6552 goto error;
6553 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6554 struct btrfs_chunk *chunk;
6555 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6556 ret = read_one_chunk(root, &found_key, leaf, chunk);
6557 if (ret)
6558 goto error;
6559 }
6560 path->slots[0]++;
6561 }
6562 ret = 0;
6563 error:
6564 unlock_chunks(root);
6565 mutex_unlock(&uuid_mutex);
6566
6567 btrfs_free_path(path);
6568 return ret;
6569 }
6570
6571 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6572 {
6573 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6574 struct btrfs_device *device;
6575
6576 while (fs_devices) {
6577 mutex_lock(&fs_devices->device_list_mutex);
6578 list_for_each_entry(device, &fs_devices->devices, dev_list)
6579 device->dev_root = fs_info->dev_root;
6580 mutex_unlock(&fs_devices->device_list_mutex);
6581
6582 fs_devices = fs_devices->seed;
6583 }
6584 }
6585
6586 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6587 {
6588 int i;
6589
6590 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6591 btrfs_dev_stat_reset(dev, i);
6592 }
6593
6594 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6595 {
6596 struct btrfs_key key;
6597 struct btrfs_key found_key;
6598 struct btrfs_root *dev_root = fs_info->dev_root;
6599 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6600 struct extent_buffer *eb;
6601 int slot;
6602 int ret = 0;
6603 struct btrfs_device *device;
6604 struct btrfs_path *path = NULL;
6605 int i;
6606
6607 path = btrfs_alloc_path();
6608 if (!path) {
6609 ret = -ENOMEM;
6610 goto out;
6611 }
6612
6613 mutex_lock(&fs_devices->device_list_mutex);
6614 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6615 int item_size;
6616 struct btrfs_dev_stats_item *ptr;
6617
6618 key.objectid = 0;
6619 key.type = BTRFS_DEV_STATS_KEY;
6620 key.offset = device->devid;
6621 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6622 if (ret) {
6623 __btrfs_reset_dev_stats(device);
6624 device->dev_stats_valid = 1;
6625 btrfs_release_path(path);
6626 continue;
6627 }
6628 slot = path->slots[0];
6629 eb = path->nodes[0];
6630 btrfs_item_key_to_cpu(eb, &found_key, slot);
6631 item_size = btrfs_item_size_nr(eb, slot);
6632
6633 ptr = btrfs_item_ptr(eb, slot,
6634 struct btrfs_dev_stats_item);
6635
6636 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6637 if (item_size >= (1 + i) * sizeof(__le64))
6638 btrfs_dev_stat_set(device, i,
6639 btrfs_dev_stats_value(eb, ptr, i));
6640 else
6641 btrfs_dev_stat_reset(device, i);
6642 }
6643
6644 device->dev_stats_valid = 1;
6645 btrfs_dev_stat_print_on_load(device);
6646 btrfs_release_path(path);
6647 }
6648 mutex_unlock(&fs_devices->device_list_mutex);
6649
6650 out:
6651 btrfs_free_path(path);
6652 return ret < 0 ? ret : 0;
6653 }
6654
6655 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6656 struct btrfs_root *dev_root,
6657 struct btrfs_device *device)
6658 {
6659 struct btrfs_path *path;
6660 struct btrfs_key key;
6661 struct extent_buffer *eb;
6662 struct btrfs_dev_stats_item *ptr;
6663 int ret;
6664 int i;
6665
6666 key.objectid = 0;
6667 key.type = BTRFS_DEV_STATS_KEY;
6668 key.offset = device->devid;
6669
6670 path = btrfs_alloc_path();
6671 BUG_ON(!path);
6672 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6673 if (ret < 0) {
6674 printk_in_rcu(KERN_WARNING "BTRFS: "
6675 "error %d while searching for dev_stats item for device %s!\n",
6676 ret, rcu_str_deref(device->name));
6677 goto out;
6678 }
6679
6680 if (ret == 0 &&
6681 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6682 /* need to delete old one and insert a new one */
6683 ret = btrfs_del_item(trans, dev_root, path);
6684 if (ret != 0) {
6685 printk_in_rcu(KERN_WARNING "BTRFS: "
6686 "delete too small dev_stats item for device %s failed %d!\n",
6687 rcu_str_deref(device->name), ret);
6688 goto out;
6689 }
6690 ret = 1;
6691 }
6692
6693 if (ret == 1) {
6694 /* need to insert a new item */
6695 btrfs_release_path(path);
6696 ret = btrfs_insert_empty_item(trans, dev_root, path,
6697 &key, sizeof(*ptr));
6698 if (ret < 0) {
6699 printk_in_rcu(KERN_WARNING "BTRFS: "
6700 "insert dev_stats item for device %s failed %d!\n",
6701 rcu_str_deref(device->name), ret);
6702 goto out;
6703 }
6704 }
6705
6706 eb = path->nodes[0];
6707 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6708 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6709 btrfs_set_dev_stats_value(eb, ptr, i,
6710 btrfs_dev_stat_read(device, i));
6711 btrfs_mark_buffer_dirty(eb);
6712
6713 out:
6714 btrfs_free_path(path);
6715 return ret;
6716 }
6717
6718 /*
6719 * called from commit_transaction. Writes all changed device stats to disk.
6720 */
6721 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6722 struct btrfs_fs_info *fs_info)
6723 {
6724 struct btrfs_root *dev_root = fs_info->dev_root;
6725 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6726 struct btrfs_device *device;
6727 int stats_cnt;
6728 int ret = 0;
6729
6730 mutex_lock(&fs_devices->device_list_mutex);
6731 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6732 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6733 continue;
6734
6735 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6736 ret = update_dev_stat_item(trans, dev_root, device);
6737 if (!ret)
6738 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6739 }
6740 mutex_unlock(&fs_devices->device_list_mutex);
6741
6742 return ret;
6743 }
6744
6745 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6746 {
6747 btrfs_dev_stat_inc(dev, index);
6748 btrfs_dev_stat_print_on_error(dev);
6749 }
6750
6751 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6752 {
6753 if (!dev->dev_stats_valid)
6754 return;
6755 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6756 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6757 rcu_str_deref(dev->name),
6758 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6759 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6760 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6761 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6762 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6763 }
6764
6765 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6766 {
6767 int i;
6768
6769 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6770 if (btrfs_dev_stat_read(dev, i) != 0)
6771 break;
6772 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6773 return; /* all values == 0, suppress message */
6774
6775 printk_in_rcu(KERN_INFO "BTRFS: "
6776 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6777 rcu_str_deref(dev->name),
6778 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6779 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6780 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6781 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6782 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6783 }
6784
6785 int btrfs_get_dev_stats(struct btrfs_root *root,
6786 struct btrfs_ioctl_get_dev_stats *stats)
6787 {
6788 struct btrfs_device *dev;
6789 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6790 int i;
6791
6792 mutex_lock(&fs_devices->device_list_mutex);
6793 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6794 mutex_unlock(&fs_devices->device_list_mutex);
6795
6796 if (!dev) {
6797 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6798 return -ENODEV;
6799 } else if (!dev->dev_stats_valid) {
6800 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6801 return -ENODEV;
6802 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6803 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6804 if (stats->nr_items > i)
6805 stats->values[i] =
6806 btrfs_dev_stat_read_and_reset(dev, i);
6807 else
6808 btrfs_dev_stat_reset(dev, i);
6809 }
6810 } else {
6811 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6812 if (stats->nr_items > i)
6813 stats->values[i] = btrfs_dev_stat_read(dev, i);
6814 }
6815 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6816 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6817 return 0;
6818 }
6819
6820 int btrfs_scratch_superblock(struct btrfs_device *device)
6821 {
6822 struct buffer_head *bh;
6823 struct btrfs_super_block *disk_super;
6824
6825 bh = btrfs_read_dev_super(device->bdev);
6826 if (!bh)
6827 return -EINVAL;
6828 disk_super = (struct btrfs_super_block *)bh->b_data;
6829
6830 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6831 set_buffer_dirty(bh);
6832 sync_dirty_buffer(bh);
6833 brelse(bh);
6834
6835 return 0;
6836 }
6837
6838 /*
6839 * Update the size of all devices, which is used for writing out the
6840 * super blocks.
6841 */
6842 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6843 {
6844 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6845 struct btrfs_device *curr, *next;
6846
6847 if (list_empty(&fs_devices->resized_devices))
6848 return;
6849
6850 mutex_lock(&fs_devices->device_list_mutex);
6851 lock_chunks(fs_info->dev_root);
6852 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6853 resized_list) {
6854 list_del_init(&curr->resized_list);
6855 curr->commit_total_bytes = curr->disk_total_bytes;
6856 }
6857 unlock_chunks(fs_info->dev_root);
6858 mutex_unlock(&fs_devices->device_list_mutex);
6859 }
6860
6861 /* Must be invoked during the transaction commit */
6862 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6863 struct btrfs_transaction *transaction)
6864 {
6865 struct extent_map *em;
6866 struct map_lookup *map;
6867 struct btrfs_device *dev;
6868 int i;
6869
6870 if (list_empty(&transaction->pending_chunks))
6871 return;
6872
6873 /* In order to kick the device replace finish process */
6874 lock_chunks(root);
6875 list_for_each_entry(em, &transaction->pending_chunks, list) {
6876 map = (struct map_lookup *)em->bdev;
6877
6878 for (i = 0; i < map->num_stripes; i++) {
6879 dev = map->stripes[i].dev;
6880 dev->commit_bytes_used = dev->bytes_used;
6881 }
6882 }
6883 unlock_chunks(root);
6884 }
6885
6886 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6887 {
6888 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6889 while (fs_devices) {
6890 fs_devices->fs_info = fs_info;
6891 fs_devices = fs_devices->seed;
6892 }
6893 }
6894
6895 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6896 {
6897 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6898 while (fs_devices) {
6899 fs_devices->fs_info = NULL;
6900 fs_devices = fs_devices->seed;
6901 }
6902 }
Linux with added FPC-III support