]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/volumes.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: comment the rest of implicit barriers before waitqueue_active
[mirror_ubuntu-artful-kernel.git] / fs / btrfs / volumes.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <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 /*
349 * atomic_dec_return implies a barrier for waitqueue_active
350 */
351 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
352 waitqueue_active(&fs_info->async_submit_wait))
353 wake_up(&fs_info->async_submit_wait);
354
355 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
356
357 /*
358 * if we're doing the sync list, record that our
359 * plug has some sync requests on it
360 *
361 * If we're doing the regular list and there are
362 * sync requests sitting around, unplug before
363 * we add more
364 */
365 if (pending_bios == &device->pending_sync_bios) {
366 sync_pending = 1;
367 } else if (sync_pending) {
368 blk_finish_plug(&plug);
369 blk_start_plug(&plug);
370 sync_pending = 0;
371 }
372
373 btrfsic_submit_bio(cur->bi_rw, cur);
374 num_run++;
375 batch_run++;
376
377 cond_resched();
378
379 /*
380 * we made progress, there is more work to do and the bdi
381 * is now congested. Back off and let other work structs
382 * run instead
383 */
384 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
385 fs_info->fs_devices->open_devices > 1) {
386 struct io_context *ioc;
387
388 ioc = current->io_context;
389
390 /*
391 * the main goal here is that we don't want to
392 * block if we're going to be able to submit
393 * more requests without blocking.
394 *
395 * This code does two great things, it pokes into
396 * the elevator code from a filesystem _and_
397 * it makes assumptions about how batching works.
398 */
399 if (ioc && ioc->nr_batch_requests > 0 &&
400 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
401 (last_waited == 0 ||
402 ioc->last_waited == last_waited)) {
403 /*
404 * we want to go through our batch of
405 * requests and stop. So, we copy out
406 * the ioc->last_waited time and test
407 * against it before looping
408 */
409 last_waited = ioc->last_waited;
410 cond_resched();
411 continue;
412 }
413 spin_lock(&device->io_lock);
414 requeue_list(pending_bios, pending, tail);
415 device->running_pending = 1;
416
417 spin_unlock(&device->io_lock);
418 btrfs_queue_work(fs_info->submit_workers,
419 &device->work);
420 goto done;
421 }
422 /* unplug every 64 requests just for good measure */
423 if (batch_run % 64 == 0) {
424 blk_finish_plug(&plug);
425 blk_start_plug(&plug);
426 sync_pending = 0;
427 }
428 }
429
430 cond_resched();
431 if (again)
432 goto loop;
433
434 spin_lock(&device->io_lock);
435 if (device->pending_bios.head || device->pending_sync_bios.head)
436 goto loop_lock;
437 spin_unlock(&device->io_lock);
438
439 done:
440 blk_finish_plug(&plug);
441 }
442
443 static void pending_bios_fn(struct btrfs_work *work)
444 {
445 struct btrfs_device *device;
446
447 device = container_of(work, struct btrfs_device, work);
448 run_scheduled_bios(device);
449 }
450
451
452 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
453 {
454 struct btrfs_fs_devices *fs_devs;
455 struct btrfs_device *dev;
456
457 if (!cur_dev->name)
458 return;
459
460 list_for_each_entry(fs_devs, &fs_uuids, list) {
461 int del = 1;
462
463 if (fs_devs->opened)
464 continue;
465 if (fs_devs->seeding)
466 continue;
467
468 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
469
470 if (dev == cur_dev)
471 continue;
472 if (!dev->name)
473 continue;
474
475 /*
476 * Todo: This won't be enough. What if the same device
477 * comes back (with new uuid and) with its mapper path?
478 * But for now, this does help as mostly an admin will
479 * either use mapper or non mapper path throughout.
480 */
481 rcu_read_lock();
482 del = strcmp(rcu_str_deref(dev->name),
483 rcu_str_deref(cur_dev->name));
484 rcu_read_unlock();
485 if (!del)
486 break;
487 }
488
489 if (!del) {
490 /* delete the stale device */
491 if (fs_devs->num_devices == 1) {
492 btrfs_sysfs_remove_fsid(fs_devs);
493 list_del(&fs_devs->list);
494 free_fs_devices(fs_devs);
495 } else {
496 fs_devs->num_devices--;
497 list_del(&dev->dev_list);
498 rcu_string_free(dev->name);
499 kfree(dev);
500 }
501 break;
502 }
503 }
504 }
505
506 /*
507 * Add new device to list of registered devices
508 *
509 * Returns:
510 * 1 - first time device is seen
511 * 0 - device already known
512 * < 0 - error
513 */
514 static noinline int device_list_add(const char *path,
515 struct btrfs_super_block *disk_super,
516 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
517 {
518 struct btrfs_device *device;
519 struct btrfs_fs_devices *fs_devices;
520 struct rcu_string *name;
521 int ret = 0;
522 u64 found_transid = btrfs_super_generation(disk_super);
523
524 fs_devices = find_fsid(disk_super->fsid);
525 if (!fs_devices) {
526 fs_devices = alloc_fs_devices(disk_super->fsid);
527 if (IS_ERR(fs_devices))
528 return PTR_ERR(fs_devices);
529
530 list_add(&fs_devices->list, &fs_uuids);
531
532 device = NULL;
533 } else {
534 device = __find_device(&fs_devices->devices, devid,
535 disk_super->dev_item.uuid);
536 }
537
538 if (!device) {
539 if (fs_devices->opened)
540 return -EBUSY;
541
542 device = btrfs_alloc_device(NULL, &devid,
543 disk_super->dev_item.uuid);
544 if (IS_ERR(device)) {
545 /* we can safely leave the fs_devices entry around */
546 return PTR_ERR(device);
547 }
548
549 name = rcu_string_strdup(path, GFP_NOFS);
550 if (!name) {
551 kfree(device);
552 return -ENOMEM;
553 }
554 rcu_assign_pointer(device->name, name);
555
556 mutex_lock(&fs_devices->device_list_mutex);
557 list_add_rcu(&device->dev_list, &fs_devices->devices);
558 fs_devices->num_devices++;
559 mutex_unlock(&fs_devices->device_list_mutex);
560
561 ret = 1;
562 device->fs_devices = fs_devices;
563 } else if (!device->name || strcmp(device->name->str, path)) {
564 /*
565 * When FS is already mounted.
566 * 1. If you are here and if the device->name is NULL that
567 * means this device was missing at time of FS mount.
568 * 2. If you are here and if the device->name is different
569 * from 'path' that means either
570 * a. The same device disappeared and reappeared with
571 * different name. or
572 * b. The missing-disk-which-was-replaced, has
573 * reappeared now.
574 *
575 * We must allow 1 and 2a above. But 2b would be a spurious
576 * and unintentional.
577 *
578 * Further in case of 1 and 2a above, the disk at 'path'
579 * would have missed some transaction when it was away and
580 * in case of 2a the stale bdev has to be updated as well.
581 * 2b must not be allowed at all time.
582 */
583
584 /*
585 * For now, we do allow update to btrfs_fs_device through the
586 * btrfs dev scan cli after FS has been mounted. We're still
587 * tracking a problem where systems fail mount by subvolume id
588 * when we reject replacement on a mounted FS.
589 */
590 if (!fs_devices->opened && found_transid < device->generation) {
591 /*
592 * That is if the FS is _not_ mounted and if you
593 * are here, that means there is more than one
594 * disk with same uuid and devid.We keep the one
595 * with larger generation number or the last-in if
596 * generation are equal.
597 */
598 return -EEXIST;
599 }
600
601 name = rcu_string_strdup(path, GFP_NOFS);
602 if (!name)
603 return -ENOMEM;
604 rcu_string_free(device->name);
605 rcu_assign_pointer(device->name, name);
606 if (device->missing) {
607 fs_devices->missing_devices--;
608 device->missing = 0;
609 }
610 }
611
612 /*
613 * Unmount does not free the btrfs_device struct but would zero
614 * generation along with most of the other members. So just update
615 * it back. We need it to pick the disk with largest generation
616 * (as above).
617 */
618 if (!fs_devices->opened)
619 device->generation = found_transid;
620
621 /*
622 * if there is new btrfs on an already registered device,
623 * then remove the stale device entry.
624 */
625 btrfs_free_stale_device(device);
626
627 *fs_devices_ret = fs_devices;
628
629 return ret;
630 }
631
632 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
633 {
634 struct btrfs_fs_devices *fs_devices;
635 struct btrfs_device *device;
636 struct btrfs_device *orig_dev;
637
638 fs_devices = alloc_fs_devices(orig->fsid);
639 if (IS_ERR(fs_devices))
640 return fs_devices;
641
642 mutex_lock(&orig->device_list_mutex);
643 fs_devices->total_devices = orig->total_devices;
644
645 /* We have held the volume lock, it is safe to get the devices. */
646 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
647 struct rcu_string *name;
648
649 device = btrfs_alloc_device(NULL, &orig_dev->devid,
650 orig_dev->uuid);
651 if (IS_ERR(device))
652 goto error;
653
654 /*
655 * This is ok to do without rcu read locked because we hold the
656 * uuid mutex so nothing we touch in here is going to disappear.
657 */
658 if (orig_dev->name) {
659 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
660 if (!name) {
661 kfree(device);
662 goto error;
663 }
664 rcu_assign_pointer(device->name, name);
665 }
666
667 list_add(&device->dev_list, &fs_devices->devices);
668 device->fs_devices = fs_devices;
669 fs_devices->num_devices++;
670 }
671 mutex_unlock(&orig->device_list_mutex);
672 return fs_devices;
673 error:
674 mutex_unlock(&orig->device_list_mutex);
675 free_fs_devices(fs_devices);
676 return ERR_PTR(-ENOMEM);
677 }
678
679 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
680 {
681 struct btrfs_device *device, *next;
682 struct btrfs_device *latest_dev = NULL;
683
684 mutex_lock(&uuid_mutex);
685 again:
686 /* This is the initialized path, it is safe to release the devices. */
687 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
688 if (device->in_fs_metadata) {
689 if (!device->is_tgtdev_for_dev_replace &&
690 (!latest_dev ||
691 device->generation > latest_dev->generation)) {
692 latest_dev = device;
693 }
694 continue;
695 }
696
697 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
698 /*
699 * In the first step, keep the device which has
700 * the correct fsid and the devid that is used
701 * for the dev_replace procedure.
702 * In the second step, the dev_replace state is
703 * read from the device tree and it is known
704 * whether the procedure is really active or
705 * not, which means whether this device is
706 * used or whether it should be removed.
707 */
708 if (step == 0 || device->is_tgtdev_for_dev_replace) {
709 continue;
710 }
711 }
712 if (device->bdev) {
713 blkdev_put(device->bdev, device->mode);
714 device->bdev = NULL;
715 fs_devices->open_devices--;
716 }
717 if (device->writeable) {
718 list_del_init(&device->dev_alloc_list);
719 device->writeable = 0;
720 if (!device->is_tgtdev_for_dev_replace)
721 fs_devices->rw_devices--;
722 }
723 list_del_init(&device->dev_list);
724 fs_devices->num_devices--;
725 rcu_string_free(device->name);
726 kfree(device);
727 }
728
729 if (fs_devices->seed) {
730 fs_devices = fs_devices->seed;
731 goto again;
732 }
733
734 fs_devices->latest_bdev = latest_dev->bdev;
735
736 mutex_unlock(&uuid_mutex);
737 }
738
739 static void __free_device(struct work_struct *work)
740 {
741 struct btrfs_device *device;
742
743 device = container_of(work, struct btrfs_device, rcu_work);
744
745 if (device->bdev)
746 blkdev_put(device->bdev, device->mode);
747
748 rcu_string_free(device->name);
749 kfree(device);
750 }
751
752 static void free_device(struct rcu_head *head)
753 {
754 struct btrfs_device *device;
755
756 device = container_of(head, struct btrfs_device, rcu);
757
758 INIT_WORK(&device->rcu_work, __free_device);
759 schedule_work(&device->rcu_work);
760 }
761
762 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
763 {
764 struct btrfs_device *device, *tmp;
765
766 if (--fs_devices->opened > 0)
767 return 0;
768
769 mutex_lock(&fs_devices->device_list_mutex);
770 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
771 struct btrfs_device *new_device;
772 struct rcu_string *name;
773
774 if (device->bdev)
775 fs_devices->open_devices--;
776
777 if (device->writeable &&
778 device->devid != BTRFS_DEV_REPLACE_DEVID) {
779 list_del_init(&device->dev_alloc_list);
780 fs_devices->rw_devices--;
781 }
782
783 if (device->missing)
784 fs_devices->missing_devices--;
785
786 new_device = btrfs_alloc_device(NULL, &device->devid,
787 device->uuid);
788 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
789
790 /* Safe because we are under uuid_mutex */
791 if (device->name) {
792 name = rcu_string_strdup(device->name->str, GFP_NOFS);
793 BUG_ON(!name); /* -ENOMEM */
794 rcu_assign_pointer(new_device->name, name);
795 }
796
797 list_replace_rcu(&device->dev_list, &new_device->dev_list);
798 new_device->fs_devices = device->fs_devices;
799
800 call_rcu(&device->rcu, free_device);
801 }
802 mutex_unlock(&fs_devices->device_list_mutex);
803
804 WARN_ON(fs_devices->open_devices);
805 WARN_ON(fs_devices->rw_devices);
806 fs_devices->opened = 0;
807 fs_devices->seeding = 0;
808
809 return 0;
810 }
811
812 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
813 {
814 struct btrfs_fs_devices *seed_devices = NULL;
815 int ret;
816
817 mutex_lock(&uuid_mutex);
818 ret = __btrfs_close_devices(fs_devices);
819 if (!fs_devices->opened) {
820 seed_devices = fs_devices->seed;
821 fs_devices->seed = NULL;
822 }
823 mutex_unlock(&uuid_mutex);
824
825 while (seed_devices) {
826 fs_devices = seed_devices;
827 seed_devices = fs_devices->seed;
828 __btrfs_close_devices(fs_devices);
829 free_fs_devices(fs_devices);
830 }
831 /*
832 * Wait for rcu kworkers under __btrfs_close_devices
833 * to finish all blkdev_puts so device is really
834 * free when umount is done.
835 */
836 rcu_barrier();
837 return ret;
838 }
839
840 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
841 fmode_t flags, void *holder)
842 {
843 struct request_queue *q;
844 struct block_device *bdev;
845 struct list_head *head = &fs_devices->devices;
846 struct btrfs_device *device;
847 struct btrfs_device *latest_dev = NULL;
848 struct buffer_head *bh;
849 struct btrfs_super_block *disk_super;
850 u64 devid;
851 int seeding = 1;
852 int ret = 0;
853
854 flags |= FMODE_EXCL;
855
856 list_for_each_entry(device, head, dev_list) {
857 if (device->bdev)
858 continue;
859 if (!device->name)
860 continue;
861
862 /* Just open everything we can; ignore failures here */
863 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
864 &bdev, &bh))
865 continue;
866
867 disk_super = (struct btrfs_super_block *)bh->b_data;
868 devid = btrfs_stack_device_id(&disk_super->dev_item);
869 if (devid != device->devid)
870 goto error_brelse;
871
872 if (memcmp(device->uuid, disk_super->dev_item.uuid,
873 BTRFS_UUID_SIZE))
874 goto error_brelse;
875
876 device->generation = btrfs_super_generation(disk_super);
877 if (!latest_dev ||
878 device->generation > latest_dev->generation)
879 latest_dev = device;
880
881 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
882 device->writeable = 0;
883 } else {
884 device->writeable = !bdev_read_only(bdev);
885 seeding = 0;
886 }
887
888 q = bdev_get_queue(bdev);
889 if (blk_queue_discard(q))
890 device->can_discard = 1;
891
892 device->bdev = bdev;
893 device->in_fs_metadata = 0;
894 device->mode = flags;
895
896 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
897 fs_devices->rotating = 1;
898
899 fs_devices->open_devices++;
900 if (device->writeable &&
901 device->devid != BTRFS_DEV_REPLACE_DEVID) {
902 fs_devices->rw_devices++;
903 list_add(&device->dev_alloc_list,
904 &fs_devices->alloc_list);
905 }
906 brelse(bh);
907 continue;
908
909 error_brelse:
910 brelse(bh);
911 blkdev_put(bdev, flags);
912 continue;
913 }
914 if (fs_devices->open_devices == 0) {
915 ret = -EINVAL;
916 goto out;
917 }
918 fs_devices->seeding = seeding;
919 fs_devices->opened = 1;
920 fs_devices->latest_bdev = latest_dev->bdev;
921 fs_devices->total_rw_bytes = 0;
922 out:
923 return ret;
924 }
925
926 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
927 fmode_t flags, void *holder)
928 {
929 int ret;
930
931 mutex_lock(&uuid_mutex);
932 if (fs_devices->opened) {
933 fs_devices->opened++;
934 ret = 0;
935 } else {
936 ret = __btrfs_open_devices(fs_devices, flags, holder);
937 }
938 mutex_unlock(&uuid_mutex);
939 return ret;
940 }
941
942 /*
943 * Look for a btrfs signature on a device. This may be called out of the mount path
944 * and we are not allowed to call set_blocksize during the scan. The superblock
945 * is read via pagecache
946 */
947 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
948 struct btrfs_fs_devices **fs_devices_ret)
949 {
950 struct btrfs_super_block *disk_super;
951 struct block_device *bdev;
952 struct page *page;
953 void *p;
954 int ret = -EINVAL;
955 u64 devid;
956 u64 transid;
957 u64 total_devices;
958 u64 bytenr;
959 pgoff_t index;
960
961 /*
962 * we would like to check all the supers, but that would make
963 * a btrfs mount succeed after a mkfs from a different FS.
964 * So, we need to add a special mount option to scan for
965 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
966 */
967 bytenr = btrfs_sb_offset(0);
968 flags |= FMODE_EXCL;
969 mutex_lock(&uuid_mutex);
970
971 bdev = blkdev_get_by_path(path, flags, holder);
972
973 if (IS_ERR(bdev)) {
974 ret = PTR_ERR(bdev);
975 goto error;
976 }
977
978 /* make sure our super fits in the device */
979 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
980 goto error_bdev_put;
981
982 /* make sure our super fits in the page */
983 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
984 goto error_bdev_put;
985
986 /* make sure our super doesn't straddle pages on disk */
987 index = bytenr >> PAGE_CACHE_SHIFT;
988 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
989 goto error_bdev_put;
990
991 /* pull in the page with our super */
992 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
993 index, GFP_NOFS);
994
995 if (IS_ERR_OR_NULL(page))
996 goto error_bdev_put;
997
998 p = kmap(page);
999
1000 /* align our pointer to the offset of the super block */
1001 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1002
1003 if (btrfs_super_bytenr(disk_super) != bytenr ||
1004 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1005 goto error_unmap;
1006
1007 devid = btrfs_stack_device_id(&disk_super->dev_item);
1008 transid = btrfs_super_generation(disk_super);
1009 total_devices = btrfs_super_num_devices(disk_super);
1010
1011 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1012 if (ret > 0) {
1013 if (disk_super->label[0]) {
1014 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1015 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1016 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1017 } else {
1018 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1019 }
1020
1021 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1022 ret = 0;
1023 }
1024 if (!ret && fs_devices_ret)
1025 (*fs_devices_ret)->total_devices = total_devices;
1026
1027 error_unmap:
1028 kunmap(page);
1029 page_cache_release(page);
1030
1031 error_bdev_put:
1032 blkdev_put(bdev, flags);
1033 error:
1034 mutex_unlock(&uuid_mutex);
1035 return ret;
1036 }
1037
1038 /* helper to account the used device space in the range */
1039 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1040 u64 end, u64 *length)
1041 {
1042 struct btrfs_key key;
1043 struct btrfs_root *root = device->dev_root;
1044 struct btrfs_dev_extent *dev_extent;
1045 struct btrfs_path *path;
1046 u64 extent_end;
1047 int ret;
1048 int slot;
1049 struct extent_buffer *l;
1050
1051 *length = 0;
1052
1053 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1054 return 0;
1055
1056 path = btrfs_alloc_path();
1057 if (!path)
1058 return -ENOMEM;
1059 path->reada = 2;
1060
1061 key.objectid = device->devid;
1062 key.offset = start;
1063 key.type = BTRFS_DEV_EXTENT_KEY;
1064
1065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1066 if (ret < 0)
1067 goto out;
1068 if (ret > 0) {
1069 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1070 if (ret < 0)
1071 goto out;
1072 }
1073
1074 while (1) {
1075 l = path->nodes[0];
1076 slot = path->slots[0];
1077 if (slot >= btrfs_header_nritems(l)) {
1078 ret = btrfs_next_leaf(root, path);
1079 if (ret == 0)
1080 continue;
1081 if (ret < 0)
1082 goto out;
1083
1084 break;
1085 }
1086 btrfs_item_key_to_cpu(l, &key, slot);
1087
1088 if (key.objectid < device->devid)
1089 goto next;
1090
1091 if (key.objectid > device->devid)
1092 break;
1093
1094 if (key.type != BTRFS_DEV_EXTENT_KEY)
1095 goto next;
1096
1097 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1098 extent_end = key.offset + btrfs_dev_extent_length(l,
1099 dev_extent);
1100 if (key.offset <= start && extent_end > end) {
1101 *length = end - start + 1;
1102 break;
1103 } else if (key.offset <= start && extent_end > start)
1104 *length += extent_end - start;
1105 else if (key.offset > start && extent_end <= end)
1106 *length += extent_end - key.offset;
1107 else if (key.offset > start && key.offset <= end) {
1108 *length += end - key.offset + 1;
1109 break;
1110 } else if (key.offset > end)
1111 break;
1112
1113 next:
1114 path->slots[0]++;
1115 }
1116 ret = 0;
1117 out:
1118 btrfs_free_path(path);
1119 return ret;
1120 }
1121
1122 static int contains_pending_extent(struct btrfs_transaction *transaction,
1123 struct btrfs_device *device,
1124 u64 *start, u64 len)
1125 {
1126 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1127 struct extent_map *em;
1128 struct list_head *search_list = &fs_info->pinned_chunks;
1129 int ret = 0;
1130 u64 physical_start = *start;
1131
1132 if (transaction)
1133 search_list = &transaction->pending_chunks;
1134 again:
1135 list_for_each_entry(em, search_list, list) {
1136 struct map_lookup *map;
1137 int i;
1138
1139 map = (struct map_lookup *)em->bdev;
1140 for (i = 0; i < map->num_stripes; i++) {
1141 u64 end;
1142
1143 if (map->stripes[i].dev != device)
1144 continue;
1145 if (map->stripes[i].physical >= physical_start + len ||
1146 map->stripes[i].physical + em->orig_block_len <=
1147 physical_start)
1148 continue;
1149 /*
1150 * Make sure that while processing the pinned list we do
1151 * not override our *start with a lower value, because
1152 * we can have pinned chunks that fall within this
1153 * device hole and that have lower physical addresses
1154 * than the pending chunks we processed before. If we
1155 * do not take this special care we can end up getting
1156 * 2 pending chunks that start at the same physical
1157 * device offsets because the end offset of a pinned
1158 * chunk can be equal to the start offset of some
1159 * pending chunk.
1160 */
1161 end = map->stripes[i].physical + em->orig_block_len;
1162 if (end > *start) {
1163 *start = end;
1164 ret = 1;
1165 }
1166 }
1167 }
1168 if (search_list != &fs_info->pinned_chunks) {
1169 search_list = &fs_info->pinned_chunks;
1170 goto again;
1171 }
1172
1173 return ret;
1174 }
1175
1176
1177 /*
1178 * find_free_dev_extent_start - find free space in the specified device
1179 * @device: the device which we search the free space in
1180 * @num_bytes: the size of the free space that we need
1181 * @search_start: the position from which to begin the search
1182 * @start: store the start of the free space.
1183 * @len: the size of the free space. that we find, or the size
1184 * of the max free space if we don't find suitable free space
1185 *
1186 * this uses a pretty simple search, the expectation is that it is
1187 * called very infrequently and that a given device has a small number
1188 * of extents
1189 *
1190 * @start is used to store the start of the free space if we find. But if we
1191 * don't find suitable free space, it will be used to store the start position
1192 * of the max free space.
1193 *
1194 * @len is used to store the size of the free space that we find.
1195 * But if we don't find suitable free space, it is used to store the size of
1196 * the max free space.
1197 */
1198 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1199 struct btrfs_device *device, u64 num_bytes,
1200 u64 search_start, u64 *start, u64 *len)
1201 {
1202 struct btrfs_key key;
1203 struct btrfs_root *root = device->dev_root;
1204 struct btrfs_dev_extent *dev_extent;
1205 struct btrfs_path *path;
1206 u64 hole_size;
1207 u64 max_hole_start;
1208 u64 max_hole_size;
1209 u64 extent_end;
1210 u64 search_end = device->total_bytes;
1211 int ret;
1212 int slot;
1213 struct extent_buffer *l;
1214
1215 path = btrfs_alloc_path();
1216 if (!path)
1217 return -ENOMEM;
1218
1219 max_hole_start = search_start;
1220 max_hole_size = 0;
1221
1222 again:
1223 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1224 ret = -ENOSPC;
1225 goto out;
1226 }
1227
1228 path->reada = 2;
1229 path->search_commit_root = 1;
1230 path->skip_locking = 1;
1231
1232 key.objectid = device->devid;
1233 key.offset = search_start;
1234 key.type = BTRFS_DEV_EXTENT_KEY;
1235
1236 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1237 if (ret < 0)
1238 goto out;
1239 if (ret > 0) {
1240 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1241 if (ret < 0)
1242 goto out;
1243 }
1244
1245 while (1) {
1246 l = path->nodes[0];
1247 slot = path->slots[0];
1248 if (slot >= btrfs_header_nritems(l)) {
1249 ret = btrfs_next_leaf(root, path);
1250 if (ret == 0)
1251 continue;
1252 if (ret < 0)
1253 goto out;
1254
1255 break;
1256 }
1257 btrfs_item_key_to_cpu(l, &key, slot);
1258
1259 if (key.objectid < device->devid)
1260 goto next;
1261
1262 if (key.objectid > device->devid)
1263 break;
1264
1265 if (key.type != BTRFS_DEV_EXTENT_KEY)
1266 goto next;
1267
1268 if (key.offset > search_start) {
1269 hole_size = key.offset - search_start;
1270
1271 /*
1272 * Have to check before we set max_hole_start, otherwise
1273 * we could end up sending back this offset anyway.
1274 */
1275 if (contains_pending_extent(transaction, device,
1276 &search_start,
1277 hole_size)) {
1278 if (key.offset >= search_start) {
1279 hole_size = key.offset - search_start;
1280 } else {
1281 WARN_ON_ONCE(1);
1282 hole_size = 0;
1283 }
1284 }
1285
1286 if (hole_size > max_hole_size) {
1287 max_hole_start = search_start;
1288 max_hole_size = hole_size;
1289 }
1290
1291 /*
1292 * If this free space is greater than which we need,
1293 * it must be the max free space that we have found
1294 * until now, so max_hole_start must point to the start
1295 * of this free space and the length of this free space
1296 * is stored in max_hole_size. Thus, we return
1297 * max_hole_start and max_hole_size and go back to the
1298 * caller.
1299 */
1300 if (hole_size >= num_bytes) {
1301 ret = 0;
1302 goto out;
1303 }
1304 }
1305
1306 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1307 extent_end = key.offset + btrfs_dev_extent_length(l,
1308 dev_extent);
1309 if (extent_end > search_start)
1310 search_start = extent_end;
1311 next:
1312 path->slots[0]++;
1313 cond_resched();
1314 }
1315
1316 /*
1317 * At this point, search_start should be the end of
1318 * allocated dev extents, and when shrinking the device,
1319 * search_end may be smaller than search_start.
1320 */
1321 if (search_end > search_start) {
1322 hole_size = search_end - search_start;
1323
1324 if (contains_pending_extent(transaction, device, &search_start,
1325 hole_size)) {
1326 btrfs_release_path(path);
1327 goto again;
1328 }
1329
1330 if (hole_size > max_hole_size) {
1331 max_hole_start = search_start;
1332 max_hole_size = hole_size;
1333 }
1334 }
1335
1336 /* See above. */
1337 if (max_hole_size < num_bytes)
1338 ret = -ENOSPC;
1339 else
1340 ret = 0;
1341
1342 out:
1343 btrfs_free_path(path);
1344 *start = max_hole_start;
1345 if (len)
1346 *len = max_hole_size;
1347 return ret;
1348 }
1349
1350 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1351 struct btrfs_device *device, u64 num_bytes,
1352 u64 *start, u64 *len)
1353 {
1354 struct btrfs_root *root = device->dev_root;
1355 u64 search_start;
1356
1357 /* FIXME use last free of some kind */
1358
1359 /*
1360 * we don't want to overwrite the superblock on the drive,
1361 * so we make sure to start at an offset of at least 1MB
1362 */
1363 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1364 return find_free_dev_extent_start(trans->transaction, device,
1365 num_bytes, search_start, start, len);
1366 }
1367
1368 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1369 struct btrfs_device *device,
1370 u64 start, u64 *dev_extent_len)
1371 {
1372 int ret;
1373 struct btrfs_path *path;
1374 struct btrfs_root *root = device->dev_root;
1375 struct btrfs_key key;
1376 struct btrfs_key found_key;
1377 struct extent_buffer *leaf = NULL;
1378 struct btrfs_dev_extent *extent = NULL;
1379
1380 path = btrfs_alloc_path();
1381 if (!path)
1382 return -ENOMEM;
1383
1384 key.objectid = device->devid;
1385 key.offset = start;
1386 key.type = BTRFS_DEV_EXTENT_KEY;
1387 again:
1388 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1389 if (ret > 0) {
1390 ret = btrfs_previous_item(root, path, key.objectid,
1391 BTRFS_DEV_EXTENT_KEY);
1392 if (ret)
1393 goto out;
1394 leaf = path->nodes[0];
1395 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1396 extent = btrfs_item_ptr(leaf, path->slots[0],
1397 struct btrfs_dev_extent);
1398 BUG_ON(found_key.offset > start || found_key.offset +
1399 btrfs_dev_extent_length(leaf, extent) < start);
1400 key = found_key;
1401 btrfs_release_path(path);
1402 goto again;
1403 } else if (ret == 0) {
1404 leaf = path->nodes[0];
1405 extent = btrfs_item_ptr(leaf, path->slots[0],
1406 struct btrfs_dev_extent);
1407 } else {
1408 btrfs_error(root->fs_info, ret, "Slot search failed");
1409 goto out;
1410 }
1411
1412 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1413
1414 ret = btrfs_del_item(trans, root, path);
1415 if (ret) {
1416 btrfs_error(root->fs_info, ret,
1417 "Failed to remove dev extent item");
1418 } else {
1419 trans->transaction->have_free_bgs = 1;
1420 }
1421 out:
1422 btrfs_free_path(path);
1423 return ret;
1424 }
1425
1426 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1427 struct btrfs_device *device,
1428 u64 chunk_tree, u64 chunk_objectid,
1429 u64 chunk_offset, u64 start, u64 num_bytes)
1430 {
1431 int ret;
1432 struct btrfs_path *path;
1433 struct btrfs_root *root = device->dev_root;
1434 struct btrfs_dev_extent *extent;
1435 struct extent_buffer *leaf;
1436 struct btrfs_key key;
1437
1438 WARN_ON(!device->in_fs_metadata);
1439 WARN_ON(device->is_tgtdev_for_dev_replace);
1440 path = btrfs_alloc_path();
1441 if (!path)
1442 return -ENOMEM;
1443
1444 key.objectid = device->devid;
1445 key.offset = start;
1446 key.type = BTRFS_DEV_EXTENT_KEY;
1447 ret = btrfs_insert_empty_item(trans, root, path, &key,
1448 sizeof(*extent));
1449 if (ret)
1450 goto out;
1451
1452 leaf = path->nodes[0];
1453 extent = btrfs_item_ptr(leaf, path->slots[0],
1454 struct btrfs_dev_extent);
1455 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1456 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1457 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1458
1459 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1460 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1461
1462 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1463 btrfs_mark_buffer_dirty(leaf);
1464 out:
1465 btrfs_free_path(path);
1466 return ret;
1467 }
1468
1469 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1470 {
1471 struct extent_map_tree *em_tree;
1472 struct extent_map *em;
1473 struct rb_node *n;
1474 u64 ret = 0;
1475
1476 em_tree = &fs_info->mapping_tree.map_tree;
1477 read_lock(&em_tree->lock);
1478 n = rb_last(&em_tree->map);
1479 if (n) {
1480 em = rb_entry(n, struct extent_map, rb_node);
1481 ret = em->start + em->len;
1482 }
1483 read_unlock(&em_tree->lock);
1484
1485 return ret;
1486 }
1487
1488 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1489 u64 *devid_ret)
1490 {
1491 int ret;
1492 struct btrfs_key key;
1493 struct btrfs_key found_key;
1494 struct btrfs_path *path;
1495
1496 path = btrfs_alloc_path();
1497 if (!path)
1498 return -ENOMEM;
1499
1500 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1501 key.type = BTRFS_DEV_ITEM_KEY;
1502 key.offset = (u64)-1;
1503
1504 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1505 if (ret < 0)
1506 goto error;
1507
1508 BUG_ON(ret == 0); /* Corruption */
1509
1510 ret = btrfs_previous_item(fs_info->chunk_root, path,
1511 BTRFS_DEV_ITEMS_OBJECTID,
1512 BTRFS_DEV_ITEM_KEY);
1513 if (ret) {
1514 *devid_ret = 1;
1515 } else {
1516 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1517 path->slots[0]);
1518 *devid_ret = found_key.offset + 1;
1519 }
1520 ret = 0;
1521 error:
1522 btrfs_free_path(path);
1523 return ret;
1524 }
1525
1526 /*
1527 * the device information is stored in the chunk root
1528 * the btrfs_device struct should be fully filled in
1529 */
1530 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1531 struct btrfs_root *root,
1532 struct btrfs_device *device)
1533 {
1534 int ret;
1535 struct btrfs_path *path;
1536 struct btrfs_dev_item *dev_item;
1537 struct extent_buffer *leaf;
1538 struct btrfs_key key;
1539 unsigned long ptr;
1540
1541 root = root->fs_info->chunk_root;
1542
1543 path = btrfs_alloc_path();
1544 if (!path)
1545 return -ENOMEM;
1546
1547 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1548 key.type = BTRFS_DEV_ITEM_KEY;
1549 key.offset = device->devid;
1550
1551 ret = btrfs_insert_empty_item(trans, root, path, &key,
1552 sizeof(*dev_item));
1553 if (ret)
1554 goto out;
1555
1556 leaf = path->nodes[0];
1557 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1558
1559 btrfs_set_device_id(leaf, dev_item, device->devid);
1560 btrfs_set_device_generation(leaf, dev_item, 0);
1561 btrfs_set_device_type(leaf, dev_item, device->type);
1562 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1563 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1564 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1565 btrfs_set_device_total_bytes(leaf, dev_item,
1566 btrfs_device_get_disk_total_bytes(device));
1567 btrfs_set_device_bytes_used(leaf, dev_item,
1568 btrfs_device_get_bytes_used(device));
1569 btrfs_set_device_group(leaf, dev_item, 0);
1570 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1571 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1572 btrfs_set_device_start_offset(leaf, dev_item, 0);
1573
1574 ptr = btrfs_device_uuid(dev_item);
1575 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1576 ptr = btrfs_device_fsid(dev_item);
1577 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1578 btrfs_mark_buffer_dirty(leaf);
1579
1580 ret = 0;
1581 out:
1582 btrfs_free_path(path);
1583 return ret;
1584 }
1585
1586 /*
1587 * Function to update ctime/mtime for a given device path.
1588 * Mainly used for ctime/mtime based probe like libblkid.
1589 */
1590 static void update_dev_time(char *path_name)
1591 {
1592 struct file *filp;
1593
1594 filp = filp_open(path_name, O_RDWR, 0);
1595 if (IS_ERR(filp))
1596 return;
1597 file_update_time(filp);
1598 filp_close(filp, NULL);
1599 return;
1600 }
1601
1602 static int btrfs_rm_dev_item(struct btrfs_root *root,
1603 struct btrfs_device *device)
1604 {
1605 int ret;
1606 struct btrfs_path *path;
1607 struct btrfs_key key;
1608 struct btrfs_trans_handle *trans;
1609
1610 root = root->fs_info->chunk_root;
1611
1612 path = btrfs_alloc_path();
1613 if (!path)
1614 return -ENOMEM;
1615
1616 trans = btrfs_start_transaction(root, 0);
1617 if (IS_ERR(trans)) {
1618 btrfs_free_path(path);
1619 return PTR_ERR(trans);
1620 }
1621 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1622 key.type = BTRFS_DEV_ITEM_KEY;
1623 key.offset = device->devid;
1624
1625 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1626 if (ret < 0)
1627 goto out;
1628
1629 if (ret > 0) {
1630 ret = -ENOENT;
1631 goto out;
1632 }
1633
1634 ret = btrfs_del_item(trans, root, path);
1635 if (ret)
1636 goto out;
1637 out:
1638 btrfs_free_path(path);
1639 btrfs_commit_transaction(trans, root);
1640 return ret;
1641 }
1642
1643 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1644 {
1645 struct btrfs_device *device;
1646 struct btrfs_device *next_device;
1647 struct block_device *bdev;
1648 struct buffer_head *bh = NULL;
1649 struct btrfs_super_block *disk_super;
1650 struct btrfs_fs_devices *cur_devices;
1651 u64 all_avail;
1652 u64 devid;
1653 u64 num_devices;
1654 u8 *dev_uuid;
1655 unsigned seq;
1656 int ret = 0;
1657 bool clear_super = false;
1658
1659 mutex_lock(&uuid_mutex);
1660
1661 do {
1662 seq = read_seqbegin(&root->fs_info->profiles_lock);
1663
1664 all_avail = root->fs_info->avail_data_alloc_bits |
1665 root->fs_info->avail_system_alloc_bits |
1666 root->fs_info->avail_metadata_alloc_bits;
1667 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1668
1669 num_devices = root->fs_info->fs_devices->num_devices;
1670 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1671 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1672 WARN_ON(num_devices < 1);
1673 num_devices--;
1674 }
1675 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1676
1677 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1678 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1679 goto out;
1680 }
1681
1682 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1683 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1684 goto out;
1685 }
1686
1687 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1688 root->fs_info->fs_devices->rw_devices <= 2) {
1689 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1690 goto out;
1691 }
1692 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1693 root->fs_info->fs_devices->rw_devices <= 3) {
1694 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1695 goto out;
1696 }
1697
1698 if (strcmp(device_path, "missing") == 0) {
1699 struct list_head *devices;
1700 struct btrfs_device *tmp;
1701
1702 device = NULL;
1703 devices = &root->fs_info->fs_devices->devices;
1704 /*
1705 * It is safe to read the devices since the volume_mutex
1706 * is held.
1707 */
1708 list_for_each_entry(tmp, devices, dev_list) {
1709 if (tmp->in_fs_metadata &&
1710 !tmp->is_tgtdev_for_dev_replace &&
1711 !tmp->bdev) {
1712 device = tmp;
1713 break;
1714 }
1715 }
1716 bdev = NULL;
1717 bh = NULL;
1718 disk_super = NULL;
1719 if (!device) {
1720 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1721 goto out;
1722 }
1723 } else {
1724 ret = btrfs_get_bdev_and_sb(device_path,
1725 FMODE_WRITE | FMODE_EXCL,
1726 root->fs_info->bdev_holder, 0,
1727 &bdev, &bh);
1728 if (ret)
1729 goto out;
1730 disk_super = (struct btrfs_super_block *)bh->b_data;
1731 devid = btrfs_stack_device_id(&disk_super->dev_item);
1732 dev_uuid = disk_super->dev_item.uuid;
1733 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1734 disk_super->fsid);
1735 if (!device) {
1736 ret = -ENOENT;
1737 goto error_brelse;
1738 }
1739 }
1740
1741 if (device->is_tgtdev_for_dev_replace) {
1742 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1743 goto error_brelse;
1744 }
1745
1746 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1747 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1748 goto error_brelse;
1749 }
1750
1751 if (device->writeable) {
1752 lock_chunks(root);
1753 list_del_init(&device->dev_alloc_list);
1754 device->fs_devices->rw_devices--;
1755 unlock_chunks(root);
1756 clear_super = true;
1757 }
1758
1759 mutex_unlock(&uuid_mutex);
1760 ret = btrfs_shrink_device(device, 0);
1761 mutex_lock(&uuid_mutex);
1762 if (ret)
1763 goto error_undo;
1764
1765 /*
1766 * TODO: the superblock still includes this device in its num_devices
1767 * counter although write_all_supers() is not locked out. This
1768 * could give a filesystem state which requires a degraded mount.
1769 */
1770 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1771 if (ret)
1772 goto error_undo;
1773
1774 device->in_fs_metadata = 0;
1775 btrfs_scrub_cancel_dev(root->fs_info, device);
1776
1777 /*
1778 * the device list mutex makes sure that we don't change
1779 * the device list while someone else is writing out all
1780 * the device supers. Whoever is writing all supers, should
1781 * lock the device list mutex before getting the number of
1782 * devices in the super block (super_copy). Conversely,
1783 * whoever updates the number of devices in the super block
1784 * (super_copy) should hold the device list mutex.
1785 */
1786
1787 cur_devices = device->fs_devices;
1788 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1789 list_del_rcu(&device->dev_list);
1790
1791 device->fs_devices->num_devices--;
1792 device->fs_devices->total_devices--;
1793
1794 if (device->missing)
1795 device->fs_devices->missing_devices--;
1796
1797 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1798 struct btrfs_device, dev_list);
1799 if (device->bdev == root->fs_info->sb->s_bdev)
1800 root->fs_info->sb->s_bdev = next_device->bdev;
1801 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1802 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1803
1804 if (device->bdev) {
1805 device->fs_devices->open_devices--;
1806 /* remove sysfs entry */
1807 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
1808 }
1809
1810 call_rcu(&device->rcu, free_device);
1811
1812 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1813 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1814 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1815
1816 if (cur_devices->open_devices == 0) {
1817 struct btrfs_fs_devices *fs_devices;
1818 fs_devices = root->fs_info->fs_devices;
1819 while (fs_devices) {
1820 if (fs_devices->seed == cur_devices) {
1821 fs_devices->seed = cur_devices->seed;
1822 break;
1823 }
1824 fs_devices = fs_devices->seed;
1825 }
1826 cur_devices->seed = NULL;
1827 __btrfs_close_devices(cur_devices);
1828 free_fs_devices(cur_devices);
1829 }
1830
1831 root->fs_info->num_tolerated_disk_barrier_failures =
1832 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1833
1834 /*
1835 * at this point, the device is zero sized. We want to
1836 * remove it from the devices list and zero out the old super
1837 */
1838 if (clear_super && disk_super) {
1839 u64 bytenr;
1840 int i;
1841
1842 /* make sure this device isn't detected as part of
1843 * the FS anymore
1844 */
1845 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1846 set_buffer_dirty(bh);
1847 sync_dirty_buffer(bh);
1848
1849 /* clear the mirror copies of super block on the disk
1850 * being removed, 0th copy is been taken care above and
1851 * the below would take of the rest
1852 */
1853 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1854 bytenr = btrfs_sb_offset(i);
1855 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1856 i_size_read(bdev->bd_inode))
1857 break;
1858
1859 brelse(bh);
1860 bh = __bread(bdev, bytenr / 4096,
1861 BTRFS_SUPER_INFO_SIZE);
1862 if (!bh)
1863 continue;
1864
1865 disk_super = (struct btrfs_super_block *)bh->b_data;
1866
1867 if (btrfs_super_bytenr(disk_super) != bytenr ||
1868 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1869 continue;
1870 }
1871 memset(&disk_super->magic, 0,
1872 sizeof(disk_super->magic));
1873 set_buffer_dirty(bh);
1874 sync_dirty_buffer(bh);
1875 }
1876 }
1877
1878 ret = 0;
1879
1880 if (bdev) {
1881 /* Notify udev that device has changed */
1882 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1883
1884 /* Update ctime/mtime for device path for libblkid */
1885 update_dev_time(device_path);
1886 }
1887
1888 error_brelse:
1889 brelse(bh);
1890 if (bdev)
1891 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1892 out:
1893 mutex_unlock(&uuid_mutex);
1894 return ret;
1895 error_undo:
1896 if (device->writeable) {
1897 lock_chunks(root);
1898 list_add(&device->dev_alloc_list,
1899 &root->fs_info->fs_devices->alloc_list);
1900 device->fs_devices->rw_devices++;
1901 unlock_chunks(root);
1902 }
1903 goto error_brelse;
1904 }
1905
1906 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1907 struct btrfs_device *srcdev)
1908 {
1909 struct btrfs_fs_devices *fs_devices;
1910
1911 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1912
1913 /*
1914 * in case of fs with no seed, srcdev->fs_devices will point
1915 * to fs_devices of fs_info. However when the dev being replaced is
1916 * a seed dev it will point to the seed's local fs_devices. In short
1917 * srcdev will have its correct fs_devices in both the cases.
1918 */
1919 fs_devices = srcdev->fs_devices;
1920
1921 list_del_rcu(&srcdev->dev_list);
1922 list_del_rcu(&srcdev->dev_alloc_list);
1923 fs_devices->num_devices--;
1924 if (srcdev->missing)
1925 fs_devices->missing_devices--;
1926
1927 if (srcdev->writeable) {
1928 fs_devices->rw_devices--;
1929 /* zero out the old super if it is writable */
1930 btrfs_scratch_superblock(srcdev);
1931 }
1932
1933 if (srcdev->bdev)
1934 fs_devices->open_devices--;
1935 }
1936
1937 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1938 struct btrfs_device *srcdev)
1939 {
1940 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1941
1942 call_rcu(&srcdev->rcu, free_device);
1943
1944 /*
1945 * unless fs_devices is seed fs, num_devices shouldn't go
1946 * zero
1947 */
1948 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1949
1950 /* if this is no devs we rather delete the fs_devices */
1951 if (!fs_devices->num_devices) {
1952 struct btrfs_fs_devices *tmp_fs_devices;
1953
1954 tmp_fs_devices = fs_info->fs_devices;
1955 while (tmp_fs_devices) {
1956 if (tmp_fs_devices->seed == fs_devices) {
1957 tmp_fs_devices->seed = fs_devices->seed;
1958 break;
1959 }
1960 tmp_fs_devices = tmp_fs_devices->seed;
1961 }
1962 fs_devices->seed = NULL;
1963 __btrfs_close_devices(fs_devices);
1964 free_fs_devices(fs_devices);
1965 }
1966 }
1967
1968 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1969 struct btrfs_device *tgtdev)
1970 {
1971 struct btrfs_device *next_device;
1972
1973 mutex_lock(&uuid_mutex);
1974 WARN_ON(!tgtdev);
1975 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1976
1977 btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev);
1978
1979 if (tgtdev->bdev) {
1980 btrfs_scratch_superblock(tgtdev);
1981 fs_info->fs_devices->open_devices--;
1982 }
1983 fs_info->fs_devices->num_devices--;
1984
1985 next_device = list_entry(fs_info->fs_devices->devices.next,
1986 struct btrfs_device, dev_list);
1987 if (tgtdev->bdev == fs_info->sb->s_bdev)
1988 fs_info->sb->s_bdev = next_device->bdev;
1989 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1990 fs_info->fs_devices->latest_bdev = next_device->bdev;
1991 list_del_rcu(&tgtdev->dev_list);
1992
1993 call_rcu(&tgtdev->rcu, free_device);
1994
1995 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1996 mutex_unlock(&uuid_mutex);
1997 }
1998
1999 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2000 struct btrfs_device **device)
2001 {
2002 int ret = 0;
2003 struct btrfs_super_block *disk_super;
2004 u64 devid;
2005 u8 *dev_uuid;
2006 struct block_device *bdev;
2007 struct buffer_head *bh;
2008
2009 *device = NULL;
2010 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2011 root->fs_info->bdev_holder, 0, &bdev, &bh);
2012 if (ret)
2013 return ret;
2014 disk_super = (struct btrfs_super_block *)bh->b_data;
2015 devid = btrfs_stack_device_id(&disk_super->dev_item);
2016 dev_uuid = disk_super->dev_item.uuid;
2017 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2018 disk_super->fsid);
2019 brelse(bh);
2020 if (!*device)
2021 ret = -ENOENT;
2022 blkdev_put(bdev, FMODE_READ);
2023 return ret;
2024 }
2025
2026 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2027 char *device_path,
2028 struct btrfs_device **device)
2029 {
2030 *device = NULL;
2031 if (strcmp(device_path, "missing") == 0) {
2032 struct list_head *devices;
2033 struct btrfs_device *tmp;
2034
2035 devices = &root->fs_info->fs_devices->devices;
2036 /*
2037 * It is safe to read the devices since the volume_mutex
2038 * is held by the caller.
2039 */
2040 list_for_each_entry(tmp, devices, dev_list) {
2041 if (tmp->in_fs_metadata && !tmp->bdev) {
2042 *device = tmp;
2043 break;
2044 }
2045 }
2046
2047 if (!*device) {
2048 btrfs_err(root->fs_info, "no missing device found");
2049 return -ENOENT;
2050 }
2051
2052 return 0;
2053 } else {
2054 return btrfs_find_device_by_path(root, device_path, device);
2055 }
2056 }
2057
2058 /*
2059 * does all the dirty work required for changing file system's UUID.
2060 */
2061 static int btrfs_prepare_sprout(struct btrfs_root *root)
2062 {
2063 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2064 struct btrfs_fs_devices *old_devices;
2065 struct btrfs_fs_devices *seed_devices;
2066 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2067 struct btrfs_device *device;
2068 u64 super_flags;
2069
2070 BUG_ON(!mutex_is_locked(&uuid_mutex));
2071 if (!fs_devices->seeding)
2072 return -EINVAL;
2073
2074 seed_devices = __alloc_fs_devices();
2075 if (IS_ERR(seed_devices))
2076 return PTR_ERR(seed_devices);
2077
2078 old_devices = clone_fs_devices(fs_devices);
2079 if (IS_ERR(old_devices)) {
2080 kfree(seed_devices);
2081 return PTR_ERR(old_devices);
2082 }
2083
2084 list_add(&old_devices->list, &fs_uuids);
2085
2086 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2087 seed_devices->opened = 1;
2088 INIT_LIST_HEAD(&seed_devices->devices);
2089 INIT_LIST_HEAD(&seed_devices->alloc_list);
2090 mutex_init(&seed_devices->device_list_mutex);
2091
2092 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2093 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2094 synchronize_rcu);
2095 list_for_each_entry(device, &seed_devices->devices, dev_list)
2096 device->fs_devices = seed_devices;
2097
2098 lock_chunks(root);
2099 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2100 unlock_chunks(root);
2101
2102 fs_devices->seeding = 0;
2103 fs_devices->num_devices = 0;
2104 fs_devices->open_devices = 0;
2105 fs_devices->missing_devices = 0;
2106 fs_devices->rotating = 0;
2107 fs_devices->seed = seed_devices;
2108
2109 generate_random_uuid(fs_devices->fsid);
2110 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2111 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2112 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2113
2114 super_flags = btrfs_super_flags(disk_super) &
2115 ~BTRFS_SUPER_FLAG_SEEDING;
2116 btrfs_set_super_flags(disk_super, super_flags);
2117
2118 return 0;
2119 }
2120
2121 /*
2122 * strore the expected generation for seed devices in device items.
2123 */
2124 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2125 struct btrfs_root *root)
2126 {
2127 struct btrfs_path *path;
2128 struct extent_buffer *leaf;
2129 struct btrfs_dev_item *dev_item;
2130 struct btrfs_device *device;
2131 struct btrfs_key key;
2132 u8 fs_uuid[BTRFS_UUID_SIZE];
2133 u8 dev_uuid[BTRFS_UUID_SIZE];
2134 u64 devid;
2135 int ret;
2136
2137 path = btrfs_alloc_path();
2138 if (!path)
2139 return -ENOMEM;
2140
2141 root = root->fs_info->chunk_root;
2142 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2143 key.offset = 0;
2144 key.type = BTRFS_DEV_ITEM_KEY;
2145
2146 while (1) {
2147 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2148 if (ret < 0)
2149 goto error;
2150
2151 leaf = path->nodes[0];
2152 next_slot:
2153 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2154 ret = btrfs_next_leaf(root, path);
2155 if (ret > 0)
2156 break;
2157 if (ret < 0)
2158 goto error;
2159 leaf = path->nodes[0];
2160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2161 btrfs_release_path(path);
2162 continue;
2163 }
2164
2165 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2166 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2167 key.type != BTRFS_DEV_ITEM_KEY)
2168 break;
2169
2170 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2171 struct btrfs_dev_item);
2172 devid = btrfs_device_id(leaf, dev_item);
2173 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2174 BTRFS_UUID_SIZE);
2175 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2176 BTRFS_UUID_SIZE);
2177 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2178 fs_uuid);
2179 BUG_ON(!device); /* Logic error */
2180
2181 if (device->fs_devices->seeding) {
2182 btrfs_set_device_generation(leaf, dev_item,
2183 device->generation);
2184 btrfs_mark_buffer_dirty(leaf);
2185 }
2186
2187 path->slots[0]++;
2188 goto next_slot;
2189 }
2190 ret = 0;
2191 error:
2192 btrfs_free_path(path);
2193 return ret;
2194 }
2195
2196 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2197 {
2198 struct request_queue *q;
2199 struct btrfs_trans_handle *trans;
2200 struct btrfs_device *device;
2201 struct block_device *bdev;
2202 struct list_head *devices;
2203 struct super_block *sb = root->fs_info->sb;
2204 struct rcu_string *name;
2205 u64 tmp;
2206 int seeding_dev = 0;
2207 int ret = 0;
2208
2209 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2210 return -EROFS;
2211
2212 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2213 root->fs_info->bdev_holder);
2214 if (IS_ERR(bdev))
2215 return PTR_ERR(bdev);
2216
2217 if (root->fs_info->fs_devices->seeding) {
2218 seeding_dev = 1;
2219 down_write(&sb->s_umount);
2220 mutex_lock(&uuid_mutex);
2221 }
2222
2223 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2224
2225 devices = &root->fs_info->fs_devices->devices;
2226
2227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2228 list_for_each_entry(device, devices, dev_list) {
2229 if (device->bdev == bdev) {
2230 ret = -EEXIST;
2231 mutex_unlock(
2232 &root->fs_info->fs_devices->device_list_mutex);
2233 goto error;
2234 }
2235 }
2236 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2237
2238 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2239 if (IS_ERR(device)) {
2240 /* we can safely leave the fs_devices entry around */
2241 ret = PTR_ERR(device);
2242 goto error;
2243 }
2244
2245 name = rcu_string_strdup(device_path, GFP_NOFS);
2246 if (!name) {
2247 kfree(device);
2248 ret = -ENOMEM;
2249 goto error;
2250 }
2251 rcu_assign_pointer(device->name, name);
2252
2253 trans = btrfs_start_transaction(root, 0);
2254 if (IS_ERR(trans)) {
2255 rcu_string_free(device->name);
2256 kfree(device);
2257 ret = PTR_ERR(trans);
2258 goto error;
2259 }
2260
2261 q = bdev_get_queue(bdev);
2262 if (blk_queue_discard(q))
2263 device->can_discard = 1;
2264 device->writeable = 1;
2265 device->generation = trans->transid;
2266 device->io_width = root->sectorsize;
2267 device->io_align = root->sectorsize;
2268 device->sector_size = root->sectorsize;
2269 device->total_bytes = i_size_read(bdev->bd_inode);
2270 device->disk_total_bytes = device->total_bytes;
2271 device->commit_total_bytes = device->total_bytes;
2272 device->dev_root = root->fs_info->dev_root;
2273 device->bdev = bdev;
2274 device->in_fs_metadata = 1;
2275 device->is_tgtdev_for_dev_replace = 0;
2276 device->mode = FMODE_EXCL;
2277 device->dev_stats_valid = 1;
2278 set_blocksize(device->bdev, 4096);
2279
2280 if (seeding_dev) {
2281 sb->s_flags &= ~MS_RDONLY;
2282 ret = btrfs_prepare_sprout(root);
2283 BUG_ON(ret); /* -ENOMEM */
2284 }
2285
2286 device->fs_devices = root->fs_info->fs_devices;
2287
2288 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2289 lock_chunks(root);
2290 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2291 list_add(&device->dev_alloc_list,
2292 &root->fs_info->fs_devices->alloc_list);
2293 root->fs_info->fs_devices->num_devices++;
2294 root->fs_info->fs_devices->open_devices++;
2295 root->fs_info->fs_devices->rw_devices++;
2296 root->fs_info->fs_devices->total_devices++;
2297 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2298
2299 spin_lock(&root->fs_info->free_chunk_lock);
2300 root->fs_info->free_chunk_space += device->total_bytes;
2301 spin_unlock(&root->fs_info->free_chunk_lock);
2302
2303 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2304 root->fs_info->fs_devices->rotating = 1;
2305
2306 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2307 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2308 tmp + device->total_bytes);
2309
2310 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2311 btrfs_set_super_num_devices(root->fs_info->super_copy,
2312 tmp + 1);
2313
2314 /* add sysfs device entry */
2315 btrfs_kobj_add_device(root->fs_info->fs_devices, device);
2316
2317 /*
2318 * we've got more storage, clear any full flags on the space
2319 * infos
2320 */
2321 btrfs_clear_space_info_full(root->fs_info);
2322
2323 unlock_chunks(root);
2324 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2325
2326 if (seeding_dev) {
2327 lock_chunks(root);
2328 ret = init_first_rw_device(trans, root, device);
2329 unlock_chunks(root);
2330 if (ret) {
2331 btrfs_abort_transaction(trans, root, ret);
2332 goto error_trans;
2333 }
2334 }
2335
2336 ret = btrfs_add_device(trans, root, device);
2337 if (ret) {
2338 btrfs_abort_transaction(trans, root, ret);
2339 goto error_trans;
2340 }
2341
2342 if (seeding_dev) {
2343 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2344
2345 ret = btrfs_finish_sprout(trans, root);
2346 if (ret) {
2347 btrfs_abort_transaction(trans, root, ret);
2348 goto error_trans;
2349 }
2350
2351 /* Sprouting would change fsid of the mounted root,
2352 * so rename the fsid on the sysfs
2353 */
2354 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2355 root->fs_info->fsid);
2356 if (kobject_rename(&root->fs_info->fs_devices->super_kobj,
2357 fsid_buf))
2358 pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n");
2359 }
2360
2361 root->fs_info->num_tolerated_disk_barrier_failures =
2362 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2363 ret = btrfs_commit_transaction(trans, root);
2364
2365 if (seeding_dev) {
2366 mutex_unlock(&uuid_mutex);
2367 up_write(&sb->s_umount);
2368
2369 if (ret) /* transaction commit */
2370 return ret;
2371
2372 ret = btrfs_relocate_sys_chunks(root);
2373 if (ret < 0)
2374 btrfs_error(root->fs_info, ret,
2375 "Failed to relocate sys chunks after "
2376 "device initialization. This can be fixed "
2377 "using the \"btrfs balance\" command.");
2378 trans = btrfs_attach_transaction(root);
2379 if (IS_ERR(trans)) {
2380 if (PTR_ERR(trans) == -ENOENT)
2381 return 0;
2382 return PTR_ERR(trans);
2383 }
2384 ret = btrfs_commit_transaction(trans, root);
2385 }
2386
2387 /* Update ctime/mtime for libblkid */
2388 update_dev_time(device_path);
2389 return ret;
2390
2391 error_trans:
2392 btrfs_end_transaction(trans, root);
2393 rcu_string_free(device->name);
2394 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
2395 kfree(device);
2396 error:
2397 blkdev_put(bdev, FMODE_EXCL);
2398 if (seeding_dev) {
2399 mutex_unlock(&uuid_mutex);
2400 up_write(&sb->s_umount);
2401 }
2402 return ret;
2403 }
2404
2405 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2406 struct btrfs_device *srcdev,
2407 struct btrfs_device **device_out)
2408 {
2409 struct request_queue *q;
2410 struct btrfs_device *device;
2411 struct block_device *bdev;
2412 struct btrfs_fs_info *fs_info = root->fs_info;
2413 struct list_head *devices;
2414 struct rcu_string *name;
2415 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2416 int ret = 0;
2417
2418 *device_out = NULL;
2419 if (fs_info->fs_devices->seeding) {
2420 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2421 return -EINVAL;
2422 }
2423
2424 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2425 fs_info->bdev_holder);
2426 if (IS_ERR(bdev)) {
2427 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2428 return PTR_ERR(bdev);
2429 }
2430
2431 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2432
2433 devices = &fs_info->fs_devices->devices;
2434 list_for_each_entry(device, devices, dev_list) {
2435 if (device->bdev == bdev) {
2436 btrfs_err(fs_info, "target device is in the filesystem!");
2437 ret = -EEXIST;
2438 goto error;
2439 }
2440 }
2441
2442
2443 if (i_size_read(bdev->bd_inode) <
2444 btrfs_device_get_total_bytes(srcdev)) {
2445 btrfs_err(fs_info, "target device is smaller than source device!");
2446 ret = -EINVAL;
2447 goto error;
2448 }
2449
2450
2451 device = btrfs_alloc_device(NULL, &devid, NULL);
2452 if (IS_ERR(device)) {
2453 ret = PTR_ERR(device);
2454 goto error;
2455 }
2456
2457 name = rcu_string_strdup(device_path, GFP_NOFS);
2458 if (!name) {
2459 kfree(device);
2460 ret = -ENOMEM;
2461 goto error;
2462 }
2463 rcu_assign_pointer(device->name, name);
2464
2465 q = bdev_get_queue(bdev);
2466 if (blk_queue_discard(q))
2467 device->can_discard = 1;
2468 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2469 device->writeable = 1;
2470 device->generation = 0;
2471 device->io_width = root->sectorsize;
2472 device->io_align = root->sectorsize;
2473 device->sector_size = root->sectorsize;
2474 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2475 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2476 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2477 ASSERT(list_empty(&srcdev->resized_list));
2478 device->commit_total_bytes = srcdev->commit_total_bytes;
2479 device->commit_bytes_used = device->bytes_used;
2480 device->dev_root = fs_info->dev_root;
2481 device->bdev = bdev;
2482 device->in_fs_metadata = 1;
2483 device->is_tgtdev_for_dev_replace = 1;
2484 device->mode = FMODE_EXCL;
2485 device->dev_stats_valid = 1;
2486 set_blocksize(device->bdev, 4096);
2487 device->fs_devices = fs_info->fs_devices;
2488 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2489 fs_info->fs_devices->num_devices++;
2490 fs_info->fs_devices->open_devices++;
2491 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2492
2493 *device_out = device;
2494 return ret;
2495
2496 error:
2497 blkdev_put(bdev, FMODE_EXCL);
2498 return ret;
2499 }
2500
2501 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2502 struct btrfs_device *tgtdev)
2503 {
2504 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2505 tgtdev->io_width = fs_info->dev_root->sectorsize;
2506 tgtdev->io_align = fs_info->dev_root->sectorsize;
2507 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2508 tgtdev->dev_root = fs_info->dev_root;
2509 tgtdev->in_fs_metadata = 1;
2510 }
2511
2512 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2513 struct btrfs_device *device)
2514 {
2515 int ret;
2516 struct btrfs_path *path;
2517 struct btrfs_root *root;
2518 struct btrfs_dev_item *dev_item;
2519 struct extent_buffer *leaf;
2520 struct btrfs_key key;
2521
2522 root = device->dev_root->fs_info->chunk_root;
2523
2524 path = btrfs_alloc_path();
2525 if (!path)
2526 return -ENOMEM;
2527
2528 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2529 key.type = BTRFS_DEV_ITEM_KEY;
2530 key.offset = device->devid;
2531
2532 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2533 if (ret < 0)
2534 goto out;
2535
2536 if (ret > 0) {
2537 ret = -ENOENT;
2538 goto out;
2539 }
2540
2541 leaf = path->nodes[0];
2542 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2543
2544 btrfs_set_device_id(leaf, dev_item, device->devid);
2545 btrfs_set_device_type(leaf, dev_item, device->type);
2546 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2547 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2548 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2549 btrfs_set_device_total_bytes(leaf, dev_item,
2550 btrfs_device_get_disk_total_bytes(device));
2551 btrfs_set_device_bytes_used(leaf, dev_item,
2552 btrfs_device_get_bytes_used(device));
2553 btrfs_mark_buffer_dirty(leaf);
2554
2555 out:
2556 btrfs_free_path(path);
2557 return ret;
2558 }
2559
2560 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2561 struct btrfs_device *device, u64 new_size)
2562 {
2563 struct btrfs_super_block *super_copy =
2564 device->dev_root->fs_info->super_copy;
2565 struct btrfs_fs_devices *fs_devices;
2566 u64 old_total;
2567 u64 diff;
2568
2569 if (!device->writeable)
2570 return -EACCES;
2571
2572 lock_chunks(device->dev_root);
2573 old_total = btrfs_super_total_bytes(super_copy);
2574 diff = new_size - device->total_bytes;
2575
2576 if (new_size <= device->total_bytes ||
2577 device->is_tgtdev_for_dev_replace) {
2578 unlock_chunks(device->dev_root);
2579 return -EINVAL;
2580 }
2581
2582 fs_devices = device->dev_root->fs_info->fs_devices;
2583
2584 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2585 device->fs_devices->total_rw_bytes += diff;
2586
2587 btrfs_device_set_total_bytes(device, new_size);
2588 btrfs_device_set_disk_total_bytes(device, new_size);
2589 btrfs_clear_space_info_full(device->dev_root->fs_info);
2590 if (list_empty(&device->resized_list))
2591 list_add_tail(&device->resized_list,
2592 &fs_devices->resized_devices);
2593 unlock_chunks(device->dev_root);
2594
2595 return btrfs_update_device(trans, device);
2596 }
2597
2598 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2599 struct btrfs_root *root, u64 chunk_objectid,
2600 u64 chunk_offset)
2601 {
2602 int ret;
2603 struct btrfs_path *path;
2604 struct btrfs_key key;
2605
2606 root = root->fs_info->chunk_root;
2607 path = btrfs_alloc_path();
2608 if (!path)
2609 return -ENOMEM;
2610
2611 key.objectid = chunk_objectid;
2612 key.offset = chunk_offset;
2613 key.type = BTRFS_CHUNK_ITEM_KEY;
2614
2615 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2616 if (ret < 0)
2617 goto out;
2618 else if (ret > 0) { /* Logic error or corruption */
2619 btrfs_error(root->fs_info, -ENOENT,
2620 "Failed lookup while freeing chunk.");
2621 ret = -ENOENT;
2622 goto out;
2623 }
2624
2625 ret = btrfs_del_item(trans, root, path);
2626 if (ret < 0)
2627 btrfs_error(root->fs_info, ret,
2628 "Failed to delete chunk item.");
2629 out:
2630 btrfs_free_path(path);
2631 return ret;
2632 }
2633
2634 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2635 chunk_offset)
2636 {
2637 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2638 struct btrfs_disk_key *disk_key;
2639 struct btrfs_chunk *chunk;
2640 u8 *ptr;
2641 int ret = 0;
2642 u32 num_stripes;
2643 u32 array_size;
2644 u32 len = 0;
2645 u32 cur;
2646 struct btrfs_key key;
2647
2648 lock_chunks(root);
2649 array_size = btrfs_super_sys_array_size(super_copy);
2650
2651 ptr = super_copy->sys_chunk_array;
2652 cur = 0;
2653
2654 while (cur < array_size) {
2655 disk_key = (struct btrfs_disk_key *)ptr;
2656 btrfs_disk_key_to_cpu(&key, disk_key);
2657
2658 len = sizeof(*disk_key);
2659
2660 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2661 chunk = (struct btrfs_chunk *)(ptr + len);
2662 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2663 len += btrfs_chunk_item_size(num_stripes);
2664 } else {
2665 ret = -EIO;
2666 break;
2667 }
2668 if (key.objectid == chunk_objectid &&
2669 key.offset == chunk_offset) {
2670 memmove(ptr, ptr + len, array_size - (cur + len));
2671 array_size -= len;
2672 btrfs_set_super_sys_array_size(super_copy, array_size);
2673 } else {
2674 ptr += len;
2675 cur += len;
2676 }
2677 }
2678 unlock_chunks(root);
2679 return ret;
2680 }
2681
2682 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root, u64 chunk_offset)
2684 {
2685 struct extent_map_tree *em_tree;
2686 struct extent_map *em;
2687 struct btrfs_root *extent_root = root->fs_info->extent_root;
2688 struct map_lookup *map;
2689 u64 dev_extent_len = 0;
2690 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2691 int i, ret = 0;
2692
2693 /* Just in case */
2694 root = root->fs_info->chunk_root;
2695 em_tree = &root->fs_info->mapping_tree.map_tree;
2696
2697 read_lock(&em_tree->lock);
2698 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2699 read_unlock(&em_tree->lock);
2700
2701 if (!em || em->start > chunk_offset ||
2702 em->start + em->len < chunk_offset) {
2703 /*
2704 * This is a logic error, but we don't want to just rely on the
2705 * user having built with ASSERT enabled, so if ASSERT doens't
2706 * do anything we still error out.
2707 */
2708 ASSERT(0);
2709 if (em)
2710 free_extent_map(em);
2711 return -EINVAL;
2712 }
2713 map = (struct map_lookup *)em->bdev;
2714 lock_chunks(root->fs_info->chunk_root);
2715 check_system_chunk(trans, extent_root, map->type);
2716 unlock_chunks(root->fs_info->chunk_root);
2717
2718 for (i = 0; i < map->num_stripes; i++) {
2719 struct btrfs_device *device = map->stripes[i].dev;
2720 ret = btrfs_free_dev_extent(trans, device,
2721 map->stripes[i].physical,
2722 &dev_extent_len);
2723 if (ret) {
2724 btrfs_abort_transaction(trans, root, ret);
2725 goto out;
2726 }
2727
2728 if (device->bytes_used > 0) {
2729 lock_chunks(root);
2730 btrfs_device_set_bytes_used(device,
2731 device->bytes_used - dev_extent_len);
2732 spin_lock(&root->fs_info->free_chunk_lock);
2733 root->fs_info->free_chunk_space += dev_extent_len;
2734 spin_unlock(&root->fs_info->free_chunk_lock);
2735 btrfs_clear_space_info_full(root->fs_info);
2736 unlock_chunks(root);
2737 }
2738
2739 if (map->stripes[i].dev) {
2740 ret = btrfs_update_device(trans, map->stripes[i].dev);
2741 if (ret) {
2742 btrfs_abort_transaction(trans, root, ret);
2743 goto out;
2744 }
2745 }
2746 }
2747 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2748 if (ret) {
2749 btrfs_abort_transaction(trans, root, ret);
2750 goto out;
2751 }
2752
2753 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2754
2755 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2756 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2757 if (ret) {
2758 btrfs_abort_transaction(trans, root, ret);
2759 goto out;
2760 }
2761 }
2762
2763 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2764 if (ret) {
2765 btrfs_abort_transaction(trans, extent_root, ret);
2766 goto out;
2767 }
2768
2769 out:
2770 /* once for us */
2771 free_extent_map(em);
2772 return ret;
2773 }
2774
2775 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2776 {
2777 struct btrfs_root *extent_root;
2778 struct btrfs_trans_handle *trans;
2779 int ret;
2780
2781 root = root->fs_info->chunk_root;
2782 extent_root = root->fs_info->extent_root;
2783
2784 /*
2785 * Prevent races with automatic removal of unused block groups.
2786 * After we relocate and before we remove the chunk with offset
2787 * chunk_offset, automatic removal of the block group can kick in,
2788 * resulting in a failure when calling btrfs_remove_chunk() below.
2789 *
2790 * Make sure to acquire this mutex before doing a tree search (dev
2791 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2792 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2793 * we release the path used to search the chunk/dev tree and before
2794 * the current task acquires this mutex and calls us.
2795 */
2796 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2797
2798 ret = btrfs_can_relocate(extent_root, chunk_offset);
2799 if (ret)
2800 return -ENOSPC;
2801
2802 /* step one, relocate all the extents inside this chunk */
2803 btrfs_scrub_pause(root);
2804 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2805 btrfs_scrub_continue(root);
2806 if (ret)
2807 return ret;
2808
2809 trans = btrfs_start_transaction(root, 0);
2810 if (IS_ERR(trans)) {
2811 ret = PTR_ERR(trans);
2812 btrfs_std_error(root->fs_info, ret);
2813 return ret;
2814 }
2815
2816 /*
2817 * step two, delete the device extents and the
2818 * chunk tree entries
2819 */
2820 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2821 btrfs_end_transaction(trans, root);
2822 return ret;
2823 }
2824
2825 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2826 {
2827 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2828 struct btrfs_path *path;
2829 struct extent_buffer *leaf;
2830 struct btrfs_chunk *chunk;
2831 struct btrfs_key key;
2832 struct btrfs_key found_key;
2833 u64 chunk_type;
2834 bool retried = false;
2835 int failed = 0;
2836 int ret;
2837
2838 path = btrfs_alloc_path();
2839 if (!path)
2840 return -ENOMEM;
2841
2842 again:
2843 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2844 key.offset = (u64)-1;
2845 key.type = BTRFS_CHUNK_ITEM_KEY;
2846
2847 while (1) {
2848 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2849 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2850 if (ret < 0) {
2851 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2852 goto error;
2853 }
2854 BUG_ON(ret == 0); /* Corruption */
2855
2856 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2857 key.type);
2858 if (ret)
2859 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2860 if (ret < 0)
2861 goto error;
2862 if (ret > 0)
2863 break;
2864
2865 leaf = path->nodes[0];
2866 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2867
2868 chunk = btrfs_item_ptr(leaf, path->slots[0],
2869 struct btrfs_chunk);
2870 chunk_type = btrfs_chunk_type(leaf, chunk);
2871 btrfs_release_path(path);
2872
2873 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2874 ret = btrfs_relocate_chunk(chunk_root,
2875 found_key.offset);
2876 if (ret == -ENOSPC)
2877 failed++;
2878 else
2879 BUG_ON(ret);
2880 }
2881 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2882
2883 if (found_key.offset == 0)
2884 break;
2885 key.offset = found_key.offset - 1;
2886 }
2887 ret = 0;
2888 if (failed && !retried) {
2889 failed = 0;
2890 retried = true;
2891 goto again;
2892 } else if (WARN_ON(failed && retried)) {
2893 ret = -ENOSPC;
2894 }
2895 error:
2896 btrfs_free_path(path);
2897 return ret;
2898 }
2899
2900 static int insert_balance_item(struct btrfs_root *root,
2901 struct btrfs_balance_control *bctl)
2902 {
2903 struct btrfs_trans_handle *trans;
2904 struct btrfs_balance_item *item;
2905 struct btrfs_disk_balance_args disk_bargs;
2906 struct btrfs_path *path;
2907 struct extent_buffer *leaf;
2908 struct btrfs_key key;
2909 int ret, err;
2910
2911 path = btrfs_alloc_path();
2912 if (!path)
2913 return -ENOMEM;
2914
2915 trans = btrfs_start_transaction(root, 0);
2916 if (IS_ERR(trans)) {
2917 btrfs_free_path(path);
2918 return PTR_ERR(trans);
2919 }
2920
2921 key.objectid = BTRFS_BALANCE_OBJECTID;
2922 key.type = BTRFS_BALANCE_ITEM_KEY;
2923 key.offset = 0;
2924
2925 ret = btrfs_insert_empty_item(trans, root, path, &key,
2926 sizeof(*item));
2927 if (ret)
2928 goto out;
2929
2930 leaf = path->nodes[0];
2931 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2932
2933 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2934
2935 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2936 btrfs_set_balance_data(leaf, item, &disk_bargs);
2937 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2938 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2939 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2940 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2941
2942 btrfs_set_balance_flags(leaf, item, bctl->flags);
2943
2944 btrfs_mark_buffer_dirty(leaf);
2945 out:
2946 btrfs_free_path(path);
2947 err = btrfs_commit_transaction(trans, root);
2948 if (err && !ret)
2949 ret = err;
2950 return ret;
2951 }
2952
2953 static int del_balance_item(struct btrfs_root *root)
2954 {
2955 struct btrfs_trans_handle *trans;
2956 struct btrfs_path *path;
2957 struct btrfs_key key;
2958 int ret, err;
2959
2960 path = btrfs_alloc_path();
2961 if (!path)
2962 return -ENOMEM;
2963
2964 trans = btrfs_start_transaction(root, 0);
2965 if (IS_ERR(trans)) {
2966 btrfs_free_path(path);
2967 return PTR_ERR(trans);
2968 }
2969
2970 key.objectid = BTRFS_BALANCE_OBJECTID;
2971 key.type = BTRFS_BALANCE_ITEM_KEY;
2972 key.offset = 0;
2973
2974 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2975 if (ret < 0)
2976 goto out;
2977 if (ret > 0) {
2978 ret = -ENOENT;
2979 goto out;
2980 }
2981
2982 ret = btrfs_del_item(trans, root, path);
2983 out:
2984 btrfs_free_path(path);
2985 err = btrfs_commit_transaction(trans, root);
2986 if (err && !ret)
2987 ret = err;
2988 return ret;
2989 }
2990
2991 /*
2992 * This is a heuristic used to reduce the number of chunks balanced on
2993 * resume after balance was interrupted.
2994 */
2995 static void update_balance_args(struct btrfs_balance_control *bctl)
2996 {
2997 /*
2998 * Turn on soft mode for chunk types that were being converted.
2999 */
3000 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3001 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3002 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3003 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3004 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3005 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3006
3007 /*
3008 * Turn on usage filter if is not already used. The idea is
3009 * that chunks that we have already balanced should be
3010 * reasonably full. Don't do it for chunks that are being
3011 * converted - that will keep us from relocating unconverted
3012 * (albeit full) chunks.
3013 */
3014 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3015 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3016 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3017 bctl->data.usage = 90;
3018 }
3019 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3020 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3021 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3022 bctl->sys.usage = 90;
3023 }
3024 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3025 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3026 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3027 bctl->meta.usage = 90;
3028 }
3029 }
3030
3031 /*
3032 * Should be called with both balance and volume mutexes held to
3033 * serialize other volume operations (add_dev/rm_dev/resize) with
3034 * restriper. Same goes for unset_balance_control.
3035 */
3036 static void set_balance_control(struct btrfs_balance_control *bctl)
3037 {
3038 struct btrfs_fs_info *fs_info = bctl->fs_info;
3039
3040 BUG_ON(fs_info->balance_ctl);
3041
3042 spin_lock(&fs_info->balance_lock);
3043 fs_info->balance_ctl = bctl;
3044 spin_unlock(&fs_info->balance_lock);
3045 }
3046
3047 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3048 {
3049 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3050
3051 BUG_ON(!fs_info->balance_ctl);
3052
3053 spin_lock(&fs_info->balance_lock);
3054 fs_info->balance_ctl = NULL;
3055 spin_unlock(&fs_info->balance_lock);
3056
3057 kfree(bctl);
3058 }
3059
3060 /*
3061 * Balance filters. Return 1 if chunk should be filtered out
3062 * (should not be balanced).
3063 */
3064 static int chunk_profiles_filter(u64 chunk_type,
3065 struct btrfs_balance_args *bargs)
3066 {
3067 chunk_type = chunk_to_extended(chunk_type) &
3068 BTRFS_EXTENDED_PROFILE_MASK;
3069
3070 if (bargs->profiles & chunk_type)
3071 return 0;
3072
3073 return 1;
3074 }
3075
3076 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3077 struct btrfs_balance_args *bargs)
3078 {
3079 struct btrfs_block_group_cache *cache;
3080 u64 chunk_used, user_thresh;
3081 int ret = 1;
3082
3083 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3084 chunk_used = btrfs_block_group_used(&cache->item);
3085
3086 if (bargs->usage == 0)
3087 user_thresh = 1;
3088 else if (bargs->usage > 100)
3089 user_thresh = cache->key.offset;
3090 else
3091 user_thresh = div_factor_fine(cache->key.offset,
3092 bargs->usage);
3093
3094 if (chunk_used < user_thresh)
3095 ret = 0;
3096
3097 btrfs_put_block_group(cache);
3098 return ret;
3099 }
3100
3101 static int chunk_devid_filter(struct extent_buffer *leaf,
3102 struct btrfs_chunk *chunk,
3103 struct btrfs_balance_args *bargs)
3104 {
3105 struct btrfs_stripe *stripe;
3106 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3107 int i;
3108
3109 for (i = 0; i < num_stripes; i++) {
3110 stripe = btrfs_stripe_nr(chunk, i);
3111 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3112 return 0;
3113 }
3114
3115 return 1;
3116 }
3117
3118 /* [pstart, pend) */
3119 static int chunk_drange_filter(struct extent_buffer *leaf,
3120 struct btrfs_chunk *chunk,
3121 u64 chunk_offset,
3122 struct btrfs_balance_args *bargs)
3123 {
3124 struct btrfs_stripe *stripe;
3125 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3126 u64 stripe_offset;
3127 u64 stripe_length;
3128 int factor;
3129 int i;
3130
3131 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3132 return 0;
3133
3134 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3135 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3136 factor = num_stripes / 2;
3137 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3138 factor = num_stripes - 1;
3139 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3140 factor = num_stripes - 2;
3141 } else {
3142 factor = num_stripes;
3143 }
3144
3145 for (i = 0; i < num_stripes; i++) {
3146 stripe = btrfs_stripe_nr(chunk, i);
3147 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3148 continue;
3149
3150 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3151 stripe_length = btrfs_chunk_length(leaf, chunk);
3152 stripe_length = div_u64(stripe_length, factor);
3153
3154 if (stripe_offset < bargs->pend &&
3155 stripe_offset + stripe_length > bargs->pstart)
3156 return 0;
3157 }
3158
3159 return 1;
3160 }
3161
3162 /* [vstart, vend) */
3163 static int chunk_vrange_filter(struct extent_buffer *leaf,
3164 struct btrfs_chunk *chunk,
3165 u64 chunk_offset,
3166 struct btrfs_balance_args *bargs)
3167 {
3168 if (chunk_offset < bargs->vend &&
3169 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3170 /* at least part of the chunk is inside this vrange */
3171 return 0;
3172
3173 return 1;
3174 }
3175
3176 static int chunk_soft_convert_filter(u64 chunk_type,
3177 struct btrfs_balance_args *bargs)
3178 {
3179 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3180 return 0;
3181
3182 chunk_type = chunk_to_extended(chunk_type) &
3183 BTRFS_EXTENDED_PROFILE_MASK;
3184
3185 if (bargs->target == chunk_type)
3186 return 1;
3187
3188 return 0;
3189 }
3190
3191 static int should_balance_chunk(struct btrfs_root *root,
3192 struct extent_buffer *leaf,
3193 struct btrfs_chunk *chunk, u64 chunk_offset)
3194 {
3195 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3196 struct btrfs_balance_args *bargs = NULL;
3197 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3198
3199 /* type filter */
3200 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3201 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3202 return 0;
3203 }
3204
3205 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3206 bargs = &bctl->data;
3207 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3208 bargs = &bctl->sys;
3209 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3210 bargs = &bctl->meta;
3211
3212 /* profiles filter */
3213 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3214 chunk_profiles_filter(chunk_type, bargs)) {
3215 return 0;
3216 }
3217
3218 /* usage filter */
3219 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3220 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3221 return 0;
3222 }
3223
3224 /* devid filter */
3225 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3226 chunk_devid_filter(leaf, chunk, bargs)) {
3227 return 0;
3228 }
3229
3230 /* drange filter, makes sense only with devid filter */
3231 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3232 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3233 return 0;
3234 }
3235
3236 /* vrange filter */
3237 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3238 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3239 return 0;
3240 }
3241
3242 /* soft profile changing mode */
3243 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3244 chunk_soft_convert_filter(chunk_type, bargs)) {
3245 return 0;
3246 }
3247
3248 /*
3249 * limited by count, must be the last filter
3250 */
3251 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3252 if (bargs->limit == 0)
3253 return 0;
3254 else
3255 bargs->limit--;
3256 }
3257
3258 return 1;
3259 }
3260
3261 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3262 {
3263 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3264 struct btrfs_root *chunk_root = fs_info->chunk_root;
3265 struct btrfs_root *dev_root = fs_info->dev_root;
3266 struct list_head *devices;
3267 struct btrfs_device *device;
3268 u64 old_size;
3269 u64 size_to_free;
3270 struct btrfs_chunk *chunk;
3271 struct btrfs_path *path;
3272 struct btrfs_key key;
3273 struct btrfs_key found_key;
3274 struct btrfs_trans_handle *trans;
3275 struct extent_buffer *leaf;
3276 int slot;
3277 int ret;
3278 int enospc_errors = 0;
3279 bool counting = true;
3280 u64 limit_data = bctl->data.limit;
3281 u64 limit_meta = bctl->meta.limit;
3282 u64 limit_sys = bctl->sys.limit;
3283
3284 /* step one make some room on all the devices */
3285 devices = &fs_info->fs_devices->devices;
3286 list_for_each_entry(device, devices, dev_list) {
3287 old_size = btrfs_device_get_total_bytes(device);
3288 size_to_free = div_factor(old_size, 1);
3289 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3290 if (!device->writeable ||
3291 btrfs_device_get_total_bytes(device) -
3292 btrfs_device_get_bytes_used(device) > size_to_free ||
3293 device->is_tgtdev_for_dev_replace)
3294 continue;
3295
3296 ret = btrfs_shrink_device(device, old_size - size_to_free);
3297 if (ret == -ENOSPC)
3298 break;
3299 BUG_ON(ret);
3300
3301 trans = btrfs_start_transaction(dev_root, 0);
3302 BUG_ON(IS_ERR(trans));
3303
3304 ret = btrfs_grow_device(trans, device, old_size);
3305 BUG_ON(ret);
3306
3307 btrfs_end_transaction(trans, dev_root);
3308 }
3309
3310 /* step two, relocate all the chunks */
3311 path = btrfs_alloc_path();
3312 if (!path) {
3313 ret = -ENOMEM;
3314 goto error;
3315 }
3316
3317 /* zero out stat counters */
3318 spin_lock(&fs_info->balance_lock);
3319 memset(&bctl->stat, 0, sizeof(bctl->stat));
3320 spin_unlock(&fs_info->balance_lock);
3321 again:
3322 if (!counting) {
3323 bctl->data.limit = limit_data;
3324 bctl->meta.limit = limit_meta;
3325 bctl->sys.limit = limit_sys;
3326 }
3327 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3328 key.offset = (u64)-1;
3329 key.type = BTRFS_CHUNK_ITEM_KEY;
3330
3331 while (1) {
3332 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3333 atomic_read(&fs_info->balance_cancel_req)) {
3334 ret = -ECANCELED;
3335 goto error;
3336 }
3337
3338 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3339 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3340 if (ret < 0) {
3341 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3342 goto error;
3343 }
3344
3345 /*
3346 * this shouldn't happen, it means the last relocate
3347 * failed
3348 */
3349 if (ret == 0)
3350 BUG(); /* FIXME break ? */
3351
3352 ret = btrfs_previous_item(chunk_root, path, 0,
3353 BTRFS_CHUNK_ITEM_KEY);
3354 if (ret) {
3355 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3356 ret = 0;
3357 break;
3358 }
3359
3360 leaf = path->nodes[0];
3361 slot = path->slots[0];
3362 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3363
3364 if (found_key.objectid != key.objectid) {
3365 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3366 break;
3367 }
3368
3369 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3370
3371 if (!counting) {
3372 spin_lock(&fs_info->balance_lock);
3373 bctl->stat.considered++;
3374 spin_unlock(&fs_info->balance_lock);
3375 }
3376
3377 ret = should_balance_chunk(chunk_root, leaf, chunk,
3378 found_key.offset);
3379 btrfs_release_path(path);
3380 if (!ret) {
3381 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3382 goto loop;
3383 }
3384
3385 if (counting) {
3386 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3387 spin_lock(&fs_info->balance_lock);
3388 bctl->stat.expected++;
3389 spin_unlock(&fs_info->balance_lock);
3390 goto loop;
3391 }
3392
3393 ret = btrfs_relocate_chunk(chunk_root,
3394 found_key.offset);
3395 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3396 if (ret && ret != -ENOSPC)
3397 goto error;
3398 if (ret == -ENOSPC) {
3399 enospc_errors++;
3400 } else {
3401 spin_lock(&fs_info->balance_lock);
3402 bctl->stat.completed++;
3403 spin_unlock(&fs_info->balance_lock);
3404 }
3405 loop:
3406 if (found_key.offset == 0)
3407 break;
3408 key.offset = found_key.offset - 1;
3409 }
3410
3411 if (counting) {
3412 btrfs_release_path(path);
3413 counting = false;
3414 goto again;
3415 }
3416 error:
3417 btrfs_free_path(path);
3418 if (enospc_errors) {
3419 btrfs_info(fs_info, "%d enospc errors during balance",
3420 enospc_errors);
3421 if (!ret)
3422 ret = -ENOSPC;
3423 }
3424
3425 return ret;
3426 }
3427
3428 /**
3429 * alloc_profile_is_valid - see if a given profile is valid and reduced
3430 * @flags: profile to validate
3431 * @extended: if true @flags is treated as an extended profile
3432 */
3433 static int alloc_profile_is_valid(u64 flags, int extended)
3434 {
3435 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3436 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3437
3438 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3439
3440 /* 1) check that all other bits are zeroed */
3441 if (flags & ~mask)
3442 return 0;
3443
3444 /* 2) see if profile is reduced */
3445 if (flags == 0)
3446 return !extended; /* "0" is valid for usual profiles */
3447
3448 /* true if exactly one bit set */
3449 return (flags & (flags - 1)) == 0;
3450 }
3451
3452 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3453 {
3454 /* cancel requested || normal exit path */
3455 return atomic_read(&fs_info->balance_cancel_req) ||
3456 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3457 atomic_read(&fs_info->balance_cancel_req) == 0);
3458 }
3459
3460 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3461 {
3462 int ret;
3463
3464 unset_balance_control(fs_info);
3465 ret = del_balance_item(fs_info->tree_root);
3466 if (ret)
3467 btrfs_std_error(fs_info, ret);
3468
3469 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3470 }
3471
3472 /*
3473 * Should be called with both balance and volume mutexes held
3474 */
3475 int btrfs_balance(struct btrfs_balance_control *bctl,
3476 struct btrfs_ioctl_balance_args *bargs)
3477 {
3478 struct btrfs_fs_info *fs_info = bctl->fs_info;
3479 u64 allowed;
3480 int mixed = 0;
3481 int ret;
3482 u64 num_devices;
3483 unsigned seq;
3484
3485 if (btrfs_fs_closing(fs_info) ||
3486 atomic_read(&fs_info->balance_pause_req) ||
3487 atomic_read(&fs_info->balance_cancel_req)) {
3488 ret = -EINVAL;
3489 goto out;
3490 }
3491
3492 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3493 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3494 mixed = 1;
3495
3496 /*
3497 * In case of mixed groups both data and meta should be picked,
3498 * and identical options should be given for both of them.
3499 */
3500 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3501 if (mixed && (bctl->flags & allowed)) {
3502 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3503 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3504 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3505 btrfs_err(fs_info, "with mixed groups data and "
3506 "metadata balance options must be the same");
3507 ret = -EINVAL;
3508 goto out;
3509 }
3510 }
3511
3512 num_devices = fs_info->fs_devices->num_devices;
3513 btrfs_dev_replace_lock(&fs_info->dev_replace);
3514 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3515 BUG_ON(num_devices < 1);
3516 num_devices--;
3517 }
3518 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3519 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3520 if (num_devices == 1)
3521 allowed |= BTRFS_BLOCK_GROUP_DUP;
3522 else if (num_devices > 1)
3523 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3524 if (num_devices > 2)
3525 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3526 if (num_devices > 3)
3527 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3528 BTRFS_BLOCK_GROUP_RAID6);
3529 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3530 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3531 (bctl->data.target & ~allowed))) {
3532 btrfs_err(fs_info, "unable to start balance with target "
3533 "data profile %llu",
3534 bctl->data.target);
3535 ret = -EINVAL;
3536 goto out;
3537 }
3538 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3539 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3540 (bctl->meta.target & ~allowed))) {
3541 btrfs_err(fs_info,
3542 "unable to start balance with target metadata profile %llu",
3543 bctl->meta.target);
3544 ret = -EINVAL;
3545 goto out;
3546 }
3547 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3548 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3549 (bctl->sys.target & ~allowed))) {
3550 btrfs_err(fs_info,
3551 "unable to start balance with target system profile %llu",
3552 bctl->sys.target);
3553 ret = -EINVAL;
3554 goto out;
3555 }
3556
3557 /* allow dup'ed data chunks only in mixed mode */
3558 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3559 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3560 btrfs_err(fs_info, "dup for data is not allowed");
3561 ret = -EINVAL;
3562 goto out;
3563 }
3564
3565 /* allow to reduce meta or sys integrity only if force set */
3566 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3567 BTRFS_BLOCK_GROUP_RAID10 |
3568 BTRFS_BLOCK_GROUP_RAID5 |
3569 BTRFS_BLOCK_GROUP_RAID6;
3570 do {
3571 seq = read_seqbegin(&fs_info->profiles_lock);
3572
3573 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3574 (fs_info->avail_system_alloc_bits & allowed) &&
3575 !(bctl->sys.target & allowed)) ||
3576 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3577 (fs_info->avail_metadata_alloc_bits & allowed) &&
3578 !(bctl->meta.target & allowed))) {
3579 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3580 btrfs_info(fs_info, "force reducing metadata integrity");
3581 } else {
3582 btrfs_err(fs_info, "balance will reduce metadata "
3583 "integrity, use force if you want this");
3584 ret = -EINVAL;
3585 goto out;
3586 }
3587 }
3588 } while (read_seqretry(&fs_info->profiles_lock, seq));
3589
3590 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3591 fs_info->num_tolerated_disk_barrier_failures = min(
3592 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3593 btrfs_get_num_tolerated_disk_barrier_failures(
3594 bctl->sys.target));
3595 }
3596
3597 ret = insert_balance_item(fs_info->tree_root, bctl);
3598 if (ret && ret != -EEXIST)
3599 goto out;
3600
3601 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3602 BUG_ON(ret == -EEXIST);
3603 set_balance_control(bctl);
3604 } else {
3605 BUG_ON(ret != -EEXIST);
3606 spin_lock(&fs_info->balance_lock);
3607 update_balance_args(bctl);
3608 spin_unlock(&fs_info->balance_lock);
3609 }
3610
3611 atomic_inc(&fs_info->balance_running);
3612 mutex_unlock(&fs_info->balance_mutex);
3613
3614 ret = __btrfs_balance(fs_info);
3615
3616 mutex_lock(&fs_info->balance_mutex);
3617 atomic_dec(&fs_info->balance_running);
3618
3619 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3620 fs_info->num_tolerated_disk_barrier_failures =
3621 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3622 }
3623
3624 if (bargs) {
3625 memset(bargs, 0, sizeof(*bargs));
3626 update_ioctl_balance_args(fs_info, 0, bargs);
3627 }
3628
3629 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3630 balance_need_close(fs_info)) {
3631 __cancel_balance(fs_info);
3632 }
3633
3634 wake_up(&fs_info->balance_wait_q);
3635
3636 return ret;
3637 out:
3638 if (bctl->flags & BTRFS_BALANCE_RESUME)
3639 __cancel_balance(fs_info);
3640 else {
3641 kfree(bctl);
3642 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3643 }
3644 return ret;
3645 }
3646
3647 static int balance_kthread(void *data)
3648 {
3649 struct btrfs_fs_info *fs_info = data;
3650 int ret = 0;
3651
3652 mutex_lock(&fs_info->volume_mutex);
3653 mutex_lock(&fs_info->balance_mutex);
3654
3655 if (fs_info->balance_ctl) {
3656 btrfs_info(fs_info, "continuing balance");
3657 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3658 }
3659
3660 mutex_unlock(&fs_info->balance_mutex);
3661 mutex_unlock(&fs_info->volume_mutex);
3662
3663 return ret;
3664 }
3665
3666 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3667 {
3668 struct task_struct *tsk;
3669
3670 spin_lock(&fs_info->balance_lock);
3671 if (!fs_info->balance_ctl) {
3672 spin_unlock(&fs_info->balance_lock);
3673 return 0;
3674 }
3675 spin_unlock(&fs_info->balance_lock);
3676
3677 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3678 btrfs_info(fs_info, "force skipping balance");
3679 return 0;
3680 }
3681
3682 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3683 return PTR_ERR_OR_ZERO(tsk);
3684 }
3685
3686 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3687 {
3688 struct btrfs_balance_control *bctl;
3689 struct btrfs_balance_item *item;
3690 struct btrfs_disk_balance_args disk_bargs;
3691 struct btrfs_path *path;
3692 struct extent_buffer *leaf;
3693 struct btrfs_key key;
3694 int ret;
3695
3696 path = btrfs_alloc_path();
3697 if (!path)
3698 return -ENOMEM;
3699
3700 key.objectid = BTRFS_BALANCE_OBJECTID;
3701 key.type = BTRFS_BALANCE_ITEM_KEY;
3702 key.offset = 0;
3703
3704 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3705 if (ret < 0)
3706 goto out;
3707 if (ret > 0) { /* ret = -ENOENT; */
3708 ret = 0;
3709 goto out;
3710 }
3711
3712 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3713 if (!bctl) {
3714 ret = -ENOMEM;
3715 goto out;
3716 }
3717
3718 leaf = path->nodes[0];
3719 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3720
3721 bctl->fs_info = fs_info;
3722 bctl->flags = btrfs_balance_flags(leaf, item);
3723 bctl->flags |= BTRFS_BALANCE_RESUME;
3724
3725 btrfs_balance_data(leaf, item, &disk_bargs);
3726 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3727 btrfs_balance_meta(leaf, item, &disk_bargs);
3728 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3729 btrfs_balance_sys(leaf, item, &disk_bargs);
3730 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3731
3732 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3733
3734 mutex_lock(&fs_info->volume_mutex);
3735 mutex_lock(&fs_info->balance_mutex);
3736
3737 set_balance_control(bctl);
3738
3739 mutex_unlock(&fs_info->balance_mutex);
3740 mutex_unlock(&fs_info->volume_mutex);
3741 out:
3742 btrfs_free_path(path);
3743 return ret;
3744 }
3745
3746 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3747 {
3748 int ret = 0;
3749
3750 mutex_lock(&fs_info->balance_mutex);
3751 if (!fs_info->balance_ctl) {
3752 mutex_unlock(&fs_info->balance_mutex);
3753 return -ENOTCONN;
3754 }
3755
3756 if (atomic_read(&fs_info->balance_running)) {
3757 atomic_inc(&fs_info->balance_pause_req);
3758 mutex_unlock(&fs_info->balance_mutex);
3759
3760 wait_event(fs_info->balance_wait_q,
3761 atomic_read(&fs_info->balance_running) == 0);
3762
3763 mutex_lock(&fs_info->balance_mutex);
3764 /* we are good with balance_ctl ripped off from under us */
3765 BUG_ON(atomic_read(&fs_info->balance_running));
3766 atomic_dec(&fs_info->balance_pause_req);
3767 } else {
3768 ret = -ENOTCONN;
3769 }
3770
3771 mutex_unlock(&fs_info->balance_mutex);
3772 return ret;
3773 }
3774
3775 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3776 {
3777 if (fs_info->sb->s_flags & MS_RDONLY)
3778 return -EROFS;
3779
3780 mutex_lock(&fs_info->balance_mutex);
3781 if (!fs_info->balance_ctl) {
3782 mutex_unlock(&fs_info->balance_mutex);
3783 return -ENOTCONN;
3784 }
3785
3786 atomic_inc(&fs_info->balance_cancel_req);
3787 /*
3788 * if we are running just wait and return, balance item is
3789 * deleted in btrfs_balance in this case
3790 */
3791 if (atomic_read(&fs_info->balance_running)) {
3792 mutex_unlock(&fs_info->balance_mutex);
3793 wait_event(fs_info->balance_wait_q,
3794 atomic_read(&fs_info->balance_running) == 0);
3795 mutex_lock(&fs_info->balance_mutex);
3796 } else {
3797 /* __cancel_balance needs volume_mutex */
3798 mutex_unlock(&fs_info->balance_mutex);
3799 mutex_lock(&fs_info->volume_mutex);
3800 mutex_lock(&fs_info->balance_mutex);
3801
3802 if (fs_info->balance_ctl)
3803 __cancel_balance(fs_info);
3804
3805 mutex_unlock(&fs_info->volume_mutex);
3806 }
3807
3808 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3809 atomic_dec(&fs_info->balance_cancel_req);
3810 mutex_unlock(&fs_info->balance_mutex);
3811 return 0;
3812 }
3813
3814 static int btrfs_uuid_scan_kthread(void *data)
3815 {
3816 struct btrfs_fs_info *fs_info = data;
3817 struct btrfs_root *root = fs_info->tree_root;
3818 struct btrfs_key key;
3819 struct btrfs_key max_key;
3820 struct btrfs_path *path = NULL;
3821 int ret = 0;
3822 struct extent_buffer *eb;
3823 int slot;
3824 struct btrfs_root_item root_item;
3825 u32 item_size;
3826 struct btrfs_trans_handle *trans = NULL;
3827
3828 path = btrfs_alloc_path();
3829 if (!path) {
3830 ret = -ENOMEM;
3831 goto out;
3832 }
3833
3834 key.objectid = 0;
3835 key.type = BTRFS_ROOT_ITEM_KEY;
3836 key.offset = 0;
3837
3838 max_key.objectid = (u64)-1;
3839 max_key.type = BTRFS_ROOT_ITEM_KEY;
3840 max_key.offset = (u64)-1;
3841
3842 while (1) {
3843 ret = btrfs_search_forward(root, &key, path, 0);
3844 if (ret) {
3845 if (ret > 0)
3846 ret = 0;
3847 break;
3848 }
3849
3850 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3851 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3852 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3853 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3854 goto skip;
3855
3856 eb = path->nodes[0];
3857 slot = path->slots[0];
3858 item_size = btrfs_item_size_nr(eb, slot);
3859 if (item_size < sizeof(root_item))
3860 goto skip;
3861
3862 read_extent_buffer(eb, &root_item,
3863 btrfs_item_ptr_offset(eb, slot),
3864 (int)sizeof(root_item));
3865 if (btrfs_root_refs(&root_item) == 0)
3866 goto skip;
3867
3868 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3869 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3870 if (trans)
3871 goto update_tree;
3872
3873 btrfs_release_path(path);
3874 /*
3875 * 1 - subvol uuid item
3876 * 1 - received_subvol uuid item
3877 */
3878 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3879 if (IS_ERR(trans)) {
3880 ret = PTR_ERR(trans);
3881 break;
3882 }
3883 continue;
3884 } else {
3885 goto skip;
3886 }
3887 update_tree:
3888 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3889 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3890 root_item.uuid,
3891 BTRFS_UUID_KEY_SUBVOL,
3892 key.objectid);
3893 if (ret < 0) {
3894 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3895 ret);
3896 break;
3897 }
3898 }
3899
3900 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3901 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3902 root_item.received_uuid,
3903 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3904 key.objectid);
3905 if (ret < 0) {
3906 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3907 ret);
3908 break;
3909 }
3910 }
3911
3912 skip:
3913 if (trans) {
3914 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3915 trans = NULL;
3916 if (ret)
3917 break;
3918 }
3919
3920 btrfs_release_path(path);
3921 if (key.offset < (u64)-1) {
3922 key.offset++;
3923 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3924 key.offset = 0;
3925 key.type = BTRFS_ROOT_ITEM_KEY;
3926 } else if (key.objectid < (u64)-1) {
3927 key.offset = 0;
3928 key.type = BTRFS_ROOT_ITEM_KEY;
3929 key.objectid++;
3930 } else {
3931 break;
3932 }
3933 cond_resched();
3934 }
3935
3936 out:
3937 btrfs_free_path(path);
3938 if (trans && !IS_ERR(trans))
3939 btrfs_end_transaction(trans, fs_info->uuid_root);
3940 if (ret)
3941 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3942 else
3943 fs_info->update_uuid_tree_gen = 1;
3944 up(&fs_info->uuid_tree_rescan_sem);
3945 return 0;
3946 }
3947
3948 /*
3949 * Callback for btrfs_uuid_tree_iterate().
3950 * returns:
3951 * 0 check succeeded, the entry is not outdated.
3952 * < 0 if an error occured.
3953 * > 0 if the check failed, which means the caller shall remove the entry.
3954 */
3955 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3956 u8 *uuid, u8 type, u64 subid)
3957 {
3958 struct btrfs_key key;
3959 int ret = 0;
3960 struct btrfs_root *subvol_root;
3961
3962 if (type != BTRFS_UUID_KEY_SUBVOL &&
3963 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3964 goto out;
3965
3966 key.objectid = subid;
3967 key.type = BTRFS_ROOT_ITEM_KEY;
3968 key.offset = (u64)-1;
3969 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3970 if (IS_ERR(subvol_root)) {
3971 ret = PTR_ERR(subvol_root);
3972 if (ret == -ENOENT)
3973 ret = 1;
3974 goto out;
3975 }
3976
3977 switch (type) {
3978 case BTRFS_UUID_KEY_SUBVOL:
3979 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3980 ret = 1;
3981 break;
3982 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3983 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3984 BTRFS_UUID_SIZE))
3985 ret = 1;
3986 break;
3987 }
3988
3989 out:
3990 return ret;
3991 }
3992
3993 static int btrfs_uuid_rescan_kthread(void *data)
3994 {
3995 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3996 int ret;
3997
3998 /*
3999 * 1st step is to iterate through the existing UUID tree and
4000 * to delete all entries that contain outdated data.
4001 * 2nd step is to add all missing entries to the UUID tree.
4002 */
4003 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4004 if (ret < 0) {
4005 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4006 up(&fs_info->uuid_tree_rescan_sem);
4007 return ret;
4008 }
4009 return btrfs_uuid_scan_kthread(data);
4010 }
4011
4012 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4013 {
4014 struct btrfs_trans_handle *trans;
4015 struct btrfs_root *tree_root = fs_info->tree_root;
4016 struct btrfs_root *uuid_root;
4017 struct task_struct *task;
4018 int ret;
4019
4020 /*
4021 * 1 - root node
4022 * 1 - root item
4023 */
4024 trans = btrfs_start_transaction(tree_root, 2);
4025 if (IS_ERR(trans))
4026 return PTR_ERR(trans);
4027
4028 uuid_root = btrfs_create_tree(trans, fs_info,
4029 BTRFS_UUID_TREE_OBJECTID);
4030 if (IS_ERR(uuid_root)) {
4031 ret = PTR_ERR(uuid_root);
4032 btrfs_abort_transaction(trans, tree_root, ret);
4033 return ret;
4034 }
4035
4036 fs_info->uuid_root = uuid_root;
4037
4038 ret = btrfs_commit_transaction(trans, tree_root);
4039 if (ret)
4040 return ret;
4041
4042 down(&fs_info->uuid_tree_rescan_sem);
4043 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4044 if (IS_ERR(task)) {
4045 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4046 btrfs_warn(fs_info, "failed to start uuid_scan task");
4047 up(&fs_info->uuid_tree_rescan_sem);
4048 return PTR_ERR(task);
4049 }
4050
4051 return 0;
4052 }
4053
4054 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4055 {
4056 struct task_struct *task;
4057
4058 down(&fs_info->uuid_tree_rescan_sem);
4059 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4060 if (IS_ERR(task)) {
4061 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4062 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4063 up(&fs_info->uuid_tree_rescan_sem);
4064 return PTR_ERR(task);
4065 }
4066
4067 return 0;
4068 }
4069
4070 /*
4071 * shrinking a device means finding all of the device extents past
4072 * the new size, and then following the back refs to the chunks.
4073 * The chunk relocation code actually frees the device extent
4074 */
4075 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4076 {
4077 struct btrfs_trans_handle *trans;
4078 struct btrfs_root *root = device->dev_root;
4079 struct btrfs_dev_extent *dev_extent = NULL;
4080 struct btrfs_path *path;
4081 u64 length;
4082 u64 chunk_offset;
4083 int ret;
4084 int slot;
4085 int failed = 0;
4086 bool retried = false;
4087 bool checked_pending_chunks = false;
4088 struct extent_buffer *l;
4089 struct btrfs_key key;
4090 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4091 u64 old_total = btrfs_super_total_bytes(super_copy);
4092 u64 old_size = btrfs_device_get_total_bytes(device);
4093 u64 diff = old_size - new_size;
4094
4095 if (device->is_tgtdev_for_dev_replace)
4096 return -EINVAL;
4097
4098 path = btrfs_alloc_path();
4099 if (!path)
4100 return -ENOMEM;
4101
4102 path->reada = 2;
4103
4104 lock_chunks(root);
4105
4106 btrfs_device_set_total_bytes(device, new_size);
4107 if (device->writeable) {
4108 device->fs_devices->total_rw_bytes -= diff;
4109 spin_lock(&root->fs_info->free_chunk_lock);
4110 root->fs_info->free_chunk_space -= diff;
4111 spin_unlock(&root->fs_info->free_chunk_lock);
4112 }
4113 unlock_chunks(root);
4114
4115 again:
4116 key.objectid = device->devid;
4117 key.offset = (u64)-1;
4118 key.type = BTRFS_DEV_EXTENT_KEY;
4119
4120 do {
4121 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4122 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4123 if (ret < 0) {
4124 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4125 goto done;
4126 }
4127
4128 ret = btrfs_previous_item(root, path, 0, key.type);
4129 if (ret)
4130 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4131 if (ret < 0)
4132 goto done;
4133 if (ret) {
4134 ret = 0;
4135 btrfs_release_path(path);
4136 break;
4137 }
4138
4139 l = path->nodes[0];
4140 slot = path->slots[0];
4141 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4142
4143 if (key.objectid != device->devid) {
4144 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4145 btrfs_release_path(path);
4146 break;
4147 }
4148
4149 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4150 length = btrfs_dev_extent_length(l, dev_extent);
4151
4152 if (key.offset + length <= new_size) {
4153 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4154 btrfs_release_path(path);
4155 break;
4156 }
4157
4158 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4159 btrfs_release_path(path);
4160
4161 ret = btrfs_relocate_chunk(root, 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->transaction, device,
4204 &start, len)) {
4205 unlock_chunks(root);
4206 checked_pending_chunks = true;
4207 failed = 0;
4208 retried = false;
4209 ret = btrfs_commit_transaction(trans, root);
4210 if (ret)
4211 goto done;
4212 goto again;
4213 }
4214 }
4215
4216 btrfs_device_set_disk_total_bytes(device, new_size);
4217 if (list_empty(&device->resized_list))
4218 list_add_tail(&device->resized_list,
4219 &root->fs_info->fs_devices->resized_devices);
4220
4221 WARN_ON(diff > old_total);
4222 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4223 unlock_chunks(root);
4224
4225 /* Now btrfs_update_device() will change the on-disk size. */
4226 ret = btrfs_update_device(trans, device);
4227 btrfs_end_transaction(trans, root);
4228 done:
4229 btrfs_free_path(path);
4230 if (ret) {
4231 lock_chunks(root);
4232 btrfs_device_set_total_bytes(device, old_size);
4233 if (device->writeable)
4234 device->fs_devices->total_rw_bytes += diff;
4235 spin_lock(&root->fs_info->free_chunk_lock);
4236 root->fs_info->free_chunk_space += diff;
4237 spin_unlock(&root->fs_info->free_chunk_lock);
4238 unlock_chunks(root);
4239 }
4240 return ret;
4241 }
4242
4243 static int btrfs_add_system_chunk(struct btrfs_root *root,
4244 struct btrfs_key *key,
4245 struct btrfs_chunk *chunk, int item_size)
4246 {
4247 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4248 struct btrfs_disk_key disk_key;
4249 u32 array_size;
4250 u8 *ptr;
4251
4252 lock_chunks(root);
4253 array_size = btrfs_super_sys_array_size(super_copy);
4254 if (array_size + item_size + sizeof(disk_key)
4255 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4256 unlock_chunks(root);
4257 return -EFBIG;
4258 }
4259
4260 ptr = super_copy->sys_chunk_array + array_size;
4261 btrfs_cpu_key_to_disk(&disk_key, key);
4262 memcpy(ptr, &disk_key, sizeof(disk_key));
4263 ptr += sizeof(disk_key);
4264 memcpy(ptr, chunk, item_size);
4265 item_size += sizeof(disk_key);
4266 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4267 unlock_chunks(root);
4268
4269 return 0;
4270 }
4271
4272 /*
4273 * sort the devices in descending order by max_avail, total_avail
4274 */
4275 static int btrfs_cmp_device_info(const void *a, const void *b)
4276 {
4277 const struct btrfs_device_info *di_a = a;
4278 const struct btrfs_device_info *di_b = b;
4279
4280 if (di_a->max_avail > di_b->max_avail)
4281 return -1;
4282 if (di_a->max_avail < di_b->max_avail)
4283 return 1;
4284 if (di_a->total_avail > di_b->total_avail)
4285 return -1;
4286 if (di_a->total_avail < di_b->total_avail)
4287 return 1;
4288 return 0;
4289 }
4290
4291 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4292 [BTRFS_RAID_RAID10] = {
4293 .sub_stripes = 2,
4294 .dev_stripes = 1,
4295 .devs_max = 0, /* 0 == as many as possible */
4296 .devs_min = 4,
4297 .devs_increment = 2,
4298 .ncopies = 2,
4299 },
4300 [BTRFS_RAID_RAID1] = {
4301 .sub_stripes = 1,
4302 .dev_stripes = 1,
4303 .devs_max = 2,
4304 .devs_min = 2,
4305 .devs_increment = 2,
4306 .ncopies = 2,
4307 },
4308 [BTRFS_RAID_DUP] = {
4309 .sub_stripes = 1,
4310 .dev_stripes = 2,
4311 .devs_max = 1,
4312 .devs_min = 1,
4313 .devs_increment = 1,
4314 .ncopies = 2,
4315 },
4316 [BTRFS_RAID_RAID0] = {
4317 .sub_stripes = 1,
4318 .dev_stripes = 1,
4319 .devs_max = 0,
4320 .devs_min = 2,
4321 .devs_increment = 1,
4322 .ncopies = 1,
4323 },
4324 [BTRFS_RAID_SINGLE] = {
4325 .sub_stripes = 1,
4326 .dev_stripes = 1,
4327 .devs_max = 1,
4328 .devs_min = 1,
4329 .devs_increment = 1,
4330 .ncopies = 1,
4331 },
4332 [BTRFS_RAID_RAID5] = {
4333 .sub_stripes = 1,
4334 .dev_stripes = 1,
4335 .devs_max = 0,
4336 .devs_min = 2,
4337 .devs_increment = 1,
4338 .ncopies = 2,
4339 },
4340 [BTRFS_RAID_RAID6] = {
4341 .sub_stripes = 1,
4342 .dev_stripes = 1,
4343 .devs_max = 0,
4344 .devs_min = 3,
4345 .devs_increment = 1,
4346 .ncopies = 3,
4347 },
4348 };
4349
4350 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4351 {
4352 /* TODO allow them to set a preferred stripe size */
4353 return 64 * 1024;
4354 }
4355
4356 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4357 {
4358 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4359 return;
4360
4361 btrfs_set_fs_incompat(info, RAID56);
4362 }
4363
4364 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4365 - sizeof(struct btrfs_item) \
4366 - sizeof(struct btrfs_chunk)) \
4367 / sizeof(struct btrfs_stripe) + 1)
4368
4369 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4370 - 2 * sizeof(struct btrfs_disk_key) \
4371 - 2 * sizeof(struct btrfs_chunk)) \
4372 / sizeof(struct btrfs_stripe) + 1)
4373
4374 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4375 struct btrfs_root *extent_root, u64 start,
4376 u64 type)
4377 {
4378 struct btrfs_fs_info *info = extent_root->fs_info;
4379 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4380 struct list_head *cur;
4381 struct map_lookup *map = NULL;
4382 struct extent_map_tree *em_tree;
4383 struct extent_map *em;
4384 struct btrfs_device_info *devices_info = NULL;
4385 u64 total_avail;
4386 int num_stripes; /* total number of stripes to allocate */
4387 int data_stripes; /* number of stripes that count for
4388 block group size */
4389 int sub_stripes; /* sub_stripes info for map */
4390 int dev_stripes; /* stripes per dev */
4391 int devs_max; /* max devs to use */
4392 int devs_min; /* min devs needed */
4393 int devs_increment; /* ndevs has to be a multiple of this */
4394 int ncopies; /* how many copies to data has */
4395 int ret;
4396 u64 max_stripe_size;
4397 u64 max_chunk_size;
4398 u64 stripe_size;
4399 u64 num_bytes;
4400 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4401 int ndevs;
4402 int i;
4403 int j;
4404 int index;
4405
4406 BUG_ON(!alloc_profile_is_valid(type, 0));
4407
4408 if (list_empty(&fs_devices->alloc_list))
4409 return -ENOSPC;
4410
4411 index = __get_raid_index(type);
4412
4413 sub_stripes = btrfs_raid_array[index].sub_stripes;
4414 dev_stripes = btrfs_raid_array[index].dev_stripes;
4415 devs_max = btrfs_raid_array[index].devs_max;
4416 devs_min = btrfs_raid_array[index].devs_min;
4417 devs_increment = btrfs_raid_array[index].devs_increment;
4418 ncopies = btrfs_raid_array[index].ncopies;
4419
4420 if (type & BTRFS_BLOCK_GROUP_DATA) {
4421 max_stripe_size = 1024 * 1024 * 1024;
4422 max_chunk_size = 10 * max_stripe_size;
4423 if (!devs_max)
4424 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4425 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4426 /* for larger filesystems, use larger metadata chunks */
4427 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4428 max_stripe_size = 1024 * 1024 * 1024;
4429 else
4430 max_stripe_size = 256 * 1024 * 1024;
4431 max_chunk_size = max_stripe_size;
4432 if (!devs_max)
4433 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4434 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4435 max_stripe_size = 32 * 1024 * 1024;
4436 max_chunk_size = 2 * max_stripe_size;
4437 if (!devs_max)
4438 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4439 } else {
4440 btrfs_err(info, "invalid chunk type 0x%llx requested",
4441 type);
4442 BUG_ON(1);
4443 }
4444
4445 /* we don't want a chunk larger than 10% of writeable space */
4446 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4447 max_chunk_size);
4448
4449 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4450 GFP_NOFS);
4451 if (!devices_info)
4452 return -ENOMEM;
4453
4454 cur = fs_devices->alloc_list.next;
4455
4456 /*
4457 * in the first pass through the devices list, we gather information
4458 * about the available holes on each device.
4459 */
4460 ndevs = 0;
4461 while (cur != &fs_devices->alloc_list) {
4462 struct btrfs_device *device;
4463 u64 max_avail;
4464 u64 dev_offset;
4465
4466 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4467
4468 cur = cur->next;
4469
4470 if (!device->writeable) {
4471 WARN(1, KERN_ERR
4472 "BTRFS: read-only device in alloc_list\n");
4473 continue;
4474 }
4475
4476 if (!device->in_fs_metadata ||
4477 device->is_tgtdev_for_dev_replace)
4478 continue;
4479
4480 if (device->total_bytes > device->bytes_used)
4481 total_avail = device->total_bytes - device->bytes_used;
4482 else
4483 total_avail = 0;
4484
4485 /* If there is no space on this device, skip it. */
4486 if (total_avail == 0)
4487 continue;
4488
4489 ret = find_free_dev_extent(trans, device,
4490 max_stripe_size * dev_stripes,
4491 &dev_offset, &max_avail);
4492 if (ret && ret != -ENOSPC)
4493 goto error;
4494
4495 if (ret == 0)
4496 max_avail = max_stripe_size * dev_stripes;
4497
4498 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4499 continue;
4500
4501 if (ndevs == fs_devices->rw_devices) {
4502 WARN(1, "%s: found more than %llu devices\n",
4503 __func__, fs_devices->rw_devices);
4504 break;
4505 }
4506 devices_info[ndevs].dev_offset = dev_offset;
4507 devices_info[ndevs].max_avail = max_avail;
4508 devices_info[ndevs].total_avail = total_avail;
4509 devices_info[ndevs].dev = device;
4510 ++ndevs;
4511 }
4512
4513 /*
4514 * now sort the devices by hole size / available space
4515 */
4516 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4517 btrfs_cmp_device_info, NULL);
4518
4519 /* round down to number of usable stripes */
4520 ndevs -= ndevs % devs_increment;
4521
4522 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4523 ret = -ENOSPC;
4524 goto error;
4525 }
4526
4527 if (devs_max && ndevs > devs_max)
4528 ndevs = devs_max;
4529 /*
4530 * the primary goal is to maximize the number of stripes, so use as many
4531 * devices as possible, even if the stripes are not maximum sized.
4532 */
4533 stripe_size = devices_info[ndevs-1].max_avail;
4534 num_stripes = ndevs * dev_stripes;
4535
4536 /*
4537 * this will have to be fixed for RAID1 and RAID10 over
4538 * more drives
4539 */
4540 data_stripes = num_stripes / ncopies;
4541
4542 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4543 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4544 btrfs_super_stripesize(info->super_copy));
4545 data_stripes = num_stripes - 1;
4546 }
4547 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4548 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4549 btrfs_super_stripesize(info->super_copy));
4550 data_stripes = num_stripes - 2;
4551 }
4552
4553 /*
4554 * Use the number of data stripes to figure out how big this chunk
4555 * is really going to be in terms of logical address space,
4556 * and compare that answer with the max chunk size
4557 */
4558 if (stripe_size * data_stripes > max_chunk_size) {
4559 u64 mask = (1ULL << 24) - 1;
4560
4561 stripe_size = div_u64(max_chunk_size, data_stripes);
4562
4563 /* bump the answer up to a 16MB boundary */
4564 stripe_size = (stripe_size + mask) & ~mask;
4565
4566 /* but don't go higher than the limits we found
4567 * while searching for free extents
4568 */
4569 if (stripe_size > devices_info[ndevs-1].max_avail)
4570 stripe_size = devices_info[ndevs-1].max_avail;
4571 }
4572
4573 stripe_size = div_u64(stripe_size, dev_stripes);
4574
4575 /* align to BTRFS_STRIPE_LEN */
4576 stripe_size = div_u64(stripe_size, raid_stripe_len);
4577 stripe_size *= raid_stripe_len;
4578
4579 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4580 if (!map) {
4581 ret = -ENOMEM;
4582 goto error;
4583 }
4584 map->num_stripes = num_stripes;
4585
4586 for (i = 0; i < ndevs; ++i) {
4587 for (j = 0; j < dev_stripes; ++j) {
4588 int s = i * dev_stripes + j;
4589 map->stripes[s].dev = devices_info[i].dev;
4590 map->stripes[s].physical = devices_info[i].dev_offset +
4591 j * stripe_size;
4592 }
4593 }
4594 map->sector_size = extent_root->sectorsize;
4595 map->stripe_len = raid_stripe_len;
4596 map->io_align = raid_stripe_len;
4597 map->io_width = raid_stripe_len;
4598 map->type = type;
4599 map->sub_stripes = sub_stripes;
4600
4601 num_bytes = stripe_size * data_stripes;
4602
4603 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4604
4605 em = alloc_extent_map();
4606 if (!em) {
4607 kfree(map);
4608 ret = -ENOMEM;
4609 goto error;
4610 }
4611 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4612 em->bdev = (struct block_device *)map;
4613 em->start = start;
4614 em->len = num_bytes;
4615 em->block_start = 0;
4616 em->block_len = em->len;
4617 em->orig_block_len = stripe_size;
4618
4619 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4620 write_lock(&em_tree->lock);
4621 ret = add_extent_mapping(em_tree, em, 0);
4622 if (!ret) {
4623 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4624 atomic_inc(&em->refs);
4625 }
4626 write_unlock(&em_tree->lock);
4627 if (ret) {
4628 free_extent_map(em);
4629 goto error;
4630 }
4631
4632 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4633 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4634 start, num_bytes);
4635 if (ret)
4636 goto error_del_extent;
4637
4638 for (i = 0; i < map->num_stripes; i++) {
4639 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4640 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4641 }
4642
4643 spin_lock(&extent_root->fs_info->free_chunk_lock);
4644 extent_root->fs_info->free_chunk_space -= (stripe_size *
4645 map->num_stripes);
4646 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4647
4648 free_extent_map(em);
4649 check_raid56_incompat_flag(extent_root->fs_info, type);
4650
4651 kfree(devices_info);
4652 return 0;
4653
4654 error_del_extent:
4655 write_lock(&em_tree->lock);
4656 remove_extent_mapping(em_tree, em);
4657 write_unlock(&em_tree->lock);
4658
4659 /* One for our allocation */
4660 free_extent_map(em);
4661 /* One for the tree reference */
4662 free_extent_map(em);
4663 /* One for the pending_chunks list reference */
4664 free_extent_map(em);
4665 error:
4666 kfree(devices_info);
4667 return ret;
4668 }
4669
4670 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4671 struct btrfs_root *extent_root,
4672 u64 chunk_offset, u64 chunk_size)
4673 {
4674 struct btrfs_key key;
4675 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4676 struct btrfs_device *device;
4677 struct btrfs_chunk *chunk;
4678 struct btrfs_stripe *stripe;
4679 struct extent_map_tree *em_tree;
4680 struct extent_map *em;
4681 struct map_lookup *map;
4682 size_t item_size;
4683 u64 dev_offset;
4684 u64 stripe_size;
4685 int i = 0;
4686 int ret;
4687
4688 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4689 read_lock(&em_tree->lock);
4690 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4691 read_unlock(&em_tree->lock);
4692
4693 if (!em) {
4694 btrfs_crit(extent_root->fs_info, "unable to find logical "
4695 "%Lu len %Lu", chunk_offset, chunk_size);
4696 return -EINVAL;
4697 }
4698
4699 if (em->start != chunk_offset || em->len != chunk_size) {
4700 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4701 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4702 chunk_size, em->start, em->len);
4703 free_extent_map(em);
4704 return -EINVAL;
4705 }
4706
4707 map = (struct map_lookup *)em->bdev;
4708 item_size = btrfs_chunk_item_size(map->num_stripes);
4709 stripe_size = em->orig_block_len;
4710
4711 chunk = kzalloc(item_size, GFP_NOFS);
4712 if (!chunk) {
4713 ret = -ENOMEM;
4714 goto out;
4715 }
4716
4717 for (i = 0; i < map->num_stripes; i++) {
4718 device = map->stripes[i].dev;
4719 dev_offset = map->stripes[i].physical;
4720
4721 ret = btrfs_update_device(trans, device);
4722 if (ret)
4723 goto out;
4724 ret = btrfs_alloc_dev_extent(trans, device,
4725 chunk_root->root_key.objectid,
4726 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4727 chunk_offset, dev_offset,
4728 stripe_size);
4729 if (ret)
4730 goto out;
4731 }
4732
4733 stripe = &chunk->stripe;
4734 for (i = 0; i < map->num_stripes; i++) {
4735 device = map->stripes[i].dev;
4736 dev_offset = map->stripes[i].physical;
4737
4738 btrfs_set_stack_stripe_devid(stripe, device->devid);
4739 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4740 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4741 stripe++;
4742 }
4743
4744 btrfs_set_stack_chunk_length(chunk, chunk_size);
4745 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4746 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4747 btrfs_set_stack_chunk_type(chunk, map->type);
4748 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4749 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4750 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4751 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4752 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4753
4754 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4755 key.type = BTRFS_CHUNK_ITEM_KEY;
4756 key.offset = chunk_offset;
4757
4758 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4759 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4760 /*
4761 * TODO: Cleanup of inserted chunk root in case of
4762 * failure.
4763 */
4764 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4765 item_size);
4766 }
4767
4768 out:
4769 kfree(chunk);
4770 free_extent_map(em);
4771 return ret;
4772 }
4773
4774 /*
4775 * Chunk allocation falls into two parts. The first part does works
4776 * that make the new allocated chunk useable, but not do any operation
4777 * that modifies the chunk tree. The second part does the works that
4778 * require modifying the chunk tree. This division is important for the
4779 * bootstrap process of adding storage to a seed btrfs.
4780 */
4781 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4782 struct btrfs_root *extent_root, u64 type)
4783 {
4784 u64 chunk_offset;
4785
4786 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4787 chunk_offset = find_next_chunk(extent_root->fs_info);
4788 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4789 }
4790
4791 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4792 struct btrfs_root *root,
4793 struct btrfs_device *device)
4794 {
4795 u64 chunk_offset;
4796 u64 sys_chunk_offset;
4797 u64 alloc_profile;
4798 struct btrfs_fs_info *fs_info = root->fs_info;
4799 struct btrfs_root *extent_root = fs_info->extent_root;
4800 int ret;
4801
4802 chunk_offset = find_next_chunk(fs_info);
4803 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4804 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4805 alloc_profile);
4806 if (ret)
4807 return ret;
4808
4809 sys_chunk_offset = find_next_chunk(root->fs_info);
4810 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4811 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4812 alloc_profile);
4813 return ret;
4814 }
4815
4816 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4817 {
4818 int max_errors;
4819
4820 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4821 BTRFS_BLOCK_GROUP_RAID10 |
4822 BTRFS_BLOCK_GROUP_RAID5 |
4823 BTRFS_BLOCK_GROUP_DUP)) {
4824 max_errors = 1;
4825 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4826 max_errors = 2;
4827 } else {
4828 max_errors = 0;
4829 }
4830
4831 return max_errors;
4832 }
4833
4834 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4835 {
4836 struct extent_map *em;
4837 struct map_lookup *map;
4838 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4839 int readonly = 0;
4840 int miss_ndevs = 0;
4841 int i;
4842
4843 read_lock(&map_tree->map_tree.lock);
4844 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4845 read_unlock(&map_tree->map_tree.lock);
4846 if (!em)
4847 return 1;
4848
4849 map = (struct map_lookup *)em->bdev;
4850 for (i = 0; i < map->num_stripes; i++) {
4851 if (map->stripes[i].dev->missing) {
4852 miss_ndevs++;
4853 continue;
4854 }
4855
4856 if (!map->stripes[i].dev->writeable) {
4857 readonly = 1;
4858 goto end;
4859 }
4860 }
4861
4862 /*
4863 * If the number of missing devices is larger than max errors,
4864 * we can not write the data into that chunk successfully, so
4865 * set it readonly.
4866 */
4867 if (miss_ndevs > btrfs_chunk_max_errors(map))
4868 readonly = 1;
4869 end:
4870 free_extent_map(em);
4871 return readonly;
4872 }
4873
4874 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4875 {
4876 extent_map_tree_init(&tree->map_tree);
4877 }
4878
4879 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4880 {
4881 struct extent_map *em;
4882
4883 while (1) {
4884 write_lock(&tree->map_tree.lock);
4885 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4886 if (em)
4887 remove_extent_mapping(&tree->map_tree, em);
4888 write_unlock(&tree->map_tree.lock);
4889 if (!em)
4890 break;
4891 /* once for us */
4892 free_extent_map(em);
4893 /* once for the tree */
4894 free_extent_map(em);
4895 }
4896 }
4897
4898 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4899 {
4900 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4901 struct extent_map *em;
4902 struct map_lookup *map;
4903 struct extent_map_tree *em_tree = &map_tree->map_tree;
4904 int ret;
4905
4906 read_lock(&em_tree->lock);
4907 em = lookup_extent_mapping(em_tree, logical, len);
4908 read_unlock(&em_tree->lock);
4909
4910 /*
4911 * We could return errors for these cases, but that could get ugly and
4912 * we'd probably do the same thing which is just not do anything else
4913 * and exit, so return 1 so the callers don't try to use other copies.
4914 */
4915 if (!em) {
4916 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4917 logical+len);
4918 return 1;
4919 }
4920
4921 if (em->start > logical || em->start + em->len < logical) {
4922 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4923 "%Lu-%Lu", logical, logical+len, em->start,
4924 em->start + em->len);
4925 free_extent_map(em);
4926 return 1;
4927 }
4928
4929 map = (struct map_lookup *)em->bdev;
4930 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4931 ret = map->num_stripes;
4932 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4933 ret = map->sub_stripes;
4934 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4935 ret = 2;
4936 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4937 ret = 3;
4938 else
4939 ret = 1;
4940 free_extent_map(em);
4941
4942 btrfs_dev_replace_lock(&fs_info->dev_replace);
4943 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4944 ret++;
4945 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4946
4947 return ret;
4948 }
4949
4950 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4951 struct btrfs_mapping_tree *map_tree,
4952 u64 logical)
4953 {
4954 struct extent_map *em;
4955 struct map_lookup *map;
4956 struct extent_map_tree *em_tree = &map_tree->map_tree;
4957 unsigned long len = root->sectorsize;
4958
4959 read_lock(&em_tree->lock);
4960 em = lookup_extent_mapping(em_tree, logical, len);
4961 read_unlock(&em_tree->lock);
4962 BUG_ON(!em);
4963
4964 BUG_ON(em->start > logical || em->start + em->len < logical);
4965 map = (struct map_lookup *)em->bdev;
4966 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4967 len = map->stripe_len * nr_data_stripes(map);
4968 free_extent_map(em);
4969 return len;
4970 }
4971
4972 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4973 u64 logical, u64 len, int mirror_num)
4974 {
4975 struct extent_map *em;
4976 struct map_lookup *map;
4977 struct extent_map_tree *em_tree = &map_tree->map_tree;
4978 int ret = 0;
4979
4980 read_lock(&em_tree->lock);
4981 em = lookup_extent_mapping(em_tree, logical, len);
4982 read_unlock(&em_tree->lock);
4983 BUG_ON(!em);
4984
4985 BUG_ON(em->start > logical || em->start + em->len < logical);
4986 map = (struct map_lookup *)em->bdev;
4987 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4988 ret = 1;
4989 free_extent_map(em);
4990 return ret;
4991 }
4992
4993 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4994 struct map_lookup *map, int first, int num,
4995 int optimal, int dev_replace_is_ongoing)
4996 {
4997 int i;
4998 int tolerance;
4999 struct btrfs_device *srcdev;
5000
5001 if (dev_replace_is_ongoing &&
5002 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5003 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5004 srcdev = fs_info->dev_replace.srcdev;
5005 else
5006 srcdev = NULL;
5007
5008 /*
5009 * try to avoid the drive that is the source drive for a
5010 * dev-replace procedure, only choose it if no other non-missing
5011 * mirror is available
5012 */
5013 for (tolerance = 0; tolerance < 2; tolerance++) {
5014 if (map->stripes[optimal].dev->bdev &&
5015 (tolerance || map->stripes[optimal].dev != srcdev))
5016 return optimal;
5017 for (i = first; i < first + num; i++) {
5018 if (map->stripes[i].dev->bdev &&
5019 (tolerance || map->stripes[i].dev != srcdev))
5020 return i;
5021 }
5022 }
5023
5024 /* we couldn't find one that doesn't fail. Just return something
5025 * and the io error handling code will clean up eventually
5026 */
5027 return optimal;
5028 }
5029
5030 static inline int parity_smaller(u64 a, u64 b)
5031 {
5032 return a > b;
5033 }
5034
5035 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5036 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5037 {
5038 struct btrfs_bio_stripe s;
5039 int i;
5040 u64 l;
5041 int again = 1;
5042
5043 while (again) {
5044 again = 0;
5045 for (i = 0; i < num_stripes - 1; i++) {
5046 if (parity_smaller(bbio->raid_map[i],
5047 bbio->raid_map[i+1])) {
5048 s = bbio->stripes[i];
5049 l = bbio->raid_map[i];
5050 bbio->stripes[i] = bbio->stripes[i+1];
5051 bbio->raid_map[i] = bbio->raid_map[i+1];
5052 bbio->stripes[i+1] = s;
5053 bbio->raid_map[i+1] = l;
5054
5055 again = 1;
5056 }
5057 }
5058 }
5059 }
5060
5061 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5062 {
5063 struct btrfs_bio *bbio = kzalloc(
5064 /* the size of the btrfs_bio */
5065 sizeof(struct btrfs_bio) +
5066 /* plus the variable array for the stripes */
5067 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5068 /* plus the variable array for the tgt dev */
5069 sizeof(int) * (real_stripes) +
5070 /*
5071 * plus the raid_map, which includes both the tgt dev
5072 * and the stripes
5073 */
5074 sizeof(u64) * (total_stripes),
5075 GFP_NOFS|__GFP_NOFAIL);
5076
5077 atomic_set(&bbio->error, 0);
5078 atomic_set(&bbio->refs, 1);
5079
5080 return bbio;
5081 }
5082
5083 void btrfs_get_bbio(struct btrfs_bio *bbio)
5084 {
5085 WARN_ON(!atomic_read(&bbio->refs));
5086 atomic_inc(&bbio->refs);
5087 }
5088
5089 void btrfs_put_bbio(struct btrfs_bio *bbio)
5090 {
5091 if (!bbio)
5092 return;
5093 if (atomic_dec_and_test(&bbio->refs))
5094 kfree(bbio);
5095 }
5096
5097 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5098 u64 logical, u64 *length,
5099 struct btrfs_bio **bbio_ret,
5100 int mirror_num, int need_raid_map)
5101 {
5102 struct extent_map *em;
5103 struct map_lookup *map;
5104 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5105 struct extent_map_tree *em_tree = &map_tree->map_tree;
5106 u64 offset;
5107 u64 stripe_offset;
5108 u64 stripe_end_offset;
5109 u64 stripe_nr;
5110 u64 stripe_nr_orig;
5111 u64 stripe_nr_end;
5112 u64 stripe_len;
5113 u32 stripe_index;
5114 int i;
5115 int ret = 0;
5116 int num_stripes;
5117 int max_errors = 0;
5118 int tgtdev_indexes = 0;
5119 struct btrfs_bio *bbio = NULL;
5120 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5121 int dev_replace_is_ongoing = 0;
5122 int num_alloc_stripes;
5123 int patch_the_first_stripe_for_dev_replace = 0;
5124 u64 physical_to_patch_in_first_stripe = 0;
5125 u64 raid56_full_stripe_start = (u64)-1;
5126
5127 read_lock(&em_tree->lock);
5128 em = lookup_extent_mapping(em_tree, logical, *length);
5129 read_unlock(&em_tree->lock);
5130
5131 if (!em) {
5132 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5133 logical, *length);
5134 return -EINVAL;
5135 }
5136
5137 if (em->start > logical || em->start + em->len < logical) {
5138 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5139 "found %Lu-%Lu", logical, em->start,
5140 em->start + em->len);
5141 free_extent_map(em);
5142 return -EINVAL;
5143 }
5144
5145 map = (struct map_lookup *)em->bdev;
5146 offset = logical - em->start;
5147
5148 stripe_len = map->stripe_len;
5149 stripe_nr = offset;
5150 /*
5151 * stripe_nr counts the total number of stripes we have to stride
5152 * to get to this block
5153 */
5154 stripe_nr = div64_u64(stripe_nr, stripe_len);
5155
5156 stripe_offset = stripe_nr * stripe_len;
5157 BUG_ON(offset < stripe_offset);
5158
5159 /* stripe_offset is the offset of this block in its stripe*/
5160 stripe_offset = offset - stripe_offset;
5161
5162 /* if we're here for raid56, we need to know the stripe aligned start */
5163 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5164 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5165 raid56_full_stripe_start = offset;
5166
5167 /* allow a write of a full stripe, but make sure we don't
5168 * allow straddling of stripes
5169 */
5170 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5171 full_stripe_len);
5172 raid56_full_stripe_start *= full_stripe_len;
5173 }
5174
5175 if (rw & REQ_DISCARD) {
5176 /* we don't discard raid56 yet */
5177 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5178 ret = -EOPNOTSUPP;
5179 goto out;
5180 }
5181 *length = min_t(u64, em->len - offset, *length);
5182 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5183 u64 max_len;
5184 /* For writes to RAID[56], allow a full stripeset across all disks.
5185 For other RAID types and for RAID[56] reads, just allow a single
5186 stripe (on a single disk). */
5187 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5188 (rw & REQ_WRITE)) {
5189 max_len = stripe_len * nr_data_stripes(map) -
5190 (offset - raid56_full_stripe_start);
5191 } else {
5192 /* we limit the length of each bio to what fits in a stripe */
5193 max_len = stripe_len - stripe_offset;
5194 }
5195 *length = min_t(u64, em->len - offset, max_len);
5196 } else {
5197 *length = em->len - offset;
5198 }
5199
5200 /* This is for when we're called from btrfs_merge_bio_hook() and all
5201 it cares about is the length */
5202 if (!bbio_ret)
5203 goto out;
5204
5205 btrfs_dev_replace_lock(dev_replace);
5206 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5207 if (!dev_replace_is_ongoing)
5208 btrfs_dev_replace_unlock(dev_replace);
5209
5210 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5211 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5212 dev_replace->tgtdev != NULL) {
5213 /*
5214 * in dev-replace case, for repair case (that's the only
5215 * case where the mirror is selected explicitly when
5216 * calling btrfs_map_block), blocks left of the left cursor
5217 * can also be read from the target drive.
5218 * For REQ_GET_READ_MIRRORS, the target drive is added as
5219 * the last one to the array of stripes. For READ, it also
5220 * needs to be supported using the same mirror number.
5221 * If the requested block is not left of the left cursor,
5222 * EIO is returned. This can happen because btrfs_num_copies()
5223 * returns one more in the dev-replace case.
5224 */
5225 u64 tmp_length = *length;
5226 struct btrfs_bio *tmp_bbio = NULL;
5227 int tmp_num_stripes;
5228 u64 srcdev_devid = dev_replace->srcdev->devid;
5229 int index_srcdev = 0;
5230 int found = 0;
5231 u64 physical_of_found = 0;
5232
5233 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5234 logical, &tmp_length, &tmp_bbio, 0, 0);
5235 if (ret) {
5236 WARN_ON(tmp_bbio != NULL);
5237 goto out;
5238 }
5239
5240 tmp_num_stripes = tmp_bbio->num_stripes;
5241 if (mirror_num > tmp_num_stripes) {
5242 /*
5243 * REQ_GET_READ_MIRRORS does not contain this
5244 * mirror, that means that the requested area
5245 * is not left of the left cursor
5246 */
5247 ret = -EIO;
5248 btrfs_put_bbio(tmp_bbio);
5249 goto out;
5250 }
5251
5252 /*
5253 * process the rest of the function using the mirror_num
5254 * of the source drive. Therefore look it up first.
5255 * At the end, patch the device pointer to the one of the
5256 * target drive.
5257 */
5258 for (i = 0; i < tmp_num_stripes; i++) {
5259 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5260 /*
5261 * In case of DUP, in order to keep it
5262 * simple, only add the mirror with the
5263 * lowest physical address
5264 */
5265 if (found &&
5266 physical_of_found <=
5267 tmp_bbio->stripes[i].physical)
5268 continue;
5269 index_srcdev = i;
5270 found = 1;
5271 physical_of_found =
5272 tmp_bbio->stripes[i].physical;
5273 }
5274 }
5275
5276 if (found) {
5277 mirror_num = index_srcdev + 1;
5278 patch_the_first_stripe_for_dev_replace = 1;
5279 physical_to_patch_in_first_stripe = physical_of_found;
5280 } else {
5281 WARN_ON(1);
5282 ret = -EIO;
5283 btrfs_put_bbio(tmp_bbio);
5284 goto out;
5285 }
5286
5287 btrfs_put_bbio(tmp_bbio);
5288 } else if (mirror_num > map->num_stripes) {
5289 mirror_num = 0;
5290 }
5291
5292 num_stripes = 1;
5293 stripe_index = 0;
5294 stripe_nr_orig = stripe_nr;
5295 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5296 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5297 stripe_end_offset = stripe_nr_end * map->stripe_len -
5298 (offset + *length);
5299
5300 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5301 if (rw & REQ_DISCARD)
5302 num_stripes = min_t(u64, map->num_stripes,
5303 stripe_nr_end - stripe_nr_orig);
5304 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5305 &stripe_index);
5306 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5307 mirror_num = 1;
5308 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5309 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5310 num_stripes = map->num_stripes;
5311 else if (mirror_num)
5312 stripe_index = mirror_num - 1;
5313 else {
5314 stripe_index = find_live_mirror(fs_info, map, 0,
5315 map->num_stripes,
5316 current->pid % map->num_stripes,
5317 dev_replace_is_ongoing);
5318 mirror_num = stripe_index + 1;
5319 }
5320
5321 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5322 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5323 num_stripes = map->num_stripes;
5324 } else if (mirror_num) {
5325 stripe_index = mirror_num - 1;
5326 } else {
5327 mirror_num = 1;
5328 }
5329
5330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5331 u32 factor = map->num_stripes / map->sub_stripes;
5332
5333 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5334 stripe_index *= map->sub_stripes;
5335
5336 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5337 num_stripes = map->sub_stripes;
5338 else if (rw & REQ_DISCARD)
5339 num_stripes = min_t(u64, map->sub_stripes *
5340 (stripe_nr_end - stripe_nr_orig),
5341 map->num_stripes);
5342 else if (mirror_num)
5343 stripe_index += mirror_num - 1;
5344 else {
5345 int old_stripe_index = stripe_index;
5346 stripe_index = find_live_mirror(fs_info, map,
5347 stripe_index,
5348 map->sub_stripes, stripe_index +
5349 current->pid % map->sub_stripes,
5350 dev_replace_is_ongoing);
5351 mirror_num = stripe_index - old_stripe_index + 1;
5352 }
5353
5354 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5355 if (need_raid_map &&
5356 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5357 mirror_num > 1)) {
5358 /* push stripe_nr back to the start of the full stripe */
5359 stripe_nr = div_u64(raid56_full_stripe_start,
5360 stripe_len * nr_data_stripes(map));
5361
5362 /* RAID[56] write or recovery. Return all stripes */
5363 num_stripes = map->num_stripes;
5364 max_errors = nr_parity_stripes(map);
5365
5366 *length = map->stripe_len;
5367 stripe_index = 0;
5368 stripe_offset = 0;
5369 } else {
5370 /*
5371 * Mirror #0 or #1 means the original data block.
5372 * Mirror #2 is RAID5 parity block.
5373 * Mirror #3 is RAID6 Q block.
5374 */
5375 stripe_nr = div_u64_rem(stripe_nr,
5376 nr_data_stripes(map), &stripe_index);
5377 if (mirror_num > 1)
5378 stripe_index = nr_data_stripes(map) +
5379 mirror_num - 2;
5380
5381 /* We distribute the parity blocks across stripes */
5382 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5383 &stripe_index);
5384 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5385 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5386 mirror_num = 1;
5387 }
5388 } else {
5389 /*
5390 * after this, stripe_nr is the number of stripes on this
5391 * device we have to walk to find the data, and stripe_index is
5392 * the number of our device in the stripe array
5393 */
5394 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5395 &stripe_index);
5396 mirror_num = stripe_index + 1;
5397 }
5398 BUG_ON(stripe_index >= map->num_stripes);
5399
5400 num_alloc_stripes = num_stripes;
5401 if (dev_replace_is_ongoing) {
5402 if (rw & (REQ_WRITE | REQ_DISCARD))
5403 num_alloc_stripes <<= 1;
5404 if (rw & REQ_GET_READ_MIRRORS)
5405 num_alloc_stripes++;
5406 tgtdev_indexes = num_stripes;
5407 }
5408
5409 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5410 if (!bbio) {
5411 ret = -ENOMEM;
5412 goto out;
5413 }
5414 if (dev_replace_is_ongoing)
5415 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5416
5417 /* build raid_map */
5418 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5419 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5420 mirror_num > 1)) {
5421 u64 tmp;
5422 unsigned rot;
5423
5424 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5425 sizeof(struct btrfs_bio_stripe) *
5426 num_alloc_stripes +
5427 sizeof(int) * tgtdev_indexes);
5428
5429 /* Work out the disk rotation on this stripe-set */
5430 div_u64_rem(stripe_nr, num_stripes, &rot);
5431
5432 /* Fill in the logical address of each stripe */
5433 tmp = stripe_nr * nr_data_stripes(map);
5434 for (i = 0; i < nr_data_stripes(map); i++)
5435 bbio->raid_map[(i+rot) % num_stripes] =
5436 em->start + (tmp + i) * map->stripe_len;
5437
5438 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5439 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5440 bbio->raid_map[(i+rot+1) % num_stripes] =
5441 RAID6_Q_STRIPE;
5442 }
5443
5444 if (rw & REQ_DISCARD) {
5445 u32 factor = 0;
5446 u32 sub_stripes = 0;
5447 u64 stripes_per_dev = 0;
5448 u32 remaining_stripes = 0;
5449 u32 last_stripe = 0;
5450
5451 if (map->type &
5452 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5453 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5454 sub_stripes = 1;
5455 else
5456 sub_stripes = map->sub_stripes;
5457
5458 factor = map->num_stripes / sub_stripes;
5459 stripes_per_dev = div_u64_rem(stripe_nr_end -
5460 stripe_nr_orig,
5461 factor,
5462 &remaining_stripes);
5463 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5464 last_stripe *= sub_stripes;
5465 }
5466
5467 for (i = 0; i < num_stripes; i++) {
5468 bbio->stripes[i].physical =
5469 map->stripes[stripe_index].physical +
5470 stripe_offset + stripe_nr * map->stripe_len;
5471 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5472
5473 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5474 BTRFS_BLOCK_GROUP_RAID10)) {
5475 bbio->stripes[i].length = stripes_per_dev *
5476 map->stripe_len;
5477
5478 if (i / sub_stripes < remaining_stripes)
5479 bbio->stripes[i].length +=
5480 map->stripe_len;
5481
5482 /*
5483 * Special for the first stripe and
5484 * the last stripe:
5485 *
5486 * |-------|...|-------|
5487 * |----------|
5488 * off end_off
5489 */
5490 if (i < sub_stripes)
5491 bbio->stripes[i].length -=
5492 stripe_offset;
5493
5494 if (stripe_index >= last_stripe &&
5495 stripe_index <= (last_stripe +
5496 sub_stripes - 1))
5497 bbio->stripes[i].length -=
5498 stripe_end_offset;
5499
5500 if (i == sub_stripes - 1)
5501 stripe_offset = 0;
5502 } else
5503 bbio->stripes[i].length = *length;
5504
5505 stripe_index++;
5506 if (stripe_index == map->num_stripes) {
5507 /* This could only happen for RAID0/10 */
5508 stripe_index = 0;
5509 stripe_nr++;
5510 }
5511 }
5512 } else {
5513 for (i = 0; i < num_stripes; i++) {
5514 bbio->stripes[i].physical =
5515 map->stripes[stripe_index].physical +
5516 stripe_offset +
5517 stripe_nr * map->stripe_len;
5518 bbio->stripes[i].dev =
5519 map->stripes[stripe_index].dev;
5520 stripe_index++;
5521 }
5522 }
5523
5524 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5525 max_errors = btrfs_chunk_max_errors(map);
5526
5527 if (bbio->raid_map)
5528 sort_parity_stripes(bbio, num_stripes);
5529
5530 tgtdev_indexes = 0;
5531 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5532 dev_replace->tgtdev != NULL) {
5533 int index_where_to_add;
5534 u64 srcdev_devid = dev_replace->srcdev->devid;
5535
5536 /*
5537 * duplicate the write operations while the dev replace
5538 * procedure is running. Since the copying of the old disk
5539 * to the new disk takes place at run time while the
5540 * filesystem is mounted writable, the regular write
5541 * operations to the old disk have to be duplicated to go
5542 * to the new disk as well.
5543 * Note that device->missing is handled by the caller, and
5544 * that the write to the old disk is already set up in the
5545 * stripes array.
5546 */
5547 index_where_to_add = num_stripes;
5548 for (i = 0; i < num_stripes; i++) {
5549 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5550 /* write to new disk, too */
5551 struct btrfs_bio_stripe *new =
5552 bbio->stripes + index_where_to_add;
5553 struct btrfs_bio_stripe *old =
5554 bbio->stripes + i;
5555
5556 new->physical = old->physical;
5557 new->length = old->length;
5558 new->dev = dev_replace->tgtdev;
5559 bbio->tgtdev_map[i] = index_where_to_add;
5560 index_where_to_add++;
5561 max_errors++;
5562 tgtdev_indexes++;
5563 }
5564 }
5565 num_stripes = index_where_to_add;
5566 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5567 dev_replace->tgtdev != NULL) {
5568 u64 srcdev_devid = dev_replace->srcdev->devid;
5569 int index_srcdev = 0;
5570 int found = 0;
5571 u64 physical_of_found = 0;
5572
5573 /*
5574 * During the dev-replace procedure, the target drive can
5575 * also be used to read data in case it is needed to repair
5576 * a corrupt block elsewhere. This is possible if the
5577 * requested area is left of the left cursor. In this area,
5578 * the target drive is a full copy of the source drive.
5579 */
5580 for (i = 0; i < num_stripes; i++) {
5581 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5582 /*
5583 * In case of DUP, in order to keep it
5584 * simple, only add the mirror with the
5585 * lowest physical address
5586 */
5587 if (found &&
5588 physical_of_found <=
5589 bbio->stripes[i].physical)
5590 continue;
5591 index_srcdev = i;
5592 found = 1;
5593 physical_of_found = bbio->stripes[i].physical;
5594 }
5595 }
5596 if (found) {
5597 if (physical_of_found + map->stripe_len <=
5598 dev_replace->cursor_left) {
5599 struct btrfs_bio_stripe *tgtdev_stripe =
5600 bbio->stripes + num_stripes;
5601
5602 tgtdev_stripe->physical = physical_of_found;
5603 tgtdev_stripe->length =
5604 bbio->stripes[index_srcdev].length;
5605 tgtdev_stripe->dev = dev_replace->tgtdev;
5606 bbio->tgtdev_map[index_srcdev] = num_stripes;
5607
5608 tgtdev_indexes++;
5609 num_stripes++;
5610 }
5611 }
5612 }
5613
5614 *bbio_ret = bbio;
5615 bbio->map_type = map->type;
5616 bbio->num_stripes = num_stripes;
5617 bbio->max_errors = max_errors;
5618 bbio->mirror_num = mirror_num;
5619 bbio->num_tgtdevs = tgtdev_indexes;
5620
5621 /*
5622 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5623 * mirror_num == num_stripes + 1 && dev_replace target drive is
5624 * available as a mirror
5625 */
5626 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5627 WARN_ON(num_stripes > 1);
5628 bbio->stripes[0].dev = dev_replace->tgtdev;
5629 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5630 bbio->mirror_num = map->num_stripes + 1;
5631 }
5632 out:
5633 if (dev_replace_is_ongoing)
5634 btrfs_dev_replace_unlock(dev_replace);
5635 free_extent_map(em);
5636 return ret;
5637 }
5638
5639 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5640 u64 logical, u64 *length,
5641 struct btrfs_bio **bbio_ret, int mirror_num)
5642 {
5643 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5644 mirror_num, 0);
5645 }
5646
5647 /* For Scrub/replace */
5648 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5649 u64 logical, u64 *length,
5650 struct btrfs_bio **bbio_ret, int mirror_num,
5651 int need_raid_map)
5652 {
5653 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5654 mirror_num, need_raid_map);
5655 }
5656
5657 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5658 u64 chunk_start, u64 physical, u64 devid,
5659 u64 **logical, int *naddrs, int *stripe_len)
5660 {
5661 struct extent_map_tree *em_tree = &map_tree->map_tree;
5662 struct extent_map *em;
5663 struct map_lookup *map;
5664 u64 *buf;
5665 u64 bytenr;
5666 u64 length;
5667 u64 stripe_nr;
5668 u64 rmap_len;
5669 int i, j, nr = 0;
5670
5671 read_lock(&em_tree->lock);
5672 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5673 read_unlock(&em_tree->lock);
5674
5675 if (!em) {
5676 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5677 chunk_start);
5678 return -EIO;
5679 }
5680
5681 if (em->start != chunk_start) {
5682 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5683 em->start, chunk_start);
5684 free_extent_map(em);
5685 return -EIO;
5686 }
5687 map = (struct map_lookup *)em->bdev;
5688
5689 length = em->len;
5690 rmap_len = map->stripe_len;
5691
5692 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5693 length = div_u64(length, map->num_stripes / map->sub_stripes);
5694 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5695 length = div_u64(length, map->num_stripes);
5696 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5697 length = div_u64(length, nr_data_stripes(map));
5698 rmap_len = map->stripe_len * nr_data_stripes(map);
5699 }
5700
5701 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5702 BUG_ON(!buf); /* -ENOMEM */
5703
5704 for (i = 0; i < map->num_stripes; i++) {
5705 if (devid && map->stripes[i].dev->devid != devid)
5706 continue;
5707 if (map->stripes[i].physical > physical ||
5708 map->stripes[i].physical + length <= physical)
5709 continue;
5710
5711 stripe_nr = physical - map->stripes[i].physical;
5712 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5713
5714 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5715 stripe_nr = stripe_nr * map->num_stripes + i;
5716 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5717 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5718 stripe_nr = stripe_nr * map->num_stripes + i;
5719 } /* else if RAID[56], multiply by nr_data_stripes().
5720 * Alternatively, just use rmap_len below instead of
5721 * map->stripe_len */
5722
5723 bytenr = chunk_start + stripe_nr * rmap_len;
5724 WARN_ON(nr >= map->num_stripes);
5725 for (j = 0; j < nr; j++) {
5726 if (buf[j] == bytenr)
5727 break;
5728 }
5729 if (j == nr) {
5730 WARN_ON(nr >= map->num_stripes);
5731 buf[nr++] = bytenr;
5732 }
5733 }
5734
5735 *logical = buf;
5736 *naddrs = nr;
5737 *stripe_len = rmap_len;
5738
5739 free_extent_map(em);
5740 return 0;
5741 }
5742
5743 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5744 {
5745 bio->bi_private = bbio->private;
5746 bio->bi_end_io = bbio->end_io;
5747 bio_endio(bio);
5748
5749 btrfs_put_bbio(bbio);
5750 }
5751
5752 static void btrfs_end_bio(struct bio *bio)
5753 {
5754 struct btrfs_bio *bbio = bio->bi_private;
5755 int is_orig_bio = 0;
5756
5757 if (bio->bi_error) {
5758 atomic_inc(&bbio->error);
5759 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5760 unsigned int stripe_index =
5761 btrfs_io_bio(bio)->stripe_index;
5762 struct btrfs_device *dev;
5763
5764 BUG_ON(stripe_index >= bbio->num_stripes);
5765 dev = bbio->stripes[stripe_index].dev;
5766 if (dev->bdev) {
5767 if (bio->bi_rw & WRITE)
5768 btrfs_dev_stat_inc(dev,
5769 BTRFS_DEV_STAT_WRITE_ERRS);
5770 else
5771 btrfs_dev_stat_inc(dev,
5772 BTRFS_DEV_STAT_READ_ERRS);
5773 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5774 btrfs_dev_stat_inc(dev,
5775 BTRFS_DEV_STAT_FLUSH_ERRS);
5776 btrfs_dev_stat_print_on_error(dev);
5777 }
5778 }
5779 }
5780
5781 if (bio == bbio->orig_bio)
5782 is_orig_bio = 1;
5783
5784 btrfs_bio_counter_dec(bbio->fs_info);
5785
5786 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5787 if (!is_orig_bio) {
5788 bio_put(bio);
5789 bio = bbio->orig_bio;
5790 }
5791
5792 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5793 /* only send an error to the higher layers if it is
5794 * beyond the tolerance of the btrfs bio
5795 */
5796 if (atomic_read(&bbio->error) > bbio->max_errors) {
5797 bio->bi_error = -EIO;
5798 } else {
5799 /*
5800 * this bio is actually up to date, we didn't
5801 * go over the max number of errors
5802 */
5803 bio->bi_error = 0;
5804 }
5805
5806 btrfs_end_bbio(bbio, bio);
5807 } else if (!is_orig_bio) {
5808 bio_put(bio);
5809 }
5810 }
5811
5812 /*
5813 * see run_scheduled_bios for a description of why bios are collected for
5814 * async submit.
5815 *
5816 * This will add one bio to the pending list for a device and make sure
5817 * the work struct is scheduled.
5818 */
5819 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5820 struct btrfs_device *device,
5821 int rw, struct bio *bio)
5822 {
5823 int should_queue = 1;
5824 struct btrfs_pending_bios *pending_bios;
5825
5826 if (device->missing || !device->bdev) {
5827 bio_io_error(bio);
5828 return;
5829 }
5830
5831 /* don't bother with additional async steps for reads, right now */
5832 if (!(rw & REQ_WRITE)) {
5833 bio_get(bio);
5834 btrfsic_submit_bio(rw, bio);
5835 bio_put(bio);
5836 return;
5837 }
5838
5839 /*
5840 * nr_async_bios allows us to reliably return congestion to the
5841 * higher layers. Otherwise, the async bio makes it appear we have
5842 * made progress against dirty pages when we've really just put it
5843 * on a queue for later
5844 */
5845 atomic_inc(&root->fs_info->nr_async_bios);
5846 WARN_ON(bio->bi_next);
5847 bio->bi_next = NULL;
5848 bio->bi_rw |= rw;
5849
5850 spin_lock(&device->io_lock);
5851 if (bio->bi_rw & REQ_SYNC)
5852 pending_bios = &device->pending_sync_bios;
5853 else
5854 pending_bios = &device->pending_bios;
5855
5856 if (pending_bios->tail)
5857 pending_bios->tail->bi_next = bio;
5858
5859 pending_bios->tail = bio;
5860 if (!pending_bios->head)
5861 pending_bios->head = bio;
5862 if (device->running_pending)
5863 should_queue = 0;
5864
5865 spin_unlock(&device->io_lock);
5866
5867 if (should_queue)
5868 btrfs_queue_work(root->fs_info->submit_workers,
5869 &device->work);
5870 }
5871
5872 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5873 struct bio *bio, u64 physical, int dev_nr,
5874 int rw, int async)
5875 {
5876 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5877
5878 bio->bi_private = bbio;
5879 btrfs_io_bio(bio)->stripe_index = dev_nr;
5880 bio->bi_end_io = btrfs_end_bio;
5881 bio->bi_iter.bi_sector = physical >> 9;
5882 #ifdef DEBUG
5883 {
5884 struct rcu_string *name;
5885
5886 rcu_read_lock();
5887 name = rcu_dereference(dev->name);
5888 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5889 "(%s id %llu), size=%u\n", rw,
5890 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5891 name->str, dev->devid, bio->bi_iter.bi_size);
5892 rcu_read_unlock();
5893 }
5894 #endif
5895 bio->bi_bdev = dev->bdev;
5896
5897 btrfs_bio_counter_inc_noblocked(root->fs_info);
5898
5899 if (async)
5900 btrfs_schedule_bio(root, dev, rw, bio);
5901 else
5902 btrfsic_submit_bio(rw, bio);
5903 }
5904
5905 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5906 {
5907 atomic_inc(&bbio->error);
5908 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5909 /* Shoud be the original bio. */
5910 WARN_ON(bio != bbio->orig_bio);
5911
5912 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5913 bio->bi_iter.bi_sector = logical >> 9;
5914 bio->bi_error = -EIO;
5915 btrfs_end_bbio(bbio, bio);
5916 }
5917 }
5918
5919 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5920 int mirror_num, int async_submit)
5921 {
5922 struct btrfs_device *dev;
5923 struct bio *first_bio = bio;
5924 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5925 u64 length = 0;
5926 u64 map_length;
5927 int ret;
5928 int dev_nr;
5929 int total_devs;
5930 struct btrfs_bio *bbio = NULL;
5931
5932 length = bio->bi_iter.bi_size;
5933 map_length = length;
5934
5935 btrfs_bio_counter_inc_blocked(root->fs_info);
5936 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5937 mirror_num, 1);
5938 if (ret) {
5939 btrfs_bio_counter_dec(root->fs_info);
5940 return ret;
5941 }
5942
5943 total_devs = bbio->num_stripes;
5944 bbio->orig_bio = first_bio;
5945 bbio->private = first_bio->bi_private;
5946 bbio->end_io = first_bio->bi_end_io;
5947 bbio->fs_info = root->fs_info;
5948 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5949
5950 if (bbio->raid_map) {
5951 /* In this case, map_length has been set to the length of
5952 a single stripe; not the whole write */
5953 if (rw & WRITE) {
5954 ret = raid56_parity_write(root, bio, bbio, map_length);
5955 } else {
5956 ret = raid56_parity_recover(root, bio, bbio, map_length,
5957 mirror_num, 1);
5958 }
5959
5960 btrfs_bio_counter_dec(root->fs_info);
5961 return ret;
5962 }
5963
5964 if (map_length < length) {
5965 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5966 logical, length, map_length);
5967 BUG();
5968 }
5969
5970 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
5971 dev = bbio->stripes[dev_nr].dev;
5972 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5973 bbio_error(bbio, first_bio, logical);
5974 continue;
5975 }
5976
5977 if (dev_nr < total_devs - 1) {
5978 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5979 BUG_ON(!bio); /* -ENOMEM */
5980 } else
5981 bio = first_bio;
5982
5983 submit_stripe_bio(root, bbio, bio,
5984 bbio->stripes[dev_nr].physical, dev_nr, rw,
5985 async_submit);
5986 }
5987 btrfs_bio_counter_dec(root->fs_info);
5988 return 0;
5989 }
5990
5991 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5992 u8 *uuid, u8 *fsid)
5993 {
5994 struct btrfs_device *device;
5995 struct btrfs_fs_devices *cur_devices;
5996
5997 cur_devices = fs_info->fs_devices;
5998 while (cur_devices) {
5999 if (!fsid ||
6000 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6001 device = __find_device(&cur_devices->devices,
6002 devid, uuid);
6003 if (device)
6004 return device;
6005 }
6006 cur_devices = cur_devices->seed;
6007 }
6008 return NULL;
6009 }
6010
6011 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6012 struct btrfs_fs_devices *fs_devices,
6013 u64 devid, u8 *dev_uuid)
6014 {
6015 struct btrfs_device *device;
6016
6017 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6018 if (IS_ERR(device))
6019 return NULL;
6020
6021 list_add(&device->dev_list, &fs_devices->devices);
6022 device->fs_devices = fs_devices;
6023 fs_devices->num_devices++;
6024
6025 device->missing = 1;
6026 fs_devices->missing_devices++;
6027
6028 return device;
6029 }
6030
6031 /**
6032 * btrfs_alloc_device - allocate struct btrfs_device
6033 * @fs_info: used only for generating a new devid, can be NULL if
6034 * devid is provided (i.e. @devid != NULL).
6035 * @devid: a pointer to devid for this device. If NULL a new devid
6036 * is generated.
6037 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6038 * is generated.
6039 *
6040 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6041 * on error. Returned struct is not linked onto any lists and can be
6042 * destroyed with kfree() right away.
6043 */
6044 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6045 const u64 *devid,
6046 const u8 *uuid)
6047 {
6048 struct btrfs_device *dev;
6049 u64 tmp;
6050
6051 if (WARN_ON(!devid && !fs_info))
6052 return ERR_PTR(-EINVAL);
6053
6054 dev = __alloc_device();
6055 if (IS_ERR(dev))
6056 return dev;
6057
6058 if (devid)
6059 tmp = *devid;
6060 else {
6061 int ret;
6062
6063 ret = find_next_devid(fs_info, &tmp);
6064 if (ret) {
6065 kfree(dev);
6066 return ERR_PTR(ret);
6067 }
6068 }
6069 dev->devid = tmp;
6070
6071 if (uuid)
6072 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6073 else
6074 generate_random_uuid(dev->uuid);
6075
6076 btrfs_init_work(&dev->work, btrfs_submit_helper,
6077 pending_bios_fn, NULL, NULL);
6078
6079 return dev;
6080 }
6081
6082 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6083 struct extent_buffer *leaf,
6084 struct btrfs_chunk *chunk)
6085 {
6086 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6087 struct map_lookup *map;
6088 struct extent_map *em;
6089 u64 logical;
6090 u64 length;
6091 u64 devid;
6092 u8 uuid[BTRFS_UUID_SIZE];
6093 int num_stripes;
6094 int ret;
6095 int i;
6096
6097 logical = key->offset;
6098 length = btrfs_chunk_length(leaf, chunk);
6099
6100 read_lock(&map_tree->map_tree.lock);
6101 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6102 read_unlock(&map_tree->map_tree.lock);
6103
6104 /* already mapped? */
6105 if (em && em->start <= logical && em->start + em->len > logical) {
6106 free_extent_map(em);
6107 return 0;
6108 } else if (em) {
6109 free_extent_map(em);
6110 }
6111
6112 em = alloc_extent_map();
6113 if (!em)
6114 return -ENOMEM;
6115 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6116 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6117 if (!map) {
6118 free_extent_map(em);
6119 return -ENOMEM;
6120 }
6121
6122 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6123 em->bdev = (struct block_device *)map;
6124 em->start = logical;
6125 em->len = length;
6126 em->orig_start = 0;
6127 em->block_start = 0;
6128 em->block_len = em->len;
6129
6130 map->num_stripes = num_stripes;
6131 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6132 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6133 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6134 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6135 map->type = btrfs_chunk_type(leaf, chunk);
6136 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6137 for (i = 0; i < num_stripes; i++) {
6138 map->stripes[i].physical =
6139 btrfs_stripe_offset_nr(leaf, chunk, i);
6140 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6141 read_extent_buffer(leaf, uuid, (unsigned long)
6142 btrfs_stripe_dev_uuid_nr(chunk, i),
6143 BTRFS_UUID_SIZE);
6144 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6145 uuid, NULL);
6146 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6147 free_extent_map(em);
6148 return -EIO;
6149 }
6150 if (!map->stripes[i].dev) {
6151 map->stripes[i].dev =
6152 add_missing_dev(root, root->fs_info->fs_devices,
6153 devid, uuid);
6154 if (!map->stripes[i].dev) {
6155 free_extent_map(em);
6156 return -EIO;
6157 }
6158 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6159 devid, uuid);
6160 }
6161 map->stripes[i].dev->in_fs_metadata = 1;
6162 }
6163
6164 write_lock(&map_tree->map_tree.lock);
6165 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6166 write_unlock(&map_tree->map_tree.lock);
6167 BUG_ON(ret); /* Tree corruption */
6168 free_extent_map(em);
6169
6170 return 0;
6171 }
6172
6173 static void fill_device_from_item(struct extent_buffer *leaf,
6174 struct btrfs_dev_item *dev_item,
6175 struct btrfs_device *device)
6176 {
6177 unsigned long ptr;
6178
6179 device->devid = btrfs_device_id(leaf, dev_item);
6180 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6181 device->total_bytes = device->disk_total_bytes;
6182 device->commit_total_bytes = device->disk_total_bytes;
6183 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6184 device->commit_bytes_used = device->bytes_used;
6185 device->type = btrfs_device_type(leaf, dev_item);
6186 device->io_align = btrfs_device_io_align(leaf, dev_item);
6187 device->io_width = btrfs_device_io_width(leaf, dev_item);
6188 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6189 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6190 device->is_tgtdev_for_dev_replace = 0;
6191
6192 ptr = btrfs_device_uuid(dev_item);
6193 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6194 }
6195
6196 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6197 u8 *fsid)
6198 {
6199 struct btrfs_fs_devices *fs_devices;
6200 int ret;
6201
6202 BUG_ON(!mutex_is_locked(&uuid_mutex));
6203
6204 fs_devices = root->fs_info->fs_devices->seed;
6205 while (fs_devices) {
6206 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6207 return fs_devices;
6208
6209 fs_devices = fs_devices->seed;
6210 }
6211
6212 fs_devices = find_fsid(fsid);
6213 if (!fs_devices) {
6214 if (!btrfs_test_opt(root, DEGRADED))
6215 return ERR_PTR(-ENOENT);
6216
6217 fs_devices = alloc_fs_devices(fsid);
6218 if (IS_ERR(fs_devices))
6219 return fs_devices;
6220
6221 fs_devices->seeding = 1;
6222 fs_devices->opened = 1;
6223 return fs_devices;
6224 }
6225
6226 fs_devices = clone_fs_devices(fs_devices);
6227 if (IS_ERR(fs_devices))
6228 return fs_devices;
6229
6230 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6231 root->fs_info->bdev_holder);
6232 if (ret) {
6233 free_fs_devices(fs_devices);
6234 fs_devices = ERR_PTR(ret);
6235 goto out;
6236 }
6237
6238 if (!fs_devices->seeding) {
6239 __btrfs_close_devices(fs_devices);
6240 free_fs_devices(fs_devices);
6241 fs_devices = ERR_PTR(-EINVAL);
6242 goto out;
6243 }
6244
6245 fs_devices->seed = root->fs_info->fs_devices->seed;
6246 root->fs_info->fs_devices->seed = fs_devices;
6247 out:
6248 return fs_devices;
6249 }
6250
6251 static int read_one_dev(struct btrfs_root *root,
6252 struct extent_buffer *leaf,
6253 struct btrfs_dev_item *dev_item)
6254 {
6255 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6256 struct btrfs_device *device;
6257 u64 devid;
6258 int ret;
6259 u8 fs_uuid[BTRFS_UUID_SIZE];
6260 u8 dev_uuid[BTRFS_UUID_SIZE];
6261
6262 devid = btrfs_device_id(leaf, dev_item);
6263 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6264 BTRFS_UUID_SIZE);
6265 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6266 BTRFS_UUID_SIZE);
6267
6268 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6269 fs_devices = open_seed_devices(root, fs_uuid);
6270 if (IS_ERR(fs_devices))
6271 return PTR_ERR(fs_devices);
6272 }
6273
6274 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6275 if (!device) {
6276 if (!btrfs_test_opt(root, DEGRADED))
6277 return -EIO;
6278
6279 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6280 if (!device)
6281 return -ENOMEM;
6282 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6283 devid, dev_uuid);
6284 } else {
6285 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6286 return -EIO;
6287
6288 if(!device->bdev && !device->missing) {
6289 /*
6290 * this happens when a device that was properly setup
6291 * in the device info lists suddenly goes bad.
6292 * device->bdev is NULL, and so we have to set
6293 * device->missing to one here
6294 */
6295 device->fs_devices->missing_devices++;
6296 device->missing = 1;
6297 }
6298
6299 /* Move the device to its own fs_devices */
6300 if (device->fs_devices != fs_devices) {
6301 ASSERT(device->missing);
6302
6303 list_move(&device->dev_list, &fs_devices->devices);
6304 device->fs_devices->num_devices--;
6305 fs_devices->num_devices++;
6306
6307 device->fs_devices->missing_devices--;
6308 fs_devices->missing_devices++;
6309
6310 device->fs_devices = fs_devices;
6311 }
6312 }
6313
6314 if (device->fs_devices != root->fs_info->fs_devices) {
6315 BUG_ON(device->writeable);
6316 if (device->generation !=
6317 btrfs_device_generation(leaf, dev_item))
6318 return -EINVAL;
6319 }
6320
6321 fill_device_from_item(leaf, dev_item, device);
6322 device->in_fs_metadata = 1;
6323 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6324 device->fs_devices->total_rw_bytes += device->total_bytes;
6325 spin_lock(&root->fs_info->free_chunk_lock);
6326 root->fs_info->free_chunk_space += device->total_bytes -
6327 device->bytes_used;
6328 spin_unlock(&root->fs_info->free_chunk_lock);
6329 }
6330 ret = 0;
6331 return ret;
6332 }
6333
6334 int btrfs_read_sys_array(struct btrfs_root *root)
6335 {
6336 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6337 struct extent_buffer *sb;
6338 struct btrfs_disk_key *disk_key;
6339 struct btrfs_chunk *chunk;
6340 u8 *array_ptr;
6341 unsigned long sb_array_offset;
6342 int ret = 0;
6343 u32 num_stripes;
6344 u32 array_size;
6345 u32 len = 0;
6346 u32 cur_offset;
6347 struct btrfs_key key;
6348
6349 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6350 /*
6351 * This will create extent buffer of nodesize, superblock size is
6352 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6353 * overallocate but we can keep it as-is, only the first page is used.
6354 */
6355 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6356 if (!sb)
6357 return -ENOMEM;
6358 btrfs_set_buffer_uptodate(sb);
6359 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6360 /*
6361 * The sb extent buffer is artifical and just used to read the system array.
6362 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6363 * pages up-to-date when the page is larger: extent does not cover the
6364 * whole page and consequently check_page_uptodate does not find all
6365 * the page's extents up-to-date (the hole beyond sb),
6366 * write_extent_buffer then triggers a WARN_ON.
6367 *
6368 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6369 * but sb spans only this function. Add an explicit SetPageUptodate call
6370 * to silence the warning eg. on PowerPC 64.
6371 */
6372 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6373 SetPageUptodate(sb->pages[0]);
6374
6375 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6376 array_size = btrfs_super_sys_array_size(super_copy);
6377
6378 array_ptr = super_copy->sys_chunk_array;
6379 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6380 cur_offset = 0;
6381
6382 while (cur_offset < array_size) {
6383 disk_key = (struct btrfs_disk_key *)array_ptr;
6384 len = sizeof(*disk_key);
6385 if (cur_offset + len > array_size)
6386 goto out_short_read;
6387
6388 btrfs_disk_key_to_cpu(&key, disk_key);
6389
6390 array_ptr += len;
6391 sb_array_offset += len;
6392 cur_offset += len;
6393
6394 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6395 chunk = (struct btrfs_chunk *)sb_array_offset;
6396 /*
6397 * At least one btrfs_chunk with one stripe must be
6398 * present, exact stripe count check comes afterwards
6399 */
6400 len = btrfs_chunk_item_size(1);
6401 if (cur_offset + len > array_size)
6402 goto out_short_read;
6403
6404 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6405 len = btrfs_chunk_item_size(num_stripes);
6406 if (cur_offset + len > array_size)
6407 goto out_short_read;
6408
6409 ret = read_one_chunk(root, &key, sb, chunk);
6410 if (ret)
6411 break;
6412 } else {
6413 ret = -EIO;
6414 break;
6415 }
6416 array_ptr += len;
6417 sb_array_offset += len;
6418 cur_offset += len;
6419 }
6420 free_extent_buffer(sb);
6421 return ret;
6422
6423 out_short_read:
6424 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6425 len, cur_offset);
6426 free_extent_buffer(sb);
6427 return -EIO;
6428 }
6429
6430 int btrfs_read_chunk_tree(struct btrfs_root *root)
6431 {
6432 struct btrfs_path *path;
6433 struct extent_buffer *leaf;
6434 struct btrfs_key key;
6435 struct btrfs_key found_key;
6436 int ret;
6437 int slot;
6438
6439 root = root->fs_info->chunk_root;
6440
6441 path = btrfs_alloc_path();
6442 if (!path)
6443 return -ENOMEM;
6444
6445 mutex_lock(&uuid_mutex);
6446 lock_chunks(root);
6447
6448 /*
6449 * Read all device items, and then all the chunk items. All
6450 * device items are found before any chunk item (their object id
6451 * is smaller than the lowest possible object id for a chunk
6452 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6453 */
6454 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6455 key.offset = 0;
6456 key.type = 0;
6457 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6458 if (ret < 0)
6459 goto error;
6460 while (1) {
6461 leaf = path->nodes[0];
6462 slot = path->slots[0];
6463 if (slot >= btrfs_header_nritems(leaf)) {
6464 ret = btrfs_next_leaf(root, path);
6465 if (ret == 0)
6466 continue;
6467 if (ret < 0)
6468 goto error;
6469 break;
6470 }
6471 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6472 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6473 struct btrfs_dev_item *dev_item;
6474 dev_item = btrfs_item_ptr(leaf, slot,
6475 struct btrfs_dev_item);
6476 ret = read_one_dev(root, leaf, dev_item);
6477 if (ret)
6478 goto error;
6479 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6480 struct btrfs_chunk *chunk;
6481 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6482 ret = read_one_chunk(root, &found_key, leaf, chunk);
6483 if (ret)
6484 goto error;
6485 }
6486 path->slots[0]++;
6487 }
6488 ret = 0;
6489 error:
6490 unlock_chunks(root);
6491 mutex_unlock(&uuid_mutex);
6492
6493 btrfs_free_path(path);
6494 return ret;
6495 }
6496
6497 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6498 {
6499 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6500 struct btrfs_device *device;
6501
6502 while (fs_devices) {
6503 mutex_lock(&fs_devices->device_list_mutex);
6504 list_for_each_entry(device, &fs_devices->devices, dev_list)
6505 device->dev_root = fs_info->dev_root;
6506 mutex_unlock(&fs_devices->device_list_mutex);
6507
6508 fs_devices = fs_devices->seed;
6509 }
6510 }
6511
6512 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6513 {
6514 int i;
6515
6516 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6517 btrfs_dev_stat_reset(dev, i);
6518 }
6519
6520 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6521 {
6522 struct btrfs_key key;
6523 struct btrfs_key found_key;
6524 struct btrfs_root *dev_root = fs_info->dev_root;
6525 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6526 struct extent_buffer *eb;
6527 int slot;
6528 int ret = 0;
6529 struct btrfs_device *device;
6530 struct btrfs_path *path = NULL;
6531 int i;
6532
6533 path = btrfs_alloc_path();
6534 if (!path) {
6535 ret = -ENOMEM;
6536 goto out;
6537 }
6538
6539 mutex_lock(&fs_devices->device_list_mutex);
6540 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6541 int item_size;
6542 struct btrfs_dev_stats_item *ptr;
6543
6544 key.objectid = 0;
6545 key.type = BTRFS_DEV_STATS_KEY;
6546 key.offset = device->devid;
6547 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6548 if (ret) {
6549 __btrfs_reset_dev_stats(device);
6550 device->dev_stats_valid = 1;
6551 btrfs_release_path(path);
6552 continue;
6553 }
6554 slot = path->slots[0];
6555 eb = path->nodes[0];
6556 btrfs_item_key_to_cpu(eb, &found_key, slot);
6557 item_size = btrfs_item_size_nr(eb, slot);
6558
6559 ptr = btrfs_item_ptr(eb, slot,
6560 struct btrfs_dev_stats_item);
6561
6562 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6563 if (item_size >= (1 + i) * sizeof(__le64))
6564 btrfs_dev_stat_set(device, i,
6565 btrfs_dev_stats_value(eb, ptr, i));
6566 else
6567 btrfs_dev_stat_reset(device, i);
6568 }
6569
6570 device->dev_stats_valid = 1;
6571 btrfs_dev_stat_print_on_load(device);
6572 btrfs_release_path(path);
6573 }
6574 mutex_unlock(&fs_devices->device_list_mutex);
6575
6576 out:
6577 btrfs_free_path(path);
6578 return ret < 0 ? ret : 0;
6579 }
6580
6581 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6582 struct btrfs_root *dev_root,
6583 struct btrfs_device *device)
6584 {
6585 struct btrfs_path *path;
6586 struct btrfs_key key;
6587 struct extent_buffer *eb;
6588 struct btrfs_dev_stats_item *ptr;
6589 int ret;
6590 int i;
6591
6592 key.objectid = 0;
6593 key.type = BTRFS_DEV_STATS_KEY;
6594 key.offset = device->devid;
6595
6596 path = btrfs_alloc_path();
6597 BUG_ON(!path);
6598 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6599 if (ret < 0) {
6600 printk_in_rcu(KERN_WARNING "BTRFS: "
6601 "error %d while searching for dev_stats item for device %s!\n",
6602 ret, rcu_str_deref(device->name));
6603 goto out;
6604 }
6605
6606 if (ret == 0 &&
6607 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6608 /* need to delete old one and insert a new one */
6609 ret = btrfs_del_item(trans, dev_root, path);
6610 if (ret != 0) {
6611 printk_in_rcu(KERN_WARNING "BTRFS: "
6612 "delete too small dev_stats item for device %s failed %d!\n",
6613 rcu_str_deref(device->name), ret);
6614 goto out;
6615 }
6616 ret = 1;
6617 }
6618
6619 if (ret == 1) {
6620 /* need to insert a new item */
6621 btrfs_release_path(path);
6622 ret = btrfs_insert_empty_item(trans, dev_root, path,
6623 &key, sizeof(*ptr));
6624 if (ret < 0) {
6625 printk_in_rcu(KERN_WARNING "BTRFS: "
6626 "insert dev_stats item for device %s failed %d!\n",
6627 rcu_str_deref(device->name), ret);
6628 goto out;
6629 }
6630 }
6631
6632 eb = path->nodes[0];
6633 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6634 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6635 btrfs_set_dev_stats_value(eb, ptr, i,
6636 btrfs_dev_stat_read(device, i));
6637 btrfs_mark_buffer_dirty(eb);
6638
6639 out:
6640 btrfs_free_path(path);
6641 return ret;
6642 }
6643
6644 /*
6645 * called from commit_transaction. Writes all changed device stats to disk.
6646 */
6647 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6648 struct btrfs_fs_info *fs_info)
6649 {
6650 struct btrfs_root *dev_root = fs_info->dev_root;
6651 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6652 struct btrfs_device *device;
6653 int stats_cnt;
6654 int ret = 0;
6655
6656 mutex_lock(&fs_devices->device_list_mutex);
6657 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6658 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6659 continue;
6660
6661 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6662 ret = update_dev_stat_item(trans, dev_root, device);
6663 if (!ret)
6664 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6665 }
6666 mutex_unlock(&fs_devices->device_list_mutex);
6667
6668 return ret;
6669 }
6670
6671 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6672 {
6673 btrfs_dev_stat_inc(dev, index);
6674 btrfs_dev_stat_print_on_error(dev);
6675 }
6676
6677 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6678 {
6679 if (!dev->dev_stats_valid)
6680 return;
6681 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6682 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6683 rcu_str_deref(dev->name),
6684 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6685 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6686 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6687 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6688 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6689 }
6690
6691 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6692 {
6693 int i;
6694
6695 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6696 if (btrfs_dev_stat_read(dev, i) != 0)
6697 break;
6698 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6699 return; /* all values == 0, suppress message */
6700
6701 printk_in_rcu(KERN_INFO "BTRFS: "
6702 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6703 rcu_str_deref(dev->name),
6704 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6705 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6706 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6707 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6708 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6709 }
6710
6711 int btrfs_get_dev_stats(struct btrfs_root *root,
6712 struct btrfs_ioctl_get_dev_stats *stats)
6713 {
6714 struct btrfs_device *dev;
6715 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6716 int i;
6717
6718 mutex_lock(&fs_devices->device_list_mutex);
6719 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6720 mutex_unlock(&fs_devices->device_list_mutex);
6721
6722 if (!dev) {
6723 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6724 return -ENODEV;
6725 } else if (!dev->dev_stats_valid) {
6726 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6727 return -ENODEV;
6728 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6729 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6730 if (stats->nr_items > i)
6731 stats->values[i] =
6732 btrfs_dev_stat_read_and_reset(dev, i);
6733 else
6734 btrfs_dev_stat_reset(dev, i);
6735 }
6736 } else {
6737 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6738 if (stats->nr_items > i)
6739 stats->values[i] = btrfs_dev_stat_read(dev, i);
6740 }
6741 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6742 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6743 return 0;
6744 }
6745
6746 int btrfs_scratch_superblock(struct btrfs_device *device)
6747 {
6748 struct buffer_head *bh;
6749 struct btrfs_super_block *disk_super;
6750
6751 bh = btrfs_read_dev_super(device->bdev);
6752 if (!bh)
6753 return -EINVAL;
6754 disk_super = (struct btrfs_super_block *)bh->b_data;
6755
6756 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6757 set_buffer_dirty(bh);
6758 sync_dirty_buffer(bh);
6759 brelse(bh);
6760
6761 return 0;
6762 }
6763
6764 /*
6765 * Update the size of all devices, which is used for writing out the
6766 * super blocks.
6767 */
6768 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6769 {
6770 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6771 struct btrfs_device *curr, *next;
6772
6773 if (list_empty(&fs_devices->resized_devices))
6774 return;
6775
6776 mutex_lock(&fs_devices->device_list_mutex);
6777 lock_chunks(fs_info->dev_root);
6778 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6779 resized_list) {
6780 list_del_init(&curr->resized_list);
6781 curr->commit_total_bytes = curr->disk_total_bytes;
6782 }
6783 unlock_chunks(fs_info->dev_root);
6784 mutex_unlock(&fs_devices->device_list_mutex);
6785 }
6786
6787 /* Must be invoked during the transaction commit */
6788 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6789 struct btrfs_transaction *transaction)
6790 {
6791 struct extent_map *em;
6792 struct map_lookup *map;
6793 struct btrfs_device *dev;
6794 int i;
6795
6796 if (list_empty(&transaction->pending_chunks))
6797 return;
6798
6799 /* In order to kick the device replace finish process */
6800 lock_chunks(root);
6801 list_for_each_entry(em, &transaction->pending_chunks, list) {
6802 map = (struct map_lookup *)em->bdev;
6803
6804 for (i = 0; i < map->num_stripes; i++) {
6805 dev = map->stripes[i].dev;
6806 dev->commit_bytes_used = dev->bytes_used;
6807 }
6808 }
6809 unlock_chunks(root);
6810 }
6811
6812 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6813 {
6814 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6815 while (fs_devices) {
6816 fs_devices->fs_info = fs_info;
6817 fs_devices = fs_devices->seed;
6818 }
6819 }
6820
6821 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6822 {
6823 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6824 while (fs_devices) {
6825 fs_devices->fs_info = NULL;
6826 fs_devices = fs_devices->seed;
6827 }
6828 }
Ubuntu Artful kernel mirror