]> 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: iterate over unused chunk space in FITRIM
[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 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
349 waitqueue_active(&fs_info->async_submit_wait))
350 wake_up(&fs_info->async_submit_wait);
351
352 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
353
354 /*
355 * if we're doing the sync list, record that our
356 * plug has some sync requests on it
357 *
358 * If we're doing the regular list and there are
359 * sync requests sitting around, unplug before
360 * we add more
361 */
362 if (pending_bios == &device->pending_sync_bios) {
363 sync_pending = 1;
364 } else if (sync_pending) {
365 blk_finish_plug(&plug);
366 blk_start_plug(&plug);
367 sync_pending = 0;
368 }
369
370 btrfsic_submit_bio(cur->bi_rw, cur);
371 num_run++;
372 batch_run++;
373
374 cond_resched();
375
376 /*
377 * we made progress, there is more work to do and the bdi
378 * is now congested. Back off and let other work structs
379 * run instead
380 */
381 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
382 fs_info->fs_devices->open_devices > 1) {
383 struct io_context *ioc;
384
385 ioc = current->io_context;
386
387 /*
388 * the main goal here is that we don't want to
389 * block if we're going to be able to submit
390 * more requests without blocking.
391 *
392 * This code does two great things, it pokes into
393 * the elevator code from a filesystem _and_
394 * it makes assumptions about how batching works.
395 */
396 if (ioc && ioc->nr_batch_requests > 0 &&
397 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
398 (last_waited == 0 ||
399 ioc->last_waited == last_waited)) {
400 /*
401 * we want to go through our batch of
402 * requests and stop. So, we copy out
403 * the ioc->last_waited time and test
404 * against it before looping
405 */
406 last_waited = ioc->last_waited;
407 cond_resched();
408 continue;
409 }
410 spin_lock(&device->io_lock);
411 requeue_list(pending_bios, pending, tail);
412 device->running_pending = 1;
413
414 spin_unlock(&device->io_lock);
415 btrfs_queue_work(fs_info->submit_workers,
416 &device->work);
417 goto done;
418 }
419 /* unplug every 64 requests just for good measure */
420 if (batch_run % 64 == 0) {
421 blk_finish_plug(&plug);
422 blk_start_plug(&plug);
423 sync_pending = 0;
424 }
425 }
426
427 cond_resched();
428 if (again)
429 goto loop;
430
431 spin_lock(&device->io_lock);
432 if (device->pending_bios.head || device->pending_sync_bios.head)
433 goto loop_lock;
434 spin_unlock(&device->io_lock);
435
436 done:
437 blk_finish_plug(&plug);
438 }
439
440 static void pending_bios_fn(struct btrfs_work *work)
441 {
442 struct btrfs_device *device;
443
444 device = container_of(work, struct btrfs_device, work);
445 run_scheduled_bios(device);
446 }
447
448
449 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
450 {
451 struct btrfs_fs_devices *fs_devs;
452 struct btrfs_device *dev;
453
454 if (!cur_dev->name)
455 return;
456
457 list_for_each_entry(fs_devs, &fs_uuids, list) {
458 int del = 1;
459
460 if (fs_devs->opened)
461 continue;
462 if (fs_devs->seeding)
463 continue;
464
465 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
466
467 if (dev == cur_dev)
468 continue;
469 if (!dev->name)
470 continue;
471
472 /*
473 * Todo: This won't be enough. What if the same device
474 * comes back (with new uuid and) with its mapper path?
475 * But for now, this does help as mostly an admin will
476 * either use mapper or non mapper path throughout.
477 */
478 rcu_read_lock();
479 del = strcmp(rcu_str_deref(dev->name),
480 rcu_str_deref(cur_dev->name));
481 rcu_read_unlock();
482 if (!del)
483 break;
484 }
485
486 if (!del) {
487 /* delete the stale device */
488 if (fs_devs->num_devices == 1) {
489 btrfs_sysfs_remove_fsid(fs_devs);
490 list_del(&fs_devs->list);
491 free_fs_devices(fs_devs);
492 } else {
493 fs_devs->num_devices--;
494 list_del(&dev->dev_list);
495 rcu_string_free(dev->name);
496 kfree(dev);
497 }
498 break;
499 }
500 }
501 }
502
503 /*
504 * Add new device to list of registered devices
505 *
506 * Returns:
507 * 1 - first time device is seen
508 * 0 - device already known
509 * < 0 - error
510 */
511 static noinline int device_list_add(const char *path,
512 struct btrfs_super_block *disk_super,
513 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
514 {
515 struct btrfs_device *device;
516 struct btrfs_fs_devices *fs_devices;
517 struct rcu_string *name;
518 int ret = 0;
519 u64 found_transid = btrfs_super_generation(disk_super);
520
521 fs_devices = find_fsid(disk_super->fsid);
522 if (!fs_devices) {
523 fs_devices = alloc_fs_devices(disk_super->fsid);
524 if (IS_ERR(fs_devices))
525 return PTR_ERR(fs_devices);
526
527 list_add(&fs_devices->list, &fs_uuids);
528
529 device = NULL;
530 } else {
531 device = __find_device(&fs_devices->devices, devid,
532 disk_super->dev_item.uuid);
533 }
534
535 if (!device) {
536 if (fs_devices->opened)
537 return -EBUSY;
538
539 device = btrfs_alloc_device(NULL, &devid,
540 disk_super->dev_item.uuid);
541 if (IS_ERR(device)) {
542 /* we can safely leave the fs_devices entry around */
543 return PTR_ERR(device);
544 }
545
546 name = rcu_string_strdup(path, GFP_NOFS);
547 if (!name) {
548 kfree(device);
549 return -ENOMEM;
550 }
551 rcu_assign_pointer(device->name, name);
552
553 mutex_lock(&fs_devices->device_list_mutex);
554 list_add_rcu(&device->dev_list, &fs_devices->devices);
555 fs_devices->num_devices++;
556 mutex_unlock(&fs_devices->device_list_mutex);
557
558 ret = 1;
559 device->fs_devices = fs_devices;
560 } else if (!device->name || strcmp(device->name->str, path)) {
561 /*
562 * When FS is already mounted.
563 * 1. If you are here and if the device->name is NULL that
564 * means this device was missing at time of FS mount.
565 * 2. If you are here and if the device->name is different
566 * from 'path' that means either
567 * a. The same device disappeared and reappeared with
568 * different name. or
569 * b. The missing-disk-which-was-replaced, has
570 * reappeared now.
571 *
572 * We must allow 1 and 2a above. But 2b would be a spurious
573 * and unintentional.
574 *
575 * Further in case of 1 and 2a above, the disk at 'path'
576 * would have missed some transaction when it was away and
577 * in case of 2a the stale bdev has to be updated as well.
578 * 2b must not be allowed at all time.
579 */
580
581 /*
582 * For now, we do allow update to btrfs_fs_device through the
583 * btrfs dev scan cli after FS has been mounted. We're still
584 * tracking a problem where systems fail mount by subvolume id
585 * when we reject replacement on a mounted FS.
586 */
587 if (!fs_devices->opened && found_transid < device->generation) {
588 /*
589 * That is if the FS is _not_ mounted and if you
590 * are here, that means there is more than one
591 * disk with same uuid and devid.We keep the one
592 * with larger generation number or the last-in if
593 * generation are equal.
594 */
595 return -EEXIST;
596 }
597
598 name = rcu_string_strdup(path, GFP_NOFS);
599 if (!name)
600 return -ENOMEM;
601 rcu_string_free(device->name);
602 rcu_assign_pointer(device->name, name);
603 if (device->missing) {
604 fs_devices->missing_devices--;
605 device->missing = 0;
606 }
607 }
608
609 /*
610 * Unmount does not free the btrfs_device struct but would zero
611 * generation along with most of the other members. So just update
612 * it back. We need it to pick the disk with largest generation
613 * (as above).
614 */
615 if (!fs_devices->opened)
616 device->generation = found_transid;
617
618 /*
619 * if there is new btrfs on an already registered device,
620 * then remove the stale device entry.
621 */
622 btrfs_free_stale_device(device);
623
624 *fs_devices_ret = fs_devices;
625
626 return ret;
627 }
628
629 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
630 {
631 struct btrfs_fs_devices *fs_devices;
632 struct btrfs_device *device;
633 struct btrfs_device *orig_dev;
634
635 fs_devices = alloc_fs_devices(orig->fsid);
636 if (IS_ERR(fs_devices))
637 return fs_devices;
638
639 mutex_lock(&orig->device_list_mutex);
640 fs_devices->total_devices = orig->total_devices;
641
642 /* We have held the volume lock, it is safe to get the devices. */
643 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
644 struct rcu_string *name;
645
646 device = btrfs_alloc_device(NULL, &orig_dev->devid,
647 orig_dev->uuid);
648 if (IS_ERR(device))
649 goto error;
650
651 /*
652 * This is ok to do without rcu read locked because we hold the
653 * uuid mutex so nothing we touch in here is going to disappear.
654 */
655 if (orig_dev->name) {
656 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
657 if (!name) {
658 kfree(device);
659 goto error;
660 }
661 rcu_assign_pointer(device->name, name);
662 }
663
664 list_add(&device->dev_list, &fs_devices->devices);
665 device->fs_devices = fs_devices;
666 fs_devices->num_devices++;
667 }
668 mutex_unlock(&orig->device_list_mutex);
669 return fs_devices;
670 error:
671 mutex_unlock(&orig->device_list_mutex);
672 free_fs_devices(fs_devices);
673 return ERR_PTR(-ENOMEM);
674 }
675
676 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
677 {
678 struct btrfs_device *device, *next;
679 struct btrfs_device *latest_dev = NULL;
680
681 mutex_lock(&uuid_mutex);
682 again:
683 /* This is the initialized path, it is safe to release the devices. */
684 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
685 if (device->in_fs_metadata) {
686 if (!device->is_tgtdev_for_dev_replace &&
687 (!latest_dev ||
688 device->generation > latest_dev->generation)) {
689 latest_dev = device;
690 }
691 continue;
692 }
693
694 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
695 /*
696 * In the first step, keep the device which has
697 * the correct fsid and the devid that is used
698 * for the dev_replace procedure.
699 * In the second step, the dev_replace state is
700 * read from the device tree and it is known
701 * whether the procedure is really active or
702 * not, which means whether this device is
703 * used or whether it should be removed.
704 */
705 if (step == 0 || device->is_tgtdev_for_dev_replace) {
706 continue;
707 }
708 }
709 if (device->bdev) {
710 blkdev_put(device->bdev, device->mode);
711 device->bdev = NULL;
712 fs_devices->open_devices--;
713 }
714 if (device->writeable) {
715 list_del_init(&device->dev_alloc_list);
716 device->writeable = 0;
717 if (!device->is_tgtdev_for_dev_replace)
718 fs_devices->rw_devices--;
719 }
720 list_del_init(&device->dev_list);
721 fs_devices->num_devices--;
722 rcu_string_free(device->name);
723 kfree(device);
724 }
725
726 if (fs_devices->seed) {
727 fs_devices = fs_devices->seed;
728 goto again;
729 }
730
731 fs_devices->latest_bdev = latest_dev->bdev;
732
733 mutex_unlock(&uuid_mutex);
734 }
735
736 static void __free_device(struct work_struct *work)
737 {
738 struct btrfs_device *device;
739
740 device = container_of(work, struct btrfs_device, rcu_work);
741
742 if (device->bdev)
743 blkdev_put(device->bdev, device->mode);
744
745 rcu_string_free(device->name);
746 kfree(device);
747 }
748
749 static void free_device(struct rcu_head *head)
750 {
751 struct btrfs_device *device;
752
753 device = container_of(head, struct btrfs_device, rcu);
754
755 INIT_WORK(&device->rcu_work, __free_device);
756 schedule_work(&device->rcu_work);
757 }
758
759 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
760 {
761 struct btrfs_device *device, *tmp;
762
763 if (--fs_devices->opened > 0)
764 return 0;
765
766 mutex_lock(&fs_devices->device_list_mutex);
767 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
768 struct btrfs_device *new_device;
769 struct rcu_string *name;
770
771 if (device->bdev)
772 fs_devices->open_devices--;
773
774 if (device->writeable &&
775 device->devid != BTRFS_DEV_REPLACE_DEVID) {
776 list_del_init(&device->dev_alloc_list);
777 fs_devices->rw_devices--;
778 }
779
780 if (device->missing)
781 fs_devices->missing_devices--;
782
783 new_device = btrfs_alloc_device(NULL, &device->devid,
784 device->uuid);
785 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
786
787 /* Safe because we are under uuid_mutex */
788 if (device->name) {
789 name = rcu_string_strdup(device->name->str, GFP_NOFS);
790 BUG_ON(!name); /* -ENOMEM */
791 rcu_assign_pointer(new_device->name, name);
792 }
793
794 list_replace_rcu(&device->dev_list, &new_device->dev_list);
795 new_device->fs_devices = device->fs_devices;
796
797 call_rcu(&device->rcu, free_device);
798 }
799 mutex_unlock(&fs_devices->device_list_mutex);
800
801 WARN_ON(fs_devices->open_devices);
802 WARN_ON(fs_devices->rw_devices);
803 fs_devices->opened = 0;
804 fs_devices->seeding = 0;
805
806 return 0;
807 }
808
809 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
810 {
811 struct btrfs_fs_devices *seed_devices = NULL;
812 int ret;
813
814 mutex_lock(&uuid_mutex);
815 ret = __btrfs_close_devices(fs_devices);
816 if (!fs_devices->opened) {
817 seed_devices = fs_devices->seed;
818 fs_devices->seed = NULL;
819 }
820 mutex_unlock(&uuid_mutex);
821
822 while (seed_devices) {
823 fs_devices = seed_devices;
824 seed_devices = fs_devices->seed;
825 __btrfs_close_devices(fs_devices);
826 free_fs_devices(fs_devices);
827 }
828 /*
829 * Wait for rcu kworkers under __btrfs_close_devices
830 * to finish all blkdev_puts so device is really
831 * free when umount is done.
832 */
833 rcu_barrier();
834 return ret;
835 }
836
837 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
838 fmode_t flags, void *holder)
839 {
840 struct request_queue *q;
841 struct block_device *bdev;
842 struct list_head *head = &fs_devices->devices;
843 struct btrfs_device *device;
844 struct btrfs_device *latest_dev = NULL;
845 struct buffer_head *bh;
846 struct btrfs_super_block *disk_super;
847 u64 devid;
848 int seeding = 1;
849 int ret = 0;
850
851 flags |= FMODE_EXCL;
852
853 list_for_each_entry(device, head, dev_list) {
854 if (device->bdev)
855 continue;
856 if (!device->name)
857 continue;
858
859 /* Just open everything we can; ignore failures here */
860 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
861 &bdev, &bh))
862 continue;
863
864 disk_super = (struct btrfs_super_block *)bh->b_data;
865 devid = btrfs_stack_device_id(&disk_super->dev_item);
866 if (devid != device->devid)
867 goto error_brelse;
868
869 if (memcmp(device->uuid, disk_super->dev_item.uuid,
870 BTRFS_UUID_SIZE))
871 goto error_brelse;
872
873 device->generation = btrfs_super_generation(disk_super);
874 if (!latest_dev ||
875 device->generation > latest_dev->generation)
876 latest_dev = device;
877
878 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
879 device->writeable = 0;
880 } else {
881 device->writeable = !bdev_read_only(bdev);
882 seeding = 0;
883 }
884
885 q = bdev_get_queue(bdev);
886 if (blk_queue_discard(q))
887 device->can_discard = 1;
888
889 device->bdev = bdev;
890 device->in_fs_metadata = 0;
891 device->mode = flags;
892
893 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
894 fs_devices->rotating = 1;
895
896 fs_devices->open_devices++;
897 if (device->writeable &&
898 device->devid != BTRFS_DEV_REPLACE_DEVID) {
899 fs_devices->rw_devices++;
900 list_add(&device->dev_alloc_list,
901 &fs_devices->alloc_list);
902 }
903 brelse(bh);
904 continue;
905
906 error_brelse:
907 brelse(bh);
908 blkdev_put(bdev, flags);
909 continue;
910 }
911 if (fs_devices->open_devices == 0) {
912 ret = -EINVAL;
913 goto out;
914 }
915 fs_devices->seeding = seeding;
916 fs_devices->opened = 1;
917 fs_devices->latest_bdev = latest_dev->bdev;
918 fs_devices->total_rw_bytes = 0;
919 out:
920 return ret;
921 }
922
923 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
924 fmode_t flags, void *holder)
925 {
926 int ret;
927
928 mutex_lock(&uuid_mutex);
929 if (fs_devices->opened) {
930 fs_devices->opened++;
931 ret = 0;
932 } else {
933 ret = __btrfs_open_devices(fs_devices, flags, holder);
934 }
935 mutex_unlock(&uuid_mutex);
936 return ret;
937 }
938
939 /*
940 * Look for a btrfs signature on a device. This may be called out of the mount path
941 * and we are not allowed to call set_blocksize during the scan. The superblock
942 * is read via pagecache
943 */
944 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
945 struct btrfs_fs_devices **fs_devices_ret)
946 {
947 struct btrfs_super_block *disk_super;
948 struct block_device *bdev;
949 struct page *page;
950 void *p;
951 int ret = -EINVAL;
952 u64 devid;
953 u64 transid;
954 u64 total_devices;
955 u64 bytenr;
956 pgoff_t index;
957
958 /*
959 * we would like to check all the supers, but that would make
960 * a btrfs mount succeed after a mkfs from a different FS.
961 * So, we need to add a special mount option to scan for
962 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
963 */
964 bytenr = btrfs_sb_offset(0);
965 flags |= FMODE_EXCL;
966 mutex_lock(&uuid_mutex);
967
968 bdev = blkdev_get_by_path(path, flags, holder);
969
970 if (IS_ERR(bdev)) {
971 ret = PTR_ERR(bdev);
972 goto error;
973 }
974
975 /* make sure our super fits in the device */
976 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
977 goto error_bdev_put;
978
979 /* make sure our super fits in the page */
980 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
981 goto error_bdev_put;
982
983 /* make sure our super doesn't straddle pages on disk */
984 index = bytenr >> PAGE_CACHE_SHIFT;
985 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
986 goto error_bdev_put;
987
988 /* pull in the page with our super */
989 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
990 index, GFP_NOFS);
991
992 if (IS_ERR_OR_NULL(page))
993 goto error_bdev_put;
994
995 p = kmap(page);
996
997 /* align our pointer to the offset of the super block */
998 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
999
1000 if (btrfs_super_bytenr(disk_super) != bytenr ||
1001 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1002 goto error_unmap;
1003
1004 devid = btrfs_stack_device_id(&disk_super->dev_item);
1005 transid = btrfs_super_generation(disk_super);
1006 total_devices = btrfs_super_num_devices(disk_super);
1007
1008 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1009 if (ret > 0) {
1010 if (disk_super->label[0]) {
1011 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1012 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1013 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1014 } else {
1015 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1016 }
1017
1018 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1019 ret = 0;
1020 }
1021 if (!ret && fs_devices_ret)
1022 (*fs_devices_ret)->total_devices = total_devices;
1023
1024 error_unmap:
1025 kunmap(page);
1026 page_cache_release(page);
1027
1028 error_bdev_put:
1029 blkdev_put(bdev, flags);
1030 error:
1031 mutex_unlock(&uuid_mutex);
1032 return ret;
1033 }
1034
1035 /* helper to account the used device space in the range */
1036 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1037 u64 end, u64 *length)
1038 {
1039 struct btrfs_key key;
1040 struct btrfs_root *root = device->dev_root;
1041 struct btrfs_dev_extent *dev_extent;
1042 struct btrfs_path *path;
1043 u64 extent_end;
1044 int ret;
1045 int slot;
1046 struct extent_buffer *l;
1047
1048 *length = 0;
1049
1050 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1051 return 0;
1052
1053 path = btrfs_alloc_path();
1054 if (!path)
1055 return -ENOMEM;
1056 path->reada = 2;
1057
1058 key.objectid = device->devid;
1059 key.offset = start;
1060 key.type = BTRFS_DEV_EXTENT_KEY;
1061
1062 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1063 if (ret < 0)
1064 goto out;
1065 if (ret > 0) {
1066 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1067 if (ret < 0)
1068 goto out;
1069 }
1070
1071 while (1) {
1072 l = path->nodes[0];
1073 slot = path->slots[0];
1074 if (slot >= btrfs_header_nritems(l)) {
1075 ret = btrfs_next_leaf(root, path);
1076 if (ret == 0)
1077 continue;
1078 if (ret < 0)
1079 goto out;
1080
1081 break;
1082 }
1083 btrfs_item_key_to_cpu(l, &key, slot);
1084
1085 if (key.objectid < device->devid)
1086 goto next;
1087
1088 if (key.objectid > device->devid)
1089 break;
1090
1091 if (key.type != BTRFS_DEV_EXTENT_KEY)
1092 goto next;
1093
1094 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1095 extent_end = key.offset + btrfs_dev_extent_length(l,
1096 dev_extent);
1097 if (key.offset <= start && extent_end > end) {
1098 *length = end - start + 1;
1099 break;
1100 } else if (key.offset <= start && extent_end > start)
1101 *length += extent_end - start;
1102 else if (key.offset > start && extent_end <= end)
1103 *length += extent_end - key.offset;
1104 else if (key.offset > start && key.offset <= end) {
1105 *length += end - key.offset + 1;
1106 break;
1107 } else if (key.offset > end)
1108 break;
1109
1110 next:
1111 path->slots[0]++;
1112 }
1113 ret = 0;
1114 out:
1115 btrfs_free_path(path);
1116 return ret;
1117 }
1118
1119 static int contains_pending_extent(struct btrfs_transaction *transaction,
1120 struct btrfs_device *device,
1121 u64 *start, u64 len)
1122 {
1123 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1124 struct extent_map *em;
1125 struct list_head *search_list = &fs_info->pinned_chunks;
1126 int ret = 0;
1127 u64 physical_start = *start;
1128
1129 if (transaction)
1130 search_list = &transaction->pending_chunks;
1131 again:
1132 list_for_each_entry(em, search_list, list) {
1133 struct map_lookup *map;
1134 int i;
1135
1136 map = (struct map_lookup *)em->bdev;
1137 for (i = 0; i < map->num_stripes; i++) {
1138 u64 end;
1139
1140 if (map->stripes[i].dev != device)
1141 continue;
1142 if (map->stripes[i].physical >= physical_start + len ||
1143 map->stripes[i].physical + em->orig_block_len <=
1144 physical_start)
1145 continue;
1146 /*
1147 * Make sure that while processing the pinned list we do
1148 * not override our *start with a lower value, because
1149 * we can have pinned chunks that fall within this
1150 * device hole and that have lower physical addresses
1151 * than the pending chunks we processed before. If we
1152 * do not take this special care we can end up getting
1153 * 2 pending chunks that start at the same physical
1154 * device offsets because the end offset of a pinned
1155 * chunk can be equal to the start offset of some
1156 * pending chunk.
1157 */
1158 end = map->stripes[i].physical + em->orig_block_len;
1159 if (end > *start) {
1160 *start = end;
1161 ret = 1;
1162 }
1163 }
1164 }
1165 if (search_list != &fs_info->pinned_chunks) {
1166 search_list = &fs_info->pinned_chunks;
1167 goto again;
1168 }
1169
1170 return ret;
1171 }
1172
1173
1174 /*
1175 * find_free_dev_extent_start - find free space in the specified device
1176 * @device: the device which we search the free space in
1177 * @num_bytes: the size of the free space that we need
1178 * @search_start: the position from which to begin the search
1179 * @start: store the start of the free space.
1180 * @len: the size of the free space. that we find, or the size
1181 * of the max free space if we don't find suitable free space
1182 *
1183 * this uses a pretty simple search, the expectation is that it is
1184 * called very infrequently and that a given device has a small number
1185 * of extents
1186 *
1187 * @start is used to store the start of the free space if we find. But if we
1188 * don't find suitable free space, it will be used to store the start position
1189 * of the max free space.
1190 *
1191 * @len is used to store the size of the free space that we find.
1192 * But if we don't find suitable free space, it is used to store the size of
1193 * the max free space.
1194 */
1195 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1196 struct btrfs_device *device, u64 num_bytes,
1197 u64 search_start, u64 *start, u64 *len)
1198 {
1199 struct btrfs_key key;
1200 struct btrfs_root *root = device->dev_root;
1201 struct btrfs_dev_extent *dev_extent;
1202 struct btrfs_path *path;
1203 u64 hole_size;
1204 u64 max_hole_start;
1205 u64 max_hole_size;
1206 u64 extent_end;
1207 u64 search_end = device->total_bytes;
1208 int ret;
1209 int slot;
1210 struct extent_buffer *l;
1211
1212 path = btrfs_alloc_path();
1213 if (!path)
1214 return -ENOMEM;
1215
1216 max_hole_start = search_start;
1217 max_hole_size = 0;
1218
1219 again:
1220 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1221 ret = -ENOSPC;
1222 goto out;
1223 }
1224
1225 path->reada = 2;
1226 path->search_commit_root = 1;
1227 path->skip_locking = 1;
1228
1229 key.objectid = device->devid;
1230 key.offset = search_start;
1231 key.type = BTRFS_DEV_EXTENT_KEY;
1232
1233 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1234 if (ret < 0)
1235 goto out;
1236 if (ret > 0) {
1237 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1238 if (ret < 0)
1239 goto out;
1240 }
1241
1242 while (1) {
1243 l = path->nodes[0];
1244 slot = path->slots[0];
1245 if (slot >= btrfs_header_nritems(l)) {
1246 ret = btrfs_next_leaf(root, path);
1247 if (ret == 0)
1248 continue;
1249 if (ret < 0)
1250 goto out;
1251
1252 break;
1253 }
1254 btrfs_item_key_to_cpu(l, &key, slot);
1255
1256 if (key.objectid < device->devid)
1257 goto next;
1258
1259 if (key.objectid > device->devid)
1260 break;
1261
1262 if (key.type != BTRFS_DEV_EXTENT_KEY)
1263 goto next;
1264
1265 if (key.offset > search_start) {
1266 hole_size = key.offset - search_start;
1267
1268 /*
1269 * Have to check before we set max_hole_start, otherwise
1270 * we could end up sending back this offset anyway.
1271 */
1272 if (contains_pending_extent(transaction, device,
1273 &search_start,
1274 hole_size)) {
1275 if (key.offset >= search_start) {
1276 hole_size = key.offset - search_start;
1277 } else {
1278 WARN_ON_ONCE(1);
1279 hole_size = 0;
1280 }
1281 }
1282
1283 if (hole_size > max_hole_size) {
1284 max_hole_start = search_start;
1285 max_hole_size = hole_size;
1286 }
1287
1288 /*
1289 * If this free space is greater than which we need,
1290 * it must be the max free space that we have found
1291 * until now, so max_hole_start must point to the start
1292 * of this free space and the length of this free space
1293 * is stored in max_hole_size. Thus, we return
1294 * max_hole_start and max_hole_size and go back to the
1295 * caller.
1296 */
1297 if (hole_size >= num_bytes) {
1298 ret = 0;
1299 goto out;
1300 }
1301 }
1302
1303 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1304 extent_end = key.offset + btrfs_dev_extent_length(l,
1305 dev_extent);
1306 if (extent_end > search_start)
1307 search_start = extent_end;
1308 next:
1309 path->slots[0]++;
1310 cond_resched();
1311 }
1312
1313 /*
1314 * At this point, search_start should be the end of
1315 * allocated dev extents, and when shrinking the device,
1316 * search_end may be smaller than search_start.
1317 */
1318 if (search_end > search_start) {
1319 hole_size = search_end - search_start;
1320
1321 if (contains_pending_extent(transaction, device, &search_start,
1322 hole_size)) {
1323 btrfs_release_path(path);
1324 goto again;
1325 }
1326
1327 if (hole_size > max_hole_size) {
1328 max_hole_start = search_start;
1329 max_hole_size = hole_size;
1330 }
1331 }
1332
1333 /* See above. */
1334 if (max_hole_size < num_bytes)
1335 ret = -ENOSPC;
1336 else
1337 ret = 0;
1338
1339 out:
1340 btrfs_free_path(path);
1341 *start = max_hole_start;
1342 if (len)
1343 *len = max_hole_size;
1344 return ret;
1345 }
1346
1347 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1348 struct btrfs_device *device, u64 num_bytes,
1349 u64 *start, u64 *len)
1350 {
1351 struct btrfs_root *root = device->dev_root;
1352 u64 search_start;
1353
1354 /* FIXME use last free of some kind */
1355
1356 /*
1357 * we don't want to overwrite the superblock on the drive,
1358 * so we make sure to start at an offset of at least 1MB
1359 */
1360 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1361 return find_free_dev_extent_start(trans->transaction, device,
1362 num_bytes, search_start, start, len);
1363 }
1364
1365 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1366 struct btrfs_device *device,
1367 u64 start, u64 *dev_extent_len)
1368 {
1369 int ret;
1370 struct btrfs_path *path;
1371 struct btrfs_root *root = device->dev_root;
1372 struct btrfs_key key;
1373 struct btrfs_key found_key;
1374 struct extent_buffer *leaf = NULL;
1375 struct btrfs_dev_extent *extent = NULL;
1376
1377 path = btrfs_alloc_path();
1378 if (!path)
1379 return -ENOMEM;
1380
1381 key.objectid = device->devid;
1382 key.offset = start;
1383 key.type = BTRFS_DEV_EXTENT_KEY;
1384 again:
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1386 if (ret > 0) {
1387 ret = btrfs_previous_item(root, path, key.objectid,
1388 BTRFS_DEV_EXTENT_KEY);
1389 if (ret)
1390 goto out;
1391 leaf = path->nodes[0];
1392 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1393 extent = btrfs_item_ptr(leaf, path->slots[0],
1394 struct btrfs_dev_extent);
1395 BUG_ON(found_key.offset > start || found_key.offset +
1396 btrfs_dev_extent_length(leaf, extent) < start);
1397 key = found_key;
1398 btrfs_release_path(path);
1399 goto again;
1400 } else if (ret == 0) {
1401 leaf = path->nodes[0];
1402 extent = btrfs_item_ptr(leaf, path->slots[0],
1403 struct btrfs_dev_extent);
1404 } else {
1405 btrfs_error(root->fs_info, ret, "Slot search failed");
1406 goto out;
1407 }
1408
1409 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1410
1411 ret = btrfs_del_item(trans, root, path);
1412 if (ret) {
1413 btrfs_error(root->fs_info, ret,
1414 "Failed to remove dev extent item");
1415 } else {
1416 trans->transaction->have_free_bgs = 1;
1417 }
1418 out:
1419 btrfs_free_path(path);
1420 return ret;
1421 }
1422
1423 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1424 struct btrfs_device *device,
1425 u64 chunk_tree, u64 chunk_objectid,
1426 u64 chunk_offset, u64 start, u64 num_bytes)
1427 {
1428 int ret;
1429 struct btrfs_path *path;
1430 struct btrfs_root *root = device->dev_root;
1431 struct btrfs_dev_extent *extent;
1432 struct extent_buffer *leaf;
1433 struct btrfs_key key;
1434
1435 WARN_ON(!device->in_fs_metadata);
1436 WARN_ON(device->is_tgtdev_for_dev_replace);
1437 path = btrfs_alloc_path();
1438 if (!path)
1439 return -ENOMEM;
1440
1441 key.objectid = device->devid;
1442 key.offset = start;
1443 key.type = BTRFS_DEV_EXTENT_KEY;
1444 ret = btrfs_insert_empty_item(trans, root, path, &key,
1445 sizeof(*extent));
1446 if (ret)
1447 goto out;
1448
1449 leaf = path->nodes[0];
1450 extent = btrfs_item_ptr(leaf, path->slots[0],
1451 struct btrfs_dev_extent);
1452 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1453 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1454 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1455
1456 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1457 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1458
1459 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1460 btrfs_mark_buffer_dirty(leaf);
1461 out:
1462 btrfs_free_path(path);
1463 return ret;
1464 }
1465
1466 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1467 {
1468 struct extent_map_tree *em_tree;
1469 struct extent_map *em;
1470 struct rb_node *n;
1471 u64 ret = 0;
1472
1473 em_tree = &fs_info->mapping_tree.map_tree;
1474 read_lock(&em_tree->lock);
1475 n = rb_last(&em_tree->map);
1476 if (n) {
1477 em = rb_entry(n, struct extent_map, rb_node);
1478 ret = em->start + em->len;
1479 }
1480 read_unlock(&em_tree->lock);
1481
1482 return ret;
1483 }
1484
1485 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1486 u64 *devid_ret)
1487 {
1488 int ret;
1489 struct btrfs_key key;
1490 struct btrfs_key found_key;
1491 struct btrfs_path *path;
1492
1493 path = btrfs_alloc_path();
1494 if (!path)
1495 return -ENOMEM;
1496
1497 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1498 key.type = BTRFS_DEV_ITEM_KEY;
1499 key.offset = (u64)-1;
1500
1501 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1502 if (ret < 0)
1503 goto error;
1504
1505 BUG_ON(ret == 0); /* Corruption */
1506
1507 ret = btrfs_previous_item(fs_info->chunk_root, path,
1508 BTRFS_DEV_ITEMS_OBJECTID,
1509 BTRFS_DEV_ITEM_KEY);
1510 if (ret) {
1511 *devid_ret = 1;
1512 } else {
1513 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1514 path->slots[0]);
1515 *devid_ret = found_key.offset + 1;
1516 }
1517 ret = 0;
1518 error:
1519 btrfs_free_path(path);
1520 return ret;
1521 }
1522
1523 /*
1524 * the device information is stored in the chunk root
1525 * the btrfs_device struct should be fully filled in
1526 */
1527 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1528 struct btrfs_root *root,
1529 struct btrfs_device *device)
1530 {
1531 int ret;
1532 struct btrfs_path *path;
1533 struct btrfs_dev_item *dev_item;
1534 struct extent_buffer *leaf;
1535 struct btrfs_key key;
1536 unsigned long ptr;
1537
1538 root = root->fs_info->chunk_root;
1539
1540 path = btrfs_alloc_path();
1541 if (!path)
1542 return -ENOMEM;
1543
1544 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1545 key.type = BTRFS_DEV_ITEM_KEY;
1546 key.offset = device->devid;
1547
1548 ret = btrfs_insert_empty_item(trans, root, path, &key,
1549 sizeof(*dev_item));
1550 if (ret)
1551 goto out;
1552
1553 leaf = path->nodes[0];
1554 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1555
1556 btrfs_set_device_id(leaf, dev_item, device->devid);
1557 btrfs_set_device_generation(leaf, dev_item, 0);
1558 btrfs_set_device_type(leaf, dev_item, device->type);
1559 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1560 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1561 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1562 btrfs_set_device_total_bytes(leaf, dev_item,
1563 btrfs_device_get_disk_total_bytes(device));
1564 btrfs_set_device_bytes_used(leaf, dev_item,
1565 btrfs_device_get_bytes_used(device));
1566 btrfs_set_device_group(leaf, dev_item, 0);
1567 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1568 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1569 btrfs_set_device_start_offset(leaf, dev_item, 0);
1570
1571 ptr = btrfs_device_uuid(dev_item);
1572 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1573 ptr = btrfs_device_fsid(dev_item);
1574 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1575 btrfs_mark_buffer_dirty(leaf);
1576
1577 ret = 0;
1578 out:
1579 btrfs_free_path(path);
1580 return ret;
1581 }
1582
1583 /*
1584 * Function to update ctime/mtime for a given device path.
1585 * Mainly used for ctime/mtime based probe like libblkid.
1586 */
1587 static void update_dev_time(char *path_name)
1588 {
1589 struct file *filp;
1590
1591 filp = filp_open(path_name, O_RDWR, 0);
1592 if (IS_ERR(filp))
1593 return;
1594 file_update_time(filp);
1595 filp_close(filp, NULL);
1596 return;
1597 }
1598
1599 static int btrfs_rm_dev_item(struct btrfs_root *root,
1600 struct btrfs_device *device)
1601 {
1602 int ret;
1603 struct btrfs_path *path;
1604 struct btrfs_key key;
1605 struct btrfs_trans_handle *trans;
1606
1607 root = root->fs_info->chunk_root;
1608
1609 path = btrfs_alloc_path();
1610 if (!path)
1611 return -ENOMEM;
1612
1613 trans = btrfs_start_transaction(root, 0);
1614 if (IS_ERR(trans)) {
1615 btrfs_free_path(path);
1616 return PTR_ERR(trans);
1617 }
1618 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1619 key.type = BTRFS_DEV_ITEM_KEY;
1620 key.offset = device->devid;
1621
1622 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1623 if (ret < 0)
1624 goto out;
1625
1626 if (ret > 0) {
1627 ret = -ENOENT;
1628 goto out;
1629 }
1630
1631 ret = btrfs_del_item(trans, root, path);
1632 if (ret)
1633 goto out;
1634 out:
1635 btrfs_free_path(path);
1636 btrfs_commit_transaction(trans, root);
1637 return ret;
1638 }
1639
1640 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1641 {
1642 struct btrfs_device *device;
1643 struct btrfs_device *next_device;
1644 struct block_device *bdev;
1645 struct buffer_head *bh = NULL;
1646 struct btrfs_super_block *disk_super;
1647 struct btrfs_fs_devices *cur_devices;
1648 u64 all_avail;
1649 u64 devid;
1650 u64 num_devices;
1651 u8 *dev_uuid;
1652 unsigned seq;
1653 int ret = 0;
1654 bool clear_super = false;
1655
1656 mutex_lock(&uuid_mutex);
1657
1658 do {
1659 seq = read_seqbegin(&root->fs_info->profiles_lock);
1660
1661 all_avail = root->fs_info->avail_data_alloc_bits |
1662 root->fs_info->avail_system_alloc_bits |
1663 root->fs_info->avail_metadata_alloc_bits;
1664 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1665
1666 num_devices = root->fs_info->fs_devices->num_devices;
1667 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1668 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1669 WARN_ON(num_devices < 1);
1670 num_devices--;
1671 }
1672 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1673
1674 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1675 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1676 goto out;
1677 }
1678
1679 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1680 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1681 goto out;
1682 }
1683
1684 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1685 root->fs_info->fs_devices->rw_devices <= 2) {
1686 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1687 goto out;
1688 }
1689 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1690 root->fs_info->fs_devices->rw_devices <= 3) {
1691 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1692 goto out;
1693 }
1694
1695 if (strcmp(device_path, "missing") == 0) {
1696 struct list_head *devices;
1697 struct btrfs_device *tmp;
1698
1699 device = NULL;
1700 devices = &root->fs_info->fs_devices->devices;
1701 /*
1702 * It is safe to read the devices since the volume_mutex
1703 * is held.
1704 */
1705 list_for_each_entry(tmp, devices, dev_list) {
1706 if (tmp->in_fs_metadata &&
1707 !tmp->is_tgtdev_for_dev_replace &&
1708 !tmp->bdev) {
1709 device = tmp;
1710 break;
1711 }
1712 }
1713 bdev = NULL;
1714 bh = NULL;
1715 disk_super = NULL;
1716 if (!device) {
1717 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1718 goto out;
1719 }
1720 } else {
1721 ret = btrfs_get_bdev_and_sb(device_path,
1722 FMODE_WRITE | FMODE_EXCL,
1723 root->fs_info->bdev_holder, 0,
1724 &bdev, &bh);
1725 if (ret)
1726 goto out;
1727 disk_super = (struct btrfs_super_block *)bh->b_data;
1728 devid = btrfs_stack_device_id(&disk_super->dev_item);
1729 dev_uuid = disk_super->dev_item.uuid;
1730 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1731 disk_super->fsid);
1732 if (!device) {
1733 ret = -ENOENT;
1734 goto error_brelse;
1735 }
1736 }
1737
1738 if (device->is_tgtdev_for_dev_replace) {
1739 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1740 goto error_brelse;
1741 }
1742
1743 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1744 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1745 goto error_brelse;
1746 }
1747
1748 if (device->writeable) {
1749 lock_chunks(root);
1750 list_del_init(&device->dev_alloc_list);
1751 device->fs_devices->rw_devices--;
1752 unlock_chunks(root);
1753 clear_super = true;
1754 }
1755
1756 mutex_unlock(&uuid_mutex);
1757 ret = btrfs_shrink_device(device, 0);
1758 mutex_lock(&uuid_mutex);
1759 if (ret)
1760 goto error_undo;
1761
1762 /*
1763 * TODO: the superblock still includes this device in its num_devices
1764 * counter although write_all_supers() is not locked out. This
1765 * could give a filesystem state which requires a degraded mount.
1766 */
1767 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1768 if (ret)
1769 goto error_undo;
1770
1771 device->in_fs_metadata = 0;
1772 btrfs_scrub_cancel_dev(root->fs_info, device);
1773
1774 /*
1775 * the device list mutex makes sure that we don't change
1776 * the device list while someone else is writing out all
1777 * the device supers. Whoever is writing all supers, should
1778 * lock the device list mutex before getting the number of
1779 * devices in the super block (super_copy). Conversely,
1780 * whoever updates the number of devices in the super block
1781 * (super_copy) should hold the device list mutex.
1782 */
1783
1784 cur_devices = device->fs_devices;
1785 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1786 list_del_rcu(&device->dev_list);
1787
1788 device->fs_devices->num_devices--;
1789 device->fs_devices->total_devices--;
1790
1791 if (device->missing)
1792 device->fs_devices->missing_devices--;
1793
1794 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1795 struct btrfs_device, dev_list);
1796 if (device->bdev == root->fs_info->sb->s_bdev)
1797 root->fs_info->sb->s_bdev = next_device->bdev;
1798 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1799 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1800
1801 if (device->bdev) {
1802 device->fs_devices->open_devices--;
1803 /* remove sysfs entry */
1804 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
1805 }
1806
1807 call_rcu(&device->rcu, free_device);
1808
1809 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1810 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1811 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1812
1813 if (cur_devices->open_devices == 0) {
1814 struct btrfs_fs_devices *fs_devices;
1815 fs_devices = root->fs_info->fs_devices;
1816 while (fs_devices) {
1817 if (fs_devices->seed == cur_devices) {
1818 fs_devices->seed = cur_devices->seed;
1819 break;
1820 }
1821 fs_devices = fs_devices->seed;
1822 }
1823 cur_devices->seed = NULL;
1824 __btrfs_close_devices(cur_devices);
1825 free_fs_devices(cur_devices);
1826 }
1827
1828 root->fs_info->num_tolerated_disk_barrier_failures =
1829 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1830
1831 /*
1832 * at this point, the device is zero sized. We want to
1833 * remove it from the devices list and zero out the old super
1834 */
1835 if (clear_super && disk_super) {
1836 u64 bytenr;
1837 int i;
1838
1839 /* make sure this device isn't detected as part of
1840 * the FS anymore
1841 */
1842 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1843 set_buffer_dirty(bh);
1844 sync_dirty_buffer(bh);
1845
1846 /* clear the mirror copies of super block on the disk
1847 * being removed, 0th copy is been taken care above and
1848 * the below would take of the rest
1849 */
1850 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1851 bytenr = btrfs_sb_offset(i);
1852 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1853 i_size_read(bdev->bd_inode))
1854 break;
1855
1856 brelse(bh);
1857 bh = __bread(bdev, bytenr / 4096,
1858 BTRFS_SUPER_INFO_SIZE);
1859 if (!bh)
1860 continue;
1861
1862 disk_super = (struct btrfs_super_block *)bh->b_data;
1863
1864 if (btrfs_super_bytenr(disk_super) != bytenr ||
1865 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1866 continue;
1867 }
1868 memset(&disk_super->magic, 0,
1869 sizeof(disk_super->magic));
1870 set_buffer_dirty(bh);
1871 sync_dirty_buffer(bh);
1872 }
1873 }
1874
1875 ret = 0;
1876
1877 if (bdev) {
1878 /* Notify udev that device has changed */
1879 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1880
1881 /* Update ctime/mtime for device path for libblkid */
1882 update_dev_time(device_path);
1883 }
1884
1885 error_brelse:
1886 brelse(bh);
1887 if (bdev)
1888 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1889 out:
1890 mutex_unlock(&uuid_mutex);
1891 return ret;
1892 error_undo:
1893 if (device->writeable) {
1894 lock_chunks(root);
1895 list_add(&device->dev_alloc_list,
1896 &root->fs_info->fs_devices->alloc_list);
1897 device->fs_devices->rw_devices++;
1898 unlock_chunks(root);
1899 }
1900 goto error_brelse;
1901 }
1902
1903 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1904 struct btrfs_device *srcdev)
1905 {
1906 struct btrfs_fs_devices *fs_devices;
1907
1908 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1909
1910 /*
1911 * in case of fs with no seed, srcdev->fs_devices will point
1912 * to fs_devices of fs_info. However when the dev being replaced is
1913 * a seed dev it will point to the seed's local fs_devices. In short
1914 * srcdev will have its correct fs_devices in both the cases.
1915 */
1916 fs_devices = srcdev->fs_devices;
1917
1918 list_del_rcu(&srcdev->dev_list);
1919 list_del_rcu(&srcdev->dev_alloc_list);
1920 fs_devices->num_devices--;
1921 if (srcdev->missing)
1922 fs_devices->missing_devices--;
1923
1924 if (srcdev->writeable) {
1925 fs_devices->rw_devices--;
1926 /* zero out the old super if it is writable */
1927 btrfs_scratch_superblock(srcdev);
1928 }
1929
1930 if (srcdev->bdev)
1931 fs_devices->open_devices--;
1932 }
1933
1934 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1935 struct btrfs_device *srcdev)
1936 {
1937 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1938
1939 call_rcu(&srcdev->rcu, free_device);
1940
1941 /*
1942 * unless fs_devices is seed fs, num_devices shouldn't go
1943 * zero
1944 */
1945 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1946
1947 /* if this is no devs we rather delete the fs_devices */
1948 if (!fs_devices->num_devices) {
1949 struct btrfs_fs_devices *tmp_fs_devices;
1950
1951 tmp_fs_devices = fs_info->fs_devices;
1952 while (tmp_fs_devices) {
1953 if (tmp_fs_devices->seed == fs_devices) {
1954 tmp_fs_devices->seed = fs_devices->seed;
1955 break;
1956 }
1957 tmp_fs_devices = tmp_fs_devices->seed;
1958 }
1959 fs_devices->seed = NULL;
1960 __btrfs_close_devices(fs_devices);
1961 free_fs_devices(fs_devices);
1962 }
1963 }
1964
1965 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1966 struct btrfs_device *tgtdev)
1967 {
1968 struct btrfs_device *next_device;
1969
1970 mutex_lock(&uuid_mutex);
1971 WARN_ON(!tgtdev);
1972 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1973
1974 btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev);
1975
1976 if (tgtdev->bdev) {
1977 btrfs_scratch_superblock(tgtdev);
1978 fs_info->fs_devices->open_devices--;
1979 }
1980 fs_info->fs_devices->num_devices--;
1981
1982 next_device = list_entry(fs_info->fs_devices->devices.next,
1983 struct btrfs_device, dev_list);
1984 if (tgtdev->bdev == fs_info->sb->s_bdev)
1985 fs_info->sb->s_bdev = next_device->bdev;
1986 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1987 fs_info->fs_devices->latest_bdev = next_device->bdev;
1988 list_del_rcu(&tgtdev->dev_list);
1989
1990 call_rcu(&tgtdev->rcu, free_device);
1991
1992 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1993 mutex_unlock(&uuid_mutex);
1994 }
1995
1996 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1997 struct btrfs_device **device)
1998 {
1999 int ret = 0;
2000 struct btrfs_super_block *disk_super;
2001 u64 devid;
2002 u8 *dev_uuid;
2003 struct block_device *bdev;
2004 struct buffer_head *bh;
2005
2006 *device = NULL;
2007 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2008 root->fs_info->bdev_holder, 0, &bdev, &bh);
2009 if (ret)
2010 return ret;
2011 disk_super = (struct btrfs_super_block *)bh->b_data;
2012 devid = btrfs_stack_device_id(&disk_super->dev_item);
2013 dev_uuid = disk_super->dev_item.uuid;
2014 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2015 disk_super->fsid);
2016 brelse(bh);
2017 if (!*device)
2018 ret = -ENOENT;
2019 blkdev_put(bdev, FMODE_READ);
2020 return ret;
2021 }
2022
2023 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2024 char *device_path,
2025 struct btrfs_device **device)
2026 {
2027 *device = NULL;
2028 if (strcmp(device_path, "missing") == 0) {
2029 struct list_head *devices;
2030 struct btrfs_device *tmp;
2031
2032 devices = &root->fs_info->fs_devices->devices;
2033 /*
2034 * It is safe to read the devices since the volume_mutex
2035 * is held by the caller.
2036 */
2037 list_for_each_entry(tmp, devices, dev_list) {
2038 if (tmp->in_fs_metadata && !tmp->bdev) {
2039 *device = tmp;
2040 break;
2041 }
2042 }
2043
2044 if (!*device) {
2045 btrfs_err(root->fs_info, "no missing device found");
2046 return -ENOENT;
2047 }
2048
2049 return 0;
2050 } else {
2051 return btrfs_find_device_by_path(root, device_path, device);
2052 }
2053 }
2054
2055 /*
2056 * does all the dirty work required for changing file system's UUID.
2057 */
2058 static int btrfs_prepare_sprout(struct btrfs_root *root)
2059 {
2060 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2061 struct btrfs_fs_devices *old_devices;
2062 struct btrfs_fs_devices *seed_devices;
2063 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2064 struct btrfs_device *device;
2065 u64 super_flags;
2066
2067 BUG_ON(!mutex_is_locked(&uuid_mutex));
2068 if (!fs_devices->seeding)
2069 return -EINVAL;
2070
2071 seed_devices = __alloc_fs_devices();
2072 if (IS_ERR(seed_devices))
2073 return PTR_ERR(seed_devices);
2074
2075 old_devices = clone_fs_devices(fs_devices);
2076 if (IS_ERR(old_devices)) {
2077 kfree(seed_devices);
2078 return PTR_ERR(old_devices);
2079 }
2080
2081 list_add(&old_devices->list, &fs_uuids);
2082
2083 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2084 seed_devices->opened = 1;
2085 INIT_LIST_HEAD(&seed_devices->devices);
2086 INIT_LIST_HEAD(&seed_devices->alloc_list);
2087 mutex_init(&seed_devices->device_list_mutex);
2088
2089 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2090 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2091 synchronize_rcu);
2092 list_for_each_entry(device, &seed_devices->devices, dev_list)
2093 device->fs_devices = seed_devices;
2094
2095 lock_chunks(root);
2096 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2097 unlock_chunks(root);
2098
2099 fs_devices->seeding = 0;
2100 fs_devices->num_devices = 0;
2101 fs_devices->open_devices = 0;
2102 fs_devices->missing_devices = 0;
2103 fs_devices->rotating = 0;
2104 fs_devices->seed = seed_devices;
2105
2106 generate_random_uuid(fs_devices->fsid);
2107 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2108 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2109 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2110
2111 super_flags = btrfs_super_flags(disk_super) &
2112 ~BTRFS_SUPER_FLAG_SEEDING;
2113 btrfs_set_super_flags(disk_super, super_flags);
2114
2115 return 0;
2116 }
2117
2118 /*
2119 * strore the expected generation for seed devices in device items.
2120 */
2121 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2122 struct btrfs_root *root)
2123 {
2124 struct btrfs_path *path;
2125 struct extent_buffer *leaf;
2126 struct btrfs_dev_item *dev_item;
2127 struct btrfs_device *device;
2128 struct btrfs_key key;
2129 u8 fs_uuid[BTRFS_UUID_SIZE];
2130 u8 dev_uuid[BTRFS_UUID_SIZE];
2131 u64 devid;
2132 int ret;
2133
2134 path = btrfs_alloc_path();
2135 if (!path)
2136 return -ENOMEM;
2137
2138 root = root->fs_info->chunk_root;
2139 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2140 key.offset = 0;
2141 key.type = BTRFS_DEV_ITEM_KEY;
2142
2143 while (1) {
2144 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2145 if (ret < 0)
2146 goto error;
2147
2148 leaf = path->nodes[0];
2149 next_slot:
2150 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2151 ret = btrfs_next_leaf(root, path);
2152 if (ret > 0)
2153 break;
2154 if (ret < 0)
2155 goto error;
2156 leaf = path->nodes[0];
2157 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2158 btrfs_release_path(path);
2159 continue;
2160 }
2161
2162 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2163 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2164 key.type != BTRFS_DEV_ITEM_KEY)
2165 break;
2166
2167 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2168 struct btrfs_dev_item);
2169 devid = btrfs_device_id(leaf, dev_item);
2170 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2171 BTRFS_UUID_SIZE);
2172 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2173 BTRFS_UUID_SIZE);
2174 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2175 fs_uuid);
2176 BUG_ON(!device); /* Logic error */
2177
2178 if (device->fs_devices->seeding) {
2179 btrfs_set_device_generation(leaf, dev_item,
2180 device->generation);
2181 btrfs_mark_buffer_dirty(leaf);
2182 }
2183
2184 path->slots[0]++;
2185 goto next_slot;
2186 }
2187 ret = 0;
2188 error:
2189 btrfs_free_path(path);
2190 return ret;
2191 }
2192
2193 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2194 {
2195 struct request_queue *q;
2196 struct btrfs_trans_handle *trans;
2197 struct btrfs_device *device;
2198 struct block_device *bdev;
2199 struct list_head *devices;
2200 struct super_block *sb = root->fs_info->sb;
2201 struct rcu_string *name;
2202 u64 tmp;
2203 int seeding_dev = 0;
2204 int ret = 0;
2205
2206 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2207 return -EROFS;
2208
2209 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2210 root->fs_info->bdev_holder);
2211 if (IS_ERR(bdev))
2212 return PTR_ERR(bdev);
2213
2214 if (root->fs_info->fs_devices->seeding) {
2215 seeding_dev = 1;
2216 down_write(&sb->s_umount);
2217 mutex_lock(&uuid_mutex);
2218 }
2219
2220 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2221
2222 devices = &root->fs_info->fs_devices->devices;
2223
2224 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2225 list_for_each_entry(device, devices, dev_list) {
2226 if (device->bdev == bdev) {
2227 ret = -EEXIST;
2228 mutex_unlock(
2229 &root->fs_info->fs_devices->device_list_mutex);
2230 goto error;
2231 }
2232 }
2233 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2234
2235 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2236 if (IS_ERR(device)) {
2237 /* we can safely leave the fs_devices entry around */
2238 ret = PTR_ERR(device);
2239 goto error;
2240 }
2241
2242 name = rcu_string_strdup(device_path, GFP_NOFS);
2243 if (!name) {
2244 kfree(device);
2245 ret = -ENOMEM;
2246 goto error;
2247 }
2248 rcu_assign_pointer(device->name, name);
2249
2250 trans = btrfs_start_transaction(root, 0);
2251 if (IS_ERR(trans)) {
2252 rcu_string_free(device->name);
2253 kfree(device);
2254 ret = PTR_ERR(trans);
2255 goto error;
2256 }
2257
2258 q = bdev_get_queue(bdev);
2259 if (blk_queue_discard(q))
2260 device->can_discard = 1;
2261 device->writeable = 1;
2262 device->generation = trans->transid;
2263 device->io_width = root->sectorsize;
2264 device->io_align = root->sectorsize;
2265 device->sector_size = root->sectorsize;
2266 device->total_bytes = i_size_read(bdev->bd_inode);
2267 device->disk_total_bytes = device->total_bytes;
2268 device->commit_total_bytes = device->total_bytes;
2269 device->dev_root = root->fs_info->dev_root;
2270 device->bdev = bdev;
2271 device->in_fs_metadata = 1;
2272 device->is_tgtdev_for_dev_replace = 0;
2273 device->mode = FMODE_EXCL;
2274 device->dev_stats_valid = 1;
2275 set_blocksize(device->bdev, 4096);
2276
2277 if (seeding_dev) {
2278 sb->s_flags &= ~MS_RDONLY;
2279 ret = btrfs_prepare_sprout(root);
2280 BUG_ON(ret); /* -ENOMEM */
2281 }
2282
2283 device->fs_devices = root->fs_info->fs_devices;
2284
2285 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2286 lock_chunks(root);
2287 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2288 list_add(&device->dev_alloc_list,
2289 &root->fs_info->fs_devices->alloc_list);
2290 root->fs_info->fs_devices->num_devices++;
2291 root->fs_info->fs_devices->open_devices++;
2292 root->fs_info->fs_devices->rw_devices++;
2293 root->fs_info->fs_devices->total_devices++;
2294 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2295
2296 spin_lock(&root->fs_info->free_chunk_lock);
2297 root->fs_info->free_chunk_space += device->total_bytes;
2298 spin_unlock(&root->fs_info->free_chunk_lock);
2299
2300 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2301 root->fs_info->fs_devices->rotating = 1;
2302
2303 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2304 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2305 tmp + device->total_bytes);
2306
2307 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2308 btrfs_set_super_num_devices(root->fs_info->super_copy,
2309 tmp + 1);
2310
2311 /* add sysfs device entry */
2312 btrfs_kobj_add_device(root->fs_info->fs_devices, device);
2313
2314 /*
2315 * we've got more storage, clear any full flags on the space
2316 * infos
2317 */
2318 btrfs_clear_space_info_full(root->fs_info);
2319
2320 unlock_chunks(root);
2321 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2322
2323 if (seeding_dev) {
2324 lock_chunks(root);
2325 ret = init_first_rw_device(trans, root, device);
2326 unlock_chunks(root);
2327 if (ret) {
2328 btrfs_abort_transaction(trans, root, ret);
2329 goto error_trans;
2330 }
2331 }
2332
2333 ret = btrfs_add_device(trans, root, device);
2334 if (ret) {
2335 btrfs_abort_transaction(trans, root, ret);
2336 goto error_trans;
2337 }
2338
2339 if (seeding_dev) {
2340 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2341
2342 ret = btrfs_finish_sprout(trans, root);
2343 if (ret) {
2344 btrfs_abort_transaction(trans, root, ret);
2345 goto error_trans;
2346 }
2347
2348 /* Sprouting would change fsid of the mounted root,
2349 * so rename the fsid on the sysfs
2350 */
2351 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2352 root->fs_info->fsid);
2353 if (kobject_rename(&root->fs_info->fs_devices->super_kobj,
2354 fsid_buf))
2355 pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n");
2356 }
2357
2358 root->fs_info->num_tolerated_disk_barrier_failures =
2359 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2360 ret = btrfs_commit_transaction(trans, root);
2361
2362 if (seeding_dev) {
2363 mutex_unlock(&uuid_mutex);
2364 up_write(&sb->s_umount);
2365
2366 if (ret) /* transaction commit */
2367 return ret;
2368
2369 ret = btrfs_relocate_sys_chunks(root);
2370 if (ret < 0)
2371 btrfs_error(root->fs_info, ret,
2372 "Failed to relocate sys chunks after "
2373 "device initialization. This can be fixed "
2374 "using the \"btrfs balance\" command.");
2375 trans = btrfs_attach_transaction(root);
2376 if (IS_ERR(trans)) {
2377 if (PTR_ERR(trans) == -ENOENT)
2378 return 0;
2379 return PTR_ERR(trans);
2380 }
2381 ret = btrfs_commit_transaction(trans, root);
2382 }
2383
2384 /* Update ctime/mtime for libblkid */
2385 update_dev_time(device_path);
2386 return ret;
2387
2388 error_trans:
2389 btrfs_end_transaction(trans, root);
2390 rcu_string_free(device->name);
2391 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
2392 kfree(device);
2393 error:
2394 blkdev_put(bdev, FMODE_EXCL);
2395 if (seeding_dev) {
2396 mutex_unlock(&uuid_mutex);
2397 up_write(&sb->s_umount);
2398 }
2399 return ret;
2400 }
2401
2402 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2403 struct btrfs_device *srcdev,
2404 struct btrfs_device **device_out)
2405 {
2406 struct request_queue *q;
2407 struct btrfs_device *device;
2408 struct block_device *bdev;
2409 struct btrfs_fs_info *fs_info = root->fs_info;
2410 struct list_head *devices;
2411 struct rcu_string *name;
2412 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2413 int ret = 0;
2414
2415 *device_out = NULL;
2416 if (fs_info->fs_devices->seeding) {
2417 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2418 return -EINVAL;
2419 }
2420
2421 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2422 fs_info->bdev_holder);
2423 if (IS_ERR(bdev)) {
2424 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2425 return PTR_ERR(bdev);
2426 }
2427
2428 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2429
2430 devices = &fs_info->fs_devices->devices;
2431 list_for_each_entry(device, devices, dev_list) {
2432 if (device->bdev == bdev) {
2433 btrfs_err(fs_info, "target device is in the filesystem!");
2434 ret = -EEXIST;
2435 goto error;
2436 }
2437 }
2438
2439
2440 if (i_size_read(bdev->bd_inode) <
2441 btrfs_device_get_total_bytes(srcdev)) {
2442 btrfs_err(fs_info, "target device is smaller than source device!");
2443 ret = -EINVAL;
2444 goto error;
2445 }
2446
2447
2448 device = btrfs_alloc_device(NULL, &devid, NULL);
2449 if (IS_ERR(device)) {
2450 ret = PTR_ERR(device);
2451 goto error;
2452 }
2453
2454 name = rcu_string_strdup(device_path, GFP_NOFS);
2455 if (!name) {
2456 kfree(device);
2457 ret = -ENOMEM;
2458 goto error;
2459 }
2460 rcu_assign_pointer(device->name, name);
2461
2462 q = bdev_get_queue(bdev);
2463 if (blk_queue_discard(q))
2464 device->can_discard = 1;
2465 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2466 device->writeable = 1;
2467 device->generation = 0;
2468 device->io_width = root->sectorsize;
2469 device->io_align = root->sectorsize;
2470 device->sector_size = root->sectorsize;
2471 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2472 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2473 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2474 ASSERT(list_empty(&srcdev->resized_list));
2475 device->commit_total_bytes = srcdev->commit_total_bytes;
2476 device->commit_bytes_used = device->bytes_used;
2477 device->dev_root = fs_info->dev_root;
2478 device->bdev = bdev;
2479 device->in_fs_metadata = 1;
2480 device->is_tgtdev_for_dev_replace = 1;
2481 device->mode = FMODE_EXCL;
2482 device->dev_stats_valid = 1;
2483 set_blocksize(device->bdev, 4096);
2484 device->fs_devices = fs_info->fs_devices;
2485 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2486 fs_info->fs_devices->num_devices++;
2487 fs_info->fs_devices->open_devices++;
2488 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2489
2490 *device_out = device;
2491 return ret;
2492
2493 error:
2494 blkdev_put(bdev, FMODE_EXCL);
2495 return ret;
2496 }
2497
2498 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2499 struct btrfs_device *tgtdev)
2500 {
2501 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2502 tgtdev->io_width = fs_info->dev_root->sectorsize;
2503 tgtdev->io_align = fs_info->dev_root->sectorsize;
2504 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2505 tgtdev->dev_root = fs_info->dev_root;
2506 tgtdev->in_fs_metadata = 1;
2507 }
2508
2509 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2510 struct btrfs_device *device)
2511 {
2512 int ret;
2513 struct btrfs_path *path;
2514 struct btrfs_root *root;
2515 struct btrfs_dev_item *dev_item;
2516 struct extent_buffer *leaf;
2517 struct btrfs_key key;
2518
2519 root = device->dev_root->fs_info->chunk_root;
2520
2521 path = btrfs_alloc_path();
2522 if (!path)
2523 return -ENOMEM;
2524
2525 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2526 key.type = BTRFS_DEV_ITEM_KEY;
2527 key.offset = device->devid;
2528
2529 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2530 if (ret < 0)
2531 goto out;
2532
2533 if (ret > 0) {
2534 ret = -ENOENT;
2535 goto out;
2536 }
2537
2538 leaf = path->nodes[0];
2539 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2540
2541 btrfs_set_device_id(leaf, dev_item, device->devid);
2542 btrfs_set_device_type(leaf, dev_item, device->type);
2543 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2544 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2545 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2546 btrfs_set_device_total_bytes(leaf, dev_item,
2547 btrfs_device_get_disk_total_bytes(device));
2548 btrfs_set_device_bytes_used(leaf, dev_item,
2549 btrfs_device_get_bytes_used(device));
2550 btrfs_mark_buffer_dirty(leaf);
2551
2552 out:
2553 btrfs_free_path(path);
2554 return ret;
2555 }
2556
2557 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2558 struct btrfs_device *device, u64 new_size)
2559 {
2560 struct btrfs_super_block *super_copy =
2561 device->dev_root->fs_info->super_copy;
2562 struct btrfs_fs_devices *fs_devices;
2563 u64 old_total;
2564 u64 diff;
2565
2566 if (!device->writeable)
2567 return -EACCES;
2568
2569 lock_chunks(device->dev_root);
2570 old_total = btrfs_super_total_bytes(super_copy);
2571 diff = new_size - device->total_bytes;
2572
2573 if (new_size <= device->total_bytes ||
2574 device->is_tgtdev_for_dev_replace) {
2575 unlock_chunks(device->dev_root);
2576 return -EINVAL;
2577 }
2578
2579 fs_devices = device->dev_root->fs_info->fs_devices;
2580
2581 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2582 device->fs_devices->total_rw_bytes += diff;
2583
2584 btrfs_device_set_total_bytes(device, new_size);
2585 btrfs_device_set_disk_total_bytes(device, new_size);
2586 btrfs_clear_space_info_full(device->dev_root->fs_info);
2587 if (list_empty(&device->resized_list))
2588 list_add_tail(&device->resized_list,
2589 &fs_devices->resized_devices);
2590 unlock_chunks(device->dev_root);
2591
2592 return btrfs_update_device(trans, device);
2593 }
2594
2595 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2596 struct btrfs_root *root, u64 chunk_objectid,
2597 u64 chunk_offset)
2598 {
2599 int ret;
2600 struct btrfs_path *path;
2601 struct btrfs_key key;
2602
2603 root = root->fs_info->chunk_root;
2604 path = btrfs_alloc_path();
2605 if (!path)
2606 return -ENOMEM;
2607
2608 key.objectid = chunk_objectid;
2609 key.offset = chunk_offset;
2610 key.type = BTRFS_CHUNK_ITEM_KEY;
2611
2612 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2613 if (ret < 0)
2614 goto out;
2615 else if (ret > 0) { /* Logic error or corruption */
2616 btrfs_error(root->fs_info, -ENOENT,
2617 "Failed lookup while freeing chunk.");
2618 ret = -ENOENT;
2619 goto out;
2620 }
2621
2622 ret = btrfs_del_item(trans, root, path);
2623 if (ret < 0)
2624 btrfs_error(root->fs_info, ret,
2625 "Failed to delete chunk item.");
2626 out:
2627 btrfs_free_path(path);
2628 return ret;
2629 }
2630
2631 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2632 chunk_offset)
2633 {
2634 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2635 struct btrfs_disk_key *disk_key;
2636 struct btrfs_chunk *chunk;
2637 u8 *ptr;
2638 int ret = 0;
2639 u32 num_stripes;
2640 u32 array_size;
2641 u32 len = 0;
2642 u32 cur;
2643 struct btrfs_key key;
2644
2645 lock_chunks(root);
2646 array_size = btrfs_super_sys_array_size(super_copy);
2647
2648 ptr = super_copy->sys_chunk_array;
2649 cur = 0;
2650
2651 while (cur < array_size) {
2652 disk_key = (struct btrfs_disk_key *)ptr;
2653 btrfs_disk_key_to_cpu(&key, disk_key);
2654
2655 len = sizeof(*disk_key);
2656
2657 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2658 chunk = (struct btrfs_chunk *)(ptr + len);
2659 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2660 len += btrfs_chunk_item_size(num_stripes);
2661 } else {
2662 ret = -EIO;
2663 break;
2664 }
2665 if (key.objectid == chunk_objectid &&
2666 key.offset == chunk_offset) {
2667 memmove(ptr, ptr + len, array_size - (cur + len));
2668 array_size -= len;
2669 btrfs_set_super_sys_array_size(super_copy, array_size);
2670 } else {
2671 ptr += len;
2672 cur += len;
2673 }
2674 }
2675 unlock_chunks(root);
2676 return ret;
2677 }
2678
2679 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2680 struct btrfs_root *root, u64 chunk_offset)
2681 {
2682 struct extent_map_tree *em_tree;
2683 struct extent_map *em;
2684 struct btrfs_root *extent_root = root->fs_info->extent_root;
2685 struct map_lookup *map;
2686 u64 dev_extent_len = 0;
2687 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2688 int i, ret = 0;
2689
2690 /* Just in case */
2691 root = root->fs_info->chunk_root;
2692 em_tree = &root->fs_info->mapping_tree.map_tree;
2693
2694 read_lock(&em_tree->lock);
2695 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2696 read_unlock(&em_tree->lock);
2697
2698 if (!em || em->start > chunk_offset ||
2699 em->start + em->len < chunk_offset) {
2700 /*
2701 * This is a logic error, but we don't want to just rely on the
2702 * user having built with ASSERT enabled, so if ASSERT doens't
2703 * do anything we still error out.
2704 */
2705 ASSERT(0);
2706 if (em)
2707 free_extent_map(em);
2708 return -EINVAL;
2709 }
2710 map = (struct map_lookup *)em->bdev;
2711 lock_chunks(root->fs_info->chunk_root);
2712 check_system_chunk(trans, extent_root, map->type);
2713 unlock_chunks(root->fs_info->chunk_root);
2714
2715 for (i = 0; i < map->num_stripes; i++) {
2716 struct btrfs_device *device = map->stripes[i].dev;
2717 ret = btrfs_free_dev_extent(trans, device,
2718 map->stripes[i].physical,
2719 &dev_extent_len);
2720 if (ret) {
2721 btrfs_abort_transaction(trans, root, ret);
2722 goto out;
2723 }
2724
2725 if (device->bytes_used > 0) {
2726 lock_chunks(root);
2727 btrfs_device_set_bytes_used(device,
2728 device->bytes_used - dev_extent_len);
2729 spin_lock(&root->fs_info->free_chunk_lock);
2730 root->fs_info->free_chunk_space += dev_extent_len;
2731 spin_unlock(&root->fs_info->free_chunk_lock);
2732 btrfs_clear_space_info_full(root->fs_info);
2733 unlock_chunks(root);
2734 }
2735
2736 if (map->stripes[i].dev) {
2737 ret = btrfs_update_device(trans, map->stripes[i].dev);
2738 if (ret) {
2739 btrfs_abort_transaction(trans, root, ret);
2740 goto out;
2741 }
2742 }
2743 }
2744 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2745 if (ret) {
2746 btrfs_abort_transaction(trans, root, ret);
2747 goto out;
2748 }
2749
2750 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2751
2752 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2753 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2754 if (ret) {
2755 btrfs_abort_transaction(trans, root, ret);
2756 goto out;
2757 }
2758 }
2759
2760 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2761 if (ret) {
2762 btrfs_abort_transaction(trans, extent_root, ret);
2763 goto out;
2764 }
2765
2766 out:
2767 /* once for us */
2768 free_extent_map(em);
2769 return ret;
2770 }
2771
2772 static int btrfs_relocate_chunk(struct btrfs_root *root,
2773 u64 chunk_objectid,
2774 u64 chunk_offset)
2775 {
2776 struct btrfs_root *extent_root;
2777 struct btrfs_trans_handle *trans;
2778 int ret;
2779
2780 root = root->fs_info->chunk_root;
2781 extent_root = root->fs_info->extent_root;
2782
2783 /*
2784 * Prevent races with automatic removal of unused block groups.
2785 * After we relocate and before we remove the chunk with offset
2786 * chunk_offset, automatic removal of the block group can kick in,
2787 * resulting in a failure when calling btrfs_remove_chunk() below.
2788 *
2789 * Make sure to acquire this mutex before doing a tree search (dev
2790 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2791 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2792 * we release the path used to search the chunk/dev tree and before
2793 * the current task acquires this mutex and calls us.
2794 */
2795 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2796
2797 ret = btrfs_can_relocate(extent_root, chunk_offset);
2798 if (ret)
2799 return -ENOSPC;
2800
2801 /* step one, relocate all the extents inside this chunk */
2802 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2803 if (ret)
2804 return ret;
2805
2806 trans = btrfs_start_transaction(root, 0);
2807 if (IS_ERR(trans)) {
2808 ret = PTR_ERR(trans);
2809 btrfs_std_error(root->fs_info, ret);
2810 return ret;
2811 }
2812
2813 /*
2814 * step two, delete the device extents and the
2815 * chunk tree entries
2816 */
2817 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2818 btrfs_end_transaction(trans, root);
2819 return ret;
2820 }
2821
2822 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2823 {
2824 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2825 struct btrfs_path *path;
2826 struct extent_buffer *leaf;
2827 struct btrfs_chunk *chunk;
2828 struct btrfs_key key;
2829 struct btrfs_key found_key;
2830 u64 chunk_type;
2831 bool retried = false;
2832 int failed = 0;
2833 int ret;
2834
2835 path = btrfs_alloc_path();
2836 if (!path)
2837 return -ENOMEM;
2838
2839 again:
2840 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2841 key.offset = (u64)-1;
2842 key.type = BTRFS_CHUNK_ITEM_KEY;
2843
2844 while (1) {
2845 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2846 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2847 if (ret < 0) {
2848 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2849 goto error;
2850 }
2851 BUG_ON(ret == 0); /* Corruption */
2852
2853 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2854 key.type);
2855 if (ret)
2856 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2857 if (ret < 0)
2858 goto error;
2859 if (ret > 0)
2860 break;
2861
2862 leaf = path->nodes[0];
2863 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2864
2865 chunk = btrfs_item_ptr(leaf, path->slots[0],
2866 struct btrfs_chunk);
2867 chunk_type = btrfs_chunk_type(leaf, chunk);
2868 btrfs_release_path(path);
2869
2870 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2871 ret = btrfs_relocate_chunk(chunk_root,
2872 found_key.objectid,
2873 found_key.offset);
2874 if (ret == -ENOSPC)
2875 failed++;
2876 else
2877 BUG_ON(ret);
2878 }
2879 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2880
2881 if (found_key.offset == 0)
2882 break;
2883 key.offset = found_key.offset - 1;
2884 }
2885 ret = 0;
2886 if (failed && !retried) {
2887 failed = 0;
2888 retried = true;
2889 goto again;
2890 } else if (WARN_ON(failed && retried)) {
2891 ret = -ENOSPC;
2892 }
2893 error:
2894 btrfs_free_path(path);
2895 return ret;
2896 }
2897
2898 static int insert_balance_item(struct btrfs_root *root,
2899 struct btrfs_balance_control *bctl)
2900 {
2901 struct btrfs_trans_handle *trans;
2902 struct btrfs_balance_item *item;
2903 struct btrfs_disk_balance_args disk_bargs;
2904 struct btrfs_path *path;
2905 struct extent_buffer *leaf;
2906 struct btrfs_key key;
2907 int ret, err;
2908
2909 path = btrfs_alloc_path();
2910 if (!path)
2911 return -ENOMEM;
2912
2913 trans = btrfs_start_transaction(root, 0);
2914 if (IS_ERR(trans)) {
2915 btrfs_free_path(path);
2916 return PTR_ERR(trans);
2917 }
2918
2919 key.objectid = BTRFS_BALANCE_OBJECTID;
2920 key.type = BTRFS_BALANCE_ITEM_KEY;
2921 key.offset = 0;
2922
2923 ret = btrfs_insert_empty_item(trans, root, path, &key,
2924 sizeof(*item));
2925 if (ret)
2926 goto out;
2927
2928 leaf = path->nodes[0];
2929 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2930
2931 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2932
2933 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2934 btrfs_set_balance_data(leaf, item, &disk_bargs);
2935 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2936 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2937 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2938 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2939
2940 btrfs_set_balance_flags(leaf, item, bctl->flags);
2941
2942 btrfs_mark_buffer_dirty(leaf);
2943 out:
2944 btrfs_free_path(path);
2945 err = btrfs_commit_transaction(trans, root);
2946 if (err && !ret)
2947 ret = err;
2948 return ret;
2949 }
2950
2951 static int del_balance_item(struct btrfs_root *root)
2952 {
2953 struct btrfs_trans_handle *trans;
2954 struct btrfs_path *path;
2955 struct btrfs_key key;
2956 int ret, err;
2957
2958 path = btrfs_alloc_path();
2959 if (!path)
2960 return -ENOMEM;
2961
2962 trans = btrfs_start_transaction(root, 0);
2963 if (IS_ERR(trans)) {
2964 btrfs_free_path(path);
2965 return PTR_ERR(trans);
2966 }
2967
2968 key.objectid = BTRFS_BALANCE_OBJECTID;
2969 key.type = BTRFS_BALANCE_ITEM_KEY;
2970 key.offset = 0;
2971
2972 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2973 if (ret < 0)
2974 goto out;
2975 if (ret > 0) {
2976 ret = -ENOENT;
2977 goto out;
2978 }
2979
2980 ret = btrfs_del_item(trans, root, path);
2981 out:
2982 btrfs_free_path(path);
2983 err = btrfs_commit_transaction(trans, root);
2984 if (err && !ret)
2985 ret = err;
2986 return ret;
2987 }
2988
2989 /*
2990 * This is a heuristic used to reduce the number of chunks balanced on
2991 * resume after balance was interrupted.
2992 */
2993 static void update_balance_args(struct btrfs_balance_control *bctl)
2994 {
2995 /*
2996 * Turn on soft mode for chunk types that were being converted.
2997 */
2998 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2999 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3000 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3001 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3002 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3003 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3004
3005 /*
3006 * Turn on usage filter if is not already used. The idea is
3007 * that chunks that we have already balanced should be
3008 * reasonably full. Don't do it for chunks that are being
3009 * converted - that will keep us from relocating unconverted
3010 * (albeit full) chunks.
3011 */
3012 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3013 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3014 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3015 bctl->data.usage = 90;
3016 }
3017 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3018 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3019 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3020 bctl->sys.usage = 90;
3021 }
3022 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3023 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3024 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3025 bctl->meta.usage = 90;
3026 }
3027 }
3028
3029 /*
3030 * Should be called with both balance and volume mutexes held to
3031 * serialize other volume operations (add_dev/rm_dev/resize) with
3032 * restriper. Same goes for unset_balance_control.
3033 */
3034 static void set_balance_control(struct btrfs_balance_control *bctl)
3035 {
3036 struct btrfs_fs_info *fs_info = bctl->fs_info;
3037
3038 BUG_ON(fs_info->balance_ctl);
3039
3040 spin_lock(&fs_info->balance_lock);
3041 fs_info->balance_ctl = bctl;
3042 spin_unlock(&fs_info->balance_lock);
3043 }
3044
3045 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3046 {
3047 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3048
3049 BUG_ON(!fs_info->balance_ctl);
3050
3051 spin_lock(&fs_info->balance_lock);
3052 fs_info->balance_ctl = NULL;
3053 spin_unlock(&fs_info->balance_lock);
3054
3055 kfree(bctl);
3056 }
3057
3058 /*
3059 * Balance filters. Return 1 if chunk should be filtered out
3060 * (should not be balanced).
3061 */
3062 static int chunk_profiles_filter(u64 chunk_type,
3063 struct btrfs_balance_args *bargs)
3064 {
3065 chunk_type = chunk_to_extended(chunk_type) &
3066 BTRFS_EXTENDED_PROFILE_MASK;
3067
3068 if (bargs->profiles & chunk_type)
3069 return 0;
3070
3071 return 1;
3072 }
3073
3074 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3075 struct btrfs_balance_args *bargs)
3076 {
3077 struct btrfs_block_group_cache *cache;
3078 u64 chunk_used, user_thresh;
3079 int ret = 1;
3080
3081 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3082 chunk_used = btrfs_block_group_used(&cache->item);
3083
3084 if (bargs->usage == 0)
3085 user_thresh = 1;
3086 else if (bargs->usage > 100)
3087 user_thresh = cache->key.offset;
3088 else
3089 user_thresh = div_factor_fine(cache->key.offset,
3090 bargs->usage);
3091
3092 if (chunk_used < user_thresh)
3093 ret = 0;
3094
3095 btrfs_put_block_group(cache);
3096 return ret;
3097 }
3098
3099 static int chunk_devid_filter(struct extent_buffer *leaf,
3100 struct btrfs_chunk *chunk,
3101 struct btrfs_balance_args *bargs)
3102 {
3103 struct btrfs_stripe *stripe;
3104 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3105 int i;
3106
3107 for (i = 0; i < num_stripes; i++) {
3108 stripe = btrfs_stripe_nr(chunk, i);
3109 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3110 return 0;
3111 }
3112
3113 return 1;
3114 }
3115
3116 /* [pstart, pend) */
3117 static int chunk_drange_filter(struct extent_buffer *leaf,
3118 struct btrfs_chunk *chunk,
3119 u64 chunk_offset,
3120 struct btrfs_balance_args *bargs)
3121 {
3122 struct btrfs_stripe *stripe;
3123 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3124 u64 stripe_offset;
3125 u64 stripe_length;
3126 int factor;
3127 int i;
3128
3129 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3130 return 0;
3131
3132 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3133 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3134 factor = num_stripes / 2;
3135 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3136 factor = num_stripes - 1;
3137 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3138 factor = num_stripes - 2;
3139 } else {
3140 factor = num_stripes;
3141 }
3142
3143 for (i = 0; i < num_stripes; i++) {
3144 stripe = btrfs_stripe_nr(chunk, i);
3145 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3146 continue;
3147
3148 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3149 stripe_length = btrfs_chunk_length(leaf, chunk);
3150 stripe_length = div_u64(stripe_length, factor);
3151
3152 if (stripe_offset < bargs->pend &&
3153 stripe_offset + stripe_length > bargs->pstart)
3154 return 0;
3155 }
3156
3157 return 1;
3158 }
3159
3160 /* [vstart, vend) */
3161 static int chunk_vrange_filter(struct extent_buffer *leaf,
3162 struct btrfs_chunk *chunk,
3163 u64 chunk_offset,
3164 struct btrfs_balance_args *bargs)
3165 {
3166 if (chunk_offset < bargs->vend &&
3167 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3168 /* at least part of the chunk is inside this vrange */
3169 return 0;
3170
3171 return 1;
3172 }
3173
3174 static int chunk_soft_convert_filter(u64 chunk_type,
3175 struct btrfs_balance_args *bargs)
3176 {
3177 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3178 return 0;
3179
3180 chunk_type = chunk_to_extended(chunk_type) &
3181 BTRFS_EXTENDED_PROFILE_MASK;
3182
3183 if (bargs->target == chunk_type)
3184 return 1;
3185
3186 return 0;
3187 }
3188
3189 static int should_balance_chunk(struct btrfs_root *root,
3190 struct extent_buffer *leaf,
3191 struct btrfs_chunk *chunk, u64 chunk_offset)
3192 {
3193 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3194 struct btrfs_balance_args *bargs = NULL;
3195 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3196
3197 /* type filter */
3198 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3199 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3200 return 0;
3201 }
3202
3203 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3204 bargs = &bctl->data;
3205 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3206 bargs = &bctl->sys;
3207 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3208 bargs = &bctl->meta;
3209
3210 /* profiles filter */
3211 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3212 chunk_profiles_filter(chunk_type, bargs)) {
3213 return 0;
3214 }
3215
3216 /* usage filter */
3217 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3218 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3219 return 0;
3220 }
3221
3222 /* devid filter */
3223 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3224 chunk_devid_filter(leaf, chunk, bargs)) {
3225 return 0;
3226 }
3227
3228 /* drange filter, makes sense only with devid filter */
3229 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3230 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3231 return 0;
3232 }
3233
3234 /* vrange filter */
3235 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3236 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3237 return 0;
3238 }
3239
3240 /* soft profile changing mode */
3241 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3242 chunk_soft_convert_filter(chunk_type, bargs)) {
3243 return 0;
3244 }
3245
3246 /*
3247 * limited by count, must be the last filter
3248 */
3249 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3250 if (bargs->limit == 0)
3251 return 0;
3252 else
3253 bargs->limit--;
3254 }
3255
3256 return 1;
3257 }
3258
3259 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3260 {
3261 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3262 struct btrfs_root *chunk_root = fs_info->chunk_root;
3263 struct btrfs_root *dev_root = fs_info->dev_root;
3264 struct list_head *devices;
3265 struct btrfs_device *device;
3266 u64 old_size;
3267 u64 size_to_free;
3268 struct btrfs_chunk *chunk;
3269 struct btrfs_path *path;
3270 struct btrfs_key key;
3271 struct btrfs_key found_key;
3272 struct btrfs_trans_handle *trans;
3273 struct extent_buffer *leaf;
3274 int slot;
3275 int ret;
3276 int enospc_errors = 0;
3277 bool counting = true;
3278 u64 limit_data = bctl->data.limit;
3279 u64 limit_meta = bctl->meta.limit;
3280 u64 limit_sys = bctl->sys.limit;
3281
3282 /* step one make some room on all the devices */
3283 devices = &fs_info->fs_devices->devices;
3284 list_for_each_entry(device, devices, dev_list) {
3285 old_size = btrfs_device_get_total_bytes(device);
3286 size_to_free = div_factor(old_size, 1);
3287 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3288 if (!device->writeable ||
3289 btrfs_device_get_total_bytes(device) -
3290 btrfs_device_get_bytes_used(device) > size_to_free ||
3291 device->is_tgtdev_for_dev_replace)
3292 continue;
3293
3294 ret = btrfs_shrink_device(device, old_size - size_to_free);
3295 if (ret == -ENOSPC)
3296 break;
3297 BUG_ON(ret);
3298
3299 trans = btrfs_start_transaction(dev_root, 0);
3300 BUG_ON(IS_ERR(trans));
3301
3302 ret = btrfs_grow_device(trans, device, old_size);
3303 BUG_ON(ret);
3304
3305 btrfs_end_transaction(trans, dev_root);
3306 }
3307
3308 /* step two, relocate all the chunks */
3309 path = btrfs_alloc_path();
3310 if (!path) {
3311 ret = -ENOMEM;
3312 goto error;
3313 }
3314
3315 /* zero out stat counters */
3316 spin_lock(&fs_info->balance_lock);
3317 memset(&bctl->stat, 0, sizeof(bctl->stat));
3318 spin_unlock(&fs_info->balance_lock);
3319 again:
3320 if (!counting) {
3321 bctl->data.limit = limit_data;
3322 bctl->meta.limit = limit_meta;
3323 bctl->sys.limit = limit_sys;
3324 }
3325 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3326 key.offset = (u64)-1;
3327 key.type = BTRFS_CHUNK_ITEM_KEY;
3328
3329 while (1) {
3330 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3331 atomic_read(&fs_info->balance_cancel_req)) {
3332 ret = -ECANCELED;
3333 goto error;
3334 }
3335
3336 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3337 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3338 if (ret < 0) {
3339 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3340 goto error;
3341 }
3342
3343 /*
3344 * this shouldn't happen, it means the last relocate
3345 * failed
3346 */
3347 if (ret == 0)
3348 BUG(); /* FIXME break ? */
3349
3350 ret = btrfs_previous_item(chunk_root, path, 0,
3351 BTRFS_CHUNK_ITEM_KEY);
3352 if (ret) {
3353 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3354 ret = 0;
3355 break;
3356 }
3357
3358 leaf = path->nodes[0];
3359 slot = path->slots[0];
3360 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3361
3362 if (found_key.objectid != key.objectid) {
3363 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3364 break;
3365 }
3366
3367 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3368
3369 if (!counting) {
3370 spin_lock(&fs_info->balance_lock);
3371 bctl->stat.considered++;
3372 spin_unlock(&fs_info->balance_lock);
3373 }
3374
3375 ret = should_balance_chunk(chunk_root, leaf, chunk,
3376 found_key.offset);
3377 btrfs_release_path(path);
3378 if (!ret) {
3379 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3380 goto loop;
3381 }
3382
3383 if (counting) {
3384 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3385 spin_lock(&fs_info->balance_lock);
3386 bctl->stat.expected++;
3387 spin_unlock(&fs_info->balance_lock);
3388 goto loop;
3389 }
3390
3391 ret = btrfs_relocate_chunk(chunk_root,
3392 found_key.objectid,
3393 found_key.offset);
3394 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3395 if (ret && ret != -ENOSPC)
3396 goto error;
3397 if (ret == -ENOSPC) {
3398 enospc_errors++;
3399 } else {
3400 spin_lock(&fs_info->balance_lock);
3401 bctl->stat.completed++;
3402 spin_unlock(&fs_info->balance_lock);
3403 }
3404 loop:
3405 if (found_key.offset == 0)
3406 break;
3407 key.offset = found_key.offset - 1;
3408 }
3409
3410 if (counting) {
3411 btrfs_release_path(path);
3412 counting = false;
3413 goto again;
3414 }
3415 error:
3416 btrfs_free_path(path);
3417 if (enospc_errors) {
3418 btrfs_info(fs_info, "%d enospc errors during balance",
3419 enospc_errors);
3420 if (!ret)
3421 ret = -ENOSPC;
3422 }
3423
3424 return ret;
3425 }
3426
3427 /**
3428 * alloc_profile_is_valid - see if a given profile is valid and reduced
3429 * @flags: profile to validate
3430 * @extended: if true @flags is treated as an extended profile
3431 */
3432 static int alloc_profile_is_valid(u64 flags, int extended)
3433 {
3434 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3435 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3436
3437 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3438
3439 /* 1) check that all other bits are zeroed */
3440 if (flags & ~mask)
3441 return 0;
3442
3443 /* 2) see if profile is reduced */
3444 if (flags == 0)
3445 return !extended; /* "0" is valid for usual profiles */
3446
3447 /* true if exactly one bit set */
3448 return (flags & (flags - 1)) == 0;
3449 }
3450
3451 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3452 {
3453 /* cancel requested || normal exit path */
3454 return atomic_read(&fs_info->balance_cancel_req) ||
3455 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3456 atomic_read(&fs_info->balance_cancel_req) == 0);
3457 }
3458
3459 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3460 {
3461 int ret;
3462
3463 unset_balance_control(fs_info);
3464 ret = del_balance_item(fs_info->tree_root);
3465 if (ret)
3466 btrfs_std_error(fs_info, ret);
3467
3468 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3469 }
3470
3471 /*
3472 * Should be called with both balance and volume mutexes held
3473 */
3474 int btrfs_balance(struct btrfs_balance_control *bctl,
3475 struct btrfs_ioctl_balance_args *bargs)
3476 {
3477 struct btrfs_fs_info *fs_info = bctl->fs_info;
3478 u64 allowed;
3479 int mixed = 0;
3480 int ret;
3481 u64 num_devices;
3482 unsigned seq;
3483
3484 if (btrfs_fs_closing(fs_info) ||
3485 atomic_read(&fs_info->balance_pause_req) ||
3486 atomic_read(&fs_info->balance_cancel_req)) {
3487 ret = -EINVAL;
3488 goto out;
3489 }
3490
3491 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3492 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3493 mixed = 1;
3494
3495 /*
3496 * In case of mixed groups both data and meta should be picked,
3497 * and identical options should be given for both of them.
3498 */
3499 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3500 if (mixed && (bctl->flags & allowed)) {
3501 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3502 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3503 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3504 btrfs_err(fs_info, "with mixed groups data and "
3505 "metadata balance options must be the same");
3506 ret = -EINVAL;
3507 goto out;
3508 }
3509 }
3510
3511 num_devices = fs_info->fs_devices->num_devices;
3512 btrfs_dev_replace_lock(&fs_info->dev_replace);
3513 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3514 BUG_ON(num_devices < 1);
3515 num_devices--;
3516 }
3517 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3518 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3519 if (num_devices == 1)
3520 allowed |= BTRFS_BLOCK_GROUP_DUP;
3521 else if (num_devices > 1)
3522 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3523 if (num_devices > 2)
3524 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3525 if (num_devices > 3)
3526 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3527 BTRFS_BLOCK_GROUP_RAID6);
3528 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3529 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3530 (bctl->data.target & ~allowed))) {
3531 btrfs_err(fs_info, "unable to start balance with target "
3532 "data profile %llu",
3533 bctl->data.target);
3534 ret = -EINVAL;
3535 goto out;
3536 }
3537 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3538 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3539 (bctl->meta.target & ~allowed))) {
3540 btrfs_err(fs_info,
3541 "unable to start balance with target metadata profile %llu",
3542 bctl->meta.target);
3543 ret = -EINVAL;
3544 goto out;
3545 }
3546 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3547 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3548 (bctl->sys.target & ~allowed))) {
3549 btrfs_err(fs_info,
3550 "unable to start balance with target system profile %llu",
3551 bctl->sys.target);
3552 ret = -EINVAL;
3553 goto out;
3554 }
3555
3556 /* allow dup'ed data chunks only in mixed mode */
3557 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3558 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3559 btrfs_err(fs_info, "dup for data is not allowed");
3560 ret = -EINVAL;
3561 goto out;
3562 }
3563
3564 /* allow to reduce meta or sys integrity only if force set */
3565 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3566 BTRFS_BLOCK_GROUP_RAID10 |
3567 BTRFS_BLOCK_GROUP_RAID5 |
3568 BTRFS_BLOCK_GROUP_RAID6;
3569 do {
3570 seq = read_seqbegin(&fs_info->profiles_lock);
3571
3572 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3573 (fs_info->avail_system_alloc_bits & allowed) &&
3574 !(bctl->sys.target & allowed)) ||
3575 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3576 (fs_info->avail_metadata_alloc_bits & allowed) &&
3577 !(bctl->meta.target & allowed))) {
3578 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3579 btrfs_info(fs_info, "force reducing metadata integrity");
3580 } else {
3581 btrfs_err(fs_info, "balance will reduce metadata "
3582 "integrity, use force if you want this");
3583 ret = -EINVAL;
3584 goto out;
3585 }
3586 }
3587 } while (read_seqretry(&fs_info->profiles_lock, seq));
3588
3589 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3590 int num_tolerated_disk_barrier_failures;
3591 u64 target = bctl->sys.target;
3592
3593 num_tolerated_disk_barrier_failures =
3594 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3595 if (num_tolerated_disk_barrier_failures > 0 &&
3596 (target &
3597 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3598 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3599 num_tolerated_disk_barrier_failures = 0;
3600 else if (num_tolerated_disk_barrier_failures > 1 &&
3601 (target &
3602 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3603 num_tolerated_disk_barrier_failures = 1;
3604
3605 fs_info->num_tolerated_disk_barrier_failures =
3606 num_tolerated_disk_barrier_failures;
3607 }
3608
3609 ret = insert_balance_item(fs_info->tree_root, bctl);
3610 if (ret && ret != -EEXIST)
3611 goto out;
3612
3613 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3614 BUG_ON(ret == -EEXIST);
3615 set_balance_control(bctl);
3616 } else {
3617 BUG_ON(ret != -EEXIST);
3618 spin_lock(&fs_info->balance_lock);
3619 update_balance_args(bctl);
3620 spin_unlock(&fs_info->balance_lock);
3621 }
3622
3623 atomic_inc(&fs_info->balance_running);
3624 mutex_unlock(&fs_info->balance_mutex);
3625
3626 ret = __btrfs_balance(fs_info);
3627
3628 mutex_lock(&fs_info->balance_mutex);
3629 atomic_dec(&fs_info->balance_running);
3630
3631 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3632 fs_info->num_tolerated_disk_barrier_failures =
3633 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3634 }
3635
3636 if (bargs) {
3637 memset(bargs, 0, sizeof(*bargs));
3638 update_ioctl_balance_args(fs_info, 0, bargs);
3639 }
3640
3641 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3642 balance_need_close(fs_info)) {
3643 __cancel_balance(fs_info);
3644 }
3645
3646 wake_up(&fs_info->balance_wait_q);
3647
3648 return ret;
3649 out:
3650 if (bctl->flags & BTRFS_BALANCE_RESUME)
3651 __cancel_balance(fs_info);
3652 else {
3653 kfree(bctl);
3654 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3655 }
3656 return ret;
3657 }
3658
3659 static int balance_kthread(void *data)
3660 {
3661 struct btrfs_fs_info *fs_info = data;
3662 int ret = 0;
3663
3664 mutex_lock(&fs_info->volume_mutex);
3665 mutex_lock(&fs_info->balance_mutex);
3666
3667 if (fs_info->balance_ctl) {
3668 btrfs_info(fs_info, "continuing balance");
3669 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3670 }
3671
3672 mutex_unlock(&fs_info->balance_mutex);
3673 mutex_unlock(&fs_info->volume_mutex);
3674
3675 return ret;
3676 }
3677
3678 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3679 {
3680 struct task_struct *tsk;
3681
3682 spin_lock(&fs_info->balance_lock);
3683 if (!fs_info->balance_ctl) {
3684 spin_unlock(&fs_info->balance_lock);
3685 return 0;
3686 }
3687 spin_unlock(&fs_info->balance_lock);
3688
3689 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3690 btrfs_info(fs_info, "force skipping balance");
3691 return 0;
3692 }
3693
3694 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3695 return PTR_ERR_OR_ZERO(tsk);
3696 }
3697
3698 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3699 {
3700 struct btrfs_balance_control *bctl;
3701 struct btrfs_balance_item *item;
3702 struct btrfs_disk_balance_args disk_bargs;
3703 struct btrfs_path *path;
3704 struct extent_buffer *leaf;
3705 struct btrfs_key key;
3706 int ret;
3707
3708 path = btrfs_alloc_path();
3709 if (!path)
3710 return -ENOMEM;
3711
3712 key.objectid = BTRFS_BALANCE_OBJECTID;
3713 key.type = BTRFS_BALANCE_ITEM_KEY;
3714 key.offset = 0;
3715
3716 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3717 if (ret < 0)
3718 goto out;
3719 if (ret > 0) { /* ret = -ENOENT; */
3720 ret = 0;
3721 goto out;
3722 }
3723
3724 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3725 if (!bctl) {
3726 ret = -ENOMEM;
3727 goto out;
3728 }
3729
3730 leaf = path->nodes[0];
3731 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3732
3733 bctl->fs_info = fs_info;
3734 bctl->flags = btrfs_balance_flags(leaf, item);
3735 bctl->flags |= BTRFS_BALANCE_RESUME;
3736
3737 btrfs_balance_data(leaf, item, &disk_bargs);
3738 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3739 btrfs_balance_meta(leaf, item, &disk_bargs);
3740 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3741 btrfs_balance_sys(leaf, item, &disk_bargs);
3742 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3743
3744 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3745
3746 mutex_lock(&fs_info->volume_mutex);
3747 mutex_lock(&fs_info->balance_mutex);
3748
3749 set_balance_control(bctl);
3750
3751 mutex_unlock(&fs_info->balance_mutex);
3752 mutex_unlock(&fs_info->volume_mutex);
3753 out:
3754 btrfs_free_path(path);
3755 return ret;
3756 }
3757
3758 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3759 {
3760 int ret = 0;
3761
3762 mutex_lock(&fs_info->balance_mutex);
3763 if (!fs_info->balance_ctl) {
3764 mutex_unlock(&fs_info->balance_mutex);
3765 return -ENOTCONN;
3766 }
3767
3768 if (atomic_read(&fs_info->balance_running)) {
3769 atomic_inc(&fs_info->balance_pause_req);
3770 mutex_unlock(&fs_info->balance_mutex);
3771
3772 wait_event(fs_info->balance_wait_q,
3773 atomic_read(&fs_info->balance_running) == 0);
3774
3775 mutex_lock(&fs_info->balance_mutex);
3776 /* we are good with balance_ctl ripped off from under us */
3777 BUG_ON(atomic_read(&fs_info->balance_running));
3778 atomic_dec(&fs_info->balance_pause_req);
3779 } else {
3780 ret = -ENOTCONN;
3781 }
3782
3783 mutex_unlock(&fs_info->balance_mutex);
3784 return ret;
3785 }
3786
3787 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3788 {
3789 if (fs_info->sb->s_flags & MS_RDONLY)
3790 return -EROFS;
3791
3792 mutex_lock(&fs_info->balance_mutex);
3793 if (!fs_info->balance_ctl) {
3794 mutex_unlock(&fs_info->balance_mutex);
3795 return -ENOTCONN;
3796 }
3797
3798 atomic_inc(&fs_info->balance_cancel_req);
3799 /*
3800 * if we are running just wait and return, balance item is
3801 * deleted in btrfs_balance in this case
3802 */
3803 if (atomic_read(&fs_info->balance_running)) {
3804 mutex_unlock(&fs_info->balance_mutex);
3805 wait_event(fs_info->balance_wait_q,
3806 atomic_read(&fs_info->balance_running) == 0);
3807 mutex_lock(&fs_info->balance_mutex);
3808 } else {
3809 /* __cancel_balance needs volume_mutex */
3810 mutex_unlock(&fs_info->balance_mutex);
3811 mutex_lock(&fs_info->volume_mutex);
3812 mutex_lock(&fs_info->balance_mutex);
3813
3814 if (fs_info->balance_ctl)
3815 __cancel_balance(fs_info);
3816
3817 mutex_unlock(&fs_info->volume_mutex);
3818 }
3819
3820 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3821 atomic_dec(&fs_info->balance_cancel_req);
3822 mutex_unlock(&fs_info->balance_mutex);
3823 return 0;
3824 }
3825
3826 static int btrfs_uuid_scan_kthread(void *data)
3827 {
3828 struct btrfs_fs_info *fs_info = data;
3829 struct btrfs_root *root = fs_info->tree_root;
3830 struct btrfs_key key;
3831 struct btrfs_key max_key;
3832 struct btrfs_path *path = NULL;
3833 int ret = 0;
3834 struct extent_buffer *eb;
3835 int slot;
3836 struct btrfs_root_item root_item;
3837 u32 item_size;
3838 struct btrfs_trans_handle *trans = NULL;
3839
3840 path = btrfs_alloc_path();
3841 if (!path) {
3842 ret = -ENOMEM;
3843 goto out;
3844 }
3845
3846 key.objectid = 0;
3847 key.type = BTRFS_ROOT_ITEM_KEY;
3848 key.offset = 0;
3849
3850 max_key.objectid = (u64)-1;
3851 max_key.type = BTRFS_ROOT_ITEM_KEY;
3852 max_key.offset = (u64)-1;
3853
3854 while (1) {
3855 ret = btrfs_search_forward(root, &key, path, 0);
3856 if (ret) {
3857 if (ret > 0)
3858 ret = 0;
3859 break;
3860 }
3861
3862 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3863 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3864 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3865 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3866 goto skip;
3867
3868 eb = path->nodes[0];
3869 slot = path->slots[0];
3870 item_size = btrfs_item_size_nr(eb, slot);
3871 if (item_size < sizeof(root_item))
3872 goto skip;
3873
3874 read_extent_buffer(eb, &root_item,
3875 btrfs_item_ptr_offset(eb, slot),
3876 (int)sizeof(root_item));
3877 if (btrfs_root_refs(&root_item) == 0)
3878 goto skip;
3879
3880 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3881 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3882 if (trans)
3883 goto update_tree;
3884
3885 btrfs_release_path(path);
3886 /*
3887 * 1 - subvol uuid item
3888 * 1 - received_subvol uuid item
3889 */
3890 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3891 if (IS_ERR(trans)) {
3892 ret = PTR_ERR(trans);
3893 break;
3894 }
3895 continue;
3896 } else {
3897 goto skip;
3898 }
3899 update_tree:
3900 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3901 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3902 root_item.uuid,
3903 BTRFS_UUID_KEY_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 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3913 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3914 root_item.received_uuid,
3915 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3916 key.objectid);
3917 if (ret < 0) {
3918 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3919 ret);
3920 break;
3921 }
3922 }
3923
3924 skip:
3925 if (trans) {
3926 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3927 trans = NULL;
3928 if (ret)
3929 break;
3930 }
3931
3932 btrfs_release_path(path);
3933 if (key.offset < (u64)-1) {
3934 key.offset++;
3935 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3936 key.offset = 0;
3937 key.type = BTRFS_ROOT_ITEM_KEY;
3938 } else if (key.objectid < (u64)-1) {
3939 key.offset = 0;
3940 key.type = BTRFS_ROOT_ITEM_KEY;
3941 key.objectid++;
3942 } else {
3943 break;
3944 }
3945 cond_resched();
3946 }
3947
3948 out:
3949 btrfs_free_path(path);
3950 if (trans && !IS_ERR(trans))
3951 btrfs_end_transaction(trans, fs_info->uuid_root);
3952 if (ret)
3953 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3954 else
3955 fs_info->update_uuid_tree_gen = 1;
3956 up(&fs_info->uuid_tree_rescan_sem);
3957 return 0;
3958 }
3959
3960 /*
3961 * Callback for btrfs_uuid_tree_iterate().
3962 * returns:
3963 * 0 check succeeded, the entry is not outdated.
3964 * < 0 if an error occured.
3965 * > 0 if the check failed, which means the caller shall remove the entry.
3966 */
3967 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3968 u8 *uuid, u8 type, u64 subid)
3969 {
3970 struct btrfs_key key;
3971 int ret = 0;
3972 struct btrfs_root *subvol_root;
3973
3974 if (type != BTRFS_UUID_KEY_SUBVOL &&
3975 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3976 goto out;
3977
3978 key.objectid = subid;
3979 key.type = BTRFS_ROOT_ITEM_KEY;
3980 key.offset = (u64)-1;
3981 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3982 if (IS_ERR(subvol_root)) {
3983 ret = PTR_ERR(subvol_root);
3984 if (ret == -ENOENT)
3985 ret = 1;
3986 goto out;
3987 }
3988
3989 switch (type) {
3990 case BTRFS_UUID_KEY_SUBVOL:
3991 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3992 ret = 1;
3993 break;
3994 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3995 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3996 BTRFS_UUID_SIZE))
3997 ret = 1;
3998 break;
3999 }
4000
4001 out:
4002 return ret;
4003 }
4004
4005 static int btrfs_uuid_rescan_kthread(void *data)
4006 {
4007 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4008 int ret;
4009
4010 /*
4011 * 1st step is to iterate through the existing UUID tree and
4012 * to delete all entries that contain outdated data.
4013 * 2nd step is to add all missing entries to the UUID tree.
4014 */
4015 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4016 if (ret < 0) {
4017 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4018 up(&fs_info->uuid_tree_rescan_sem);
4019 return ret;
4020 }
4021 return btrfs_uuid_scan_kthread(data);
4022 }
4023
4024 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4025 {
4026 struct btrfs_trans_handle *trans;
4027 struct btrfs_root *tree_root = fs_info->tree_root;
4028 struct btrfs_root *uuid_root;
4029 struct task_struct *task;
4030 int ret;
4031
4032 /*
4033 * 1 - root node
4034 * 1 - root item
4035 */
4036 trans = btrfs_start_transaction(tree_root, 2);
4037 if (IS_ERR(trans))
4038 return PTR_ERR(trans);
4039
4040 uuid_root = btrfs_create_tree(trans, fs_info,
4041 BTRFS_UUID_TREE_OBJECTID);
4042 if (IS_ERR(uuid_root)) {
4043 ret = PTR_ERR(uuid_root);
4044 btrfs_abort_transaction(trans, tree_root, ret);
4045 return ret;
4046 }
4047
4048 fs_info->uuid_root = uuid_root;
4049
4050 ret = btrfs_commit_transaction(trans, tree_root);
4051 if (ret)
4052 return ret;
4053
4054 down(&fs_info->uuid_tree_rescan_sem);
4055 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4056 if (IS_ERR(task)) {
4057 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4058 btrfs_warn(fs_info, "failed to start uuid_scan task");
4059 up(&fs_info->uuid_tree_rescan_sem);
4060 return PTR_ERR(task);
4061 }
4062
4063 return 0;
4064 }
4065
4066 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4067 {
4068 struct task_struct *task;
4069
4070 down(&fs_info->uuid_tree_rescan_sem);
4071 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4072 if (IS_ERR(task)) {
4073 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4074 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4075 up(&fs_info->uuid_tree_rescan_sem);
4076 return PTR_ERR(task);
4077 }
4078
4079 return 0;
4080 }
4081
4082 /*
4083 * shrinking a device means finding all of the device extents past
4084 * the new size, and then following the back refs to the chunks.
4085 * The chunk relocation code actually frees the device extent
4086 */
4087 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4088 {
4089 struct btrfs_trans_handle *trans;
4090 struct btrfs_root *root = device->dev_root;
4091 struct btrfs_dev_extent *dev_extent = NULL;
4092 struct btrfs_path *path;
4093 u64 length;
4094 u64 chunk_objectid;
4095 u64 chunk_offset;
4096 int ret;
4097 int slot;
4098 int failed = 0;
4099 bool retried = false;
4100 bool checked_pending_chunks = false;
4101 struct extent_buffer *l;
4102 struct btrfs_key key;
4103 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4104 u64 old_total = btrfs_super_total_bytes(super_copy);
4105 u64 old_size = btrfs_device_get_total_bytes(device);
4106 u64 diff = old_size - new_size;
4107
4108 if (device->is_tgtdev_for_dev_replace)
4109 return -EINVAL;
4110
4111 path = btrfs_alloc_path();
4112 if (!path)
4113 return -ENOMEM;
4114
4115 path->reada = 2;
4116
4117 lock_chunks(root);
4118
4119 btrfs_device_set_total_bytes(device, new_size);
4120 if (device->writeable) {
4121 device->fs_devices->total_rw_bytes -= diff;
4122 spin_lock(&root->fs_info->free_chunk_lock);
4123 root->fs_info->free_chunk_space -= diff;
4124 spin_unlock(&root->fs_info->free_chunk_lock);
4125 }
4126 unlock_chunks(root);
4127
4128 again:
4129 key.objectid = device->devid;
4130 key.offset = (u64)-1;
4131 key.type = BTRFS_DEV_EXTENT_KEY;
4132
4133 do {
4134 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4135 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4136 if (ret < 0) {
4137 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4138 goto done;
4139 }
4140
4141 ret = btrfs_previous_item(root, path, 0, key.type);
4142 if (ret)
4143 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4144 if (ret < 0)
4145 goto done;
4146 if (ret) {
4147 ret = 0;
4148 btrfs_release_path(path);
4149 break;
4150 }
4151
4152 l = path->nodes[0];
4153 slot = path->slots[0];
4154 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4155
4156 if (key.objectid != device->devid) {
4157 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4158 btrfs_release_path(path);
4159 break;
4160 }
4161
4162 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4163 length = btrfs_dev_extent_length(l, dev_extent);
4164
4165 if (key.offset + length <= new_size) {
4166 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4167 btrfs_release_path(path);
4168 break;
4169 }
4170
4171 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4172 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4173 btrfs_release_path(path);
4174
4175 ret = btrfs_relocate_chunk(root, chunk_objectid, chunk_offset);
4176 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4177 if (ret && ret != -ENOSPC)
4178 goto done;
4179 if (ret == -ENOSPC)
4180 failed++;
4181 } while (key.offset-- > 0);
4182
4183 if (failed && !retried) {
4184 failed = 0;
4185 retried = true;
4186 goto again;
4187 } else if (failed && retried) {
4188 ret = -ENOSPC;
4189 goto done;
4190 }
4191
4192 /* Shrinking succeeded, else we would be at "done". */
4193 trans = btrfs_start_transaction(root, 0);
4194 if (IS_ERR(trans)) {
4195 ret = PTR_ERR(trans);
4196 goto done;
4197 }
4198
4199 lock_chunks(root);
4200
4201 /*
4202 * We checked in the above loop all device extents that were already in
4203 * the device tree. However before we have updated the device's
4204 * total_bytes to the new size, we might have had chunk allocations that
4205 * have not complete yet (new block groups attached to transaction
4206 * handles), and therefore their device extents were not yet in the
4207 * device tree and we missed them in the loop above. So if we have any
4208 * pending chunk using a device extent that overlaps the device range
4209 * that we can not use anymore, commit the current transaction and
4210 * repeat the search on the device tree - this way we guarantee we will
4211 * not have chunks using device extents that end beyond 'new_size'.
4212 */
4213 if (!checked_pending_chunks) {
4214 u64 start = new_size;
4215 u64 len = old_size - new_size;
4216
4217 if (contains_pending_extent(trans->transaction, device,
4218 &start, len)) {
4219 unlock_chunks(root);
4220 checked_pending_chunks = true;
4221 failed = 0;
4222 retried = false;
4223 ret = btrfs_commit_transaction(trans, root);
4224 if (ret)
4225 goto done;
4226 goto again;
4227 }
4228 }
4229
4230 btrfs_device_set_disk_total_bytes(device, new_size);
4231 if (list_empty(&device->resized_list))
4232 list_add_tail(&device->resized_list,
4233 &root->fs_info->fs_devices->resized_devices);
4234
4235 WARN_ON(diff > old_total);
4236 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4237 unlock_chunks(root);
4238
4239 /* Now btrfs_update_device() will change the on-disk size. */
4240 ret = btrfs_update_device(trans, device);
4241 btrfs_end_transaction(trans, root);
4242 done:
4243 btrfs_free_path(path);
4244 if (ret) {
4245 lock_chunks(root);
4246 btrfs_device_set_total_bytes(device, old_size);
4247 if (device->writeable)
4248 device->fs_devices->total_rw_bytes += diff;
4249 spin_lock(&root->fs_info->free_chunk_lock);
4250 root->fs_info->free_chunk_space += diff;
4251 spin_unlock(&root->fs_info->free_chunk_lock);
4252 unlock_chunks(root);
4253 }
4254 return ret;
4255 }
4256
4257 static int btrfs_add_system_chunk(struct btrfs_root *root,
4258 struct btrfs_key *key,
4259 struct btrfs_chunk *chunk, int item_size)
4260 {
4261 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4262 struct btrfs_disk_key disk_key;
4263 u32 array_size;
4264 u8 *ptr;
4265
4266 lock_chunks(root);
4267 array_size = btrfs_super_sys_array_size(super_copy);
4268 if (array_size + item_size + sizeof(disk_key)
4269 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4270 unlock_chunks(root);
4271 return -EFBIG;
4272 }
4273
4274 ptr = super_copy->sys_chunk_array + array_size;
4275 btrfs_cpu_key_to_disk(&disk_key, key);
4276 memcpy(ptr, &disk_key, sizeof(disk_key));
4277 ptr += sizeof(disk_key);
4278 memcpy(ptr, chunk, item_size);
4279 item_size += sizeof(disk_key);
4280 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4281 unlock_chunks(root);
4282
4283 return 0;
4284 }
4285
4286 /*
4287 * sort the devices in descending order by max_avail, total_avail
4288 */
4289 static int btrfs_cmp_device_info(const void *a, const void *b)
4290 {
4291 const struct btrfs_device_info *di_a = a;
4292 const struct btrfs_device_info *di_b = b;
4293
4294 if (di_a->max_avail > di_b->max_avail)
4295 return -1;
4296 if (di_a->max_avail < di_b->max_avail)
4297 return 1;
4298 if (di_a->total_avail > di_b->total_avail)
4299 return -1;
4300 if (di_a->total_avail < di_b->total_avail)
4301 return 1;
4302 return 0;
4303 }
4304
4305 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4306 [BTRFS_RAID_RAID10] = {
4307 .sub_stripes = 2,
4308 .dev_stripes = 1,
4309 .devs_max = 0, /* 0 == as many as possible */
4310 .devs_min = 4,
4311 .devs_increment = 2,
4312 .ncopies = 2,
4313 },
4314 [BTRFS_RAID_RAID1] = {
4315 .sub_stripes = 1,
4316 .dev_stripes = 1,
4317 .devs_max = 2,
4318 .devs_min = 2,
4319 .devs_increment = 2,
4320 .ncopies = 2,
4321 },
4322 [BTRFS_RAID_DUP] = {
4323 .sub_stripes = 1,
4324 .dev_stripes = 2,
4325 .devs_max = 1,
4326 .devs_min = 1,
4327 .devs_increment = 1,
4328 .ncopies = 2,
4329 },
4330 [BTRFS_RAID_RAID0] = {
4331 .sub_stripes = 1,
4332 .dev_stripes = 1,
4333 .devs_max = 0,
4334 .devs_min = 2,
4335 .devs_increment = 1,
4336 .ncopies = 1,
4337 },
4338 [BTRFS_RAID_SINGLE] = {
4339 .sub_stripes = 1,
4340 .dev_stripes = 1,
4341 .devs_max = 1,
4342 .devs_min = 1,
4343 .devs_increment = 1,
4344 .ncopies = 1,
4345 },
4346 [BTRFS_RAID_RAID5] = {
4347 .sub_stripes = 1,
4348 .dev_stripes = 1,
4349 .devs_max = 0,
4350 .devs_min = 2,
4351 .devs_increment = 1,
4352 .ncopies = 2,
4353 },
4354 [BTRFS_RAID_RAID6] = {
4355 .sub_stripes = 1,
4356 .dev_stripes = 1,
4357 .devs_max = 0,
4358 .devs_min = 3,
4359 .devs_increment = 1,
4360 .ncopies = 3,
4361 },
4362 };
4363
4364 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4365 {
4366 /* TODO allow them to set a preferred stripe size */
4367 return 64 * 1024;
4368 }
4369
4370 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4371 {
4372 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4373 return;
4374
4375 btrfs_set_fs_incompat(info, RAID56);
4376 }
4377
4378 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4379 - sizeof(struct btrfs_item) \
4380 - sizeof(struct btrfs_chunk)) \
4381 / sizeof(struct btrfs_stripe) + 1)
4382
4383 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4384 - 2 * sizeof(struct btrfs_disk_key) \
4385 - 2 * sizeof(struct btrfs_chunk)) \
4386 / sizeof(struct btrfs_stripe) + 1)
4387
4388 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4389 struct btrfs_root *extent_root, u64 start,
4390 u64 type)
4391 {
4392 struct btrfs_fs_info *info = extent_root->fs_info;
4393 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4394 struct list_head *cur;
4395 struct map_lookup *map = NULL;
4396 struct extent_map_tree *em_tree;
4397 struct extent_map *em;
4398 struct btrfs_device_info *devices_info = NULL;
4399 u64 total_avail;
4400 int num_stripes; /* total number of stripes to allocate */
4401 int data_stripes; /* number of stripes that count for
4402 block group size */
4403 int sub_stripes; /* sub_stripes info for map */
4404 int dev_stripes; /* stripes per dev */
4405 int devs_max; /* max devs to use */
4406 int devs_min; /* min devs needed */
4407 int devs_increment; /* ndevs has to be a multiple of this */
4408 int ncopies; /* how many copies to data has */
4409 int ret;
4410 u64 max_stripe_size;
4411 u64 max_chunk_size;
4412 u64 stripe_size;
4413 u64 num_bytes;
4414 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4415 int ndevs;
4416 int i;
4417 int j;
4418 int index;
4419
4420 BUG_ON(!alloc_profile_is_valid(type, 0));
4421
4422 if (list_empty(&fs_devices->alloc_list))
4423 return -ENOSPC;
4424
4425 index = __get_raid_index(type);
4426
4427 sub_stripes = btrfs_raid_array[index].sub_stripes;
4428 dev_stripes = btrfs_raid_array[index].dev_stripes;
4429 devs_max = btrfs_raid_array[index].devs_max;
4430 devs_min = btrfs_raid_array[index].devs_min;
4431 devs_increment = btrfs_raid_array[index].devs_increment;
4432 ncopies = btrfs_raid_array[index].ncopies;
4433
4434 if (type & BTRFS_BLOCK_GROUP_DATA) {
4435 max_stripe_size = 1024 * 1024 * 1024;
4436 max_chunk_size = 10 * max_stripe_size;
4437 if (!devs_max)
4438 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4439 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4440 /* for larger filesystems, use larger metadata chunks */
4441 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4442 max_stripe_size = 1024 * 1024 * 1024;
4443 else
4444 max_stripe_size = 256 * 1024 * 1024;
4445 max_chunk_size = max_stripe_size;
4446 if (!devs_max)
4447 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4448 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4449 max_stripe_size = 32 * 1024 * 1024;
4450 max_chunk_size = 2 * max_stripe_size;
4451 if (!devs_max)
4452 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4453 } else {
4454 btrfs_err(info, "invalid chunk type 0x%llx requested",
4455 type);
4456 BUG_ON(1);
4457 }
4458
4459 /* we don't want a chunk larger than 10% of writeable space */
4460 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4461 max_chunk_size);
4462
4463 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4464 GFP_NOFS);
4465 if (!devices_info)
4466 return -ENOMEM;
4467
4468 cur = fs_devices->alloc_list.next;
4469
4470 /*
4471 * in the first pass through the devices list, we gather information
4472 * about the available holes on each device.
4473 */
4474 ndevs = 0;
4475 while (cur != &fs_devices->alloc_list) {
4476 struct btrfs_device *device;
4477 u64 max_avail;
4478 u64 dev_offset;
4479
4480 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4481
4482 cur = cur->next;
4483
4484 if (!device->writeable) {
4485 WARN(1, KERN_ERR
4486 "BTRFS: read-only device in alloc_list\n");
4487 continue;
4488 }
4489
4490 if (!device->in_fs_metadata ||
4491 device->is_tgtdev_for_dev_replace)
4492 continue;
4493
4494 if (device->total_bytes > device->bytes_used)
4495 total_avail = device->total_bytes - device->bytes_used;
4496 else
4497 total_avail = 0;
4498
4499 /* If there is no space on this device, skip it. */
4500 if (total_avail == 0)
4501 continue;
4502
4503 ret = find_free_dev_extent(trans, device,
4504 max_stripe_size * dev_stripes,
4505 &dev_offset, &max_avail);
4506 if (ret && ret != -ENOSPC)
4507 goto error;
4508
4509 if (ret == 0)
4510 max_avail = max_stripe_size * dev_stripes;
4511
4512 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4513 continue;
4514
4515 if (ndevs == fs_devices->rw_devices) {
4516 WARN(1, "%s: found more than %llu devices\n",
4517 __func__, fs_devices->rw_devices);
4518 break;
4519 }
4520 devices_info[ndevs].dev_offset = dev_offset;
4521 devices_info[ndevs].max_avail = max_avail;
4522 devices_info[ndevs].total_avail = total_avail;
4523 devices_info[ndevs].dev = device;
4524 ++ndevs;
4525 }
4526
4527 /*
4528 * now sort the devices by hole size / available space
4529 */
4530 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4531 btrfs_cmp_device_info, NULL);
4532
4533 /* round down to number of usable stripes */
4534 ndevs -= ndevs % devs_increment;
4535
4536 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4537 ret = -ENOSPC;
4538 goto error;
4539 }
4540
4541 if (devs_max && ndevs > devs_max)
4542 ndevs = devs_max;
4543 /*
4544 * the primary goal is to maximize the number of stripes, so use as many
4545 * devices as possible, even if the stripes are not maximum sized.
4546 */
4547 stripe_size = devices_info[ndevs-1].max_avail;
4548 num_stripes = ndevs * dev_stripes;
4549
4550 /*
4551 * this will have to be fixed for RAID1 and RAID10 over
4552 * more drives
4553 */
4554 data_stripes = num_stripes / ncopies;
4555
4556 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4557 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4558 btrfs_super_stripesize(info->super_copy));
4559 data_stripes = num_stripes - 1;
4560 }
4561 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4562 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4563 btrfs_super_stripesize(info->super_copy));
4564 data_stripes = num_stripes - 2;
4565 }
4566
4567 /*
4568 * Use the number of data stripes to figure out how big this chunk
4569 * is really going to be in terms of logical address space,
4570 * and compare that answer with the max chunk size
4571 */
4572 if (stripe_size * data_stripes > max_chunk_size) {
4573 u64 mask = (1ULL << 24) - 1;
4574
4575 stripe_size = div_u64(max_chunk_size, data_stripes);
4576
4577 /* bump the answer up to a 16MB boundary */
4578 stripe_size = (stripe_size + mask) & ~mask;
4579
4580 /* but don't go higher than the limits we found
4581 * while searching for free extents
4582 */
4583 if (stripe_size > devices_info[ndevs-1].max_avail)
4584 stripe_size = devices_info[ndevs-1].max_avail;
4585 }
4586
4587 stripe_size = div_u64(stripe_size, dev_stripes);
4588
4589 /* align to BTRFS_STRIPE_LEN */
4590 stripe_size = div_u64(stripe_size, raid_stripe_len);
4591 stripe_size *= raid_stripe_len;
4592
4593 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4594 if (!map) {
4595 ret = -ENOMEM;
4596 goto error;
4597 }
4598 map->num_stripes = num_stripes;
4599
4600 for (i = 0; i < ndevs; ++i) {
4601 for (j = 0; j < dev_stripes; ++j) {
4602 int s = i * dev_stripes + j;
4603 map->stripes[s].dev = devices_info[i].dev;
4604 map->stripes[s].physical = devices_info[i].dev_offset +
4605 j * stripe_size;
4606 }
4607 }
4608 map->sector_size = extent_root->sectorsize;
4609 map->stripe_len = raid_stripe_len;
4610 map->io_align = raid_stripe_len;
4611 map->io_width = raid_stripe_len;
4612 map->type = type;
4613 map->sub_stripes = sub_stripes;
4614
4615 num_bytes = stripe_size * data_stripes;
4616
4617 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4618
4619 em = alloc_extent_map();
4620 if (!em) {
4621 kfree(map);
4622 ret = -ENOMEM;
4623 goto error;
4624 }
4625 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4626 em->bdev = (struct block_device *)map;
4627 em->start = start;
4628 em->len = num_bytes;
4629 em->block_start = 0;
4630 em->block_len = em->len;
4631 em->orig_block_len = stripe_size;
4632
4633 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4634 write_lock(&em_tree->lock);
4635 ret = add_extent_mapping(em_tree, em, 0);
4636 if (!ret) {
4637 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4638 atomic_inc(&em->refs);
4639 }
4640 write_unlock(&em_tree->lock);
4641 if (ret) {
4642 free_extent_map(em);
4643 goto error;
4644 }
4645
4646 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4647 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4648 start, num_bytes);
4649 if (ret)
4650 goto error_del_extent;
4651
4652 for (i = 0; i < map->num_stripes; i++) {
4653 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4654 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4655 }
4656
4657 spin_lock(&extent_root->fs_info->free_chunk_lock);
4658 extent_root->fs_info->free_chunk_space -= (stripe_size *
4659 map->num_stripes);
4660 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4661
4662 free_extent_map(em);
4663 check_raid56_incompat_flag(extent_root->fs_info, type);
4664
4665 kfree(devices_info);
4666 return 0;
4667
4668 error_del_extent:
4669 write_lock(&em_tree->lock);
4670 remove_extent_mapping(em_tree, em);
4671 write_unlock(&em_tree->lock);
4672
4673 /* One for our allocation */
4674 free_extent_map(em);
4675 /* One for the tree reference */
4676 free_extent_map(em);
4677 /* One for the pending_chunks list reference */
4678 free_extent_map(em);
4679 error:
4680 kfree(devices_info);
4681 return ret;
4682 }
4683
4684 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4685 struct btrfs_root *extent_root,
4686 u64 chunk_offset, u64 chunk_size)
4687 {
4688 struct btrfs_key key;
4689 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4690 struct btrfs_device *device;
4691 struct btrfs_chunk *chunk;
4692 struct btrfs_stripe *stripe;
4693 struct extent_map_tree *em_tree;
4694 struct extent_map *em;
4695 struct map_lookup *map;
4696 size_t item_size;
4697 u64 dev_offset;
4698 u64 stripe_size;
4699 int i = 0;
4700 int ret;
4701
4702 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4703 read_lock(&em_tree->lock);
4704 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4705 read_unlock(&em_tree->lock);
4706
4707 if (!em) {
4708 btrfs_crit(extent_root->fs_info, "unable to find logical "
4709 "%Lu len %Lu", chunk_offset, chunk_size);
4710 return -EINVAL;
4711 }
4712
4713 if (em->start != chunk_offset || em->len != chunk_size) {
4714 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4715 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4716 chunk_size, em->start, em->len);
4717 free_extent_map(em);
4718 return -EINVAL;
4719 }
4720
4721 map = (struct map_lookup *)em->bdev;
4722 item_size = btrfs_chunk_item_size(map->num_stripes);
4723 stripe_size = em->orig_block_len;
4724
4725 chunk = kzalloc(item_size, GFP_NOFS);
4726 if (!chunk) {
4727 ret = -ENOMEM;
4728 goto out;
4729 }
4730
4731 for (i = 0; i < map->num_stripes; i++) {
4732 device = map->stripes[i].dev;
4733 dev_offset = map->stripes[i].physical;
4734
4735 ret = btrfs_update_device(trans, device);
4736 if (ret)
4737 goto out;
4738 ret = btrfs_alloc_dev_extent(trans, device,
4739 chunk_root->root_key.objectid,
4740 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4741 chunk_offset, dev_offset,
4742 stripe_size);
4743 if (ret)
4744 goto out;
4745 }
4746
4747 stripe = &chunk->stripe;
4748 for (i = 0; i < map->num_stripes; i++) {
4749 device = map->stripes[i].dev;
4750 dev_offset = map->stripes[i].physical;
4751
4752 btrfs_set_stack_stripe_devid(stripe, device->devid);
4753 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4754 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4755 stripe++;
4756 }
4757
4758 btrfs_set_stack_chunk_length(chunk, chunk_size);
4759 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4760 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4761 btrfs_set_stack_chunk_type(chunk, map->type);
4762 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4763 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4764 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4765 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4766 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4767
4768 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4769 key.type = BTRFS_CHUNK_ITEM_KEY;
4770 key.offset = chunk_offset;
4771
4772 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4773 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4774 /*
4775 * TODO: Cleanup of inserted chunk root in case of
4776 * failure.
4777 */
4778 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4779 item_size);
4780 }
4781
4782 out:
4783 kfree(chunk);
4784 free_extent_map(em);
4785 return ret;
4786 }
4787
4788 /*
4789 * Chunk allocation falls into two parts. The first part does works
4790 * that make the new allocated chunk useable, but not do any operation
4791 * that modifies the chunk tree. The second part does the works that
4792 * require modifying the chunk tree. This division is important for the
4793 * bootstrap process of adding storage to a seed btrfs.
4794 */
4795 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4796 struct btrfs_root *extent_root, u64 type)
4797 {
4798 u64 chunk_offset;
4799
4800 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4801 chunk_offset = find_next_chunk(extent_root->fs_info);
4802 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4803 }
4804
4805 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4806 struct btrfs_root *root,
4807 struct btrfs_device *device)
4808 {
4809 u64 chunk_offset;
4810 u64 sys_chunk_offset;
4811 u64 alloc_profile;
4812 struct btrfs_fs_info *fs_info = root->fs_info;
4813 struct btrfs_root *extent_root = fs_info->extent_root;
4814 int ret;
4815
4816 chunk_offset = find_next_chunk(fs_info);
4817 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4818 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4819 alloc_profile);
4820 if (ret)
4821 return ret;
4822
4823 sys_chunk_offset = find_next_chunk(root->fs_info);
4824 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4825 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4826 alloc_profile);
4827 return ret;
4828 }
4829
4830 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4831 {
4832 int max_errors;
4833
4834 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4835 BTRFS_BLOCK_GROUP_RAID10 |
4836 BTRFS_BLOCK_GROUP_RAID5 |
4837 BTRFS_BLOCK_GROUP_DUP)) {
4838 max_errors = 1;
4839 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4840 max_errors = 2;
4841 } else {
4842 max_errors = 0;
4843 }
4844
4845 return max_errors;
4846 }
4847
4848 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4849 {
4850 struct extent_map *em;
4851 struct map_lookup *map;
4852 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4853 int readonly = 0;
4854 int miss_ndevs = 0;
4855 int i;
4856
4857 read_lock(&map_tree->map_tree.lock);
4858 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4859 read_unlock(&map_tree->map_tree.lock);
4860 if (!em)
4861 return 1;
4862
4863 map = (struct map_lookup *)em->bdev;
4864 for (i = 0; i < map->num_stripes; i++) {
4865 if (map->stripes[i].dev->missing) {
4866 miss_ndevs++;
4867 continue;
4868 }
4869
4870 if (!map->stripes[i].dev->writeable) {
4871 readonly = 1;
4872 goto end;
4873 }
4874 }
4875
4876 /*
4877 * If the number of missing devices is larger than max errors,
4878 * we can not write the data into that chunk successfully, so
4879 * set it readonly.
4880 */
4881 if (miss_ndevs > btrfs_chunk_max_errors(map))
4882 readonly = 1;
4883 end:
4884 free_extent_map(em);
4885 return readonly;
4886 }
4887
4888 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4889 {
4890 extent_map_tree_init(&tree->map_tree);
4891 }
4892
4893 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4894 {
4895 struct extent_map *em;
4896
4897 while (1) {
4898 write_lock(&tree->map_tree.lock);
4899 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4900 if (em)
4901 remove_extent_mapping(&tree->map_tree, em);
4902 write_unlock(&tree->map_tree.lock);
4903 if (!em)
4904 break;
4905 /* once for us */
4906 free_extent_map(em);
4907 /* once for the tree */
4908 free_extent_map(em);
4909 }
4910 }
4911
4912 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4913 {
4914 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4915 struct extent_map *em;
4916 struct map_lookup *map;
4917 struct extent_map_tree *em_tree = &map_tree->map_tree;
4918 int ret;
4919
4920 read_lock(&em_tree->lock);
4921 em = lookup_extent_mapping(em_tree, logical, len);
4922 read_unlock(&em_tree->lock);
4923
4924 /*
4925 * We could return errors for these cases, but that could get ugly and
4926 * we'd probably do the same thing which is just not do anything else
4927 * and exit, so return 1 so the callers don't try to use other copies.
4928 */
4929 if (!em) {
4930 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4931 logical+len);
4932 return 1;
4933 }
4934
4935 if (em->start > logical || em->start + em->len < logical) {
4936 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4937 "%Lu-%Lu", logical, logical+len, em->start,
4938 em->start + em->len);
4939 free_extent_map(em);
4940 return 1;
4941 }
4942
4943 map = (struct map_lookup *)em->bdev;
4944 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4945 ret = map->num_stripes;
4946 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4947 ret = map->sub_stripes;
4948 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4949 ret = 2;
4950 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4951 ret = 3;
4952 else
4953 ret = 1;
4954 free_extent_map(em);
4955
4956 btrfs_dev_replace_lock(&fs_info->dev_replace);
4957 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4958 ret++;
4959 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4960
4961 return ret;
4962 }
4963
4964 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4965 struct btrfs_mapping_tree *map_tree,
4966 u64 logical)
4967 {
4968 struct extent_map *em;
4969 struct map_lookup *map;
4970 struct extent_map_tree *em_tree = &map_tree->map_tree;
4971 unsigned long len = root->sectorsize;
4972
4973 read_lock(&em_tree->lock);
4974 em = lookup_extent_mapping(em_tree, logical, len);
4975 read_unlock(&em_tree->lock);
4976 BUG_ON(!em);
4977
4978 BUG_ON(em->start > logical || em->start + em->len < logical);
4979 map = (struct map_lookup *)em->bdev;
4980 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4981 len = map->stripe_len * nr_data_stripes(map);
4982 free_extent_map(em);
4983 return len;
4984 }
4985
4986 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4987 u64 logical, u64 len, int mirror_num)
4988 {
4989 struct extent_map *em;
4990 struct map_lookup *map;
4991 struct extent_map_tree *em_tree = &map_tree->map_tree;
4992 int ret = 0;
4993
4994 read_lock(&em_tree->lock);
4995 em = lookup_extent_mapping(em_tree, logical, len);
4996 read_unlock(&em_tree->lock);
4997 BUG_ON(!em);
4998
4999 BUG_ON(em->start > logical || em->start + em->len < logical);
5000 map = (struct map_lookup *)em->bdev;
5001 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5002 ret = 1;
5003 free_extent_map(em);
5004 return ret;
5005 }
5006
5007 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5008 struct map_lookup *map, int first, int num,
5009 int optimal, int dev_replace_is_ongoing)
5010 {
5011 int i;
5012 int tolerance;
5013 struct btrfs_device *srcdev;
5014
5015 if (dev_replace_is_ongoing &&
5016 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5017 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5018 srcdev = fs_info->dev_replace.srcdev;
5019 else
5020 srcdev = NULL;
5021
5022 /*
5023 * try to avoid the drive that is the source drive for a
5024 * dev-replace procedure, only choose it if no other non-missing
5025 * mirror is available
5026 */
5027 for (tolerance = 0; tolerance < 2; tolerance++) {
5028 if (map->stripes[optimal].dev->bdev &&
5029 (tolerance || map->stripes[optimal].dev != srcdev))
5030 return optimal;
5031 for (i = first; i < first + num; i++) {
5032 if (map->stripes[i].dev->bdev &&
5033 (tolerance || map->stripes[i].dev != srcdev))
5034 return i;
5035 }
5036 }
5037
5038 /* we couldn't find one that doesn't fail. Just return something
5039 * and the io error handling code will clean up eventually
5040 */
5041 return optimal;
5042 }
5043
5044 static inline int parity_smaller(u64 a, u64 b)
5045 {
5046 return a > b;
5047 }
5048
5049 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5050 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5051 {
5052 struct btrfs_bio_stripe s;
5053 int i;
5054 u64 l;
5055 int again = 1;
5056
5057 while (again) {
5058 again = 0;
5059 for (i = 0; i < num_stripes - 1; i++) {
5060 if (parity_smaller(bbio->raid_map[i],
5061 bbio->raid_map[i+1])) {
5062 s = bbio->stripes[i];
5063 l = bbio->raid_map[i];
5064 bbio->stripes[i] = bbio->stripes[i+1];
5065 bbio->raid_map[i] = bbio->raid_map[i+1];
5066 bbio->stripes[i+1] = s;
5067 bbio->raid_map[i+1] = l;
5068
5069 again = 1;
5070 }
5071 }
5072 }
5073 }
5074
5075 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5076 {
5077 struct btrfs_bio *bbio = kzalloc(
5078 /* the size of the btrfs_bio */
5079 sizeof(struct btrfs_bio) +
5080 /* plus the variable array for the stripes */
5081 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5082 /* plus the variable array for the tgt dev */
5083 sizeof(int) * (real_stripes) +
5084 /*
5085 * plus the raid_map, which includes both the tgt dev
5086 * and the stripes
5087 */
5088 sizeof(u64) * (total_stripes),
5089 GFP_NOFS);
5090 if (!bbio)
5091 return NULL;
5092
5093 atomic_set(&bbio->error, 0);
5094 atomic_set(&bbio->refs, 1);
5095
5096 return bbio;
5097 }
5098
5099 void btrfs_get_bbio(struct btrfs_bio *bbio)
5100 {
5101 WARN_ON(!atomic_read(&bbio->refs));
5102 atomic_inc(&bbio->refs);
5103 }
5104
5105 void btrfs_put_bbio(struct btrfs_bio *bbio)
5106 {
5107 if (!bbio)
5108 return;
5109 if (atomic_dec_and_test(&bbio->refs))
5110 kfree(bbio);
5111 }
5112
5113 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5114 u64 logical, u64 *length,
5115 struct btrfs_bio **bbio_ret,
5116 int mirror_num, int need_raid_map)
5117 {
5118 struct extent_map *em;
5119 struct map_lookup *map;
5120 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5121 struct extent_map_tree *em_tree = &map_tree->map_tree;
5122 u64 offset;
5123 u64 stripe_offset;
5124 u64 stripe_end_offset;
5125 u64 stripe_nr;
5126 u64 stripe_nr_orig;
5127 u64 stripe_nr_end;
5128 u64 stripe_len;
5129 u32 stripe_index;
5130 int i;
5131 int ret = 0;
5132 int num_stripes;
5133 int max_errors = 0;
5134 int tgtdev_indexes = 0;
5135 struct btrfs_bio *bbio = NULL;
5136 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5137 int dev_replace_is_ongoing = 0;
5138 int num_alloc_stripes;
5139 int patch_the_first_stripe_for_dev_replace = 0;
5140 u64 physical_to_patch_in_first_stripe = 0;
5141 u64 raid56_full_stripe_start = (u64)-1;
5142
5143 read_lock(&em_tree->lock);
5144 em = lookup_extent_mapping(em_tree, logical, *length);
5145 read_unlock(&em_tree->lock);
5146
5147 if (!em) {
5148 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5149 logical, *length);
5150 return -EINVAL;
5151 }
5152
5153 if (em->start > logical || em->start + em->len < logical) {
5154 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5155 "found %Lu-%Lu", logical, em->start,
5156 em->start + em->len);
5157 free_extent_map(em);
5158 return -EINVAL;
5159 }
5160
5161 map = (struct map_lookup *)em->bdev;
5162 offset = logical - em->start;
5163
5164 stripe_len = map->stripe_len;
5165 stripe_nr = offset;
5166 /*
5167 * stripe_nr counts the total number of stripes we have to stride
5168 * to get to this block
5169 */
5170 stripe_nr = div64_u64(stripe_nr, stripe_len);
5171
5172 stripe_offset = stripe_nr * stripe_len;
5173 BUG_ON(offset < stripe_offset);
5174
5175 /* stripe_offset is the offset of this block in its stripe*/
5176 stripe_offset = offset - stripe_offset;
5177
5178 /* if we're here for raid56, we need to know the stripe aligned start */
5179 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5180 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5181 raid56_full_stripe_start = offset;
5182
5183 /* allow a write of a full stripe, but make sure we don't
5184 * allow straddling of stripes
5185 */
5186 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5187 full_stripe_len);
5188 raid56_full_stripe_start *= full_stripe_len;
5189 }
5190
5191 if (rw & REQ_DISCARD) {
5192 /* we don't discard raid56 yet */
5193 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5194 ret = -EOPNOTSUPP;
5195 goto out;
5196 }
5197 *length = min_t(u64, em->len - offset, *length);
5198 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5199 u64 max_len;
5200 /* For writes to RAID[56], allow a full stripeset across all disks.
5201 For other RAID types and for RAID[56] reads, just allow a single
5202 stripe (on a single disk). */
5203 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5204 (rw & REQ_WRITE)) {
5205 max_len = stripe_len * nr_data_stripes(map) -
5206 (offset - raid56_full_stripe_start);
5207 } else {
5208 /* we limit the length of each bio to what fits in a stripe */
5209 max_len = stripe_len - stripe_offset;
5210 }
5211 *length = min_t(u64, em->len - offset, max_len);
5212 } else {
5213 *length = em->len - offset;
5214 }
5215
5216 /* This is for when we're called from btrfs_merge_bio_hook() and all
5217 it cares about is the length */
5218 if (!bbio_ret)
5219 goto out;
5220
5221 btrfs_dev_replace_lock(dev_replace);
5222 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5223 if (!dev_replace_is_ongoing)
5224 btrfs_dev_replace_unlock(dev_replace);
5225
5226 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5227 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5228 dev_replace->tgtdev != NULL) {
5229 /*
5230 * in dev-replace case, for repair case (that's the only
5231 * case where the mirror is selected explicitly when
5232 * calling btrfs_map_block), blocks left of the left cursor
5233 * can also be read from the target drive.
5234 * For REQ_GET_READ_MIRRORS, the target drive is added as
5235 * the last one to the array of stripes. For READ, it also
5236 * needs to be supported using the same mirror number.
5237 * If the requested block is not left of the left cursor,
5238 * EIO is returned. This can happen because btrfs_num_copies()
5239 * returns one more in the dev-replace case.
5240 */
5241 u64 tmp_length = *length;
5242 struct btrfs_bio *tmp_bbio = NULL;
5243 int tmp_num_stripes;
5244 u64 srcdev_devid = dev_replace->srcdev->devid;
5245 int index_srcdev = 0;
5246 int found = 0;
5247 u64 physical_of_found = 0;
5248
5249 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5250 logical, &tmp_length, &tmp_bbio, 0, 0);
5251 if (ret) {
5252 WARN_ON(tmp_bbio != NULL);
5253 goto out;
5254 }
5255
5256 tmp_num_stripes = tmp_bbio->num_stripes;
5257 if (mirror_num > tmp_num_stripes) {
5258 /*
5259 * REQ_GET_READ_MIRRORS does not contain this
5260 * mirror, that means that the requested area
5261 * is not left of the left cursor
5262 */
5263 ret = -EIO;
5264 btrfs_put_bbio(tmp_bbio);
5265 goto out;
5266 }
5267
5268 /*
5269 * process the rest of the function using the mirror_num
5270 * of the source drive. Therefore look it up first.
5271 * At the end, patch the device pointer to the one of the
5272 * target drive.
5273 */
5274 for (i = 0; i < tmp_num_stripes; i++) {
5275 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5276 /*
5277 * In case of DUP, in order to keep it
5278 * simple, only add the mirror with the
5279 * lowest physical address
5280 */
5281 if (found &&
5282 physical_of_found <=
5283 tmp_bbio->stripes[i].physical)
5284 continue;
5285 index_srcdev = i;
5286 found = 1;
5287 physical_of_found =
5288 tmp_bbio->stripes[i].physical;
5289 }
5290 }
5291
5292 if (found) {
5293 mirror_num = index_srcdev + 1;
5294 patch_the_first_stripe_for_dev_replace = 1;
5295 physical_to_patch_in_first_stripe = physical_of_found;
5296 } else {
5297 WARN_ON(1);
5298 ret = -EIO;
5299 btrfs_put_bbio(tmp_bbio);
5300 goto out;
5301 }
5302
5303 btrfs_put_bbio(tmp_bbio);
5304 } else if (mirror_num > map->num_stripes) {
5305 mirror_num = 0;
5306 }
5307
5308 num_stripes = 1;
5309 stripe_index = 0;
5310 stripe_nr_orig = stripe_nr;
5311 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5312 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5313 stripe_end_offset = stripe_nr_end * map->stripe_len -
5314 (offset + *length);
5315
5316 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5317 if (rw & REQ_DISCARD)
5318 num_stripes = min_t(u64, map->num_stripes,
5319 stripe_nr_end - stripe_nr_orig);
5320 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5321 &stripe_index);
5322 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5323 mirror_num = 1;
5324 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5325 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5326 num_stripes = map->num_stripes;
5327 else if (mirror_num)
5328 stripe_index = mirror_num - 1;
5329 else {
5330 stripe_index = find_live_mirror(fs_info, map, 0,
5331 map->num_stripes,
5332 current->pid % map->num_stripes,
5333 dev_replace_is_ongoing);
5334 mirror_num = stripe_index + 1;
5335 }
5336
5337 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5338 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5339 num_stripes = map->num_stripes;
5340 } else if (mirror_num) {
5341 stripe_index = mirror_num - 1;
5342 } else {
5343 mirror_num = 1;
5344 }
5345
5346 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5347 u32 factor = map->num_stripes / map->sub_stripes;
5348
5349 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5350 stripe_index *= map->sub_stripes;
5351
5352 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5353 num_stripes = map->sub_stripes;
5354 else if (rw & REQ_DISCARD)
5355 num_stripes = min_t(u64, map->sub_stripes *
5356 (stripe_nr_end - stripe_nr_orig),
5357 map->num_stripes);
5358 else if (mirror_num)
5359 stripe_index += mirror_num - 1;
5360 else {
5361 int old_stripe_index = stripe_index;
5362 stripe_index = find_live_mirror(fs_info, map,
5363 stripe_index,
5364 map->sub_stripes, stripe_index +
5365 current->pid % map->sub_stripes,
5366 dev_replace_is_ongoing);
5367 mirror_num = stripe_index - old_stripe_index + 1;
5368 }
5369
5370 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5371 if (need_raid_map &&
5372 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5373 mirror_num > 1)) {
5374 /* push stripe_nr back to the start of the full stripe */
5375 stripe_nr = div_u64(raid56_full_stripe_start,
5376 stripe_len * nr_data_stripes(map));
5377
5378 /* RAID[56] write or recovery. Return all stripes */
5379 num_stripes = map->num_stripes;
5380 max_errors = nr_parity_stripes(map);
5381
5382 *length = map->stripe_len;
5383 stripe_index = 0;
5384 stripe_offset = 0;
5385 } else {
5386 /*
5387 * Mirror #0 or #1 means the original data block.
5388 * Mirror #2 is RAID5 parity block.
5389 * Mirror #3 is RAID6 Q block.
5390 */
5391 stripe_nr = div_u64_rem(stripe_nr,
5392 nr_data_stripes(map), &stripe_index);
5393 if (mirror_num > 1)
5394 stripe_index = nr_data_stripes(map) +
5395 mirror_num - 2;
5396
5397 /* We distribute the parity blocks across stripes */
5398 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5399 &stripe_index);
5400 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5401 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5402 mirror_num = 1;
5403 }
5404 } else {
5405 /*
5406 * after this, stripe_nr is the number of stripes on this
5407 * device we have to walk to find the data, and stripe_index is
5408 * the number of our device in the stripe array
5409 */
5410 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5411 &stripe_index);
5412 mirror_num = stripe_index + 1;
5413 }
5414 BUG_ON(stripe_index >= map->num_stripes);
5415
5416 num_alloc_stripes = num_stripes;
5417 if (dev_replace_is_ongoing) {
5418 if (rw & (REQ_WRITE | REQ_DISCARD))
5419 num_alloc_stripes <<= 1;
5420 if (rw & REQ_GET_READ_MIRRORS)
5421 num_alloc_stripes++;
5422 tgtdev_indexes = num_stripes;
5423 }
5424
5425 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5426 if (!bbio) {
5427 ret = -ENOMEM;
5428 goto out;
5429 }
5430 if (dev_replace_is_ongoing)
5431 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5432
5433 /* build raid_map */
5434 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5435 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5436 mirror_num > 1)) {
5437 u64 tmp;
5438 unsigned rot;
5439
5440 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5441 sizeof(struct btrfs_bio_stripe) *
5442 num_alloc_stripes +
5443 sizeof(int) * tgtdev_indexes);
5444
5445 /* Work out the disk rotation on this stripe-set */
5446 div_u64_rem(stripe_nr, num_stripes, &rot);
5447
5448 /* Fill in the logical address of each stripe */
5449 tmp = stripe_nr * nr_data_stripes(map);
5450 for (i = 0; i < nr_data_stripes(map); i++)
5451 bbio->raid_map[(i+rot) % num_stripes] =
5452 em->start + (tmp + i) * map->stripe_len;
5453
5454 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5455 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5456 bbio->raid_map[(i+rot+1) % num_stripes] =
5457 RAID6_Q_STRIPE;
5458 }
5459
5460 if (rw & REQ_DISCARD) {
5461 u32 factor = 0;
5462 u32 sub_stripes = 0;
5463 u64 stripes_per_dev = 0;
5464 u32 remaining_stripes = 0;
5465 u32 last_stripe = 0;
5466
5467 if (map->type &
5468 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5469 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5470 sub_stripes = 1;
5471 else
5472 sub_stripes = map->sub_stripes;
5473
5474 factor = map->num_stripes / sub_stripes;
5475 stripes_per_dev = div_u64_rem(stripe_nr_end -
5476 stripe_nr_orig,
5477 factor,
5478 &remaining_stripes);
5479 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5480 last_stripe *= sub_stripes;
5481 }
5482
5483 for (i = 0; i < num_stripes; i++) {
5484 bbio->stripes[i].physical =
5485 map->stripes[stripe_index].physical +
5486 stripe_offset + stripe_nr * map->stripe_len;
5487 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5488
5489 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5490 BTRFS_BLOCK_GROUP_RAID10)) {
5491 bbio->stripes[i].length = stripes_per_dev *
5492 map->stripe_len;
5493
5494 if (i / sub_stripes < remaining_stripes)
5495 bbio->stripes[i].length +=
5496 map->stripe_len;
5497
5498 /*
5499 * Special for the first stripe and
5500 * the last stripe:
5501 *
5502 * |-------|...|-------|
5503 * |----------|
5504 * off end_off
5505 */
5506 if (i < sub_stripes)
5507 bbio->stripes[i].length -=
5508 stripe_offset;
5509
5510 if (stripe_index >= last_stripe &&
5511 stripe_index <= (last_stripe +
5512 sub_stripes - 1))
5513 bbio->stripes[i].length -=
5514 stripe_end_offset;
5515
5516 if (i == sub_stripes - 1)
5517 stripe_offset = 0;
5518 } else
5519 bbio->stripes[i].length = *length;
5520
5521 stripe_index++;
5522 if (stripe_index == map->num_stripes) {
5523 /* This could only happen for RAID0/10 */
5524 stripe_index = 0;
5525 stripe_nr++;
5526 }
5527 }
5528 } else {
5529 for (i = 0; i < num_stripes; i++) {
5530 bbio->stripes[i].physical =
5531 map->stripes[stripe_index].physical +
5532 stripe_offset +
5533 stripe_nr * map->stripe_len;
5534 bbio->stripes[i].dev =
5535 map->stripes[stripe_index].dev;
5536 stripe_index++;
5537 }
5538 }
5539
5540 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5541 max_errors = btrfs_chunk_max_errors(map);
5542
5543 if (bbio->raid_map)
5544 sort_parity_stripes(bbio, num_stripes);
5545
5546 tgtdev_indexes = 0;
5547 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5548 dev_replace->tgtdev != NULL) {
5549 int index_where_to_add;
5550 u64 srcdev_devid = dev_replace->srcdev->devid;
5551
5552 /*
5553 * duplicate the write operations while the dev replace
5554 * procedure is running. Since the copying of the old disk
5555 * to the new disk takes place at run time while the
5556 * filesystem is mounted writable, the regular write
5557 * operations to the old disk have to be duplicated to go
5558 * to the new disk as well.
5559 * Note that device->missing is handled by the caller, and
5560 * that the write to the old disk is already set up in the
5561 * stripes array.
5562 */
5563 index_where_to_add = num_stripes;
5564 for (i = 0; i < num_stripes; i++) {
5565 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5566 /* write to new disk, too */
5567 struct btrfs_bio_stripe *new =
5568 bbio->stripes + index_where_to_add;
5569 struct btrfs_bio_stripe *old =
5570 bbio->stripes + i;
5571
5572 new->physical = old->physical;
5573 new->length = old->length;
5574 new->dev = dev_replace->tgtdev;
5575 bbio->tgtdev_map[i] = index_where_to_add;
5576 index_where_to_add++;
5577 max_errors++;
5578 tgtdev_indexes++;
5579 }
5580 }
5581 num_stripes = index_where_to_add;
5582 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5583 dev_replace->tgtdev != NULL) {
5584 u64 srcdev_devid = dev_replace->srcdev->devid;
5585 int index_srcdev = 0;
5586 int found = 0;
5587 u64 physical_of_found = 0;
5588
5589 /*
5590 * During the dev-replace procedure, the target drive can
5591 * also be used to read data in case it is needed to repair
5592 * a corrupt block elsewhere. This is possible if the
5593 * requested area is left of the left cursor. In this area,
5594 * the target drive is a full copy of the source drive.
5595 */
5596 for (i = 0; i < num_stripes; i++) {
5597 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5598 /*
5599 * In case of DUP, in order to keep it
5600 * simple, only add the mirror with the
5601 * lowest physical address
5602 */
5603 if (found &&
5604 physical_of_found <=
5605 bbio->stripes[i].physical)
5606 continue;
5607 index_srcdev = i;
5608 found = 1;
5609 physical_of_found = bbio->stripes[i].physical;
5610 }
5611 }
5612 if (found) {
5613 if (physical_of_found + map->stripe_len <=
5614 dev_replace->cursor_left) {
5615 struct btrfs_bio_stripe *tgtdev_stripe =
5616 bbio->stripes + num_stripes;
5617
5618 tgtdev_stripe->physical = physical_of_found;
5619 tgtdev_stripe->length =
5620 bbio->stripes[index_srcdev].length;
5621 tgtdev_stripe->dev = dev_replace->tgtdev;
5622 bbio->tgtdev_map[index_srcdev] = num_stripes;
5623
5624 tgtdev_indexes++;
5625 num_stripes++;
5626 }
5627 }
5628 }
5629
5630 *bbio_ret = bbio;
5631 bbio->map_type = map->type;
5632 bbio->num_stripes = num_stripes;
5633 bbio->max_errors = max_errors;
5634 bbio->mirror_num = mirror_num;
5635 bbio->num_tgtdevs = tgtdev_indexes;
5636
5637 /*
5638 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5639 * mirror_num == num_stripes + 1 && dev_replace target drive is
5640 * available as a mirror
5641 */
5642 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5643 WARN_ON(num_stripes > 1);
5644 bbio->stripes[0].dev = dev_replace->tgtdev;
5645 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5646 bbio->mirror_num = map->num_stripes + 1;
5647 }
5648 out:
5649 if (dev_replace_is_ongoing)
5650 btrfs_dev_replace_unlock(dev_replace);
5651 free_extent_map(em);
5652 return ret;
5653 }
5654
5655 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5656 u64 logical, u64 *length,
5657 struct btrfs_bio **bbio_ret, int mirror_num)
5658 {
5659 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5660 mirror_num, 0);
5661 }
5662
5663 /* For Scrub/replace */
5664 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5665 u64 logical, u64 *length,
5666 struct btrfs_bio **bbio_ret, int mirror_num,
5667 int need_raid_map)
5668 {
5669 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5670 mirror_num, need_raid_map);
5671 }
5672
5673 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5674 u64 chunk_start, u64 physical, u64 devid,
5675 u64 **logical, int *naddrs, int *stripe_len)
5676 {
5677 struct extent_map_tree *em_tree = &map_tree->map_tree;
5678 struct extent_map *em;
5679 struct map_lookup *map;
5680 u64 *buf;
5681 u64 bytenr;
5682 u64 length;
5683 u64 stripe_nr;
5684 u64 rmap_len;
5685 int i, j, nr = 0;
5686
5687 read_lock(&em_tree->lock);
5688 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5689 read_unlock(&em_tree->lock);
5690
5691 if (!em) {
5692 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5693 chunk_start);
5694 return -EIO;
5695 }
5696
5697 if (em->start != chunk_start) {
5698 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5699 em->start, chunk_start);
5700 free_extent_map(em);
5701 return -EIO;
5702 }
5703 map = (struct map_lookup *)em->bdev;
5704
5705 length = em->len;
5706 rmap_len = map->stripe_len;
5707
5708 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5709 length = div_u64(length, map->num_stripes / map->sub_stripes);
5710 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5711 length = div_u64(length, map->num_stripes);
5712 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5713 length = div_u64(length, nr_data_stripes(map));
5714 rmap_len = map->stripe_len * nr_data_stripes(map);
5715 }
5716
5717 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5718 BUG_ON(!buf); /* -ENOMEM */
5719
5720 for (i = 0; i < map->num_stripes; i++) {
5721 if (devid && map->stripes[i].dev->devid != devid)
5722 continue;
5723 if (map->stripes[i].physical > physical ||
5724 map->stripes[i].physical + length <= physical)
5725 continue;
5726
5727 stripe_nr = physical - map->stripes[i].physical;
5728 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5729
5730 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5731 stripe_nr = stripe_nr * map->num_stripes + i;
5732 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5733 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5734 stripe_nr = stripe_nr * map->num_stripes + i;
5735 } /* else if RAID[56], multiply by nr_data_stripes().
5736 * Alternatively, just use rmap_len below instead of
5737 * map->stripe_len */
5738
5739 bytenr = chunk_start + stripe_nr * rmap_len;
5740 WARN_ON(nr >= map->num_stripes);
5741 for (j = 0; j < nr; j++) {
5742 if (buf[j] == bytenr)
5743 break;
5744 }
5745 if (j == nr) {
5746 WARN_ON(nr >= map->num_stripes);
5747 buf[nr++] = bytenr;
5748 }
5749 }
5750
5751 *logical = buf;
5752 *naddrs = nr;
5753 *stripe_len = rmap_len;
5754
5755 free_extent_map(em);
5756 return 0;
5757 }
5758
5759 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5760 {
5761 if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5762 bio_endio_nodec(bio, err);
5763 else
5764 bio_endio(bio, err);
5765 btrfs_put_bbio(bbio);
5766 }
5767
5768 static void btrfs_end_bio(struct bio *bio, int err)
5769 {
5770 struct btrfs_bio *bbio = bio->bi_private;
5771 int is_orig_bio = 0;
5772
5773 if (err) {
5774 atomic_inc(&bbio->error);
5775 if (err == -EIO || err == -EREMOTEIO) {
5776 unsigned int stripe_index =
5777 btrfs_io_bio(bio)->stripe_index;
5778 struct btrfs_device *dev;
5779
5780 BUG_ON(stripe_index >= bbio->num_stripes);
5781 dev = bbio->stripes[stripe_index].dev;
5782 if (dev->bdev) {
5783 if (bio->bi_rw & WRITE)
5784 btrfs_dev_stat_inc(dev,
5785 BTRFS_DEV_STAT_WRITE_ERRS);
5786 else
5787 btrfs_dev_stat_inc(dev,
5788 BTRFS_DEV_STAT_READ_ERRS);
5789 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5790 btrfs_dev_stat_inc(dev,
5791 BTRFS_DEV_STAT_FLUSH_ERRS);
5792 btrfs_dev_stat_print_on_error(dev);
5793 }
5794 }
5795 }
5796
5797 if (bio == bbio->orig_bio)
5798 is_orig_bio = 1;
5799
5800 btrfs_bio_counter_dec(bbio->fs_info);
5801
5802 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5803 if (!is_orig_bio) {
5804 bio_put(bio);
5805 bio = bbio->orig_bio;
5806 }
5807
5808 bio->bi_private = bbio->private;
5809 bio->bi_end_io = bbio->end_io;
5810 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5811 /* only send an error to the higher layers if it is
5812 * beyond the tolerance of the btrfs bio
5813 */
5814 if (atomic_read(&bbio->error) > bbio->max_errors) {
5815 err = -EIO;
5816 } else {
5817 /*
5818 * this bio is actually up to date, we didn't
5819 * go over the max number of errors
5820 */
5821 set_bit(BIO_UPTODATE, &bio->bi_flags);
5822 err = 0;
5823 }
5824
5825 btrfs_end_bbio(bbio, bio, err);
5826 } else if (!is_orig_bio) {
5827 bio_put(bio);
5828 }
5829 }
5830
5831 /*
5832 * see run_scheduled_bios for a description of why bios are collected for
5833 * async submit.
5834 *
5835 * This will add one bio to the pending list for a device and make sure
5836 * the work struct is scheduled.
5837 */
5838 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5839 struct btrfs_device *device,
5840 int rw, struct bio *bio)
5841 {
5842 int should_queue = 1;
5843 struct btrfs_pending_bios *pending_bios;
5844
5845 if (device->missing || !device->bdev) {
5846 bio_endio(bio, -EIO);
5847 return;
5848 }
5849
5850 /* don't bother with additional async steps for reads, right now */
5851 if (!(rw & REQ_WRITE)) {
5852 bio_get(bio);
5853 btrfsic_submit_bio(rw, bio);
5854 bio_put(bio);
5855 return;
5856 }
5857
5858 /*
5859 * nr_async_bios allows us to reliably return congestion to the
5860 * higher layers. Otherwise, the async bio makes it appear we have
5861 * made progress against dirty pages when we've really just put it
5862 * on a queue for later
5863 */
5864 atomic_inc(&root->fs_info->nr_async_bios);
5865 WARN_ON(bio->bi_next);
5866 bio->bi_next = NULL;
5867 bio->bi_rw |= rw;
5868
5869 spin_lock(&device->io_lock);
5870 if (bio->bi_rw & REQ_SYNC)
5871 pending_bios = &device->pending_sync_bios;
5872 else
5873 pending_bios = &device->pending_bios;
5874
5875 if (pending_bios->tail)
5876 pending_bios->tail->bi_next = bio;
5877
5878 pending_bios->tail = bio;
5879 if (!pending_bios->head)
5880 pending_bios->head = bio;
5881 if (device->running_pending)
5882 should_queue = 0;
5883
5884 spin_unlock(&device->io_lock);
5885
5886 if (should_queue)
5887 btrfs_queue_work(root->fs_info->submit_workers,
5888 &device->work);
5889 }
5890
5891 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5892 sector_t sector)
5893 {
5894 struct bio_vec *prev;
5895 struct request_queue *q = bdev_get_queue(bdev);
5896 unsigned int max_sectors = queue_max_sectors(q);
5897 struct bvec_merge_data bvm = {
5898 .bi_bdev = bdev,
5899 .bi_sector = sector,
5900 .bi_rw = bio->bi_rw,
5901 };
5902
5903 if (WARN_ON(bio->bi_vcnt == 0))
5904 return 1;
5905
5906 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5907 if (bio_sectors(bio) > max_sectors)
5908 return 0;
5909
5910 if (!q->merge_bvec_fn)
5911 return 1;
5912
5913 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5914 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5915 return 0;
5916 return 1;
5917 }
5918
5919 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5920 struct bio *bio, u64 physical, int dev_nr,
5921 int rw, int async)
5922 {
5923 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5924
5925 bio->bi_private = bbio;
5926 btrfs_io_bio(bio)->stripe_index = dev_nr;
5927 bio->bi_end_io = btrfs_end_bio;
5928 bio->bi_iter.bi_sector = physical >> 9;
5929 #ifdef DEBUG
5930 {
5931 struct rcu_string *name;
5932
5933 rcu_read_lock();
5934 name = rcu_dereference(dev->name);
5935 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5936 "(%s id %llu), size=%u\n", rw,
5937 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5938 name->str, dev->devid, bio->bi_iter.bi_size);
5939 rcu_read_unlock();
5940 }
5941 #endif
5942 bio->bi_bdev = dev->bdev;
5943
5944 btrfs_bio_counter_inc_noblocked(root->fs_info);
5945
5946 if (async)
5947 btrfs_schedule_bio(root, dev, rw, bio);
5948 else
5949 btrfsic_submit_bio(rw, bio);
5950 }
5951
5952 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5953 struct bio *first_bio, struct btrfs_device *dev,
5954 int dev_nr, int rw, int async)
5955 {
5956 struct bio_vec *bvec = first_bio->bi_io_vec;
5957 struct bio *bio;
5958 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5959 u64 physical = bbio->stripes[dev_nr].physical;
5960
5961 again:
5962 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5963 if (!bio)
5964 return -ENOMEM;
5965
5966 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5967 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5968 bvec->bv_offset) < bvec->bv_len) {
5969 u64 len = bio->bi_iter.bi_size;
5970
5971 atomic_inc(&bbio->stripes_pending);
5972 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5973 rw, async);
5974 physical += len;
5975 goto again;
5976 }
5977 bvec++;
5978 }
5979
5980 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5981 return 0;
5982 }
5983
5984 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5985 {
5986 atomic_inc(&bbio->error);
5987 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5988 /* Shoud be the original bio. */
5989 WARN_ON(bio != bbio->orig_bio);
5990
5991 bio->bi_private = bbio->private;
5992 bio->bi_end_io = bbio->end_io;
5993 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5994 bio->bi_iter.bi_sector = logical >> 9;
5995
5996 btrfs_end_bbio(bbio, bio, -EIO);
5997 }
5998 }
5999
6000 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6001 int mirror_num, int async_submit)
6002 {
6003 struct btrfs_device *dev;
6004 struct bio *first_bio = bio;
6005 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6006 u64 length = 0;
6007 u64 map_length;
6008 int ret;
6009 int dev_nr;
6010 int total_devs;
6011 struct btrfs_bio *bbio = NULL;
6012
6013 length = bio->bi_iter.bi_size;
6014 map_length = length;
6015
6016 btrfs_bio_counter_inc_blocked(root->fs_info);
6017 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6018 mirror_num, 1);
6019 if (ret) {
6020 btrfs_bio_counter_dec(root->fs_info);
6021 return ret;
6022 }
6023
6024 total_devs = bbio->num_stripes;
6025 bbio->orig_bio = first_bio;
6026 bbio->private = first_bio->bi_private;
6027 bbio->end_io = first_bio->bi_end_io;
6028 bbio->fs_info = root->fs_info;
6029 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6030
6031 if (bbio->raid_map) {
6032 /* In this case, map_length has been set to the length of
6033 a single stripe; not the whole write */
6034 if (rw & WRITE) {
6035 ret = raid56_parity_write(root, bio, bbio, map_length);
6036 } else {
6037 ret = raid56_parity_recover(root, bio, bbio, map_length,
6038 mirror_num, 1);
6039 }
6040
6041 btrfs_bio_counter_dec(root->fs_info);
6042 return ret;
6043 }
6044
6045 if (map_length < length) {
6046 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6047 logical, length, map_length);
6048 BUG();
6049 }
6050
6051 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6052 dev = bbio->stripes[dev_nr].dev;
6053 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6054 bbio_error(bbio, first_bio, logical);
6055 continue;
6056 }
6057
6058 /*
6059 * Check and see if we're ok with this bio based on it's size
6060 * and offset with the given device.
6061 */
6062 if (!bio_size_ok(dev->bdev, first_bio,
6063 bbio->stripes[dev_nr].physical >> 9)) {
6064 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
6065 dev_nr, rw, async_submit);
6066 BUG_ON(ret);
6067 continue;
6068 }
6069
6070 if (dev_nr < total_devs - 1) {
6071 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6072 BUG_ON(!bio); /* -ENOMEM */
6073 } else {
6074 bio = first_bio;
6075 bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
6076 }
6077
6078 submit_stripe_bio(root, bbio, bio,
6079 bbio->stripes[dev_nr].physical, dev_nr, rw,
6080 async_submit);
6081 }
6082 btrfs_bio_counter_dec(root->fs_info);
6083 return 0;
6084 }
6085
6086 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6087 u8 *uuid, u8 *fsid)
6088 {
6089 struct btrfs_device *device;
6090 struct btrfs_fs_devices *cur_devices;
6091
6092 cur_devices = fs_info->fs_devices;
6093 while (cur_devices) {
6094 if (!fsid ||
6095 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6096 device = __find_device(&cur_devices->devices,
6097 devid, uuid);
6098 if (device)
6099 return device;
6100 }
6101 cur_devices = cur_devices->seed;
6102 }
6103 return NULL;
6104 }
6105
6106 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6107 struct btrfs_fs_devices *fs_devices,
6108 u64 devid, u8 *dev_uuid)
6109 {
6110 struct btrfs_device *device;
6111
6112 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6113 if (IS_ERR(device))
6114 return NULL;
6115
6116 list_add(&device->dev_list, &fs_devices->devices);
6117 device->fs_devices = fs_devices;
6118 fs_devices->num_devices++;
6119
6120 device->missing = 1;
6121 fs_devices->missing_devices++;
6122
6123 return device;
6124 }
6125
6126 /**
6127 * btrfs_alloc_device - allocate struct btrfs_device
6128 * @fs_info: used only for generating a new devid, can be NULL if
6129 * devid is provided (i.e. @devid != NULL).
6130 * @devid: a pointer to devid for this device. If NULL a new devid
6131 * is generated.
6132 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6133 * is generated.
6134 *
6135 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6136 * on error. Returned struct is not linked onto any lists and can be
6137 * destroyed with kfree() right away.
6138 */
6139 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6140 const u64 *devid,
6141 const u8 *uuid)
6142 {
6143 struct btrfs_device *dev;
6144 u64 tmp;
6145
6146 if (WARN_ON(!devid && !fs_info))
6147 return ERR_PTR(-EINVAL);
6148
6149 dev = __alloc_device();
6150 if (IS_ERR(dev))
6151 return dev;
6152
6153 if (devid)
6154 tmp = *devid;
6155 else {
6156 int ret;
6157
6158 ret = find_next_devid(fs_info, &tmp);
6159 if (ret) {
6160 kfree(dev);
6161 return ERR_PTR(ret);
6162 }
6163 }
6164 dev->devid = tmp;
6165
6166 if (uuid)
6167 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6168 else
6169 generate_random_uuid(dev->uuid);
6170
6171 btrfs_init_work(&dev->work, btrfs_submit_helper,
6172 pending_bios_fn, NULL, NULL);
6173
6174 return dev;
6175 }
6176
6177 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6178 struct extent_buffer *leaf,
6179 struct btrfs_chunk *chunk)
6180 {
6181 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6182 struct map_lookup *map;
6183 struct extent_map *em;
6184 u64 logical;
6185 u64 length;
6186 u64 devid;
6187 u8 uuid[BTRFS_UUID_SIZE];
6188 int num_stripes;
6189 int ret;
6190 int i;
6191
6192 logical = key->offset;
6193 length = btrfs_chunk_length(leaf, chunk);
6194
6195 read_lock(&map_tree->map_tree.lock);
6196 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6197 read_unlock(&map_tree->map_tree.lock);
6198
6199 /* already mapped? */
6200 if (em && em->start <= logical && em->start + em->len > logical) {
6201 free_extent_map(em);
6202 return 0;
6203 } else if (em) {
6204 free_extent_map(em);
6205 }
6206
6207 em = alloc_extent_map();
6208 if (!em)
6209 return -ENOMEM;
6210 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6211 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6212 if (!map) {
6213 free_extent_map(em);
6214 return -ENOMEM;
6215 }
6216
6217 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6218 em->bdev = (struct block_device *)map;
6219 em->start = logical;
6220 em->len = length;
6221 em->orig_start = 0;
6222 em->block_start = 0;
6223 em->block_len = em->len;
6224
6225 map->num_stripes = num_stripes;
6226 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6227 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6228 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6229 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6230 map->type = btrfs_chunk_type(leaf, chunk);
6231 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6232 for (i = 0; i < num_stripes; i++) {
6233 map->stripes[i].physical =
6234 btrfs_stripe_offset_nr(leaf, chunk, i);
6235 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6236 read_extent_buffer(leaf, uuid, (unsigned long)
6237 btrfs_stripe_dev_uuid_nr(chunk, i),
6238 BTRFS_UUID_SIZE);
6239 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6240 uuid, NULL);
6241 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6242 free_extent_map(em);
6243 return -EIO;
6244 }
6245 if (!map->stripes[i].dev) {
6246 map->stripes[i].dev =
6247 add_missing_dev(root, root->fs_info->fs_devices,
6248 devid, uuid);
6249 if (!map->stripes[i].dev) {
6250 free_extent_map(em);
6251 return -EIO;
6252 }
6253 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6254 devid, uuid);
6255 }
6256 map->stripes[i].dev->in_fs_metadata = 1;
6257 }
6258
6259 write_lock(&map_tree->map_tree.lock);
6260 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6261 write_unlock(&map_tree->map_tree.lock);
6262 BUG_ON(ret); /* Tree corruption */
6263 free_extent_map(em);
6264
6265 return 0;
6266 }
6267
6268 static void fill_device_from_item(struct extent_buffer *leaf,
6269 struct btrfs_dev_item *dev_item,
6270 struct btrfs_device *device)
6271 {
6272 unsigned long ptr;
6273
6274 device->devid = btrfs_device_id(leaf, dev_item);
6275 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6276 device->total_bytes = device->disk_total_bytes;
6277 device->commit_total_bytes = device->disk_total_bytes;
6278 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6279 device->commit_bytes_used = device->bytes_used;
6280 device->type = btrfs_device_type(leaf, dev_item);
6281 device->io_align = btrfs_device_io_align(leaf, dev_item);
6282 device->io_width = btrfs_device_io_width(leaf, dev_item);
6283 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6284 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6285 device->is_tgtdev_for_dev_replace = 0;
6286
6287 ptr = btrfs_device_uuid(dev_item);
6288 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6289 }
6290
6291 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6292 u8 *fsid)
6293 {
6294 struct btrfs_fs_devices *fs_devices;
6295 int ret;
6296
6297 BUG_ON(!mutex_is_locked(&uuid_mutex));
6298
6299 fs_devices = root->fs_info->fs_devices->seed;
6300 while (fs_devices) {
6301 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6302 return fs_devices;
6303
6304 fs_devices = fs_devices->seed;
6305 }
6306
6307 fs_devices = find_fsid(fsid);
6308 if (!fs_devices) {
6309 if (!btrfs_test_opt(root, DEGRADED))
6310 return ERR_PTR(-ENOENT);
6311
6312 fs_devices = alloc_fs_devices(fsid);
6313 if (IS_ERR(fs_devices))
6314 return fs_devices;
6315
6316 fs_devices->seeding = 1;
6317 fs_devices->opened = 1;
6318 return fs_devices;
6319 }
6320
6321 fs_devices = clone_fs_devices(fs_devices);
6322 if (IS_ERR(fs_devices))
6323 return fs_devices;
6324
6325 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6326 root->fs_info->bdev_holder);
6327 if (ret) {
6328 free_fs_devices(fs_devices);
6329 fs_devices = ERR_PTR(ret);
6330 goto out;
6331 }
6332
6333 if (!fs_devices->seeding) {
6334 __btrfs_close_devices(fs_devices);
6335 free_fs_devices(fs_devices);
6336 fs_devices = ERR_PTR(-EINVAL);
6337 goto out;
6338 }
6339
6340 fs_devices->seed = root->fs_info->fs_devices->seed;
6341 root->fs_info->fs_devices->seed = fs_devices;
6342 out:
6343 return fs_devices;
6344 }
6345
6346 static int read_one_dev(struct btrfs_root *root,
6347 struct extent_buffer *leaf,
6348 struct btrfs_dev_item *dev_item)
6349 {
6350 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6351 struct btrfs_device *device;
6352 u64 devid;
6353 int ret;
6354 u8 fs_uuid[BTRFS_UUID_SIZE];
6355 u8 dev_uuid[BTRFS_UUID_SIZE];
6356
6357 devid = btrfs_device_id(leaf, dev_item);
6358 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6359 BTRFS_UUID_SIZE);
6360 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6361 BTRFS_UUID_SIZE);
6362
6363 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6364 fs_devices = open_seed_devices(root, fs_uuid);
6365 if (IS_ERR(fs_devices))
6366 return PTR_ERR(fs_devices);
6367 }
6368
6369 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6370 if (!device) {
6371 if (!btrfs_test_opt(root, DEGRADED))
6372 return -EIO;
6373
6374 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6375 if (!device)
6376 return -ENOMEM;
6377 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6378 devid, dev_uuid);
6379 } else {
6380 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6381 return -EIO;
6382
6383 if(!device->bdev && !device->missing) {
6384 /*
6385 * this happens when a device that was properly setup
6386 * in the device info lists suddenly goes bad.
6387 * device->bdev is NULL, and so we have to set
6388 * device->missing to one here
6389 */
6390 device->fs_devices->missing_devices++;
6391 device->missing = 1;
6392 }
6393
6394 /* Move the device to its own fs_devices */
6395 if (device->fs_devices != fs_devices) {
6396 ASSERT(device->missing);
6397
6398 list_move(&device->dev_list, &fs_devices->devices);
6399 device->fs_devices->num_devices--;
6400 fs_devices->num_devices++;
6401
6402 device->fs_devices->missing_devices--;
6403 fs_devices->missing_devices++;
6404
6405 device->fs_devices = fs_devices;
6406 }
6407 }
6408
6409 if (device->fs_devices != root->fs_info->fs_devices) {
6410 BUG_ON(device->writeable);
6411 if (device->generation !=
6412 btrfs_device_generation(leaf, dev_item))
6413 return -EINVAL;
6414 }
6415
6416 fill_device_from_item(leaf, dev_item, device);
6417 device->in_fs_metadata = 1;
6418 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6419 device->fs_devices->total_rw_bytes += device->total_bytes;
6420 spin_lock(&root->fs_info->free_chunk_lock);
6421 root->fs_info->free_chunk_space += device->total_bytes -
6422 device->bytes_used;
6423 spin_unlock(&root->fs_info->free_chunk_lock);
6424 }
6425 ret = 0;
6426 return ret;
6427 }
6428
6429 int btrfs_read_sys_array(struct btrfs_root *root)
6430 {
6431 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6432 struct extent_buffer *sb;
6433 struct btrfs_disk_key *disk_key;
6434 struct btrfs_chunk *chunk;
6435 u8 *array_ptr;
6436 unsigned long sb_array_offset;
6437 int ret = 0;
6438 u32 num_stripes;
6439 u32 array_size;
6440 u32 len = 0;
6441 u32 cur_offset;
6442 struct btrfs_key key;
6443
6444 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6445 /*
6446 * This will create extent buffer of nodesize, superblock size is
6447 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6448 * overallocate but we can keep it as-is, only the first page is used.
6449 */
6450 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6451 if (!sb)
6452 return -ENOMEM;
6453 btrfs_set_buffer_uptodate(sb);
6454 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6455 /*
6456 * The sb extent buffer is artifical and just used to read the system array.
6457 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6458 * pages up-to-date when the page is larger: extent does not cover the
6459 * whole page and consequently check_page_uptodate does not find all
6460 * the page's extents up-to-date (the hole beyond sb),
6461 * write_extent_buffer then triggers a WARN_ON.
6462 *
6463 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6464 * but sb spans only this function. Add an explicit SetPageUptodate call
6465 * to silence the warning eg. on PowerPC 64.
6466 */
6467 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6468 SetPageUptodate(sb->pages[0]);
6469
6470 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6471 array_size = btrfs_super_sys_array_size(super_copy);
6472
6473 array_ptr = super_copy->sys_chunk_array;
6474 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6475 cur_offset = 0;
6476
6477 while (cur_offset < array_size) {
6478 disk_key = (struct btrfs_disk_key *)array_ptr;
6479 len = sizeof(*disk_key);
6480 if (cur_offset + len > array_size)
6481 goto out_short_read;
6482
6483 btrfs_disk_key_to_cpu(&key, disk_key);
6484
6485 array_ptr += len;
6486 sb_array_offset += len;
6487 cur_offset += len;
6488
6489 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6490 chunk = (struct btrfs_chunk *)sb_array_offset;
6491 /*
6492 * At least one btrfs_chunk with one stripe must be
6493 * present, exact stripe count check comes afterwards
6494 */
6495 len = btrfs_chunk_item_size(1);
6496 if (cur_offset + len > array_size)
6497 goto out_short_read;
6498
6499 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6500 len = btrfs_chunk_item_size(num_stripes);
6501 if (cur_offset + len > array_size)
6502 goto out_short_read;
6503
6504 ret = read_one_chunk(root, &key, sb, chunk);
6505 if (ret)
6506 break;
6507 } else {
6508 ret = -EIO;
6509 break;
6510 }
6511 array_ptr += len;
6512 sb_array_offset += len;
6513 cur_offset += len;
6514 }
6515 free_extent_buffer(sb);
6516 return ret;
6517
6518 out_short_read:
6519 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6520 len, cur_offset);
6521 free_extent_buffer(sb);
6522 return -EIO;
6523 }
6524
6525 int btrfs_read_chunk_tree(struct btrfs_root *root)
6526 {
6527 struct btrfs_path *path;
6528 struct extent_buffer *leaf;
6529 struct btrfs_key key;
6530 struct btrfs_key found_key;
6531 int ret;
6532 int slot;
6533
6534 root = root->fs_info->chunk_root;
6535
6536 path = btrfs_alloc_path();
6537 if (!path)
6538 return -ENOMEM;
6539
6540 mutex_lock(&uuid_mutex);
6541 lock_chunks(root);
6542
6543 /*
6544 * Read all device items, and then all the chunk items. All
6545 * device items are found before any chunk item (their object id
6546 * is smaller than the lowest possible object id for a chunk
6547 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6548 */
6549 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6550 key.offset = 0;
6551 key.type = 0;
6552 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6553 if (ret < 0)
6554 goto error;
6555 while (1) {
6556 leaf = path->nodes[0];
6557 slot = path->slots[0];
6558 if (slot >= btrfs_header_nritems(leaf)) {
6559 ret = btrfs_next_leaf(root, path);
6560 if (ret == 0)
6561 continue;
6562 if (ret < 0)
6563 goto error;
6564 break;
6565 }
6566 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6567 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6568 struct btrfs_dev_item *dev_item;
6569 dev_item = btrfs_item_ptr(leaf, slot,
6570 struct btrfs_dev_item);
6571 ret = read_one_dev(root, leaf, dev_item);
6572 if (ret)
6573 goto error;
6574 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6575 struct btrfs_chunk *chunk;
6576 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6577 ret = read_one_chunk(root, &found_key, leaf, chunk);
6578 if (ret)
6579 goto error;
6580 }
6581 path->slots[0]++;
6582 }
6583 ret = 0;
6584 error:
6585 unlock_chunks(root);
6586 mutex_unlock(&uuid_mutex);
6587
6588 btrfs_free_path(path);
6589 return ret;
6590 }
6591
6592 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6593 {
6594 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6595 struct btrfs_device *device;
6596
6597 while (fs_devices) {
6598 mutex_lock(&fs_devices->device_list_mutex);
6599 list_for_each_entry(device, &fs_devices->devices, dev_list)
6600 device->dev_root = fs_info->dev_root;
6601 mutex_unlock(&fs_devices->device_list_mutex);
6602
6603 fs_devices = fs_devices->seed;
6604 }
6605 }
6606
6607 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6608 {
6609 int i;
6610
6611 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6612 btrfs_dev_stat_reset(dev, i);
6613 }
6614
6615 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6616 {
6617 struct btrfs_key key;
6618 struct btrfs_key found_key;
6619 struct btrfs_root *dev_root = fs_info->dev_root;
6620 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6621 struct extent_buffer *eb;
6622 int slot;
6623 int ret = 0;
6624 struct btrfs_device *device;
6625 struct btrfs_path *path = NULL;
6626 int i;
6627
6628 path = btrfs_alloc_path();
6629 if (!path) {
6630 ret = -ENOMEM;
6631 goto out;
6632 }
6633
6634 mutex_lock(&fs_devices->device_list_mutex);
6635 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6636 int item_size;
6637 struct btrfs_dev_stats_item *ptr;
6638
6639 key.objectid = 0;
6640 key.type = BTRFS_DEV_STATS_KEY;
6641 key.offset = device->devid;
6642 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6643 if (ret) {
6644 __btrfs_reset_dev_stats(device);
6645 device->dev_stats_valid = 1;
6646 btrfs_release_path(path);
6647 continue;
6648 }
6649 slot = path->slots[0];
6650 eb = path->nodes[0];
6651 btrfs_item_key_to_cpu(eb, &found_key, slot);
6652 item_size = btrfs_item_size_nr(eb, slot);
6653
6654 ptr = btrfs_item_ptr(eb, slot,
6655 struct btrfs_dev_stats_item);
6656
6657 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6658 if (item_size >= (1 + i) * sizeof(__le64))
6659 btrfs_dev_stat_set(device, i,
6660 btrfs_dev_stats_value(eb, ptr, i));
6661 else
6662 btrfs_dev_stat_reset(device, i);
6663 }
6664
6665 device->dev_stats_valid = 1;
6666 btrfs_dev_stat_print_on_load(device);
6667 btrfs_release_path(path);
6668 }
6669 mutex_unlock(&fs_devices->device_list_mutex);
6670
6671 out:
6672 btrfs_free_path(path);
6673 return ret < 0 ? ret : 0;
6674 }
6675
6676 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6677 struct btrfs_root *dev_root,
6678 struct btrfs_device *device)
6679 {
6680 struct btrfs_path *path;
6681 struct btrfs_key key;
6682 struct extent_buffer *eb;
6683 struct btrfs_dev_stats_item *ptr;
6684 int ret;
6685 int i;
6686
6687 key.objectid = 0;
6688 key.type = BTRFS_DEV_STATS_KEY;
6689 key.offset = device->devid;
6690
6691 path = btrfs_alloc_path();
6692 BUG_ON(!path);
6693 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6694 if (ret < 0) {
6695 printk_in_rcu(KERN_WARNING "BTRFS: "
6696 "error %d while searching for dev_stats item for device %s!\n",
6697 ret, rcu_str_deref(device->name));
6698 goto out;
6699 }
6700
6701 if (ret == 0 &&
6702 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6703 /* need to delete old one and insert a new one */
6704 ret = btrfs_del_item(trans, dev_root, path);
6705 if (ret != 0) {
6706 printk_in_rcu(KERN_WARNING "BTRFS: "
6707 "delete too small dev_stats item for device %s failed %d!\n",
6708 rcu_str_deref(device->name), ret);
6709 goto out;
6710 }
6711 ret = 1;
6712 }
6713
6714 if (ret == 1) {
6715 /* need to insert a new item */
6716 btrfs_release_path(path);
6717 ret = btrfs_insert_empty_item(trans, dev_root, path,
6718 &key, sizeof(*ptr));
6719 if (ret < 0) {
6720 printk_in_rcu(KERN_WARNING "BTRFS: "
6721 "insert dev_stats item for device %s failed %d!\n",
6722 rcu_str_deref(device->name), ret);
6723 goto out;
6724 }
6725 }
6726
6727 eb = path->nodes[0];
6728 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6729 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6730 btrfs_set_dev_stats_value(eb, ptr, i,
6731 btrfs_dev_stat_read(device, i));
6732 btrfs_mark_buffer_dirty(eb);
6733
6734 out:
6735 btrfs_free_path(path);
6736 return ret;
6737 }
6738
6739 /*
6740 * called from commit_transaction. Writes all changed device stats to disk.
6741 */
6742 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6743 struct btrfs_fs_info *fs_info)
6744 {
6745 struct btrfs_root *dev_root = fs_info->dev_root;
6746 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6747 struct btrfs_device *device;
6748 int stats_cnt;
6749 int ret = 0;
6750
6751 mutex_lock(&fs_devices->device_list_mutex);
6752 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6753 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6754 continue;
6755
6756 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6757 ret = update_dev_stat_item(trans, dev_root, device);
6758 if (!ret)
6759 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6760 }
6761 mutex_unlock(&fs_devices->device_list_mutex);
6762
6763 return ret;
6764 }
6765
6766 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6767 {
6768 btrfs_dev_stat_inc(dev, index);
6769 btrfs_dev_stat_print_on_error(dev);
6770 }
6771
6772 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6773 {
6774 if (!dev->dev_stats_valid)
6775 return;
6776 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6777 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6778 rcu_str_deref(dev->name),
6779 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6780 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6781 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6782 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6783 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6784 }
6785
6786 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6787 {
6788 int i;
6789
6790 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6791 if (btrfs_dev_stat_read(dev, i) != 0)
6792 break;
6793 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6794 return; /* all values == 0, suppress message */
6795
6796 printk_in_rcu(KERN_INFO "BTRFS: "
6797 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6798 rcu_str_deref(dev->name),
6799 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6800 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6801 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6802 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6803 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6804 }
6805
6806 int btrfs_get_dev_stats(struct btrfs_root *root,
6807 struct btrfs_ioctl_get_dev_stats *stats)
6808 {
6809 struct btrfs_device *dev;
6810 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6811 int i;
6812
6813 mutex_lock(&fs_devices->device_list_mutex);
6814 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6815 mutex_unlock(&fs_devices->device_list_mutex);
6816
6817 if (!dev) {
6818 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6819 return -ENODEV;
6820 } else if (!dev->dev_stats_valid) {
6821 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6822 return -ENODEV;
6823 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6824 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6825 if (stats->nr_items > i)
6826 stats->values[i] =
6827 btrfs_dev_stat_read_and_reset(dev, i);
6828 else
6829 btrfs_dev_stat_reset(dev, i);
6830 }
6831 } else {
6832 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6833 if (stats->nr_items > i)
6834 stats->values[i] = btrfs_dev_stat_read(dev, i);
6835 }
6836 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6837 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6838 return 0;
6839 }
6840
6841 int btrfs_scratch_superblock(struct btrfs_device *device)
6842 {
6843 struct buffer_head *bh;
6844 struct btrfs_super_block *disk_super;
6845
6846 bh = btrfs_read_dev_super(device->bdev);
6847 if (!bh)
6848 return -EINVAL;
6849 disk_super = (struct btrfs_super_block *)bh->b_data;
6850
6851 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6852 set_buffer_dirty(bh);
6853 sync_dirty_buffer(bh);
6854 brelse(bh);
6855
6856 return 0;
6857 }
6858
6859 /*
6860 * Update the size of all devices, which is used for writing out the
6861 * super blocks.
6862 */
6863 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6864 {
6865 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6866 struct btrfs_device *curr, *next;
6867
6868 if (list_empty(&fs_devices->resized_devices))
6869 return;
6870
6871 mutex_lock(&fs_devices->device_list_mutex);
6872 lock_chunks(fs_info->dev_root);
6873 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6874 resized_list) {
6875 list_del_init(&curr->resized_list);
6876 curr->commit_total_bytes = curr->disk_total_bytes;
6877 }
6878 unlock_chunks(fs_info->dev_root);
6879 mutex_unlock(&fs_devices->device_list_mutex);
6880 }
6881
6882 /* Must be invoked during the transaction commit */
6883 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6884 struct btrfs_transaction *transaction)
6885 {
6886 struct extent_map *em;
6887 struct map_lookup *map;
6888 struct btrfs_device *dev;
6889 int i;
6890
6891 if (list_empty(&transaction->pending_chunks))
6892 return;
6893
6894 /* In order to kick the device replace finish process */
6895 lock_chunks(root);
6896 list_for_each_entry(em, &transaction->pending_chunks, list) {
6897 map = (struct map_lookup *)em->bdev;
6898
6899 for (i = 0; i < map->num_stripes; i++) {
6900 dev = map->stripes[i].dev;
6901 dev->commit_bytes_used = dev->bytes_used;
6902 }
6903 }
6904 unlock_chunks(root);
6905 }
6906
6907 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6908 {
6909 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6910 while (fs_devices) {
6911 fs_devices->fs_info = fs_info;
6912 fs_devices = fs_devices->seed;
6913 }
6914 }
6915
6916 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6917 {
6918 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6919 while (fs_devices) {
6920 fs_devices->fs_info = NULL;
6921 fs_devices = fs_devices->seed;
6922 }
6923 }
Ubuntu Artful kernel mirror