]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/btrfs/volumes.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: cleanup, stop casting for extent_map->lookup everywhere
[mirror_ubuntu-jammy-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 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
47 .sub_stripes = 2,
48 .dev_stripes = 1,
49 .devs_max = 0, /* 0 == as many as possible */
50 .devs_min = 4,
51 .tolerated_failures = 1,
52 .devs_increment = 2,
53 .ncopies = 2,
54 },
55 [BTRFS_RAID_RAID1] = {
56 .sub_stripes = 1,
57 .dev_stripes = 1,
58 .devs_max = 2,
59 .devs_min = 2,
60 .tolerated_failures = 1,
61 .devs_increment = 2,
62 .ncopies = 2,
63 },
64 [BTRFS_RAID_DUP] = {
65 .sub_stripes = 1,
66 .dev_stripes = 2,
67 .devs_max = 1,
68 .devs_min = 1,
69 .tolerated_failures = 0,
70 .devs_increment = 1,
71 .ncopies = 2,
72 },
73 [BTRFS_RAID_RAID0] = {
74 .sub_stripes = 1,
75 .dev_stripes = 1,
76 .devs_max = 0,
77 .devs_min = 2,
78 .tolerated_failures = 0,
79 .devs_increment = 1,
80 .ncopies = 1,
81 },
82 [BTRFS_RAID_SINGLE] = {
83 .sub_stripes = 1,
84 .dev_stripes = 1,
85 .devs_max = 1,
86 .devs_min = 1,
87 .tolerated_failures = 0,
88 .devs_increment = 1,
89 .ncopies = 1,
90 },
91 [BTRFS_RAID_RAID5] = {
92 .sub_stripes = 1,
93 .dev_stripes = 1,
94 .devs_max = 0,
95 .devs_min = 2,
96 .tolerated_failures = 1,
97 .devs_increment = 1,
98 .ncopies = 2,
99 },
100 [BTRFS_RAID_RAID6] = {
101 .sub_stripes = 1,
102 .dev_stripes = 1,
103 .devs_max = 0,
104 .devs_min = 3,
105 .tolerated_failures = 2,
106 .devs_increment = 1,
107 .ncopies = 3,
108 },
109 };
110
111 const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
119 };
120
121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root,
123 struct btrfs_device *device);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
128
129 DEFINE_MUTEX(uuid_mutex);
130 static LIST_HEAD(fs_uuids);
131 struct list_head *btrfs_get_fs_uuids(void)
132 {
133 return &fs_uuids;
134 }
135
136 static struct btrfs_fs_devices *__alloc_fs_devices(void)
137 {
138 struct btrfs_fs_devices *fs_devs;
139
140 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
141 if (!fs_devs)
142 return ERR_PTR(-ENOMEM);
143
144 mutex_init(&fs_devs->device_list_mutex);
145
146 INIT_LIST_HEAD(&fs_devs->devices);
147 INIT_LIST_HEAD(&fs_devs->resized_devices);
148 INIT_LIST_HEAD(&fs_devs->alloc_list);
149 INIT_LIST_HEAD(&fs_devs->list);
150
151 return fs_devs;
152 }
153
154 /**
155 * alloc_fs_devices - allocate struct btrfs_fs_devices
156 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
157 * generated.
158 *
159 * Return: a pointer to a new &struct btrfs_fs_devices on success;
160 * ERR_PTR() on error. Returned struct is not linked onto any lists and
161 * can be destroyed with kfree() right away.
162 */
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
164 {
165 struct btrfs_fs_devices *fs_devs;
166
167 fs_devs = __alloc_fs_devices();
168 if (IS_ERR(fs_devs))
169 return fs_devs;
170
171 if (fsid)
172 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
173 else
174 generate_random_uuid(fs_devs->fsid);
175
176 return fs_devs;
177 }
178
179 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
180 {
181 struct btrfs_device *device;
182 WARN_ON(fs_devices->opened);
183 while (!list_empty(&fs_devices->devices)) {
184 device = list_entry(fs_devices->devices.next,
185 struct btrfs_device, dev_list);
186 list_del(&device->dev_list);
187 rcu_string_free(device->name);
188 kfree(device);
189 }
190 kfree(fs_devices);
191 }
192
193 static void btrfs_kobject_uevent(struct block_device *bdev,
194 enum kobject_action action)
195 {
196 int ret;
197
198 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
199 if (ret)
200 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
201 action,
202 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
203 &disk_to_dev(bdev->bd_disk)->kobj);
204 }
205
206 void btrfs_cleanup_fs_uuids(void)
207 {
208 struct btrfs_fs_devices *fs_devices;
209
210 while (!list_empty(&fs_uuids)) {
211 fs_devices = list_entry(fs_uuids.next,
212 struct btrfs_fs_devices, list);
213 list_del(&fs_devices->list);
214 free_fs_devices(fs_devices);
215 }
216 }
217
218 static struct btrfs_device *__alloc_device(void)
219 {
220 struct btrfs_device *dev;
221
222 dev = kzalloc(sizeof(*dev), GFP_NOFS);
223 if (!dev)
224 return ERR_PTR(-ENOMEM);
225
226 INIT_LIST_HEAD(&dev->dev_list);
227 INIT_LIST_HEAD(&dev->dev_alloc_list);
228 INIT_LIST_HEAD(&dev->resized_list);
229
230 spin_lock_init(&dev->io_lock);
231
232 spin_lock_init(&dev->reada_lock);
233 atomic_set(&dev->reada_in_flight, 0);
234 atomic_set(&dev->dev_stats_ccnt, 0);
235 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
236 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
237
238 return dev;
239 }
240
241 static noinline struct btrfs_device *__find_device(struct list_head *head,
242 u64 devid, u8 *uuid)
243 {
244 struct btrfs_device *dev;
245
246 list_for_each_entry(dev, head, dev_list) {
247 if (dev->devid == devid &&
248 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
249 return dev;
250 }
251 }
252 return NULL;
253 }
254
255 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
256 {
257 struct btrfs_fs_devices *fs_devices;
258
259 list_for_each_entry(fs_devices, &fs_uuids, list) {
260 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
261 return fs_devices;
262 }
263 return NULL;
264 }
265
266 static int
267 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
268 int flush, struct block_device **bdev,
269 struct buffer_head **bh)
270 {
271 int ret;
272
273 *bdev = blkdev_get_by_path(device_path, flags, holder);
274
275 if (IS_ERR(*bdev)) {
276 ret = PTR_ERR(*bdev);
277 goto error;
278 }
279
280 if (flush)
281 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
282 ret = set_blocksize(*bdev, 4096);
283 if (ret) {
284 blkdev_put(*bdev, flags);
285 goto error;
286 }
287 invalidate_bdev(*bdev);
288 *bh = btrfs_read_dev_super(*bdev);
289 if (IS_ERR(*bh)) {
290 ret = PTR_ERR(*bh);
291 blkdev_put(*bdev, flags);
292 goto error;
293 }
294
295 return 0;
296
297 error:
298 *bdev = NULL;
299 *bh = NULL;
300 return ret;
301 }
302
303 static void requeue_list(struct btrfs_pending_bios *pending_bios,
304 struct bio *head, struct bio *tail)
305 {
306
307 struct bio *old_head;
308
309 old_head = pending_bios->head;
310 pending_bios->head = head;
311 if (pending_bios->tail)
312 tail->bi_next = old_head;
313 else
314 pending_bios->tail = tail;
315 }
316
317 /*
318 * we try to collect pending bios for a device so we don't get a large
319 * number of procs sending bios down to the same device. This greatly
320 * improves the schedulers ability to collect and merge the bios.
321 *
322 * But, it also turns into a long list of bios to process and that is sure
323 * to eventually make the worker thread block. The solution here is to
324 * make some progress and then put this work struct back at the end of
325 * the list if the block device is congested. This way, multiple devices
326 * can make progress from a single worker thread.
327 */
328 static noinline void run_scheduled_bios(struct btrfs_device *device)
329 {
330 struct bio *pending;
331 struct backing_dev_info *bdi;
332 struct btrfs_fs_info *fs_info;
333 struct btrfs_pending_bios *pending_bios;
334 struct bio *tail;
335 struct bio *cur;
336 int again = 0;
337 unsigned long num_run;
338 unsigned long batch_run = 0;
339 unsigned long limit;
340 unsigned long last_waited = 0;
341 int force_reg = 0;
342 int sync_pending = 0;
343 struct blk_plug plug;
344
345 /*
346 * this function runs all the bios we've collected for
347 * a particular device. We don't want to wander off to
348 * another device without first sending all of these down.
349 * So, setup a plug here and finish it off before we return
350 */
351 blk_start_plug(&plug);
352
353 bdi = blk_get_backing_dev_info(device->bdev);
354 fs_info = device->dev_root->fs_info;
355 limit = btrfs_async_submit_limit(fs_info);
356 limit = limit * 2 / 3;
357
358 loop:
359 spin_lock(&device->io_lock);
360
361 loop_lock:
362 num_run = 0;
363
364 /* take all the bios off the list at once and process them
365 * later on (without the lock held). But, remember the
366 * tail and other pointers so the bios can be properly reinserted
367 * into the list if we hit congestion
368 */
369 if (!force_reg && device->pending_sync_bios.head) {
370 pending_bios = &device->pending_sync_bios;
371 force_reg = 1;
372 } else {
373 pending_bios = &device->pending_bios;
374 force_reg = 0;
375 }
376
377 pending = pending_bios->head;
378 tail = pending_bios->tail;
379 WARN_ON(pending && !tail);
380
381 /*
382 * if pending was null this time around, no bios need processing
383 * at all and we can stop. Otherwise it'll loop back up again
384 * and do an additional check so no bios are missed.
385 *
386 * device->running_pending is used to synchronize with the
387 * schedule_bio code.
388 */
389 if (device->pending_sync_bios.head == NULL &&
390 device->pending_bios.head == NULL) {
391 again = 0;
392 device->running_pending = 0;
393 } else {
394 again = 1;
395 device->running_pending = 1;
396 }
397
398 pending_bios->head = NULL;
399 pending_bios->tail = NULL;
400
401 spin_unlock(&device->io_lock);
402
403 while (pending) {
404
405 rmb();
406 /* we want to work on both lists, but do more bios on the
407 * sync list than the regular list
408 */
409 if ((num_run > 32 &&
410 pending_bios != &device->pending_sync_bios &&
411 device->pending_sync_bios.head) ||
412 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
413 device->pending_bios.head)) {
414 spin_lock(&device->io_lock);
415 requeue_list(pending_bios, pending, tail);
416 goto loop_lock;
417 }
418
419 cur = pending;
420 pending = pending->bi_next;
421 cur->bi_next = NULL;
422
423 /*
424 * atomic_dec_return implies a barrier for waitqueue_active
425 */
426 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
427 waitqueue_active(&fs_info->async_submit_wait))
428 wake_up(&fs_info->async_submit_wait);
429
430 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
431
432 /*
433 * if we're doing the sync list, record that our
434 * plug has some sync requests on it
435 *
436 * If we're doing the regular list and there are
437 * sync requests sitting around, unplug before
438 * we add more
439 */
440 if (pending_bios == &device->pending_sync_bios) {
441 sync_pending = 1;
442 } else if (sync_pending) {
443 blk_finish_plug(&plug);
444 blk_start_plug(&plug);
445 sync_pending = 0;
446 }
447
448 btrfsic_submit_bio(cur->bi_rw, cur);
449 num_run++;
450 batch_run++;
451
452 cond_resched();
453
454 /*
455 * we made progress, there is more work to do and the bdi
456 * is now congested. Back off and let other work structs
457 * run instead
458 */
459 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
460 fs_info->fs_devices->open_devices > 1) {
461 struct io_context *ioc;
462
463 ioc = current->io_context;
464
465 /*
466 * the main goal here is that we don't want to
467 * block if we're going to be able to submit
468 * more requests without blocking.
469 *
470 * This code does two great things, it pokes into
471 * the elevator code from a filesystem _and_
472 * it makes assumptions about how batching works.
473 */
474 if (ioc && ioc->nr_batch_requests > 0 &&
475 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
476 (last_waited == 0 ||
477 ioc->last_waited == last_waited)) {
478 /*
479 * we want to go through our batch of
480 * requests and stop. So, we copy out
481 * the ioc->last_waited time and test
482 * against it before looping
483 */
484 last_waited = ioc->last_waited;
485 cond_resched();
486 continue;
487 }
488 spin_lock(&device->io_lock);
489 requeue_list(pending_bios, pending, tail);
490 device->running_pending = 1;
491
492 spin_unlock(&device->io_lock);
493 btrfs_queue_work(fs_info->submit_workers,
494 &device->work);
495 goto done;
496 }
497 /* unplug every 64 requests just for good measure */
498 if (batch_run % 64 == 0) {
499 blk_finish_plug(&plug);
500 blk_start_plug(&plug);
501 sync_pending = 0;
502 }
503 }
504
505 cond_resched();
506 if (again)
507 goto loop;
508
509 spin_lock(&device->io_lock);
510 if (device->pending_bios.head || device->pending_sync_bios.head)
511 goto loop_lock;
512 spin_unlock(&device->io_lock);
513
514 done:
515 blk_finish_plug(&plug);
516 }
517
518 static void pending_bios_fn(struct btrfs_work *work)
519 {
520 struct btrfs_device *device;
521
522 device = container_of(work, struct btrfs_device, work);
523 run_scheduled_bios(device);
524 }
525
526
527 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
528 {
529 struct btrfs_fs_devices *fs_devs;
530 struct btrfs_device *dev;
531
532 if (!cur_dev->name)
533 return;
534
535 list_for_each_entry(fs_devs, &fs_uuids, list) {
536 int del = 1;
537
538 if (fs_devs->opened)
539 continue;
540 if (fs_devs->seeding)
541 continue;
542
543 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
544
545 if (dev == cur_dev)
546 continue;
547 if (!dev->name)
548 continue;
549
550 /*
551 * Todo: This won't be enough. What if the same device
552 * comes back (with new uuid and) with its mapper path?
553 * But for now, this does help as mostly an admin will
554 * either use mapper or non mapper path throughout.
555 */
556 rcu_read_lock();
557 del = strcmp(rcu_str_deref(dev->name),
558 rcu_str_deref(cur_dev->name));
559 rcu_read_unlock();
560 if (!del)
561 break;
562 }
563
564 if (!del) {
565 /* delete the stale device */
566 if (fs_devs->num_devices == 1) {
567 btrfs_sysfs_remove_fsid(fs_devs);
568 list_del(&fs_devs->list);
569 free_fs_devices(fs_devs);
570 } else {
571 fs_devs->num_devices--;
572 list_del(&dev->dev_list);
573 rcu_string_free(dev->name);
574 kfree(dev);
575 }
576 break;
577 }
578 }
579 }
580
581 /*
582 * Add new device to list of registered devices
583 *
584 * Returns:
585 * 1 - first time device is seen
586 * 0 - device already known
587 * < 0 - error
588 */
589 static noinline int device_list_add(const char *path,
590 struct btrfs_super_block *disk_super,
591 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
592 {
593 struct btrfs_device *device;
594 struct btrfs_fs_devices *fs_devices;
595 struct rcu_string *name;
596 int ret = 0;
597 u64 found_transid = btrfs_super_generation(disk_super);
598
599 fs_devices = find_fsid(disk_super->fsid);
600 if (!fs_devices) {
601 fs_devices = alloc_fs_devices(disk_super->fsid);
602 if (IS_ERR(fs_devices))
603 return PTR_ERR(fs_devices);
604
605 list_add(&fs_devices->list, &fs_uuids);
606
607 device = NULL;
608 } else {
609 device = __find_device(&fs_devices->devices, devid,
610 disk_super->dev_item.uuid);
611 }
612
613 if (!device) {
614 if (fs_devices->opened)
615 return -EBUSY;
616
617 device = btrfs_alloc_device(NULL, &devid,
618 disk_super->dev_item.uuid);
619 if (IS_ERR(device)) {
620 /* we can safely leave the fs_devices entry around */
621 return PTR_ERR(device);
622 }
623
624 name = rcu_string_strdup(path, GFP_NOFS);
625 if (!name) {
626 kfree(device);
627 return -ENOMEM;
628 }
629 rcu_assign_pointer(device->name, name);
630
631 mutex_lock(&fs_devices->device_list_mutex);
632 list_add_rcu(&device->dev_list, &fs_devices->devices);
633 fs_devices->num_devices++;
634 mutex_unlock(&fs_devices->device_list_mutex);
635
636 ret = 1;
637 device->fs_devices = fs_devices;
638 } else if (!device->name || strcmp(device->name->str, path)) {
639 /*
640 * When FS is already mounted.
641 * 1. If you are here and if the device->name is NULL that
642 * means this device was missing at time of FS mount.
643 * 2. If you are here and if the device->name is different
644 * from 'path' that means either
645 * a. The same device disappeared and reappeared with
646 * different name. or
647 * b. The missing-disk-which-was-replaced, has
648 * reappeared now.
649 *
650 * We must allow 1 and 2a above. But 2b would be a spurious
651 * and unintentional.
652 *
653 * Further in case of 1 and 2a above, the disk at 'path'
654 * would have missed some transaction when it was away and
655 * in case of 2a the stale bdev has to be updated as well.
656 * 2b must not be allowed at all time.
657 */
658
659 /*
660 * For now, we do allow update to btrfs_fs_device through the
661 * btrfs dev scan cli after FS has been mounted. We're still
662 * tracking a problem where systems fail mount by subvolume id
663 * when we reject replacement on a mounted FS.
664 */
665 if (!fs_devices->opened && found_transid < device->generation) {
666 /*
667 * That is if the FS is _not_ mounted and if you
668 * are here, that means there is more than one
669 * disk with same uuid and devid.We keep the one
670 * with larger generation number or the last-in if
671 * generation are equal.
672 */
673 return -EEXIST;
674 }
675
676 name = rcu_string_strdup(path, GFP_NOFS);
677 if (!name)
678 return -ENOMEM;
679 rcu_string_free(device->name);
680 rcu_assign_pointer(device->name, name);
681 if (device->missing) {
682 fs_devices->missing_devices--;
683 device->missing = 0;
684 }
685 }
686
687 /*
688 * Unmount does not free the btrfs_device struct but would zero
689 * generation along with most of the other members. So just update
690 * it back. We need it to pick the disk with largest generation
691 * (as above).
692 */
693 if (!fs_devices->opened)
694 device->generation = found_transid;
695
696 /*
697 * if there is new btrfs on an already registered device,
698 * then remove the stale device entry.
699 */
700 btrfs_free_stale_device(device);
701
702 *fs_devices_ret = fs_devices;
703
704 return ret;
705 }
706
707 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
708 {
709 struct btrfs_fs_devices *fs_devices;
710 struct btrfs_device *device;
711 struct btrfs_device *orig_dev;
712
713 fs_devices = alloc_fs_devices(orig->fsid);
714 if (IS_ERR(fs_devices))
715 return fs_devices;
716
717 mutex_lock(&orig->device_list_mutex);
718 fs_devices->total_devices = orig->total_devices;
719
720 /* We have held the volume lock, it is safe to get the devices. */
721 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
722 struct rcu_string *name;
723
724 device = btrfs_alloc_device(NULL, &orig_dev->devid,
725 orig_dev->uuid);
726 if (IS_ERR(device))
727 goto error;
728
729 /*
730 * This is ok to do without rcu read locked because we hold the
731 * uuid mutex so nothing we touch in here is going to disappear.
732 */
733 if (orig_dev->name) {
734 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
735 if (!name) {
736 kfree(device);
737 goto error;
738 }
739 rcu_assign_pointer(device->name, name);
740 }
741
742 list_add(&device->dev_list, &fs_devices->devices);
743 device->fs_devices = fs_devices;
744 fs_devices->num_devices++;
745 }
746 mutex_unlock(&orig->device_list_mutex);
747 return fs_devices;
748 error:
749 mutex_unlock(&orig->device_list_mutex);
750 free_fs_devices(fs_devices);
751 return ERR_PTR(-ENOMEM);
752 }
753
754 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
755 {
756 struct btrfs_device *device, *next;
757 struct btrfs_device *latest_dev = NULL;
758
759 mutex_lock(&uuid_mutex);
760 again:
761 /* This is the initialized path, it is safe to release the devices. */
762 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
763 if (device->in_fs_metadata) {
764 if (!device->is_tgtdev_for_dev_replace &&
765 (!latest_dev ||
766 device->generation > latest_dev->generation)) {
767 latest_dev = device;
768 }
769 continue;
770 }
771
772 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
773 /*
774 * In the first step, keep the device which has
775 * the correct fsid and the devid that is used
776 * for the dev_replace procedure.
777 * In the second step, the dev_replace state is
778 * read from the device tree and it is known
779 * whether the procedure is really active or
780 * not, which means whether this device is
781 * used or whether it should be removed.
782 */
783 if (step == 0 || device->is_tgtdev_for_dev_replace) {
784 continue;
785 }
786 }
787 if (device->bdev) {
788 blkdev_put(device->bdev, device->mode);
789 device->bdev = NULL;
790 fs_devices->open_devices--;
791 }
792 if (device->writeable) {
793 list_del_init(&device->dev_alloc_list);
794 device->writeable = 0;
795 if (!device->is_tgtdev_for_dev_replace)
796 fs_devices->rw_devices--;
797 }
798 list_del_init(&device->dev_list);
799 fs_devices->num_devices--;
800 rcu_string_free(device->name);
801 kfree(device);
802 }
803
804 if (fs_devices->seed) {
805 fs_devices = fs_devices->seed;
806 goto again;
807 }
808
809 fs_devices->latest_bdev = latest_dev->bdev;
810
811 mutex_unlock(&uuid_mutex);
812 }
813
814 static void __free_device(struct work_struct *work)
815 {
816 struct btrfs_device *device;
817
818 device = container_of(work, struct btrfs_device, rcu_work);
819
820 if (device->bdev)
821 blkdev_put(device->bdev, device->mode);
822
823 rcu_string_free(device->name);
824 kfree(device);
825 }
826
827 static void free_device(struct rcu_head *head)
828 {
829 struct btrfs_device *device;
830
831 device = container_of(head, struct btrfs_device, rcu);
832
833 INIT_WORK(&device->rcu_work, __free_device);
834 schedule_work(&device->rcu_work);
835 }
836
837 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
838 {
839 struct btrfs_device *device, *tmp;
840
841 if (--fs_devices->opened > 0)
842 return 0;
843
844 mutex_lock(&fs_devices->device_list_mutex);
845 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
846 btrfs_close_one_device(device);
847 }
848 mutex_unlock(&fs_devices->device_list_mutex);
849
850 WARN_ON(fs_devices->open_devices);
851 WARN_ON(fs_devices->rw_devices);
852 fs_devices->opened = 0;
853 fs_devices->seeding = 0;
854
855 return 0;
856 }
857
858 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
859 {
860 struct btrfs_fs_devices *seed_devices = NULL;
861 int ret;
862
863 mutex_lock(&uuid_mutex);
864 ret = __btrfs_close_devices(fs_devices);
865 if (!fs_devices->opened) {
866 seed_devices = fs_devices->seed;
867 fs_devices->seed = NULL;
868 }
869 mutex_unlock(&uuid_mutex);
870
871 while (seed_devices) {
872 fs_devices = seed_devices;
873 seed_devices = fs_devices->seed;
874 __btrfs_close_devices(fs_devices);
875 free_fs_devices(fs_devices);
876 }
877 /*
878 * Wait for rcu kworkers under __btrfs_close_devices
879 * to finish all blkdev_puts so device is really
880 * free when umount is done.
881 */
882 rcu_barrier();
883 return ret;
884 }
885
886 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
887 fmode_t flags, void *holder)
888 {
889 struct request_queue *q;
890 struct block_device *bdev;
891 struct list_head *head = &fs_devices->devices;
892 struct btrfs_device *device;
893 struct btrfs_device *latest_dev = NULL;
894 struct buffer_head *bh;
895 struct btrfs_super_block *disk_super;
896 u64 devid;
897 int seeding = 1;
898 int ret = 0;
899
900 flags |= FMODE_EXCL;
901
902 list_for_each_entry(device, head, dev_list) {
903 if (device->bdev)
904 continue;
905 if (!device->name)
906 continue;
907
908 /* Just open everything we can; ignore failures here */
909 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
910 &bdev, &bh))
911 continue;
912
913 disk_super = (struct btrfs_super_block *)bh->b_data;
914 devid = btrfs_stack_device_id(&disk_super->dev_item);
915 if (devid != device->devid)
916 goto error_brelse;
917
918 if (memcmp(device->uuid, disk_super->dev_item.uuid,
919 BTRFS_UUID_SIZE))
920 goto error_brelse;
921
922 device->generation = btrfs_super_generation(disk_super);
923 if (!latest_dev ||
924 device->generation > latest_dev->generation)
925 latest_dev = device;
926
927 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
928 device->writeable = 0;
929 } else {
930 device->writeable = !bdev_read_only(bdev);
931 seeding = 0;
932 }
933
934 q = bdev_get_queue(bdev);
935 if (blk_queue_discard(q))
936 device->can_discard = 1;
937
938 device->bdev = bdev;
939 device->in_fs_metadata = 0;
940 device->mode = flags;
941
942 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
943 fs_devices->rotating = 1;
944
945 fs_devices->open_devices++;
946 if (device->writeable &&
947 device->devid != BTRFS_DEV_REPLACE_DEVID) {
948 fs_devices->rw_devices++;
949 list_add(&device->dev_alloc_list,
950 &fs_devices->alloc_list);
951 }
952 brelse(bh);
953 continue;
954
955 error_brelse:
956 brelse(bh);
957 blkdev_put(bdev, flags);
958 continue;
959 }
960 if (fs_devices->open_devices == 0) {
961 ret = -EINVAL;
962 goto out;
963 }
964 fs_devices->seeding = seeding;
965 fs_devices->opened = 1;
966 fs_devices->latest_bdev = latest_dev->bdev;
967 fs_devices->total_rw_bytes = 0;
968 out:
969 return ret;
970 }
971
972 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
973 fmode_t flags, void *holder)
974 {
975 int ret;
976
977 mutex_lock(&uuid_mutex);
978 if (fs_devices->opened) {
979 fs_devices->opened++;
980 ret = 0;
981 } else {
982 ret = __btrfs_open_devices(fs_devices, flags, holder);
983 }
984 mutex_unlock(&uuid_mutex);
985 return ret;
986 }
987
988 /*
989 * Look for a btrfs signature on a device. This may be called out of the mount path
990 * and we are not allowed to call set_blocksize during the scan. The superblock
991 * is read via pagecache
992 */
993 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
994 struct btrfs_fs_devices **fs_devices_ret)
995 {
996 struct btrfs_super_block *disk_super;
997 struct block_device *bdev;
998 struct page *page;
999 void *p;
1000 int ret = -EINVAL;
1001 u64 devid;
1002 u64 transid;
1003 u64 total_devices;
1004 u64 bytenr;
1005 pgoff_t index;
1006
1007 /*
1008 * we would like to check all the supers, but that would make
1009 * a btrfs mount succeed after a mkfs from a different FS.
1010 * So, we need to add a special mount option to scan for
1011 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1012 */
1013 bytenr = btrfs_sb_offset(0);
1014 flags |= FMODE_EXCL;
1015 mutex_lock(&uuid_mutex);
1016
1017 bdev = blkdev_get_by_path(path, flags, holder);
1018
1019 if (IS_ERR(bdev)) {
1020 ret = PTR_ERR(bdev);
1021 goto error;
1022 }
1023
1024 /* make sure our super fits in the device */
1025 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1026 goto error_bdev_put;
1027
1028 /* make sure our super fits in the page */
1029 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1030 goto error_bdev_put;
1031
1032 /* make sure our super doesn't straddle pages on disk */
1033 index = bytenr >> PAGE_CACHE_SHIFT;
1034 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1035 goto error_bdev_put;
1036
1037 /* pull in the page with our super */
1038 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1039 index, GFP_NOFS);
1040
1041 if (IS_ERR_OR_NULL(page))
1042 goto error_bdev_put;
1043
1044 p = kmap(page);
1045
1046 /* align our pointer to the offset of the super block */
1047 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1048
1049 if (btrfs_super_bytenr(disk_super) != bytenr ||
1050 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1051 goto error_unmap;
1052
1053 devid = btrfs_stack_device_id(&disk_super->dev_item);
1054 transid = btrfs_super_generation(disk_super);
1055 total_devices = btrfs_super_num_devices(disk_super);
1056
1057 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1058 if (ret > 0) {
1059 if (disk_super->label[0]) {
1060 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1061 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1062 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1063 } else {
1064 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1065 }
1066
1067 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1068 ret = 0;
1069 }
1070 if (!ret && fs_devices_ret)
1071 (*fs_devices_ret)->total_devices = total_devices;
1072
1073 error_unmap:
1074 kunmap(page);
1075 page_cache_release(page);
1076
1077 error_bdev_put:
1078 blkdev_put(bdev, flags);
1079 error:
1080 mutex_unlock(&uuid_mutex);
1081 return ret;
1082 }
1083
1084 /* helper to account the used device space in the range */
1085 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1086 u64 end, u64 *length)
1087 {
1088 struct btrfs_key key;
1089 struct btrfs_root *root = device->dev_root;
1090 struct btrfs_dev_extent *dev_extent;
1091 struct btrfs_path *path;
1092 u64 extent_end;
1093 int ret;
1094 int slot;
1095 struct extent_buffer *l;
1096
1097 *length = 0;
1098
1099 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1100 return 0;
1101
1102 path = btrfs_alloc_path();
1103 if (!path)
1104 return -ENOMEM;
1105 path->reada = 2;
1106
1107 key.objectid = device->devid;
1108 key.offset = start;
1109 key.type = BTRFS_DEV_EXTENT_KEY;
1110
1111 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1112 if (ret < 0)
1113 goto out;
1114 if (ret > 0) {
1115 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1116 if (ret < 0)
1117 goto out;
1118 }
1119
1120 while (1) {
1121 l = path->nodes[0];
1122 slot = path->slots[0];
1123 if (slot >= btrfs_header_nritems(l)) {
1124 ret = btrfs_next_leaf(root, path);
1125 if (ret == 0)
1126 continue;
1127 if (ret < 0)
1128 goto out;
1129
1130 break;
1131 }
1132 btrfs_item_key_to_cpu(l, &key, slot);
1133
1134 if (key.objectid < device->devid)
1135 goto next;
1136
1137 if (key.objectid > device->devid)
1138 break;
1139
1140 if (key.type != BTRFS_DEV_EXTENT_KEY)
1141 goto next;
1142
1143 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1144 extent_end = key.offset + btrfs_dev_extent_length(l,
1145 dev_extent);
1146 if (key.offset <= start && extent_end > end) {
1147 *length = end - start + 1;
1148 break;
1149 } else if (key.offset <= start && extent_end > start)
1150 *length += extent_end - start;
1151 else if (key.offset > start && extent_end <= end)
1152 *length += extent_end - key.offset;
1153 else if (key.offset > start && key.offset <= end) {
1154 *length += end - key.offset + 1;
1155 break;
1156 } else if (key.offset > end)
1157 break;
1158
1159 next:
1160 path->slots[0]++;
1161 }
1162 ret = 0;
1163 out:
1164 btrfs_free_path(path);
1165 return ret;
1166 }
1167
1168 static int contains_pending_extent(struct btrfs_transaction *transaction,
1169 struct btrfs_device *device,
1170 u64 *start, u64 len)
1171 {
1172 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1173 struct extent_map *em;
1174 struct list_head *search_list = &fs_info->pinned_chunks;
1175 int ret = 0;
1176 u64 physical_start = *start;
1177
1178 if (transaction)
1179 search_list = &transaction->pending_chunks;
1180 again:
1181 list_for_each_entry(em, search_list, list) {
1182 struct map_lookup *map;
1183 int i;
1184
1185 map = em->map_lookup;
1186 for (i = 0; i < map->num_stripes; i++) {
1187 u64 end;
1188
1189 if (map->stripes[i].dev != device)
1190 continue;
1191 if (map->stripes[i].physical >= physical_start + len ||
1192 map->stripes[i].physical + em->orig_block_len <=
1193 physical_start)
1194 continue;
1195 /*
1196 * Make sure that while processing the pinned list we do
1197 * not override our *start with a lower value, because
1198 * we can have pinned chunks that fall within this
1199 * device hole and that have lower physical addresses
1200 * than the pending chunks we processed before. If we
1201 * do not take this special care we can end up getting
1202 * 2 pending chunks that start at the same physical
1203 * device offsets because the end offset of a pinned
1204 * chunk can be equal to the start offset of some
1205 * pending chunk.
1206 */
1207 end = map->stripes[i].physical + em->orig_block_len;
1208 if (end > *start) {
1209 *start = end;
1210 ret = 1;
1211 }
1212 }
1213 }
1214 if (search_list != &fs_info->pinned_chunks) {
1215 search_list = &fs_info->pinned_chunks;
1216 goto again;
1217 }
1218
1219 return ret;
1220 }
1221
1222
1223 /*
1224 * find_free_dev_extent_start - find free space in the specified device
1225 * @device: the device which we search the free space in
1226 * @num_bytes: the size of the free space that we need
1227 * @search_start: the position from which to begin the search
1228 * @start: store the start of the free space.
1229 * @len: the size of the free space. that we find, or the size
1230 * of the max free space if we don't find suitable free space
1231 *
1232 * this uses a pretty simple search, the expectation is that it is
1233 * called very infrequently and that a given device has a small number
1234 * of extents
1235 *
1236 * @start is used to store the start of the free space if we find. But if we
1237 * don't find suitable free space, it will be used to store the start position
1238 * of the max free space.
1239 *
1240 * @len is used to store the size of the free space that we find.
1241 * But if we don't find suitable free space, it is used to store the size of
1242 * the max free space.
1243 */
1244 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1245 struct btrfs_device *device, u64 num_bytes,
1246 u64 search_start, u64 *start, u64 *len)
1247 {
1248 struct btrfs_key key;
1249 struct btrfs_root *root = device->dev_root;
1250 struct btrfs_dev_extent *dev_extent;
1251 struct btrfs_path *path;
1252 u64 hole_size;
1253 u64 max_hole_start;
1254 u64 max_hole_size;
1255 u64 extent_end;
1256 u64 search_end = device->total_bytes;
1257 int ret;
1258 int slot;
1259 struct extent_buffer *l;
1260
1261 path = btrfs_alloc_path();
1262 if (!path)
1263 return -ENOMEM;
1264
1265 max_hole_start = search_start;
1266 max_hole_size = 0;
1267
1268 again:
1269 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1270 ret = -ENOSPC;
1271 goto out;
1272 }
1273
1274 path->reada = 2;
1275 path->search_commit_root = 1;
1276 path->skip_locking = 1;
1277
1278 key.objectid = device->devid;
1279 key.offset = search_start;
1280 key.type = BTRFS_DEV_EXTENT_KEY;
1281
1282 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1283 if (ret < 0)
1284 goto out;
1285 if (ret > 0) {
1286 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1287 if (ret < 0)
1288 goto out;
1289 }
1290
1291 while (1) {
1292 l = path->nodes[0];
1293 slot = path->slots[0];
1294 if (slot >= btrfs_header_nritems(l)) {
1295 ret = btrfs_next_leaf(root, path);
1296 if (ret == 0)
1297 continue;
1298 if (ret < 0)
1299 goto out;
1300
1301 break;
1302 }
1303 btrfs_item_key_to_cpu(l, &key, slot);
1304
1305 if (key.objectid < device->devid)
1306 goto next;
1307
1308 if (key.objectid > device->devid)
1309 break;
1310
1311 if (key.type != BTRFS_DEV_EXTENT_KEY)
1312 goto next;
1313
1314 if (key.offset > search_start) {
1315 hole_size = key.offset - search_start;
1316
1317 /*
1318 * Have to check before we set max_hole_start, otherwise
1319 * we could end up sending back this offset anyway.
1320 */
1321 if (contains_pending_extent(transaction, device,
1322 &search_start,
1323 hole_size)) {
1324 if (key.offset >= search_start) {
1325 hole_size = key.offset - search_start;
1326 } else {
1327 WARN_ON_ONCE(1);
1328 hole_size = 0;
1329 }
1330 }
1331
1332 if (hole_size > max_hole_size) {
1333 max_hole_start = search_start;
1334 max_hole_size = hole_size;
1335 }
1336
1337 /*
1338 * If this free space is greater than which we need,
1339 * it must be the max free space that we have found
1340 * until now, so max_hole_start must point to the start
1341 * of this free space and the length of this free space
1342 * is stored in max_hole_size. Thus, we return
1343 * max_hole_start and max_hole_size and go back to the
1344 * caller.
1345 */
1346 if (hole_size >= num_bytes) {
1347 ret = 0;
1348 goto out;
1349 }
1350 }
1351
1352 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1353 extent_end = key.offset + btrfs_dev_extent_length(l,
1354 dev_extent);
1355 if (extent_end > search_start)
1356 search_start = extent_end;
1357 next:
1358 path->slots[0]++;
1359 cond_resched();
1360 }
1361
1362 /*
1363 * At this point, search_start should be the end of
1364 * allocated dev extents, and when shrinking the device,
1365 * search_end may be smaller than search_start.
1366 */
1367 if (search_end > search_start) {
1368 hole_size = search_end - search_start;
1369
1370 if (contains_pending_extent(transaction, device, &search_start,
1371 hole_size)) {
1372 btrfs_release_path(path);
1373 goto again;
1374 }
1375
1376 if (hole_size > max_hole_size) {
1377 max_hole_start = search_start;
1378 max_hole_size = hole_size;
1379 }
1380 }
1381
1382 /* See above. */
1383 if (max_hole_size < num_bytes)
1384 ret = -ENOSPC;
1385 else
1386 ret = 0;
1387
1388 out:
1389 btrfs_free_path(path);
1390 *start = max_hole_start;
1391 if (len)
1392 *len = max_hole_size;
1393 return ret;
1394 }
1395
1396 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1397 struct btrfs_device *device, u64 num_bytes,
1398 u64 *start, u64 *len)
1399 {
1400 struct btrfs_root *root = device->dev_root;
1401 u64 search_start;
1402
1403 /* FIXME use last free of some kind */
1404
1405 /*
1406 * we don't want to overwrite the superblock on the drive,
1407 * so we make sure to start at an offset of at least 1MB
1408 */
1409 search_start = max_t(u64, root->fs_info->alloc_start, SZ_1M);
1410 return find_free_dev_extent_start(trans->transaction, device,
1411 num_bytes, search_start, start, len);
1412 }
1413
1414 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1415 struct btrfs_device *device,
1416 u64 start, u64 *dev_extent_len)
1417 {
1418 int ret;
1419 struct btrfs_path *path;
1420 struct btrfs_root *root = device->dev_root;
1421 struct btrfs_key key;
1422 struct btrfs_key found_key;
1423 struct extent_buffer *leaf = NULL;
1424 struct btrfs_dev_extent *extent = NULL;
1425
1426 path = btrfs_alloc_path();
1427 if (!path)
1428 return -ENOMEM;
1429
1430 key.objectid = device->devid;
1431 key.offset = start;
1432 key.type = BTRFS_DEV_EXTENT_KEY;
1433 again:
1434 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1435 if (ret > 0) {
1436 ret = btrfs_previous_item(root, path, key.objectid,
1437 BTRFS_DEV_EXTENT_KEY);
1438 if (ret)
1439 goto out;
1440 leaf = path->nodes[0];
1441 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1442 extent = btrfs_item_ptr(leaf, path->slots[0],
1443 struct btrfs_dev_extent);
1444 BUG_ON(found_key.offset > start || found_key.offset +
1445 btrfs_dev_extent_length(leaf, extent) < start);
1446 key = found_key;
1447 btrfs_release_path(path);
1448 goto again;
1449 } else if (ret == 0) {
1450 leaf = path->nodes[0];
1451 extent = btrfs_item_ptr(leaf, path->slots[0],
1452 struct btrfs_dev_extent);
1453 } else {
1454 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1455 goto out;
1456 }
1457
1458 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1459
1460 ret = btrfs_del_item(trans, root, path);
1461 if (ret) {
1462 btrfs_std_error(root->fs_info, ret,
1463 "Failed to remove dev extent item");
1464 } else {
1465 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1466 }
1467 out:
1468 btrfs_free_path(path);
1469 return ret;
1470 }
1471
1472 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1473 struct btrfs_device *device,
1474 u64 chunk_tree, u64 chunk_objectid,
1475 u64 chunk_offset, u64 start, u64 num_bytes)
1476 {
1477 int ret;
1478 struct btrfs_path *path;
1479 struct btrfs_root *root = device->dev_root;
1480 struct btrfs_dev_extent *extent;
1481 struct extent_buffer *leaf;
1482 struct btrfs_key key;
1483
1484 WARN_ON(!device->in_fs_metadata);
1485 WARN_ON(device->is_tgtdev_for_dev_replace);
1486 path = btrfs_alloc_path();
1487 if (!path)
1488 return -ENOMEM;
1489
1490 key.objectid = device->devid;
1491 key.offset = start;
1492 key.type = BTRFS_DEV_EXTENT_KEY;
1493 ret = btrfs_insert_empty_item(trans, root, path, &key,
1494 sizeof(*extent));
1495 if (ret)
1496 goto out;
1497
1498 leaf = path->nodes[0];
1499 extent = btrfs_item_ptr(leaf, path->slots[0],
1500 struct btrfs_dev_extent);
1501 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1502 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1503 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1504
1505 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1506 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1507
1508 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1509 btrfs_mark_buffer_dirty(leaf);
1510 out:
1511 btrfs_free_path(path);
1512 return ret;
1513 }
1514
1515 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1516 {
1517 struct extent_map_tree *em_tree;
1518 struct extent_map *em;
1519 struct rb_node *n;
1520 u64 ret = 0;
1521
1522 em_tree = &fs_info->mapping_tree.map_tree;
1523 read_lock(&em_tree->lock);
1524 n = rb_last(&em_tree->map);
1525 if (n) {
1526 em = rb_entry(n, struct extent_map, rb_node);
1527 ret = em->start + em->len;
1528 }
1529 read_unlock(&em_tree->lock);
1530
1531 return ret;
1532 }
1533
1534 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1535 u64 *devid_ret)
1536 {
1537 int ret;
1538 struct btrfs_key key;
1539 struct btrfs_key found_key;
1540 struct btrfs_path *path;
1541
1542 path = btrfs_alloc_path();
1543 if (!path)
1544 return -ENOMEM;
1545
1546 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1547 key.type = BTRFS_DEV_ITEM_KEY;
1548 key.offset = (u64)-1;
1549
1550 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1551 if (ret < 0)
1552 goto error;
1553
1554 BUG_ON(ret == 0); /* Corruption */
1555
1556 ret = btrfs_previous_item(fs_info->chunk_root, path,
1557 BTRFS_DEV_ITEMS_OBJECTID,
1558 BTRFS_DEV_ITEM_KEY);
1559 if (ret) {
1560 *devid_ret = 1;
1561 } else {
1562 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1563 path->slots[0]);
1564 *devid_ret = found_key.offset + 1;
1565 }
1566 ret = 0;
1567 error:
1568 btrfs_free_path(path);
1569 return ret;
1570 }
1571
1572 /*
1573 * the device information is stored in the chunk root
1574 * the btrfs_device struct should be fully filled in
1575 */
1576 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1577 struct btrfs_root *root,
1578 struct btrfs_device *device)
1579 {
1580 int ret;
1581 struct btrfs_path *path;
1582 struct btrfs_dev_item *dev_item;
1583 struct extent_buffer *leaf;
1584 struct btrfs_key key;
1585 unsigned long ptr;
1586
1587 root = root->fs_info->chunk_root;
1588
1589 path = btrfs_alloc_path();
1590 if (!path)
1591 return -ENOMEM;
1592
1593 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594 key.type = BTRFS_DEV_ITEM_KEY;
1595 key.offset = device->devid;
1596
1597 ret = btrfs_insert_empty_item(trans, root, path, &key,
1598 sizeof(*dev_item));
1599 if (ret)
1600 goto out;
1601
1602 leaf = path->nodes[0];
1603 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1604
1605 btrfs_set_device_id(leaf, dev_item, device->devid);
1606 btrfs_set_device_generation(leaf, dev_item, 0);
1607 btrfs_set_device_type(leaf, dev_item, device->type);
1608 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1609 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1610 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1611 btrfs_set_device_total_bytes(leaf, dev_item,
1612 btrfs_device_get_disk_total_bytes(device));
1613 btrfs_set_device_bytes_used(leaf, dev_item,
1614 btrfs_device_get_bytes_used(device));
1615 btrfs_set_device_group(leaf, dev_item, 0);
1616 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1617 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1618 btrfs_set_device_start_offset(leaf, dev_item, 0);
1619
1620 ptr = btrfs_device_uuid(dev_item);
1621 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1622 ptr = btrfs_device_fsid(dev_item);
1623 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1624 btrfs_mark_buffer_dirty(leaf);
1625
1626 ret = 0;
1627 out:
1628 btrfs_free_path(path);
1629 return ret;
1630 }
1631
1632 /*
1633 * Function to update ctime/mtime for a given device path.
1634 * Mainly used for ctime/mtime based probe like libblkid.
1635 */
1636 static void update_dev_time(char *path_name)
1637 {
1638 struct file *filp;
1639
1640 filp = filp_open(path_name, O_RDWR, 0);
1641 if (IS_ERR(filp))
1642 return;
1643 file_update_time(filp);
1644 filp_close(filp, NULL);
1645 }
1646
1647 static int btrfs_rm_dev_item(struct btrfs_root *root,
1648 struct btrfs_device *device)
1649 {
1650 int ret;
1651 struct btrfs_path *path;
1652 struct btrfs_key key;
1653 struct btrfs_trans_handle *trans;
1654
1655 root = root->fs_info->chunk_root;
1656
1657 path = btrfs_alloc_path();
1658 if (!path)
1659 return -ENOMEM;
1660
1661 trans = btrfs_start_transaction(root, 0);
1662 if (IS_ERR(trans)) {
1663 btrfs_free_path(path);
1664 return PTR_ERR(trans);
1665 }
1666 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1667 key.type = BTRFS_DEV_ITEM_KEY;
1668 key.offset = device->devid;
1669
1670 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1671 if (ret < 0)
1672 goto out;
1673
1674 if (ret > 0) {
1675 ret = -ENOENT;
1676 goto out;
1677 }
1678
1679 ret = btrfs_del_item(trans, root, path);
1680 if (ret)
1681 goto out;
1682 out:
1683 btrfs_free_path(path);
1684 btrfs_commit_transaction(trans, root);
1685 return ret;
1686 }
1687
1688 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1689 {
1690 struct btrfs_device *device;
1691 struct btrfs_device *next_device;
1692 struct block_device *bdev;
1693 struct buffer_head *bh = NULL;
1694 struct btrfs_super_block *disk_super;
1695 struct btrfs_fs_devices *cur_devices;
1696 u64 all_avail;
1697 u64 devid;
1698 u64 num_devices;
1699 u8 *dev_uuid;
1700 unsigned seq;
1701 int ret = 0;
1702 bool clear_super = false;
1703
1704 mutex_lock(&uuid_mutex);
1705
1706 do {
1707 seq = read_seqbegin(&root->fs_info->profiles_lock);
1708
1709 all_avail = root->fs_info->avail_data_alloc_bits |
1710 root->fs_info->avail_system_alloc_bits |
1711 root->fs_info->avail_metadata_alloc_bits;
1712 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1713
1714 num_devices = root->fs_info->fs_devices->num_devices;
1715 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1716 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1717 WARN_ON(num_devices < 1);
1718 num_devices--;
1719 }
1720 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1721
1722 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1723 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1724 goto out;
1725 }
1726
1727 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1728 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1729 goto out;
1730 }
1731
1732 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1733 root->fs_info->fs_devices->rw_devices <= 2) {
1734 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1735 goto out;
1736 }
1737 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1738 root->fs_info->fs_devices->rw_devices <= 3) {
1739 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1740 goto out;
1741 }
1742
1743 if (strcmp(device_path, "missing") == 0) {
1744 struct list_head *devices;
1745 struct btrfs_device *tmp;
1746
1747 device = NULL;
1748 devices = &root->fs_info->fs_devices->devices;
1749 /*
1750 * It is safe to read the devices since the volume_mutex
1751 * is held.
1752 */
1753 list_for_each_entry(tmp, devices, dev_list) {
1754 if (tmp->in_fs_metadata &&
1755 !tmp->is_tgtdev_for_dev_replace &&
1756 !tmp->bdev) {
1757 device = tmp;
1758 break;
1759 }
1760 }
1761 bdev = NULL;
1762 bh = NULL;
1763 disk_super = NULL;
1764 if (!device) {
1765 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1766 goto out;
1767 }
1768 } else {
1769 ret = btrfs_get_bdev_and_sb(device_path,
1770 FMODE_WRITE | FMODE_EXCL,
1771 root->fs_info->bdev_holder, 0,
1772 &bdev, &bh);
1773 if (ret)
1774 goto out;
1775 disk_super = (struct btrfs_super_block *)bh->b_data;
1776 devid = btrfs_stack_device_id(&disk_super->dev_item);
1777 dev_uuid = disk_super->dev_item.uuid;
1778 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1779 disk_super->fsid);
1780 if (!device) {
1781 ret = -ENOENT;
1782 goto error_brelse;
1783 }
1784 }
1785
1786 if (device->is_tgtdev_for_dev_replace) {
1787 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1788 goto error_brelse;
1789 }
1790
1791 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1792 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1793 goto error_brelse;
1794 }
1795
1796 if (device->writeable) {
1797 lock_chunks(root);
1798 list_del_init(&device->dev_alloc_list);
1799 device->fs_devices->rw_devices--;
1800 unlock_chunks(root);
1801 clear_super = true;
1802 }
1803
1804 mutex_unlock(&uuid_mutex);
1805 ret = btrfs_shrink_device(device, 0);
1806 mutex_lock(&uuid_mutex);
1807 if (ret)
1808 goto error_undo;
1809
1810 /*
1811 * TODO: the superblock still includes this device in its num_devices
1812 * counter although write_all_supers() is not locked out. This
1813 * could give a filesystem state which requires a degraded mount.
1814 */
1815 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1816 if (ret)
1817 goto error_undo;
1818
1819 device->in_fs_metadata = 0;
1820 btrfs_scrub_cancel_dev(root->fs_info, device);
1821
1822 /*
1823 * the device list mutex makes sure that we don't change
1824 * the device list while someone else is writing out all
1825 * the device supers. Whoever is writing all supers, should
1826 * lock the device list mutex before getting the number of
1827 * devices in the super block (super_copy). Conversely,
1828 * whoever updates the number of devices in the super block
1829 * (super_copy) should hold the device list mutex.
1830 */
1831
1832 cur_devices = device->fs_devices;
1833 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1834 list_del_rcu(&device->dev_list);
1835
1836 device->fs_devices->num_devices--;
1837 device->fs_devices->total_devices--;
1838
1839 if (device->missing)
1840 device->fs_devices->missing_devices--;
1841
1842 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1843 struct btrfs_device, dev_list);
1844 if (device->bdev == root->fs_info->sb->s_bdev)
1845 root->fs_info->sb->s_bdev = next_device->bdev;
1846 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1847 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1848
1849 if (device->bdev) {
1850 device->fs_devices->open_devices--;
1851 /* remove sysfs entry */
1852 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1853 }
1854
1855 call_rcu(&device->rcu, free_device);
1856
1857 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1858 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1859 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1860
1861 if (cur_devices->open_devices == 0) {
1862 struct btrfs_fs_devices *fs_devices;
1863 fs_devices = root->fs_info->fs_devices;
1864 while (fs_devices) {
1865 if (fs_devices->seed == cur_devices) {
1866 fs_devices->seed = cur_devices->seed;
1867 break;
1868 }
1869 fs_devices = fs_devices->seed;
1870 }
1871 cur_devices->seed = NULL;
1872 __btrfs_close_devices(cur_devices);
1873 free_fs_devices(cur_devices);
1874 }
1875
1876 root->fs_info->num_tolerated_disk_barrier_failures =
1877 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1878
1879 /*
1880 * at this point, the device is zero sized. We want to
1881 * remove it from the devices list and zero out the old super
1882 */
1883 if (clear_super && disk_super) {
1884 u64 bytenr;
1885 int i;
1886
1887 /* make sure this device isn't detected as part of
1888 * the FS anymore
1889 */
1890 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1891 set_buffer_dirty(bh);
1892 sync_dirty_buffer(bh);
1893
1894 /* clear the mirror copies of super block on the disk
1895 * being removed, 0th copy is been taken care above and
1896 * the below would take of the rest
1897 */
1898 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1899 bytenr = btrfs_sb_offset(i);
1900 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1901 i_size_read(bdev->bd_inode))
1902 break;
1903
1904 brelse(bh);
1905 bh = __bread(bdev, bytenr / 4096,
1906 BTRFS_SUPER_INFO_SIZE);
1907 if (!bh)
1908 continue;
1909
1910 disk_super = (struct btrfs_super_block *)bh->b_data;
1911
1912 if (btrfs_super_bytenr(disk_super) != bytenr ||
1913 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1914 continue;
1915 }
1916 memset(&disk_super->magic, 0,
1917 sizeof(disk_super->magic));
1918 set_buffer_dirty(bh);
1919 sync_dirty_buffer(bh);
1920 }
1921 }
1922
1923 ret = 0;
1924
1925 if (bdev) {
1926 /* Notify udev that device has changed */
1927 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1928
1929 /* Update ctime/mtime for device path for libblkid */
1930 update_dev_time(device_path);
1931 }
1932
1933 error_brelse:
1934 brelse(bh);
1935 if (bdev)
1936 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1937 out:
1938 mutex_unlock(&uuid_mutex);
1939 return ret;
1940 error_undo:
1941 if (device->writeable) {
1942 lock_chunks(root);
1943 list_add(&device->dev_alloc_list,
1944 &root->fs_info->fs_devices->alloc_list);
1945 device->fs_devices->rw_devices++;
1946 unlock_chunks(root);
1947 }
1948 goto error_brelse;
1949 }
1950
1951 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1952 struct btrfs_device *srcdev)
1953 {
1954 struct btrfs_fs_devices *fs_devices;
1955
1956 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1957
1958 /*
1959 * in case of fs with no seed, srcdev->fs_devices will point
1960 * to fs_devices of fs_info. However when the dev being replaced is
1961 * a seed dev it will point to the seed's local fs_devices. In short
1962 * srcdev will have its correct fs_devices in both the cases.
1963 */
1964 fs_devices = srcdev->fs_devices;
1965
1966 list_del_rcu(&srcdev->dev_list);
1967 list_del_rcu(&srcdev->dev_alloc_list);
1968 fs_devices->num_devices--;
1969 if (srcdev->missing)
1970 fs_devices->missing_devices--;
1971
1972 if (srcdev->writeable) {
1973 fs_devices->rw_devices--;
1974 /* zero out the old super if it is writable */
1975 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1976 }
1977
1978 if (srcdev->bdev)
1979 fs_devices->open_devices--;
1980 }
1981
1982 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1983 struct btrfs_device *srcdev)
1984 {
1985 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1986
1987 call_rcu(&srcdev->rcu, free_device);
1988
1989 /*
1990 * unless fs_devices is seed fs, num_devices shouldn't go
1991 * zero
1992 */
1993 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1994
1995 /* if this is no devs we rather delete the fs_devices */
1996 if (!fs_devices->num_devices) {
1997 struct btrfs_fs_devices *tmp_fs_devices;
1998
1999 tmp_fs_devices = fs_info->fs_devices;
2000 while (tmp_fs_devices) {
2001 if (tmp_fs_devices->seed == fs_devices) {
2002 tmp_fs_devices->seed = fs_devices->seed;
2003 break;
2004 }
2005 tmp_fs_devices = tmp_fs_devices->seed;
2006 }
2007 fs_devices->seed = NULL;
2008 __btrfs_close_devices(fs_devices);
2009 free_fs_devices(fs_devices);
2010 }
2011 }
2012
2013 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2014 struct btrfs_device *tgtdev)
2015 {
2016 struct btrfs_device *next_device;
2017
2018 mutex_lock(&uuid_mutex);
2019 WARN_ON(!tgtdev);
2020 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2021
2022 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2023
2024 if (tgtdev->bdev) {
2025 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2026 fs_info->fs_devices->open_devices--;
2027 }
2028 fs_info->fs_devices->num_devices--;
2029
2030 next_device = list_entry(fs_info->fs_devices->devices.next,
2031 struct btrfs_device, dev_list);
2032 if (tgtdev->bdev == fs_info->sb->s_bdev)
2033 fs_info->sb->s_bdev = next_device->bdev;
2034 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2035 fs_info->fs_devices->latest_bdev = next_device->bdev;
2036 list_del_rcu(&tgtdev->dev_list);
2037
2038 call_rcu(&tgtdev->rcu, free_device);
2039
2040 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2041 mutex_unlock(&uuid_mutex);
2042 }
2043
2044 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2045 struct btrfs_device **device)
2046 {
2047 int ret = 0;
2048 struct btrfs_super_block *disk_super;
2049 u64 devid;
2050 u8 *dev_uuid;
2051 struct block_device *bdev;
2052 struct buffer_head *bh;
2053
2054 *device = NULL;
2055 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2056 root->fs_info->bdev_holder, 0, &bdev, &bh);
2057 if (ret)
2058 return ret;
2059 disk_super = (struct btrfs_super_block *)bh->b_data;
2060 devid = btrfs_stack_device_id(&disk_super->dev_item);
2061 dev_uuid = disk_super->dev_item.uuid;
2062 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2063 disk_super->fsid);
2064 brelse(bh);
2065 if (!*device)
2066 ret = -ENOENT;
2067 blkdev_put(bdev, FMODE_READ);
2068 return ret;
2069 }
2070
2071 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2072 char *device_path,
2073 struct btrfs_device **device)
2074 {
2075 *device = NULL;
2076 if (strcmp(device_path, "missing") == 0) {
2077 struct list_head *devices;
2078 struct btrfs_device *tmp;
2079
2080 devices = &root->fs_info->fs_devices->devices;
2081 /*
2082 * It is safe to read the devices since the volume_mutex
2083 * is held by the caller.
2084 */
2085 list_for_each_entry(tmp, devices, dev_list) {
2086 if (tmp->in_fs_metadata && !tmp->bdev) {
2087 *device = tmp;
2088 break;
2089 }
2090 }
2091
2092 if (!*device)
2093 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2094
2095 return 0;
2096 } else {
2097 return btrfs_find_device_by_path(root, device_path, device);
2098 }
2099 }
2100
2101 /*
2102 * does all the dirty work required for changing file system's UUID.
2103 */
2104 static int btrfs_prepare_sprout(struct btrfs_root *root)
2105 {
2106 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2107 struct btrfs_fs_devices *old_devices;
2108 struct btrfs_fs_devices *seed_devices;
2109 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2110 struct btrfs_device *device;
2111 u64 super_flags;
2112
2113 BUG_ON(!mutex_is_locked(&uuid_mutex));
2114 if (!fs_devices->seeding)
2115 return -EINVAL;
2116
2117 seed_devices = __alloc_fs_devices();
2118 if (IS_ERR(seed_devices))
2119 return PTR_ERR(seed_devices);
2120
2121 old_devices = clone_fs_devices(fs_devices);
2122 if (IS_ERR(old_devices)) {
2123 kfree(seed_devices);
2124 return PTR_ERR(old_devices);
2125 }
2126
2127 list_add(&old_devices->list, &fs_uuids);
2128
2129 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2130 seed_devices->opened = 1;
2131 INIT_LIST_HEAD(&seed_devices->devices);
2132 INIT_LIST_HEAD(&seed_devices->alloc_list);
2133 mutex_init(&seed_devices->device_list_mutex);
2134
2135 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2136 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2137 synchronize_rcu);
2138 list_for_each_entry(device, &seed_devices->devices, dev_list)
2139 device->fs_devices = seed_devices;
2140
2141 lock_chunks(root);
2142 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2143 unlock_chunks(root);
2144
2145 fs_devices->seeding = 0;
2146 fs_devices->num_devices = 0;
2147 fs_devices->open_devices = 0;
2148 fs_devices->missing_devices = 0;
2149 fs_devices->rotating = 0;
2150 fs_devices->seed = seed_devices;
2151
2152 generate_random_uuid(fs_devices->fsid);
2153 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2154 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2155 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2156
2157 super_flags = btrfs_super_flags(disk_super) &
2158 ~BTRFS_SUPER_FLAG_SEEDING;
2159 btrfs_set_super_flags(disk_super, super_flags);
2160
2161 return 0;
2162 }
2163
2164 /*
2165 * strore the expected generation for seed devices in device items.
2166 */
2167 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root)
2169 {
2170 struct btrfs_path *path;
2171 struct extent_buffer *leaf;
2172 struct btrfs_dev_item *dev_item;
2173 struct btrfs_device *device;
2174 struct btrfs_key key;
2175 u8 fs_uuid[BTRFS_UUID_SIZE];
2176 u8 dev_uuid[BTRFS_UUID_SIZE];
2177 u64 devid;
2178 int ret;
2179
2180 path = btrfs_alloc_path();
2181 if (!path)
2182 return -ENOMEM;
2183
2184 root = root->fs_info->chunk_root;
2185 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2186 key.offset = 0;
2187 key.type = BTRFS_DEV_ITEM_KEY;
2188
2189 while (1) {
2190 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2191 if (ret < 0)
2192 goto error;
2193
2194 leaf = path->nodes[0];
2195 next_slot:
2196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2197 ret = btrfs_next_leaf(root, path);
2198 if (ret > 0)
2199 break;
2200 if (ret < 0)
2201 goto error;
2202 leaf = path->nodes[0];
2203 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2204 btrfs_release_path(path);
2205 continue;
2206 }
2207
2208 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2209 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2210 key.type != BTRFS_DEV_ITEM_KEY)
2211 break;
2212
2213 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2214 struct btrfs_dev_item);
2215 devid = btrfs_device_id(leaf, dev_item);
2216 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2217 BTRFS_UUID_SIZE);
2218 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2219 BTRFS_UUID_SIZE);
2220 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2221 fs_uuid);
2222 BUG_ON(!device); /* Logic error */
2223
2224 if (device->fs_devices->seeding) {
2225 btrfs_set_device_generation(leaf, dev_item,
2226 device->generation);
2227 btrfs_mark_buffer_dirty(leaf);
2228 }
2229
2230 path->slots[0]++;
2231 goto next_slot;
2232 }
2233 ret = 0;
2234 error:
2235 btrfs_free_path(path);
2236 return ret;
2237 }
2238
2239 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2240 {
2241 struct request_queue *q;
2242 struct btrfs_trans_handle *trans;
2243 struct btrfs_device *device;
2244 struct block_device *bdev;
2245 struct list_head *devices;
2246 struct super_block *sb = root->fs_info->sb;
2247 struct rcu_string *name;
2248 u64 tmp;
2249 int seeding_dev = 0;
2250 int ret = 0;
2251
2252 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2253 return -EROFS;
2254
2255 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2256 root->fs_info->bdev_holder);
2257 if (IS_ERR(bdev))
2258 return PTR_ERR(bdev);
2259
2260 if (root->fs_info->fs_devices->seeding) {
2261 seeding_dev = 1;
2262 down_write(&sb->s_umount);
2263 mutex_lock(&uuid_mutex);
2264 }
2265
2266 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2267
2268 devices = &root->fs_info->fs_devices->devices;
2269
2270 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2271 list_for_each_entry(device, devices, dev_list) {
2272 if (device->bdev == bdev) {
2273 ret = -EEXIST;
2274 mutex_unlock(
2275 &root->fs_info->fs_devices->device_list_mutex);
2276 goto error;
2277 }
2278 }
2279 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2280
2281 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2282 if (IS_ERR(device)) {
2283 /* we can safely leave the fs_devices entry around */
2284 ret = PTR_ERR(device);
2285 goto error;
2286 }
2287
2288 name = rcu_string_strdup(device_path, GFP_NOFS);
2289 if (!name) {
2290 kfree(device);
2291 ret = -ENOMEM;
2292 goto error;
2293 }
2294 rcu_assign_pointer(device->name, name);
2295
2296 trans = btrfs_start_transaction(root, 0);
2297 if (IS_ERR(trans)) {
2298 rcu_string_free(device->name);
2299 kfree(device);
2300 ret = PTR_ERR(trans);
2301 goto error;
2302 }
2303
2304 q = bdev_get_queue(bdev);
2305 if (blk_queue_discard(q))
2306 device->can_discard = 1;
2307 device->writeable = 1;
2308 device->generation = trans->transid;
2309 device->io_width = root->sectorsize;
2310 device->io_align = root->sectorsize;
2311 device->sector_size = root->sectorsize;
2312 device->total_bytes = i_size_read(bdev->bd_inode);
2313 device->disk_total_bytes = device->total_bytes;
2314 device->commit_total_bytes = device->total_bytes;
2315 device->dev_root = root->fs_info->dev_root;
2316 device->bdev = bdev;
2317 device->in_fs_metadata = 1;
2318 device->is_tgtdev_for_dev_replace = 0;
2319 device->mode = FMODE_EXCL;
2320 device->dev_stats_valid = 1;
2321 set_blocksize(device->bdev, 4096);
2322
2323 if (seeding_dev) {
2324 sb->s_flags &= ~MS_RDONLY;
2325 ret = btrfs_prepare_sprout(root);
2326 BUG_ON(ret); /* -ENOMEM */
2327 }
2328
2329 device->fs_devices = root->fs_info->fs_devices;
2330
2331 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2332 lock_chunks(root);
2333 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2334 list_add(&device->dev_alloc_list,
2335 &root->fs_info->fs_devices->alloc_list);
2336 root->fs_info->fs_devices->num_devices++;
2337 root->fs_info->fs_devices->open_devices++;
2338 root->fs_info->fs_devices->rw_devices++;
2339 root->fs_info->fs_devices->total_devices++;
2340 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2341
2342 spin_lock(&root->fs_info->free_chunk_lock);
2343 root->fs_info->free_chunk_space += device->total_bytes;
2344 spin_unlock(&root->fs_info->free_chunk_lock);
2345
2346 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2347 root->fs_info->fs_devices->rotating = 1;
2348
2349 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2350 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2351 tmp + device->total_bytes);
2352
2353 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2354 btrfs_set_super_num_devices(root->fs_info->super_copy,
2355 tmp + 1);
2356
2357 /* add sysfs device entry */
2358 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2359
2360 /*
2361 * we've got more storage, clear any full flags on the space
2362 * infos
2363 */
2364 btrfs_clear_space_info_full(root->fs_info);
2365
2366 unlock_chunks(root);
2367 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2368
2369 if (seeding_dev) {
2370 lock_chunks(root);
2371 ret = init_first_rw_device(trans, root, device);
2372 unlock_chunks(root);
2373 if (ret) {
2374 btrfs_abort_transaction(trans, root, ret);
2375 goto error_trans;
2376 }
2377 }
2378
2379 ret = btrfs_add_device(trans, root, device);
2380 if (ret) {
2381 btrfs_abort_transaction(trans, root, ret);
2382 goto error_trans;
2383 }
2384
2385 if (seeding_dev) {
2386 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2387
2388 ret = btrfs_finish_sprout(trans, root);
2389 if (ret) {
2390 btrfs_abort_transaction(trans, root, ret);
2391 goto error_trans;
2392 }
2393
2394 /* Sprouting would change fsid of the mounted root,
2395 * so rename the fsid on the sysfs
2396 */
2397 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2398 root->fs_info->fsid);
2399 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2400 fsid_buf))
2401 btrfs_warn(root->fs_info,
2402 "sysfs: failed to create fsid for sprout");
2403 }
2404
2405 root->fs_info->num_tolerated_disk_barrier_failures =
2406 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2407 ret = btrfs_commit_transaction(trans, root);
2408
2409 if (seeding_dev) {
2410 mutex_unlock(&uuid_mutex);
2411 up_write(&sb->s_umount);
2412
2413 if (ret) /* transaction commit */
2414 return ret;
2415
2416 ret = btrfs_relocate_sys_chunks(root);
2417 if (ret < 0)
2418 btrfs_std_error(root->fs_info, ret,
2419 "Failed to relocate sys chunks after "
2420 "device initialization. This can be fixed "
2421 "using the \"btrfs balance\" command.");
2422 trans = btrfs_attach_transaction(root);
2423 if (IS_ERR(trans)) {
2424 if (PTR_ERR(trans) == -ENOENT)
2425 return 0;
2426 return PTR_ERR(trans);
2427 }
2428 ret = btrfs_commit_transaction(trans, root);
2429 }
2430
2431 /* Update ctime/mtime for libblkid */
2432 update_dev_time(device_path);
2433 return ret;
2434
2435 error_trans:
2436 btrfs_end_transaction(trans, root);
2437 rcu_string_free(device->name);
2438 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2439 kfree(device);
2440 error:
2441 blkdev_put(bdev, FMODE_EXCL);
2442 if (seeding_dev) {
2443 mutex_unlock(&uuid_mutex);
2444 up_write(&sb->s_umount);
2445 }
2446 return ret;
2447 }
2448
2449 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2450 struct btrfs_device *srcdev,
2451 struct btrfs_device **device_out)
2452 {
2453 struct request_queue *q;
2454 struct btrfs_device *device;
2455 struct block_device *bdev;
2456 struct btrfs_fs_info *fs_info = root->fs_info;
2457 struct list_head *devices;
2458 struct rcu_string *name;
2459 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2460 int ret = 0;
2461
2462 *device_out = NULL;
2463 if (fs_info->fs_devices->seeding) {
2464 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2465 return -EINVAL;
2466 }
2467
2468 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2469 fs_info->bdev_holder);
2470 if (IS_ERR(bdev)) {
2471 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2472 return PTR_ERR(bdev);
2473 }
2474
2475 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2476
2477 devices = &fs_info->fs_devices->devices;
2478 list_for_each_entry(device, devices, dev_list) {
2479 if (device->bdev == bdev) {
2480 btrfs_err(fs_info, "target device is in the filesystem!");
2481 ret = -EEXIST;
2482 goto error;
2483 }
2484 }
2485
2486
2487 if (i_size_read(bdev->bd_inode) <
2488 btrfs_device_get_total_bytes(srcdev)) {
2489 btrfs_err(fs_info, "target device is smaller than source device!");
2490 ret = -EINVAL;
2491 goto error;
2492 }
2493
2494
2495 device = btrfs_alloc_device(NULL, &devid, NULL);
2496 if (IS_ERR(device)) {
2497 ret = PTR_ERR(device);
2498 goto error;
2499 }
2500
2501 name = rcu_string_strdup(device_path, GFP_NOFS);
2502 if (!name) {
2503 kfree(device);
2504 ret = -ENOMEM;
2505 goto error;
2506 }
2507 rcu_assign_pointer(device->name, name);
2508
2509 q = bdev_get_queue(bdev);
2510 if (blk_queue_discard(q))
2511 device->can_discard = 1;
2512 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2513 device->writeable = 1;
2514 device->generation = 0;
2515 device->io_width = root->sectorsize;
2516 device->io_align = root->sectorsize;
2517 device->sector_size = root->sectorsize;
2518 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2519 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2520 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2521 ASSERT(list_empty(&srcdev->resized_list));
2522 device->commit_total_bytes = srcdev->commit_total_bytes;
2523 device->commit_bytes_used = device->bytes_used;
2524 device->dev_root = fs_info->dev_root;
2525 device->bdev = bdev;
2526 device->in_fs_metadata = 1;
2527 device->is_tgtdev_for_dev_replace = 1;
2528 device->mode = FMODE_EXCL;
2529 device->dev_stats_valid = 1;
2530 set_blocksize(device->bdev, 4096);
2531 device->fs_devices = fs_info->fs_devices;
2532 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2533 fs_info->fs_devices->num_devices++;
2534 fs_info->fs_devices->open_devices++;
2535 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2536
2537 *device_out = device;
2538 return ret;
2539
2540 error:
2541 blkdev_put(bdev, FMODE_EXCL);
2542 return ret;
2543 }
2544
2545 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2546 struct btrfs_device *tgtdev)
2547 {
2548 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2549 tgtdev->io_width = fs_info->dev_root->sectorsize;
2550 tgtdev->io_align = fs_info->dev_root->sectorsize;
2551 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2552 tgtdev->dev_root = fs_info->dev_root;
2553 tgtdev->in_fs_metadata = 1;
2554 }
2555
2556 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2557 struct btrfs_device *device)
2558 {
2559 int ret;
2560 struct btrfs_path *path;
2561 struct btrfs_root *root;
2562 struct btrfs_dev_item *dev_item;
2563 struct extent_buffer *leaf;
2564 struct btrfs_key key;
2565
2566 root = device->dev_root->fs_info->chunk_root;
2567
2568 path = btrfs_alloc_path();
2569 if (!path)
2570 return -ENOMEM;
2571
2572 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2573 key.type = BTRFS_DEV_ITEM_KEY;
2574 key.offset = device->devid;
2575
2576 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2577 if (ret < 0)
2578 goto out;
2579
2580 if (ret > 0) {
2581 ret = -ENOENT;
2582 goto out;
2583 }
2584
2585 leaf = path->nodes[0];
2586 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2587
2588 btrfs_set_device_id(leaf, dev_item, device->devid);
2589 btrfs_set_device_type(leaf, dev_item, device->type);
2590 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2591 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2592 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2593 btrfs_set_device_total_bytes(leaf, dev_item,
2594 btrfs_device_get_disk_total_bytes(device));
2595 btrfs_set_device_bytes_used(leaf, dev_item,
2596 btrfs_device_get_bytes_used(device));
2597 btrfs_mark_buffer_dirty(leaf);
2598
2599 out:
2600 btrfs_free_path(path);
2601 return ret;
2602 }
2603
2604 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2605 struct btrfs_device *device, u64 new_size)
2606 {
2607 struct btrfs_super_block *super_copy =
2608 device->dev_root->fs_info->super_copy;
2609 struct btrfs_fs_devices *fs_devices;
2610 u64 old_total;
2611 u64 diff;
2612
2613 if (!device->writeable)
2614 return -EACCES;
2615
2616 lock_chunks(device->dev_root);
2617 old_total = btrfs_super_total_bytes(super_copy);
2618 diff = new_size - device->total_bytes;
2619
2620 if (new_size <= device->total_bytes ||
2621 device->is_tgtdev_for_dev_replace) {
2622 unlock_chunks(device->dev_root);
2623 return -EINVAL;
2624 }
2625
2626 fs_devices = device->dev_root->fs_info->fs_devices;
2627
2628 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2629 device->fs_devices->total_rw_bytes += diff;
2630
2631 btrfs_device_set_total_bytes(device, new_size);
2632 btrfs_device_set_disk_total_bytes(device, new_size);
2633 btrfs_clear_space_info_full(device->dev_root->fs_info);
2634 if (list_empty(&device->resized_list))
2635 list_add_tail(&device->resized_list,
2636 &fs_devices->resized_devices);
2637 unlock_chunks(device->dev_root);
2638
2639 return btrfs_update_device(trans, device);
2640 }
2641
2642 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *root, u64 chunk_objectid,
2644 u64 chunk_offset)
2645 {
2646 int ret;
2647 struct btrfs_path *path;
2648 struct btrfs_key key;
2649
2650 root = root->fs_info->chunk_root;
2651 path = btrfs_alloc_path();
2652 if (!path)
2653 return -ENOMEM;
2654
2655 key.objectid = chunk_objectid;
2656 key.offset = chunk_offset;
2657 key.type = BTRFS_CHUNK_ITEM_KEY;
2658
2659 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2660 if (ret < 0)
2661 goto out;
2662 else if (ret > 0) { /* Logic error or corruption */
2663 btrfs_std_error(root->fs_info, -ENOENT,
2664 "Failed lookup while freeing chunk.");
2665 ret = -ENOENT;
2666 goto out;
2667 }
2668
2669 ret = btrfs_del_item(trans, root, path);
2670 if (ret < 0)
2671 btrfs_std_error(root->fs_info, ret,
2672 "Failed to delete chunk item.");
2673 out:
2674 btrfs_free_path(path);
2675 return ret;
2676 }
2677
2678 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2679 chunk_offset)
2680 {
2681 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2682 struct btrfs_disk_key *disk_key;
2683 struct btrfs_chunk *chunk;
2684 u8 *ptr;
2685 int ret = 0;
2686 u32 num_stripes;
2687 u32 array_size;
2688 u32 len = 0;
2689 u32 cur;
2690 struct btrfs_key key;
2691
2692 lock_chunks(root);
2693 array_size = btrfs_super_sys_array_size(super_copy);
2694
2695 ptr = super_copy->sys_chunk_array;
2696 cur = 0;
2697
2698 while (cur < array_size) {
2699 disk_key = (struct btrfs_disk_key *)ptr;
2700 btrfs_disk_key_to_cpu(&key, disk_key);
2701
2702 len = sizeof(*disk_key);
2703
2704 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2705 chunk = (struct btrfs_chunk *)(ptr + len);
2706 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2707 len += btrfs_chunk_item_size(num_stripes);
2708 } else {
2709 ret = -EIO;
2710 break;
2711 }
2712 if (key.objectid == chunk_objectid &&
2713 key.offset == chunk_offset) {
2714 memmove(ptr, ptr + len, array_size - (cur + len));
2715 array_size -= len;
2716 btrfs_set_super_sys_array_size(super_copy, array_size);
2717 } else {
2718 ptr += len;
2719 cur += len;
2720 }
2721 }
2722 unlock_chunks(root);
2723 return ret;
2724 }
2725
2726 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2727 struct btrfs_root *root, u64 chunk_offset)
2728 {
2729 struct extent_map_tree *em_tree;
2730 struct extent_map *em;
2731 struct btrfs_root *extent_root = root->fs_info->extent_root;
2732 struct map_lookup *map;
2733 u64 dev_extent_len = 0;
2734 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2735 int i, ret = 0;
2736
2737 /* Just in case */
2738 root = root->fs_info->chunk_root;
2739 em_tree = &root->fs_info->mapping_tree.map_tree;
2740
2741 read_lock(&em_tree->lock);
2742 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2743 read_unlock(&em_tree->lock);
2744
2745 if (!em || em->start > chunk_offset ||
2746 em->start + em->len < chunk_offset) {
2747 /*
2748 * This is a logic error, but we don't want to just rely on the
2749 * user having built with ASSERT enabled, so if ASSERT doens't
2750 * do anything we still error out.
2751 */
2752 ASSERT(0);
2753 if (em)
2754 free_extent_map(em);
2755 return -EINVAL;
2756 }
2757 map = em->map_lookup;
2758 lock_chunks(root->fs_info->chunk_root);
2759 check_system_chunk(trans, extent_root, map->type);
2760 unlock_chunks(root->fs_info->chunk_root);
2761
2762 for (i = 0; i < map->num_stripes; i++) {
2763 struct btrfs_device *device = map->stripes[i].dev;
2764 ret = btrfs_free_dev_extent(trans, device,
2765 map->stripes[i].physical,
2766 &dev_extent_len);
2767 if (ret) {
2768 btrfs_abort_transaction(trans, root, ret);
2769 goto out;
2770 }
2771
2772 if (device->bytes_used > 0) {
2773 lock_chunks(root);
2774 btrfs_device_set_bytes_used(device,
2775 device->bytes_used - dev_extent_len);
2776 spin_lock(&root->fs_info->free_chunk_lock);
2777 root->fs_info->free_chunk_space += dev_extent_len;
2778 spin_unlock(&root->fs_info->free_chunk_lock);
2779 btrfs_clear_space_info_full(root->fs_info);
2780 unlock_chunks(root);
2781 }
2782
2783 if (map->stripes[i].dev) {
2784 ret = btrfs_update_device(trans, map->stripes[i].dev);
2785 if (ret) {
2786 btrfs_abort_transaction(trans, root, ret);
2787 goto out;
2788 }
2789 }
2790 }
2791 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2792 if (ret) {
2793 btrfs_abort_transaction(trans, root, ret);
2794 goto out;
2795 }
2796
2797 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2798
2799 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2800 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2801 if (ret) {
2802 btrfs_abort_transaction(trans, root, ret);
2803 goto out;
2804 }
2805 }
2806
2807 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2808 if (ret) {
2809 btrfs_abort_transaction(trans, extent_root, ret);
2810 goto out;
2811 }
2812
2813 out:
2814 /* once for us */
2815 free_extent_map(em);
2816 return ret;
2817 }
2818
2819 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2820 {
2821 struct btrfs_root *extent_root;
2822 struct btrfs_trans_handle *trans;
2823 int ret;
2824
2825 root = root->fs_info->chunk_root;
2826 extent_root = root->fs_info->extent_root;
2827
2828 /*
2829 * Prevent races with automatic removal of unused block groups.
2830 * After we relocate and before we remove the chunk with offset
2831 * chunk_offset, automatic removal of the block group can kick in,
2832 * resulting in a failure when calling btrfs_remove_chunk() below.
2833 *
2834 * Make sure to acquire this mutex before doing a tree search (dev
2835 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2836 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2837 * we release the path used to search the chunk/dev tree and before
2838 * the current task acquires this mutex and calls us.
2839 */
2840 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2841
2842 ret = btrfs_can_relocate(extent_root, chunk_offset);
2843 if (ret)
2844 return -ENOSPC;
2845
2846 /* step one, relocate all the extents inside this chunk */
2847 btrfs_scrub_pause(root);
2848 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2849 btrfs_scrub_continue(root);
2850 if (ret)
2851 return ret;
2852
2853 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2854 chunk_offset);
2855 if (IS_ERR(trans)) {
2856 ret = PTR_ERR(trans);
2857 btrfs_std_error(root->fs_info, ret, NULL);
2858 return ret;
2859 }
2860
2861 /*
2862 * step two, delete the device extents and the
2863 * chunk tree entries
2864 */
2865 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2866 btrfs_end_transaction(trans, root);
2867 return ret;
2868 }
2869
2870 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2871 {
2872 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2873 struct btrfs_path *path;
2874 struct extent_buffer *leaf;
2875 struct btrfs_chunk *chunk;
2876 struct btrfs_key key;
2877 struct btrfs_key found_key;
2878 u64 chunk_type;
2879 bool retried = false;
2880 int failed = 0;
2881 int ret;
2882
2883 path = btrfs_alloc_path();
2884 if (!path)
2885 return -ENOMEM;
2886
2887 again:
2888 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2889 key.offset = (u64)-1;
2890 key.type = BTRFS_CHUNK_ITEM_KEY;
2891
2892 while (1) {
2893 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2894 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2895 if (ret < 0) {
2896 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2897 goto error;
2898 }
2899 BUG_ON(ret == 0); /* Corruption */
2900
2901 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2902 key.type);
2903 if (ret)
2904 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2905 if (ret < 0)
2906 goto error;
2907 if (ret > 0)
2908 break;
2909
2910 leaf = path->nodes[0];
2911 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2912
2913 chunk = btrfs_item_ptr(leaf, path->slots[0],
2914 struct btrfs_chunk);
2915 chunk_type = btrfs_chunk_type(leaf, chunk);
2916 btrfs_release_path(path);
2917
2918 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2919 ret = btrfs_relocate_chunk(chunk_root,
2920 found_key.offset);
2921 if (ret == -ENOSPC)
2922 failed++;
2923 else
2924 BUG_ON(ret);
2925 }
2926 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2927
2928 if (found_key.offset == 0)
2929 break;
2930 key.offset = found_key.offset - 1;
2931 }
2932 ret = 0;
2933 if (failed && !retried) {
2934 failed = 0;
2935 retried = true;
2936 goto again;
2937 } else if (WARN_ON(failed && retried)) {
2938 ret = -ENOSPC;
2939 }
2940 error:
2941 btrfs_free_path(path);
2942 return ret;
2943 }
2944
2945 static int insert_balance_item(struct btrfs_root *root,
2946 struct btrfs_balance_control *bctl)
2947 {
2948 struct btrfs_trans_handle *trans;
2949 struct btrfs_balance_item *item;
2950 struct btrfs_disk_balance_args disk_bargs;
2951 struct btrfs_path *path;
2952 struct extent_buffer *leaf;
2953 struct btrfs_key key;
2954 int ret, err;
2955
2956 path = btrfs_alloc_path();
2957 if (!path)
2958 return -ENOMEM;
2959
2960 trans = btrfs_start_transaction(root, 0);
2961 if (IS_ERR(trans)) {
2962 btrfs_free_path(path);
2963 return PTR_ERR(trans);
2964 }
2965
2966 key.objectid = BTRFS_BALANCE_OBJECTID;
2967 key.type = BTRFS_BALANCE_ITEM_KEY;
2968 key.offset = 0;
2969
2970 ret = btrfs_insert_empty_item(trans, root, path, &key,
2971 sizeof(*item));
2972 if (ret)
2973 goto out;
2974
2975 leaf = path->nodes[0];
2976 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2977
2978 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2979
2980 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2981 btrfs_set_balance_data(leaf, item, &disk_bargs);
2982 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2983 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2984 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2985 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2986
2987 btrfs_set_balance_flags(leaf, item, bctl->flags);
2988
2989 btrfs_mark_buffer_dirty(leaf);
2990 out:
2991 btrfs_free_path(path);
2992 err = btrfs_commit_transaction(trans, root);
2993 if (err && !ret)
2994 ret = err;
2995 return ret;
2996 }
2997
2998 static int del_balance_item(struct btrfs_root *root)
2999 {
3000 struct btrfs_trans_handle *trans;
3001 struct btrfs_path *path;
3002 struct btrfs_key key;
3003 int ret, err;
3004
3005 path = btrfs_alloc_path();
3006 if (!path)
3007 return -ENOMEM;
3008
3009 trans = btrfs_start_transaction(root, 0);
3010 if (IS_ERR(trans)) {
3011 btrfs_free_path(path);
3012 return PTR_ERR(trans);
3013 }
3014
3015 key.objectid = BTRFS_BALANCE_OBJECTID;
3016 key.type = BTRFS_BALANCE_ITEM_KEY;
3017 key.offset = 0;
3018
3019 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3020 if (ret < 0)
3021 goto out;
3022 if (ret > 0) {
3023 ret = -ENOENT;
3024 goto out;
3025 }
3026
3027 ret = btrfs_del_item(trans, root, path);
3028 out:
3029 btrfs_free_path(path);
3030 err = btrfs_commit_transaction(trans, root);
3031 if (err && !ret)
3032 ret = err;
3033 return ret;
3034 }
3035
3036 /*
3037 * This is a heuristic used to reduce the number of chunks balanced on
3038 * resume after balance was interrupted.
3039 */
3040 static void update_balance_args(struct btrfs_balance_control *bctl)
3041 {
3042 /*
3043 * Turn on soft mode for chunk types that were being converted.
3044 */
3045 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3046 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3047 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3048 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3049 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051
3052 /*
3053 * Turn on usage filter if is not already used. The idea is
3054 * that chunks that we have already balanced should be
3055 * reasonably full. Don't do it for chunks that are being
3056 * converted - that will keep us from relocating unconverted
3057 * (albeit full) chunks.
3058 */
3059 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3060 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3061 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3062 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3063 bctl->data.usage = 90;
3064 }
3065 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3066 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3067 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3068 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3069 bctl->sys.usage = 90;
3070 }
3071 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3072 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3073 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3074 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3075 bctl->meta.usage = 90;
3076 }
3077 }
3078
3079 /*
3080 * Should be called with both balance and volume mutexes held to
3081 * serialize other volume operations (add_dev/rm_dev/resize) with
3082 * restriper. Same goes for unset_balance_control.
3083 */
3084 static void set_balance_control(struct btrfs_balance_control *bctl)
3085 {
3086 struct btrfs_fs_info *fs_info = bctl->fs_info;
3087
3088 BUG_ON(fs_info->balance_ctl);
3089
3090 spin_lock(&fs_info->balance_lock);
3091 fs_info->balance_ctl = bctl;
3092 spin_unlock(&fs_info->balance_lock);
3093 }
3094
3095 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3096 {
3097 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3098
3099 BUG_ON(!fs_info->balance_ctl);
3100
3101 spin_lock(&fs_info->balance_lock);
3102 fs_info->balance_ctl = NULL;
3103 spin_unlock(&fs_info->balance_lock);
3104
3105 kfree(bctl);
3106 }
3107
3108 /*
3109 * Balance filters. Return 1 if chunk should be filtered out
3110 * (should not be balanced).
3111 */
3112 static int chunk_profiles_filter(u64 chunk_type,
3113 struct btrfs_balance_args *bargs)
3114 {
3115 chunk_type = chunk_to_extended(chunk_type) &
3116 BTRFS_EXTENDED_PROFILE_MASK;
3117
3118 if (bargs->profiles & chunk_type)
3119 return 0;
3120
3121 return 1;
3122 }
3123
3124 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3125 struct btrfs_balance_args *bargs)
3126 {
3127 struct btrfs_block_group_cache *cache;
3128 u64 chunk_used;
3129 u64 user_thresh_min;
3130 u64 user_thresh_max;
3131 int ret = 1;
3132
3133 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3134 chunk_used = btrfs_block_group_used(&cache->item);
3135
3136 if (bargs->usage_min == 0)
3137 user_thresh_min = 0;
3138 else
3139 user_thresh_min = div_factor_fine(cache->key.offset,
3140 bargs->usage_min);
3141
3142 if (bargs->usage_max == 0)
3143 user_thresh_max = 1;
3144 else if (bargs->usage_max > 100)
3145 user_thresh_max = cache->key.offset;
3146 else
3147 user_thresh_max = div_factor_fine(cache->key.offset,
3148 bargs->usage_max);
3149
3150 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3151 ret = 0;
3152
3153 btrfs_put_block_group(cache);
3154 return ret;
3155 }
3156
3157 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3158 u64 chunk_offset, struct btrfs_balance_args *bargs)
3159 {
3160 struct btrfs_block_group_cache *cache;
3161 u64 chunk_used, user_thresh;
3162 int ret = 1;
3163
3164 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3165 chunk_used = btrfs_block_group_used(&cache->item);
3166
3167 if (bargs->usage_min == 0)
3168 user_thresh = 1;
3169 else if (bargs->usage > 100)
3170 user_thresh = cache->key.offset;
3171 else
3172 user_thresh = div_factor_fine(cache->key.offset,
3173 bargs->usage);
3174
3175 if (chunk_used < user_thresh)
3176 ret = 0;
3177
3178 btrfs_put_block_group(cache);
3179 return ret;
3180 }
3181
3182 static int chunk_devid_filter(struct extent_buffer *leaf,
3183 struct btrfs_chunk *chunk,
3184 struct btrfs_balance_args *bargs)
3185 {
3186 struct btrfs_stripe *stripe;
3187 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3188 int i;
3189
3190 for (i = 0; i < num_stripes; i++) {
3191 stripe = btrfs_stripe_nr(chunk, i);
3192 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3193 return 0;
3194 }
3195
3196 return 1;
3197 }
3198
3199 /* [pstart, pend) */
3200 static int chunk_drange_filter(struct extent_buffer *leaf,
3201 struct btrfs_chunk *chunk,
3202 u64 chunk_offset,
3203 struct btrfs_balance_args *bargs)
3204 {
3205 struct btrfs_stripe *stripe;
3206 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3207 u64 stripe_offset;
3208 u64 stripe_length;
3209 int factor;
3210 int i;
3211
3212 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3213 return 0;
3214
3215 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3216 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3217 factor = num_stripes / 2;
3218 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3219 factor = num_stripes - 1;
3220 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3221 factor = num_stripes - 2;
3222 } else {
3223 factor = num_stripes;
3224 }
3225
3226 for (i = 0; i < num_stripes; i++) {
3227 stripe = btrfs_stripe_nr(chunk, i);
3228 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3229 continue;
3230
3231 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3232 stripe_length = btrfs_chunk_length(leaf, chunk);
3233 stripe_length = div_u64(stripe_length, factor);
3234
3235 if (stripe_offset < bargs->pend &&
3236 stripe_offset + stripe_length > bargs->pstart)
3237 return 0;
3238 }
3239
3240 return 1;
3241 }
3242
3243 /* [vstart, vend) */
3244 static int chunk_vrange_filter(struct extent_buffer *leaf,
3245 struct btrfs_chunk *chunk,
3246 u64 chunk_offset,
3247 struct btrfs_balance_args *bargs)
3248 {
3249 if (chunk_offset < bargs->vend &&
3250 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3251 /* at least part of the chunk is inside this vrange */
3252 return 0;
3253
3254 return 1;
3255 }
3256
3257 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3258 struct btrfs_chunk *chunk,
3259 struct btrfs_balance_args *bargs)
3260 {
3261 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3262
3263 if (bargs->stripes_min <= num_stripes
3264 && num_stripes <= bargs->stripes_max)
3265 return 0;
3266
3267 return 1;
3268 }
3269
3270 static int chunk_soft_convert_filter(u64 chunk_type,
3271 struct btrfs_balance_args *bargs)
3272 {
3273 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3274 return 0;
3275
3276 chunk_type = chunk_to_extended(chunk_type) &
3277 BTRFS_EXTENDED_PROFILE_MASK;
3278
3279 if (bargs->target == chunk_type)
3280 return 1;
3281
3282 return 0;
3283 }
3284
3285 static int should_balance_chunk(struct btrfs_root *root,
3286 struct extent_buffer *leaf,
3287 struct btrfs_chunk *chunk, u64 chunk_offset)
3288 {
3289 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3290 struct btrfs_balance_args *bargs = NULL;
3291 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3292
3293 /* type filter */
3294 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3295 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3296 return 0;
3297 }
3298
3299 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3300 bargs = &bctl->data;
3301 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3302 bargs = &bctl->sys;
3303 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3304 bargs = &bctl->meta;
3305
3306 /* profiles filter */
3307 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3308 chunk_profiles_filter(chunk_type, bargs)) {
3309 return 0;
3310 }
3311
3312 /* usage filter */
3313 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3314 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3315 return 0;
3316 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3317 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3318 return 0;
3319 }
3320
3321 /* devid filter */
3322 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3323 chunk_devid_filter(leaf, chunk, bargs)) {
3324 return 0;
3325 }
3326
3327 /* drange filter, makes sense only with devid filter */
3328 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3329 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3330 return 0;
3331 }
3332
3333 /* vrange filter */
3334 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3335 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3336 return 0;
3337 }
3338
3339 /* stripes filter */
3340 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3341 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3342 return 0;
3343 }
3344
3345 /* soft profile changing mode */
3346 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3347 chunk_soft_convert_filter(chunk_type, bargs)) {
3348 return 0;
3349 }
3350
3351 /*
3352 * limited by count, must be the last filter
3353 */
3354 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3355 if (bargs->limit == 0)
3356 return 0;
3357 else
3358 bargs->limit--;
3359 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3360 /*
3361 * Same logic as the 'limit' filter; the minimum cannot be
3362 * determined here because we do not have the global informatoin
3363 * about the count of all chunks that satisfy the filters.
3364 */
3365 if (bargs->limit_max == 0)
3366 return 0;
3367 else
3368 bargs->limit_max--;
3369 }
3370
3371 return 1;
3372 }
3373
3374 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3375 {
3376 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3377 struct btrfs_root *chunk_root = fs_info->chunk_root;
3378 struct btrfs_root *dev_root = fs_info->dev_root;
3379 struct list_head *devices;
3380 struct btrfs_device *device;
3381 u64 old_size;
3382 u64 size_to_free;
3383 u64 chunk_type;
3384 struct btrfs_chunk *chunk;
3385 struct btrfs_path *path;
3386 struct btrfs_key key;
3387 struct btrfs_key found_key;
3388 struct btrfs_trans_handle *trans;
3389 struct extent_buffer *leaf;
3390 int slot;
3391 int ret;
3392 int enospc_errors = 0;
3393 bool counting = true;
3394 /* The single value limit and min/max limits use the same bytes in the */
3395 u64 limit_data = bctl->data.limit;
3396 u64 limit_meta = bctl->meta.limit;
3397 u64 limit_sys = bctl->sys.limit;
3398 u32 count_data = 0;
3399 u32 count_meta = 0;
3400 u32 count_sys = 0;
3401 int chunk_reserved = 0;
3402
3403 /* step one make some room on all the devices */
3404 devices = &fs_info->fs_devices->devices;
3405 list_for_each_entry(device, devices, dev_list) {
3406 old_size = btrfs_device_get_total_bytes(device);
3407 size_to_free = div_factor(old_size, 1);
3408 size_to_free = min_t(u64, size_to_free, SZ_1M);
3409 if (!device->writeable ||
3410 btrfs_device_get_total_bytes(device) -
3411 btrfs_device_get_bytes_used(device) > size_to_free ||
3412 device->is_tgtdev_for_dev_replace)
3413 continue;
3414
3415 ret = btrfs_shrink_device(device, old_size - size_to_free);
3416 if (ret == -ENOSPC)
3417 break;
3418 BUG_ON(ret);
3419
3420 trans = btrfs_start_transaction(dev_root, 0);
3421 BUG_ON(IS_ERR(trans));
3422
3423 ret = btrfs_grow_device(trans, device, old_size);
3424 BUG_ON(ret);
3425
3426 btrfs_end_transaction(trans, dev_root);
3427 }
3428
3429 /* step two, relocate all the chunks */
3430 path = btrfs_alloc_path();
3431 if (!path) {
3432 ret = -ENOMEM;
3433 goto error;
3434 }
3435
3436 /* zero out stat counters */
3437 spin_lock(&fs_info->balance_lock);
3438 memset(&bctl->stat, 0, sizeof(bctl->stat));
3439 spin_unlock(&fs_info->balance_lock);
3440 again:
3441 if (!counting) {
3442 /*
3443 * The single value limit and min/max limits use the same bytes
3444 * in the
3445 */
3446 bctl->data.limit = limit_data;
3447 bctl->meta.limit = limit_meta;
3448 bctl->sys.limit = limit_sys;
3449 }
3450 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3451 key.offset = (u64)-1;
3452 key.type = BTRFS_CHUNK_ITEM_KEY;
3453
3454 while (1) {
3455 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3456 atomic_read(&fs_info->balance_cancel_req)) {
3457 ret = -ECANCELED;
3458 goto error;
3459 }
3460
3461 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3462 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3463 if (ret < 0) {
3464 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3465 goto error;
3466 }
3467
3468 /*
3469 * this shouldn't happen, it means the last relocate
3470 * failed
3471 */
3472 if (ret == 0)
3473 BUG(); /* FIXME break ? */
3474
3475 ret = btrfs_previous_item(chunk_root, path, 0,
3476 BTRFS_CHUNK_ITEM_KEY);
3477 if (ret) {
3478 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3479 ret = 0;
3480 break;
3481 }
3482
3483 leaf = path->nodes[0];
3484 slot = path->slots[0];
3485 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3486
3487 if (found_key.objectid != key.objectid) {
3488 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3489 break;
3490 }
3491
3492 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3493 chunk_type = btrfs_chunk_type(leaf, chunk);
3494
3495 if (!counting) {
3496 spin_lock(&fs_info->balance_lock);
3497 bctl->stat.considered++;
3498 spin_unlock(&fs_info->balance_lock);
3499 }
3500
3501 ret = should_balance_chunk(chunk_root, leaf, chunk,
3502 found_key.offset);
3503
3504 btrfs_release_path(path);
3505 if (!ret) {
3506 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3507 goto loop;
3508 }
3509
3510 if (counting) {
3511 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3512 spin_lock(&fs_info->balance_lock);
3513 bctl->stat.expected++;
3514 spin_unlock(&fs_info->balance_lock);
3515
3516 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3517 count_data++;
3518 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3519 count_sys++;
3520 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3521 count_meta++;
3522
3523 goto loop;
3524 }
3525
3526 /*
3527 * Apply limit_min filter, no need to check if the LIMITS
3528 * filter is used, limit_min is 0 by default
3529 */
3530 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3531 count_data < bctl->data.limit_min)
3532 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3533 count_meta < bctl->meta.limit_min)
3534 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3535 count_sys < bctl->sys.limit_min)) {
3536 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3537 goto loop;
3538 }
3539
3540 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3541 trans = btrfs_start_transaction(chunk_root, 0);
3542 if (IS_ERR(trans)) {
3543 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3544 ret = PTR_ERR(trans);
3545 goto error;
3546 }
3547
3548 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3549 BTRFS_BLOCK_GROUP_DATA);
3550 btrfs_end_transaction(trans, chunk_root);
3551 if (ret < 0) {
3552 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3553 goto error;
3554 }
3555 chunk_reserved = 1;
3556 }
3557
3558 ret = btrfs_relocate_chunk(chunk_root,
3559 found_key.offset);
3560 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3561 if (ret && ret != -ENOSPC)
3562 goto error;
3563 if (ret == -ENOSPC) {
3564 enospc_errors++;
3565 } else {
3566 spin_lock(&fs_info->balance_lock);
3567 bctl->stat.completed++;
3568 spin_unlock(&fs_info->balance_lock);
3569 }
3570 loop:
3571 if (found_key.offset == 0)
3572 break;
3573 key.offset = found_key.offset - 1;
3574 }
3575
3576 if (counting) {
3577 btrfs_release_path(path);
3578 counting = false;
3579 goto again;
3580 }
3581 error:
3582 btrfs_free_path(path);
3583 if (enospc_errors) {
3584 btrfs_info(fs_info, "%d enospc errors during balance",
3585 enospc_errors);
3586 if (!ret)
3587 ret = -ENOSPC;
3588 }
3589
3590 return ret;
3591 }
3592
3593 /**
3594 * alloc_profile_is_valid - see if a given profile is valid and reduced
3595 * @flags: profile to validate
3596 * @extended: if true @flags is treated as an extended profile
3597 */
3598 static int alloc_profile_is_valid(u64 flags, int extended)
3599 {
3600 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3601 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3602
3603 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3604
3605 /* 1) check that all other bits are zeroed */
3606 if (flags & ~mask)
3607 return 0;
3608
3609 /* 2) see if profile is reduced */
3610 if (flags == 0)
3611 return !extended; /* "0" is valid for usual profiles */
3612
3613 /* true if exactly one bit set */
3614 return (flags & (flags - 1)) == 0;
3615 }
3616
3617 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3618 {
3619 /* cancel requested || normal exit path */
3620 return atomic_read(&fs_info->balance_cancel_req) ||
3621 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3622 atomic_read(&fs_info->balance_cancel_req) == 0);
3623 }
3624
3625 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3626 {
3627 int ret;
3628
3629 unset_balance_control(fs_info);
3630 ret = del_balance_item(fs_info->tree_root);
3631 if (ret)
3632 btrfs_std_error(fs_info, ret, NULL);
3633
3634 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3635 }
3636
3637 /* Non-zero return value signifies invalidity */
3638 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3639 u64 allowed)
3640 {
3641 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3642 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3643 (bctl_arg->target & ~allowed)));
3644 }
3645
3646 /*
3647 * Should be called with both balance and volume mutexes held
3648 */
3649 int btrfs_balance(struct btrfs_balance_control *bctl,
3650 struct btrfs_ioctl_balance_args *bargs)
3651 {
3652 struct btrfs_fs_info *fs_info = bctl->fs_info;
3653 u64 allowed;
3654 int mixed = 0;
3655 int ret;
3656 u64 num_devices;
3657 unsigned seq;
3658
3659 if (btrfs_fs_closing(fs_info) ||
3660 atomic_read(&fs_info->balance_pause_req) ||
3661 atomic_read(&fs_info->balance_cancel_req)) {
3662 ret = -EINVAL;
3663 goto out;
3664 }
3665
3666 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3667 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3668 mixed = 1;
3669
3670 /*
3671 * In case of mixed groups both data and meta should be picked,
3672 * and identical options should be given for both of them.
3673 */
3674 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3675 if (mixed && (bctl->flags & allowed)) {
3676 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3677 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3678 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3679 btrfs_err(fs_info, "with mixed groups data and "
3680 "metadata balance options must be the same");
3681 ret = -EINVAL;
3682 goto out;
3683 }
3684 }
3685
3686 num_devices = fs_info->fs_devices->num_devices;
3687 btrfs_dev_replace_lock(&fs_info->dev_replace);
3688 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3689 BUG_ON(num_devices < 1);
3690 num_devices--;
3691 }
3692 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3693 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3694 if (num_devices == 1)
3695 allowed |= BTRFS_BLOCK_GROUP_DUP;
3696 else if (num_devices > 1)
3697 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3698 if (num_devices > 2)
3699 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3700 if (num_devices > 3)
3701 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3702 BTRFS_BLOCK_GROUP_RAID6);
3703 if (validate_convert_profile(&bctl->data, allowed)) {
3704 btrfs_err(fs_info, "unable to start balance with target "
3705 "data profile %llu",
3706 bctl->data.target);
3707 ret = -EINVAL;
3708 goto out;
3709 }
3710 if (validate_convert_profile(&bctl->meta, allowed)) {
3711 btrfs_err(fs_info,
3712 "unable to start balance with target metadata profile %llu",
3713 bctl->meta.target);
3714 ret = -EINVAL;
3715 goto out;
3716 }
3717 if (validate_convert_profile(&bctl->sys, allowed)) {
3718 btrfs_err(fs_info,
3719 "unable to start balance with target system profile %llu",
3720 bctl->sys.target);
3721 ret = -EINVAL;
3722 goto out;
3723 }
3724
3725 /* allow dup'ed data chunks only in mixed mode */
3726 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3727 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3728 btrfs_err(fs_info, "dup for data is not allowed");
3729 ret = -EINVAL;
3730 goto out;
3731 }
3732
3733 /* allow to reduce meta or sys integrity only if force set */
3734 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3735 BTRFS_BLOCK_GROUP_RAID10 |
3736 BTRFS_BLOCK_GROUP_RAID5 |
3737 BTRFS_BLOCK_GROUP_RAID6;
3738 do {
3739 seq = read_seqbegin(&fs_info->profiles_lock);
3740
3741 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3742 (fs_info->avail_system_alloc_bits & allowed) &&
3743 !(bctl->sys.target & allowed)) ||
3744 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3745 (fs_info->avail_metadata_alloc_bits & allowed) &&
3746 !(bctl->meta.target & allowed))) {
3747 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3748 btrfs_info(fs_info, "force reducing metadata integrity");
3749 } else {
3750 btrfs_err(fs_info, "balance will reduce metadata "
3751 "integrity, use force if you want this");
3752 ret = -EINVAL;
3753 goto out;
3754 }
3755 }
3756 } while (read_seqretry(&fs_info->profiles_lock, seq));
3757
3758 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3759 fs_info->num_tolerated_disk_barrier_failures = min(
3760 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3761 btrfs_get_num_tolerated_disk_barrier_failures(
3762 bctl->sys.target));
3763 }
3764
3765 ret = insert_balance_item(fs_info->tree_root, bctl);
3766 if (ret && ret != -EEXIST)
3767 goto out;
3768
3769 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3770 BUG_ON(ret == -EEXIST);
3771 set_balance_control(bctl);
3772 } else {
3773 BUG_ON(ret != -EEXIST);
3774 spin_lock(&fs_info->balance_lock);
3775 update_balance_args(bctl);
3776 spin_unlock(&fs_info->balance_lock);
3777 }
3778
3779 atomic_inc(&fs_info->balance_running);
3780 mutex_unlock(&fs_info->balance_mutex);
3781
3782 ret = __btrfs_balance(fs_info);
3783
3784 mutex_lock(&fs_info->balance_mutex);
3785 atomic_dec(&fs_info->balance_running);
3786
3787 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3788 fs_info->num_tolerated_disk_barrier_failures =
3789 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3790 }
3791
3792 if (bargs) {
3793 memset(bargs, 0, sizeof(*bargs));
3794 update_ioctl_balance_args(fs_info, 0, bargs);
3795 }
3796
3797 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3798 balance_need_close(fs_info)) {
3799 __cancel_balance(fs_info);
3800 }
3801
3802 wake_up(&fs_info->balance_wait_q);
3803
3804 return ret;
3805 out:
3806 if (bctl->flags & BTRFS_BALANCE_RESUME)
3807 __cancel_balance(fs_info);
3808 else {
3809 kfree(bctl);
3810 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3811 }
3812 return ret;
3813 }
3814
3815 static int balance_kthread(void *data)
3816 {
3817 struct btrfs_fs_info *fs_info = data;
3818 int ret = 0;
3819
3820 mutex_lock(&fs_info->volume_mutex);
3821 mutex_lock(&fs_info->balance_mutex);
3822
3823 if (fs_info->balance_ctl) {
3824 btrfs_info(fs_info, "continuing balance");
3825 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3826 }
3827
3828 mutex_unlock(&fs_info->balance_mutex);
3829 mutex_unlock(&fs_info->volume_mutex);
3830
3831 return ret;
3832 }
3833
3834 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3835 {
3836 struct task_struct *tsk;
3837
3838 spin_lock(&fs_info->balance_lock);
3839 if (!fs_info->balance_ctl) {
3840 spin_unlock(&fs_info->balance_lock);
3841 return 0;
3842 }
3843 spin_unlock(&fs_info->balance_lock);
3844
3845 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3846 btrfs_info(fs_info, "force skipping balance");
3847 return 0;
3848 }
3849
3850 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3851 return PTR_ERR_OR_ZERO(tsk);
3852 }
3853
3854 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3855 {
3856 struct btrfs_balance_control *bctl;
3857 struct btrfs_balance_item *item;
3858 struct btrfs_disk_balance_args disk_bargs;
3859 struct btrfs_path *path;
3860 struct extent_buffer *leaf;
3861 struct btrfs_key key;
3862 int ret;
3863
3864 path = btrfs_alloc_path();
3865 if (!path)
3866 return -ENOMEM;
3867
3868 key.objectid = BTRFS_BALANCE_OBJECTID;
3869 key.type = BTRFS_BALANCE_ITEM_KEY;
3870 key.offset = 0;
3871
3872 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3873 if (ret < 0)
3874 goto out;
3875 if (ret > 0) { /* ret = -ENOENT; */
3876 ret = 0;
3877 goto out;
3878 }
3879
3880 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3881 if (!bctl) {
3882 ret = -ENOMEM;
3883 goto out;
3884 }
3885
3886 leaf = path->nodes[0];
3887 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3888
3889 bctl->fs_info = fs_info;
3890 bctl->flags = btrfs_balance_flags(leaf, item);
3891 bctl->flags |= BTRFS_BALANCE_RESUME;
3892
3893 btrfs_balance_data(leaf, item, &disk_bargs);
3894 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3895 btrfs_balance_meta(leaf, item, &disk_bargs);
3896 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3897 btrfs_balance_sys(leaf, item, &disk_bargs);
3898 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3899
3900 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3901
3902 mutex_lock(&fs_info->volume_mutex);
3903 mutex_lock(&fs_info->balance_mutex);
3904
3905 set_balance_control(bctl);
3906
3907 mutex_unlock(&fs_info->balance_mutex);
3908 mutex_unlock(&fs_info->volume_mutex);
3909 out:
3910 btrfs_free_path(path);
3911 return ret;
3912 }
3913
3914 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3915 {
3916 int ret = 0;
3917
3918 mutex_lock(&fs_info->balance_mutex);
3919 if (!fs_info->balance_ctl) {
3920 mutex_unlock(&fs_info->balance_mutex);
3921 return -ENOTCONN;
3922 }
3923
3924 if (atomic_read(&fs_info->balance_running)) {
3925 atomic_inc(&fs_info->balance_pause_req);
3926 mutex_unlock(&fs_info->balance_mutex);
3927
3928 wait_event(fs_info->balance_wait_q,
3929 atomic_read(&fs_info->balance_running) == 0);
3930
3931 mutex_lock(&fs_info->balance_mutex);
3932 /* we are good with balance_ctl ripped off from under us */
3933 BUG_ON(atomic_read(&fs_info->balance_running));
3934 atomic_dec(&fs_info->balance_pause_req);
3935 } else {
3936 ret = -ENOTCONN;
3937 }
3938
3939 mutex_unlock(&fs_info->balance_mutex);
3940 return ret;
3941 }
3942
3943 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3944 {
3945 if (fs_info->sb->s_flags & MS_RDONLY)
3946 return -EROFS;
3947
3948 mutex_lock(&fs_info->balance_mutex);
3949 if (!fs_info->balance_ctl) {
3950 mutex_unlock(&fs_info->balance_mutex);
3951 return -ENOTCONN;
3952 }
3953
3954 atomic_inc(&fs_info->balance_cancel_req);
3955 /*
3956 * if we are running just wait and return, balance item is
3957 * deleted in btrfs_balance in this case
3958 */
3959 if (atomic_read(&fs_info->balance_running)) {
3960 mutex_unlock(&fs_info->balance_mutex);
3961 wait_event(fs_info->balance_wait_q,
3962 atomic_read(&fs_info->balance_running) == 0);
3963 mutex_lock(&fs_info->balance_mutex);
3964 } else {
3965 /* __cancel_balance needs volume_mutex */
3966 mutex_unlock(&fs_info->balance_mutex);
3967 mutex_lock(&fs_info->volume_mutex);
3968 mutex_lock(&fs_info->balance_mutex);
3969
3970 if (fs_info->balance_ctl)
3971 __cancel_balance(fs_info);
3972
3973 mutex_unlock(&fs_info->volume_mutex);
3974 }
3975
3976 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3977 atomic_dec(&fs_info->balance_cancel_req);
3978 mutex_unlock(&fs_info->balance_mutex);
3979 return 0;
3980 }
3981
3982 static int btrfs_uuid_scan_kthread(void *data)
3983 {
3984 struct btrfs_fs_info *fs_info = data;
3985 struct btrfs_root *root = fs_info->tree_root;
3986 struct btrfs_key key;
3987 struct btrfs_key max_key;
3988 struct btrfs_path *path = NULL;
3989 int ret = 0;
3990 struct extent_buffer *eb;
3991 int slot;
3992 struct btrfs_root_item root_item;
3993 u32 item_size;
3994 struct btrfs_trans_handle *trans = NULL;
3995
3996 path = btrfs_alloc_path();
3997 if (!path) {
3998 ret = -ENOMEM;
3999 goto out;
4000 }
4001
4002 key.objectid = 0;
4003 key.type = BTRFS_ROOT_ITEM_KEY;
4004 key.offset = 0;
4005
4006 max_key.objectid = (u64)-1;
4007 max_key.type = BTRFS_ROOT_ITEM_KEY;
4008 max_key.offset = (u64)-1;
4009
4010 while (1) {
4011 ret = btrfs_search_forward(root, &key, path, 0);
4012 if (ret) {
4013 if (ret > 0)
4014 ret = 0;
4015 break;
4016 }
4017
4018 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4019 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4020 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4021 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4022 goto skip;
4023
4024 eb = path->nodes[0];
4025 slot = path->slots[0];
4026 item_size = btrfs_item_size_nr(eb, slot);
4027 if (item_size < sizeof(root_item))
4028 goto skip;
4029
4030 read_extent_buffer(eb, &root_item,
4031 btrfs_item_ptr_offset(eb, slot),
4032 (int)sizeof(root_item));
4033 if (btrfs_root_refs(&root_item) == 0)
4034 goto skip;
4035
4036 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4037 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4038 if (trans)
4039 goto update_tree;
4040
4041 btrfs_release_path(path);
4042 /*
4043 * 1 - subvol uuid item
4044 * 1 - received_subvol uuid item
4045 */
4046 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4047 if (IS_ERR(trans)) {
4048 ret = PTR_ERR(trans);
4049 break;
4050 }
4051 continue;
4052 } else {
4053 goto skip;
4054 }
4055 update_tree:
4056 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4057 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4058 root_item.uuid,
4059 BTRFS_UUID_KEY_SUBVOL,
4060 key.objectid);
4061 if (ret < 0) {
4062 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4063 ret);
4064 break;
4065 }
4066 }
4067
4068 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4069 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4070 root_item.received_uuid,
4071 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4072 key.objectid);
4073 if (ret < 0) {
4074 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4075 ret);
4076 break;
4077 }
4078 }
4079
4080 skip:
4081 if (trans) {
4082 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4083 trans = NULL;
4084 if (ret)
4085 break;
4086 }
4087
4088 btrfs_release_path(path);
4089 if (key.offset < (u64)-1) {
4090 key.offset++;
4091 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4092 key.offset = 0;
4093 key.type = BTRFS_ROOT_ITEM_KEY;
4094 } else if (key.objectid < (u64)-1) {
4095 key.offset = 0;
4096 key.type = BTRFS_ROOT_ITEM_KEY;
4097 key.objectid++;
4098 } else {
4099 break;
4100 }
4101 cond_resched();
4102 }
4103
4104 out:
4105 btrfs_free_path(path);
4106 if (trans && !IS_ERR(trans))
4107 btrfs_end_transaction(trans, fs_info->uuid_root);
4108 if (ret)
4109 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4110 else
4111 fs_info->update_uuid_tree_gen = 1;
4112 up(&fs_info->uuid_tree_rescan_sem);
4113 return 0;
4114 }
4115
4116 /*
4117 * Callback for btrfs_uuid_tree_iterate().
4118 * returns:
4119 * 0 check succeeded, the entry is not outdated.
4120 * < 0 if an error occured.
4121 * > 0 if the check failed, which means the caller shall remove the entry.
4122 */
4123 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4124 u8 *uuid, u8 type, u64 subid)
4125 {
4126 struct btrfs_key key;
4127 int ret = 0;
4128 struct btrfs_root *subvol_root;
4129
4130 if (type != BTRFS_UUID_KEY_SUBVOL &&
4131 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4132 goto out;
4133
4134 key.objectid = subid;
4135 key.type = BTRFS_ROOT_ITEM_KEY;
4136 key.offset = (u64)-1;
4137 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4138 if (IS_ERR(subvol_root)) {
4139 ret = PTR_ERR(subvol_root);
4140 if (ret == -ENOENT)
4141 ret = 1;
4142 goto out;
4143 }
4144
4145 switch (type) {
4146 case BTRFS_UUID_KEY_SUBVOL:
4147 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4148 ret = 1;
4149 break;
4150 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4151 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4152 BTRFS_UUID_SIZE))
4153 ret = 1;
4154 break;
4155 }
4156
4157 out:
4158 return ret;
4159 }
4160
4161 static int btrfs_uuid_rescan_kthread(void *data)
4162 {
4163 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4164 int ret;
4165
4166 /*
4167 * 1st step is to iterate through the existing UUID tree and
4168 * to delete all entries that contain outdated data.
4169 * 2nd step is to add all missing entries to the UUID tree.
4170 */
4171 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4172 if (ret < 0) {
4173 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4174 up(&fs_info->uuid_tree_rescan_sem);
4175 return ret;
4176 }
4177 return btrfs_uuid_scan_kthread(data);
4178 }
4179
4180 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4181 {
4182 struct btrfs_trans_handle *trans;
4183 struct btrfs_root *tree_root = fs_info->tree_root;
4184 struct btrfs_root *uuid_root;
4185 struct task_struct *task;
4186 int ret;
4187
4188 /*
4189 * 1 - root node
4190 * 1 - root item
4191 */
4192 trans = btrfs_start_transaction(tree_root, 2);
4193 if (IS_ERR(trans))
4194 return PTR_ERR(trans);
4195
4196 uuid_root = btrfs_create_tree(trans, fs_info,
4197 BTRFS_UUID_TREE_OBJECTID);
4198 if (IS_ERR(uuid_root)) {
4199 ret = PTR_ERR(uuid_root);
4200 btrfs_abort_transaction(trans, tree_root, ret);
4201 return ret;
4202 }
4203
4204 fs_info->uuid_root = uuid_root;
4205
4206 ret = btrfs_commit_transaction(trans, tree_root);
4207 if (ret)
4208 return ret;
4209
4210 down(&fs_info->uuid_tree_rescan_sem);
4211 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4212 if (IS_ERR(task)) {
4213 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4214 btrfs_warn(fs_info, "failed to start uuid_scan task");
4215 up(&fs_info->uuid_tree_rescan_sem);
4216 return PTR_ERR(task);
4217 }
4218
4219 return 0;
4220 }
4221
4222 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4223 {
4224 struct task_struct *task;
4225
4226 down(&fs_info->uuid_tree_rescan_sem);
4227 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4228 if (IS_ERR(task)) {
4229 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4230 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4231 up(&fs_info->uuid_tree_rescan_sem);
4232 return PTR_ERR(task);
4233 }
4234
4235 return 0;
4236 }
4237
4238 /*
4239 * shrinking a device means finding all of the device extents past
4240 * the new size, and then following the back refs to the chunks.
4241 * The chunk relocation code actually frees the device extent
4242 */
4243 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4244 {
4245 struct btrfs_trans_handle *trans;
4246 struct btrfs_root *root = device->dev_root;
4247 struct btrfs_dev_extent *dev_extent = NULL;
4248 struct btrfs_path *path;
4249 u64 length;
4250 u64 chunk_offset;
4251 int ret;
4252 int slot;
4253 int failed = 0;
4254 bool retried = false;
4255 bool checked_pending_chunks = false;
4256 struct extent_buffer *l;
4257 struct btrfs_key key;
4258 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4259 u64 old_total = btrfs_super_total_bytes(super_copy);
4260 u64 old_size = btrfs_device_get_total_bytes(device);
4261 u64 diff = old_size - new_size;
4262
4263 if (device->is_tgtdev_for_dev_replace)
4264 return -EINVAL;
4265
4266 path = btrfs_alloc_path();
4267 if (!path)
4268 return -ENOMEM;
4269
4270 path->reada = 2;
4271
4272 lock_chunks(root);
4273
4274 btrfs_device_set_total_bytes(device, new_size);
4275 if (device->writeable) {
4276 device->fs_devices->total_rw_bytes -= diff;
4277 spin_lock(&root->fs_info->free_chunk_lock);
4278 root->fs_info->free_chunk_space -= diff;
4279 spin_unlock(&root->fs_info->free_chunk_lock);
4280 }
4281 unlock_chunks(root);
4282
4283 again:
4284 key.objectid = device->devid;
4285 key.offset = (u64)-1;
4286 key.type = BTRFS_DEV_EXTENT_KEY;
4287
4288 do {
4289 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4290 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4291 if (ret < 0) {
4292 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4293 goto done;
4294 }
4295
4296 ret = btrfs_previous_item(root, path, 0, key.type);
4297 if (ret)
4298 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4299 if (ret < 0)
4300 goto done;
4301 if (ret) {
4302 ret = 0;
4303 btrfs_release_path(path);
4304 break;
4305 }
4306
4307 l = path->nodes[0];
4308 slot = path->slots[0];
4309 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4310
4311 if (key.objectid != device->devid) {
4312 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4313 btrfs_release_path(path);
4314 break;
4315 }
4316
4317 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4318 length = btrfs_dev_extent_length(l, dev_extent);
4319
4320 if (key.offset + length <= new_size) {
4321 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4322 btrfs_release_path(path);
4323 break;
4324 }
4325
4326 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4327 btrfs_release_path(path);
4328
4329 ret = btrfs_relocate_chunk(root, chunk_offset);
4330 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4331 if (ret && ret != -ENOSPC)
4332 goto done;
4333 if (ret == -ENOSPC)
4334 failed++;
4335 } while (key.offset-- > 0);
4336
4337 if (failed && !retried) {
4338 failed = 0;
4339 retried = true;
4340 goto again;
4341 } else if (failed && retried) {
4342 ret = -ENOSPC;
4343 goto done;
4344 }
4345
4346 /* Shrinking succeeded, else we would be at "done". */
4347 trans = btrfs_start_transaction(root, 0);
4348 if (IS_ERR(trans)) {
4349 ret = PTR_ERR(trans);
4350 goto done;
4351 }
4352
4353 lock_chunks(root);
4354
4355 /*
4356 * We checked in the above loop all device extents that were already in
4357 * the device tree. However before we have updated the device's
4358 * total_bytes to the new size, we might have had chunk allocations that
4359 * have not complete yet (new block groups attached to transaction
4360 * handles), and therefore their device extents were not yet in the
4361 * device tree and we missed them in the loop above. So if we have any
4362 * pending chunk using a device extent that overlaps the device range
4363 * that we can not use anymore, commit the current transaction and
4364 * repeat the search on the device tree - this way we guarantee we will
4365 * not have chunks using device extents that end beyond 'new_size'.
4366 */
4367 if (!checked_pending_chunks) {
4368 u64 start = new_size;
4369 u64 len = old_size - new_size;
4370
4371 if (contains_pending_extent(trans->transaction, device,
4372 &start, len)) {
4373 unlock_chunks(root);
4374 checked_pending_chunks = true;
4375 failed = 0;
4376 retried = false;
4377 ret = btrfs_commit_transaction(trans, root);
4378 if (ret)
4379 goto done;
4380 goto again;
4381 }
4382 }
4383
4384 btrfs_device_set_disk_total_bytes(device, new_size);
4385 if (list_empty(&device->resized_list))
4386 list_add_tail(&device->resized_list,
4387 &root->fs_info->fs_devices->resized_devices);
4388
4389 WARN_ON(diff > old_total);
4390 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4391 unlock_chunks(root);
4392
4393 /* Now btrfs_update_device() will change the on-disk size. */
4394 ret = btrfs_update_device(trans, device);
4395 btrfs_end_transaction(trans, root);
4396 done:
4397 btrfs_free_path(path);
4398 if (ret) {
4399 lock_chunks(root);
4400 btrfs_device_set_total_bytes(device, old_size);
4401 if (device->writeable)
4402 device->fs_devices->total_rw_bytes += diff;
4403 spin_lock(&root->fs_info->free_chunk_lock);
4404 root->fs_info->free_chunk_space += diff;
4405 spin_unlock(&root->fs_info->free_chunk_lock);
4406 unlock_chunks(root);
4407 }
4408 return ret;
4409 }
4410
4411 static int btrfs_add_system_chunk(struct btrfs_root *root,
4412 struct btrfs_key *key,
4413 struct btrfs_chunk *chunk, int item_size)
4414 {
4415 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4416 struct btrfs_disk_key disk_key;
4417 u32 array_size;
4418 u8 *ptr;
4419
4420 lock_chunks(root);
4421 array_size = btrfs_super_sys_array_size(super_copy);
4422 if (array_size + item_size + sizeof(disk_key)
4423 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4424 unlock_chunks(root);
4425 return -EFBIG;
4426 }
4427
4428 ptr = super_copy->sys_chunk_array + array_size;
4429 btrfs_cpu_key_to_disk(&disk_key, key);
4430 memcpy(ptr, &disk_key, sizeof(disk_key));
4431 ptr += sizeof(disk_key);
4432 memcpy(ptr, chunk, item_size);
4433 item_size += sizeof(disk_key);
4434 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4435 unlock_chunks(root);
4436
4437 return 0;
4438 }
4439
4440 /*
4441 * sort the devices in descending order by max_avail, total_avail
4442 */
4443 static int btrfs_cmp_device_info(const void *a, const void *b)
4444 {
4445 const struct btrfs_device_info *di_a = a;
4446 const struct btrfs_device_info *di_b = b;
4447
4448 if (di_a->max_avail > di_b->max_avail)
4449 return -1;
4450 if (di_a->max_avail < di_b->max_avail)
4451 return 1;
4452 if (di_a->total_avail > di_b->total_avail)
4453 return -1;
4454 if (di_a->total_avail < di_b->total_avail)
4455 return 1;
4456 return 0;
4457 }
4458
4459 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4460 {
4461 /* TODO allow them to set a preferred stripe size */
4462 return SZ_64K;
4463 }
4464
4465 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4466 {
4467 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4468 return;
4469
4470 btrfs_set_fs_incompat(info, RAID56);
4471 }
4472
4473 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4474 - sizeof(struct btrfs_item) \
4475 - sizeof(struct btrfs_chunk)) \
4476 / sizeof(struct btrfs_stripe) + 1)
4477
4478 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4479 - 2 * sizeof(struct btrfs_disk_key) \
4480 - 2 * sizeof(struct btrfs_chunk)) \
4481 / sizeof(struct btrfs_stripe) + 1)
4482
4483 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4484 struct btrfs_root *extent_root, u64 start,
4485 u64 type)
4486 {
4487 struct btrfs_fs_info *info = extent_root->fs_info;
4488 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4489 struct list_head *cur;
4490 struct map_lookup *map = NULL;
4491 struct extent_map_tree *em_tree;
4492 struct extent_map *em;
4493 struct btrfs_device_info *devices_info = NULL;
4494 u64 total_avail;
4495 int num_stripes; /* total number of stripes to allocate */
4496 int data_stripes; /* number of stripes that count for
4497 block group size */
4498 int sub_stripes; /* sub_stripes info for map */
4499 int dev_stripes; /* stripes per dev */
4500 int devs_max; /* max devs to use */
4501 int devs_min; /* min devs needed */
4502 int devs_increment; /* ndevs has to be a multiple of this */
4503 int ncopies; /* how many copies to data has */
4504 int ret;
4505 u64 max_stripe_size;
4506 u64 max_chunk_size;
4507 u64 stripe_size;
4508 u64 num_bytes;
4509 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4510 int ndevs;
4511 int i;
4512 int j;
4513 int index;
4514
4515 BUG_ON(!alloc_profile_is_valid(type, 0));
4516
4517 if (list_empty(&fs_devices->alloc_list))
4518 return -ENOSPC;
4519
4520 index = __get_raid_index(type);
4521
4522 sub_stripes = btrfs_raid_array[index].sub_stripes;
4523 dev_stripes = btrfs_raid_array[index].dev_stripes;
4524 devs_max = btrfs_raid_array[index].devs_max;
4525 devs_min = btrfs_raid_array[index].devs_min;
4526 devs_increment = btrfs_raid_array[index].devs_increment;
4527 ncopies = btrfs_raid_array[index].ncopies;
4528
4529 if (type & BTRFS_BLOCK_GROUP_DATA) {
4530 max_stripe_size = SZ_1G;
4531 max_chunk_size = 10 * max_stripe_size;
4532 if (!devs_max)
4533 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4534 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4535 /* for larger filesystems, use larger metadata chunks */
4536 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4537 max_stripe_size = SZ_1G;
4538 else
4539 max_stripe_size = SZ_256M;
4540 max_chunk_size = max_stripe_size;
4541 if (!devs_max)
4542 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4543 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4544 max_stripe_size = SZ_32M;
4545 max_chunk_size = 2 * max_stripe_size;
4546 if (!devs_max)
4547 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4548 } else {
4549 btrfs_err(info, "invalid chunk type 0x%llx requested",
4550 type);
4551 BUG_ON(1);
4552 }
4553
4554 /* we don't want a chunk larger than 10% of writeable space */
4555 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4556 max_chunk_size);
4557
4558 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4559 GFP_NOFS);
4560 if (!devices_info)
4561 return -ENOMEM;
4562
4563 cur = fs_devices->alloc_list.next;
4564
4565 /*
4566 * in the first pass through the devices list, we gather information
4567 * about the available holes on each device.
4568 */
4569 ndevs = 0;
4570 while (cur != &fs_devices->alloc_list) {
4571 struct btrfs_device *device;
4572 u64 max_avail;
4573 u64 dev_offset;
4574
4575 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4576
4577 cur = cur->next;
4578
4579 if (!device->writeable) {
4580 WARN(1, KERN_ERR
4581 "BTRFS: read-only device in alloc_list\n");
4582 continue;
4583 }
4584
4585 if (!device->in_fs_metadata ||
4586 device->is_tgtdev_for_dev_replace)
4587 continue;
4588
4589 if (device->total_bytes > device->bytes_used)
4590 total_avail = device->total_bytes - device->bytes_used;
4591 else
4592 total_avail = 0;
4593
4594 /* If there is no space on this device, skip it. */
4595 if (total_avail == 0)
4596 continue;
4597
4598 ret = find_free_dev_extent(trans, device,
4599 max_stripe_size * dev_stripes,
4600 &dev_offset, &max_avail);
4601 if (ret && ret != -ENOSPC)
4602 goto error;
4603
4604 if (ret == 0)
4605 max_avail = max_stripe_size * dev_stripes;
4606
4607 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4608 continue;
4609
4610 if (ndevs == fs_devices->rw_devices) {
4611 WARN(1, "%s: found more than %llu devices\n",
4612 __func__, fs_devices->rw_devices);
4613 break;
4614 }
4615 devices_info[ndevs].dev_offset = dev_offset;
4616 devices_info[ndevs].max_avail = max_avail;
4617 devices_info[ndevs].total_avail = total_avail;
4618 devices_info[ndevs].dev = device;
4619 ++ndevs;
4620 }
4621
4622 /*
4623 * now sort the devices by hole size / available space
4624 */
4625 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4626 btrfs_cmp_device_info, NULL);
4627
4628 /* round down to number of usable stripes */
4629 ndevs -= ndevs % devs_increment;
4630
4631 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4632 ret = -ENOSPC;
4633 goto error;
4634 }
4635
4636 if (devs_max && ndevs > devs_max)
4637 ndevs = devs_max;
4638 /*
4639 * the primary goal is to maximize the number of stripes, so use as many
4640 * devices as possible, even if the stripes are not maximum sized.
4641 */
4642 stripe_size = devices_info[ndevs-1].max_avail;
4643 num_stripes = ndevs * dev_stripes;
4644
4645 /*
4646 * this will have to be fixed for RAID1 and RAID10 over
4647 * more drives
4648 */
4649 data_stripes = num_stripes / ncopies;
4650
4651 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4652 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4653 btrfs_super_stripesize(info->super_copy));
4654 data_stripes = num_stripes - 1;
4655 }
4656 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4657 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4658 btrfs_super_stripesize(info->super_copy));
4659 data_stripes = num_stripes - 2;
4660 }
4661
4662 /*
4663 * Use the number of data stripes to figure out how big this chunk
4664 * is really going to be in terms of logical address space,
4665 * and compare that answer with the max chunk size
4666 */
4667 if (stripe_size * data_stripes > max_chunk_size) {
4668 u64 mask = (1ULL << 24) - 1;
4669
4670 stripe_size = div_u64(max_chunk_size, data_stripes);
4671
4672 /* bump the answer up to a 16MB boundary */
4673 stripe_size = (stripe_size + mask) & ~mask;
4674
4675 /* but don't go higher than the limits we found
4676 * while searching for free extents
4677 */
4678 if (stripe_size > devices_info[ndevs-1].max_avail)
4679 stripe_size = devices_info[ndevs-1].max_avail;
4680 }
4681
4682 stripe_size = div_u64(stripe_size, dev_stripes);
4683
4684 /* align to BTRFS_STRIPE_LEN */
4685 stripe_size = div_u64(stripe_size, raid_stripe_len);
4686 stripe_size *= raid_stripe_len;
4687
4688 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4689 if (!map) {
4690 ret = -ENOMEM;
4691 goto error;
4692 }
4693 map->num_stripes = num_stripes;
4694
4695 for (i = 0; i < ndevs; ++i) {
4696 for (j = 0; j < dev_stripes; ++j) {
4697 int s = i * dev_stripes + j;
4698 map->stripes[s].dev = devices_info[i].dev;
4699 map->stripes[s].physical = devices_info[i].dev_offset +
4700 j * stripe_size;
4701 }
4702 }
4703 map->sector_size = extent_root->sectorsize;
4704 map->stripe_len = raid_stripe_len;
4705 map->io_align = raid_stripe_len;
4706 map->io_width = raid_stripe_len;
4707 map->type = type;
4708 map->sub_stripes = sub_stripes;
4709
4710 num_bytes = stripe_size * data_stripes;
4711
4712 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4713
4714 em = alloc_extent_map();
4715 if (!em) {
4716 kfree(map);
4717 ret = -ENOMEM;
4718 goto error;
4719 }
4720 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4721 em->map_lookup = map;
4722 em->start = start;
4723 em->len = num_bytes;
4724 em->block_start = 0;
4725 em->block_len = em->len;
4726 em->orig_block_len = stripe_size;
4727
4728 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4729 write_lock(&em_tree->lock);
4730 ret = add_extent_mapping(em_tree, em, 0);
4731 if (!ret) {
4732 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4733 atomic_inc(&em->refs);
4734 }
4735 write_unlock(&em_tree->lock);
4736 if (ret) {
4737 free_extent_map(em);
4738 goto error;
4739 }
4740
4741 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4742 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4743 start, num_bytes);
4744 if (ret)
4745 goto error_del_extent;
4746
4747 for (i = 0; i < map->num_stripes; i++) {
4748 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4749 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4750 }
4751
4752 spin_lock(&extent_root->fs_info->free_chunk_lock);
4753 extent_root->fs_info->free_chunk_space -= (stripe_size *
4754 map->num_stripes);
4755 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4756
4757 free_extent_map(em);
4758 check_raid56_incompat_flag(extent_root->fs_info, type);
4759
4760 kfree(devices_info);
4761 return 0;
4762
4763 error_del_extent:
4764 write_lock(&em_tree->lock);
4765 remove_extent_mapping(em_tree, em);
4766 write_unlock(&em_tree->lock);
4767
4768 /* One for our allocation */
4769 free_extent_map(em);
4770 /* One for the tree reference */
4771 free_extent_map(em);
4772 /* One for the pending_chunks list reference */
4773 free_extent_map(em);
4774 error:
4775 kfree(devices_info);
4776 return ret;
4777 }
4778
4779 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4780 struct btrfs_root *extent_root,
4781 u64 chunk_offset, u64 chunk_size)
4782 {
4783 struct btrfs_key key;
4784 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4785 struct btrfs_device *device;
4786 struct btrfs_chunk *chunk;
4787 struct btrfs_stripe *stripe;
4788 struct extent_map_tree *em_tree;
4789 struct extent_map *em;
4790 struct map_lookup *map;
4791 size_t item_size;
4792 u64 dev_offset;
4793 u64 stripe_size;
4794 int i = 0;
4795 int ret;
4796
4797 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4798 read_lock(&em_tree->lock);
4799 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4800 read_unlock(&em_tree->lock);
4801
4802 if (!em) {
4803 btrfs_crit(extent_root->fs_info, "unable to find logical "
4804 "%Lu len %Lu", chunk_offset, chunk_size);
4805 return -EINVAL;
4806 }
4807
4808 if (em->start != chunk_offset || em->len != chunk_size) {
4809 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4810 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4811 chunk_size, em->start, em->len);
4812 free_extent_map(em);
4813 return -EINVAL;
4814 }
4815
4816 map = em->map_lookup;
4817 item_size = btrfs_chunk_item_size(map->num_stripes);
4818 stripe_size = em->orig_block_len;
4819
4820 chunk = kzalloc(item_size, GFP_NOFS);
4821 if (!chunk) {
4822 ret = -ENOMEM;
4823 goto out;
4824 }
4825
4826 for (i = 0; i < map->num_stripes; i++) {
4827 device = map->stripes[i].dev;
4828 dev_offset = map->stripes[i].physical;
4829
4830 ret = btrfs_update_device(trans, device);
4831 if (ret)
4832 goto out;
4833 ret = btrfs_alloc_dev_extent(trans, device,
4834 chunk_root->root_key.objectid,
4835 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4836 chunk_offset, dev_offset,
4837 stripe_size);
4838 if (ret)
4839 goto out;
4840 }
4841
4842 stripe = &chunk->stripe;
4843 for (i = 0; i < map->num_stripes; i++) {
4844 device = map->stripes[i].dev;
4845 dev_offset = map->stripes[i].physical;
4846
4847 btrfs_set_stack_stripe_devid(stripe, device->devid);
4848 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4849 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4850 stripe++;
4851 }
4852
4853 btrfs_set_stack_chunk_length(chunk, chunk_size);
4854 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4855 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4856 btrfs_set_stack_chunk_type(chunk, map->type);
4857 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4858 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4859 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4860 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4861 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4862
4863 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4864 key.type = BTRFS_CHUNK_ITEM_KEY;
4865 key.offset = chunk_offset;
4866
4867 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4868 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4869 /*
4870 * TODO: Cleanup of inserted chunk root in case of
4871 * failure.
4872 */
4873 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4874 item_size);
4875 }
4876
4877 out:
4878 kfree(chunk);
4879 free_extent_map(em);
4880 return ret;
4881 }
4882
4883 /*
4884 * Chunk allocation falls into two parts. The first part does works
4885 * that make the new allocated chunk useable, but not do any operation
4886 * that modifies the chunk tree. The second part does the works that
4887 * require modifying the chunk tree. This division is important for the
4888 * bootstrap process of adding storage to a seed btrfs.
4889 */
4890 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4891 struct btrfs_root *extent_root, u64 type)
4892 {
4893 u64 chunk_offset;
4894
4895 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4896 chunk_offset = find_next_chunk(extent_root->fs_info);
4897 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4898 }
4899
4900 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4901 struct btrfs_root *root,
4902 struct btrfs_device *device)
4903 {
4904 u64 chunk_offset;
4905 u64 sys_chunk_offset;
4906 u64 alloc_profile;
4907 struct btrfs_fs_info *fs_info = root->fs_info;
4908 struct btrfs_root *extent_root = fs_info->extent_root;
4909 int ret;
4910
4911 chunk_offset = find_next_chunk(fs_info);
4912 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4913 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4914 alloc_profile);
4915 if (ret)
4916 return ret;
4917
4918 sys_chunk_offset = find_next_chunk(root->fs_info);
4919 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4920 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4921 alloc_profile);
4922 return ret;
4923 }
4924
4925 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4926 {
4927 int max_errors;
4928
4929 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4930 BTRFS_BLOCK_GROUP_RAID10 |
4931 BTRFS_BLOCK_GROUP_RAID5 |
4932 BTRFS_BLOCK_GROUP_DUP)) {
4933 max_errors = 1;
4934 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4935 max_errors = 2;
4936 } else {
4937 max_errors = 0;
4938 }
4939
4940 return max_errors;
4941 }
4942
4943 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4944 {
4945 struct extent_map *em;
4946 struct map_lookup *map;
4947 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4948 int readonly = 0;
4949 int miss_ndevs = 0;
4950 int i;
4951
4952 read_lock(&map_tree->map_tree.lock);
4953 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4954 read_unlock(&map_tree->map_tree.lock);
4955 if (!em)
4956 return 1;
4957
4958 map = em->map_lookup;
4959 for (i = 0; i < map->num_stripes; i++) {
4960 if (map->stripes[i].dev->missing) {
4961 miss_ndevs++;
4962 continue;
4963 }
4964
4965 if (!map->stripes[i].dev->writeable) {
4966 readonly = 1;
4967 goto end;
4968 }
4969 }
4970
4971 /*
4972 * If the number of missing devices is larger than max errors,
4973 * we can not write the data into that chunk successfully, so
4974 * set it readonly.
4975 */
4976 if (miss_ndevs > btrfs_chunk_max_errors(map))
4977 readonly = 1;
4978 end:
4979 free_extent_map(em);
4980 return readonly;
4981 }
4982
4983 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4984 {
4985 extent_map_tree_init(&tree->map_tree);
4986 }
4987
4988 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4989 {
4990 struct extent_map *em;
4991
4992 while (1) {
4993 write_lock(&tree->map_tree.lock);
4994 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4995 if (em)
4996 remove_extent_mapping(&tree->map_tree, em);
4997 write_unlock(&tree->map_tree.lock);
4998 if (!em)
4999 break;
5000 /* once for us */
5001 free_extent_map(em);
5002 /* once for the tree */
5003 free_extent_map(em);
5004 }
5005 }
5006
5007 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5008 {
5009 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5010 struct extent_map *em;
5011 struct map_lookup *map;
5012 struct extent_map_tree *em_tree = &map_tree->map_tree;
5013 int ret;
5014
5015 read_lock(&em_tree->lock);
5016 em = lookup_extent_mapping(em_tree, logical, len);
5017 read_unlock(&em_tree->lock);
5018
5019 /*
5020 * We could return errors for these cases, but that could get ugly and
5021 * we'd probably do the same thing which is just not do anything else
5022 * and exit, so return 1 so the callers don't try to use other copies.
5023 */
5024 if (!em) {
5025 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5026 logical+len);
5027 return 1;
5028 }
5029
5030 if (em->start > logical || em->start + em->len < logical) {
5031 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5032 "%Lu-%Lu", logical, logical+len, em->start,
5033 em->start + em->len);
5034 free_extent_map(em);
5035 return 1;
5036 }
5037
5038 map = em->map_lookup;
5039 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5040 ret = map->num_stripes;
5041 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5042 ret = map->sub_stripes;
5043 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5044 ret = 2;
5045 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5046 ret = 3;
5047 else
5048 ret = 1;
5049 free_extent_map(em);
5050
5051 btrfs_dev_replace_lock(&fs_info->dev_replace);
5052 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5053 ret++;
5054 btrfs_dev_replace_unlock(&fs_info->dev_replace);
5055
5056 return ret;
5057 }
5058
5059 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5060 struct btrfs_mapping_tree *map_tree,
5061 u64 logical)
5062 {
5063 struct extent_map *em;
5064 struct map_lookup *map;
5065 struct extent_map_tree *em_tree = &map_tree->map_tree;
5066 unsigned long len = root->sectorsize;
5067
5068 read_lock(&em_tree->lock);
5069 em = lookup_extent_mapping(em_tree, logical, len);
5070 read_unlock(&em_tree->lock);
5071 BUG_ON(!em);
5072
5073 BUG_ON(em->start > logical || em->start + em->len < logical);
5074 map = em->map_lookup;
5075 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5076 len = map->stripe_len * nr_data_stripes(map);
5077 free_extent_map(em);
5078 return len;
5079 }
5080
5081 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5082 u64 logical, u64 len, int mirror_num)
5083 {
5084 struct extent_map *em;
5085 struct map_lookup *map;
5086 struct extent_map_tree *em_tree = &map_tree->map_tree;
5087 int ret = 0;
5088
5089 read_lock(&em_tree->lock);
5090 em = lookup_extent_mapping(em_tree, logical, len);
5091 read_unlock(&em_tree->lock);
5092 BUG_ON(!em);
5093
5094 BUG_ON(em->start > logical || em->start + em->len < logical);
5095 map = em->map_lookup;
5096 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5097 ret = 1;
5098 free_extent_map(em);
5099 return ret;
5100 }
5101
5102 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5103 struct map_lookup *map, int first, int num,
5104 int optimal, int dev_replace_is_ongoing)
5105 {
5106 int i;
5107 int tolerance;
5108 struct btrfs_device *srcdev;
5109
5110 if (dev_replace_is_ongoing &&
5111 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5112 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5113 srcdev = fs_info->dev_replace.srcdev;
5114 else
5115 srcdev = NULL;
5116
5117 /*
5118 * try to avoid the drive that is the source drive for a
5119 * dev-replace procedure, only choose it if no other non-missing
5120 * mirror is available
5121 */
5122 for (tolerance = 0; tolerance < 2; tolerance++) {
5123 if (map->stripes[optimal].dev->bdev &&
5124 (tolerance || map->stripes[optimal].dev != srcdev))
5125 return optimal;
5126 for (i = first; i < first + num; i++) {
5127 if (map->stripes[i].dev->bdev &&
5128 (tolerance || map->stripes[i].dev != srcdev))
5129 return i;
5130 }
5131 }
5132
5133 /* we couldn't find one that doesn't fail. Just return something
5134 * and the io error handling code will clean up eventually
5135 */
5136 return optimal;
5137 }
5138
5139 static inline int parity_smaller(u64 a, u64 b)
5140 {
5141 return a > b;
5142 }
5143
5144 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5145 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5146 {
5147 struct btrfs_bio_stripe s;
5148 int i;
5149 u64 l;
5150 int again = 1;
5151
5152 while (again) {
5153 again = 0;
5154 for (i = 0; i < num_stripes - 1; i++) {
5155 if (parity_smaller(bbio->raid_map[i],
5156 bbio->raid_map[i+1])) {
5157 s = bbio->stripes[i];
5158 l = bbio->raid_map[i];
5159 bbio->stripes[i] = bbio->stripes[i+1];
5160 bbio->raid_map[i] = bbio->raid_map[i+1];
5161 bbio->stripes[i+1] = s;
5162 bbio->raid_map[i+1] = l;
5163
5164 again = 1;
5165 }
5166 }
5167 }
5168 }
5169
5170 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5171 {
5172 struct btrfs_bio *bbio = kzalloc(
5173 /* the size of the btrfs_bio */
5174 sizeof(struct btrfs_bio) +
5175 /* plus the variable array for the stripes */
5176 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5177 /* plus the variable array for the tgt dev */
5178 sizeof(int) * (real_stripes) +
5179 /*
5180 * plus the raid_map, which includes both the tgt dev
5181 * and the stripes
5182 */
5183 sizeof(u64) * (total_stripes),
5184 GFP_NOFS|__GFP_NOFAIL);
5185
5186 atomic_set(&bbio->error, 0);
5187 atomic_set(&bbio->refs, 1);
5188
5189 return bbio;
5190 }
5191
5192 void btrfs_get_bbio(struct btrfs_bio *bbio)
5193 {
5194 WARN_ON(!atomic_read(&bbio->refs));
5195 atomic_inc(&bbio->refs);
5196 }
5197
5198 void btrfs_put_bbio(struct btrfs_bio *bbio)
5199 {
5200 if (!bbio)
5201 return;
5202 if (atomic_dec_and_test(&bbio->refs))
5203 kfree(bbio);
5204 }
5205
5206 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5207 u64 logical, u64 *length,
5208 struct btrfs_bio **bbio_ret,
5209 int mirror_num, int need_raid_map)
5210 {
5211 struct extent_map *em;
5212 struct map_lookup *map;
5213 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5214 struct extent_map_tree *em_tree = &map_tree->map_tree;
5215 u64 offset;
5216 u64 stripe_offset;
5217 u64 stripe_end_offset;
5218 u64 stripe_nr;
5219 u64 stripe_nr_orig;
5220 u64 stripe_nr_end;
5221 u64 stripe_len;
5222 u32 stripe_index;
5223 int i;
5224 int ret = 0;
5225 int num_stripes;
5226 int max_errors = 0;
5227 int tgtdev_indexes = 0;
5228 struct btrfs_bio *bbio = NULL;
5229 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5230 int dev_replace_is_ongoing = 0;
5231 int num_alloc_stripes;
5232 int patch_the_first_stripe_for_dev_replace = 0;
5233 u64 physical_to_patch_in_first_stripe = 0;
5234 u64 raid56_full_stripe_start = (u64)-1;
5235
5236 read_lock(&em_tree->lock);
5237 em = lookup_extent_mapping(em_tree, logical, *length);
5238 read_unlock(&em_tree->lock);
5239
5240 if (!em) {
5241 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5242 logical, *length);
5243 return -EINVAL;
5244 }
5245
5246 if (em->start > logical || em->start + em->len < logical) {
5247 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5248 "found %Lu-%Lu", logical, em->start,
5249 em->start + em->len);
5250 free_extent_map(em);
5251 return -EINVAL;
5252 }
5253
5254 map = em->map_lookup;
5255 offset = logical - em->start;
5256
5257 stripe_len = map->stripe_len;
5258 stripe_nr = offset;
5259 /*
5260 * stripe_nr counts the total number of stripes we have to stride
5261 * to get to this block
5262 */
5263 stripe_nr = div64_u64(stripe_nr, stripe_len);
5264
5265 stripe_offset = stripe_nr * stripe_len;
5266 BUG_ON(offset < stripe_offset);
5267
5268 /* stripe_offset is the offset of this block in its stripe*/
5269 stripe_offset = offset - stripe_offset;
5270
5271 /* if we're here for raid56, we need to know the stripe aligned start */
5272 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5273 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5274 raid56_full_stripe_start = offset;
5275
5276 /* allow a write of a full stripe, but make sure we don't
5277 * allow straddling of stripes
5278 */
5279 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5280 full_stripe_len);
5281 raid56_full_stripe_start *= full_stripe_len;
5282 }
5283
5284 if (rw & REQ_DISCARD) {
5285 /* we don't discard raid56 yet */
5286 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5287 ret = -EOPNOTSUPP;
5288 goto out;
5289 }
5290 *length = min_t(u64, em->len - offset, *length);
5291 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5292 u64 max_len;
5293 /* For writes to RAID[56], allow a full stripeset across all disks.
5294 For other RAID types and for RAID[56] reads, just allow a single
5295 stripe (on a single disk). */
5296 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5297 (rw & REQ_WRITE)) {
5298 max_len = stripe_len * nr_data_stripes(map) -
5299 (offset - raid56_full_stripe_start);
5300 } else {
5301 /* we limit the length of each bio to what fits in a stripe */
5302 max_len = stripe_len - stripe_offset;
5303 }
5304 *length = min_t(u64, em->len - offset, max_len);
5305 } else {
5306 *length = em->len - offset;
5307 }
5308
5309 /* This is for when we're called from btrfs_merge_bio_hook() and all
5310 it cares about is the length */
5311 if (!bbio_ret)
5312 goto out;
5313
5314 btrfs_dev_replace_lock(dev_replace);
5315 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5316 if (!dev_replace_is_ongoing)
5317 btrfs_dev_replace_unlock(dev_replace);
5318
5319 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5320 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5321 dev_replace->tgtdev != NULL) {
5322 /*
5323 * in dev-replace case, for repair case (that's the only
5324 * case where the mirror is selected explicitly when
5325 * calling btrfs_map_block), blocks left of the left cursor
5326 * can also be read from the target drive.
5327 * For REQ_GET_READ_MIRRORS, the target drive is added as
5328 * the last one to the array of stripes. For READ, it also
5329 * needs to be supported using the same mirror number.
5330 * If the requested block is not left of the left cursor,
5331 * EIO is returned. This can happen because btrfs_num_copies()
5332 * returns one more in the dev-replace case.
5333 */
5334 u64 tmp_length = *length;
5335 struct btrfs_bio *tmp_bbio = NULL;
5336 int tmp_num_stripes;
5337 u64 srcdev_devid = dev_replace->srcdev->devid;
5338 int index_srcdev = 0;
5339 int found = 0;
5340 u64 physical_of_found = 0;
5341
5342 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5343 logical, &tmp_length, &tmp_bbio, 0, 0);
5344 if (ret) {
5345 WARN_ON(tmp_bbio != NULL);
5346 goto out;
5347 }
5348
5349 tmp_num_stripes = tmp_bbio->num_stripes;
5350 if (mirror_num > tmp_num_stripes) {
5351 /*
5352 * REQ_GET_READ_MIRRORS does not contain this
5353 * mirror, that means that the requested area
5354 * is not left of the left cursor
5355 */
5356 ret = -EIO;
5357 btrfs_put_bbio(tmp_bbio);
5358 goto out;
5359 }
5360
5361 /*
5362 * process the rest of the function using the mirror_num
5363 * of the source drive. Therefore look it up first.
5364 * At the end, patch the device pointer to the one of the
5365 * target drive.
5366 */
5367 for (i = 0; i < tmp_num_stripes; i++) {
5368 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5369 /*
5370 * In case of DUP, in order to keep it
5371 * simple, only add the mirror with the
5372 * lowest physical address
5373 */
5374 if (found &&
5375 physical_of_found <=
5376 tmp_bbio->stripes[i].physical)
5377 continue;
5378 index_srcdev = i;
5379 found = 1;
5380 physical_of_found =
5381 tmp_bbio->stripes[i].physical;
5382 }
5383 }
5384
5385 if (found) {
5386 mirror_num = index_srcdev + 1;
5387 patch_the_first_stripe_for_dev_replace = 1;
5388 physical_to_patch_in_first_stripe = physical_of_found;
5389 } else {
5390 WARN_ON(1);
5391 ret = -EIO;
5392 btrfs_put_bbio(tmp_bbio);
5393 goto out;
5394 }
5395
5396 btrfs_put_bbio(tmp_bbio);
5397 } else if (mirror_num > map->num_stripes) {
5398 mirror_num = 0;
5399 }
5400
5401 num_stripes = 1;
5402 stripe_index = 0;
5403 stripe_nr_orig = stripe_nr;
5404 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5405 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5406 stripe_end_offset = stripe_nr_end * map->stripe_len -
5407 (offset + *length);
5408
5409 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5410 if (rw & REQ_DISCARD)
5411 num_stripes = min_t(u64, map->num_stripes,
5412 stripe_nr_end - stripe_nr_orig);
5413 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5414 &stripe_index);
5415 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5416 mirror_num = 1;
5417 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5418 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5419 num_stripes = map->num_stripes;
5420 else if (mirror_num)
5421 stripe_index = mirror_num - 1;
5422 else {
5423 stripe_index = find_live_mirror(fs_info, map, 0,
5424 map->num_stripes,
5425 current->pid % map->num_stripes,
5426 dev_replace_is_ongoing);
5427 mirror_num = stripe_index + 1;
5428 }
5429
5430 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5431 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5432 num_stripes = map->num_stripes;
5433 } else if (mirror_num) {
5434 stripe_index = mirror_num - 1;
5435 } else {
5436 mirror_num = 1;
5437 }
5438
5439 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5440 u32 factor = map->num_stripes / map->sub_stripes;
5441
5442 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5443 stripe_index *= map->sub_stripes;
5444
5445 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5446 num_stripes = map->sub_stripes;
5447 else if (rw & REQ_DISCARD)
5448 num_stripes = min_t(u64, map->sub_stripes *
5449 (stripe_nr_end - stripe_nr_orig),
5450 map->num_stripes);
5451 else if (mirror_num)
5452 stripe_index += mirror_num - 1;
5453 else {
5454 int old_stripe_index = stripe_index;
5455 stripe_index = find_live_mirror(fs_info, map,
5456 stripe_index,
5457 map->sub_stripes, stripe_index +
5458 current->pid % map->sub_stripes,
5459 dev_replace_is_ongoing);
5460 mirror_num = stripe_index - old_stripe_index + 1;
5461 }
5462
5463 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5464 if (need_raid_map &&
5465 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5466 mirror_num > 1)) {
5467 /* push stripe_nr back to the start of the full stripe */
5468 stripe_nr = div_u64(raid56_full_stripe_start,
5469 stripe_len * nr_data_stripes(map));
5470
5471 /* RAID[56] write or recovery. Return all stripes */
5472 num_stripes = map->num_stripes;
5473 max_errors = nr_parity_stripes(map);
5474
5475 *length = map->stripe_len;
5476 stripe_index = 0;
5477 stripe_offset = 0;
5478 } else {
5479 /*
5480 * Mirror #0 or #1 means the original data block.
5481 * Mirror #2 is RAID5 parity block.
5482 * Mirror #3 is RAID6 Q block.
5483 */
5484 stripe_nr = div_u64_rem(stripe_nr,
5485 nr_data_stripes(map), &stripe_index);
5486 if (mirror_num > 1)
5487 stripe_index = nr_data_stripes(map) +
5488 mirror_num - 2;
5489
5490 /* We distribute the parity blocks across stripes */
5491 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5492 &stripe_index);
5493 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5494 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5495 mirror_num = 1;
5496 }
5497 } else {
5498 /*
5499 * after this, stripe_nr is the number of stripes on this
5500 * device we have to walk to find the data, and stripe_index is
5501 * the number of our device in the stripe array
5502 */
5503 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5504 &stripe_index);
5505 mirror_num = stripe_index + 1;
5506 }
5507 BUG_ON(stripe_index >= map->num_stripes);
5508
5509 num_alloc_stripes = num_stripes;
5510 if (dev_replace_is_ongoing) {
5511 if (rw & (REQ_WRITE | REQ_DISCARD))
5512 num_alloc_stripes <<= 1;
5513 if (rw & REQ_GET_READ_MIRRORS)
5514 num_alloc_stripes++;
5515 tgtdev_indexes = num_stripes;
5516 }
5517
5518 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5519 if (!bbio) {
5520 ret = -ENOMEM;
5521 goto out;
5522 }
5523 if (dev_replace_is_ongoing)
5524 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5525
5526 /* build raid_map */
5527 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5528 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5529 mirror_num > 1)) {
5530 u64 tmp;
5531 unsigned rot;
5532
5533 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5534 sizeof(struct btrfs_bio_stripe) *
5535 num_alloc_stripes +
5536 sizeof(int) * tgtdev_indexes);
5537
5538 /* Work out the disk rotation on this stripe-set */
5539 div_u64_rem(stripe_nr, num_stripes, &rot);
5540
5541 /* Fill in the logical address of each stripe */
5542 tmp = stripe_nr * nr_data_stripes(map);
5543 for (i = 0; i < nr_data_stripes(map); i++)
5544 bbio->raid_map[(i+rot) % num_stripes] =
5545 em->start + (tmp + i) * map->stripe_len;
5546
5547 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5548 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5549 bbio->raid_map[(i+rot+1) % num_stripes] =
5550 RAID6_Q_STRIPE;
5551 }
5552
5553 if (rw & REQ_DISCARD) {
5554 u32 factor = 0;
5555 u32 sub_stripes = 0;
5556 u64 stripes_per_dev = 0;
5557 u32 remaining_stripes = 0;
5558 u32 last_stripe = 0;
5559
5560 if (map->type &
5561 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5562 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5563 sub_stripes = 1;
5564 else
5565 sub_stripes = map->sub_stripes;
5566
5567 factor = map->num_stripes / sub_stripes;
5568 stripes_per_dev = div_u64_rem(stripe_nr_end -
5569 stripe_nr_orig,
5570 factor,
5571 &remaining_stripes);
5572 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5573 last_stripe *= sub_stripes;
5574 }
5575
5576 for (i = 0; i < num_stripes; i++) {
5577 bbio->stripes[i].physical =
5578 map->stripes[stripe_index].physical +
5579 stripe_offset + stripe_nr * map->stripe_len;
5580 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5581
5582 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5583 BTRFS_BLOCK_GROUP_RAID10)) {
5584 bbio->stripes[i].length = stripes_per_dev *
5585 map->stripe_len;
5586
5587 if (i / sub_stripes < remaining_stripes)
5588 bbio->stripes[i].length +=
5589 map->stripe_len;
5590
5591 /*
5592 * Special for the first stripe and
5593 * the last stripe:
5594 *
5595 * |-------|...|-------|
5596 * |----------|
5597 * off end_off
5598 */
5599 if (i < sub_stripes)
5600 bbio->stripes[i].length -=
5601 stripe_offset;
5602
5603 if (stripe_index >= last_stripe &&
5604 stripe_index <= (last_stripe +
5605 sub_stripes - 1))
5606 bbio->stripes[i].length -=
5607 stripe_end_offset;
5608
5609 if (i == sub_stripes - 1)
5610 stripe_offset = 0;
5611 } else
5612 bbio->stripes[i].length = *length;
5613
5614 stripe_index++;
5615 if (stripe_index == map->num_stripes) {
5616 /* This could only happen for RAID0/10 */
5617 stripe_index = 0;
5618 stripe_nr++;
5619 }
5620 }
5621 } else {
5622 for (i = 0; i < num_stripes; i++) {
5623 bbio->stripes[i].physical =
5624 map->stripes[stripe_index].physical +
5625 stripe_offset +
5626 stripe_nr * map->stripe_len;
5627 bbio->stripes[i].dev =
5628 map->stripes[stripe_index].dev;
5629 stripe_index++;
5630 }
5631 }
5632
5633 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5634 max_errors = btrfs_chunk_max_errors(map);
5635
5636 if (bbio->raid_map)
5637 sort_parity_stripes(bbio, num_stripes);
5638
5639 tgtdev_indexes = 0;
5640 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5641 dev_replace->tgtdev != NULL) {
5642 int index_where_to_add;
5643 u64 srcdev_devid = dev_replace->srcdev->devid;
5644
5645 /*
5646 * duplicate the write operations while the dev replace
5647 * procedure is running. Since the copying of the old disk
5648 * to the new disk takes place at run time while the
5649 * filesystem is mounted writable, the regular write
5650 * operations to the old disk have to be duplicated to go
5651 * to the new disk as well.
5652 * Note that device->missing is handled by the caller, and
5653 * that the write to the old disk is already set up in the
5654 * stripes array.
5655 */
5656 index_where_to_add = num_stripes;
5657 for (i = 0; i < num_stripes; i++) {
5658 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5659 /* write to new disk, too */
5660 struct btrfs_bio_stripe *new =
5661 bbio->stripes + index_where_to_add;
5662 struct btrfs_bio_stripe *old =
5663 bbio->stripes + i;
5664
5665 new->physical = old->physical;
5666 new->length = old->length;
5667 new->dev = dev_replace->tgtdev;
5668 bbio->tgtdev_map[i] = index_where_to_add;
5669 index_where_to_add++;
5670 max_errors++;
5671 tgtdev_indexes++;
5672 }
5673 }
5674 num_stripes = index_where_to_add;
5675 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5676 dev_replace->tgtdev != NULL) {
5677 u64 srcdev_devid = dev_replace->srcdev->devid;
5678 int index_srcdev = 0;
5679 int found = 0;
5680 u64 physical_of_found = 0;
5681
5682 /*
5683 * During the dev-replace procedure, the target drive can
5684 * also be used to read data in case it is needed to repair
5685 * a corrupt block elsewhere. This is possible if the
5686 * requested area is left of the left cursor. In this area,
5687 * the target drive is a full copy of the source drive.
5688 */
5689 for (i = 0; i < num_stripes; i++) {
5690 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5691 /*
5692 * In case of DUP, in order to keep it
5693 * simple, only add the mirror with the
5694 * lowest physical address
5695 */
5696 if (found &&
5697 physical_of_found <=
5698 bbio->stripes[i].physical)
5699 continue;
5700 index_srcdev = i;
5701 found = 1;
5702 physical_of_found = bbio->stripes[i].physical;
5703 }
5704 }
5705 if (found) {
5706 if (physical_of_found + map->stripe_len <=
5707 dev_replace->cursor_left) {
5708 struct btrfs_bio_stripe *tgtdev_stripe =
5709 bbio->stripes + num_stripes;
5710
5711 tgtdev_stripe->physical = physical_of_found;
5712 tgtdev_stripe->length =
5713 bbio->stripes[index_srcdev].length;
5714 tgtdev_stripe->dev = dev_replace->tgtdev;
5715 bbio->tgtdev_map[index_srcdev] = num_stripes;
5716
5717 tgtdev_indexes++;
5718 num_stripes++;
5719 }
5720 }
5721 }
5722
5723 *bbio_ret = bbio;
5724 bbio->map_type = map->type;
5725 bbio->num_stripes = num_stripes;
5726 bbio->max_errors = max_errors;
5727 bbio->mirror_num = mirror_num;
5728 bbio->num_tgtdevs = tgtdev_indexes;
5729
5730 /*
5731 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5732 * mirror_num == num_stripes + 1 && dev_replace target drive is
5733 * available as a mirror
5734 */
5735 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5736 WARN_ON(num_stripes > 1);
5737 bbio->stripes[0].dev = dev_replace->tgtdev;
5738 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5739 bbio->mirror_num = map->num_stripes + 1;
5740 }
5741 out:
5742 if (dev_replace_is_ongoing)
5743 btrfs_dev_replace_unlock(dev_replace);
5744 free_extent_map(em);
5745 return ret;
5746 }
5747
5748 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5749 u64 logical, u64 *length,
5750 struct btrfs_bio **bbio_ret, int mirror_num)
5751 {
5752 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5753 mirror_num, 0);
5754 }
5755
5756 /* For Scrub/replace */
5757 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5758 u64 logical, u64 *length,
5759 struct btrfs_bio **bbio_ret, int mirror_num,
5760 int need_raid_map)
5761 {
5762 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5763 mirror_num, need_raid_map);
5764 }
5765
5766 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5767 u64 chunk_start, u64 physical, u64 devid,
5768 u64 **logical, int *naddrs, int *stripe_len)
5769 {
5770 struct extent_map_tree *em_tree = &map_tree->map_tree;
5771 struct extent_map *em;
5772 struct map_lookup *map;
5773 u64 *buf;
5774 u64 bytenr;
5775 u64 length;
5776 u64 stripe_nr;
5777 u64 rmap_len;
5778 int i, j, nr = 0;
5779
5780 read_lock(&em_tree->lock);
5781 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5782 read_unlock(&em_tree->lock);
5783
5784 if (!em) {
5785 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5786 chunk_start);
5787 return -EIO;
5788 }
5789
5790 if (em->start != chunk_start) {
5791 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5792 em->start, chunk_start);
5793 free_extent_map(em);
5794 return -EIO;
5795 }
5796 map = em->map_lookup;
5797
5798 length = em->len;
5799 rmap_len = map->stripe_len;
5800
5801 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5802 length = div_u64(length, map->num_stripes / map->sub_stripes);
5803 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5804 length = div_u64(length, map->num_stripes);
5805 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5806 length = div_u64(length, nr_data_stripes(map));
5807 rmap_len = map->stripe_len * nr_data_stripes(map);
5808 }
5809
5810 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5811 BUG_ON(!buf); /* -ENOMEM */
5812
5813 for (i = 0; i < map->num_stripes; i++) {
5814 if (devid && map->stripes[i].dev->devid != devid)
5815 continue;
5816 if (map->stripes[i].physical > physical ||
5817 map->stripes[i].physical + length <= physical)
5818 continue;
5819
5820 stripe_nr = physical - map->stripes[i].physical;
5821 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5822
5823 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5824 stripe_nr = stripe_nr * map->num_stripes + i;
5825 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5826 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5827 stripe_nr = stripe_nr * map->num_stripes + i;
5828 } /* else if RAID[56], multiply by nr_data_stripes().
5829 * Alternatively, just use rmap_len below instead of
5830 * map->stripe_len */
5831
5832 bytenr = chunk_start + stripe_nr * rmap_len;
5833 WARN_ON(nr >= map->num_stripes);
5834 for (j = 0; j < nr; j++) {
5835 if (buf[j] == bytenr)
5836 break;
5837 }
5838 if (j == nr) {
5839 WARN_ON(nr >= map->num_stripes);
5840 buf[nr++] = bytenr;
5841 }
5842 }
5843
5844 *logical = buf;
5845 *naddrs = nr;
5846 *stripe_len = rmap_len;
5847
5848 free_extent_map(em);
5849 return 0;
5850 }
5851
5852 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5853 {
5854 bio->bi_private = bbio->private;
5855 bio->bi_end_io = bbio->end_io;
5856 bio_endio(bio);
5857
5858 btrfs_put_bbio(bbio);
5859 }
5860
5861 static void btrfs_end_bio(struct bio *bio)
5862 {
5863 struct btrfs_bio *bbio = bio->bi_private;
5864 int is_orig_bio = 0;
5865
5866 if (bio->bi_error) {
5867 atomic_inc(&bbio->error);
5868 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5869 unsigned int stripe_index =
5870 btrfs_io_bio(bio)->stripe_index;
5871 struct btrfs_device *dev;
5872
5873 BUG_ON(stripe_index >= bbio->num_stripes);
5874 dev = bbio->stripes[stripe_index].dev;
5875 if (dev->bdev) {
5876 if (bio->bi_rw & WRITE)
5877 btrfs_dev_stat_inc(dev,
5878 BTRFS_DEV_STAT_WRITE_ERRS);
5879 else
5880 btrfs_dev_stat_inc(dev,
5881 BTRFS_DEV_STAT_READ_ERRS);
5882 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5883 btrfs_dev_stat_inc(dev,
5884 BTRFS_DEV_STAT_FLUSH_ERRS);
5885 btrfs_dev_stat_print_on_error(dev);
5886 }
5887 }
5888 }
5889
5890 if (bio == bbio->orig_bio)
5891 is_orig_bio = 1;
5892
5893 btrfs_bio_counter_dec(bbio->fs_info);
5894
5895 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5896 if (!is_orig_bio) {
5897 bio_put(bio);
5898 bio = bbio->orig_bio;
5899 }
5900
5901 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5902 /* only send an error to the higher layers if it is
5903 * beyond the tolerance of the btrfs bio
5904 */
5905 if (atomic_read(&bbio->error) > bbio->max_errors) {
5906 bio->bi_error = -EIO;
5907 } else {
5908 /*
5909 * this bio is actually up to date, we didn't
5910 * go over the max number of errors
5911 */
5912 bio->bi_error = 0;
5913 }
5914
5915 btrfs_end_bbio(bbio, bio);
5916 } else if (!is_orig_bio) {
5917 bio_put(bio);
5918 }
5919 }
5920
5921 /*
5922 * see run_scheduled_bios for a description of why bios are collected for
5923 * async submit.
5924 *
5925 * This will add one bio to the pending list for a device and make sure
5926 * the work struct is scheduled.
5927 */
5928 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5929 struct btrfs_device *device,
5930 int rw, struct bio *bio)
5931 {
5932 int should_queue = 1;
5933 struct btrfs_pending_bios *pending_bios;
5934
5935 if (device->missing || !device->bdev) {
5936 bio_io_error(bio);
5937 return;
5938 }
5939
5940 /* don't bother with additional async steps for reads, right now */
5941 if (!(rw & REQ_WRITE)) {
5942 bio_get(bio);
5943 btrfsic_submit_bio(rw, bio);
5944 bio_put(bio);
5945 return;
5946 }
5947
5948 /*
5949 * nr_async_bios allows us to reliably return congestion to the
5950 * higher layers. Otherwise, the async bio makes it appear we have
5951 * made progress against dirty pages when we've really just put it
5952 * on a queue for later
5953 */
5954 atomic_inc(&root->fs_info->nr_async_bios);
5955 WARN_ON(bio->bi_next);
5956 bio->bi_next = NULL;
5957 bio->bi_rw |= rw;
5958
5959 spin_lock(&device->io_lock);
5960 if (bio->bi_rw & REQ_SYNC)
5961 pending_bios = &device->pending_sync_bios;
5962 else
5963 pending_bios = &device->pending_bios;
5964
5965 if (pending_bios->tail)
5966 pending_bios->tail->bi_next = bio;
5967
5968 pending_bios->tail = bio;
5969 if (!pending_bios->head)
5970 pending_bios->head = bio;
5971 if (device->running_pending)
5972 should_queue = 0;
5973
5974 spin_unlock(&device->io_lock);
5975
5976 if (should_queue)
5977 btrfs_queue_work(root->fs_info->submit_workers,
5978 &device->work);
5979 }
5980
5981 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5982 struct bio *bio, u64 physical, int dev_nr,
5983 int rw, int async)
5984 {
5985 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5986
5987 bio->bi_private = bbio;
5988 btrfs_io_bio(bio)->stripe_index = dev_nr;
5989 bio->bi_end_io = btrfs_end_bio;
5990 bio->bi_iter.bi_sector = physical >> 9;
5991 #ifdef DEBUG
5992 {
5993 struct rcu_string *name;
5994
5995 rcu_read_lock();
5996 name = rcu_dereference(dev->name);
5997 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5998 "(%s id %llu), size=%u\n", rw,
5999 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6000 name->str, dev->devid, bio->bi_iter.bi_size);
6001 rcu_read_unlock();
6002 }
6003 #endif
6004 bio->bi_bdev = dev->bdev;
6005
6006 btrfs_bio_counter_inc_noblocked(root->fs_info);
6007
6008 if (async)
6009 btrfs_schedule_bio(root, dev, rw, bio);
6010 else
6011 btrfsic_submit_bio(rw, bio);
6012 }
6013
6014 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6015 {
6016 atomic_inc(&bbio->error);
6017 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6018 /* Shoud be the original bio. */
6019 WARN_ON(bio != bbio->orig_bio);
6020
6021 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6022 bio->bi_iter.bi_sector = logical >> 9;
6023 bio->bi_error = -EIO;
6024 btrfs_end_bbio(bbio, bio);
6025 }
6026 }
6027
6028 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6029 int mirror_num, int async_submit)
6030 {
6031 struct btrfs_device *dev;
6032 struct bio *first_bio = bio;
6033 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6034 u64 length = 0;
6035 u64 map_length;
6036 int ret;
6037 int dev_nr;
6038 int total_devs;
6039 struct btrfs_bio *bbio = NULL;
6040
6041 length = bio->bi_iter.bi_size;
6042 map_length = length;
6043
6044 btrfs_bio_counter_inc_blocked(root->fs_info);
6045 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6046 mirror_num, 1);
6047 if (ret) {
6048 btrfs_bio_counter_dec(root->fs_info);
6049 return ret;
6050 }
6051
6052 total_devs = bbio->num_stripes;
6053 bbio->orig_bio = first_bio;
6054 bbio->private = first_bio->bi_private;
6055 bbio->end_io = first_bio->bi_end_io;
6056 bbio->fs_info = root->fs_info;
6057 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6058
6059 if (bbio->raid_map) {
6060 /* In this case, map_length has been set to the length of
6061 a single stripe; not the whole write */
6062 if (rw & WRITE) {
6063 ret = raid56_parity_write(root, bio, bbio, map_length);
6064 } else {
6065 ret = raid56_parity_recover(root, bio, bbio, map_length,
6066 mirror_num, 1);
6067 }
6068
6069 btrfs_bio_counter_dec(root->fs_info);
6070 return ret;
6071 }
6072
6073 if (map_length < length) {
6074 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6075 logical, length, map_length);
6076 BUG();
6077 }
6078
6079 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6080 dev = bbio->stripes[dev_nr].dev;
6081 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6082 bbio_error(bbio, first_bio, logical);
6083 continue;
6084 }
6085
6086 if (dev_nr < total_devs - 1) {
6087 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6088 BUG_ON(!bio); /* -ENOMEM */
6089 } else
6090 bio = first_bio;
6091
6092 submit_stripe_bio(root, bbio, bio,
6093 bbio->stripes[dev_nr].physical, dev_nr, rw,
6094 async_submit);
6095 }
6096 btrfs_bio_counter_dec(root->fs_info);
6097 return 0;
6098 }
6099
6100 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6101 u8 *uuid, u8 *fsid)
6102 {
6103 struct btrfs_device *device;
6104 struct btrfs_fs_devices *cur_devices;
6105
6106 cur_devices = fs_info->fs_devices;
6107 while (cur_devices) {
6108 if (!fsid ||
6109 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6110 device = __find_device(&cur_devices->devices,
6111 devid, uuid);
6112 if (device)
6113 return device;
6114 }
6115 cur_devices = cur_devices->seed;
6116 }
6117 return NULL;
6118 }
6119
6120 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6121 struct btrfs_fs_devices *fs_devices,
6122 u64 devid, u8 *dev_uuid)
6123 {
6124 struct btrfs_device *device;
6125
6126 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6127 if (IS_ERR(device))
6128 return NULL;
6129
6130 list_add(&device->dev_list, &fs_devices->devices);
6131 device->fs_devices = fs_devices;
6132 fs_devices->num_devices++;
6133
6134 device->missing = 1;
6135 fs_devices->missing_devices++;
6136
6137 return device;
6138 }
6139
6140 /**
6141 * btrfs_alloc_device - allocate struct btrfs_device
6142 * @fs_info: used only for generating a new devid, can be NULL if
6143 * devid is provided (i.e. @devid != NULL).
6144 * @devid: a pointer to devid for this device. If NULL a new devid
6145 * is generated.
6146 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6147 * is generated.
6148 *
6149 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6150 * on error. Returned struct is not linked onto any lists and can be
6151 * destroyed with kfree() right away.
6152 */
6153 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6154 const u64 *devid,
6155 const u8 *uuid)
6156 {
6157 struct btrfs_device *dev;
6158 u64 tmp;
6159
6160 if (WARN_ON(!devid && !fs_info))
6161 return ERR_PTR(-EINVAL);
6162
6163 dev = __alloc_device();
6164 if (IS_ERR(dev))
6165 return dev;
6166
6167 if (devid)
6168 tmp = *devid;
6169 else {
6170 int ret;
6171
6172 ret = find_next_devid(fs_info, &tmp);
6173 if (ret) {
6174 kfree(dev);
6175 return ERR_PTR(ret);
6176 }
6177 }
6178 dev->devid = tmp;
6179
6180 if (uuid)
6181 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6182 else
6183 generate_random_uuid(dev->uuid);
6184
6185 btrfs_init_work(&dev->work, btrfs_submit_helper,
6186 pending_bios_fn, NULL, NULL);
6187
6188 return dev;
6189 }
6190
6191 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6192 struct extent_buffer *leaf,
6193 struct btrfs_chunk *chunk)
6194 {
6195 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6196 struct map_lookup *map;
6197 struct extent_map *em;
6198 u64 logical;
6199 u64 length;
6200 u64 devid;
6201 u8 uuid[BTRFS_UUID_SIZE];
6202 int num_stripes;
6203 int ret;
6204 int i;
6205
6206 logical = key->offset;
6207 length = btrfs_chunk_length(leaf, chunk);
6208
6209 read_lock(&map_tree->map_tree.lock);
6210 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6211 read_unlock(&map_tree->map_tree.lock);
6212
6213 /* already mapped? */
6214 if (em && em->start <= logical && em->start + em->len > logical) {
6215 free_extent_map(em);
6216 return 0;
6217 } else if (em) {
6218 free_extent_map(em);
6219 }
6220
6221 em = alloc_extent_map();
6222 if (!em)
6223 return -ENOMEM;
6224 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6225 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6226 if (!map) {
6227 free_extent_map(em);
6228 return -ENOMEM;
6229 }
6230
6231 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6232 em->map_lookup = map;
6233 em->start = logical;
6234 em->len = length;
6235 em->orig_start = 0;
6236 em->block_start = 0;
6237 em->block_len = em->len;
6238
6239 map->num_stripes = num_stripes;
6240 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6241 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6242 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6243 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6244 map->type = btrfs_chunk_type(leaf, chunk);
6245 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6246 for (i = 0; i < num_stripes; i++) {
6247 map->stripes[i].physical =
6248 btrfs_stripe_offset_nr(leaf, chunk, i);
6249 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6250 read_extent_buffer(leaf, uuid, (unsigned long)
6251 btrfs_stripe_dev_uuid_nr(chunk, i),
6252 BTRFS_UUID_SIZE);
6253 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6254 uuid, NULL);
6255 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6256 free_extent_map(em);
6257 return -EIO;
6258 }
6259 if (!map->stripes[i].dev) {
6260 map->stripes[i].dev =
6261 add_missing_dev(root, root->fs_info->fs_devices,
6262 devid, uuid);
6263 if (!map->stripes[i].dev) {
6264 free_extent_map(em);
6265 return -EIO;
6266 }
6267 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6268 devid, uuid);
6269 }
6270 map->stripes[i].dev->in_fs_metadata = 1;
6271 }
6272
6273 write_lock(&map_tree->map_tree.lock);
6274 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6275 write_unlock(&map_tree->map_tree.lock);
6276 BUG_ON(ret); /* Tree corruption */
6277 free_extent_map(em);
6278
6279 return 0;
6280 }
6281
6282 static void fill_device_from_item(struct extent_buffer *leaf,
6283 struct btrfs_dev_item *dev_item,
6284 struct btrfs_device *device)
6285 {
6286 unsigned long ptr;
6287
6288 device->devid = btrfs_device_id(leaf, dev_item);
6289 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6290 device->total_bytes = device->disk_total_bytes;
6291 device->commit_total_bytes = device->disk_total_bytes;
6292 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6293 device->commit_bytes_used = device->bytes_used;
6294 device->type = btrfs_device_type(leaf, dev_item);
6295 device->io_align = btrfs_device_io_align(leaf, dev_item);
6296 device->io_width = btrfs_device_io_width(leaf, dev_item);
6297 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6298 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6299 device->is_tgtdev_for_dev_replace = 0;
6300
6301 ptr = btrfs_device_uuid(dev_item);
6302 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6303 }
6304
6305 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6306 u8 *fsid)
6307 {
6308 struct btrfs_fs_devices *fs_devices;
6309 int ret;
6310
6311 BUG_ON(!mutex_is_locked(&uuid_mutex));
6312
6313 fs_devices = root->fs_info->fs_devices->seed;
6314 while (fs_devices) {
6315 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6316 return fs_devices;
6317
6318 fs_devices = fs_devices->seed;
6319 }
6320
6321 fs_devices = find_fsid(fsid);
6322 if (!fs_devices) {
6323 if (!btrfs_test_opt(root, DEGRADED))
6324 return ERR_PTR(-ENOENT);
6325
6326 fs_devices = alloc_fs_devices(fsid);
6327 if (IS_ERR(fs_devices))
6328 return fs_devices;
6329
6330 fs_devices->seeding = 1;
6331 fs_devices->opened = 1;
6332 return fs_devices;
6333 }
6334
6335 fs_devices = clone_fs_devices(fs_devices);
6336 if (IS_ERR(fs_devices))
6337 return fs_devices;
6338
6339 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6340 root->fs_info->bdev_holder);
6341 if (ret) {
6342 free_fs_devices(fs_devices);
6343 fs_devices = ERR_PTR(ret);
6344 goto out;
6345 }
6346
6347 if (!fs_devices->seeding) {
6348 __btrfs_close_devices(fs_devices);
6349 free_fs_devices(fs_devices);
6350 fs_devices = ERR_PTR(-EINVAL);
6351 goto out;
6352 }
6353
6354 fs_devices->seed = root->fs_info->fs_devices->seed;
6355 root->fs_info->fs_devices->seed = fs_devices;
6356 out:
6357 return fs_devices;
6358 }
6359
6360 static int read_one_dev(struct btrfs_root *root,
6361 struct extent_buffer *leaf,
6362 struct btrfs_dev_item *dev_item)
6363 {
6364 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6365 struct btrfs_device *device;
6366 u64 devid;
6367 int ret;
6368 u8 fs_uuid[BTRFS_UUID_SIZE];
6369 u8 dev_uuid[BTRFS_UUID_SIZE];
6370
6371 devid = btrfs_device_id(leaf, dev_item);
6372 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6373 BTRFS_UUID_SIZE);
6374 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6375 BTRFS_UUID_SIZE);
6376
6377 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6378 fs_devices = open_seed_devices(root, fs_uuid);
6379 if (IS_ERR(fs_devices))
6380 return PTR_ERR(fs_devices);
6381 }
6382
6383 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6384 if (!device) {
6385 if (!btrfs_test_opt(root, DEGRADED))
6386 return -EIO;
6387
6388 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6389 if (!device)
6390 return -ENOMEM;
6391 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6392 devid, dev_uuid);
6393 } else {
6394 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6395 return -EIO;
6396
6397 if(!device->bdev && !device->missing) {
6398 /*
6399 * this happens when a device that was properly setup
6400 * in the device info lists suddenly goes bad.
6401 * device->bdev is NULL, and so we have to set
6402 * device->missing to one here
6403 */
6404 device->fs_devices->missing_devices++;
6405 device->missing = 1;
6406 }
6407
6408 /* Move the device to its own fs_devices */
6409 if (device->fs_devices != fs_devices) {
6410 ASSERT(device->missing);
6411
6412 list_move(&device->dev_list, &fs_devices->devices);
6413 device->fs_devices->num_devices--;
6414 fs_devices->num_devices++;
6415
6416 device->fs_devices->missing_devices--;
6417 fs_devices->missing_devices++;
6418
6419 device->fs_devices = fs_devices;
6420 }
6421 }
6422
6423 if (device->fs_devices != root->fs_info->fs_devices) {
6424 BUG_ON(device->writeable);
6425 if (device->generation !=
6426 btrfs_device_generation(leaf, dev_item))
6427 return -EINVAL;
6428 }
6429
6430 fill_device_from_item(leaf, dev_item, device);
6431 device->in_fs_metadata = 1;
6432 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6433 device->fs_devices->total_rw_bytes += device->total_bytes;
6434 spin_lock(&root->fs_info->free_chunk_lock);
6435 root->fs_info->free_chunk_space += device->total_bytes -
6436 device->bytes_used;
6437 spin_unlock(&root->fs_info->free_chunk_lock);
6438 }
6439 ret = 0;
6440 return ret;
6441 }
6442
6443 int btrfs_read_sys_array(struct btrfs_root *root)
6444 {
6445 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6446 struct extent_buffer *sb;
6447 struct btrfs_disk_key *disk_key;
6448 struct btrfs_chunk *chunk;
6449 u8 *array_ptr;
6450 unsigned long sb_array_offset;
6451 int ret = 0;
6452 u32 num_stripes;
6453 u32 array_size;
6454 u32 len = 0;
6455 u32 cur_offset;
6456 struct btrfs_key key;
6457
6458 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6459 /*
6460 * This will create extent buffer of nodesize, superblock size is
6461 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6462 * overallocate but we can keep it as-is, only the first page is used.
6463 */
6464 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6465 if (!sb)
6466 return -ENOMEM;
6467 btrfs_set_buffer_uptodate(sb);
6468 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6469 /*
6470 * The sb extent buffer is artifical and just used to read the system array.
6471 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6472 * pages up-to-date when the page is larger: extent does not cover the
6473 * whole page and consequently check_page_uptodate does not find all
6474 * the page's extents up-to-date (the hole beyond sb),
6475 * write_extent_buffer then triggers a WARN_ON.
6476 *
6477 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6478 * but sb spans only this function. Add an explicit SetPageUptodate call
6479 * to silence the warning eg. on PowerPC 64.
6480 */
6481 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6482 SetPageUptodate(sb->pages[0]);
6483
6484 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6485 array_size = btrfs_super_sys_array_size(super_copy);
6486
6487 array_ptr = super_copy->sys_chunk_array;
6488 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6489 cur_offset = 0;
6490
6491 while (cur_offset < array_size) {
6492 disk_key = (struct btrfs_disk_key *)array_ptr;
6493 len = sizeof(*disk_key);
6494 if (cur_offset + len > array_size)
6495 goto out_short_read;
6496
6497 btrfs_disk_key_to_cpu(&key, disk_key);
6498
6499 array_ptr += len;
6500 sb_array_offset += len;
6501 cur_offset += len;
6502
6503 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6504 chunk = (struct btrfs_chunk *)sb_array_offset;
6505 /*
6506 * At least one btrfs_chunk with one stripe must be
6507 * present, exact stripe count check comes afterwards
6508 */
6509 len = btrfs_chunk_item_size(1);
6510 if (cur_offset + len > array_size)
6511 goto out_short_read;
6512
6513 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6514 len = btrfs_chunk_item_size(num_stripes);
6515 if (cur_offset + len > array_size)
6516 goto out_short_read;
6517
6518 ret = read_one_chunk(root, &key, sb, chunk);
6519 if (ret)
6520 break;
6521 } else {
6522 ret = -EIO;
6523 break;
6524 }
6525 array_ptr += len;
6526 sb_array_offset += len;
6527 cur_offset += len;
6528 }
6529 free_extent_buffer(sb);
6530 return ret;
6531
6532 out_short_read:
6533 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6534 len, cur_offset);
6535 free_extent_buffer(sb);
6536 return -EIO;
6537 }
6538
6539 int btrfs_read_chunk_tree(struct btrfs_root *root)
6540 {
6541 struct btrfs_path *path;
6542 struct extent_buffer *leaf;
6543 struct btrfs_key key;
6544 struct btrfs_key found_key;
6545 int ret;
6546 int slot;
6547
6548 root = root->fs_info->chunk_root;
6549
6550 path = btrfs_alloc_path();
6551 if (!path)
6552 return -ENOMEM;
6553
6554 mutex_lock(&uuid_mutex);
6555 lock_chunks(root);
6556
6557 /*
6558 * Read all device items, and then all the chunk items. All
6559 * device items are found before any chunk item (their object id
6560 * is smaller than the lowest possible object id for a chunk
6561 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6562 */
6563 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6564 key.offset = 0;
6565 key.type = 0;
6566 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6567 if (ret < 0)
6568 goto error;
6569 while (1) {
6570 leaf = path->nodes[0];
6571 slot = path->slots[0];
6572 if (slot >= btrfs_header_nritems(leaf)) {
6573 ret = btrfs_next_leaf(root, path);
6574 if (ret == 0)
6575 continue;
6576 if (ret < 0)
6577 goto error;
6578 break;
6579 }
6580 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6581 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6582 struct btrfs_dev_item *dev_item;
6583 dev_item = btrfs_item_ptr(leaf, slot,
6584 struct btrfs_dev_item);
6585 ret = read_one_dev(root, leaf, dev_item);
6586 if (ret)
6587 goto error;
6588 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6589 struct btrfs_chunk *chunk;
6590 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6591 ret = read_one_chunk(root, &found_key, leaf, chunk);
6592 if (ret)
6593 goto error;
6594 }
6595 path->slots[0]++;
6596 }
6597 ret = 0;
6598 error:
6599 unlock_chunks(root);
6600 mutex_unlock(&uuid_mutex);
6601
6602 btrfs_free_path(path);
6603 return ret;
6604 }
6605
6606 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6607 {
6608 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6609 struct btrfs_device *device;
6610
6611 while (fs_devices) {
6612 mutex_lock(&fs_devices->device_list_mutex);
6613 list_for_each_entry(device, &fs_devices->devices, dev_list)
6614 device->dev_root = fs_info->dev_root;
6615 mutex_unlock(&fs_devices->device_list_mutex);
6616
6617 fs_devices = fs_devices->seed;
6618 }
6619 }
6620
6621 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6622 {
6623 int i;
6624
6625 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6626 btrfs_dev_stat_reset(dev, i);
6627 }
6628
6629 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6630 {
6631 struct btrfs_key key;
6632 struct btrfs_key found_key;
6633 struct btrfs_root *dev_root = fs_info->dev_root;
6634 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6635 struct extent_buffer *eb;
6636 int slot;
6637 int ret = 0;
6638 struct btrfs_device *device;
6639 struct btrfs_path *path = NULL;
6640 int i;
6641
6642 path = btrfs_alloc_path();
6643 if (!path) {
6644 ret = -ENOMEM;
6645 goto out;
6646 }
6647
6648 mutex_lock(&fs_devices->device_list_mutex);
6649 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6650 int item_size;
6651 struct btrfs_dev_stats_item *ptr;
6652
6653 key.objectid = 0;
6654 key.type = BTRFS_DEV_STATS_KEY;
6655 key.offset = device->devid;
6656 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6657 if (ret) {
6658 __btrfs_reset_dev_stats(device);
6659 device->dev_stats_valid = 1;
6660 btrfs_release_path(path);
6661 continue;
6662 }
6663 slot = path->slots[0];
6664 eb = path->nodes[0];
6665 btrfs_item_key_to_cpu(eb, &found_key, slot);
6666 item_size = btrfs_item_size_nr(eb, slot);
6667
6668 ptr = btrfs_item_ptr(eb, slot,
6669 struct btrfs_dev_stats_item);
6670
6671 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6672 if (item_size >= (1 + i) * sizeof(__le64))
6673 btrfs_dev_stat_set(device, i,
6674 btrfs_dev_stats_value(eb, ptr, i));
6675 else
6676 btrfs_dev_stat_reset(device, i);
6677 }
6678
6679 device->dev_stats_valid = 1;
6680 btrfs_dev_stat_print_on_load(device);
6681 btrfs_release_path(path);
6682 }
6683 mutex_unlock(&fs_devices->device_list_mutex);
6684
6685 out:
6686 btrfs_free_path(path);
6687 return ret < 0 ? ret : 0;
6688 }
6689
6690 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6691 struct btrfs_root *dev_root,
6692 struct btrfs_device *device)
6693 {
6694 struct btrfs_path *path;
6695 struct btrfs_key key;
6696 struct extent_buffer *eb;
6697 struct btrfs_dev_stats_item *ptr;
6698 int ret;
6699 int i;
6700
6701 key.objectid = 0;
6702 key.type = BTRFS_DEV_STATS_KEY;
6703 key.offset = device->devid;
6704
6705 path = btrfs_alloc_path();
6706 BUG_ON(!path);
6707 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6708 if (ret < 0) {
6709 btrfs_warn_in_rcu(dev_root->fs_info,
6710 "error %d while searching for dev_stats item for device %s",
6711 ret, rcu_str_deref(device->name));
6712 goto out;
6713 }
6714
6715 if (ret == 0 &&
6716 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6717 /* need to delete old one and insert a new one */
6718 ret = btrfs_del_item(trans, dev_root, path);
6719 if (ret != 0) {
6720 btrfs_warn_in_rcu(dev_root->fs_info,
6721 "delete too small dev_stats item for device %s failed %d",
6722 rcu_str_deref(device->name), ret);
6723 goto out;
6724 }
6725 ret = 1;
6726 }
6727
6728 if (ret == 1) {
6729 /* need to insert a new item */
6730 btrfs_release_path(path);
6731 ret = btrfs_insert_empty_item(trans, dev_root, path,
6732 &key, sizeof(*ptr));
6733 if (ret < 0) {
6734 btrfs_warn_in_rcu(dev_root->fs_info,
6735 "insert dev_stats item for device %s failed %d",
6736 rcu_str_deref(device->name), ret);
6737 goto out;
6738 }
6739 }
6740
6741 eb = path->nodes[0];
6742 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6743 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6744 btrfs_set_dev_stats_value(eb, ptr, i,
6745 btrfs_dev_stat_read(device, i));
6746 btrfs_mark_buffer_dirty(eb);
6747
6748 out:
6749 btrfs_free_path(path);
6750 return ret;
6751 }
6752
6753 /*
6754 * called from commit_transaction. Writes all changed device stats to disk.
6755 */
6756 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6757 struct btrfs_fs_info *fs_info)
6758 {
6759 struct btrfs_root *dev_root = fs_info->dev_root;
6760 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6761 struct btrfs_device *device;
6762 int stats_cnt;
6763 int ret = 0;
6764
6765 mutex_lock(&fs_devices->device_list_mutex);
6766 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6767 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6768 continue;
6769
6770 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6771 ret = update_dev_stat_item(trans, dev_root, device);
6772 if (!ret)
6773 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6774 }
6775 mutex_unlock(&fs_devices->device_list_mutex);
6776
6777 return ret;
6778 }
6779
6780 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6781 {
6782 btrfs_dev_stat_inc(dev, index);
6783 btrfs_dev_stat_print_on_error(dev);
6784 }
6785
6786 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6787 {
6788 if (!dev->dev_stats_valid)
6789 return;
6790 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6791 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6792 rcu_str_deref(dev->name),
6793 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6794 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6795 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6796 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6797 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6798 }
6799
6800 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6801 {
6802 int i;
6803
6804 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6805 if (btrfs_dev_stat_read(dev, i) != 0)
6806 break;
6807 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6808 return; /* all values == 0, suppress message */
6809
6810 btrfs_info_in_rcu(dev->dev_root->fs_info,
6811 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6812 rcu_str_deref(dev->name),
6813 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6814 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6815 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6816 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6817 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6818 }
6819
6820 int btrfs_get_dev_stats(struct btrfs_root *root,
6821 struct btrfs_ioctl_get_dev_stats *stats)
6822 {
6823 struct btrfs_device *dev;
6824 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6825 int i;
6826
6827 mutex_lock(&fs_devices->device_list_mutex);
6828 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6829 mutex_unlock(&fs_devices->device_list_mutex);
6830
6831 if (!dev) {
6832 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6833 return -ENODEV;
6834 } else if (!dev->dev_stats_valid) {
6835 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6836 return -ENODEV;
6837 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6838 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6839 if (stats->nr_items > i)
6840 stats->values[i] =
6841 btrfs_dev_stat_read_and_reset(dev, i);
6842 else
6843 btrfs_dev_stat_reset(dev, i);
6844 }
6845 } else {
6846 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6847 if (stats->nr_items > i)
6848 stats->values[i] = btrfs_dev_stat_read(dev, i);
6849 }
6850 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6851 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6852 return 0;
6853 }
6854
6855 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6856 {
6857 struct buffer_head *bh;
6858 struct btrfs_super_block *disk_super;
6859 int copy_num;
6860
6861 if (!bdev)
6862 return;
6863
6864 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6865 copy_num++) {
6866
6867 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6868 continue;
6869
6870 disk_super = (struct btrfs_super_block *)bh->b_data;
6871
6872 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6873 set_buffer_dirty(bh);
6874 sync_dirty_buffer(bh);
6875 brelse(bh);
6876 }
6877
6878 /* Notify udev that device has changed */
6879 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6880
6881 /* Update ctime/mtime for device path for libblkid */
6882 update_dev_time(device_path);
6883 }
6884
6885 /*
6886 * Update the size of all devices, which is used for writing out the
6887 * super blocks.
6888 */
6889 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6890 {
6891 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6892 struct btrfs_device *curr, *next;
6893
6894 if (list_empty(&fs_devices->resized_devices))
6895 return;
6896
6897 mutex_lock(&fs_devices->device_list_mutex);
6898 lock_chunks(fs_info->dev_root);
6899 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6900 resized_list) {
6901 list_del_init(&curr->resized_list);
6902 curr->commit_total_bytes = curr->disk_total_bytes;
6903 }
6904 unlock_chunks(fs_info->dev_root);
6905 mutex_unlock(&fs_devices->device_list_mutex);
6906 }
6907
6908 /* Must be invoked during the transaction commit */
6909 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6910 struct btrfs_transaction *transaction)
6911 {
6912 struct extent_map *em;
6913 struct map_lookup *map;
6914 struct btrfs_device *dev;
6915 int i;
6916
6917 if (list_empty(&transaction->pending_chunks))
6918 return;
6919
6920 /* In order to kick the device replace finish process */
6921 lock_chunks(root);
6922 list_for_each_entry(em, &transaction->pending_chunks, list) {
6923 map = em->map_lookup;
6924
6925 for (i = 0; i < map->num_stripes; i++) {
6926 dev = map->stripes[i].dev;
6927 dev->commit_bytes_used = dev->bytes_used;
6928 }
6929 }
6930 unlock_chunks(root);
6931 }
6932
6933 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6934 {
6935 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6936 while (fs_devices) {
6937 fs_devices->fs_info = fs_info;
6938 fs_devices = fs_devices->seed;
6939 }
6940 }
6941
6942 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6943 {
6944 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6945 while (fs_devices) {
6946 fs_devices->fs_info = NULL;
6947 fs_devices = fs_devices->seed;
6948 }
6949 }
6950
6951 void btrfs_close_one_device(struct btrfs_device *device)
6952 {
6953 struct btrfs_fs_devices *fs_devices = device->fs_devices;
6954 struct btrfs_device *new_device;
6955 struct rcu_string *name;
6956
6957 if (device->bdev)
6958 fs_devices->open_devices--;
6959
6960 if (device->writeable &&
6961 device->devid != BTRFS_DEV_REPLACE_DEVID) {
6962 list_del_init(&device->dev_alloc_list);
6963 fs_devices->rw_devices--;
6964 }
6965
6966 if (device->missing)
6967 fs_devices->missing_devices--;
6968
6969 new_device = btrfs_alloc_device(NULL, &device->devid,
6970 device->uuid);
6971 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6972
6973 /* Safe because we are under uuid_mutex */
6974 if (device->name) {
6975 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6976 BUG_ON(!name); /* -ENOMEM */
6977 rcu_assign_pointer(new_device->name, name);
6978 }
6979
6980 list_replace_rcu(&device->dev_list, &new_device->dev_list);
6981 new_device->fs_devices = device->fs_devices;
6982
6983 call_rcu(&device->rcu, free_device);
6984 }
Ubuntu Jammy Linux kernel mirror