]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/volumes.c
Merge remote-tracking branch 'mkp-scsi/4.7/scsi-fixes' into fixes
[mirror_ubuntu-bionic-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/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.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 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 /*
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
125 */
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134 };
135
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143 static void btrfs_close_one_device(struct btrfs_device *device);
144
145 DEFINE_MUTEX(uuid_mutex);
146 static LIST_HEAD(fs_uuids);
147 struct list_head *btrfs_get_fs_uuids(void)
148 {
149 return &fs_uuids;
150 }
151
152 static struct btrfs_fs_devices *__alloc_fs_devices(void)
153 {
154 struct btrfs_fs_devices *fs_devs;
155
156 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
157 if (!fs_devs)
158 return ERR_PTR(-ENOMEM);
159
160 mutex_init(&fs_devs->device_list_mutex);
161
162 INIT_LIST_HEAD(&fs_devs->devices);
163 INIT_LIST_HEAD(&fs_devs->resized_devices);
164 INIT_LIST_HEAD(&fs_devs->alloc_list);
165 INIT_LIST_HEAD(&fs_devs->list);
166
167 return fs_devs;
168 }
169
170 /**
171 * alloc_fs_devices - allocate struct btrfs_fs_devices
172 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
173 * generated.
174 *
175 * Return: a pointer to a new &struct btrfs_fs_devices on success;
176 * ERR_PTR() on error. Returned struct is not linked onto any lists and
177 * can be destroyed with kfree() right away.
178 */
179 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
180 {
181 struct btrfs_fs_devices *fs_devs;
182
183 fs_devs = __alloc_fs_devices();
184 if (IS_ERR(fs_devs))
185 return fs_devs;
186
187 if (fsid)
188 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
189 else
190 generate_random_uuid(fs_devs->fsid);
191
192 return fs_devs;
193 }
194
195 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
196 {
197 struct btrfs_device *device;
198 WARN_ON(fs_devices->opened);
199 while (!list_empty(&fs_devices->devices)) {
200 device = list_entry(fs_devices->devices.next,
201 struct btrfs_device, dev_list);
202 list_del(&device->dev_list);
203 rcu_string_free(device->name);
204 kfree(device);
205 }
206 kfree(fs_devices);
207 }
208
209 static void btrfs_kobject_uevent(struct block_device *bdev,
210 enum kobject_action action)
211 {
212 int ret;
213
214 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
215 if (ret)
216 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
217 action,
218 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
219 &disk_to_dev(bdev->bd_disk)->kobj);
220 }
221
222 void btrfs_cleanup_fs_uuids(void)
223 {
224 struct btrfs_fs_devices *fs_devices;
225
226 while (!list_empty(&fs_uuids)) {
227 fs_devices = list_entry(fs_uuids.next,
228 struct btrfs_fs_devices, list);
229 list_del(&fs_devices->list);
230 free_fs_devices(fs_devices);
231 }
232 }
233
234 static struct btrfs_device *__alloc_device(void)
235 {
236 struct btrfs_device *dev;
237
238 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
239 if (!dev)
240 return ERR_PTR(-ENOMEM);
241
242 INIT_LIST_HEAD(&dev->dev_list);
243 INIT_LIST_HEAD(&dev->dev_alloc_list);
244 INIT_LIST_HEAD(&dev->resized_list);
245
246 spin_lock_init(&dev->io_lock);
247
248 spin_lock_init(&dev->reada_lock);
249 atomic_set(&dev->reada_in_flight, 0);
250 atomic_set(&dev->dev_stats_ccnt, 0);
251 btrfs_device_data_ordered_init(dev);
252 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
254
255 return dev;
256 }
257
258 static noinline struct btrfs_device *__find_device(struct list_head *head,
259 u64 devid, u8 *uuid)
260 {
261 struct btrfs_device *dev;
262
263 list_for_each_entry(dev, head, dev_list) {
264 if (dev->devid == devid &&
265 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
266 return dev;
267 }
268 }
269 return NULL;
270 }
271
272 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
273 {
274 struct btrfs_fs_devices *fs_devices;
275
276 list_for_each_entry(fs_devices, &fs_uuids, list) {
277 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
278 return fs_devices;
279 }
280 return NULL;
281 }
282
283 static int
284 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
285 int flush, struct block_device **bdev,
286 struct buffer_head **bh)
287 {
288 int ret;
289
290 *bdev = blkdev_get_by_path(device_path, flags, holder);
291
292 if (IS_ERR(*bdev)) {
293 ret = PTR_ERR(*bdev);
294 goto error;
295 }
296
297 if (flush)
298 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
299 ret = set_blocksize(*bdev, 4096);
300 if (ret) {
301 blkdev_put(*bdev, flags);
302 goto error;
303 }
304 invalidate_bdev(*bdev);
305 *bh = btrfs_read_dev_super(*bdev);
306 if (IS_ERR(*bh)) {
307 ret = PTR_ERR(*bh);
308 blkdev_put(*bdev, flags);
309 goto error;
310 }
311
312 return 0;
313
314 error:
315 *bdev = NULL;
316 *bh = NULL;
317 return ret;
318 }
319
320 static void requeue_list(struct btrfs_pending_bios *pending_bios,
321 struct bio *head, struct bio *tail)
322 {
323
324 struct bio *old_head;
325
326 old_head = pending_bios->head;
327 pending_bios->head = head;
328 if (pending_bios->tail)
329 tail->bi_next = old_head;
330 else
331 pending_bios->tail = tail;
332 }
333
334 /*
335 * we try to collect pending bios for a device so we don't get a large
336 * number of procs sending bios down to the same device. This greatly
337 * improves the schedulers ability to collect and merge the bios.
338 *
339 * But, it also turns into a long list of bios to process and that is sure
340 * to eventually make the worker thread block. The solution here is to
341 * make some progress and then put this work struct back at the end of
342 * the list if the block device is congested. This way, multiple devices
343 * can make progress from a single worker thread.
344 */
345 static noinline void run_scheduled_bios(struct btrfs_device *device)
346 {
347 struct bio *pending;
348 struct backing_dev_info *bdi;
349 struct btrfs_fs_info *fs_info;
350 struct btrfs_pending_bios *pending_bios;
351 struct bio *tail;
352 struct bio *cur;
353 int again = 0;
354 unsigned long num_run;
355 unsigned long batch_run = 0;
356 unsigned long limit;
357 unsigned long last_waited = 0;
358 int force_reg = 0;
359 int sync_pending = 0;
360 struct blk_plug plug;
361
362 /*
363 * this function runs all the bios we've collected for
364 * a particular device. We don't want to wander off to
365 * another device without first sending all of these down.
366 * So, setup a plug here and finish it off before we return
367 */
368 blk_start_plug(&plug);
369
370 bdi = blk_get_backing_dev_info(device->bdev);
371 fs_info = device->dev_root->fs_info;
372 limit = btrfs_async_submit_limit(fs_info);
373 limit = limit * 2 / 3;
374
375 loop:
376 spin_lock(&device->io_lock);
377
378 loop_lock:
379 num_run = 0;
380
381 /* take all the bios off the list at once and process them
382 * later on (without the lock held). But, remember the
383 * tail and other pointers so the bios can be properly reinserted
384 * into the list if we hit congestion
385 */
386 if (!force_reg && device->pending_sync_bios.head) {
387 pending_bios = &device->pending_sync_bios;
388 force_reg = 1;
389 } else {
390 pending_bios = &device->pending_bios;
391 force_reg = 0;
392 }
393
394 pending = pending_bios->head;
395 tail = pending_bios->tail;
396 WARN_ON(pending && !tail);
397
398 /*
399 * if pending was null this time around, no bios need processing
400 * at all and we can stop. Otherwise it'll loop back up again
401 * and do an additional check so no bios are missed.
402 *
403 * device->running_pending is used to synchronize with the
404 * schedule_bio code.
405 */
406 if (device->pending_sync_bios.head == NULL &&
407 device->pending_bios.head == NULL) {
408 again = 0;
409 device->running_pending = 0;
410 } else {
411 again = 1;
412 device->running_pending = 1;
413 }
414
415 pending_bios->head = NULL;
416 pending_bios->tail = NULL;
417
418 spin_unlock(&device->io_lock);
419
420 while (pending) {
421
422 rmb();
423 /* we want to work on both lists, but do more bios on the
424 * sync list than the regular list
425 */
426 if ((num_run > 32 &&
427 pending_bios != &device->pending_sync_bios &&
428 device->pending_sync_bios.head) ||
429 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
430 device->pending_bios.head)) {
431 spin_lock(&device->io_lock);
432 requeue_list(pending_bios, pending, tail);
433 goto loop_lock;
434 }
435
436 cur = pending;
437 pending = pending->bi_next;
438 cur->bi_next = NULL;
439
440 /*
441 * atomic_dec_return implies a barrier for waitqueue_active
442 */
443 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
444 waitqueue_active(&fs_info->async_submit_wait))
445 wake_up(&fs_info->async_submit_wait);
446
447 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
448
449 /*
450 * if we're doing the sync list, record that our
451 * plug has some sync requests on it
452 *
453 * If we're doing the regular list and there are
454 * sync requests sitting around, unplug before
455 * we add more
456 */
457 if (pending_bios == &device->pending_sync_bios) {
458 sync_pending = 1;
459 } else if (sync_pending) {
460 blk_finish_plug(&plug);
461 blk_start_plug(&plug);
462 sync_pending = 0;
463 }
464
465 btrfsic_submit_bio(cur->bi_rw, cur);
466 num_run++;
467 batch_run++;
468
469 cond_resched();
470
471 /*
472 * we made progress, there is more work to do and the bdi
473 * is now congested. Back off and let other work structs
474 * run instead
475 */
476 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
477 fs_info->fs_devices->open_devices > 1) {
478 struct io_context *ioc;
479
480 ioc = current->io_context;
481
482 /*
483 * the main goal here is that we don't want to
484 * block if we're going to be able to submit
485 * more requests without blocking.
486 *
487 * This code does two great things, it pokes into
488 * the elevator code from a filesystem _and_
489 * it makes assumptions about how batching works.
490 */
491 if (ioc && ioc->nr_batch_requests > 0 &&
492 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
493 (last_waited == 0 ||
494 ioc->last_waited == last_waited)) {
495 /*
496 * we want to go through our batch of
497 * requests and stop. So, we copy out
498 * the ioc->last_waited time and test
499 * against it before looping
500 */
501 last_waited = ioc->last_waited;
502 cond_resched();
503 continue;
504 }
505 spin_lock(&device->io_lock);
506 requeue_list(pending_bios, pending, tail);
507 device->running_pending = 1;
508
509 spin_unlock(&device->io_lock);
510 btrfs_queue_work(fs_info->submit_workers,
511 &device->work);
512 goto done;
513 }
514 /* unplug every 64 requests just for good measure */
515 if (batch_run % 64 == 0) {
516 blk_finish_plug(&plug);
517 blk_start_plug(&plug);
518 sync_pending = 0;
519 }
520 }
521
522 cond_resched();
523 if (again)
524 goto loop;
525
526 spin_lock(&device->io_lock);
527 if (device->pending_bios.head || device->pending_sync_bios.head)
528 goto loop_lock;
529 spin_unlock(&device->io_lock);
530
531 done:
532 blk_finish_plug(&plug);
533 }
534
535 static void pending_bios_fn(struct btrfs_work *work)
536 {
537 struct btrfs_device *device;
538
539 device = container_of(work, struct btrfs_device, work);
540 run_scheduled_bios(device);
541 }
542
543
544 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
545 {
546 struct btrfs_fs_devices *fs_devs;
547 struct btrfs_device *dev;
548
549 if (!cur_dev->name)
550 return;
551
552 list_for_each_entry(fs_devs, &fs_uuids, list) {
553 int del = 1;
554
555 if (fs_devs->opened)
556 continue;
557 if (fs_devs->seeding)
558 continue;
559
560 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
561
562 if (dev == cur_dev)
563 continue;
564 if (!dev->name)
565 continue;
566
567 /*
568 * Todo: This won't be enough. What if the same device
569 * comes back (with new uuid and) with its mapper path?
570 * But for now, this does help as mostly an admin will
571 * either use mapper or non mapper path throughout.
572 */
573 rcu_read_lock();
574 del = strcmp(rcu_str_deref(dev->name),
575 rcu_str_deref(cur_dev->name));
576 rcu_read_unlock();
577 if (!del)
578 break;
579 }
580
581 if (!del) {
582 /* delete the stale device */
583 if (fs_devs->num_devices == 1) {
584 btrfs_sysfs_remove_fsid(fs_devs);
585 list_del(&fs_devs->list);
586 free_fs_devices(fs_devs);
587 } else {
588 fs_devs->num_devices--;
589 list_del(&dev->dev_list);
590 rcu_string_free(dev->name);
591 kfree(dev);
592 }
593 break;
594 }
595 }
596 }
597
598 /*
599 * Add new device to list of registered devices
600 *
601 * Returns:
602 * 1 - first time device is seen
603 * 0 - device already known
604 * < 0 - error
605 */
606 static noinline int device_list_add(const char *path,
607 struct btrfs_super_block *disk_super,
608 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
609 {
610 struct btrfs_device *device;
611 struct btrfs_fs_devices *fs_devices;
612 struct rcu_string *name;
613 int ret = 0;
614 u64 found_transid = btrfs_super_generation(disk_super);
615
616 fs_devices = find_fsid(disk_super->fsid);
617 if (!fs_devices) {
618 fs_devices = alloc_fs_devices(disk_super->fsid);
619 if (IS_ERR(fs_devices))
620 return PTR_ERR(fs_devices);
621
622 list_add(&fs_devices->list, &fs_uuids);
623
624 device = NULL;
625 } else {
626 device = __find_device(&fs_devices->devices, devid,
627 disk_super->dev_item.uuid);
628 }
629
630 if (!device) {
631 if (fs_devices->opened)
632 return -EBUSY;
633
634 device = btrfs_alloc_device(NULL, &devid,
635 disk_super->dev_item.uuid);
636 if (IS_ERR(device)) {
637 /* we can safely leave the fs_devices entry around */
638 return PTR_ERR(device);
639 }
640
641 name = rcu_string_strdup(path, GFP_NOFS);
642 if (!name) {
643 kfree(device);
644 return -ENOMEM;
645 }
646 rcu_assign_pointer(device->name, name);
647
648 mutex_lock(&fs_devices->device_list_mutex);
649 list_add_rcu(&device->dev_list, &fs_devices->devices);
650 fs_devices->num_devices++;
651 mutex_unlock(&fs_devices->device_list_mutex);
652
653 ret = 1;
654 device->fs_devices = fs_devices;
655 } else if (!device->name || strcmp(device->name->str, path)) {
656 /*
657 * When FS is already mounted.
658 * 1. If you are here and if the device->name is NULL that
659 * means this device was missing at time of FS mount.
660 * 2. If you are here and if the device->name is different
661 * from 'path' that means either
662 * a. The same device disappeared and reappeared with
663 * different name. or
664 * b. The missing-disk-which-was-replaced, has
665 * reappeared now.
666 *
667 * We must allow 1 and 2a above. But 2b would be a spurious
668 * and unintentional.
669 *
670 * Further in case of 1 and 2a above, the disk at 'path'
671 * would have missed some transaction when it was away and
672 * in case of 2a the stale bdev has to be updated as well.
673 * 2b must not be allowed at all time.
674 */
675
676 /*
677 * For now, we do allow update to btrfs_fs_device through the
678 * btrfs dev scan cli after FS has been mounted. We're still
679 * tracking a problem where systems fail mount by subvolume id
680 * when we reject replacement on a mounted FS.
681 */
682 if (!fs_devices->opened && found_transid < device->generation) {
683 /*
684 * That is if the FS is _not_ mounted and if you
685 * are here, that means there is more than one
686 * disk with same uuid and devid.We keep the one
687 * with larger generation number or the last-in if
688 * generation are equal.
689 */
690 return -EEXIST;
691 }
692
693 name = rcu_string_strdup(path, GFP_NOFS);
694 if (!name)
695 return -ENOMEM;
696 rcu_string_free(device->name);
697 rcu_assign_pointer(device->name, name);
698 if (device->missing) {
699 fs_devices->missing_devices--;
700 device->missing = 0;
701 }
702 }
703
704 /*
705 * Unmount does not free the btrfs_device struct but would zero
706 * generation along with most of the other members. So just update
707 * it back. We need it to pick the disk with largest generation
708 * (as above).
709 */
710 if (!fs_devices->opened)
711 device->generation = found_transid;
712
713 /*
714 * if there is new btrfs on an already registered device,
715 * then remove the stale device entry.
716 */
717 if (ret > 0)
718 btrfs_free_stale_device(device);
719
720 *fs_devices_ret = fs_devices;
721
722 return ret;
723 }
724
725 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
726 {
727 struct btrfs_fs_devices *fs_devices;
728 struct btrfs_device *device;
729 struct btrfs_device *orig_dev;
730
731 fs_devices = alloc_fs_devices(orig->fsid);
732 if (IS_ERR(fs_devices))
733 return fs_devices;
734
735 mutex_lock(&orig->device_list_mutex);
736 fs_devices->total_devices = orig->total_devices;
737
738 /* We have held the volume lock, it is safe to get the devices. */
739 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
740 struct rcu_string *name;
741
742 device = btrfs_alloc_device(NULL, &orig_dev->devid,
743 orig_dev->uuid);
744 if (IS_ERR(device))
745 goto error;
746
747 /*
748 * This is ok to do without rcu read locked because we hold the
749 * uuid mutex so nothing we touch in here is going to disappear.
750 */
751 if (orig_dev->name) {
752 name = rcu_string_strdup(orig_dev->name->str,
753 GFP_KERNEL);
754 if (!name) {
755 kfree(device);
756 goto error;
757 }
758 rcu_assign_pointer(device->name, name);
759 }
760
761 list_add(&device->dev_list, &fs_devices->devices);
762 device->fs_devices = fs_devices;
763 fs_devices->num_devices++;
764 }
765 mutex_unlock(&orig->device_list_mutex);
766 return fs_devices;
767 error:
768 mutex_unlock(&orig->device_list_mutex);
769 free_fs_devices(fs_devices);
770 return ERR_PTR(-ENOMEM);
771 }
772
773 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
774 {
775 struct btrfs_device *device, *next;
776 struct btrfs_device *latest_dev = NULL;
777
778 mutex_lock(&uuid_mutex);
779 again:
780 /* This is the initialized path, it is safe to release the devices. */
781 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
782 if (device->in_fs_metadata) {
783 if (!device->is_tgtdev_for_dev_replace &&
784 (!latest_dev ||
785 device->generation > latest_dev->generation)) {
786 latest_dev = device;
787 }
788 continue;
789 }
790
791 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
792 /*
793 * In the first step, keep the device which has
794 * the correct fsid and the devid that is used
795 * for the dev_replace procedure.
796 * In the second step, the dev_replace state is
797 * read from the device tree and it is known
798 * whether the procedure is really active or
799 * not, which means whether this device is
800 * used or whether it should be removed.
801 */
802 if (step == 0 || device->is_tgtdev_for_dev_replace) {
803 continue;
804 }
805 }
806 if (device->bdev) {
807 blkdev_put(device->bdev, device->mode);
808 device->bdev = NULL;
809 fs_devices->open_devices--;
810 }
811 if (device->writeable) {
812 list_del_init(&device->dev_alloc_list);
813 device->writeable = 0;
814 if (!device->is_tgtdev_for_dev_replace)
815 fs_devices->rw_devices--;
816 }
817 list_del_init(&device->dev_list);
818 fs_devices->num_devices--;
819 rcu_string_free(device->name);
820 kfree(device);
821 }
822
823 if (fs_devices->seed) {
824 fs_devices = fs_devices->seed;
825 goto again;
826 }
827
828 fs_devices->latest_bdev = latest_dev->bdev;
829
830 mutex_unlock(&uuid_mutex);
831 }
832
833 static void __free_device(struct work_struct *work)
834 {
835 struct btrfs_device *device;
836
837 device = container_of(work, struct btrfs_device, rcu_work);
838
839 if (device->bdev)
840 blkdev_put(device->bdev, device->mode);
841
842 rcu_string_free(device->name);
843 kfree(device);
844 }
845
846 static void free_device(struct rcu_head *head)
847 {
848 struct btrfs_device *device;
849
850 device = container_of(head, struct btrfs_device, rcu);
851
852 INIT_WORK(&device->rcu_work, __free_device);
853 schedule_work(&device->rcu_work);
854 }
855
856 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
857 {
858 struct btrfs_device *device, *tmp;
859
860 if (--fs_devices->opened > 0)
861 return 0;
862
863 mutex_lock(&fs_devices->device_list_mutex);
864 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
865 btrfs_close_one_device(device);
866 }
867 mutex_unlock(&fs_devices->device_list_mutex);
868
869 WARN_ON(fs_devices->open_devices);
870 WARN_ON(fs_devices->rw_devices);
871 fs_devices->opened = 0;
872 fs_devices->seeding = 0;
873
874 return 0;
875 }
876
877 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
878 {
879 struct btrfs_fs_devices *seed_devices = NULL;
880 int ret;
881
882 mutex_lock(&uuid_mutex);
883 ret = __btrfs_close_devices(fs_devices);
884 if (!fs_devices->opened) {
885 seed_devices = fs_devices->seed;
886 fs_devices->seed = NULL;
887 }
888 mutex_unlock(&uuid_mutex);
889
890 while (seed_devices) {
891 fs_devices = seed_devices;
892 seed_devices = fs_devices->seed;
893 __btrfs_close_devices(fs_devices);
894 free_fs_devices(fs_devices);
895 }
896 /*
897 * Wait for rcu kworkers under __btrfs_close_devices
898 * to finish all blkdev_puts so device is really
899 * free when umount is done.
900 */
901 rcu_barrier();
902 return ret;
903 }
904
905 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
906 fmode_t flags, void *holder)
907 {
908 struct request_queue *q;
909 struct block_device *bdev;
910 struct list_head *head = &fs_devices->devices;
911 struct btrfs_device *device;
912 struct btrfs_device *latest_dev = NULL;
913 struct buffer_head *bh;
914 struct btrfs_super_block *disk_super;
915 u64 devid;
916 int seeding = 1;
917 int ret = 0;
918
919 flags |= FMODE_EXCL;
920
921 list_for_each_entry(device, head, dev_list) {
922 if (device->bdev)
923 continue;
924 if (!device->name)
925 continue;
926
927 /* Just open everything we can; ignore failures here */
928 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
929 &bdev, &bh))
930 continue;
931
932 disk_super = (struct btrfs_super_block *)bh->b_data;
933 devid = btrfs_stack_device_id(&disk_super->dev_item);
934 if (devid != device->devid)
935 goto error_brelse;
936
937 if (memcmp(device->uuid, disk_super->dev_item.uuid,
938 BTRFS_UUID_SIZE))
939 goto error_brelse;
940
941 device->generation = btrfs_super_generation(disk_super);
942 if (!latest_dev ||
943 device->generation > latest_dev->generation)
944 latest_dev = device;
945
946 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
947 device->writeable = 0;
948 } else {
949 device->writeable = !bdev_read_only(bdev);
950 seeding = 0;
951 }
952
953 q = bdev_get_queue(bdev);
954 if (blk_queue_discard(q))
955 device->can_discard = 1;
956
957 device->bdev = bdev;
958 device->in_fs_metadata = 0;
959 device->mode = flags;
960
961 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
962 fs_devices->rotating = 1;
963
964 fs_devices->open_devices++;
965 if (device->writeable &&
966 device->devid != BTRFS_DEV_REPLACE_DEVID) {
967 fs_devices->rw_devices++;
968 list_add(&device->dev_alloc_list,
969 &fs_devices->alloc_list);
970 }
971 brelse(bh);
972 continue;
973
974 error_brelse:
975 brelse(bh);
976 blkdev_put(bdev, flags);
977 continue;
978 }
979 if (fs_devices->open_devices == 0) {
980 ret = -EINVAL;
981 goto out;
982 }
983 fs_devices->seeding = seeding;
984 fs_devices->opened = 1;
985 fs_devices->latest_bdev = latest_dev->bdev;
986 fs_devices->total_rw_bytes = 0;
987 out:
988 return ret;
989 }
990
991 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
992 fmode_t flags, void *holder)
993 {
994 int ret;
995
996 mutex_lock(&uuid_mutex);
997 if (fs_devices->opened) {
998 fs_devices->opened++;
999 ret = 0;
1000 } else {
1001 ret = __btrfs_open_devices(fs_devices, flags, holder);
1002 }
1003 mutex_unlock(&uuid_mutex);
1004 return ret;
1005 }
1006
1007 void btrfs_release_disk_super(struct page *page)
1008 {
1009 kunmap(page);
1010 put_page(page);
1011 }
1012
1013 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1014 struct page **page, struct btrfs_super_block **disk_super)
1015 {
1016 void *p;
1017 pgoff_t index;
1018
1019 /* make sure our super fits in the device */
1020 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1021 return 1;
1022
1023 /* make sure our super fits in the page */
1024 if (sizeof(**disk_super) > PAGE_SIZE)
1025 return 1;
1026
1027 /* make sure our super doesn't straddle pages on disk */
1028 index = bytenr >> PAGE_SHIFT;
1029 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1030 return 1;
1031
1032 /* pull in the page with our super */
1033 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1034 index, GFP_KERNEL);
1035
1036 if (IS_ERR_OR_NULL(*page))
1037 return 1;
1038
1039 p = kmap(*page);
1040
1041 /* align our pointer to the offset of the super block */
1042 *disk_super = p + (bytenr & ~PAGE_MASK);
1043
1044 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1045 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1046 btrfs_release_disk_super(*page);
1047 return 1;
1048 }
1049
1050 if ((*disk_super)->label[0] &&
1051 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1052 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1053
1054 return 0;
1055 }
1056
1057 /*
1058 * Look for a btrfs signature on a device. This may be called out of the mount path
1059 * and we are not allowed to call set_blocksize during the scan. The superblock
1060 * is read via pagecache
1061 */
1062 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1063 struct btrfs_fs_devices **fs_devices_ret)
1064 {
1065 struct btrfs_super_block *disk_super;
1066 struct block_device *bdev;
1067 struct page *page;
1068 int ret = -EINVAL;
1069 u64 devid;
1070 u64 transid;
1071 u64 total_devices;
1072 u64 bytenr;
1073
1074 /*
1075 * we would like to check all the supers, but that would make
1076 * a btrfs mount succeed after a mkfs from a different FS.
1077 * So, we need to add a special mount option to scan for
1078 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1079 */
1080 bytenr = btrfs_sb_offset(0);
1081 flags |= FMODE_EXCL;
1082 mutex_lock(&uuid_mutex);
1083
1084 bdev = blkdev_get_by_path(path, flags, holder);
1085 if (IS_ERR(bdev)) {
1086 ret = PTR_ERR(bdev);
1087 goto error;
1088 }
1089
1090 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1091 goto error_bdev_put;
1092
1093 devid = btrfs_stack_device_id(&disk_super->dev_item);
1094 transid = btrfs_super_generation(disk_super);
1095 total_devices = btrfs_super_num_devices(disk_super);
1096
1097 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1098 if (ret > 0) {
1099 if (disk_super->label[0]) {
1100 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1101 } else {
1102 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1103 }
1104
1105 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1106 ret = 0;
1107 }
1108 if (!ret && fs_devices_ret)
1109 (*fs_devices_ret)->total_devices = total_devices;
1110
1111 btrfs_release_disk_super(page);
1112
1113 error_bdev_put:
1114 blkdev_put(bdev, flags);
1115 error:
1116 mutex_unlock(&uuid_mutex);
1117 return ret;
1118 }
1119
1120 /* helper to account the used device space in the range */
1121 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1122 u64 end, u64 *length)
1123 {
1124 struct btrfs_key key;
1125 struct btrfs_root *root = device->dev_root;
1126 struct btrfs_dev_extent *dev_extent;
1127 struct btrfs_path *path;
1128 u64 extent_end;
1129 int ret;
1130 int slot;
1131 struct extent_buffer *l;
1132
1133 *length = 0;
1134
1135 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1136 return 0;
1137
1138 path = btrfs_alloc_path();
1139 if (!path)
1140 return -ENOMEM;
1141 path->reada = READA_FORWARD;
1142
1143 key.objectid = device->devid;
1144 key.offset = start;
1145 key.type = BTRFS_DEV_EXTENT_KEY;
1146
1147 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1148 if (ret < 0)
1149 goto out;
1150 if (ret > 0) {
1151 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1152 if (ret < 0)
1153 goto out;
1154 }
1155
1156 while (1) {
1157 l = path->nodes[0];
1158 slot = path->slots[0];
1159 if (slot >= btrfs_header_nritems(l)) {
1160 ret = btrfs_next_leaf(root, path);
1161 if (ret == 0)
1162 continue;
1163 if (ret < 0)
1164 goto out;
1165
1166 break;
1167 }
1168 btrfs_item_key_to_cpu(l, &key, slot);
1169
1170 if (key.objectid < device->devid)
1171 goto next;
1172
1173 if (key.objectid > device->devid)
1174 break;
1175
1176 if (key.type != BTRFS_DEV_EXTENT_KEY)
1177 goto next;
1178
1179 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180 extent_end = key.offset + btrfs_dev_extent_length(l,
1181 dev_extent);
1182 if (key.offset <= start && extent_end > end) {
1183 *length = end - start + 1;
1184 break;
1185 } else if (key.offset <= start && extent_end > start)
1186 *length += extent_end - start;
1187 else if (key.offset > start && extent_end <= end)
1188 *length += extent_end - key.offset;
1189 else if (key.offset > start && key.offset <= end) {
1190 *length += end - key.offset + 1;
1191 break;
1192 } else if (key.offset > end)
1193 break;
1194
1195 next:
1196 path->slots[0]++;
1197 }
1198 ret = 0;
1199 out:
1200 btrfs_free_path(path);
1201 return ret;
1202 }
1203
1204 static int contains_pending_extent(struct btrfs_transaction *transaction,
1205 struct btrfs_device *device,
1206 u64 *start, u64 len)
1207 {
1208 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1209 struct extent_map *em;
1210 struct list_head *search_list = &fs_info->pinned_chunks;
1211 int ret = 0;
1212 u64 physical_start = *start;
1213
1214 if (transaction)
1215 search_list = &transaction->pending_chunks;
1216 again:
1217 list_for_each_entry(em, search_list, list) {
1218 struct map_lookup *map;
1219 int i;
1220
1221 map = em->map_lookup;
1222 for (i = 0; i < map->num_stripes; i++) {
1223 u64 end;
1224
1225 if (map->stripes[i].dev != device)
1226 continue;
1227 if (map->stripes[i].physical >= physical_start + len ||
1228 map->stripes[i].physical + em->orig_block_len <=
1229 physical_start)
1230 continue;
1231 /*
1232 * Make sure that while processing the pinned list we do
1233 * not override our *start with a lower value, because
1234 * we can have pinned chunks that fall within this
1235 * device hole and that have lower physical addresses
1236 * than the pending chunks we processed before. If we
1237 * do not take this special care we can end up getting
1238 * 2 pending chunks that start at the same physical
1239 * device offsets because the end offset of a pinned
1240 * chunk can be equal to the start offset of some
1241 * pending chunk.
1242 */
1243 end = map->stripes[i].physical + em->orig_block_len;
1244 if (end > *start) {
1245 *start = end;
1246 ret = 1;
1247 }
1248 }
1249 }
1250 if (search_list != &fs_info->pinned_chunks) {
1251 search_list = &fs_info->pinned_chunks;
1252 goto again;
1253 }
1254
1255 return ret;
1256 }
1257
1258
1259 /*
1260 * find_free_dev_extent_start - find free space in the specified device
1261 * @device: the device which we search the free space in
1262 * @num_bytes: the size of the free space that we need
1263 * @search_start: the position from which to begin the search
1264 * @start: store the start of the free space.
1265 * @len: the size of the free space. that we find, or the size
1266 * of the max free space if we don't find suitable free space
1267 *
1268 * this uses a pretty simple search, the expectation is that it is
1269 * called very infrequently and that a given device has a small number
1270 * of extents
1271 *
1272 * @start is used to store the start of the free space if we find. But if we
1273 * don't find suitable free space, it will be used to store the start position
1274 * of the max free space.
1275 *
1276 * @len is used to store the size of the free space that we find.
1277 * But if we don't find suitable free space, it is used to store the size of
1278 * the max free space.
1279 */
1280 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1281 struct btrfs_device *device, u64 num_bytes,
1282 u64 search_start, u64 *start, u64 *len)
1283 {
1284 struct btrfs_key key;
1285 struct btrfs_root *root = device->dev_root;
1286 struct btrfs_dev_extent *dev_extent;
1287 struct btrfs_path *path;
1288 u64 hole_size;
1289 u64 max_hole_start;
1290 u64 max_hole_size;
1291 u64 extent_end;
1292 u64 search_end = device->total_bytes;
1293 int ret;
1294 int slot;
1295 struct extent_buffer *l;
1296 u64 min_search_start;
1297
1298 /*
1299 * We don't want to overwrite the superblock on the drive nor any area
1300 * used by the boot loader (grub for example), so we make sure to start
1301 * at an offset of at least 1MB.
1302 */
1303 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1304 search_start = max(search_start, min_search_start);
1305
1306 path = btrfs_alloc_path();
1307 if (!path)
1308 return -ENOMEM;
1309
1310 max_hole_start = search_start;
1311 max_hole_size = 0;
1312
1313 again:
1314 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1315 ret = -ENOSPC;
1316 goto out;
1317 }
1318
1319 path->reada = READA_FORWARD;
1320 path->search_commit_root = 1;
1321 path->skip_locking = 1;
1322
1323 key.objectid = device->devid;
1324 key.offset = search_start;
1325 key.type = BTRFS_DEV_EXTENT_KEY;
1326
1327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1328 if (ret < 0)
1329 goto out;
1330 if (ret > 0) {
1331 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1332 if (ret < 0)
1333 goto out;
1334 }
1335
1336 while (1) {
1337 l = path->nodes[0];
1338 slot = path->slots[0];
1339 if (slot >= btrfs_header_nritems(l)) {
1340 ret = btrfs_next_leaf(root, path);
1341 if (ret == 0)
1342 continue;
1343 if (ret < 0)
1344 goto out;
1345
1346 break;
1347 }
1348 btrfs_item_key_to_cpu(l, &key, slot);
1349
1350 if (key.objectid < device->devid)
1351 goto next;
1352
1353 if (key.objectid > device->devid)
1354 break;
1355
1356 if (key.type != BTRFS_DEV_EXTENT_KEY)
1357 goto next;
1358
1359 if (key.offset > search_start) {
1360 hole_size = key.offset - search_start;
1361
1362 /*
1363 * Have to check before we set max_hole_start, otherwise
1364 * we could end up sending back this offset anyway.
1365 */
1366 if (contains_pending_extent(transaction, device,
1367 &search_start,
1368 hole_size)) {
1369 if (key.offset >= search_start) {
1370 hole_size = key.offset - search_start;
1371 } else {
1372 WARN_ON_ONCE(1);
1373 hole_size = 0;
1374 }
1375 }
1376
1377 if (hole_size > max_hole_size) {
1378 max_hole_start = search_start;
1379 max_hole_size = hole_size;
1380 }
1381
1382 /*
1383 * If this free space is greater than which we need,
1384 * it must be the max free space that we have found
1385 * until now, so max_hole_start must point to the start
1386 * of this free space and the length of this free space
1387 * is stored in max_hole_size. Thus, we return
1388 * max_hole_start and max_hole_size and go back to the
1389 * caller.
1390 */
1391 if (hole_size >= num_bytes) {
1392 ret = 0;
1393 goto out;
1394 }
1395 }
1396
1397 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1398 extent_end = key.offset + btrfs_dev_extent_length(l,
1399 dev_extent);
1400 if (extent_end > search_start)
1401 search_start = extent_end;
1402 next:
1403 path->slots[0]++;
1404 cond_resched();
1405 }
1406
1407 /*
1408 * At this point, search_start should be the end of
1409 * allocated dev extents, and when shrinking the device,
1410 * search_end may be smaller than search_start.
1411 */
1412 if (search_end > search_start) {
1413 hole_size = search_end - search_start;
1414
1415 if (contains_pending_extent(transaction, device, &search_start,
1416 hole_size)) {
1417 btrfs_release_path(path);
1418 goto again;
1419 }
1420
1421 if (hole_size > max_hole_size) {
1422 max_hole_start = search_start;
1423 max_hole_size = hole_size;
1424 }
1425 }
1426
1427 /* See above. */
1428 if (max_hole_size < num_bytes)
1429 ret = -ENOSPC;
1430 else
1431 ret = 0;
1432
1433 out:
1434 btrfs_free_path(path);
1435 *start = max_hole_start;
1436 if (len)
1437 *len = max_hole_size;
1438 return ret;
1439 }
1440
1441 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1442 struct btrfs_device *device, u64 num_bytes,
1443 u64 *start, u64 *len)
1444 {
1445 /* FIXME use last free of some kind */
1446 return find_free_dev_extent_start(trans->transaction, device,
1447 num_bytes, 0, start, len);
1448 }
1449
1450 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1451 struct btrfs_device *device,
1452 u64 start, u64 *dev_extent_len)
1453 {
1454 int ret;
1455 struct btrfs_path *path;
1456 struct btrfs_root *root = device->dev_root;
1457 struct btrfs_key key;
1458 struct btrfs_key found_key;
1459 struct extent_buffer *leaf = NULL;
1460 struct btrfs_dev_extent *extent = NULL;
1461
1462 path = btrfs_alloc_path();
1463 if (!path)
1464 return -ENOMEM;
1465
1466 key.objectid = device->devid;
1467 key.offset = start;
1468 key.type = BTRFS_DEV_EXTENT_KEY;
1469 again:
1470 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1471 if (ret > 0) {
1472 ret = btrfs_previous_item(root, path, key.objectid,
1473 BTRFS_DEV_EXTENT_KEY);
1474 if (ret)
1475 goto out;
1476 leaf = path->nodes[0];
1477 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1478 extent = btrfs_item_ptr(leaf, path->slots[0],
1479 struct btrfs_dev_extent);
1480 BUG_ON(found_key.offset > start || found_key.offset +
1481 btrfs_dev_extent_length(leaf, extent) < start);
1482 key = found_key;
1483 btrfs_release_path(path);
1484 goto again;
1485 } else if (ret == 0) {
1486 leaf = path->nodes[0];
1487 extent = btrfs_item_ptr(leaf, path->slots[0],
1488 struct btrfs_dev_extent);
1489 } else {
1490 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1491 goto out;
1492 }
1493
1494 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1495
1496 ret = btrfs_del_item(trans, root, path);
1497 if (ret) {
1498 btrfs_handle_fs_error(root->fs_info, ret,
1499 "Failed to remove dev extent item");
1500 } else {
1501 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1502 }
1503 out:
1504 btrfs_free_path(path);
1505 return ret;
1506 }
1507
1508 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1509 struct btrfs_device *device,
1510 u64 chunk_tree, u64 chunk_objectid,
1511 u64 chunk_offset, u64 start, u64 num_bytes)
1512 {
1513 int ret;
1514 struct btrfs_path *path;
1515 struct btrfs_root *root = device->dev_root;
1516 struct btrfs_dev_extent *extent;
1517 struct extent_buffer *leaf;
1518 struct btrfs_key key;
1519
1520 WARN_ON(!device->in_fs_metadata);
1521 WARN_ON(device->is_tgtdev_for_dev_replace);
1522 path = btrfs_alloc_path();
1523 if (!path)
1524 return -ENOMEM;
1525
1526 key.objectid = device->devid;
1527 key.offset = start;
1528 key.type = BTRFS_DEV_EXTENT_KEY;
1529 ret = btrfs_insert_empty_item(trans, root, path, &key,
1530 sizeof(*extent));
1531 if (ret)
1532 goto out;
1533
1534 leaf = path->nodes[0];
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1538 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1539 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1540
1541 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1542 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1543
1544 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1545 btrfs_mark_buffer_dirty(leaf);
1546 out:
1547 btrfs_free_path(path);
1548 return ret;
1549 }
1550
1551 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1552 {
1553 struct extent_map_tree *em_tree;
1554 struct extent_map *em;
1555 struct rb_node *n;
1556 u64 ret = 0;
1557
1558 em_tree = &fs_info->mapping_tree.map_tree;
1559 read_lock(&em_tree->lock);
1560 n = rb_last(&em_tree->map);
1561 if (n) {
1562 em = rb_entry(n, struct extent_map, rb_node);
1563 ret = em->start + em->len;
1564 }
1565 read_unlock(&em_tree->lock);
1566
1567 return ret;
1568 }
1569
1570 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1571 u64 *devid_ret)
1572 {
1573 int ret;
1574 struct btrfs_key key;
1575 struct btrfs_key found_key;
1576 struct btrfs_path *path;
1577
1578 path = btrfs_alloc_path();
1579 if (!path)
1580 return -ENOMEM;
1581
1582 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1583 key.type = BTRFS_DEV_ITEM_KEY;
1584 key.offset = (u64)-1;
1585
1586 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1587 if (ret < 0)
1588 goto error;
1589
1590 BUG_ON(ret == 0); /* Corruption */
1591
1592 ret = btrfs_previous_item(fs_info->chunk_root, path,
1593 BTRFS_DEV_ITEMS_OBJECTID,
1594 BTRFS_DEV_ITEM_KEY);
1595 if (ret) {
1596 *devid_ret = 1;
1597 } else {
1598 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1599 path->slots[0]);
1600 *devid_ret = found_key.offset + 1;
1601 }
1602 ret = 0;
1603 error:
1604 btrfs_free_path(path);
1605 return ret;
1606 }
1607
1608 /*
1609 * the device information is stored in the chunk root
1610 * the btrfs_device struct should be fully filled in
1611 */
1612 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1613 struct btrfs_root *root,
1614 struct btrfs_device *device)
1615 {
1616 int ret;
1617 struct btrfs_path *path;
1618 struct btrfs_dev_item *dev_item;
1619 struct extent_buffer *leaf;
1620 struct btrfs_key key;
1621 unsigned long ptr;
1622
1623 root = root->fs_info->chunk_root;
1624
1625 path = btrfs_alloc_path();
1626 if (!path)
1627 return -ENOMEM;
1628
1629 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1630 key.type = BTRFS_DEV_ITEM_KEY;
1631 key.offset = device->devid;
1632
1633 ret = btrfs_insert_empty_item(trans, root, path, &key,
1634 sizeof(*dev_item));
1635 if (ret)
1636 goto out;
1637
1638 leaf = path->nodes[0];
1639 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1640
1641 btrfs_set_device_id(leaf, dev_item, device->devid);
1642 btrfs_set_device_generation(leaf, dev_item, 0);
1643 btrfs_set_device_type(leaf, dev_item, device->type);
1644 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1645 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1646 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1647 btrfs_set_device_total_bytes(leaf, dev_item,
1648 btrfs_device_get_disk_total_bytes(device));
1649 btrfs_set_device_bytes_used(leaf, dev_item,
1650 btrfs_device_get_bytes_used(device));
1651 btrfs_set_device_group(leaf, dev_item, 0);
1652 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1653 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1654 btrfs_set_device_start_offset(leaf, dev_item, 0);
1655
1656 ptr = btrfs_device_uuid(dev_item);
1657 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1658 ptr = btrfs_device_fsid(dev_item);
1659 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1660 btrfs_mark_buffer_dirty(leaf);
1661
1662 ret = 0;
1663 out:
1664 btrfs_free_path(path);
1665 return ret;
1666 }
1667
1668 /*
1669 * Function to update ctime/mtime for a given device path.
1670 * Mainly used for ctime/mtime based probe like libblkid.
1671 */
1672 static void update_dev_time(char *path_name)
1673 {
1674 struct file *filp;
1675
1676 filp = filp_open(path_name, O_RDWR, 0);
1677 if (IS_ERR(filp))
1678 return;
1679 file_update_time(filp);
1680 filp_close(filp, NULL);
1681 }
1682
1683 static int btrfs_rm_dev_item(struct btrfs_root *root,
1684 struct btrfs_device *device)
1685 {
1686 int ret;
1687 struct btrfs_path *path;
1688 struct btrfs_key key;
1689 struct btrfs_trans_handle *trans;
1690
1691 root = root->fs_info->chunk_root;
1692
1693 path = btrfs_alloc_path();
1694 if (!path)
1695 return -ENOMEM;
1696
1697 trans = btrfs_start_transaction(root, 0);
1698 if (IS_ERR(trans)) {
1699 btrfs_free_path(path);
1700 return PTR_ERR(trans);
1701 }
1702 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1703 key.type = BTRFS_DEV_ITEM_KEY;
1704 key.offset = device->devid;
1705
1706 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1707 if (ret < 0)
1708 goto out;
1709
1710 if (ret > 0) {
1711 ret = -ENOENT;
1712 goto out;
1713 }
1714
1715 ret = btrfs_del_item(trans, root, path);
1716 if (ret)
1717 goto out;
1718 out:
1719 btrfs_free_path(path);
1720 btrfs_commit_transaction(trans, root);
1721 return ret;
1722 }
1723
1724 /*
1725 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1726 * filesystem. It's up to the caller to adjust that number regarding eg. device
1727 * replace.
1728 */
1729 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1730 u64 num_devices)
1731 {
1732 u64 all_avail;
1733 unsigned seq;
1734 int i;
1735
1736 do {
1737 seq = read_seqbegin(&fs_info->profiles_lock);
1738
1739 all_avail = fs_info->avail_data_alloc_bits |
1740 fs_info->avail_system_alloc_bits |
1741 fs_info->avail_metadata_alloc_bits;
1742 } while (read_seqretry(&fs_info->profiles_lock, seq));
1743
1744 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1745 if (!(all_avail & btrfs_raid_group[i]))
1746 continue;
1747
1748 if (num_devices < btrfs_raid_array[i].devs_min) {
1749 int ret = btrfs_raid_mindev_error[i];
1750
1751 if (ret)
1752 return ret;
1753 }
1754 }
1755
1756 return 0;
1757 }
1758
1759 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1760 struct btrfs_device *device)
1761 {
1762 struct btrfs_device *next_device;
1763
1764 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1765 if (next_device != device &&
1766 !next_device->missing && next_device->bdev)
1767 return next_device;
1768 }
1769
1770 return NULL;
1771 }
1772
1773 /*
1774 * Helper function to check if the given device is part of s_bdev / latest_bdev
1775 * and replace it with the provided or the next active device, in the context
1776 * where this function called, there should be always be another device (or
1777 * this_dev) which is active.
1778 */
1779 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1780 struct btrfs_device *device, struct btrfs_device *this_dev)
1781 {
1782 struct btrfs_device *next_device;
1783
1784 if (this_dev)
1785 next_device = this_dev;
1786 else
1787 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1788 device);
1789 ASSERT(next_device);
1790
1791 if (fs_info->sb->s_bdev &&
1792 (fs_info->sb->s_bdev == device->bdev))
1793 fs_info->sb->s_bdev = next_device->bdev;
1794
1795 if (fs_info->fs_devices->latest_bdev == device->bdev)
1796 fs_info->fs_devices->latest_bdev = next_device->bdev;
1797 }
1798
1799 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1800 {
1801 struct btrfs_device *device;
1802 struct btrfs_fs_devices *cur_devices;
1803 u64 num_devices;
1804 int ret = 0;
1805 bool clear_super = false;
1806 char *dev_name = NULL;
1807
1808 mutex_lock(&uuid_mutex);
1809
1810 num_devices = root->fs_info->fs_devices->num_devices;
1811 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1812 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1813 WARN_ON(num_devices < 1);
1814 num_devices--;
1815 }
1816 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1817
1818 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1819 if (ret)
1820 goto out;
1821
1822 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1823 &device);
1824 if (ret)
1825 goto out;
1826
1827 if (device->is_tgtdev_for_dev_replace) {
1828 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1829 goto out;
1830 }
1831
1832 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1833 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1834 goto out;
1835 }
1836
1837 if (device->writeable) {
1838 lock_chunks(root);
1839 list_del_init(&device->dev_alloc_list);
1840 device->fs_devices->rw_devices--;
1841 unlock_chunks(root);
1842 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1843 if (!dev_name) {
1844 ret = -ENOMEM;
1845 goto error_undo;
1846 }
1847 clear_super = true;
1848 }
1849
1850 mutex_unlock(&uuid_mutex);
1851 ret = btrfs_shrink_device(device, 0);
1852 mutex_lock(&uuid_mutex);
1853 if (ret)
1854 goto error_undo;
1855
1856 /*
1857 * TODO: the superblock still includes this device in its num_devices
1858 * counter although write_all_supers() is not locked out. This
1859 * could give a filesystem state which requires a degraded mount.
1860 */
1861 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1862 if (ret)
1863 goto error_undo;
1864
1865 device->in_fs_metadata = 0;
1866 btrfs_scrub_cancel_dev(root->fs_info, device);
1867
1868 /*
1869 * the device list mutex makes sure that we don't change
1870 * the device list while someone else is writing out all
1871 * the device supers. Whoever is writing all supers, should
1872 * lock the device list mutex before getting the number of
1873 * devices in the super block (super_copy). Conversely,
1874 * whoever updates the number of devices in the super block
1875 * (super_copy) should hold the device list mutex.
1876 */
1877
1878 cur_devices = device->fs_devices;
1879 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1880 list_del_rcu(&device->dev_list);
1881
1882 device->fs_devices->num_devices--;
1883 device->fs_devices->total_devices--;
1884
1885 if (device->missing)
1886 device->fs_devices->missing_devices--;
1887
1888 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1889
1890 if (device->bdev) {
1891 device->fs_devices->open_devices--;
1892 /* remove sysfs entry */
1893 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1894 }
1895
1896 call_rcu(&device->rcu, free_device);
1897
1898 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1899 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1900 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1901
1902 if (cur_devices->open_devices == 0) {
1903 struct btrfs_fs_devices *fs_devices;
1904 fs_devices = root->fs_info->fs_devices;
1905 while (fs_devices) {
1906 if (fs_devices->seed == cur_devices) {
1907 fs_devices->seed = cur_devices->seed;
1908 break;
1909 }
1910 fs_devices = fs_devices->seed;
1911 }
1912 cur_devices->seed = NULL;
1913 __btrfs_close_devices(cur_devices);
1914 free_fs_devices(cur_devices);
1915 }
1916
1917 root->fs_info->num_tolerated_disk_barrier_failures =
1918 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1919
1920 /*
1921 * at this point, the device is zero sized. We want to
1922 * remove it from the devices list and zero out the old super
1923 */
1924 if (clear_super) {
1925 struct block_device *bdev;
1926
1927 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1928 root->fs_info->bdev_holder);
1929 if (!IS_ERR(bdev)) {
1930 btrfs_scratch_superblocks(bdev, dev_name);
1931 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1932 }
1933 }
1934
1935 out:
1936 kfree(dev_name);
1937
1938 mutex_unlock(&uuid_mutex);
1939 return ret;
1940
1941 error_undo:
1942 if (device->writeable) {
1943 lock_chunks(root);
1944 list_add(&device->dev_alloc_list,
1945 &root->fs_info->fs_devices->alloc_list);
1946 device->fs_devices->rw_devices++;
1947 unlock_chunks(root);
1948 }
1949 goto out;
1950 }
1951
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953 struct btrfs_device *srcdev)
1954 {
1955 struct btrfs_fs_devices *fs_devices;
1956
1957 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1958
1959 /*
1960 * in case of fs with no seed, srcdev->fs_devices will point
1961 * to fs_devices of fs_info. However when the dev being replaced is
1962 * a seed dev it will point to the seed's local fs_devices. In short
1963 * srcdev will have its correct fs_devices in both the cases.
1964 */
1965 fs_devices = srcdev->fs_devices;
1966
1967 list_del_rcu(&srcdev->dev_list);
1968 list_del_rcu(&srcdev->dev_alloc_list);
1969 fs_devices->num_devices--;
1970 if (srcdev->missing)
1971 fs_devices->missing_devices--;
1972
1973 if (srcdev->writeable)
1974 fs_devices->rw_devices--;
1975
1976 if (srcdev->bdev)
1977 fs_devices->open_devices--;
1978 }
1979
1980 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1981 struct btrfs_device *srcdev)
1982 {
1983 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1984
1985 if (srcdev->writeable) {
1986 /* zero out the old super if it is writable */
1987 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1988 }
1989 call_rcu(&srcdev->rcu, free_device);
1990
1991 /*
1992 * unless fs_devices is seed fs, num_devices shouldn't go
1993 * zero
1994 */
1995 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996
1997 /* if this is no devs we rather delete the fs_devices */
1998 if (!fs_devices->num_devices) {
1999 struct btrfs_fs_devices *tmp_fs_devices;
2000
2001 tmp_fs_devices = fs_info->fs_devices;
2002 while (tmp_fs_devices) {
2003 if (tmp_fs_devices->seed == fs_devices) {
2004 tmp_fs_devices->seed = fs_devices->seed;
2005 break;
2006 }
2007 tmp_fs_devices = tmp_fs_devices->seed;
2008 }
2009 fs_devices->seed = NULL;
2010 __btrfs_close_devices(fs_devices);
2011 free_fs_devices(fs_devices);
2012 }
2013 }
2014
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016 struct btrfs_device *tgtdev)
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 fs_info->fs_devices->open_devices--;
2026
2027 fs_info->fs_devices->num_devices--;
2028
2029 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2030
2031 list_del_rcu(&tgtdev->dev_list);
2032
2033 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2034 mutex_unlock(&uuid_mutex);
2035
2036 /*
2037 * The update_dev_time() with in btrfs_scratch_superblocks()
2038 * may lead to a call to btrfs_show_devname() which will try
2039 * to hold device_list_mutex. And here this device
2040 * is already out of device list, so we don't have to hold
2041 * the device_list_mutex lock.
2042 */
2043 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2044 call_rcu(&tgtdev->rcu, free_device);
2045 }
2046
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048 struct btrfs_device **device)
2049 {
2050 int ret = 0;
2051 struct btrfs_super_block *disk_super;
2052 u64 devid;
2053 u8 *dev_uuid;
2054 struct block_device *bdev;
2055 struct buffer_head *bh;
2056
2057 *device = NULL;
2058 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059 root->fs_info->bdev_holder, 0, &bdev, &bh);
2060 if (ret)
2061 return ret;
2062 disk_super = (struct btrfs_super_block *)bh->b_data;
2063 devid = btrfs_stack_device_id(&disk_super->dev_item);
2064 dev_uuid = disk_super->dev_item.uuid;
2065 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066 disk_super->fsid);
2067 brelse(bh);
2068 if (!*device)
2069 ret = -ENOENT;
2070 blkdev_put(bdev, FMODE_READ);
2071 return ret;
2072 }
2073
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075 char *device_path,
2076 struct btrfs_device **device)
2077 {
2078 *device = NULL;
2079 if (strcmp(device_path, "missing") == 0) {
2080 struct list_head *devices;
2081 struct btrfs_device *tmp;
2082
2083 devices = &root->fs_info->fs_devices->devices;
2084 /*
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2087 */
2088 list_for_each_entry(tmp, devices, dev_list) {
2089 if (tmp->in_fs_metadata && !tmp->bdev) {
2090 *device = tmp;
2091 break;
2092 }
2093 }
2094
2095 if (!*device)
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097
2098 return 0;
2099 } else {
2100 return btrfs_find_device_by_path(root, device_path, device);
2101 }
2102 }
2103
2104 /*
2105 * Lookup a device given by device id, or the path if the id is 0.
2106 */
2107 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2108 char *devpath,
2109 struct btrfs_device **device)
2110 {
2111 int ret;
2112
2113 if (devid) {
2114 ret = 0;
2115 *device = btrfs_find_device(root->fs_info, devid, NULL,
2116 NULL);
2117 if (!*device)
2118 ret = -ENOENT;
2119 } else {
2120 if (!devpath || !devpath[0])
2121 return -EINVAL;
2122
2123 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2124 device);
2125 }
2126 return ret;
2127 }
2128
2129 /*
2130 * does all the dirty work required for changing file system's UUID.
2131 */
2132 static int btrfs_prepare_sprout(struct btrfs_root *root)
2133 {
2134 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2135 struct btrfs_fs_devices *old_devices;
2136 struct btrfs_fs_devices *seed_devices;
2137 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2138 struct btrfs_device *device;
2139 u64 super_flags;
2140
2141 BUG_ON(!mutex_is_locked(&uuid_mutex));
2142 if (!fs_devices->seeding)
2143 return -EINVAL;
2144
2145 seed_devices = __alloc_fs_devices();
2146 if (IS_ERR(seed_devices))
2147 return PTR_ERR(seed_devices);
2148
2149 old_devices = clone_fs_devices(fs_devices);
2150 if (IS_ERR(old_devices)) {
2151 kfree(seed_devices);
2152 return PTR_ERR(old_devices);
2153 }
2154
2155 list_add(&old_devices->list, &fs_uuids);
2156
2157 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2158 seed_devices->opened = 1;
2159 INIT_LIST_HEAD(&seed_devices->devices);
2160 INIT_LIST_HEAD(&seed_devices->alloc_list);
2161 mutex_init(&seed_devices->device_list_mutex);
2162
2163 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2164 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2165 synchronize_rcu);
2166 list_for_each_entry(device, &seed_devices->devices, dev_list)
2167 device->fs_devices = seed_devices;
2168
2169 lock_chunks(root);
2170 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2171 unlock_chunks(root);
2172
2173 fs_devices->seeding = 0;
2174 fs_devices->num_devices = 0;
2175 fs_devices->open_devices = 0;
2176 fs_devices->missing_devices = 0;
2177 fs_devices->rotating = 0;
2178 fs_devices->seed = seed_devices;
2179
2180 generate_random_uuid(fs_devices->fsid);
2181 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2182 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2183 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2184
2185 super_flags = btrfs_super_flags(disk_super) &
2186 ~BTRFS_SUPER_FLAG_SEEDING;
2187 btrfs_set_super_flags(disk_super, super_flags);
2188
2189 return 0;
2190 }
2191
2192 /*
2193 * Store the expected generation for seed devices in device items.
2194 */
2195 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2196 struct btrfs_root *root)
2197 {
2198 struct btrfs_path *path;
2199 struct extent_buffer *leaf;
2200 struct btrfs_dev_item *dev_item;
2201 struct btrfs_device *device;
2202 struct btrfs_key key;
2203 u8 fs_uuid[BTRFS_UUID_SIZE];
2204 u8 dev_uuid[BTRFS_UUID_SIZE];
2205 u64 devid;
2206 int ret;
2207
2208 path = btrfs_alloc_path();
2209 if (!path)
2210 return -ENOMEM;
2211
2212 root = root->fs_info->chunk_root;
2213 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2214 key.offset = 0;
2215 key.type = BTRFS_DEV_ITEM_KEY;
2216
2217 while (1) {
2218 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2219 if (ret < 0)
2220 goto error;
2221
2222 leaf = path->nodes[0];
2223 next_slot:
2224 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2225 ret = btrfs_next_leaf(root, path);
2226 if (ret > 0)
2227 break;
2228 if (ret < 0)
2229 goto error;
2230 leaf = path->nodes[0];
2231 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2232 btrfs_release_path(path);
2233 continue;
2234 }
2235
2236 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2237 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2238 key.type != BTRFS_DEV_ITEM_KEY)
2239 break;
2240
2241 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2242 struct btrfs_dev_item);
2243 devid = btrfs_device_id(leaf, dev_item);
2244 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2245 BTRFS_UUID_SIZE);
2246 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2247 BTRFS_UUID_SIZE);
2248 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2249 fs_uuid);
2250 BUG_ON(!device); /* Logic error */
2251
2252 if (device->fs_devices->seeding) {
2253 btrfs_set_device_generation(leaf, dev_item,
2254 device->generation);
2255 btrfs_mark_buffer_dirty(leaf);
2256 }
2257
2258 path->slots[0]++;
2259 goto next_slot;
2260 }
2261 ret = 0;
2262 error:
2263 btrfs_free_path(path);
2264 return ret;
2265 }
2266
2267 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2268 {
2269 struct request_queue *q;
2270 struct btrfs_trans_handle *trans;
2271 struct btrfs_device *device;
2272 struct block_device *bdev;
2273 struct list_head *devices;
2274 struct super_block *sb = root->fs_info->sb;
2275 struct rcu_string *name;
2276 u64 tmp;
2277 int seeding_dev = 0;
2278 int ret = 0;
2279
2280 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2281 return -EROFS;
2282
2283 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2284 root->fs_info->bdev_holder);
2285 if (IS_ERR(bdev))
2286 return PTR_ERR(bdev);
2287
2288 if (root->fs_info->fs_devices->seeding) {
2289 seeding_dev = 1;
2290 down_write(&sb->s_umount);
2291 mutex_lock(&uuid_mutex);
2292 }
2293
2294 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2295
2296 devices = &root->fs_info->fs_devices->devices;
2297
2298 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2299 list_for_each_entry(device, devices, dev_list) {
2300 if (device->bdev == bdev) {
2301 ret = -EEXIST;
2302 mutex_unlock(
2303 &root->fs_info->fs_devices->device_list_mutex);
2304 goto error;
2305 }
2306 }
2307 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308
2309 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2310 if (IS_ERR(device)) {
2311 /* we can safely leave the fs_devices entry around */
2312 ret = PTR_ERR(device);
2313 goto error;
2314 }
2315
2316 name = rcu_string_strdup(device_path, GFP_KERNEL);
2317 if (!name) {
2318 kfree(device);
2319 ret = -ENOMEM;
2320 goto error;
2321 }
2322 rcu_assign_pointer(device->name, name);
2323
2324 trans = btrfs_start_transaction(root, 0);
2325 if (IS_ERR(trans)) {
2326 rcu_string_free(device->name);
2327 kfree(device);
2328 ret = PTR_ERR(trans);
2329 goto error;
2330 }
2331
2332 q = bdev_get_queue(bdev);
2333 if (blk_queue_discard(q))
2334 device->can_discard = 1;
2335 device->writeable = 1;
2336 device->generation = trans->transid;
2337 device->io_width = root->sectorsize;
2338 device->io_align = root->sectorsize;
2339 device->sector_size = root->sectorsize;
2340 device->total_bytes = i_size_read(bdev->bd_inode);
2341 device->disk_total_bytes = device->total_bytes;
2342 device->commit_total_bytes = device->total_bytes;
2343 device->dev_root = root->fs_info->dev_root;
2344 device->bdev = bdev;
2345 device->in_fs_metadata = 1;
2346 device->is_tgtdev_for_dev_replace = 0;
2347 device->mode = FMODE_EXCL;
2348 device->dev_stats_valid = 1;
2349 set_blocksize(device->bdev, 4096);
2350
2351 if (seeding_dev) {
2352 sb->s_flags &= ~MS_RDONLY;
2353 ret = btrfs_prepare_sprout(root);
2354 BUG_ON(ret); /* -ENOMEM */
2355 }
2356
2357 device->fs_devices = root->fs_info->fs_devices;
2358
2359 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2360 lock_chunks(root);
2361 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2362 list_add(&device->dev_alloc_list,
2363 &root->fs_info->fs_devices->alloc_list);
2364 root->fs_info->fs_devices->num_devices++;
2365 root->fs_info->fs_devices->open_devices++;
2366 root->fs_info->fs_devices->rw_devices++;
2367 root->fs_info->fs_devices->total_devices++;
2368 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2369
2370 spin_lock(&root->fs_info->free_chunk_lock);
2371 root->fs_info->free_chunk_space += device->total_bytes;
2372 spin_unlock(&root->fs_info->free_chunk_lock);
2373
2374 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2375 root->fs_info->fs_devices->rotating = 1;
2376
2377 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2378 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2379 tmp + device->total_bytes);
2380
2381 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2382 btrfs_set_super_num_devices(root->fs_info->super_copy,
2383 tmp + 1);
2384
2385 /* add sysfs device entry */
2386 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2387
2388 /*
2389 * we've got more storage, clear any full flags on the space
2390 * infos
2391 */
2392 btrfs_clear_space_info_full(root->fs_info);
2393
2394 unlock_chunks(root);
2395 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2396
2397 if (seeding_dev) {
2398 lock_chunks(root);
2399 ret = init_first_rw_device(trans, root, device);
2400 unlock_chunks(root);
2401 if (ret) {
2402 btrfs_abort_transaction(trans, root, ret);
2403 goto error_trans;
2404 }
2405 }
2406
2407 ret = btrfs_add_device(trans, root, device);
2408 if (ret) {
2409 btrfs_abort_transaction(trans, root, ret);
2410 goto error_trans;
2411 }
2412
2413 if (seeding_dev) {
2414 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2415
2416 ret = btrfs_finish_sprout(trans, root);
2417 if (ret) {
2418 btrfs_abort_transaction(trans, root, ret);
2419 goto error_trans;
2420 }
2421
2422 /* Sprouting would change fsid of the mounted root,
2423 * so rename the fsid on the sysfs
2424 */
2425 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2426 root->fs_info->fsid);
2427 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2428 fsid_buf))
2429 btrfs_warn(root->fs_info,
2430 "sysfs: failed to create fsid for sprout");
2431 }
2432
2433 root->fs_info->num_tolerated_disk_barrier_failures =
2434 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2435 ret = btrfs_commit_transaction(trans, root);
2436
2437 if (seeding_dev) {
2438 mutex_unlock(&uuid_mutex);
2439 up_write(&sb->s_umount);
2440
2441 if (ret) /* transaction commit */
2442 return ret;
2443
2444 ret = btrfs_relocate_sys_chunks(root);
2445 if (ret < 0)
2446 btrfs_handle_fs_error(root->fs_info, ret,
2447 "Failed to relocate sys chunks after "
2448 "device initialization. This can be fixed "
2449 "using the \"btrfs balance\" command.");
2450 trans = btrfs_attach_transaction(root);
2451 if (IS_ERR(trans)) {
2452 if (PTR_ERR(trans) == -ENOENT)
2453 return 0;
2454 return PTR_ERR(trans);
2455 }
2456 ret = btrfs_commit_transaction(trans, root);
2457 }
2458
2459 /* Update ctime/mtime for libblkid */
2460 update_dev_time(device_path);
2461 return ret;
2462
2463 error_trans:
2464 btrfs_end_transaction(trans, root);
2465 rcu_string_free(device->name);
2466 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2467 kfree(device);
2468 error:
2469 blkdev_put(bdev, FMODE_EXCL);
2470 if (seeding_dev) {
2471 mutex_unlock(&uuid_mutex);
2472 up_write(&sb->s_umount);
2473 }
2474 return ret;
2475 }
2476
2477 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2478 struct btrfs_device *srcdev,
2479 struct btrfs_device **device_out)
2480 {
2481 struct request_queue *q;
2482 struct btrfs_device *device;
2483 struct block_device *bdev;
2484 struct btrfs_fs_info *fs_info = root->fs_info;
2485 struct list_head *devices;
2486 struct rcu_string *name;
2487 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2488 int ret = 0;
2489
2490 *device_out = NULL;
2491 if (fs_info->fs_devices->seeding) {
2492 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2493 return -EINVAL;
2494 }
2495
2496 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2497 fs_info->bdev_holder);
2498 if (IS_ERR(bdev)) {
2499 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2500 return PTR_ERR(bdev);
2501 }
2502
2503 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2504
2505 devices = &fs_info->fs_devices->devices;
2506 list_for_each_entry(device, devices, dev_list) {
2507 if (device->bdev == bdev) {
2508 btrfs_err(fs_info, "target device is in the filesystem!");
2509 ret = -EEXIST;
2510 goto error;
2511 }
2512 }
2513
2514
2515 if (i_size_read(bdev->bd_inode) <
2516 btrfs_device_get_total_bytes(srcdev)) {
2517 btrfs_err(fs_info, "target device is smaller than source device!");
2518 ret = -EINVAL;
2519 goto error;
2520 }
2521
2522
2523 device = btrfs_alloc_device(NULL, &devid, NULL);
2524 if (IS_ERR(device)) {
2525 ret = PTR_ERR(device);
2526 goto error;
2527 }
2528
2529 name = rcu_string_strdup(device_path, GFP_NOFS);
2530 if (!name) {
2531 kfree(device);
2532 ret = -ENOMEM;
2533 goto error;
2534 }
2535 rcu_assign_pointer(device->name, name);
2536
2537 q = bdev_get_queue(bdev);
2538 if (blk_queue_discard(q))
2539 device->can_discard = 1;
2540 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2541 device->writeable = 1;
2542 device->generation = 0;
2543 device->io_width = root->sectorsize;
2544 device->io_align = root->sectorsize;
2545 device->sector_size = root->sectorsize;
2546 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2547 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2548 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2549 ASSERT(list_empty(&srcdev->resized_list));
2550 device->commit_total_bytes = srcdev->commit_total_bytes;
2551 device->commit_bytes_used = device->bytes_used;
2552 device->dev_root = fs_info->dev_root;
2553 device->bdev = bdev;
2554 device->in_fs_metadata = 1;
2555 device->is_tgtdev_for_dev_replace = 1;
2556 device->mode = FMODE_EXCL;
2557 device->dev_stats_valid = 1;
2558 set_blocksize(device->bdev, 4096);
2559 device->fs_devices = fs_info->fs_devices;
2560 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2561 fs_info->fs_devices->num_devices++;
2562 fs_info->fs_devices->open_devices++;
2563 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2564
2565 *device_out = device;
2566 return ret;
2567
2568 error:
2569 blkdev_put(bdev, FMODE_EXCL);
2570 return ret;
2571 }
2572
2573 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2574 struct btrfs_device *tgtdev)
2575 {
2576 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2577 tgtdev->io_width = fs_info->dev_root->sectorsize;
2578 tgtdev->io_align = fs_info->dev_root->sectorsize;
2579 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2580 tgtdev->dev_root = fs_info->dev_root;
2581 tgtdev->in_fs_metadata = 1;
2582 }
2583
2584 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2585 struct btrfs_device *device)
2586 {
2587 int ret;
2588 struct btrfs_path *path;
2589 struct btrfs_root *root;
2590 struct btrfs_dev_item *dev_item;
2591 struct extent_buffer *leaf;
2592 struct btrfs_key key;
2593
2594 root = device->dev_root->fs_info->chunk_root;
2595
2596 path = btrfs_alloc_path();
2597 if (!path)
2598 return -ENOMEM;
2599
2600 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2601 key.type = BTRFS_DEV_ITEM_KEY;
2602 key.offset = device->devid;
2603
2604 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2605 if (ret < 0)
2606 goto out;
2607
2608 if (ret > 0) {
2609 ret = -ENOENT;
2610 goto out;
2611 }
2612
2613 leaf = path->nodes[0];
2614 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2615
2616 btrfs_set_device_id(leaf, dev_item, device->devid);
2617 btrfs_set_device_type(leaf, dev_item, device->type);
2618 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2619 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2620 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2621 btrfs_set_device_total_bytes(leaf, dev_item,
2622 btrfs_device_get_disk_total_bytes(device));
2623 btrfs_set_device_bytes_used(leaf, dev_item,
2624 btrfs_device_get_bytes_used(device));
2625 btrfs_mark_buffer_dirty(leaf);
2626
2627 out:
2628 btrfs_free_path(path);
2629 return ret;
2630 }
2631
2632 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2633 struct btrfs_device *device, u64 new_size)
2634 {
2635 struct btrfs_super_block *super_copy =
2636 device->dev_root->fs_info->super_copy;
2637 struct btrfs_fs_devices *fs_devices;
2638 u64 old_total;
2639 u64 diff;
2640
2641 if (!device->writeable)
2642 return -EACCES;
2643
2644 lock_chunks(device->dev_root);
2645 old_total = btrfs_super_total_bytes(super_copy);
2646 diff = new_size - device->total_bytes;
2647
2648 if (new_size <= device->total_bytes ||
2649 device->is_tgtdev_for_dev_replace) {
2650 unlock_chunks(device->dev_root);
2651 return -EINVAL;
2652 }
2653
2654 fs_devices = device->dev_root->fs_info->fs_devices;
2655
2656 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2657 device->fs_devices->total_rw_bytes += diff;
2658
2659 btrfs_device_set_total_bytes(device, new_size);
2660 btrfs_device_set_disk_total_bytes(device, new_size);
2661 btrfs_clear_space_info_full(device->dev_root->fs_info);
2662 if (list_empty(&device->resized_list))
2663 list_add_tail(&device->resized_list,
2664 &fs_devices->resized_devices);
2665 unlock_chunks(device->dev_root);
2666
2667 return btrfs_update_device(trans, device);
2668 }
2669
2670 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2671 struct btrfs_root *root, u64 chunk_objectid,
2672 u64 chunk_offset)
2673 {
2674 int ret;
2675 struct btrfs_path *path;
2676 struct btrfs_key key;
2677
2678 root = root->fs_info->chunk_root;
2679 path = btrfs_alloc_path();
2680 if (!path)
2681 return -ENOMEM;
2682
2683 key.objectid = chunk_objectid;
2684 key.offset = chunk_offset;
2685 key.type = BTRFS_CHUNK_ITEM_KEY;
2686
2687 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2688 if (ret < 0)
2689 goto out;
2690 else if (ret > 0) { /* Logic error or corruption */
2691 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2692 "Failed lookup while freeing chunk.");
2693 ret = -ENOENT;
2694 goto out;
2695 }
2696
2697 ret = btrfs_del_item(trans, root, path);
2698 if (ret < 0)
2699 btrfs_handle_fs_error(root->fs_info, ret,
2700 "Failed to delete chunk item.");
2701 out:
2702 btrfs_free_path(path);
2703 return ret;
2704 }
2705
2706 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2707 chunk_offset)
2708 {
2709 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2710 struct btrfs_disk_key *disk_key;
2711 struct btrfs_chunk *chunk;
2712 u8 *ptr;
2713 int ret = 0;
2714 u32 num_stripes;
2715 u32 array_size;
2716 u32 len = 0;
2717 u32 cur;
2718 struct btrfs_key key;
2719
2720 lock_chunks(root);
2721 array_size = btrfs_super_sys_array_size(super_copy);
2722
2723 ptr = super_copy->sys_chunk_array;
2724 cur = 0;
2725
2726 while (cur < array_size) {
2727 disk_key = (struct btrfs_disk_key *)ptr;
2728 btrfs_disk_key_to_cpu(&key, disk_key);
2729
2730 len = sizeof(*disk_key);
2731
2732 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2733 chunk = (struct btrfs_chunk *)(ptr + len);
2734 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2735 len += btrfs_chunk_item_size(num_stripes);
2736 } else {
2737 ret = -EIO;
2738 break;
2739 }
2740 if (key.objectid == chunk_objectid &&
2741 key.offset == chunk_offset) {
2742 memmove(ptr, ptr + len, array_size - (cur + len));
2743 array_size -= len;
2744 btrfs_set_super_sys_array_size(super_copy, array_size);
2745 } else {
2746 ptr += len;
2747 cur += len;
2748 }
2749 }
2750 unlock_chunks(root);
2751 return ret;
2752 }
2753
2754 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2755 struct btrfs_root *root, u64 chunk_offset)
2756 {
2757 struct extent_map_tree *em_tree;
2758 struct extent_map *em;
2759 struct btrfs_root *extent_root = root->fs_info->extent_root;
2760 struct map_lookup *map;
2761 u64 dev_extent_len = 0;
2762 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2763 int i, ret = 0;
2764
2765 /* Just in case */
2766 root = root->fs_info->chunk_root;
2767 em_tree = &root->fs_info->mapping_tree.map_tree;
2768
2769 read_lock(&em_tree->lock);
2770 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2771 read_unlock(&em_tree->lock);
2772
2773 if (!em || em->start > chunk_offset ||
2774 em->start + em->len < chunk_offset) {
2775 /*
2776 * This is a logic error, but we don't want to just rely on the
2777 * user having built with ASSERT enabled, so if ASSERT doesn't
2778 * do anything we still error out.
2779 */
2780 ASSERT(0);
2781 if (em)
2782 free_extent_map(em);
2783 return -EINVAL;
2784 }
2785 map = em->map_lookup;
2786 lock_chunks(root->fs_info->chunk_root);
2787 check_system_chunk(trans, extent_root, map->type);
2788 unlock_chunks(root->fs_info->chunk_root);
2789
2790 for (i = 0; i < map->num_stripes; i++) {
2791 struct btrfs_device *device = map->stripes[i].dev;
2792 ret = btrfs_free_dev_extent(trans, device,
2793 map->stripes[i].physical,
2794 &dev_extent_len);
2795 if (ret) {
2796 btrfs_abort_transaction(trans, root, ret);
2797 goto out;
2798 }
2799
2800 if (device->bytes_used > 0) {
2801 lock_chunks(root);
2802 btrfs_device_set_bytes_used(device,
2803 device->bytes_used - dev_extent_len);
2804 spin_lock(&root->fs_info->free_chunk_lock);
2805 root->fs_info->free_chunk_space += dev_extent_len;
2806 spin_unlock(&root->fs_info->free_chunk_lock);
2807 btrfs_clear_space_info_full(root->fs_info);
2808 unlock_chunks(root);
2809 }
2810
2811 if (map->stripes[i].dev) {
2812 ret = btrfs_update_device(trans, map->stripes[i].dev);
2813 if (ret) {
2814 btrfs_abort_transaction(trans, root, ret);
2815 goto out;
2816 }
2817 }
2818 }
2819 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2820 if (ret) {
2821 btrfs_abort_transaction(trans, root, ret);
2822 goto out;
2823 }
2824
2825 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2826
2827 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2828 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2829 if (ret) {
2830 btrfs_abort_transaction(trans, root, ret);
2831 goto out;
2832 }
2833 }
2834
2835 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2836 if (ret) {
2837 btrfs_abort_transaction(trans, extent_root, ret);
2838 goto out;
2839 }
2840
2841 out:
2842 /* once for us */
2843 free_extent_map(em);
2844 return ret;
2845 }
2846
2847 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2848 {
2849 struct btrfs_root *extent_root;
2850 struct btrfs_trans_handle *trans;
2851 int ret;
2852
2853 root = root->fs_info->chunk_root;
2854 extent_root = root->fs_info->extent_root;
2855
2856 /*
2857 * Prevent races with automatic removal of unused block groups.
2858 * After we relocate and before we remove the chunk with offset
2859 * chunk_offset, automatic removal of the block group can kick in,
2860 * resulting in a failure when calling btrfs_remove_chunk() below.
2861 *
2862 * Make sure to acquire this mutex before doing a tree search (dev
2863 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2864 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2865 * we release the path used to search the chunk/dev tree and before
2866 * the current task acquires this mutex and calls us.
2867 */
2868 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2869
2870 ret = btrfs_can_relocate(extent_root, chunk_offset);
2871 if (ret)
2872 return -ENOSPC;
2873
2874 /* step one, relocate all the extents inside this chunk */
2875 btrfs_scrub_pause(root);
2876 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2877 btrfs_scrub_continue(root);
2878 if (ret)
2879 return ret;
2880
2881 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2882 chunk_offset);
2883 if (IS_ERR(trans)) {
2884 ret = PTR_ERR(trans);
2885 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2886 return ret;
2887 }
2888
2889 /*
2890 * step two, delete the device extents and the
2891 * chunk tree entries
2892 */
2893 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2894 btrfs_end_transaction(trans, root);
2895 return ret;
2896 }
2897
2898 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2899 {
2900 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2901 struct btrfs_path *path;
2902 struct extent_buffer *leaf;
2903 struct btrfs_chunk *chunk;
2904 struct btrfs_key key;
2905 struct btrfs_key found_key;
2906 u64 chunk_type;
2907 bool retried = false;
2908 int failed = 0;
2909 int ret;
2910
2911 path = btrfs_alloc_path();
2912 if (!path)
2913 return -ENOMEM;
2914
2915 again:
2916 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2917 key.offset = (u64)-1;
2918 key.type = BTRFS_CHUNK_ITEM_KEY;
2919
2920 while (1) {
2921 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2922 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2923 if (ret < 0) {
2924 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2925 goto error;
2926 }
2927 BUG_ON(ret == 0); /* Corruption */
2928
2929 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2930 key.type);
2931 if (ret)
2932 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2933 if (ret < 0)
2934 goto error;
2935 if (ret > 0)
2936 break;
2937
2938 leaf = path->nodes[0];
2939 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2940
2941 chunk = btrfs_item_ptr(leaf, path->slots[0],
2942 struct btrfs_chunk);
2943 chunk_type = btrfs_chunk_type(leaf, chunk);
2944 btrfs_release_path(path);
2945
2946 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2947 ret = btrfs_relocate_chunk(chunk_root,
2948 found_key.offset);
2949 if (ret == -ENOSPC)
2950 failed++;
2951 else
2952 BUG_ON(ret);
2953 }
2954 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2955
2956 if (found_key.offset == 0)
2957 break;
2958 key.offset = found_key.offset - 1;
2959 }
2960 ret = 0;
2961 if (failed && !retried) {
2962 failed = 0;
2963 retried = true;
2964 goto again;
2965 } else if (WARN_ON(failed && retried)) {
2966 ret = -ENOSPC;
2967 }
2968 error:
2969 btrfs_free_path(path);
2970 return ret;
2971 }
2972
2973 static int insert_balance_item(struct btrfs_root *root,
2974 struct btrfs_balance_control *bctl)
2975 {
2976 struct btrfs_trans_handle *trans;
2977 struct btrfs_balance_item *item;
2978 struct btrfs_disk_balance_args disk_bargs;
2979 struct btrfs_path *path;
2980 struct extent_buffer *leaf;
2981 struct btrfs_key key;
2982 int ret, err;
2983
2984 path = btrfs_alloc_path();
2985 if (!path)
2986 return -ENOMEM;
2987
2988 trans = btrfs_start_transaction(root, 0);
2989 if (IS_ERR(trans)) {
2990 btrfs_free_path(path);
2991 return PTR_ERR(trans);
2992 }
2993
2994 key.objectid = BTRFS_BALANCE_OBJECTID;
2995 key.type = BTRFS_TEMPORARY_ITEM_KEY;
2996 key.offset = 0;
2997
2998 ret = btrfs_insert_empty_item(trans, root, path, &key,
2999 sizeof(*item));
3000 if (ret)
3001 goto out;
3002
3003 leaf = path->nodes[0];
3004 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3005
3006 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3007
3008 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3009 btrfs_set_balance_data(leaf, item, &disk_bargs);
3010 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3011 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3012 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3013 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3014
3015 btrfs_set_balance_flags(leaf, item, bctl->flags);
3016
3017 btrfs_mark_buffer_dirty(leaf);
3018 out:
3019 btrfs_free_path(path);
3020 err = btrfs_commit_transaction(trans, root);
3021 if (err && !ret)
3022 ret = err;
3023 return ret;
3024 }
3025
3026 static int del_balance_item(struct btrfs_root *root)
3027 {
3028 struct btrfs_trans_handle *trans;
3029 struct btrfs_path *path;
3030 struct btrfs_key key;
3031 int ret, err;
3032
3033 path = btrfs_alloc_path();
3034 if (!path)
3035 return -ENOMEM;
3036
3037 trans = btrfs_start_transaction(root, 0);
3038 if (IS_ERR(trans)) {
3039 btrfs_free_path(path);
3040 return PTR_ERR(trans);
3041 }
3042
3043 key.objectid = BTRFS_BALANCE_OBJECTID;
3044 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3045 key.offset = 0;
3046
3047 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3048 if (ret < 0)
3049 goto out;
3050 if (ret > 0) {
3051 ret = -ENOENT;
3052 goto out;
3053 }
3054
3055 ret = btrfs_del_item(trans, root, path);
3056 out:
3057 btrfs_free_path(path);
3058 err = btrfs_commit_transaction(trans, root);
3059 if (err && !ret)
3060 ret = err;
3061 return ret;
3062 }
3063
3064 /*
3065 * This is a heuristic used to reduce the number of chunks balanced on
3066 * resume after balance was interrupted.
3067 */
3068 static void update_balance_args(struct btrfs_balance_control *bctl)
3069 {
3070 /*
3071 * Turn on soft mode for chunk types that were being converted.
3072 */
3073 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3074 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3075 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3076 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3077 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3078 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3079
3080 /*
3081 * Turn on usage filter if is not already used. The idea is
3082 * that chunks that we have already balanced should be
3083 * reasonably full. Don't do it for chunks that are being
3084 * converted - that will keep us from relocating unconverted
3085 * (albeit full) chunks.
3086 */
3087 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3088 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3089 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3090 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3091 bctl->data.usage = 90;
3092 }
3093 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3094 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3095 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3096 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3097 bctl->sys.usage = 90;
3098 }
3099 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3100 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3101 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3102 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3103 bctl->meta.usage = 90;
3104 }
3105 }
3106
3107 /*
3108 * Should be called with both balance and volume mutexes held to
3109 * serialize other volume operations (add_dev/rm_dev/resize) with
3110 * restriper. Same goes for unset_balance_control.
3111 */
3112 static void set_balance_control(struct btrfs_balance_control *bctl)
3113 {
3114 struct btrfs_fs_info *fs_info = bctl->fs_info;
3115
3116 BUG_ON(fs_info->balance_ctl);
3117
3118 spin_lock(&fs_info->balance_lock);
3119 fs_info->balance_ctl = bctl;
3120 spin_unlock(&fs_info->balance_lock);
3121 }
3122
3123 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3124 {
3125 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3126
3127 BUG_ON(!fs_info->balance_ctl);
3128
3129 spin_lock(&fs_info->balance_lock);
3130 fs_info->balance_ctl = NULL;
3131 spin_unlock(&fs_info->balance_lock);
3132
3133 kfree(bctl);
3134 }
3135
3136 /*
3137 * Balance filters. Return 1 if chunk should be filtered out
3138 * (should not be balanced).
3139 */
3140 static int chunk_profiles_filter(u64 chunk_type,
3141 struct btrfs_balance_args *bargs)
3142 {
3143 chunk_type = chunk_to_extended(chunk_type) &
3144 BTRFS_EXTENDED_PROFILE_MASK;
3145
3146 if (bargs->profiles & chunk_type)
3147 return 0;
3148
3149 return 1;
3150 }
3151
3152 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3153 struct btrfs_balance_args *bargs)
3154 {
3155 struct btrfs_block_group_cache *cache;
3156 u64 chunk_used;
3157 u64 user_thresh_min;
3158 u64 user_thresh_max;
3159 int ret = 1;
3160
3161 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3162 chunk_used = btrfs_block_group_used(&cache->item);
3163
3164 if (bargs->usage_min == 0)
3165 user_thresh_min = 0;
3166 else
3167 user_thresh_min = div_factor_fine(cache->key.offset,
3168 bargs->usage_min);
3169
3170 if (bargs->usage_max == 0)
3171 user_thresh_max = 1;
3172 else if (bargs->usage_max > 100)
3173 user_thresh_max = cache->key.offset;
3174 else
3175 user_thresh_max = div_factor_fine(cache->key.offset,
3176 bargs->usage_max);
3177
3178 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3179 ret = 0;
3180
3181 btrfs_put_block_group(cache);
3182 return ret;
3183 }
3184
3185 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3186 u64 chunk_offset, struct btrfs_balance_args *bargs)
3187 {
3188 struct btrfs_block_group_cache *cache;
3189 u64 chunk_used, user_thresh;
3190 int ret = 1;
3191
3192 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3193 chunk_used = btrfs_block_group_used(&cache->item);
3194
3195 if (bargs->usage_min == 0)
3196 user_thresh = 1;
3197 else if (bargs->usage > 100)
3198 user_thresh = cache->key.offset;
3199 else
3200 user_thresh = div_factor_fine(cache->key.offset,
3201 bargs->usage);
3202
3203 if (chunk_used < user_thresh)
3204 ret = 0;
3205
3206 btrfs_put_block_group(cache);
3207 return ret;
3208 }
3209
3210 static int chunk_devid_filter(struct extent_buffer *leaf,
3211 struct btrfs_chunk *chunk,
3212 struct btrfs_balance_args *bargs)
3213 {
3214 struct btrfs_stripe *stripe;
3215 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3216 int i;
3217
3218 for (i = 0; i < num_stripes; i++) {
3219 stripe = btrfs_stripe_nr(chunk, i);
3220 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3221 return 0;
3222 }
3223
3224 return 1;
3225 }
3226
3227 /* [pstart, pend) */
3228 static int chunk_drange_filter(struct extent_buffer *leaf,
3229 struct btrfs_chunk *chunk,
3230 u64 chunk_offset,
3231 struct btrfs_balance_args *bargs)
3232 {
3233 struct btrfs_stripe *stripe;
3234 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3235 u64 stripe_offset;
3236 u64 stripe_length;
3237 int factor;
3238 int i;
3239
3240 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3241 return 0;
3242
3243 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3244 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3245 factor = num_stripes / 2;
3246 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3247 factor = num_stripes - 1;
3248 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3249 factor = num_stripes - 2;
3250 } else {
3251 factor = num_stripes;
3252 }
3253
3254 for (i = 0; i < num_stripes; i++) {
3255 stripe = btrfs_stripe_nr(chunk, i);
3256 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3257 continue;
3258
3259 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3260 stripe_length = btrfs_chunk_length(leaf, chunk);
3261 stripe_length = div_u64(stripe_length, factor);
3262
3263 if (stripe_offset < bargs->pend &&
3264 stripe_offset + stripe_length > bargs->pstart)
3265 return 0;
3266 }
3267
3268 return 1;
3269 }
3270
3271 /* [vstart, vend) */
3272 static int chunk_vrange_filter(struct extent_buffer *leaf,
3273 struct btrfs_chunk *chunk,
3274 u64 chunk_offset,
3275 struct btrfs_balance_args *bargs)
3276 {
3277 if (chunk_offset < bargs->vend &&
3278 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3279 /* at least part of the chunk is inside this vrange */
3280 return 0;
3281
3282 return 1;
3283 }
3284
3285 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3286 struct btrfs_chunk *chunk,
3287 struct btrfs_balance_args *bargs)
3288 {
3289 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3290
3291 if (bargs->stripes_min <= num_stripes
3292 && num_stripes <= bargs->stripes_max)
3293 return 0;
3294
3295 return 1;
3296 }
3297
3298 static int chunk_soft_convert_filter(u64 chunk_type,
3299 struct btrfs_balance_args *bargs)
3300 {
3301 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3302 return 0;
3303
3304 chunk_type = chunk_to_extended(chunk_type) &
3305 BTRFS_EXTENDED_PROFILE_MASK;
3306
3307 if (bargs->target == chunk_type)
3308 return 1;
3309
3310 return 0;
3311 }
3312
3313 static int should_balance_chunk(struct btrfs_root *root,
3314 struct extent_buffer *leaf,
3315 struct btrfs_chunk *chunk, u64 chunk_offset)
3316 {
3317 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3318 struct btrfs_balance_args *bargs = NULL;
3319 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3320
3321 /* type filter */
3322 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3323 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3324 return 0;
3325 }
3326
3327 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3328 bargs = &bctl->data;
3329 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3330 bargs = &bctl->sys;
3331 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3332 bargs = &bctl->meta;
3333
3334 /* profiles filter */
3335 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3336 chunk_profiles_filter(chunk_type, bargs)) {
3337 return 0;
3338 }
3339
3340 /* usage filter */
3341 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3342 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3343 return 0;
3344 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3345 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3346 return 0;
3347 }
3348
3349 /* devid filter */
3350 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3351 chunk_devid_filter(leaf, chunk, bargs)) {
3352 return 0;
3353 }
3354
3355 /* drange filter, makes sense only with devid filter */
3356 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3357 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3358 return 0;
3359 }
3360
3361 /* vrange filter */
3362 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3363 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3364 return 0;
3365 }
3366
3367 /* stripes filter */
3368 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3369 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3370 return 0;
3371 }
3372
3373 /* soft profile changing mode */
3374 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3375 chunk_soft_convert_filter(chunk_type, bargs)) {
3376 return 0;
3377 }
3378
3379 /*
3380 * limited by count, must be the last filter
3381 */
3382 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3383 if (bargs->limit == 0)
3384 return 0;
3385 else
3386 bargs->limit--;
3387 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3388 /*
3389 * Same logic as the 'limit' filter; the minimum cannot be
3390 * determined here because we do not have the global information
3391 * about the count of all chunks that satisfy the filters.
3392 */
3393 if (bargs->limit_max == 0)
3394 return 0;
3395 else
3396 bargs->limit_max--;
3397 }
3398
3399 return 1;
3400 }
3401
3402 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3403 {
3404 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3405 struct btrfs_root *chunk_root = fs_info->chunk_root;
3406 struct btrfs_root *dev_root = fs_info->dev_root;
3407 struct list_head *devices;
3408 struct btrfs_device *device;
3409 u64 old_size;
3410 u64 size_to_free;
3411 u64 chunk_type;
3412 struct btrfs_chunk *chunk;
3413 struct btrfs_path *path;
3414 struct btrfs_key key;
3415 struct btrfs_key found_key;
3416 struct btrfs_trans_handle *trans;
3417 struct extent_buffer *leaf;
3418 int slot;
3419 int ret;
3420 int enospc_errors = 0;
3421 bool counting = true;
3422 /* The single value limit and min/max limits use the same bytes in the */
3423 u64 limit_data = bctl->data.limit;
3424 u64 limit_meta = bctl->meta.limit;
3425 u64 limit_sys = bctl->sys.limit;
3426 u32 count_data = 0;
3427 u32 count_meta = 0;
3428 u32 count_sys = 0;
3429 int chunk_reserved = 0;
3430 u64 bytes_used = 0;
3431
3432 /* step one make some room on all the devices */
3433 devices = &fs_info->fs_devices->devices;
3434 list_for_each_entry(device, devices, dev_list) {
3435 old_size = btrfs_device_get_total_bytes(device);
3436 size_to_free = div_factor(old_size, 1);
3437 size_to_free = min_t(u64, size_to_free, SZ_1M);
3438 if (!device->writeable ||
3439 btrfs_device_get_total_bytes(device) -
3440 btrfs_device_get_bytes_used(device) > size_to_free ||
3441 device->is_tgtdev_for_dev_replace)
3442 continue;
3443
3444 ret = btrfs_shrink_device(device, old_size - size_to_free);
3445 if (ret == -ENOSPC)
3446 break;
3447 BUG_ON(ret);
3448
3449 trans = btrfs_start_transaction(dev_root, 0);
3450 BUG_ON(IS_ERR(trans));
3451
3452 ret = btrfs_grow_device(trans, device, old_size);
3453 BUG_ON(ret);
3454
3455 btrfs_end_transaction(trans, dev_root);
3456 }
3457
3458 /* step two, relocate all the chunks */
3459 path = btrfs_alloc_path();
3460 if (!path) {
3461 ret = -ENOMEM;
3462 goto error;
3463 }
3464
3465 /* zero out stat counters */
3466 spin_lock(&fs_info->balance_lock);
3467 memset(&bctl->stat, 0, sizeof(bctl->stat));
3468 spin_unlock(&fs_info->balance_lock);
3469 again:
3470 if (!counting) {
3471 /*
3472 * The single value limit and min/max limits use the same bytes
3473 * in the
3474 */
3475 bctl->data.limit = limit_data;
3476 bctl->meta.limit = limit_meta;
3477 bctl->sys.limit = limit_sys;
3478 }
3479 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3480 key.offset = (u64)-1;
3481 key.type = BTRFS_CHUNK_ITEM_KEY;
3482
3483 while (1) {
3484 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3485 atomic_read(&fs_info->balance_cancel_req)) {
3486 ret = -ECANCELED;
3487 goto error;
3488 }
3489
3490 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3491 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3492 if (ret < 0) {
3493 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3494 goto error;
3495 }
3496
3497 /*
3498 * this shouldn't happen, it means the last relocate
3499 * failed
3500 */
3501 if (ret == 0)
3502 BUG(); /* FIXME break ? */
3503
3504 ret = btrfs_previous_item(chunk_root, path, 0,
3505 BTRFS_CHUNK_ITEM_KEY);
3506 if (ret) {
3507 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3508 ret = 0;
3509 break;
3510 }
3511
3512 leaf = path->nodes[0];
3513 slot = path->slots[0];
3514 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3515
3516 if (found_key.objectid != key.objectid) {
3517 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3518 break;
3519 }
3520
3521 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3522 chunk_type = btrfs_chunk_type(leaf, chunk);
3523
3524 if (!counting) {
3525 spin_lock(&fs_info->balance_lock);
3526 bctl->stat.considered++;
3527 spin_unlock(&fs_info->balance_lock);
3528 }
3529
3530 ret = should_balance_chunk(chunk_root, leaf, chunk,
3531 found_key.offset);
3532
3533 btrfs_release_path(path);
3534 if (!ret) {
3535 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3536 goto loop;
3537 }
3538
3539 if (counting) {
3540 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3541 spin_lock(&fs_info->balance_lock);
3542 bctl->stat.expected++;
3543 spin_unlock(&fs_info->balance_lock);
3544
3545 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3546 count_data++;
3547 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3548 count_sys++;
3549 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3550 count_meta++;
3551
3552 goto loop;
3553 }
3554
3555 /*
3556 * Apply limit_min filter, no need to check if the LIMITS
3557 * filter is used, limit_min is 0 by default
3558 */
3559 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3560 count_data < bctl->data.limit_min)
3561 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3562 count_meta < bctl->meta.limit_min)
3563 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3564 count_sys < bctl->sys.limit_min)) {
3565 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3566 goto loop;
3567 }
3568
3569 ASSERT(fs_info->data_sinfo);
3570 spin_lock(&fs_info->data_sinfo->lock);
3571 bytes_used = fs_info->data_sinfo->bytes_used;
3572 spin_unlock(&fs_info->data_sinfo->lock);
3573
3574 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3575 !chunk_reserved && !bytes_used) {
3576 trans = btrfs_start_transaction(chunk_root, 0);
3577 if (IS_ERR(trans)) {
3578 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3579 ret = PTR_ERR(trans);
3580 goto error;
3581 }
3582
3583 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3584 BTRFS_BLOCK_GROUP_DATA);
3585 btrfs_end_transaction(trans, chunk_root);
3586 if (ret < 0) {
3587 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3588 goto error;
3589 }
3590 chunk_reserved = 1;
3591 }
3592
3593 ret = btrfs_relocate_chunk(chunk_root,
3594 found_key.offset);
3595 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3596 if (ret && ret != -ENOSPC)
3597 goto error;
3598 if (ret == -ENOSPC) {
3599 enospc_errors++;
3600 } else {
3601 spin_lock(&fs_info->balance_lock);
3602 bctl->stat.completed++;
3603 spin_unlock(&fs_info->balance_lock);
3604 }
3605 loop:
3606 if (found_key.offset == 0)
3607 break;
3608 key.offset = found_key.offset - 1;
3609 }
3610
3611 if (counting) {
3612 btrfs_release_path(path);
3613 counting = false;
3614 goto again;
3615 }
3616 error:
3617 btrfs_free_path(path);
3618 if (enospc_errors) {
3619 btrfs_info(fs_info, "%d enospc errors during balance",
3620 enospc_errors);
3621 if (!ret)
3622 ret = -ENOSPC;
3623 }
3624
3625 return ret;
3626 }
3627
3628 /**
3629 * alloc_profile_is_valid - see if a given profile is valid and reduced
3630 * @flags: profile to validate
3631 * @extended: if true @flags is treated as an extended profile
3632 */
3633 static int alloc_profile_is_valid(u64 flags, int extended)
3634 {
3635 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3636 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3637
3638 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3639
3640 /* 1) check that all other bits are zeroed */
3641 if (flags & ~mask)
3642 return 0;
3643
3644 /* 2) see if profile is reduced */
3645 if (flags == 0)
3646 return !extended; /* "0" is valid for usual profiles */
3647
3648 /* true if exactly one bit set */
3649 return (flags & (flags - 1)) == 0;
3650 }
3651
3652 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3653 {
3654 /* cancel requested || normal exit path */
3655 return atomic_read(&fs_info->balance_cancel_req) ||
3656 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3657 atomic_read(&fs_info->balance_cancel_req) == 0);
3658 }
3659
3660 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3661 {
3662 int ret;
3663
3664 unset_balance_control(fs_info);
3665 ret = del_balance_item(fs_info->tree_root);
3666 if (ret)
3667 btrfs_handle_fs_error(fs_info, ret, NULL);
3668
3669 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3670 }
3671
3672 /* Non-zero return value signifies invalidity */
3673 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3674 u64 allowed)
3675 {
3676 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3677 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3678 (bctl_arg->target & ~allowed)));
3679 }
3680
3681 /*
3682 * Should be called with both balance and volume mutexes held
3683 */
3684 int btrfs_balance(struct btrfs_balance_control *bctl,
3685 struct btrfs_ioctl_balance_args *bargs)
3686 {
3687 struct btrfs_fs_info *fs_info = bctl->fs_info;
3688 u64 allowed;
3689 int mixed = 0;
3690 int ret;
3691 u64 num_devices;
3692 unsigned seq;
3693
3694 if (btrfs_fs_closing(fs_info) ||
3695 atomic_read(&fs_info->balance_pause_req) ||
3696 atomic_read(&fs_info->balance_cancel_req)) {
3697 ret = -EINVAL;
3698 goto out;
3699 }
3700
3701 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3702 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3703 mixed = 1;
3704
3705 /*
3706 * In case of mixed groups both data and meta should be picked,
3707 * and identical options should be given for both of them.
3708 */
3709 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3710 if (mixed && (bctl->flags & allowed)) {
3711 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3712 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3713 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3714 btrfs_err(fs_info, "with mixed groups data and "
3715 "metadata balance options must be the same");
3716 ret = -EINVAL;
3717 goto out;
3718 }
3719 }
3720
3721 num_devices = fs_info->fs_devices->num_devices;
3722 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3723 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3724 BUG_ON(num_devices < 1);
3725 num_devices--;
3726 }
3727 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3728 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3729 if (num_devices > 1)
3730 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3731 if (num_devices > 2)
3732 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3733 if (num_devices > 3)
3734 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3735 BTRFS_BLOCK_GROUP_RAID6);
3736 if (validate_convert_profile(&bctl->data, allowed)) {
3737 btrfs_err(fs_info, "unable to start balance with target "
3738 "data profile %llu",
3739 bctl->data.target);
3740 ret = -EINVAL;
3741 goto out;
3742 }
3743 if (validate_convert_profile(&bctl->meta, allowed)) {
3744 btrfs_err(fs_info,
3745 "unable to start balance with target metadata profile %llu",
3746 bctl->meta.target);
3747 ret = -EINVAL;
3748 goto out;
3749 }
3750 if (validate_convert_profile(&bctl->sys, allowed)) {
3751 btrfs_err(fs_info,
3752 "unable to start balance with target system profile %llu",
3753 bctl->sys.target);
3754 ret = -EINVAL;
3755 goto out;
3756 }
3757
3758 /* allow to reduce meta or sys integrity only if force set */
3759 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3760 BTRFS_BLOCK_GROUP_RAID10 |
3761 BTRFS_BLOCK_GROUP_RAID5 |
3762 BTRFS_BLOCK_GROUP_RAID6;
3763 do {
3764 seq = read_seqbegin(&fs_info->profiles_lock);
3765
3766 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767 (fs_info->avail_system_alloc_bits & allowed) &&
3768 !(bctl->sys.target & allowed)) ||
3769 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3770 (fs_info->avail_metadata_alloc_bits & allowed) &&
3771 !(bctl->meta.target & allowed))) {
3772 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3773 btrfs_info(fs_info, "force reducing metadata integrity");
3774 } else {
3775 btrfs_err(fs_info, "balance will reduce metadata "
3776 "integrity, use force if you want this");
3777 ret = -EINVAL;
3778 goto out;
3779 }
3780 }
3781 } while (read_seqretry(&fs_info->profiles_lock, seq));
3782
3783 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3784 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3785 btrfs_warn(fs_info,
3786 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3787 bctl->meta.target, bctl->data.target);
3788 }
3789
3790 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3791 fs_info->num_tolerated_disk_barrier_failures = min(
3792 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3793 btrfs_get_num_tolerated_disk_barrier_failures(
3794 bctl->sys.target));
3795 }
3796
3797 ret = insert_balance_item(fs_info->tree_root, bctl);
3798 if (ret && ret != -EEXIST)
3799 goto out;
3800
3801 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3802 BUG_ON(ret == -EEXIST);
3803 set_balance_control(bctl);
3804 } else {
3805 BUG_ON(ret != -EEXIST);
3806 spin_lock(&fs_info->balance_lock);
3807 update_balance_args(bctl);
3808 spin_unlock(&fs_info->balance_lock);
3809 }
3810
3811 atomic_inc(&fs_info->balance_running);
3812 mutex_unlock(&fs_info->balance_mutex);
3813
3814 ret = __btrfs_balance(fs_info);
3815
3816 mutex_lock(&fs_info->balance_mutex);
3817 atomic_dec(&fs_info->balance_running);
3818
3819 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3820 fs_info->num_tolerated_disk_barrier_failures =
3821 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3822 }
3823
3824 if (bargs) {
3825 memset(bargs, 0, sizeof(*bargs));
3826 update_ioctl_balance_args(fs_info, 0, bargs);
3827 }
3828
3829 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3830 balance_need_close(fs_info)) {
3831 __cancel_balance(fs_info);
3832 }
3833
3834 wake_up(&fs_info->balance_wait_q);
3835
3836 return ret;
3837 out:
3838 if (bctl->flags & BTRFS_BALANCE_RESUME)
3839 __cancel_balance(fs_info);
3840 else {
3841 kfree(bctl);
3842 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3843 }
3844 return ret;
3845 }
3846
3847 static int balance_kthread(void *data)
3848 {
3849 struct btrfs_fs_info *fs_info = data;
3850 int ret = 0;
3851
3852 mutex_lock(&fs_info->volume_mutex);
3853 mutex_lock(&fs_info->balance_mutex);
3854
3855 if (fs_info->balance_ctl) {
3856 btrfs_info(fs_info, "continuing balance");
3857 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3858 }
3859
3860 mutex_unlock(&fs_info->balance_mutex);
3861 mutex_unlock(&fs_info->volume_mutex);
3862
3863 return ret;
3864 }
3865
3866 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3867 {
3868 struct task_struct *tsk;
3869
3870 spin_lock(&fs_info->balance_lock);
3871 if (!fs_info->balance_ctl) {
3872 spin_unlock(&fs_info->balance_lock);
3873 return 0;
3874 }
3875 spin_unlock(&fs_info->balance_lock);
3876
3877 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3878 btrfs_info(fs_info, "force skipping balance");
3879 return 0;
3880 }
3881
3882 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3883 return PTR_ERR_OR_ZERO(tsk);
3884 }
3885
3886 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3887 {
3888 struct btrfs_balance_control *bctl;
3889 struct btrfs_balance_item *item;
3890 struct btrfs_disk_balance_args disk_bargs;
3891 struct btrfs_path *path;
3892 struct extent_buffer *leaf;
3893 struct btrfs_key key;
3894 int ret;
3895
3896 path = btrfs_alloc_path();
3897 if (!path)
3898 return -ENOMEM;
3899
3900 key.objectid = BTRFS_BALANCE_OBJECTID;
3901 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3902 key.offset = 0;
3903
3904 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3905 if (ret < 0)
3906 goto out;
3907 if (ret > 0) { /* ret = -ENOENT; */
3908 ret = 0;
3909 goto out;
3910 }
3911
3912 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3913 if (!bctl) {
3914 ret = -ENOMEM;
3915 goto out;
3916 }
3917
3918 leaf = path->nodes[0];
3919 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3920
3921 bctl->fs_info = fs_info;
3922 bctl->flags = btrfs_balance_flags(leaf, item);
3923 bctl->flags |= BTRFS_BALANCE_RESUME;
3924
3925 btrfs_balance_data(leaf, item, &disk_bargs);
3926 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3927 btrfs_balance_meta(leaf, item, &disk_bargs);
3928 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3929 btrfs_balance_sys(leaf, item, &disk_bargs);
3930 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3931
3932 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3933
3934 mutex_lock(&fs_info->volume_mutex);
3935 mutex_lock(&fs_info->balance_mutex);
3936
3937 set_balance_control(bctl);
3938
3939 mutex_unlock(&fs_info->balance_mutex);
3940 mutex_unlock(&fs_info->volume_mutex);
3941 out:
3942 btrfs_free_path(path);
3943 return ret;
3944 }
3945
3946 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3947 {
3948 int ret = 0;
3949
3950 mutex_lock(&fs_info->balance_mutex);
3951 if (!fs_info->balance_ctl) {
3952 mutex_unlock(&fs_info->balance_mutex);
3953 return -ENOTCONN;
3954 }
3955
3956 if (atomic_read(&fs_info->balance_running)) {
3957 atomic_inc(&fs_info->balance_pause_req);
3958 mutex_unlock(&fs_info->balance_mutex);
3959
3960 wait_event(fs_info->balance_wait_q,
3961 atomic_read(&fs_info->balance_running) == 0);
3962
3963 mutex_lock(&fs_info->balance_mutex);
3964 /* we are good with balance_ctl ripped off from under us */
3965 BUG_ON(atomic_read(&fs_info->balance_running));
3966 atomic_dec(&fs_info->balance_pause_req);
3967 } else {
3968 ret = -ENOTCONN;
3969 }
3970
3971 mutex_unlock(&fs_info->balance_mutex);
3972 return ret;
3973 }
3974
3975 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3976 {
3977 if (fs_info->sb->s_flags & MS_RDONLY)
3978 return -EROFS;
3979
3980 mutex_lock(&fs_info->balance_mutex);
3981 if (!fs_info->balance_ctl) {
3982 mutex_unlock(&fs_info->balance_mutex);
3983 return -ENOTCONN;
3984 }
3985
3986 atomic_inc(&fs_info->balance_cancel_req);
3987 /*
3988 * if we are running just wait and return, balance item is
3989 * deleted in btrfs_balance in this case
3990 */
3991 if (atomic_read(&fs_info->balance_running)) {
3992 mutex_unlock(&fs_info->balance_mutex);
3993 wait_event(fs_info->balance_wait_q,
3994 atomic_read(&fs_info->balance_running) == 0);
3995 mutex_lock(&fs_info->balance_mutex);
3996 } else {
3997 /* __cancel_balance needs volume_mutex */
3998 mutex_unlock(&fs_info->balance_mutex);
3999 mutex_lock(&fs_info->volume_mutex);
4000 mutex_lock(&fs_info->balance_mutex);
4001
4002 if (fs_info->balance_ctl)
4003 __cancel_balance(fs_info);
4004
4005 mutex_unlock(&fs_info->volume_mutex);
4006 }
4007
4008 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4009 atomic_dec(&fs_info->balance_cancel_req);
4010 mutex_unlock(&fs_info->balance_mutex);
4011 return 0;
4012 }
4013
4014 static int btrfs_uuid_scan_kthread(void *data)
4015 {
4016 struct btrfs_fs_info *fs_info = data;
4017 struct btrfs_root *root = fs_info->tree_root;
4018 struct btrfs_key key;
4019 struct btrfs_key max_key;
4020 struct btrfs_path *path = NULL;
4021 int ret = 0;
4022 struct extent_buffer *eb;
4023 int slot;
4024 struct btrfs_root_item root_item;
4025 u32 item_size;
4026 struct btrfs_trans_handle *trans = NULL;
4027
4028 path = btrfs_alloc_path();
4029 if (!path) {
4030 ret = -ENOMEM;
4031 goto out;
4032 }
4033
4034 key.objectid = 0;
4035 key.type = BTRFS_ROOT_ITEM_KEY;
4036 key.offset = 0;
4037
4038 max_key.objectid = (u64)-1;
4039 max_key.type = BTRFS_ROOT_ITEM_KEY;
4040 max_key.offset = (u64)-1;
4041
4042 while (1) {
4043 ret = btrfs_search_forward(root, &key, path, 0);
4044 if (ret) {
4045 if (ret > 0)
4046 ret = 0;
4047 break;
4048 }
4049
4050 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4051 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4052 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4053 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4054 goto skip;
4055
4056 eb = path->nodes[0];
4057 slot = path->slots[0];
4058 item_size = btrfs_item_size_nr(eb, slot);
4059 if (item_size < sizeof(root_item))
4060 goto skip;
4061
4062 read_extent_buffer(eb, &root_item,
4063 btrfs_item_ptr_offset(eb, slot),
4064 (int)sizeof(root_item));
4065 if (btrfs_root_refs(&root_item) == 0)
4066 goto skip;
4067
4068 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4069 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4070 if (trans)
4071 goto update_tree;
4072
4073 btrfs_release_path(path);
4074 /*
4075 * 1 - subvol uuid item
4076 * 1 - received_subvol uuid item
4077 */
4078 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4079 if (IS_ERR(trans)) {
4080 ret = PTR_ERR(trans);
4081 break;
4082 }
4083 continue;
4084 } else {
4085 goto skip;
4086 }
4087 update_tree:
4088 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4089 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4090 root_item.uuid,
4091 BTRFS_UUID_KEY_SUBVOL,
4092 key.objectid);
4093 if (ret < 0) {
4094 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4095 ret);
4096 break;
4097 }
4098 }
4099
4100 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4101 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4102 root_item.received_uuid,
4103 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4104 key.objectid);
4105 if (ret < 0) {
4106 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4107 ret);
4108 break;
4109 }
4110 }
4111
4112 skip:
4113 if (trans) {
4114 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4115 trans = NULL;
4116 if (ret)
4117 break;
4118 }
4119
4120 btrfs_release_path(path);
4121 if (key.offset < (u64)-1) {
4122 key.offset++;
4123 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4124 key.offset = 0;
4125 key.type = BTRFS_ROOT_ITEM_KEY;
4126 } else if (key.objectid < (u64)-1) {
4127 key.offset = 0;
4128 key.type = BTRFS_ROOT_ITEM_KEY;
4129 key.objectid++;
4130 } else {
4131 break;
4132 }
4133 cond_resched();
4134 }
4135
4136 out:
4137 btrfs_free_path(path);
4138 if (trans && !IS_ERR(trans))
4139 btrfs_end_transaction(trans, fs_info->uuid_root);
4140 if (ret)
4141 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4142 else
4143 fs_info->update_uuid_tree_gen = 1;
4144 up(&fs_info->uuid_tree_rescan_sem);
4145 return 0;
4146 }
4147
4148 /*
4149 * Callback for btrfs_uuid_tree_iterate().
4150 * returns:
4151 * 0 check succeeded, the entry is not outdated.
4152 * < 0 if an error occurred.
4153 * > 0 if the check failed, which means the caller shall remove the entry.
4154 */
4155 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4156 u8 *uuid, u8 type, u64 subid)
4157 {
4158 struct btrfs_key key;
4159 int ret = 0;
4160 struct btrfs_root *subvol_root;
4161
4162 if (type != BTRFS_UUID_KEY_SUBVOL &&
4163 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4164 goto out;
4165
4166 key.objectid = subid;
4167 key.type = BTRFS_ROOT_ITEM_KEY;
4168 key.offset = (u64)-1;
4169 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4170 if (IS_ERR(subvol_root)) {
4171 ret = PTR_ERR(subvol_root);
4172 if (ret == -ENOENT)
4173 ret = 1;
4174 goto out;
4175 }
4176
4177 switch (type) {
4178 case BTRFS_UUID_KEY_SUBVOL:
4179 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4180 ret = 1;
4181 break;
4182 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4183 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4184 BTRFS_UUID_SIZE))
4185 ret = 1;
4186 break;
4187 }
4188
4189 out:
4190 return ret;
4191 }
4192
4193 static int btrfs_uuid_rescan_kthread(void *data)
4194 {
4195 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4196 int ret;
4197
4198 /*
4199 * 1st step is to iterate through the existing UUID tree and
4200 * to delete all entries that contain outdated data.
4201 * 2nd step is to add all missing entries to the UUID tree.
4202 */
4203 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4204 if (ret < 0) {
4205 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4206 up(&fs_info->uuid_tree_rescan_sem);
4207 return ret;
4208 }
4209 return btrfs_uuid_scan_kthread(data);
4210 }
4211
4212 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4213 {
4214 struct btrfs_trans_handle *trans;
4215 struct btrfs_root *tree_root = fs_info->tree_root;
4216 struct btrfs_root *uuid_root;
4217 struct task_struct *task;
4218 int ret;
4219
4220 /*
4221 * 1 - root node
4222 * 1 - root item
4223 */
4224 trans = btrfs_start_transaction(tree_root, 2);
4225 if (IS_ERR(trans))
4226 return PTR_ERR(trans);
4227
4228 uuid_root = btrfs_create_tree(trans, fs_info,
4229 BTRFS_UUID_TREE_OBJECTID);
4230 if (IS_ERR(uuid_root)) {
4231 ret = PTR_ERR(uuid_root);
4232 btrfs_abort_transaction(trans, tree_root, ret);
4233 return ret;
4234 }
4235
4236 fs_info->uuid_root = uuid_root;
4237
4238 ret = btrfs_commit_transaction(trans, tree_root);
4239 if (ret)
4240 return ret;
4241
4242 down(&fs_info->uuid_tree_rescan_sem);
4243 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4244 if (IS_ERR(task)) {
4245 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4246 btrfs_warn(fs_info, "failed to start uuid_scan task");
4247 up(&fs_info->uuid_tree_rescan_sem);
4248 return PTR_ERR(task);
4249 }
4250
4251 return 0;
4252 }
4253
4254 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4255 {
4256 struct task_struct *task;
4257
4258 down(&fs_info->uuid_tree_rescan_sem);
4259 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4260 if (IS_ERR(task)) {
4261 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4262 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4263 up(&fs_info->uuid_tree_rescan_sem);
4264 return PTR_ERR(task);
4265 }
4266
4267 return 0;
4268 }
4269
4270 /*
4271 * shrinking a device means finding all of the device extents past
4272 * the new size, and then following the back refs to the chunks.
4273 * The chunk relocation code actually frees the device extent
4274 */
4275 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4276 {
4277 struct btrfs_trans_handle *trans;
4278 struct btrfs_root *root = device->dev_root;
4279 struct btrfs_dev_extent *dev_extent = NULL;
4280 struct btrfs_path *path;
4281 u64 length;
4282 u64 chunk_offset;
4283 int ret;
4284 int slot;
4285 int failed = 0;
4286 bool retried = false;
4287 bool checked_pending_chunks = false;
4288 struct extent_buffer *l;
4289 struct btrfs_key key;
4290 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4291 u64 old_total = btrfs_super_total_bytes(super_copy);
4292 u64 old_size = btrfs_device_get_total_bytes(device);
4293 u64 diff = old_size - new_size;
4294
4295 if (device->is_tgtdev_for_dev_replace)
4296 return -EINVAL;
4297
4298 path = btrfs_alloc_path();
4299 if (!path)
4300 return -ENOMEM;
4301
4302 path->reada = READA_FORWARD;
4303
4304 lock_chunks(root);
4305
4306 btrfs_device_set_total_bytes(device, new_size);
4307 if (device->writeable) {
4308 device->fs_devices->total_rw_bytes -= diff;
4309 spin_lock(&root->fs_info->free_chunk_lock);
4310 root->fs_info->free_chunk_space -= diff;
4311 spin_unlock(&root->fs_info->free_chunk_lock);
4312 }
4313 unlock_chunks(root);
4314
4315 again:
4316 key.objectid = device->devid;
4317 key.offset = (u64)-1;
4318 key.type = BTRFS_DEV_EXTENT_KEY;
4319
4320 do {
4321 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4322 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4323 if (ret < 0) {
4324 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4325 goto done;
4326 }
4327
4328 ret = btrfs_previous_item(root, path, 0, key.type);
4329 if (ret)
4330 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4331 if (ret < 0)
4332 goto done;
4333 if (ret) {
4334 ret = 0;
4335 btrfs_release_path(path);
4336 break;
4337 }
4338
4339 l = path->nodes[0];
4340 slot = path->slots[0];
4341 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4342
4343 if (key.objectid != device->devid) {
4344 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4345 btrfs_release_path(path);
4346 break;
4347 }
4348
4349 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4350 length = btrfs_dev_extent_length(l, dev_extent);
4351
4352 if (key.offset + length <= new_size) {
4353 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4354 btrfs_release_path(path);
4355 break;
4356 }
4357
4358 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4359 btrfs_release_path(path);
4360
4361 ret = btrfs_relocate_chunk(root, chunk_offset);
4362 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4363 if (ret && ret != -ENOSPC)
4364 goto done;
4365 if (ret == -ENOSPC)
4366 failed++;
4367 } while (key.offset-- > 0);
4368
4369 if (failed && !retried) {
4370 failed = 0;
4371 retried = true;
4372 goto again;
4373 } else if (failed && retried) {
4374 ret = -ENOSPC;
4375 goto done;
4376 }
4377
4378 /* Shrinking succeeded, else we would be at "done". */
4379 trans = btrfs_start_transaction(root, 0);
4380 if (IS_ERR(trans)) {
4381 ret = PTR_ERR(trans);
4382 goto done;
4383 }
4384
4385 lock_chunks(root);
4386
4387 /*
4388 * We checked in the above loop all device extents that were already in
4389 * the device tree. However before we have updated the device's
4390 * total_bytes to the new size, we might have had chunk allocations that
4391 * have not complete yet (new block groups attached to transaction
4392 * handles), and therefore their device extents were not yet in the
4393 * device tree and we missed them in the loop above. So if we have any
4394 * pending chunk using a device extent that overlaps the device range
4395 * that we can not use anymore, commit the current transaction and
4396 * repeat the search on the device tree - this way we guarantee we will
4397 * not have chunks using device extents that end beyond 'new_size'.
4398 */
4399 if (!checked_pending_chunks) {
4400 u64 start = new_size;
4401 u64 len = old_size - new_size;
4402
4403 if (contains_pending_extent(trans->transaction, device,
4404 &start, len)) {
4405 unlock_chunks(root);
4406 checked_pending_chunks = true;
4407 failed = 0;
4408 retried = false;
4409 ret = btrfs_commit_transaction(trans, root);
4410 if (ret)
4411 goto done;
4412 goto again;
4413 }
4414 }
4415
4416 btrfs_device_set_disk_total_bytes(device, new_size);
4417 if (list_empty(&device->resized_list))
4418 list_add_tail(&device->resized_list,
4419 &root->fs_info->fs_devices->resized_devices);
4420
4421 WARN_ON(diff > old_total);
4422 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4423 unlock_chunks(root);
4424
4425 /* Now btrfs_update_device() will change the on-disk size. */
4426 ret = btrfs_update_device(trans, device);
4427 btrfs_end_transaction(trans, root);
4428 done:
4429 btrfs_free_path(path);
4430 if (ret) {
4431 lock_chunks(root);
4432 btrfs_device_set_total_bytes(device, old_size);
4433 if (device->writeable)
4434 device->fs_devices->total_rw_bytes += diff;
4435 spin_lock(&root->fs_info->free_chunk_lock);
4436 root->fs_info->free_chunk_space += diff;
4437 spin_unlock(&root->fs_info->free_chunk_lock);
4438 unlock_chunks(root);
4439 }
4440 return ret;
4441 }
4442
4443 static int btrfs_add_system_chunk(struct btrfs_root *root,
4444 struct btrfs_key *key,
4445 struct btrfs_chunk *chunk, int item_size)
4446 {
4447 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4448 struct btrfs_disk_key disk_key;
4449 u32 array_size;
4450 u8 *ptr;
4451
4452 lock_chunks(root);
4453 array_size = btrfs_super_sys_array_size(super_copy);
4454 if (array_size + item_size + sizeof(disk_key)
4455 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4456 unlock_chunks(root);
4457 return -EFBIG;
4458 }
4459
4460 ptr = super_copy->sys_chunk_array + array_size;
4461 btrfs_cpu_key_to_disk(&disk_key, key);
4462 memcpy(ptr, &disk_key, sizeof(disk_key));
4463 ptr += sizeof(disk_key);
4464 memcpy(ptr, chunk, item_size);
4465 item_size += sizeof(disk_key);
4466 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4467 unlock_chunks(root);
4468
4469 return 0;
4470 }
4471
4472 /*
4473 * sort the devices in descending order by max_avail, total_avail
4474 */
4475 static int btrfs_cmp_device_info(const void *a, const void *b)
4476 {
4477 const struct btrfs_device_info *di_a = a;
4478 const struct btrfs_device_info *di_b = b;
4479
4480 if (di_a->max_avail > di_b->max_avail)
4481 return -1;
4482 if (di_a->max_avail < di_b->max_avail)
4483 return 1;
4484 if (di_a->total_avail > di_b->total_avail)
4485 return -1;
4486 if (di_a->total_avail < di_b->total_avail)
4487 return 1;
4488 return 0;
4489 }
4490
4491 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4492 {
4493 /* TODO allow them to set a preferred stripe size */
4494 return SZ_64K;
4495 }
4496
4497 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4498 {
4499 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4500 return;
4501
4502 btrfs_set_fs_incompat(info, RAID56);
4503 }
4504
4505 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4506 - sizeof(struct btrfs_item) \
4507 - sizeof(struct btrfs_chunk)) \
4508 / sizeof(struct btrfs_stripe) + 1)
4509
4510 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4511 - 2 * sizeof(struct btrfs_disk_key) \
4512 - 2 * sizeof(struct btrfs_chunk)) \
4513 / sizeof(struct btrfs_stripe) + 1)
4514
4515 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4516 struct btrfs_root *extent_root, u64 start,
4517 u64 type)
4518 {
4519 struct btrfs_fs_info *info = extent_root->fs_info;
4520 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4521 struct list_head *cur;
4522 struct map_lookup *map = NULL;
4523 struct extent_map_tree *em_tree;
4524 struct extent_map *em;
4525 struct btrfs_device_info *devices_info = NULL;
4526 u64 total_avail;
4527 int num_stripes; /* total number of stripes to allocate */
4528 int data_stripes; /* number of stripes that count for
4529 block group size */
4530 int sub_stripes; /* sub_stripes info for map */
4531 int dev_stripes; /* stripes per dev */
4532 int devs_max; /* max devs to use */
4533 int devs_min; /* min devs needed */
4534 int devs_increment; /* ndevs has to be a multiple of this */
4535 int ncopies; /* how many copies to data has */
4536 int ret;
4537 u64 max_stripe_size;
4538 u64 max_chunk_size;
4539 u64 stripe_size;
4540 u64 num_bytes;
4541 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4542 int ndevs;
4543 int i;
4544 int j;
4545 int index;
4546
4547 BUG_ON(!alloc_profile_is_valid(type, 0));
4548
4549 if (list_empty(&fs_devices->alloc_list))
4550 return -ENOSPC;
4551
4552 index = __get_raid_index(type);
4553
4554 sub_stripes = btrfs_raid_array[index].sub_stripes;
4555 dev_stripes = btrfs_raid_array[index].dev_stripes;
4556 devs_max = btrfs_raid_array[index].devs_max;
4557 devs_min = btrfs_raid_array[index].devs_min;
4558 devs_increment = btrfs_raid_array[index].devs_increment;
4559 ncopies = btrfs_raid_array[index].ncopies;
4560
4561 if (type & BTRFS_BLOCK_GROUP_DATA) {
4562 max_stripe_size = SZ_1G;
4563 max_chunk_size = 10 * max_stripe_size;
4564 if (!devs_max)
4565 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4566 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4567 /* for larger filesystems, use larger metadata chunks */
4568 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4569 max_stripe_size = SZ_1G;
4570 else
4571 max_stripe_size = SZ_256M;
4572 max_chunk_size = max_stripe_size;
4573 if (!devs_max)
4574 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4575 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4576 max_stripe_size = SZ_32M;
4577 max_chunk_size = 2 * max_stripe_size;
4578 if (!devs_max)
4579 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4580 } else {
4581 btrfs_err(info, "invalid chunk type 0x%llx requested",
4582 type);
4583 BUG_ON(1);
4584 }
4585
4586 /* we don't want a chunk larger than 10% of writeable space */
4587 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4588 max_chunk_size);
4589
4590 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4591 GFP_NOFS);
4592 if (!devices_info)
4593 return -ENOMEM;
4594
4595 cur = fs_devices->alloc_list.next;
4596
4597 /*
4598 * in the first pass through the devices list, we gather information
4599 * about the available holes on each device.
4600 */
4601 ndevs = 0;
4602 while (cur != &fs_devices->alloc_list) {
4603 struct btrfs_device *device;
4604 u64 max_avail;
4605 u64 dev_offset;
4606
4607 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4608
4609 cur = cur->next;
4610
4611 if (!device->writeable) {
4612 WARN(1, KERN_ERR
4613 "BTRFS: read-only device in alloc_list\n");
4614 continue;
4615 }
4616
4617 if (!device->in_fs_metadata ||
4618 device->is_tgtdev_for_dev_replace)
4619 continue;
4620
4621 if (device->total_bytes > device->bytes_used)
4622 total_avail = device->total_bytes - device->bytes_used;
4623 else
4624 total_avail = 0;
4625
4626 /* If there is no space on this device, skip it. */
4627 if (total_avail == 0)
4628 continue;
4629
4630 ret = find_free_dev_extent(trans, device,
4631 max_stripe_size * dev_stripes,
4632 &dev_offset, &max_avail);
4633 if (ret && ret != -ENOSPC)
4634 goto error;
4635
4636 if (ret == 0)
4637 max_avail = max_stripe_size * dev_stripes;
4638
4639 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4640 continue;
4641
4642 if (ndevs == fs_devices->rw_devices) {
4643 WARN(1, "%s: found more than %llu devices\n",
4644 __func__, fs_devices->rw_devices);
4645 break;
4646 }
4647 devices_info[ndevs].dev_offset = dev_offset;
4648 devices_info[ndevs].max_avail = max_avail;
4649 devices_info[ndevs].total_avail = total_avail;
4650 devices_info[ndevs].dev = device;
4651 ++ndevs;
4652 }
4653
4654 /*
4655 * now sort the devices by hole size / available space
4656 */
4657 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4658 btrfs_cmp_device_info, NULL);
4659
4660 /* round down to number of usable stripes */
4661 ndevs -= ndevs % devs_increment;
4662
4663 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4664 ret = -ENOSPC;
4665 goto error;
4666 }
4667
4668 if (devs_max && ndevs > devs_max)
4669 ndevs = devs_max;
4670 /*
4671 * the primary goal is to maximize the number of stripes, so use as many
4672 * devices as possible, even if the stripes are not maximum sized.
4673 */
4674 stripe_size = devices_info[ndevs-1].max_avail;
4675 num_stripes = ndevs * dev_stripes;
4676
4677 /*
4678 * this will have to be fixed for RAID1 and RAID10 over
4679 * more drives
4680 */
4681 data_stripes = num_stripes / ncopies;
4682
4683 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4684 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4685 btrfs_super_stripesize(info->super_copy));
4686 data_stripes = num_stripes - 1;
4687 }
4688 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4689 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4690 btrfs_super_stripesize(info->super_copy));
4691 data_stripes = num_stripes - 2;
4692 }
4693
4694 /*
4695 * Use the number of data stripes to figure out how big this chunk
4696 * is really going to be in terms of logical address space,
4697 * and compare that answer with the max chunk size
4698 */
4699 if (stripe_size * data_stripes > max_chunk_size) {
4700 u64 mask = (1ULL << 24) - 1;
4701
4702 stripe_size = div_u64(max_chunk_size, data_stripes);
4703
4704 /* bump the answer up to a 16MB boundary */
4705 stripe_size = (stripe_size + mask) & ~mask;
4706
4707 /* but don't go higher than the limits we found
4708 * while searching for free extents
4709 */
4710 if (stripe_size > devices_info[ndevs-1].max_avail)
4711 stripe_size = devices_info[ndevs-1].max_avail;
4712 }
4713
4714 stripe_size = div_u64(stripe_size, dev_stripes);
4715
4716 /* align to BTRFS_STRIPE_LEN */
4717 stripe_size = div_u64(stripe_size, raid_stripe_len);
4718 stripe_size *= raid_stripe_len;
4719
4720 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4721 if (!map) {
4722 ret = -ENOMEM;
4723 goto error;
4724 }
4725 map->num_stripes = num_stripes;
4726
4727 for (i = 0; i < ndevs; ++i) {
4728 for (j = 0; j < dev_stripes; ++j) {
4729 int s = i * dev_stripes + j;
4730 map->stripes[s].dev = devices_info[i].dev;
4731 map->stripes[s].physical = devices_info[i].dev_offset +
4732 j * stripe_size;
4733 }
4734 }
4735 map->sector_size = extent_root->sectorsize;
4736 map->stripe_len = raid_stripe_len;
4737 map->io_align = raid_stripe_len;
4738 map->io_width = raid_stripe_len;
4739 map->type = type;
4740 map->sub_stripes = sub_stripes;
4741
4742 num_bytes = stripe_size * data_stripes;
4743
4744 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4745
4746 em = alloc_extent_map();
4747 if (!em) {
4748 kfree(map);
4749 ret = -ENOMEM;
4750 goto error;
4751 }
4752 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4753 em->map_lookup = map;
4754 em->start = start;
4755 em->len = num_bytes;
4756 em->block_start = 0;
4757 em->block_len = em->len;
4758 em->orig_block_len = stripe_size;
4759
4760 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4761 write_lock(&em_tree->lock);
4762 ret = add_extent_mapping(em_tree, em, 0);
4763 if (!ret) {
4764 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4765 atomic_inc(&em->refs);
4766 }
4767 write_unlock(&em_tree->lock);
4768 if (ret) {
4769 free_extent_map(em);
4770 goto error;
4771 }
4772
4773 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4774 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4775 start, num_bytes);
4776 if (ret)
4777 goto error_del_extent;
4778
4779 for (i = 0; i < map->num_stripes; i++) {
4780 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4781 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4782 }
4783
4784 spin_lock(&extent_root->fs_info->free_chunk_lock);
4785 extent_root->fs_info->free_chunk_space -= (stripe_size *
4786 map->num_stripes);
4787 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4788
4789 free_extent_map(em);
4790 check_raid56_incompat_flag(extent_root->fs_info, type);
4791
4792 kfree(devices_info);
4793 return 0;
4794
4795 error_del_extent:
4796 write_lock(&em_tree->lock);
4797 remove_extent_mapping(em_tree, em);
4798 write_unlock(&em_tree->lock);
4799
4800 /* One for our allocation */
4801 free_extent_map(em);
4802 /* One for the tree reference */
4803 free_extent_map(em);
4804 /* One for the pending_chunks list reference */
4805 free_extent_map(em);
4806 error:
4807 kfree(devices_info);
4808 return ret;
4809 }
4810
4811 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4812 struct btrfs_root *extent_root,
4813 u64 chunk_offset, u64 chunk_size)
4814 {
4815 struct btrfs_key key;
4816 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4817 struct btrfs_device *device;
4818 struct btrfs_chunk *chunk;
4819 struct btrfs_stripe *stripe;
4820 struct extent_map_tree *em_tree;
4821 struct extent_map *em;
4822 struct map_lookup *map;
4823 size_t item_size;
4824 u64 dev_offset;
4825 u64 stripe_size;
4826 int i = 0;
4827 int ret = 0;
4828
4829 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4830 read_lock(&em_tree->lock);
4831 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4832 read_unlock(&em_tree->lock);
4833
4834 if (!em) {
4835 btrfs_crit(extent_root->fs_info, "unable to find logical "
4836 "%Lu len %Lu", chunk_offset, chunk_size);
4837 return -EINVAL;
4838 }
4839
4840 if (em->start != chunk_offset || em->len != chunk_size) {
4841 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4842 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4843 chunk_size, em->start, em->len);
4844 free_extent_map(em);
4845 return -EINVAL;
4846 }
4847
4848 map = em->map_lookup;
4849 item_size = btrfs_chunk_item_size(map->num_stripes);
4850 stripe_size = em->orig_block_len;
4851
4852 chunk = kzalloc(item_size, GFP_NOFS);
4853 if (!chunk) {
4854 ret = -ENOMEM;
4855 goto out;
4856 }
4857
4858 /*
4859 * Take the device list mutex to prevent races with the final phase of
4860 * a device replace operation that replaces the device object associated
4861 * with the map's stripes, because the device object's id can change
4862 * at any time during that final phase of the device replace operation
4863 * (dev-replace.c:btrfs_dev_replace_finishing()).
4864 */
4865 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4866 for (i = 0; i < map->num_stripes; i++) {
4867 device = map->stripes[i].dev;
4868 dev_offset = map->stripes[i].physical;
4869
4870 ret = btrfs_update_device(trans, device);
4871 if (ret)
4872 break;
4873 ret = btrfs_alloc_dev_extent(trans, device,
4874 chunk_root->root_key.objectid,
4875 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4876 chunk_offset, dev_offset,
4877 stripe_size);
4878 if (ret)
4879 break;
4880 }
4881 if (ret) {
4882 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4883 goto out;
4884 }
4885
4886 stripe = &chunk->stripe;
4887 for (i = 0; i < map->num_stripes; i++) {
4888 device = map->stripes[i].dev;
4889 dev_offset = map->stripes[i].physical;
4890
4891 btrfs_set_stack_stripe_devid(stripe, device->devid);
4892 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4893 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4894 stripe++;
4895 }
4896 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4897
4898 btrfs_set_stack_chunk_length(chunk, chunk_size);
4899 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4900 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4901 btrfs_set_stack_chunk_type(chunk, map->type);
4902 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4903 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4904 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4905 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4906 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4907
4908 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4909 key.type = BTRFS_CHUNK_ITEM_KEY;
4910 key.offset = chunk_offset;
4911
4912 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4913 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4914 /*
4915 * TODO: Cleanup of inserted chunk root in case of
4916 * failure.
4917 */
4918 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4919 item_size);
4920 }
4921
4922 out:
4923 kfree(chunk);
4924 free_extent_map(em);
4925 return ret;
4926 }
4927
4928 /*
4929 * Chunk allocation falls into two parts. The first part does works
4930 * that make the new allocated chunk useable, but not do any operation
4931 * that modifies the chunk tree. The second part does the works that
4932 * require modifying the chunk tree. This division is important for the
4933 * bootstrap process of adding storage to a seed btrfs.
4934 */
4935 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4936 struct btrfs_root *extent_root, u64 type)
4937 {
4938 u64 chunk_offset;
4939
4940 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4941 chunk_offset = find_next_chunk(extent_root->fs_info);
4942 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4943 }
4944
4945 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4946 struct btrfs_root *root,
4947 struct btrfs_device *device)
4948 {
4949 u64 chunk_offset;
4950 u64 sys_chunk_offset;
4951 u64 alloc_profile;
4952 struct btrfs_fs_info *fs_info = root->fs_info;
4953 struct btrfs_root *extent_root = fs_info->extent_root;
4954 int ret;
4955
4956 chunk_offset = find_next_chunk(fs_info);
4957 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4958 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4959 alloc_profile);
4960 if (ret)
4961 return ret;
4962
4963 sys_chunk_offset = find_next_chunk(root->fs_info);
4964 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4965 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4966 alloc_profile);
4967 return ret;
4968 }
4969
4970 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4971 {
4972 int max_errors;
4973
4974 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4975 BTRFS_BLOCK_GROUP_RAID10 |
4976 BTRFS_BLOCK_GROUP_RAID5 |
4977 BTRFS_BLOCK_GROUP_DUP)) {
4978 max_errors = 1;
4979 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4980 max_errors = 2;
4981 } else {
4982 max_errors = 0;
4983 }
4984
4985 return max_errors;
4986 }
4987
4988 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4989 {
4990 struct extent_map *em;
4991 struct map_lookup *map;
4992 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4993 int readonly = 0;
4994 int miss_ndevs = 0;
4995 int i;
4996
4997 read_lock(&map_tree->map_tree.lock);
4998 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4999 read_unlock(&map_tree->map_tree.lock);
5000 if (!em)
5001 return 1;
5002
5003 map = em->map_lookup;
5004 for (i = 0; i < map->num_stripes; i++) {
5005 if (map->stripes[i].dev->missing) {
5006 miss_ndevs++;
5007 continue;
5008 }
5009
5010 if (!map->stripes[i].dev->writeable) {
5011 readonly = 1;
5012 goto end;
5013 }
5014 }
5015
5016 /*
5017 * If the number of missing devices is larger than max errors,
5018 * we can not write the data into that chunk successfully, so
5019 * set it readonly.
5020 */
5021 if (miss_ndevs > btrfs_chunk_max_errors(map))
5022 readonly = 1;
5023 end:
5024 free_extent_map(em);
5025 return readonly;
5026 }
5027
5028 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5029 {
5030 extent_map_tree_init(&tree->map_tree);
5031 }
5032
5033 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5034 {
5035 struct extent_map *em;
5036
5037 while (1) {
5038 write_lock(&tree->map_tree.lock);
5039 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5040 if (em)
5041 remove_extent_mapping(&tree->map_tree, em);
5042 write_unlock(&tree->map_tree.lock);
5043 if (!em)
5044 break;
5045 /* once for us */
5046 free_extent_map(em);
5047 /* once for the tree */
5048 free_extent_map(em);
5049 }
5050 }
5051
5052 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5053 {
5054 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5055 struct extent_map *em;
5056 struct map_lookup *map;
5057 struct extent_map_tree *em_tree = &map_tree->map_tree;
5058 int ret;
5059
5060 read_lock(&em_tree->lock);
5061 em = lookup_extent_mapping(em_tree, logical, len);
5062 read_unlock(&em_tree->lock);
5063
5064 /*
5065 * We could return errors for these cases, but that could get ugly and
5066 * we'd probably do the same thing which is just not do anything else
5067 * and exit, so return 1 so the callers don't try to use other copies.
5068 */
5069 if (!em) {
5070 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5071 logical+len);
5072 return 1;
5073 }
5074
5075 if (em->start > logical || em->start + em->len < logical) {
5076 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5077 "%Lu-%Lu", logical, logical+len, em->start,
5078 em->start + em->len);
5079 free_extent_map(em);
5080 return 1;
5081 }
5082
5083 map = em->map_lookup;
5084 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5085 ret = map->num_stripes;
5086 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5087 ret = map->sub_stripes;
5088 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5089 ret = 2;
5090 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5091 ret = 3;
5092 else
5093 ret = 1;
5094 free_extent_map(em);
5095
5096 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5097 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5098 ret++;
5099 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5100
5101 return ret;
5102 }
5103
5104 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5105 struct btrfs_mapping_tree *map_tree,
5106 u64 logical)
5107 {
5108 struct extent_map *em;
5109 struct map_lookup *map;
5110 struct extent_map_tree *em_tree = &map_tree->map_tree;
5111 unsigned long len = root->sectorsize;
5112
5113 read_lock(&em_tree->lock);
5114 em = lookup_extent_mapping(em_tree, logical, len);
5115 read_unlock(&em_tree->lock);
5116 BUG_ON(!em);
5117
5118 BUG_ON(em->start > logical || em->start + em->len < logical);
5119 map = em->map_lookup;
5120 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5121 len = map->stripe_len * nr_data_stripes(map);
5122 free_extent_map(em);
5123 return len;
5124 }
5125
5126 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5127 u64 logical, u64 len, int mirror_num)
5128 {
5129 struct extent_map *em;
5130 struct map_lookup *map;
5131 struct extent_map_tree *em_tree = &map_tree->map_tree;
5132 int ret = 0;
5133
5134 read_lock(&em_tree->lock);
5135 em = lookup_extent_mapping(em_tree, logical, len);
5136 read_unlock(&em_tree->lock);
5137 BUG_ON(!em);
5138
5139 BUG_ON(em->start > logical || em->start + em->len < logical);
5140 map = em->map_lookup;
5141 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5142 ret = 1;
5143 free_extent_map(em);
5144 return ret;
5145 }
5146
5147 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5148 struct map_lookup *map, int first, int num,
5149 int optimal, int dev_replace_is_ongoing)
5150 {
5151 int i;
5152 int tolerance;
5153 struct btrfs_device *srcdev;
5154
5155 if (dev_replace_is_ongoing &&
5156 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5157 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5158 srcdev = fs_info->dev_replace.srcdev;
5159 else
5160 srcdev = NULL;
5161
5162 /*
5163 * try to avoid the drive that is the source drive for a
5164 * dev-replace procedure, only choose it if no other non-missing
5165 * mirror is available
5166 */
5167 for (tolerance = 0; tolerance < 2; tolerance++) {
5168 if (map->stripes[optimal].dev->bdev &&
5169 (tolerance || map->stripes[optimal].dev != srcdev))
5170 return optimal;
5171 for (i = first; i < first + num; i++) {
5172 if (map->stripes[i].dev->bdev &&
5173 (tolerance || map->stripes[i].dev != srcdev))
5174 return i;
5175 }
5176 }
5177
5178 /* we couldn't find one that doesn't fail. Just return something
5179 * and the io error handling code will clean up eventually
5180 */
5181 return optimal;
5182 }
5183
5184 static inline int parity_smaller(u64 a, u64 b)
5185 {
5186 return a > b;
5187 }
5188
5189 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5190 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5191 {
5192 struct btrfs_bio_stripe s;
5193 int i;
5194 u64 l;
5195 int again = 1;
5196
5197 while (again) {
5198 again = 0;
5199 for (i = 0; i < num_stripes - 1; i++) {
5200 if (parity_smaller(bbio->raid_map[i],
5201 bbio->raid_map[i+1])) {
5202 s = bbio->stripes[i];
5203 l = bbio->raid_map[i];
5204 bbio->stripes[i] = bbio->stripes[i+1];
5205 bbio->raid_map[i] = bbio->raid_map[i+1];
5206 bbio->stripes[i+1] = s;
5207 bbio->raid_map[i+1] = l;
5208
5209 again = 1;
5210 }
5211 }
5212 }
5213 }
5214
5215 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5216 {
5217 struct btrfs_bio *bbio = kzalloc(
5218 /* the size of the btrfs_bio */
5219 sizeof(struct btrfs_bio) +
5220 /* plus the variable array for the stripes */
5221 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5222 /* plus the variable array for the tgt dev */
5223 sizeof(int) * (real_stripes) +
5224 /*
5225 * plus the raid_map, which includes both the tgt dev
5226 * and the stripes
5227 */
5228 sizeof(u64) * (total_stripes),
5229 GFP_NOFS|__GFP_NOFAIL);
5230
5231 atomic_set(&bbio->error, 0);
5232 atomic_set(&bbio->refs, 1);
5233
5234 return bbio;
5235 }
5236
5237 void btrfs_get_bbio(struct btrfs_bio *bbio)
5238 {
5239 WARN_ON(!atomic_read(&bbio->refs));
5240 atomic_inc(&bbio->refs);
5241 }
5242
5243 void btrfs_put_bbio(struct btrfs_bio *bbio)
5244 {
5245 if (!bbio)
5246 return;
5247 if (atomic_dec_and_test(&bbio->refs))
5248 kfree(bbio);
5249 }
5250
5251 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5252 u64 logical, u64 *length,
5253 struct btrfs_bio **bbio_ret,
5254 int mirror_num, int need_raid_map)
5255 {
5256 struct extent_map *em;
5257 struct map_lookup *map;
5258 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5259 struct extent_map_tree *em_tree = &map_tree->map_tree;
5260 u64 offset;
5261 u64 stripe_offset;
5262 u64 stripe_end_offset;
5263 u64 stripe_nr;
5264 u64 stripe_nr_orig;
5265 u64 stripe_nr_end;
5266 u64 stripe_len;
5267 u32 stripe_index;
5268 int i;
5269 int ret = 0;
5270 int num_stripes;
5271 int max_errors = 0;
5272 int tgtdev_indexes = 0;
5273 struct btrfs_bio *bbio = NULL;
5274 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5275 int dev_replace_is_ongoing = 0;
5276 int num_alloc_stripes;
5277 int patch_the_first_stripe_for_dev_replace = 0;
5278 u64 physical_to_patch_in_first_stripe = 0;
5279 u64 raid56_full_stripe_start = (u64)-1;
5280
5281 read_lock(&em_tree->lock);
5282 em = lookup_extent_mapping(em_tree, logical, *length);
5283 read_unlock(&em_tree->lock);
5284
5285 if (!em) {
5286 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5287 logical, *length);
5288 return -EINVAL;
5289 }
5290
5291 if (em->start > logical || em->start + em->len < logical) {
5292 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5293 "found %Lu-%Lu", logical, em->start,
5294 em->start + em->len);
5295 free_extent_map(em);
5296 return -EINVAL;
5297 }
5298
5299 map = em->map_lookup;
5300 offset = logical - em->start;
5301
5302 stripe_len = map->stripe_len;
5303 stripe_nr = offset;
5304 /*
5305 * stripe_nr counts the total number of stripes we have to stride
5306 * to get to this block
5307 */
5308 stripe_nr = div64_u64(stripe_nr, stripe_len);
5309
5310 stripe_offset = stripe_nr * stripe_len;
5311 if (offset < stripe_offset) {
5312 btrfs_crit(fs_info, "stripe math has gone wrong, "
5313 "stripe_offset=%llu, offset=%llu, start=%llu, "
5314 "logical=%llu, stripe_len=%llu",
5315 stripe_offset, offset, em->start, logical,
5316 stripe_len);
5317 free_extent_map(em);
5318 return -EINVAL;
5319 }
5320
5321 /* stripe_offset is the offset of this block in its stripe*/
5322 stripe_offset = offset - stripe_offset;
5323
5324 /* if we're here for raid56, we need to know the stripe aligned start */
5325 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5326 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5327 raid56_full_stripe_start = offset;
5328
5329 /* allow a write of a full stripe, but make sure we don't
5330 * allow straddling of stripes
5331 */
5332 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5333 full_stripe_len);
5334 raid56_full_stripe_start *= full_stripe_len;
5335 }
5336
5337 if (rw & REQ_DISCARD) {
5338 /* we don't discard raid56 yet */
5339 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5340 ret = -EOPNOTSUPP;
5341 goto out;
5342 }
5343 *length = min_t(u64, em->len - offset, *length);
5344 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5345 u64 max_len;
5346 /* For writes to RAID[56], allow a full stripeset across all disks.
5347 For other RAID types and for RAID[56] reads, just allow a single
5348 stripe (on a single disk). */
5349 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5350 (rw & REQ_WRITE)) {
5351 max_len = stripe_len * nr_data_stripes(map) -
5352 (offset - raid56_full_stripe_start);
5353 } else {
5354 /* we limit the length of each bio to what fits in a stripe */
5355 max_len = stripe_len - stripe_offset;
5356 }
5357 *length = min_t(u64, em->len - offset, max_len);
5358 } else {
5359 *length = em->len - offset;
5360 }
5361
5362 /* This is for when we're called from btrfs_merge_bio_hook() and all
5363 it cares about is the length */
5364 if (!bbio_ret)
5365 goto out;
5366
5367 btrfs_dev_replace_lock(dev_replace, 0);
5368 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5369 if (!dev_replace_is_ongoing)
5370 btrfs_dev_replace_unlock(dev_replace, 0);
5371 else
5372 btrfs_dev_replace_set_lock_blocking(dev_replace);
5373
5374 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5375 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5376 dev_replace->tgtdev != NULL) {
5377 /*
5378 * in dev-replace case, for repair case (that's the only
5379 * case where the mirror is selected explicitly when
5380 * calling btrfs_map_block), blocks left of the left cursor
5381 * can also be read from the target drive.
5382 * For REQ_GET_READ_MIRRORS, the target drive is added as
5383 * the last one to the array of stripes. For READ, it also
5384 * needs to be supported using the same mirror number.
5385 * If the requested block is not left of the left cursor,
5386 * EIO is returned. This can happen because btrfs_num_copies()
5387 * returns one more in the dev-replace case.
5388 */
5389 u64 tmp_length = *length;
5390 struct btrfs_bio *tmp_bbio = NULL;
5391 int tmp_num_stripes;
5392 u64 srcdev_devid = dev_replace->srcdev->devid;
5393 int index_srcdev = 0;
5394 int found = 0;
5395 u64 physical_of_found = 0;
5396
5397 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5398 logical, &tmp_length, &tmp_bbio, 0, 0);
5399 if (ret) {
5400 WARN_ON(tmp_bbio != NULL);
5401 goto out;
5402 }
5403
5404 tmp_num_stripes = tmp_bbio->num_stripes;
5405 if (mirror_num > tmp_num_stripes) {
5406 /*
5407 * REQ_GET_READ_MIRRORS does not contain this
5408 * mirror, that means that the requested area
5409 * is not left of the left cursor
5410 */
5411 ret = -EIO;
5412 btrfs_put_bbio(tmp_bbio);
5413 goto out;
5414 }
5415
5416 /*
5417 * process the rest of the function using the mirror_num
5418 * of the source drive. Therefore look it up first.
5419 * At the end, patch the device pointer to the one of the
5420 * target drive.
5421 */
5422 for (i = 0; i < tmp_num_stripes; i++) {
5423 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5424 continue;
5425
5426 /*
5427 * In case of DUP, in order to keep it simple, only add
5428 * the mirror with the lowest physical address
5429 */
5430 if (found &&
5431 physical_of_found <= tmp_bbio->stripes[i].physical)
5432 continue;
5433
5434 index_srcdev = i;
5435 found = 1;
5436 physical_of_found = tmp_bbio->stripes[i].physical;
5437 }
5438
5439 btrfs_put_bbio(tmp_bbio);
5440
5441 if (!found) {
5442 WARN_ON(1);
5443 ret = -EIO;
5444 goto out;
5445 }
5446
5447 mirror_num = index_srcdev + 1;
5448 patch_the_first_stripe_for_dev_replace = 1;
5449 physical_to_patch_in_first_stripe = physical_of_found;
5450 } else if (mirror_num > map->num_stripes) {
5451 mirror_num = 0;
5452 }
5453
5454 num_stripes = 1;
5455 stripe_index = 0;
5456 stripe_nr_orig = stripe_nr;
5457 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5458 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5459 stripe_end_offset = stripe_nr_end * map->stripe_len -
5460 (offset + *length);
5461
5462 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5463 if (rw & REQ_DISCARD)
5464 num_stripes = min_t(u64, map->num_stripes,
5465 stripe_nr_end - stripe_nr_orig);
5466 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5467 &stripe_index);
5468 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5469 mirror_num = 1;
5470 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5471 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5472 num_stripes = map->num_stripes;
5473 else if (mirror_num)
5474 stripe_index = mirror_num - 1;
5475 else {
5476 stripe_index = find_live_mirror(fs_info, map, 0,
5477 map->num_stripes,
5478 current->pid % map->num_stripes,
5479 dev_replace_is_ongoing);
5480 mirror_num = stripe_index + 1;
5481 }
5482
5483 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5484 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5485 num_stripes = map->num_stripes;
5486 } else if (mirror_num) {
5487 stripe_index = mirror_num - 1;
5488 } else {
5489 mirror_num = 1;
5490 }
5491
5492 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5493 u32 factor = map->num_stripes / map->sub_stripes;
5494
5495 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5496 stripe_index *= map->sub_stripes;
5497
5498 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5499 num_stripes = map->sub_stripes;
5500 else if (rw & REQ_DISCARD)
5501 num_stripes = min_t(u64, map->sub_stripes *
5502 (stripe_nr_end - stripe_nr_orig),
5503 map->num_stripes);
5504 else if (mirror_num)
5505 stripe_index += mirror_num - 1;
5506 else {
5507 int old_stripe_index = stripe_index;
5508 stripe_index = find_live_mirror(fs_info, map,
5509 stripe_index,
5510 map->sub_stripes, stripe_index +
5511 current->pid % map->sub_stripes,
5512 dev_replace_is_ongoing);
5513 mirror_num = stripe_index - old_stripe_index + 1;
5514 }
5515
5516 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5517 if (need_raid_map &&
5518 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5519 mirror_num > 1)) {
5520 /* push stripe_nr back to the start of the full stripe */
5521 stripe_nr = div_u64(raid56_full_stripe_start,
5522 stripe_len * nr_data_stripes(map));
5523
5524 /* RAID[56] write or recovery. Return all stripes */
5525 num_stripes = map->num_stripes;
5526 max_errors = nr_parity_stripes(map);
5527
5528 *length = map->stripe_len;
5529 stripe_index = 0;
5530 stripe_offset = 0;
5531 } else {
5532 /*
5533 * Mirror #0 or #1 means the original data block.
5534 * Mirror #2 is RAID5 parity block.
5535 * Mirror #3 is RAID6 Q block.
5536 */
5537 stripe_nr = div_u64_rem(stripe_nr,
5538 nr_data_stripes(map), &stripe_index);
5539 if (mirror_num > 1)
5540 stripe_index = nr_data_stripes(map) +
5541 mirror_num - 2;
5542
5543 /* We distribute the parity blocks across stripes */
5544 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5545 &stripe_index);
5546 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5547 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5548 mirror_num = 1;
5549 }
5550 } else {
5551 /*
5552 * after this, stripe_nr is the number of stripes on this
5553 * device we have to walk to find the data, and stripe_index is
5554 * the number of our device in the stripe array
5555 */
5556 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5557 &stripe_index);
5558 mirror_num = stripe_index + 1;
5559 }
5560 if (stripe_index >= map->num_stripes) {
5561 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5562 "got stripe_index=%u, num_stripes=%u",
5563 stripe_index, map->num_stripes);
5564 ret = -EINVAL;
5565 goto out;
5566 }
5567
5568 num_alloc_stripes = num_stripes;
5569 if (dev_replace_is_ongoing) {
5570 if (rw & (REQ_WRITE | REQ_DISCARD))
5571 num_alloc_stripes <<= 1;
5572 if (rw & REQ_GET_READ_MIRRORS)
5573 num_alloc_stripes++;
5574 tgtdev_indexes = num_stripes;
5575 }
5576
5577 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5578 if (!bbio) {
5579 ret = -ENOMEM;
5580 goto out;
5581 }
5582 if (dev_replace_is_ongoing)
5583 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5584
5585 /* build raid_map */
5586 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5587 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5588 mirror_num > 1)) {
5589 u64 tmp;
5590 unsigned rot;
5591
5592 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5593 sizeof(struct btrfs_bio_stripe) *
5594 num_alloc_stripes +
5595 sizeof(int) * tgtdev_indexes);
5596
5597 /* Work out the disk rotation on this stripe-set */
5598 div_u64_rem(stripe_nr, num_stripes, &rot);
5599
5600 /* Fill in the logical address of each stripe */
5601 tmp = stripe_nr * nr_data_stripes(map);
5602 for (i = 0; i < nr_data_stripes(map); i++)
5603 bbio->raid_map[(i+rot) % num_stripes] =
5604 em->start + (tmp + i) * map->stripe_len;
5605
5606 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5607 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5608 bbio->raid_map[(i+rot+1) % num_stripes] =
5609 RAID6_Q_STRIPE;
5610 }
5611
5612 if (rw & REQ_DISCARD) {
5613 u32 factor = 0;
5614 u32 sub_stripes = 0;
5615 u64 stripes_per_dev = 0;
5616 u32 remaining_stripes = 0;
5617 u32 last_stripe = 0;
5618
5619 if (map->type &
5620 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5621 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5622 sub_stripes = 1;
5623 else
5624 sub_stripes = map->sub_stripes;
5625
5626 factor = map->num_stripes / sub_stripes;
5627 stripes_per_dev = div_u64_rem(stripe_nr_end -
5628 stripe_nr_orig,
5629 factor,
5630 &remaining_stripes);
5631 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5632 last_stripe *= sub_stripes;
5633 }
5634
5635 for (i = 0; i < num_stripes; i++) {
5636 bbio->stripes[i].physical =
5637 map->stripes[stripe_index].physical +
5638 stripe_offset + stripe_nr * map->stripe_len;
5639 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5640
5641 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5642 BTRFS_BLOCK_GROUP_RAID10)) {
5643 bbio->stripes[i].length = stripes_per_dev *
5644 map->stripe_len;
5645
5646 if (i / sub_stripes < remaining_stripes)
5647 bbio->stripes[i].length +=
5648 map->stripe_len;
5649
5650 /*
5651 * Special for the first stripe and
5652 * the last stripe:
5653 *
5654 * |-------|...|-------|
5655 * |----------|
5656 * off end_off
5657 */
5658 if (i < sub_stripes)
5659 bbio->stripes[i].length -=
5660 stripe_offset;
5661
5662 if (stripe_index >= last_stripe &&
5663 stripe_index <= (last_stripe +
5664 sub_stripes - 1))
5665 bbio->stripes[i].length -=
5666 stripe_end_offset;
5667
5668 if (i == sub_stripes - 1)
5669 stripe_offset = 0;
5670 } else
5671 bbio->stripes[i].length = *length;
5672
5673 stripe_index++;
5674 if (stripe_index == map->num_stripes) {
5675 /* This could only happen for RAID0/10 */
5676 stripe_index = 0;
5677 stripe_nr++;
5678 }
5679 }
5680 } else {
5681 for (i = 0; i < num_stripes; i++) {
5682 bbio->stripes[i].physical =
5683 map->stripes[stripe_index].physical +
5684 stripe_offset +
5685 stripe_nr * map->stripe_len;
5686 bbio->stripes[i].dev =
5687 map->stripes[stripe_index].dev;
5688 stripe_index++;
5689 }
5690 }
5691
5692 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5693 max_errors = btrfs_chunk_max_errors(map);
5694
5695 if (bbio->raid_map)
5696 sort_parity_stripes(bbio, num_stripes);
5697
5698 tgtdev_indexes = 0;
5699 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5700 dev_replace->tgtdev != NULL) {
5701 int index_where_to_add;
5702 u64 srcdev_devid = dev_replace->srcdev->devid;
5703
5704 /*
5705 * duplicate the write operations while the dev replace
5706 * procedure is running. Since the copying of the old disk
5707 * to the new disk takes place at run time while the
5708 * filesystem is mounted writable, the regular write
5709 * operations to the old disk have to be duplicated to go
5710 * to the new disk as well.
5711 * Note that device->missing is handled by the caller, and
5712 * that the write to the old disk is already set up in the
5713 * stripes array.
5714 */
5715 index_where_to_add = num_stripes;
5716 for (i = 0; i < num_stripes; i++) {
5717 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5718 /* write to new disk, too */
5719 struct btrfs_bio_stripe *new =
5720 bbio->stripes + index_where_to_add;
5721 struct btrfs_bio_stripe *old =
5722 bbio->stripes + i;
5723
5724 new->physical = old->physical;
5725 new->length = old->length;
5726 new->dev = dev_replace->tgtdev;
5727 bbio->tgtdev_map[i] = index_where_to_add;
5728 index_where_to_add++;
5729 max_errors++;
5730 tgtdev_indexes++;
5731 }
5732 }
5733 num_stripes = index_where_to_add;
5734 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5735 dev_replace->tgtdev != NULL) {
5736 u64 srcdev_devid = dev_replace->srcdev->devid;
5737 int index_srcdev = 0;
5738 int found = 0;
5739 u64 physical_of_found = 0;
5740
5741 /*
5742 * During the dev-replace procedure, the target drive can
5743 * also be used to read data in case it is needed to repair
5744 * a corrupt block elsewhere. This is possible if the
5745 * requested area is left of the left cursor. In this area,
5746 * the target drive is a full copy of the source drive.
5747 */
5748 for (i = 0; i < num_stripes; i++) {
5749 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5750 /*
5751 * In case of DUP, in order to keep it
5752 * simple, only add the mirror with the
5753 * lowest physical address
5754 */
5755 if (found &&
5756 physical_of_found <=
5757 bbio->stripes[i].physical)
5758 continue;
5759 index_srcdev = i;
5760 found = 1;
5761 physical_of_found = bbio->stripes[i].physical;
5762 }
5763 }
5764 if (found) {
5765 if (physical_of_found + map->stripe_len <=
5766 dev_replace->cursor_left) {
5767 struct btrfs_bio_stripe *tgtdev_stripe =
5768 bbio->stripes + num_stripes;
5769
5770 tgtdev_stripe->physical = physical_of_found;
5771 tgtdev_stripe->length =
5772 bbio->stripes[index_srcdev].length;
5773 tgtdev_stripe->dev = dev_replace->tgtdev;
5774 bbio->tgtdev_map[index_srcdev] = num_stripes;
5775
5776 tgtdev_indexes++;
5777 num_stripes++;
5778 }
5779 }
5780 }
5781
5782 *bbio_ret = bbio;
5783 bbio->map_type = map->type;
5784 bbio->num_stripes = num_stripes;
5785 bbio->max_errors = max_errors;
5786 bbio->mirror_num = mirror_num;
5787 bbio->num_tgtdevs = tgtdev_indexes;
5788
5789 /*
5790 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5791 * mirror_num == num_stripes + 1 && dev_replace target drive is
5792 * available as a mirror
5793 */
5794 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5795 WARN_ON(num_stripes > 1);
5796 bbio->stripes[0].dev = dev_replace->tgtdev;
5797 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5798 bbio->mirror_num = map->num_stripes + 1;
5799 }
5800 out:
5801 if (dev_replace_is_ongoing) {
5802 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5803 btrfs_dev_replace_unlock(dev_replace, 0);
5804 }
5805 free_extent_map(em);
5806 return ret;
5807 }
5808
5809 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5810 u64 logical, u64 *length,
5811 struct btrfs_bio **bbio_ret, int mirror_num)
5812 {
5813 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5814 mirror_num, 0);
5815 }
5816
5817 /* For Scrub/replace */
5818 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5819 u64 logical, u64 *length,
5820 struct btrfs_bio **bbio_ret, int mirror_num,
5821 int need_raid_map)
5822 {
5823 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5824 mirror_num, need_raid_map);
5825 }
5826
5827 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5828 u64 chunk_start, u64 physical, u64 devid,
5829 u64 **logical, int *naddrs, int *stripe_len)
5830 {
5831 struct extent_map_tree *em_tree = &map_tree->map_tree;
5832 struct extent_map *em;
5833 struct map_lookup *map;
5834 u64 *buf;
5835 u64 bytenr;
5836 u64 length;
5837 u64 stripe_nr;
5838 u64 rmap_len;
5839 int i, j, nr = 0;
5840
5841 read_lock(&em_tree->lock);
5842 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5843 read_unlock(&em_tree->lock);
5844
5845 if (!em) {
5846 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5847 chunk_start);
5848 return -EIO;
5849 }
5850
5851 if (em->start != chunk_start) {
5852 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5853 em->start, chunk_start);
5854 free_extent_map(em);
5855 return -EIO;
5856 }
5857 map = em->map_lookup;
5858
5859 length = em->len;
5860 rmap_len = map->stripe_len;
5861
5862 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5863 length = div_u64(length, map->num_stripes / map->sub_stripes);
5864 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5865 length = div_u64(length, map->num_stripes);
5866 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5867 length = div_u64(length, nr_data_stripes(map));
5868 rmap_len = map->stripe_len * nr_data_stripes(map);
5869 }
5870
5871 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5872 BUG_ON(!buf); /* -ENOMEM */
5873
5874 for (i = 0; i < map->num_stripes; i++) {
5875 if (devid && map->stripes[i].dev->devid != devid)
5876 continue;
5877 if (map->stripes[i].physical > physical ||
5878 map->stripes[i].physical + length <= physical)
5879 continue;
5880
5881 stripe_nr = physical - map->stripes[i].physical;
5882 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5883
5884 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5885 stripe_nr = stripe_nr * map->num_stripes + i;
5886 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5887 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5888 stripe_nr = stripe_nr * map->num_stripes + i;
5889 } /* else if RAID[56], multiply by nr_data_stripes().
5890 * Alternatively, just use rmap_len below instead of
5891 * map->stripe_len */
5892
5893 bytenr = chunk_start + stripe_nr * rmap_len;
5894 WARN_ON(nr >= map->num_stripes);
5895 for (j = 0; j < nr; j++) {
5896 if (buf[j] == bytenr)
5897 break;
5898 }
5899 if (j == nr) {
5900 WARN_ON(nr >= map->num_stripes);
5901 buf[nr++] = bytenr;
5902 }
5903 }
5904
5905 *logical = buf;
5906 *naddrs = nr;
5907 *stripe_len = rmap_len;
5908
5909 free_extent_map(em);
5910 return 0;
5911 }
5912
5913 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5914 {
5915 bio->bi_private = bbio->private;
5916 bio->bi_end_io = bbio->end_io;
5917 bio_endio(bio);
5918
5919 btrfs_put_bbio(bbio);
5920 }
5921
5922 static void btrfs_end_bio(struct bio *bio)
5923 {
5924 struct btrfs_bio *bbio = bio->bi_private;
5925 int is_orig_bio = 0;
5926
5927 if (bio->bi_error) {
5928 atomic_inc(&bbio->error);
5929 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5930 unsigned int stripe_index =
5931 btrfs_io_bio(bio)->stripe_index;
5932 struct btrfs_device *dev;
5933
5934 BUG_ON(stripe_index >= bbio->num_stripes);
5935 dev = bbio->stripes[stripe_index].dev;
5936 if (dev->bdev) {
5937 if (bio->bi_rw & WRITE)
5938 btrfs_dev_stat_inc(dev,
5939 BTRFS_DEV_STAT_WRITE_ERRS);
5940 else
5941 btrfs_dev_stat_inc(dev,
5942 BTRFS_DEV_STAT_READ_ERRS);
5943 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5944 btrfs_dev_stat_inc(dev,
5945 BTRFS_DEV_STAT_FLUSH_ERRS);
5946 btrfs_dev_stat_print_on_error(dev);
5947 }
5948 }
5949 }
5950
5951 if (bio == bbio->orig_bio)
5952 is_orig_bio = 1;
5953
5954 btrfs_bio_counter_dec(bbio->fs_info);
5955
5956 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5957 if (!is_orig_bio) {
5958 bio_put(bio);
5959 bio = bbio->orig_bio;
5960 }
5961
5962 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5963 /* only send an error to the higher layers if it is
5964 * beyond the tolerance of the btrfs bio
5965 */
5966 if (atomic_read(&bbio->error) > bbio->max_errors) {
5967 bio->bi_error = -EIO;
5968 } else {
5969 /*
5970 * this bio is actually up to date, we didn't
5971 * go over the max number of errors
5972 */
5973 bio->bi_error = 0;
5974 }
5975
5976 btrfs_end_bbio(bbio, bio);
5977 } else if (!is_orig_bio) {
5978 bio_put(bio);
5979 }
5980 }
5981
5982 /*
5983 * see run_scheduled_bios for a description of why bios are collected for
5984 * async submit.
5985 *
5986 * This will add one bio to the pending list for a device and make sure
5987 * the work struct is scheduled.
5988 */
5989 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5990 struct btrfs_device *device,
5991 int rw, struct bio *bio)
5992 {
5993 int should_queue = 1;
5994 struct btrfs_pending_bios *pending_bios;
5995
5996 if (device->missing || !device->bdev) {
5997 bio_io_error(bio);
5998 return;
5999 }
6000
6001 /* don't bother with additional async steps for reads, right now */
6002 if (!(rw & REQ_WRITE)) {
6003 bio_get(bio);
6004 btrfsic_submit_bio(rw, bio);
6005 bio_put(bio);
6006 return;
6007 }
6008
6009 /*
6010 * nr_async_bios allows us to reliably return congestion to the
6011 * higher layers. Otherwise, the async bio makes it appear we have
6012 * made progress against dirty pages when we've really just put it
6013 * on a queue for later
6014 */
6015 atomic_inc(&root->fs_info->nr_async_bios);
6016 WARN_ON(bio->bi_next);
6017 bio->bi_next = NULL;
6018 bio->bi_rw |= rw;
6019
6020 spin_lock(&device->io_lock);
6021 if (bio->bi_rw & REQ_SYNC)
6022 pending_bios = &device->pending_sync_bios;
6023 else
6024 pending_bios = &device->pending_bios;
6025
6026 if (pending_bios->tail)
6027 pending_bios->tail->bi_next = bio;
6028
6029 pending_bios->tail = bio;
6030 if (!pending_bios->head)
6031 pending_bios->head = bio;
6032 if (device->running_pending)
6033 should_queue = 0;
6034
6035 spin_unlock(&device->io_lock);
6036
6037 if (should_queue)
6038 btrfs_queue_work(root->fs_info->submit_workers,
6039 &device->work);
6040 }
6041
6042 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6043 struct bio *bio, u64 physical, int dev_nr,
6044 int rw, int async)
6045 {
6046 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6047
6048 bio->bi_private = bbio;
6049 btrfs_io_bio(bio)->stripe_index = dev_nr;
6050 bio->bi_end_io = btrfs_end_bio;
6051 bio->bi_iter.bi_sector = physical >> 9;
6052 #ifdef DEBUG
6053 {
6054 struct rcu_string *name;
6055
6056 rcu_read_lock();
6057 name = rcu_dereference(dev->name);
6058 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6059 "(%s id %llu), size=%u\n", rw,
6060 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6061 name->str, dev->devid, bio->bi_iter.bi_size);
6062 rcu_read_unlock();
6063 }
6064 #endif
6065 bio->bi_bdev = dev->bdev;
6066
6067 btrfs_bio_counter_inc_noblocked(root->fs_info);
6068
6069 if (async)
6070 btrfs_schedule_bio(root, dev, rw, bio);
6071 else
6072 btrfsic_submit_bio(rw, bio);
6073 }
6074
6075 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6076 {
6077 atomic_inc(&bbio->error);
6078 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6079 /* Should be the original bio. */
6080 WARN_ON(bio != bbio->orig_bio);
6081
6082 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6083 bio->bi_iter.bi_sector = logical >> 9;
6084 bio->bi_error = -EIO;
6085 btrfs_end_bbio(bbio, bio);
6086 }
6087 }
6088
6089 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6090 int mirror_num, int async_submit)
6091 {
6092 struct btrfs_device *dev;
6093 struct bio *first_bio = bio;
6094 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6095 u64 length = 0;
6096 u64 map_length;
6097 int ret;
6098 int dev_nr;
6099 int total_devs;
6100 struct btrfs_bio *bbio = NULL;
6101
6102 length = bio->bi_iter.bi_size;
6103 map_length = length;
6104
6105 btrfs_bio_counter_inc_blocked(root->fs_info);
6106 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6107 mirror_num, 1);
6108 if (ret) {
6109 btrfs_bio_counter_dec(root->fs_info);
6110 return ret;
6111 }
6112
6113 total_devs = bbio->num_stripes;
6114 bbio->orig_bio = first_bio;
6115 bbio->private = first_bio->bi_private;
6116 bbio->end_io = first_bio->bi_end_io;
6117 bbio->fs_info = root->fs_info;
6118 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6119
6120 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6121 ((rw & WRITE) || (mirror_num > 1))) {
6122 /* In this case, map_length has been set to the length of
6123 a single stripe; not the whole write */
6124 if (rw & WRITE) {
6125 ret = raid56_parity_write(root, bio, bbio, map_length);
6126 } else {
6127 ret = raid56_parity_recover(root, bio, bbio, map_length,
6128 mirror_num, 1);
6129 }
6130
6131 btrfs_bio_counter_dec(root->fs_info);
6132 return ret;
6133 }
6134
6135 if (map_length < length) {
6136 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6137 logical, length, map_length);
6138 BUG();
6139 }
6140
6141 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6142 dev = bbio->stripes[dev_nr].dev;
6143 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6144 bbio_error(bbio, first_bio, logical);
6145 continue;
6146 }
6147
6148 if (dev_nr < total_devs - 1) {
6149 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6150 BUG_ON(!bio); /* -ENOMEM */
6151 } else
6152 bio = first_bio;
6153
6154 submit_stripe_bio(root, bbio, bio,
6155 bbio->stripes[dev_nr].physical, dev_nr, rw,
6156 async_submit);
6157 }
6158 btrfs_bio_counter_dec(root->fs_info);
6159 return 0;
6160 }
6161
6162 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6163 u8 *uuid, u8 *fsid)
6164 {
6165 struct btrfs_device *device;
6166 struct btrfs_fs_devices *cur_devices;
6167
6168 cur_devices = fs_info->fs_devices;
6169 while (cur_devices) {
6170 if (!fsid ||
6171 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6172 device = __find_device(&cur_devices->devices,
6173 devid, uuid);
6174 if (device)
6175 return device;
6176 }
6177 cur_devices = cur_devices->seed;
6178 }
6179 return NULL;
6180 }
6181
6182 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6183 struct btrfs_fs_devices *fs_devices,
6184 u64 devid, u8 *dev_uuid)
6185 {
6186 struct btrfs_device *device;
6187
6188 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6189 if (IS_ERR(device))
6190 return NULL;
6191
6192 list_add(&device->dev_list, &fs_devices->devices);
6193 device->fs_devices = fs_devices;
6194 fs_devices->num_devices++;
6195
6196 device->missing = 1;
6197 fs_devices->missing_devices++;
6198
6199 return device;
6200 }
6201
6202 /**
6203 * btrfs_alloc_device - allocate struct btrfs_device
6204 * @fs_info: used only for generating a new devid, can be NULL if
6205 * devid is provided (i.e. @devid != NULL).
6206 * @devid: a pointer to devid for this device. If NULL a new devid
6207 * is generated.
6208 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6209 * is generated.
6210 *
6211 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6212 * on error. Returned struct is not linked onto any lists and can be
6213 * destroyed with kfree() right away.
6214 */
6215 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6216 const u64 *devid,
6217 const u8 *uuid)
6218 {
6219 struct btrfs_device *dev;
6220 u64 tmp;
6221
6222 if (WARN_ON(!devid && !fs_info))
6223 return ERR_PTR(-EINVAL);
6224
6225 dev = __alloc_device();
6226 if (IS_ERR(dev))
6227 return dev;
6228
6229 if (devid)
6230 tmp = *devid;
6231 else {
6232 int ret;
6233
6234 ret = find_next_devid(fs_info, &tmp);
6235 if (ret) {
6236 kfree(dev);
6237 return ERR_PTR(ret);
6238 }
6239 }
6240 dev->devid = tmp;
6241
6242 if (uuid)
6243 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6244 else
6245 generate_random_uuid(dev->uuid);
6246
6247 btrfs_init_work(&dev->work, btrfs_submit_helper,
6248 pending_bios_fn, NULL, NULL);
6249
6250 return dev;
6251 }
6252
6253 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6254 struct extent_buffer *leaf,
6255 struct btrfs_chunk *chunk)
6256 {
6257 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6258 struct map_lookup *map;
6259 struct extent_map *em;
6260 u64 logical;
6261 u64 length;
6262 u64 stripe_len;
6263 u64 devid;
6264 u8 uuid[BTRFS_UUID_SIZE];
6265 int num_stripes;
6266 int ret;
6267 int i;
6268
6269 logical = key->offset;
6270 length = btrfs_chunk_length(leaf, chunk);
6271 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6272 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6273 /* Validation check */
6274 if (!num_stripes) {
6275 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6276 num_stripes);
6277 return -EIO;
6278 }
6279 if (!IS_ALIGNED(logical, root->sectorsize)) {
6280 btrfs_err(root->fs_info,
6281 "invalid chunk logical %llu", logical);
6282 return -EIO;
6283 }
6284 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6285 btrfs_err(root->fs_info,
6286 "invalid chunk length %llu", length);
6287 return -EIO;
6288 }
6289 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6290 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6291 stripe_len);
6292 return -EIO;
6293 }
6294 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6295 btrfs_chunk_type(leaf, chunk)) {
6296 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6297 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6298 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6299 btrfs_chunk_type(leaf, chunk));
6300 return -EIO;
6301 }
6302
6303 read_lock(&map_tree->map_tree.lock);
6304 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6305 read_unlock(&map_tree->map_tree.lock);
6306
6307 /* already mapped? */
6308 if (em && em->start <= logical && em->start + em->len > logical) {
6309 free_extent_map(em);
6310 return 0;
6311 } else if (em) {
6312 free_extent_map(em);
6313 }
6314
6315 em = alloc_extent_map();
6316 if (!em)
6317 return -ENOMEM;
6318 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6319 if (!map) {
6320 free_extent_map(em);
6321 return -ENOMEM;
6322 }
6323
6324 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6325 em->map_lookup = map;
6326 em->start = logical;
6327 em->len = length;
6328 em->orig_start = 0;
6329 em->block_start = 0;
6330 em->block_len = em->len;
6331
6332 map->num_stripes = num_stripes;
6333 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6334 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6335 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6336 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6337 map->type = btrfs_chunk_type(leaf, chunk);
6338 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6339 for (i = 0; i < num_stripes; i++) {
6340 map->stripes[i].physical =
6341 btrfs_stripe_offset_nr(leaf, chunk, i);
6342 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6343 read_extent_buffer(leaf, uuid, (unsigned long)
6344 btrfs_stripe_dev_uuid_nr(chunk, i),
6345 BTRFS_UUID_SIZE);
6346 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6347 uuid, NULL);
6348 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6349 free_extent_map(em);
6350 return -EIO;
6351 }
6352 if (!map->stripes[i].dev) {
6353 map->stripes[i].dev =
6354 add_missing_dev(root, root->fs_info->fs_devices,
6355 devid, uuid);
6356 if (!map->stripes[i].dev) {
6357 free_extent_map(em);
6358 return -EIO;
6359 }
6360 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6361 devid, uuid);
6362 }
6363 map->stripes[i].dev->in_fs_metadata = 1;
6364 }
6365
6366 write_lock(&map_tree->map_tree.lock);
6367 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6368 write_unlock(&map_tree->map_tree.lock);
6369 BUG_ON(ret); /* Tree corruption */
6370 free_extent_map(em);
6371
6372 return 0;
6373 }
6374
6375 static void fill_device_from_item(struct extent_buffer *leaf,
6376 struct btrfs_dev_item *dev_item,
6377 struct btrfs_device *device)
6378 {
6379 unsigned long ptr;
6380
6381 device->devid = btrfs_device_id(leaf, dev_item);
6382 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6383 device->total_bytes = device->disk_total_bytes;
6384 device->commit_total_bytes = device->disk_total_bytes;
6385 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6386 device->commit_bytes_used = device->bytes_used;
6387 device->type = btrfs_device_type(leaf, dev_item);
6388 device->io_align = btrfs_device_io_align(leaf, dev_item);
6389 device->io_width = btrfs_device_io_width(leaf, dev_item);
6390 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6391 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6392 device->is_tgtdev_for_dev_replace = 0;
6393
6394 ptr = btrfs_device_uuid(dev_item);
6395 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6396 }
6397
6398 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6399 u8 *fsid)
6400 {
6401 struct btrfs_fs_devices *fs_devices;
6402 int ret;
6403
6404 BUG_ON(!mutex_is_locked(&uuid_mutex));
6405
6406 fs_devices = root->fs_info->fs_devices->seed;
6407 while (fs_devices) {
6408 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6409 return fs_devices;
6410
6411 fs_devices = fs_devices->seed;
6412 }
6413
6414 fs_devices = find_fsid(fsid);
6415 if (!fs_devices) {
6416 if (!btrfs_test_opt(root, DEGRADED))
6417 return ERR_PTR(-ENOENT);
6418
6419 fs_devices = alloc_fs_devices(fsid);
6420 if (IS_ERR(fs_devices))
6421 return fs_devices;
6422
6423 fs_devices->seeding = 1;
6424 fs_devices->opened = 1;
6425 return fs_devices;
6426 }
6427
6428 fs_devices = clone_fs_devices(fs_devices);
6429 if (IS_ERR(fs_devices))
6430 return fs_devices;
6431
6432 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6433 root->fs_info->bdev_holder);
6434 if (ret) {
6435 free_fs_devices(fs_devices);
6436 fs_devices = ERR_PTR(ret);
6437 goto out;
6438 }
6439
6440 if (!fs_devices->seeding) {
6441 __btrfs_close_devices(fs_devices);
6442 free_fs_devices(fs_devices);
6443 fs_devices = ERR_PTR(-EINVAL);
6444 goto out;
6445 }
6446
6447 fs_devices->seed = root->fs_info->fs_devices->seed;
6448 root->fs_info->fs_devices->seed = fs_devices;
6449 out:
6450 return fs_devices;
6451 }
6452
6453 static int read_one_dev(struct btrfs_root *root,
6454 struct extent_buffer *leaf,
6455 struct btrfs_dev_item *dev_item)
6456 {
6457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6458 struct btrfs_device *device;
6459 u64 devid;
6460 int ret;
6461 u8 fs_uuid[BTRFS_UUID_SIZE];
6462 u8 dev_uuid[BTRFS_UUID_SIZE];
6463
6464 devid = btrfs_device_id(leaf, dev_item);
6465 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6466 BTRFS_UUID_SIZE);
6467 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6468 BTRFS_UUID_SIZE);
6469
6470 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6471 fs_devices = open_seed_devices(root, fs_uuid);
6472 if (IS_ERR(fs_devices))
6473 return PTR_ERR(fs_devices);
6474 }
6475
6476 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6477 if (!device) {
6478 if (!btrfs_test_opt(root, DEGRADED))
6479 return -EIO;
6480
6481 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6482 if (!device)
6483 return -ENOMEM;
6484 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6485 devid, dev_uuid);
6486 } else {
6487 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6488 return -EIO;
6489
6490 if(!device->bdev && !device->missing) {
6491 /*
6492 * this happens when a device that was properly setup
6493 * in the device info lists suddenly goes bad.
6494 * device->bdev is NULL, and so we have to set
6495 * device->missing to one here
6496 */
6497 device->fs_devices->missing_devices++;
6498 device->missing = 1;
6499 }
6500
6501 /* Move the device to its own fs_devices */
6502 if (device->fs_devices != fs_devices) {
6503 ASSERT(device->missing);
6504
6505 list_move(&device->dev_list, &fs_devices->devices);
6506 device->fs_devices->num_devices--;
6507 fs_devices->num_devices++;
6508
6509 device->fs_devices->missing_devices--;
6510 fs_devices->missing_devices++;
6511
6512 device->fs_devices = fs_devices;
6513 }
6514 }
6515
6516 if (device->fs_devices != root->fs_info->fs_devices) {
6517 BUG_ON(device->writeable);
6518 if (device->generation !=
6519 btrfs_device_generation(leaf, dev_item))
6520 return -EINVAL;
6521 }
6522
6523 fill_device_from_item(leaf, dev_item, device);
6524 device->in_fs_metadata = 1;
6525 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6526 device->fs_devices->total_rw_bytes += device->total_bytes;
6527 spin_lock(&root->fs_info->free_chunk_lock);
6528 root->fs_info->free_chunk_space += device->total_bytes -
6529 device->bytes_used;
6530 spin_unlock(&root->fs_info->free_chunk_lock);
6531 }
6532 ret = 0;
6533 return ret;
6534 }
6535
6536 int btrfs_read_sys_array(struct btrfs_root *root)
6537 {
6538 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6539 struct extent_buffer *sb;
6540 struct btrfs_disk_key *disk_key;
6541 struct btrfs_chunk *chunk;
6542 u8 *array_ptr;
6543 unsigned long sb_array_offset;
6544 int ret = 0;
6545 u32 num_stripes;
6546 u32 array_size;
6547 u32 len = 0;
6548 u32 cur_offset;
6549 struct btrfs_key key;
6550
6551 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6552 /*
6553 * This will create extent buffer of nodesize, superblock size is
6554 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6555 * overallocate but we can keep it as-is, only the first page is used.
6556 */
6557 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6558 if (!sb)
6559 return -ENOMEM;
6560 set_extent_buffer_uptodate(sb);
6561 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6562 /*
6563 * The sb extent buffer is artificial and just used to read the system array.
6564 * set_extent_buffer_uptodate() call does not properly mark all it's
6565 * pages up-to-date when the page is larger: extent does not cover the
6566 * whole page and consequently check_page_uptodate does not find all
6567 * the page's extents up-to-date (the hole beyond sb),
6568 * write_extent_buffer then triggers a WARN_ON.
6569 *
6570 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6571 * but sb spans only this function. Add an explicit SetPageUptodate call
6572 * to silence the warning eg. on PowerPC 64.
6573 */
6574 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6575 SetPageUptodate(sb->pages[0]);
6576
6577 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6578 array_size = btrfs_super_sys_array_size(super_copy);
6579
6580 array_ptr = super_copy->sys_chunk_array;
6581 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6582 cur_offset = 0;
6583
6584 while (cur_offset < array_size) {
6585 disk_key = (struct btrfs_disk_key *)array_ptr;
6586 len = sizeof(*disk_key);
6587 if (cur_offset + len > array_size)
6588 goto out_short_read;
6589
6590 btrfs_disk_key_to_cpu(&key, disk_key);
6591
6592 array_ptr += len;
6593 sb_array_offset += len;
6594 cur_offset += len;
6595
6596 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6597 chunk = (struct btrfs_chunk *)sb_array_offset;
6598 /*
6599 * At least one btrfs_chunk with one stripe must be
6600 * present, exact stripe count check comes afterwards
6601 */
6602 len = btrfs_chunk_item_size(1);
6603 if (cur_offset + len > array_size)
6604 goto out_short_read;
6605
6606 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6607 if (!num_stripes) {
6608 printk(KERN_ERR
6609 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6610 num_stripes, cur_offset);
6611 ret = -EIO;
6612 break;
6613 }
6614
6615 len = btrfs_chunk_item_size(num_stripes);
6616 if (cur_offset + len > array_size)
6617 goto out_short_read;
6618
6619 ret = read_one_chunk(root, &key, sb, chunk);
6620 if (ret)
6621 break;
6622 } else {
6623 printk(KERN_ERR
6624 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6625 (u32)key.type, cur_offset);
6626 ret = -EIO;
6627 break;
6628 }
6629 array_ptr += len;
6630 sb_array_offset += len;
6631 cur_offset += len;
6632 }
6633 free_extent_buffer_stale(sb);
6634 return ret;
6635
6636 out_short_read:
6637 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6638 len, cur_offset);
6639 free_extent_buffer_stale(sb);
6640 return -EIO;
6641 }
6642
6643 int btrfs_read_chunk_tree(struct btrfs_root *root)
6644 {
6645 struct btrfs_path *path;
6646 struct extent_buffer *leaf;
6647 struct btrfs_key key;
6648 struct btrfs_key found_key;
6649 int ret;
6650 int slot;
6651
6652 root = root->fs_info->chunk_root;
6653
6654 path = btrfs_alloc_path();
6655 if (!path)
6656 return -ENOMEM;
6657
6658 mutex_lock(&uuid_mutex);
6659 lock_chunks(root);
6660
6661 /*
6662 * Read all device items, and then all the chunk items. All
6663 * device items are found before any chunk item (their object id
6664 * is smaller than the lowest possible object id for a chunk
6665 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6666 */
6667 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6668 key.offset = 0;
6669 key.type = 0;
6670 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6671 if (ret < 0)
6672 goto error;
6673 while (1) {
6674 leaf = path->nodes[0];
6675 slot = path->slots[0];
6676 if (slot >= btrfs_header_nritems(leaf)) {
6677 ret = btrfs_next_leaf(root, path);
6678 if (ret == 0)
6679 continue;
6680 if (ret < 0)
6681 goto error;
6682 break;
6683 }
6684 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6685 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6686 struct btrfs_dev_item *dev_item;
6687 dev_item = btrfs_item_ptr(leaf, slot,
6688 struct btrfs_dev_item);
6689 ret = read_one_dev(root, leaf, dev_item);
6690 if (ret)
6691 goto error;
6692 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6693 struct btrfs_chunk *chunk;
6694 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6695 ret = read_one_chunk(root, &found_key, leaf, chunk);
6696 if (ret)
6697 goto error;
6698 }
6699 path->slots[0]++;
6700 }
6701 ret = 0;
6702 error:
6703 unlock_chunks(root);
6704 mutex_unlock(&uuid_mutex);
6705
6706 btrfs_free_path(path);
6707 return ret;
6708 }
6709
6710 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6711 {
6712 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6713 struct btrfs_device *device;
6714
6715 while (fs_devices) {
6716 mutex_lock(&fs_devices->device_list_mutex);
6717 list_for_each_entry(device, &fs_devices->devices, dev_list)
6718 device->dev_root = fs_info->dev_root;
6719 mutex_unlock(&fs_devices->device_list_mutex);
6720
6721 fs_devices = fs_devices->seed;
6722 }
6723 }
6724
6725 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6726 {
6727 int i;
6728
6729 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6730 btrfs_dev_stat_reset(dev, i);
6731 }
6732
6733 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6734 {
6735 struct btrfs_key key;
6736 struct btrfs_key found_key;
6737 struct btrfs_root *dev_root = fs_info->dev_root;
6738 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6739 struct extent_buffer *eb;
6740 int slot;
6741 int ret = 0;
6742 struct btrfs_device *device;
6743 struct btrfs_path *path = NULL;
6744 int i;
6745
6746 path = btrfs_alloc_path();
6747 if (!path) {
6748 ret = -ENOMEM;
6749 goto out;
6750 }
6751
6752 mutex_lock(&fs_devices->device_list_mutex);
6753 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6754 int item_size;
6755 struct btrfs_dev_stats_item *ptr;
6756
6757 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6758 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6759 key.offset = device->devid;
6760 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6761 if (ret) {
6762 __btrfs_reset_dev_stats(device);
6763 device->dev_stats_valid = 1;
6764 btrfs_release_path(path);
6765 continue;
6766 }
6767 slot = path->slots[0];
6768 eb = path->nodes[0];
6769 btrfs_item_key_to_cpu(eb, &found_key, slot);
6770 item_size = btrfs_item_size_nr(eb, slot);
6771
6772 ptr = btrfs_item_ptr(eb, slot,
6773 struct btrfs_dev_stats_item);
6774
6775 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6776 if (item_size >= (1 + i) * sizeof(__le64))
6777 btrfs_dev_stat_set(device, i,
6778 btrfs_dev_stats_value(eb, ptr, i));
6779 else
6780 btrfs_dev_stat_reset(device, i);
6781 }
6782
6783 device->dev_stats_valid = 1;
6784 btrfs_dev_stat_print_on_load(device);
6785 btrfs_release_path(path);
6786 }
6787 mutex_unlock(&fs_devices->device_list_mutex);
6788
6789 out:
6790 btrfs_free_path(path);
6791 return ret < 0 ? ret : 0;
6792 }
6793
6794 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6795 struct btrfs_root *dev_root,
6796 struct btrfs_device *device)
6797 {
6798 struct btrfs_path *path;
6799 struct btrfs_key key;
6800 struct extent_buffer *eb;
6801 struct btrfs_dev_stats_item *ptr;
6802 int ret;
6803 int i;
6804
6805 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6806 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6807 key.offset = device->devid;
6808
6809 path = btrfs_alloc_path();
6810 BUG_ON(!path);
6811 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6812 if (ret < 0) {
6813 btrfs_warn_in_rcu(dev_root->fs_info,
6814 "error %d while searching for dev_stats item for device %s",
6815 ret, rcu_str_deref(device->name));
6816 goto out;
6817 }
6818
6819 if (ret == 0 &&
6820 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6821 /* need to delete old one and insert a new one */
6822 ret = btrfs_del_item(trans, dev_root, path);
6823 if (ret != 0) {
6824 btrfs_warn_in_rcu(dev_root->fs_info,
6825 "delete too small dev_stats item for device %s failed %d",
6826 rcu_str_deref(device->name), ret);
6827 goto out;
6828 }
6829 ret = 1;
6830 }
6831
6832 if (ret == 1) {
6833 /* need to insert a new item */
6834 btrfs_release_path(path);
6835 ret = btrfs_insert_empty_item(trans, dev_root, path,
6836 &key, sizeof(*ptr));
6837 if (ret < 0) {
6838 btrfs_warn_in_rcu(dev_root->fs_info,
6839 "insert dev_stats item for device %s failed %d",
6840 rcu_str_deref(device->name), ret);
6841 goto out;
6842 }
6843 }
6844
6845 eb = path->nodes[0];
6846 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6847 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6848 btrfs_set_dev_stats_value(eb, ptr, i,
6849 btrfs_dev_stat_read(device, i));
6850 btrfs_mark_buffer_dirty(eb);
6851
6852 out:
6853 btrfs_free_path(path);
6854 return ret;
6855 }
6856
6857 /*
6858 * called from commit_transaction. Writes all changed device stats to disk.
6859 */
6860 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6861 struct btrfs_fs_info *fs_info)
6862 {
6863 struct btrfs_root *dev_root = fs_info->dev_root;
6864 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6865 struct btrfs_device *device;
6866 int stats_cnt;
6867 int ret = 0;
6868
6869 mutex_lock(&fs_devices->device_list_mutex);
6870 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6871 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6872 continue;
6873
6874 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6875 ret = update_dev_stat_item(trans, dev_root, device);
6876 if (!ret)
6877 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6878 }
6879 mutex_unlock(&fs_devices->device_list_mutex);
6880
6881 return ret;
6882 }
6883
6884 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6885 {
6886 btrfs_dev_stat_inc(dev, index);
6887 btrfs_dev_stat_print_on_error(dev);
6888 }
6889
6890 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6891 {
6892 if (!dev->dev_stats_valid)
6893 return;
6894 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6895 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6896 rcu_str_deref(dev->name),
6897 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6898 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6899 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6900 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6901 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6902 }
6903
6904 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6905 {
6906 int i;
6907
6908 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6909 if (btrfs_dev_stat_read(dev, i) != 0)
6910 break;
6911 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6912 return; /* all values == 0, suppress message */
6913
6914 btrfs_info_in_rcu(dev->dev_root->fs_info,
6915 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6916 rcu_str_deref(dev->name),
6917 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6918 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6919 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6920 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6921 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6922 }
6923
6924 int btrfs_get_dev_stats(struct btrfs_root *root,
6925 struct btrfs_ioctl_get_dev_stats *stats)
6926 {
6927 struct btrfs_device *dev;
6928 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6929 int i;
6930
6931 mutex_lock(&fs_devices->device_list_mutex);
6932 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6933 mutex_unlock(&fs_devices->device_list_mutex);
6934
6935 if (!dev) {
6936 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6937 return -ENODEV;
6938 } else if (!dev->dev_stats_valid) {
6939 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6940 return -ENODEV;
6941 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6942 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6943 if (stats->nr_items > i)
6944 stats->values[i] =
6945 btrfs_dev_stat_read_and_reset(dev, i);
6946 else
6947 btrfs_dev_stat_reset(dev, i);
6948 }
6949 } else {
6950 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6951 if (stats->nr_items > i)
6952 stats->values[i] = btrfs_dev_stat_read(dev, i);
6953 }
6954 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6955 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6956 return 0;
6957 }
6958
6959 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6960 {
6961 struct buffer_head *bh;
6962 struct btrfs_super_block *disk_super;
6963 int copy_num;
6964
6965 if (!bdev)
6966 return;
6967
6968 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6969 copy_num++) {
6970
6971 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6972 continue;
6973
6974 disk_super = (struct btrfs_super_block *)bh->b_data;
6975
6976 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6977 set_buffer_dirty(bh);
6978 sync_dirty_buffer(bh);
6979 brelse(bh);
6980 }
6981
6982 /* Notify udev that device has changed */
6983 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6984
6985 /* Update ctime/mtime for device path for libblkid */
6986 update_dev_time(device_path);
6987 }
6988
6989 /*
6990 * Update the size of all devices, which is used for writing out the
6991 * super blocks.
6992 */
6993 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6994 {
6995 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6996 struct btrfs_device *curr, *next;
6997
6998 if (list_empty(&fs_devices->resized_devices))
6999 return;
7000
7001 mutex_lock(&fs_devices->device_list_mutex);
7002 lock_chunks(fs_info->dev_root);
7003 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7004 resized_list) {
7005 list_del_init(&curr->resized_list);
7006 curr->commit_total_bytes = curr->disk_total_bytes;
7007 }
7008 unlock_chunks(fs_info->dev_root);
7009 mutex_unlock(&fs_devices->device_list_mutex);
7010 }
7011
7012 /* Must be invoked during the transaction commit */
7013 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7014 struct btrfs_transaction *transaction)
7015 {
7016 struct extent_map *em;
7017 struct map_lookup *map;
7018 struct btrfs_device *dev;
7019 int i;
7020
7021 if (list_empty(&transaction->pending_chunks))
7022 return;
7023
7024 /* In order to kick the device replace finish process */
7025 lock_chunks(root);
7026 list_for_each_entry(em, &transaction->pending_chunks, list) {
7027 map = em->map_lookup;
7028
7029 for (i = 0; i < map->num_stripes; i++) {
7030 dev = map->stripes[i].dev;
7031 dev->commit_bytes_used = dev->bytes_used;
7032 }
7033 }
7034 unlock_chunks(root);
7035 }
7036
7037 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7038 {
7039 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7040 while (fs_devices) {
7041 fs_devices->fs_info = fs_info;
7042 fs_devices = fs_devices->seed;
7043 }
7044 }
7045
7046 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7047 {
7048 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7049 while (fs_devices) {
7050 fs_devices->fs_info = NULL;
7051 fs_devices = fs_devices->seed;
7052 }
7053 }
7054
7055 static void btrfs_close_one_device(struct btrfs_device *device)
7056 {
7057 struct btrfs_fs_devices *fs_devices = device->fs_devices;
7058 struct btrfs_device *new_device;
7059 struct rcu_string *name;
7060
7061 if (device->bdev)
7062 fs_devices->open_devices--;
7063
7064 if (device->writeable &&
7065 device->devid != BTRFS_DEV_REPLACE_DEVID) {
7066 list_del_init(&device->dev_alloc_list);
7067 fs_devices->rw_devices--;
7068 }
7069
7070 if (device->missing)
7071 fs_devices->missing_devices--;
7072
7073 new_device = btrfs_alloc_device(NULL, &device->devid,
7074 device->uuid);
7075 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7076
7077 /* Safe because we are under uuid_mutex */
7078 if (device->name) {
7079 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7080 BUG_ON(!name); /* -ENOMEM */
7081 rcu_assign_pointer(new_device->name, name);
7082 }
7083
7084 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7085 new_device->fs_devices = device->fs_devices;
7086
7087 call_rcu(&device->rcu, free_device);
7088 }