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