?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
[linux-2.6.git] / fs / btrfs / volumes.c
blob92303f42baaa92d5d845edddff1f8600fc46518e
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 struct btrfs_root *root,
46 struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
51
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
54
55 static void lock_chunks(struct btrfs_root *root)
56 {
57 mutex_lock(&root->fs_info->chunk_mutex);
58 }
59
60 static void unlock_chunks(struct btrfs_root *root)
61 {
62 mutex_unlock(&root->fs_info->chunk_mutex);
63 }
64
65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
66 {
67 struct btrfs_fs_devices *fs_devs;
68
69 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
70 if (!fs_devs)
71 return ERR_PTR(-ENOMEM);
72
73 mutex_init(&fs_devs->device_list_mutex);
74
75 INIT_LIST_HEAD(&fs_devs->devices);
76 INIT_LIST_HEAD(&fs_devs->alloc_list);
77 INIT_LIST_HEAD(&fs_devs->list);
78
79 return fs_devs;
80 }
81
82 /**
83 * alloc_fs_devices - allocate struct btrfs_fs_devices
84 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
85 * generated.
86 *
87 * Return: a pointer to a new &struct btrfs_fs_devices on success;
88 * ERR_PTR() on error. Returned struct is not linked onto any lists and
89 * can be destroyed with kfree() right away.
90 */
91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
92 {
93 struct btrfs_fs_devices *fs_devs;
94
95 fs_devs = __alloc_fs_devices();
96 if (IS_ERR(fs_devs))
97 return fs_devs;
98
99 if (fsid)
100 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
101 else
102 generate_random_uuid(fs_devs->fsid);
103
104 return fs_devs;
105 }
106
107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
108 {
109 struct btrfs_device *device;
110 WARN_ON(fs_devices->opened);
111 while (!list_empty(&fs_devices->devices)) {
112 device = list_entry(fs_devices->devices.next,
113 struct btrfs_device, dev_list);
114 list_del(&device->dev_list);
115 rcu_string_free(device->name);
116 kfree(device);
117 }
118 kfree(fs_devices);
119 }
120
121 static void btrfs_kobject_uevent(struct block_device *bdev,
122 enum kobject_action action)
123 {
124 int ret;
125
126 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
127 if (ret)
128 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
129 action,
130 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
131 &disk_to_dev(bdev->bd_disk)->kobj);
132 }
133
134 void btrfs_cleanup_fs_uuids(void)
135 {
136 struct btrfs_fs_devices *fs_devices;
137
138 while (!list_empty(&fs_uuids)) {
139 fs_devices = list_entry(fs_uuids.next,
140 struct btrfs_fs_devices, list);
141 list_del(&fs_devices->list);
142 free_fs_devices(fs_devices);
143 }
144 }
145
146 static struct btrfs_device *__alloc_device(void)
147 {
148 struct btrfs_device *dev;
149
150 dev = kzalloc(sizeof(*dev), GFP_NOFS);
151 if (!dev)
152 return ERR_PTR(-ENOMEM);
153
154 INIT_LIST_HEAD(&dev->dev_list);
155 INIT_LIST_HEAD(&dev->dev_alloc_list);
156
157 spin_lock_init(&dev->io_lock);
158
159 spin_lock_init(&dev->reada_lock);
160 atomic_set(&dev->reada_in_flight, 0);
161 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
162 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
163
164 return dev;
165 }
166
167 static noinline struct btrfs_device *__find_device(struct list_head *head,
168 u64 devid, u8 *uuid)
169 {
170 struct btrfs_device *dev;
171
172 list_for_each_entry(dev, head, dev_list) {
173 if (dev->devid == devid &&
174 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
175 return dev;
176 }
177 }
178 return NULL;
179 }
180
181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
182 {
183 struct btrfs_fs_devices *fs_devices;
184
185 list_for_each_entry(fs_devices, &fs_uuids, list) {
186 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
187 return fs_devices;
188 }
189 return NULL;
190 }
191
192 static int
193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
194 int flush, struct block_device **bdev,
195 struct buffer_head **bh)
196 {
197 int ret;
198
199 *bdev = blkdev_get_by_path(device_path, flags, holder);
200
201 if (IS_ERR(*bdev)) {
202 ret = PTR_ERR(*bdev);
203 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
204 goto error;
205 }
206
207 if (flush)
208 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
209 ret = set_blocksize(*bdev, 4096);
210 if (ret) {
211 blkdev_put(*bdev, flags);
212 goto error;
213 }
214 invalidate_bdev(*bdev);
215 *bh = btrfs_read_dev_super(*bdev);
216 if (!*bh) {
217 ret = -EINVAL;
218 blkdev_put(*bdev, flags);
219 goto error;
220 }
221
222 return 0;
223
224 error:
225 *bdev = NULL;
226 *bh = NULL;
227 return ret;
228 }
229
230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
231 struct bio *head, struct bio *tail)
232 {
233
234 struct bio *old_head;
235
236 old_head = pending_bios->head;
237 pending_bios->head = head;
238 if (pending_bios->tail)
239 tail->bi_next = old_head;
240 else
241 pending_bios->tail = tail;
242 }
243
244 /*
245 * we try to collect pending bios for a device so we don't get a large
246 * number of procs sending bios down to the same device. This greatly
247 * improves the schedulers ability to collect and merge the bios.
248 *
249 * But, it also turns into a long list of bios to process and that is sure
250 * to eventually make the worker thread block. The solution here is to
251 * make some progress and then put this work struct back at the end of
252 * the list if the block device is congested. This way, multiple devices
253 * can make progress from a single worker thread.
254 */
255 static noinline void run_scheduled_bios(struct btrfs_device *device)
256 {
257 struct bio *pending;
258 struct backing_dev_info *bdi;
259 struct btrfs_fs_info *fs_info;
260 struct btrfs_pending_bios *pending_bios;
261 struct bio *tail;
262 struct bio *cur;
263 int again = 0;
264 unsigned long num_run;
265 unsigned long batch_run = 0;
266 unsigned long limit;
267 unsigned long last_waited = 0;
268 int force_reg = 0;
269 int sync_pending = 0;
270 struct blk_plug plug;
271
272 /*
273 * this function runs all the bios we've collected for
274 * a particular device. We don't want to wander off to
275 * another device without first sending all of these down.
276 * So, setup a plug here and finish it off before we return
277 */
278 blk_start_plug(&plug);
279
280 bdi = blk_get_backing_dev_info(device->bdev);
281 fs_info = device->dev_root->fs_info;
282 limit = btrfs_async_submit_limit(fs_info);
283 limit = limit * 2 / 3;
284
285 loop:
286 spin_lock(&device->io_lock);
287
288 loop_lock:
289 num_run = 0;
290
291 /* take all the bios off the list at once and process them
292 * later on (without the lock held). But, remember the
293 * tail and other pointers so the bios can be properly reinserted
294 * into the list if we hit congestion
295 */
296 if (!force_reg && device->pending_sync_bios.head) {
297 pending_bios = &device->pending_sync_bios;
298 force_reg = 1;
299 } else {
300 pending_bios = &device->pending_bios;
301 force_reg = 0;
302 }
303
304 pending = pending_bios->head;
305 tail = pending_bios->tail;
306 WARN_ON(pending && !tail);
307
308 /*
309 * if pending was null this time around, no bios need processing
310 * at all and we can stop. Otherwise it'll loop back up again
311 * and do an additional check so no bios are missed.
312 *
313 * device->running_pending is used to synchronize with the
314 * schedule_bio code.
315 */
316 if (device->pending_sync_bios.head == NULL &&
317 device->pending_bios.head == NULL) {
318 again = 0;
319 device->running_pending = 0;
320 } else {
321 again = 1;
322 device->running_pending = 1;
323 }
324
325 pending_bios->head = NULL;
326 pending_bios->tail = NULL;
327
328 spin_unlock(&device->io_lock);
329
330 while (pending) {
331
332 rmb();
333 /* we want to work on both lists, but do more bios on the
334 * sync list than the regular list
335 */
336 if ((num_run > 32 &&
337 pending_bios != &device->pending_sync_bios &&
338 device->pending_sync_bios.head) ||
339 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
340 device->pending_bios.head)) {
341 spin_lock(&device->io_lock);
342 requeue_list(pending_bios, pending, tail);
343 goto loop_lock;
344 }
345
346 cur = pending;
347 pending = pending->bi_next;
348 cur->bi_next = NULL;
349
350 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
351 waitqueue_active(&fs_info->async_submit_wait))
352 wake_up(&fs_info->async_submit_wait);
353
354 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
355
356 /*
357 * if we're doing the sync list, record that our
358 * plug has some sync requests on it
359 *
360 * If we're doing the regular list and there are
361 * sync requests sitting around, unplug before
362 * we add more
363 */
364 if (pending_bios == &device->pending_sync_bios) {
365 sync_pending = 1;
366 } else if (sync_pending) {
367 blk_finish_plug(&plug);
368 blk_start_plug(&plug);
369 sync_pending = 0;
370 }
371
372 btrfsic_submit_bio(cur->bi_rw, cur);
373 num_run++;
374 batch_run++;
375 if (need_resched())
376 cond_resched();
377
378 /*
379 * we made progress, there is more work to do and the bdi
380 * is now congested. Back off and let other work structs
381 * run instead
382 */
383 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
384 fs_info->fs_devices->open_devices > 1) {
385 struct io_context *ioc;
386
387 ioc = current->io_context;
388
389 /*
390 * the main goal here is that we don't want to
391 * block if we're going to be able to submit
392 * more requests without blocking.
393 *
394 * This code does two great things, it pokes into
395 * the elevator code from a filesystem _and_
396 * it makes assumptions about how batching works.
397 */
398 if (ioc && ioc->nr_batch_requests > 0 &&
399 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
400 (last_waited == 0 ||
401 ioc->last_waited == last_waited)) {
402 /*
403 * we want to go through our batch of
404 * requests and stop. So, we copy out
405 * the ioc->last_waited time and test
406 * against it before looping
407 */
408 last_waited = ioc->last_waited;
409 if (need_resched())
410 cond_resched();
411 continue;
412 }
413 spin_lock(&device->io_lock);
414 requeue_list(pending_bios, pending, tail);
415 device->running_pending = 1;
416
417 spin_unlock(&device->io_lock);
418 btrfs_requeue_work(&device->work);
419 goto done;
420 }
421 /* unplug every 64 requests just for good measure */
422 if (batch_run % 64 == 0) {
423 blk_finish_plug(&plug);
424 blk_start_plug(&plug);
425 sync_pending = 0;
426 }
427 }
428
429 cond_resched();
430 if (again)
431 goto loop;
432
433 spin_lock(&device->io_lock);
434 if (device->pending_bios.head || device->pending_sync_bios.head)
435 goto loop_lock;
436 spin_unlock(&device->io_lock);
437
438 done:
439 blk_finish_plug(&plug);
440 }
441
442 static void pending_bios_fn(struct btrfs_work *work)
443 {
444 struct btrfs_device *device;
445
446 device = container_of(work, struct btrfs_device, work);
447 run_scheduled_bios(device);
448 }
449
450 static noinline int device_list_add(const char *path,
451 struct btrfs_super_block *disk_super,
452 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
453 {
454 struct btrfs_device *device;
455 struct btrfs_fs_devices *fs_devices;
456 struct rcu_string *name;
457 u64 found_transid = btrfs_super_generation(disk_super);
458
459 fs_devices = find_fsid(disk_super->fsid);
460 if (!fs_devices) {
461 fs_devices = alloc_fs_devices(disk_super->fsid);
462 if (IS_ERR(fs_devices))
463 return PTR_ERR(fs_devices);
464
465 list_add(&fs_devices->list, &fs_uuids);
466 fs_devices->latest_devid = devid;
467 fs_devices->latest_trans = found_transid;
468
469 device = NULL;
470 } else {
471 device = __find_device(&fs_devices->devices, devid,
472 disk_super->dev_item.uuid);
473 }
474 if (!device) {
475 if (fs_devices->opened)
476 return -EBUSY;
477
478 device = btrfs_alloc_device(NULL, &devid,
479 disk_super->dev_item.uuid);
480 if (IS_ERR(device)) {
481 /* we can safely leave the fs_devices entry around */
482 return PTR_ERR(device);
483 }
484
485 name = rcu_string_strdup(path, GFP_NOFS);
486 if (!name) {
487 kfree(device);
488 return -ENOMEM;
489 }
490 rcu_assign_pointer(device->name, name);
491
492 mutex_lock(&fs_devices->device_list_mutex);
493 list_add_rcu(&device->dev_list, &fs_devices->devices);
494 fs_devices->num_devices++;
495 mutex_unlock(&fs_devices->device_list_mutex);
496
497 device->fs_devices = fs_devices;
498 } else if (!device->name || strcmp(device->name->str, path)) {
499 name = rcu_string_strdup(path, GFP_NOFS);
500 if (!name)
501 return -ENOMEM;
502 rcu_string_free(device->name);
503 rcu_assign_pointer(device->name, name);
504 if (device->missing) {
505 fs_devices->missing_devices--;
506 device->missing = 0;
507 }
508 }
509
510 if (found_transid > fs_devices->latest_trans) {
511 fs_devices->latest_devid = devid;
512 fs_devices->latest_trans = found_transid;
513 }
514 *fs_devices_ret = fs_devices;
515 return 0;
516 }
517
518 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
519 {
520 struct btrfs_fs_devices *fs_devices;
521 struct btrfs_device *device;
522 struct btrfs_device *orig_dev;
523
524 fs_devices = alloc_fs_devices(orig->fsid);
525 if (IS_ERR(fs_devices))
526 return fs_devices;
527
528 fs_devices->latest_devid = orig->latest_devid;
529 fs_devices->latest_trans = orig->latest_trans;
530 fs_devices->total_devices = orig->total_devices;
531
532 /* We have held the volume lock, it is safe to get the devices. */
533 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
534 struct rcu_string *name;
535
536 device = btrfs_alloc_device(NULL, &orig_dev->devid,
537 orig_dev->uuid);
538 if (IS_ERR(device))
539 goto error;
540
541 /*
542 * This is ok to do without rcu read locked because we hold the
543 * uuid mutex so nothing we touch in here is going to disappear.
544 */
545 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
546 if (!name) {
547 kfree(device);
548 goto error;
549 }
550 rcu_assign_pointer(device->name, name);
551
552 list_add(&device->dev_list, &fs_devices->devices);
553 device->fs_devices = fs_devices;
554 fs_devices->num_devices++;
555 }
556 return fs_devices;
557 error:
558 free_fs_devices(fs_devices);
559 return ERR_PTR(-ENOMEM);
560 }
561
562 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
563 struct btrfs_fs_devices *fs_devices, int step)
564 {
565 struct btrfs_device *device, *next;
566
567 struct block_device *latest_bdev = NULL;
568 u64 latest_devid = 0;
569 u64 latest_transid = 0;
570
571 mutex_lock(&uuid_mutex);
572 again:
573 /* This is the initialized path, it is safe to release the devices. */
574 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
575 if (device->in_fs_metadata) {
576 if (!device->is_tgtdev_for_dev_replace &&
577 (!latest_transid ||
578 device->generation > latest_transid)) {
579 latest_devid = device->devid;
580 latest_transid = device->generation;
581 latest_bdev = device->bdev;
582 }
583 continue;
584 }
585
586 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
587 /*
588 * In the first step, keep the device which has
589 * the correct fsid and the devid that is used
590 * for the dev_replace procedure.
591 * In the second step, the dev_replace state is
592 * read from the device tree and it is known
593 * whether the procedure is really active or
594 * not, which means whether this device is
595 * used or whether it should be removed.
596 */
597 if (step == 0 || device->is_tgtdev_for_dev_replace) {
598 continue;
599 }
600 }
601 if (device->bdev) {
602 blkdev_put(device->bdev, device->mode);
603 device->bdev = NULL;
604 fs_devices->open_devices--;
605 }
606 if (device->writeable) {
607 list_del_init(&device->dev_alloc_list);
608 device->writeable = 0;
609 if (!device->is_tgtdev_for_dev_replace)
610 fs_devices->rw_devices--;
611 }
612 list_del_init(&device->dev_list);
613 fs_devices->num_devices--;
614 rcu_string_free(device->name);
615 kfree(device);
616 }
617
618 if (fs_devices->seed) {
619 fs_devices = fs_devices->seed;
620 goto again;
621 }
622
623 fs_devices->latest_bdev = latest_bdev;
624 fs_devices->latest_devid = latest_devid;
625 fs_devices->latest_trans = latest_transid;
626
627 mutex_unlock(&uuid_mutex);
628 }
629
630 static void __free_device(struct work_struct *work)
631 {
632 struct btrfs_device *device;
633
634 device = container_of(work, struct btrfs_device, rcu_work);
635
636 if (device->bdev)
637 blkdev_put(device->bdev, device->mode);
638
639 rcu_string_free(device->name);
640 kfree(device);
641 }
642
643 static void free_device(struct rcu_head *head)
644 {
645 struct btrfs_device *device;
646
647 device = container_of(head, struct btrfs_device, rcu);
648
649 INIT_WORK(&device->rcu_work, __free_device);
650 schedule_work(&device->rcu_work);
651 }
652
653 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
654 {
655 struct btrfs_device *device;
656
657 if (--fs_devices->opened > 0)
658 return 0;
659
660 mutex_lock(&fs_devices->device_list_mutex);
661 list_for_each_entry(device, &fs_devices->devices, dev_list) {
662 struct btrfs_device *new_device;
663 struct rcu_string *name;
664
665 if (device->bdev)
666 fs_devices->open_devices--;
667
668 if (device->writeable &&
669 device->devid != BTRFS_DEV_REPLACE_DEVID) {
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 if (device->missing)
677 fs_devices->missing_devices--;
678
679 new_device = btrfs_alloc_device(NULL, &device->devid,
680 device->uuid);
681 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
682
683 /* Safe because we are under uuid_mutex */
684 if (device->name) {
685 name = rcu_string_strdup(device->name->str, GFP_NOFS);
686 BUG_ON(!name); /* -ENOMEM */
687 rcu_assign_pointer(new_device->name, name);
688 }
689
690 list_replace_rcu(&device->dev_list, &new_device->dev_list);
691 new_device->fs_devices = device->fs_devices;
692
693 call_rcu(&device->rcu, free_device);
694 }
695 mutex_unlock(&fs_devices->device_list_mutex);
696
697 WARN_ON(fs_devices->open_devices);
698 WARN_ON(fs_devices->rw_devices);
699 fs_devices->opened = 0;
700 fs_devices->seeding = 0;
701
702 return 0;
703 }
704
705 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
706 {
707 struct btrfs_fs_devices *seed_devices = NULL;
708 int ret;
709
710 mutex_lock(&uuid_mutex);
711 ret = __btrfs_close_devices(fs_devices);
712 if (!fs_devices->opened) {
713 seed_devices = fs_devices->seed;
714 fs_devices->seed = NULL;
715 }
716 mutex_unlock(&uuid_mutex);
717
718 while (seed_devices) {
719 fs_devices = seed_devices;
720 seed_devices = fs_devices->seed;
721 __btrfs_close_devices(fs_devices);
722 free_fs_devices(fs_devices);
723 }
724 /*
725 * Wait for rcu kworkers under __btrfs_close_devices
726 * to finish all blkdev_puts so device is really
727 * free when umount is done.
728 */
729 rcu_barrier();
730 return ret;
731 }
732
733 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
734 fmode_t flags, void *holder)
735 {
736 struct request_queue *q;
737 struct block_device *bdev;
738 struct list_head *head = &fs_devices->devices;
739 struct btrfs_device *device;
740 struct block_device *latest_bdev = NULL;
741 struct buffer_head *bh;
742 struct btrfs_super_block *disk_super;
743 u64 latest_devid = 0;
744 u64 latest_transid = 0;
745 u64 devid;
746 int seeding = 1;
747 int ret = 0;
748
749 flags |= FMODE_EXCL;
750
751 list_for_each_entry(device, head, dev_list) {
752 if (device->bdev)
753 continue;
754 if (!device->name)
755 continue;
756
757 /* Just open everything we can; ignore failures here */
758 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
759 &bdev, &bh))
760 continue;
761
762 disk_super = (struct btrfs_super_block *)bh->b_data;
763 devid = btrfs_stack_device_id(&disk_super->dev_item);
764 if (devid != device->devid)
765 goto error_brelse;
766
767 if (memcmp(device->uuid, disk_super->dev_item.uuid,
768 BTRFS_UUID_SIZE))
769 goto error_brelse;
770
771 device->generation = btrfs_super_generation(disk_super);
772 if (!latest_transid || device->generation > latest_transid) {
773 latest_devid = devid;
774 latest_transid = device->generation;
775 latest_bdev = bdev;
776 }
777
778 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
779 device->writeable = 0;
780 } else {
781 device->writeable = !bdev_read_only(bdev);
782 seeding = 0;
783 }
784
785 q = bdev_get_queue(bdev);
786 if (blk_queue_discard(q)) {
787 device->can_discard = 1;
788 fs_devices->num_can_discard++;
789 }
790
791 device->bdev = bdev;
792 device->in_fs_metadata = 0;
793 device->mode = flags;
794
795 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
796 fs_devices->rotating = 1;
797
798 fs_devices->open_devices++;
799 if (device->writeable &&
800 device->devid != BTRFS_DEV_REPLACE_DEVID) {
801 fs_devices->rw_devices++;
802 list_add(&device->dev_alloc_list,
803 &fs_devices->alloc_list);
804 }
805 brelse(bh);
806 continue;
807
808 error_brelse:
809 brelse(bh);
810 blkdev_put(bdev, flags);
811 continue;
812 }
813 if (fs_devices->open_devices == 0) {
814 ret = -EINVAL;
815 goto out;
816 }
817 fs_devices->seeding = seeding;
818 fs_devices->opened = 1;
819 fs_devices->latest_bdev = latest_bdev;
820 fs_devices->latest_devid = latest_devid;
821 fs_devices->latest_trans = latest_transid;
822 fs_devices->total_rw_bytes = 0;
823 out:
824 return ret;
825 }
826
827 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
828 fmode_t flags, void *holder)
829 {
830 int ret;
831
832 mutex_lock(&uuid_mutex);
833 if (fs_devices->opened) {
834 fs_devices->opened++;
835 ret = 0;
836 } else {
837 ret = __btrfs_open_devices(fs_devices, flags, holder);
838 }
839 mutex_unlock(&uuid_mutex);
840 return ret;
841 }
842
843 /*
844 * Look for a btrfs signature on a device. This may be called out of the mount path
845 * and we are not allowed to call set_blocksize during the scan. The superblock
846 * is read via pagecache
847 */
848 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
849 struct btrfs_fs_devices **fs_devices_ret)
850 {
851 struct btrfs_super_block *disk_super;
852 struct block_device *bdev;
853 struct page *page;
854 void *p;
855 int ret = -EINVAL;
856 u64 devid;
857 u64 transid;
858 u64 total_devices;
859 u64 bytenr;
860 pgoff_t index;
861
862 /*
863 * we would like to check all the supers, but that would make
864 * a btrfs mount succeed after a mkfs from a different FS.
865 * So, we need to add a special mount option to scan for
866 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
867 */
868 bytenr = btrfs_sb_offset(0);
869 flags |= FMODE_EXCL;
870 mutex_lock(&uuid_mutex);
871
872 bdev = blkdev_get_by_path(path, flags, holder);
873
874 if (IS_ERR(bdev)) {
875 ret = PTR_ERR(bdev);
876 goto error;
877 }
878
879 /* make sure our super fits in the device */
880 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
881 goto error_bdev_put;
882
883 /* make sure our super fits in the page */
884 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
885 goto error_bdev_put;
886
887 /* make sure our super doesn't straddle pages on disk */
888 index = bytenr >> PAGE_CACHE_SHIFT;
889 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
890 goto error_bdev_put;
891
892 /* pull in the page with our super */
893 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
894 index, GFP_NOFS);
895
896 if (IS_ERR_OR_NULL(page))
897 goto error_bdev_put;
898
899 p = kmap(page);
900
901 /* align our pointer to the offset of the super block */
902 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
903
904 if (btrfs_super_bytenr(disk_super) != bytenr ||
905 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
906 goto error_unmap;
907
908 devid = btrfs_stack_device_id(&disk_super->dev_item);
909 transid = btrfs_super_generation(disk_super);
910 total_devices = btrfs_super_num_devices(disk_super);
911
912 if (disk_super->label[0]) {
913 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
914 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
915 printk(KERN_INFO "btrfs: device label %s ", disk_super->label);
916 } else {
917 printk(KERN_INFO "btrfs: device fsid %pU ", disk_super->fsid);
918 }
919
920 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
921
922 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
923 if (!ret && fs_devices_ret)
924 (*fs_devices_ret)->total_devices = total_devices;
925
926 error_unmap:
927 kunmap(page);
928 page_cache_release(page);
929
930 error_bdev_put:
931 blkdev_put(bdev, flags);
932 error:
933 mutex_unlock(&uuid_mutex);
934 return ret;
935 }
936
937 /* helper to account the used device space in the range */
938 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
939 u64 end, u64 *length)
940 {
941 struct btrfs_key key;
942 struct btrfs_root *root = device->dev_root;
943 struct btrfs_dev_extent *dev_extent;
944 struct btrfs_path *path;
945 u64 extent_end;
946 int ret;
947 int slot;
948 struct extent_buffer *l;
949
950 *length = 0;
951
952 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
953 return 0;
954
955 path = btrfs_alloc_path();
956 if (!path)
957 return -ENOMEM;
958 path->reada = 2;
959
960 key.objectid = device->devid;
961 key.offset = start;
962 key.type = BTRFS_DEV_EXTENT_KEY;
963
964 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
965 if (ret < 0)
966 goto out;
967 if (ret > 0) {
968 ret = btrfs_previous_item(root, path, key.objectid, key.type);
969 if (ret < 0)
970 goto out;
971 }
972
973 while (1) {
974 l = path->nodes[0];
975 slot = path->slots[0];
976 if (slot >= btrfs_header_nritems(l)) {
977 ret = btrfs_next_leaf(root, path);
978 if (ret == 0)
979 continue;
980 if (ret < 0)
981 goto out;
982
983 break;
984 }
985 btrfs_item_key_to_cpu(l, &key, slot);
986
987 if (key.objectid < device->devid)
988 goto next;
989
990 if (key.objectid > device->devid)
991 break;
992
993 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
994 goto next;
995
996 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
997 extent_end = key.offset + btrfs_dev_extent_length(l,
998 dev_extent);
999 if (key.offset <= start && extent_end > end) {
1000 *length = end - start + 1;
1001 break;
1002 } else if (key.offset <= start && extent_end > start)
1003 *length += extent_end - start;
1004 else if (key.offset > start && extent_end <= end)
1005 *length += extent_end - key.offset;
1006 else if (key.offset > start && key.offset <= end) {
1007 *length += end - key.offset + 1;
1008 break;
1009 } else if (key.offset > end)
1010 break;
1011
1012 next:
1013 path->slots[0]++;
1014 }
1015 ret = 0;
1016 out:
1017 btrfs_free_path(path);
1018 return ret;
1019 }
1020
1021 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1022 struct btrfs_device *device,
1023 u64 *start, u64 len)
1024 {
1025 struct extent_map *em;
1026 int ret = 0;
1027
1028 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1029 struct map_lookup *map;
1030 int i;
1031
1032 map = (struct map_lookup *)em->bdev;
1033 for (i = 0; i < map->num_stripes; i++) {
1034 if (map->stripes[i].dev != device)
1035 continue;
1036 if (map->stripes[i].physical >= *start + len ||
1037 map->stripes[i].physical + em->orig_block_len <=
1038 *start)
1039 continue;
1040 *start = map->stripes[i].physical +
1041 em->orig_block_len;
1042 ret = 1;
1043 }
1044 }
1045
1046 return ret;
1047 }
1048
1049
1050 /*
1051 * find_free_dev_extent - find free space in the specified device
1052 * @device: the device which we search the free space in
1053 * @num_bytes: the size of the free space that we need
1054 * @start: store the start of the free space.
1055 * @len: the size of the free space. that we find, or the size of the max
1056 * free space if we don't find suitable free space
1057 *
1058 * this uses a pretty simple search, the expectation is that it is
1059 * called very infrequently and that a given device has a small number
1060 * of extents
1061 *
1062 * @start is used to store the start of the free space if we find. But if we
1063 * don't find suitable free space, it will be used to store the start position
1064 * of the max free space.
1065 *
1066 * @len is used to store the size of the free space that we find.
1067 * But if we don't find suitable free space, it is used to store the size of
1068 * the max free space.
1069 */
1070 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1071 struct btrfs_device *device, u64 num_bytes,
1072 u64 *start, u64 *len)
1073 {
1074 struct btrfs_key key;
1075 struct btrfs_root *root = device->dev_root;
1076 struct btrfs_dev_extent *dev_extent;
1077 struct btrfs_path *path;
1078 u64 hole_size;
1079 u64 max_hole_start;
1080 u64 max_hole_size;
1081 u64 extent_end;
1082 u64 search_start;
1083 u64 search_end = device->total_bytes;
1084 int ret;
1085 int slot;
1086 struct extent_buffer *l;
1087
1088 /* FIXME use last free of some kind */
1089
1090 /* we don't want to overwrite the superblock on the drive,
1091 * so we make sure to start at an offset of at least 1MB
1092 */
1093 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1094
1095 path = btrfs_alloc_path();
1096 if (!path)
1097 return -ENOMEM;
1098 again:
1099 max_hole_start = search_start;
1100 max_hole_size = 0;
1101 hole_size = 0;
1102
1103 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1104 ret = -ENOSPC;
1105 goto out;
1106 }
1107
1108 path->reada = 2;
1109 path->search_commit_root = 1;
1110 path->skip_locking = 1;
1111
1112 key.objectid = device->devid;
1113 key.offset = search_start;
1114 key.type = BTRFS_DEV_EXTENT_KEY;
1115
1116 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1117 if (ret < 0)
1118 goto out;
1119 if (ret > 0) {
1120 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1121 if (ret < 0)
1122 goto out;
1123 }
1124
1125 while (1) {
1126 l = path->nodes[0];
1127 slot = path->slots[0];
1128 if (slot >= btrfs_header_nritems(l)) {
1129 ret = btrfs_next_leaf(root, path);
1130 if (ret == 0)
1131 continue;
1132 if (ret < 0)
1133 goto out;
1134
1135 break;
1136 }
1137 btrfs_item_key_to_cpu(l, &key, slot);
1138
1139 if (key.objectid < device->devid)
1140 goto next;
1141
1142 if (key.objectid > device->devid)
1143 break;
1144
1145 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1146 goto next;
1147
1148 if (key.offset > search_start) {
1149 hole_size = key.offset - search_start;
1150
1151 /*
1152 * Have to check before we set max_hole_start, otherwise
1153 * we could end up sending back this offset anyway.
1154 */
1155 if (contains_pending_extent(trans, device,
1156 &search_start,
1157 hole_size))
1158 hole_size = 0;
1159
1160 if (hole_size > max_hole_size) {
1161 max_hole_start = search_start;
1162 max_hole_size = hole_size;
1163 }
1164
1165 /*
1166 * If this free space is greater than which we need,
1167 * it must be the max free space that we have found
1168 * until now, so max_hole_start must point to the start
1169 * of this free space and the length of this free space
1170 * is stored in max_hole_size. Thus, we return
1171 * max_hole_start and max_hole_size and go back to the
1172 * caller.
1173 */
1174 if (hole_size >= num_bytes) {
1175 ret = 0;
1176 goto out;
1177 }
1178 }
1179
1180 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1181 extent_end = key.offset + btrfs_dev_extent_length(l,
1182 dev_extent);
1183 if (extent_end > search_start)
1184 search_start = extent_end;
1185 next:
1186 path->slots[0]++;
1187 cond_resched();
1188 }
1189
1190 /*
1191 * At this point, search_start should be the end of
1192 * allocated dev extents, and when shrinking the device,
1193 * search_end may be smaller than search_start.
1194 */
1195 if (search_end > search_start)
1196 hole_size = search_end - search_start;
1197
1198 if (hole_size > max_hole_size) {
1199 max_hole_start = search_start;
1200 max_hole_size = hole_size;
1201 }
1202
1203 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1204 btrfs_release_path(path);
1205 goto again;
1206 }
1207
1208 /* See above. */
1209 if (hole_size < num_bytes)
1210 ret = -ENOSPC;
1211 else
1212 ret = 0;
1213
1214 out:
1215 btrfs_free_path(path);
1216 *start = max_hole_start;
1217 if (len)
1218 *len = max_hole_size;
1219 return ret;
1220 }
1221
1222 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1223 struct btrfs_device *device,
1224 u64 start)
1225 {
1226 int ret;
1227 struct btrfs_path *path;
1228 struct btrfs_root *root = device->dev_root;
1229 struct btrfs_key key;
1230 struct btrfs_key found_key;
1231 struct extent_buffer *leaf = NULL;
1232 struct btrfs_dev_extent *extent = NULL;
1233
1234 path = btrfs_alloc_path();
1235 if (!path)
1236 return -ENOMEM;
1237
1238 key.objectid = device->devid;
1239 key.offset = start;
1240 key.type = BTRFS_DEV_EXTENT_KEY;
1241 again:
1242 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1243 if (ret > 0) {
1244 ret = btrfs_previous_item(root, path, key.objectid,
1245 BTRFS_DEV_EXTENT_KEY);
1246 if (ret)
1247 goto out;
1248 leaf = path->nodes[0];
1249 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1250 extent = btrfs_item_ptr(leaf, path->slots[0],
1251 struct btrfs_dev_extent);
1252 BUG_ON(found_key.offset > start || found_key.offset +
1253 btrfs_dev_extent_length(leaf, extent) < start);
1254 key = found_key;
1255 btrfs_release_path(path);
1256 goto again;
1257 } else if (ret == 0) {
1258 leaf = path->nodes[0];
1259 extent = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_dev_extent);
1261 } else {
1262 btrfs_error(root->fs_info, ret, "Slot search failed");
1263 goto out;
1264 }
1265
1266 if (device->bytes_used > 0) {
1267 u64 len = btrfs_dev_extent_length(leaf, extent);
1268 device->bytes_used -= len;
1269 spin_lock(&root->fs_info->free_chunk_lock);
1270 root->fs_info->free_chunk_space += len;
1271 spin_unlock(&root->fs_info->free_chunk_lock);
1272 }
1273 ret = btrfs_del_item(trans, root, path);
1274 if (ret) {
1275 btrfs_error(root->fs_info, ret,
1276 "Failed to remove dev extent item");
1277 }
1278 out:
1279 btrfs_free_path(path);
1280 return ret;
1281 }
1282
1283 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1284 struct btrfs_device *device,
1285 u64 chunk_tree, u64 chunk_objectid,
1286 u64 chunk_offset, u64 start, u64 num_bytes)
1287 {
1288 int ret;
1289 struct btrfs_path *path;
1290 struct btrfs_root *root = device->dev_root;
1291 struct btrfs_dev_extent *extent;
1292 struct extent_buffer *leaf;
1293 struct btrfs_key key;
1294
1295 WARN_ON(!device->in_fs_metadata);
1296 WARN_ON(device->is_tgtdev_for_dev_replace);
1297 path = btrfs_alloc_path();
1298 if (!path)
1299 return -ENOMEM;
1300
1301 key.objectid = device->devid;
1302 key.offset = start;
1303 key.type = BTRFS_DEV_EXTENT_KEY;
1304 ret = btrfs_insert_empty_item(trans, root, path, &key,
1305 sizeof(*extent));
1306 if (ret)
1307 goto out;
1308
1309 leaf = path->nodes[0];
1310 extent = btrfs_item_ptr(leaf, path->slots[0],
1311 struct btrfs_dev_extent);
1312 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1313 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1314 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1315
1316 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1317 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1318
1319 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1320 btrfs_mark_buffer_dirty(leaf);
1321 out:
1322 btrfs_free_path(path);
1323 return ret;
1324 }
1325
1326 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1327 {
1328 struct extent_map_tree *em_tree;
1329 struct extent_map *em;
1330 struct rb_node *n;
1331 u64 ret = 0;
1332
1333 em_tree = &fs_info->mapping_tree.map_tree;
1334 read_lock(&em_tree->lock);
1335 n = rb_last(&em_tree->map);
1336 if (n) {
1337 em = rb_entry(n, struct extent_map, rb_node);
1338 ret = em->start + em->len;
1339 }
1340 read_unlock(&em_tree->lock);
1341
1342 return ret;
1343 }
1344
1345 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1346 u64 *devid_ret)
1347 {
1348 int ret;
1349 struct btrfs_key key;
1350 struct btrfs_key found_key;
1351 struct btrfs_path *path;
1352
1353 path = btrfs_alloc_path();
1354 if (!path)
1355 return -ENOMEM;
1356
1357 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1358 key.type = BTRFS_DEV_ITEM_KEY;
1359 key.offset = (u64)-1;
1360
1361 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1362 if (ret < 0)
1363 goto error;
1364
1365 BUG_ON(ret == 0); /* Corruption */
1366
1367 ret = btrfs_previous_item(fs_info->chunk_root, path,
1368 BTRFS_DEV_ITEMS_OBJECTID,
1369 BTRFS_DEV_ITEM_KEY);
1370 if (ret) {
1371 *devid_ret = 1;
1372 } else {
1373 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1374 path->slots[0]);
1375 *devid_ret = found_key.offset + 1;
1376 }
1377 ret = 0;
1378 error:
1379 btrfs_free_path(path);
1380 return ret;
1381 }
1382
1383 /*
1384 * the device information is stored in the chunk root
1385 * the btrfs_device struct should be fully filled in
1386 */
1387 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1388 struct btrfs_root *root,
1389 struct btrfs_device *device)
1390 {
1391 int ret;
1392 struct btrfs_path *path;
1393 struct btrfs_dev_item *dev_item;
1394 struct extent_buffer *leaf;
1395 struct btrfs_key key;
1396 unsigned long ptr;
1397
1398 root = root->fs_info->chunk_root;
1399
1400 path = btrfs_alloc_path();
1401 if (!path)
1402 return -ENOMEM;
1403
1404 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1405 key.type = BTRFS_DEV_ITEM_KEY;
1406 key.offset = device->devid;
1407
1408 ret = btrfs_insert_empty_item(trans, root, path, &key,
1409 sizeof(*dev_item));
1410 if (ret)
1411 goto out;
1412
1413 leaf = path->nodes[0];
1414 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1415
1416 btrfs_set_device_id(leaf, dev_item, device->devid);
1417 btrfs_set_device_generation(leaf, dev_item, 0);
1418 btrfs_set_device_type(leaf, dev_item, device->type);
1419 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1420 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1421 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1422 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1423 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1424 btrfs_set_device_group(leaf, dev_item, 0);
1425 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1426 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1427 btrfs_set_device_start_offset(leaf, dev_item, 0);
1428
1429 ptr = btrfs_device_uuid(dev_item);
1430 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1431 ptr = btrfs_device_fsid(dev_item);
1432 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1433 btrfs_mark_buffer_dirty(leaf);
1434
1435 ret = 0;
1436 out:
1437 btrfs_free_path(path);
1438 return ret;
1439 }
1440
1441 static int btrfs_rm_dev_item(struct btrfs_root *root,
1442 struct btrfs_device *device)
1443 {
1444 int ret;
1445 struct btrfs_path *path;
1446 struct btrfs_key key;
1447 struct btrfs_trans_handle *trans;
1448
1449 root = root->fs_info->chunk_root;
1450
1451 path = btrfs_alloc_path();
1452 if (!path)
1453 return -ENOMEM;
1454
1455 trans = btrfs_start_transaction(root, 0);
1456 if (IS_ERR(trans)) {
1457 btrfs_free_path(path);
1458 return PTR_ERR(trans);
1459 }
1460 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1461 key.type = BTRFS_DEV_ITEM_KEY;
1462 key.offset = device->devid;
1463 lock_chunks(root);
1464
1465 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1466 if (ret < 0)
1467 goto out;
1468
1469 if (ret > 0) {
1470 ret = -ENOENT;
1471 goto out;
1472 }
1473
1474 ret = btrfs_del_item(trans, root, path);
1475 if (ret)
1476 goto out;
1477 out:
1478 btrfs_free_path(path);
1479 unlock_chunks(root);
1480 btrfs_commit_transaction(trans, root);
1481 return ret;
1482 }
1483
1484 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1485 {
1486 struct btrfs_device *device;
1487 struct btrfs_device *next_device;
1488 struct block_device *bdev;
1489 struct buffer_head *bh = NULL;
1490 struct btrfs_super_block *disk_super;
1491 struct btrfs_fs_devices *cur_devices;
1492 u64 all_avail;
1493 u64 devid;
1494 u64 num_devices;
1495 u8 *dev_uuid;
1496 unsigned seq;
1497 int ret = 0;
1498 bool clear_super = false;
1499
1500 mutex_lock(&uuid_mutex);
1501
1502 do {
1503 seq = read_seqbegin(&root->fs_info->profiles_lock);
1504
1505 all_avail = root->fs_info->avail_data_alloc_bits |
1506 root->fs_info->avail_system_alloc_bits |
1507 root->fs_info->avail_metadata_alloc_bits;
1508 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1509
1510 num_devices = root->fs_info->fs_devices->num_devices;
1511 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1512 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1513 WARN_ON(num_devices < 1);
1514 num_devices--;
1515 }
1516 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1517
1518 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1519 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1520 goto out;
1521 }
1522
1523 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1524 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1525 goto out;
1526 }
1527
1528 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1529 root->fs_info->fs_devices->rw_devices <= 2) {
1530 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1531 goto out;
1532 }
1533 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1534 root->fs_info->fs_devices->rw_devices <= 3) {
1535 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1536 goto out;
1537 }
1538
1539 if (strcmp(device_path, "missing") == 0) {
1540 struct list_head *devices;
1541 struct btrfs_device *tmp;
1542
1543 device = NULL;
1544 devices = &root->fs_info->fs_devices->devices;
1545 /*
1546 * It is safe to read the devices since the volume_mutex
1547 * is held.
1548 */
1549 list_for_each_entry(tmp, devices, dev_list) {
1550 if (tmp->in_fs_metadata &&
1551 !tmp->is_tgtdev_for_dev_replace &&
1552 !tmp->bdev) {
1553 device = tmp;
1554 break;
1555 }
1556 }
1557 bdev = NULL;
1558 bh = NULL;
1559 disk_super = NULL;
1560 if (!device) {
1561 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1562 goto out;
1563 }
1564 } else {
1565 ret = btrfs_get_bdev_and_sb(device_path,
1566 FMODE_WRITE | FMODE_EXCL,
1567 root->fs_info->bdev_holder, 0,
1568 &bdev, &bh);
1569 if (ret)
1570 goto out;
1571 disk_super = (struct btrfs_super_block *)bh->b_data;
1572 devid = btrfs_stack_device_id(&disk_super->dev_item);
1573 dev_uuid = disk_super->dev_item.uuid;
1574 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1575 disk_super->fsid);
1576 if (!device) {
1577 ret = -ENOENT;
1578 goto error_brelse;
1579 }
1580 }
1581
1582 if (device->is_tgtdev_for_dev_replace) {
1583 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1584 goto error_brelse;
1585 }
1586
1587 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1588 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1589 goto error_brelse;
1590 }
1591
1592 if (device->writeable) {
1593 lock_chunks(root);
1594 list_del_init(&device->dev_alloc_list);
1595 unlock_chunks(root);
1596 root->fs_info->fs_devices->rw_devices--;
1597 clear_super = true;
1598 }
1599
1600 mutex_unlock(&uuid_mutex);
1601 ret = btrfs_shrink_device(device, 0);
1602 mutex_lock(&uuid_mutex);
1603 if (ret)
1604 goto error_undo;
1605
1606 /*
1607 * TODO: the superblock still includes this device in its num_devices
1608 * counter although write_all_supers() is not locked out. This
1609 * could give a filesystem state which requires a degraded mount.
1610 */
1611 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1612 if (ret)
1613 goto error_undo;
1614
1615 spin_lock(&root->fs_info->free_chunk_lock);
1616 root->fs_info->free_chunk_space = device->total_bytes -
1617 device->bytes_used;
1618 spin_unlock(&root->fs_info->free_chunk_lock);
1619
1620 device->in_fs_metadata = 0;
1621 btrfs_scrub_cancel_dev(root->fs_info, device);
1622
1623 /*
1624 * the device list mutex makes sure that we don't change
1625 * the device list while someone else is writing out all
1626 * the device supers. Whoever is writing all supers, should
1627 * lock the device list mutex before getting the number of
1628 * devices in the super block (super_copy). Conversely,
1629 * whoever updates the number of devices in the super block
1630 * (super_copy) should hold the device list mutex.
1631 */
1632
1633 cur_devices = device->fs_devices;
1634 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1635 list_del_rcu(&device->dev_list);
1636
1637 device->fs_devices->num_devices--;
1638 device->fs_devices->total_devices--;
1639
1640 if (device->missing)
1641 root->fs_info->fs_devices->missing_devices--;
1642
1643 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1644 struct btrfs_device, dev_list);
1645 if (device->bdev == root->fs_info->sb->s_bdev)
1646 root->fs_info->sb->s_bdev = next_device->bdev;
1647 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1648 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1649
1650 if (device->bdev)
1651 device->fs_devices->open_devices--;
1652
1653 call_rcu(&device->rcu, free_device);
1654
1655 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1656 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1657 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1658
1659 if (cur_devices->open_devices == 0) {
1660 struct btrfs_fs_devices *fs_devices;
1661 fs_devices = root->fs_info->fs_devices;
1662 while (fs_devices) {
1663 if (fs_devices->seed == cur_devices)
1664 break;
1665 fs_devices = fs_devices->seed;
1666 }
1667 fs_devices->seed = cur_devices->seed;
1668 cur_devices->seed = NULL;
1669 lock_chunks(root);
1670 __btrfs_close_devices(cur_devices);
1671 unlock_chunks(root);
1672 free_fs_devices(cur_devices);
1673 }
1674
1675 root->fs_info->num_tolerated_disk_barrier_failures =
1676 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1677
1678 /*
1679 * at this point, the device is zero sized. We want to
1680 * remove it from the devices list and zero out the old super
1681 */
1682 if (clear_super && disk_super) {
1683 /* make sure this device isn't detected as part of
1684 * the FS anymore
1685 */
1686 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1687 set_buffer_dirty(bh);
1688 sync_dirty_buffer(bh);
1689 }
1690
1691 ret = 0;
1692
1693 /* Notify udev that device has changed */
1694 if (bdev)
1695 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1696
1697 error_brelse:
1698 brelse(bh);
1699 if (bdev)
1700 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1701 out:
1702 mutex_unlock(&uuid_mutex);
1703 return ret;
1704 error_undo:
1705 if (device->writeable) {
1706 lock_chunks(root);
1707 list_add(&device->dev_alloc_list,
1708 &root->fs_info->fs_devices->alloc_list);
1709 unlock_chunks(root);
1710 root->fs_info->fs_devices->rw_devices++;
1711 }
1712 goto error_brelse;
1713 }
1714
1715 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1716 struct btrfs_device *srcdev)
1717 {
1718 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1719
1720 list_del_rcu(&srcdev->dev_list);
1721 list_del_rcu(&srcdev->dev_alloc_list);
1722 fs_info->fs_devices->num_devices--;
1723 if (srcdev->missing) {
1724 fs_info->fs_devices->missing_devices--;
1725 fs_info->fs_devices->rw_devices++;
1726 }
1727 if (srcdev->can_discard)
1728 fs_info->fs_devices->num_can_discard--;
1729 if (srcdev->bdev) {
1730 fs_info->fs_devices->open_devices--;
1731
1732 /* zero out the old super */
1733 btrfs_scratch_superblock(srcdev);
1734 }
1735
1736 call_rcu(&srcdev->rcu, free_device);
1737 }
1738
1739 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1740 struct btrfs_device *tgtdev)
1741 {
1742 struct btrfs_device *next_device;
1743
1744 WARN_ON(!tgtdev);
1745 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1746 if (tgtdev->bdev) {
1747 btrfs_scratch_superblock(tgtdev);
1748 fs_info->fs_devices->open_devices--;
1749 }
1750 fs_info->fs_devices->num_devices--;
1751 if (tgtdev->can_discard)
1752 fs_info->fs_devices->num_can_discard++;
1753
1754 next_device = list_entry(fs_info->fs_devices->devices.next,
1755 struct btrfs_device, dev_list);
1756 if (tgtdev->bdev == fs_info->sb->s_bdev)
1757 fs_info->sb->s_bdev = next_device->bdev;
1758 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1759 fs_info->fs_devices->latest_bdev = next_device->bdev;
1760 list_del_rcu(&tgtdev->dev_list);
1761
1762 call_rcu(&tgtdev->rcu, free_device);
1763
1764 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1765 }
1766
1767 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1768 struct btrfs_device **device)
1769 {
1770 int ret = 0;
1771 struct btrfs_super_block *disk_super;
1772 u64 devid;
1773 u8 *dev_uuid;
1774 struct block_device *bdev;
1775 struct buffer_head *bh;
1776
1777 *device = NULL;
1778 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1779 root->fs_info->bdev_holder, 0, &bdev, &bh);
1780 if (ret)
1781 return ret;
1782 disk_super = (struct btrfs_super_block *)bh->b_data;
1783 devid = btrfs_stack_device_id(&disk_super->dev_item);
1784 dev_uuid = disk_super->dev_item.uuid;
1785 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1786 disk_super->fsid);
1787 brelse(bh);
1788 if (!*device)
1789 ret = -ENOENT;
1790 blkdev_put(bdev, FMODE_READ);
1791 return ret;
1792 }
1793
1794 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1795 char *device_path,
1796 struct btrfs_device **device)
1797 {
1798 *device = NULL;
1799 if (strcmp(device_path, "missing") == 0) {
1800 struct list_head *devices;
1801 struct btrfs_device *tmp;
1802
1803 devices = &root->fs_info->fs_devices->devices;
1804 /*
1805 * It is safe to read the devices since the volume_mutex
1806 * is held by the caller.
1807 */
1808 list_for_each_entry(tmp, devices, dev_list) {
1809 if (tmp->in_fs_metadata && !tmp->bdev) {
1810 *device = tmp;
1811 break;
1812 }
1813 }
1814
1815 if (!*device) {
1816 pr_err("btrfs: no missing device found\n");
1817 return -ENOENT;
1818 }
1819
1820 return 0;
1821 } else {
1822 return btrfs_find_device_by_path(root, device_path, device);
1823 }
1824 }
1825
1826 /*
1827 * does all the dirty work required for changing file system's UUID.
1828 */
1829 static int btrfs_prepare_sprout(struct btrfs_root *root)
1830 {
1831 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1832 struct btrfs_fs_devices *old_devices;
1833 struct btrfs_fs_devices *seed_devices;
1834 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1835 struct btrfs_device *device;
1836 u64 super_flags;
1837
1838 BUG_ON(!mutex_is_locked(&uuid_mutex));
1839 if (!fs_devices->seeding)
1840 return -EINVAL;
1841
1842 seed_devices = __alloc_fs_devices();
1843 if (IS_ERR(seed_devices))
1844 return PTR_ERR(seed_devices);
1845
1846 old_devices = clone_fs_devices(fs_devices);
1847 if (IS_ERR(old_devices)) {
1848 kfree(seed_devices);
1849 return PTR_ERR(old_devices);
1850 }
1851
1852 list_add(&old_devices->list, &fs_uuids);
1853
1854 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1855 seed_devices->opened = 1;
1856 INIT_LIST_HEAD(&seed_devices->devices);
1857 INIT_LIST_HEAD(&seed_devices->alloc_list);
1858 mutex_init(&seed_devices->device_list_mutex);
1859
1860 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1861 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1862 synchronize_rcu);
1863
1864 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1865 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1866 device->fs_devices = seed_devices;
1867 }
1868
1869 fs_devices->seeding = 0;
1870 fs_devices->num_devices = 0;
1871 fs_devices->open_devices = 0;
1872 fs_devices->total_devices = 0;
1873 fs_devices->seed = seed_devices;
1874
1875 generate_random_uuid(fs_devices->fsid);
1876 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1877 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1878 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1879
1880 super_flags = btrfs_super_flags(disk_super) &
1881 ~BTRFS_SUPER_FLAG_SEEDING;
1882 btrfs_set_super_flags(disk_super, super_flags);
1883
1884 return 0;
1885 }
1886
1887 /*
1888 * strore the expected generation for seed devices in device items.
1889 */
1890 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1891 struct btrfs_root *root)
1892 {
1893 struct btrfs_path *path;
1894 struct extent_buffer *leaf;
1895 struct btrfs_dev_item *dev_item;
1896 struct btrfs_device *device;
1897 struct btrfs_key key;
1898 u8 fs_uuid[BTRFS_UUID_SIZE];
1899 u8 dev_uuid[BTRFS_UUID_SIZE];
1900 u64 devid;
1901 int ret;
1902
1903 path = btrfs_alloc_path();
1904 if (!path)
1905 return -ENOMEM;
1906
1907 root = root->fs_info->chunk_root;
1908 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1909 key.offset = 0;
1910 key.type = BTRFS_DEV_ITEM_KEY;
1911
1912 while (1) {
1913 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1914 if (ret < 0)
1915 goto error;
1916
1917 leaf = path->nodes[0];
1918 next_slot:
1919 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1920 ret = btrfs_next_leaf(root, path);
1921 if (ret > 0)
1922 break;
1923 if (ret < 0)
1924 goto error;
1925 leaf = path->nodes[0];
1926 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1927 btrfs_release_path(path);
1928 continue;
1929 }
1930
1931 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1932 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1933 key.type != BTRFS_DEV_ITEM_KEY)
1934 break;
1935
1936 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1937 struct btrfs_dev_item);
1938 devid = btrfs_device_id(leaf, dev_item);
1939 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
1940 BTRFS_UUID_SIZE);
1941 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
1942 BTRFS_UUID_SIZE);
1943 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1944 fs_uuid);
1945 BUG_ON(!device); /* Logic error */
1946
1947 if (device->fs_devices->seeding) {
1948 btrfs_set_device_generation(leaf, dev_item,
1949 device->generation);
1950 btrfs_mark_buffer_dirty(leaf);
1951 }
1952
1953 path->slots[0]++;
1954 goto next_slot;
1955 }
1956 ret = 0;
1957 error:
1958 btrfs_free_path(path);
1959 return ret;
1960 }
1961
1962 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1963 {
1964 struct request_queue *q;
1965 struct btrfs_trans_handle *trans;
1966 struct btrfs_device *device;
1967 struct block_device *bdev;
1968 struct list_head *devices;
1969 struct super_block *sb = root->fs_info->sb;
1970 struct rcu_string *name;
1971 u64 total_bytes;
1972 int seeding_dev = 0;
1973 int ret = 0;
1974
1975 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1976 return -EROFS;
1977
1978 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1979 root->fs_info->bdev_holder);
1980 if (IS_ERR(bdev))
1981 return PTR_ERR(bdev);
1982
1983 if (root->fs_info->fs_devices->seeding) {
1984 seeding_dev = 1;
1985 down_write(&sb->s_umount);
1986 mutex_lock(&uuid_mutex);
1987 }
1988
1989 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1990
1991 devices = &root->fs_info->fs_devices->devices;
1992
1993 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1994 list_for_each_entry(device, devices, dev_list) {
1995 if (device->bdev == bdev) {
1996 ret = -EEXIST;
1997 mutex_unlock(
1998 &root->fs_info->fs_devices->device_list_mutex);
1999 goto error;
2000 }
2001 }
2002 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2003
2004 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2005 if (IS_ERR(device)) {
2006 /* we can safely leave the fs_devices entry around */
2007 ret = PTR_ERR(device);
2008 goto error;
2009 }
2010
2011 name = rcu_string_strdup(device_path, GFP_NOFS);
2012 if (!name) {
2013 kfree(device);
2014 ret = -ENOMEM;
2015 goto error;
2016 }
2017 rcu_assign_pointer(device->name, name);
2018
2019 trans = btrfs_start_transaction(root, 0);
2020 if (IS_ERR(trans)) {
2021 rcu_string_free(device->name);
2022 kfree(device);
2023 ret = PTR_ERR(trans);
2024 goto error;
2025 }
2026
2027 lock_chunks(root);
2028
2029 q = bdev_get_queue(bdev);
2030 if (blk_queue_discard(q))
2031 device->can_discard = 1;
2032 device->writeable = 1;
2033 device->generation = trans->transid;
2034 device->io_width = root->sectorsize;
2035 device->io_align = root->sectorsize;
2036 device->sector_size = root->sectorsize;
2037 device->total_bytes = i_size_read(bdev->bd_inode);
2038 device->disk_total_bytes = device->total_bytes;
2039 device->dev_root = root->fs_info->dev_root;
2040 device->bdev = bdev;
2041 device->in_fs_metadata = 1;
2042 device->is_tgtdev_for_dev_replace = 0;
2043 device->mode = FMODE_EXCL;
2044 device->dev_stats_valid = 1;
2045 set_blocksize(device->bdev, 4096);
2046
2047 if (seeding_dev) {
2048 sb->s_flags &= ~MS_RDONLY;
2049 ret = btrfs_prepare_sprout(root);
2050 BUG_ON(ret); /* -ENOMEM */
2051 }
2052
2053 device->fs_devices = root->fs_info->fs_devices;
2054
2055 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2056 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2057 list_add(&device->dev_alloc_list,
2058 &root->fs_info->fs_devices->alloc_list);
2059 root->fs_info->fs_devices->num_devices++;
2060 root->fs_info->fs_devices->open_devices++;
2061 root->fs_info->fs_devices->rw_devices++;
2062 root->fs_info->fs_devices->total_devices++;
2063 if (device->can_discard)
2064 root->fs_info->fs_devices->num_can_discard++;
2065 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2066
2067 spin_lock(&root->fs_info->free_chunk_lock);
2068 root->fs_info->free_chunk_space += device->total_bytes;
2069 spin_unlock(&root->fs_info->free_chunk_lock);
2070
2071 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2072 root->fs_info->fs_devices->rotating = 1;
2073
2074 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2075 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2076 total_bytes + device->total_bytes);
2077
2078 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2079 btrfs_set_super_num_devices(root->fs_info->super_copy,
2080 total_bytes + 1);
2081 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2082
2083 if (seeding_dev) {
2084 ret = init_first_rw_device(trans, root, device);
2085 if (ret) {
2086 btrfs_abort_transaction(trans, root, ret);
2087 goto error_trans;
2088 }
2089 ret = btrfs_finish_sprout(trans, root);
2090 if (ret) {
2091 btrfs_abort_transaction(trans, root, ret);
2092 goto error_trans;
2093 }
2094 } else {
2095 ret = btrfs_add_device(trans, root, device);
2096 if (ret) {
2097 btrfs_abort_transaction(trans, root, ret);
2098 goto error_trans;
2099 }
2100 }
2101
2102 /*
2103 * we've got more storage, clear any full flags on the space
2104 * infos
2105 */
2106 btrfs_clear_space_info_full(root->fs_info);
2107
2108 unlock_chunks(root);
2109 root->fs_info->num_tolerated_disk_barrier_failures =
2110 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2111 ret = btrfs_commit_transaction(trans, root);
2112
2113 if (seeding_dev) {
2114 mutex_unlock(&uuid_mutex);
2115 up_write(&sb->s_umount);
2116
2117 if (ret) /* transaction commit */
2118 return ret;
2119
2120 ret = btrfs_relocate_sys_chunks(root);
2121 if (ret < 0)
2122 btrfs_error(root->fs_info, ret,
2123 "Failed to relocate sys chunks after "
2124 "device initialization. This can be fixed "
2125 "using the \"btrfs balance\" command.");
2126 trans = btrfs_attach_transaction(root);
2127 if (IS_ERR(trans)) {
2128 if (PTR_ERR(trans) == -ENOENT)
2129 return 0;
2130 return PTR_ERR(trans);
2131 }
2132 ret = btrfs_commit_transaction(trans, root);
2133 }
2134
2135 return ret;
2136
2137 error_trans:
2138 unlock_chunks(root);
2139 btrfs_end_transaction(trans, root);
2140 rcu_string_free(device->name);
2141 kfree(device);
2142 error:
2143 blkdev_put(bdev, FMODE_EXCL);
2144 if (seeding_dev) {
2145 mutex_unlock(&uuid_mutex);
2146 up_write(&sb->s_umount);
2147 }
2148 return ret;
2149 }
2150
2151 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2152 struct btrfs_device **device_out)
2153 {
2154 struct request_queue *q;
2155 struct btrfs_device *device;
2156 struct block_device *bdev;
2157 struct btrfs_fs_info *fs_info = root->fs_info;
2158 struct list_head *devices;
2159 struct rcu_string *name;
2160 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2161 int ret = 0;
2162
2163 *device_out = NULL;
2164 if (fs_info->fs_devices->seeding)
2165 return -EINVAL;
2166
2167 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2168 fs_info->bdev_holder);
2169 if (IS_ERR(bdev))
2170 return PTR_ERR(bdev);
2171
2172 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2173
2174 devices = &fs_info->fs_devices->devices;
2175 list_for_each_entry(device, devices, dev_list) {
2176 if (device->bdev == bdev) {
2177 ret = -EEXIST;
2178 goto error;
2179 }
2180 }
2181
2182 device = btrfs_alloc_device(NULL, &devid, NULL);
2183 if (IS_ERR(device)) {
2184 ret = PTR_ERR(device);
2185 goto error;
2186 }
2187
2188 name = rcu_string_strdup(device_path, GFP_NOFS);
2189 if (!name) {
2190 kfree(device);
2191 ret = -ENOMEM;
2192 goto error;
2193 }
2194 rcu_assign_pointer(device->name, name);
2195
2196 q = bdev_get_queue(bdev);
2197 if (blk_queue_discard(q))
2198 device->can_discard = 1;
2199 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2200 device->writeable = 1;
2201 device->generation = 0;
2202 device->io_width = root->sectorsize;
2203 device->io_align = root->sectorsize;
2204 device->sector_size = root->sectorsize;
2205 device->total_bytes = i_size_read(bdev->bd_inode);
2206 device->disk_total_bytes = device->total_bytes;
2207 device->dev_root = fs_info->dev_root;
2208 device->bdev = bdev;
2209 device->in_fs_metadata = 1;
2210 device->is_tgtdev_for_dev_replace = 1;
2211 device->mode = FMODE_EXCL;
2212 device->dev_stats_valid = 1;
2213 set_blocksize(device->bdev, 4096);
2214 device->fs_devices = fs_info->fs_devices;
2215 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2216 fs_info->fs_devices->num_devices++;
2217 fs_info->fs_devices->open_devices++;
2218 if (device->can_discard)
2219 fs_info->fs_devices->num_can_discard++;
2220 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2221
2222 *device_out = device;
2223 return ret;
2224
2225 error:
2226 blkdev_put(bdev, FMODE_EXCL);
2227 return ret;
2228 }
2229
2230 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2231 struct btrfs_device *tgtdev)
2232 {
2233 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2234 tgtdev->io_width = fs_info->dev_root->sectorsize;
2235 tgtdev->io_align = fs_info->dev_root->sectorsize;
2236 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2237 tgtdev->dev_root = fs_info->dev_root;
2238 tgtdev->in_fs_metadata = 1;
2239 }
2240
2241 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2242 struct btrfs_device *device)
2243 {
2244 int ret;
2245 struct btrfs_path *path;
2246 struct btrfs_root *root;
2247 struct btrfs_dev_item *dev_item;
2248 struct extent_buffer *leaf;
2249 struct btrfs_key key;
2250
2251 root = device->dev_root->fs_info->chunk_root;
2252
2253 path = btrfs_alloc_path();
2254 if (!path)
2255 return -ENOMEM;
2256
2257 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2258 key.type = BTRFS_DEV_ITEM_KEY;
2259 key.offset = device->devid;
2260
2261 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2262 if (ret < 0)
2263 goto out;
2264
2265 if (ret > 0) {
2266 ret = -ENOENT;
2267 goto out;
2268 }
2269
2270 leaf = path->nodes[0];
2271 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2272
2273 btrfs_set_device_id(leaf, dev_item, device->devid);
2274 btrfs_set_device_type(leaf, dev_item, device->type);
2275 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2276 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2277 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2278 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2279 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2280 btrfs_mark_buffer_dirty(leaf);
2281
2282 out:
2283 btrfs_free_path(path);
2284 return ret;
2285 }
2286
2287 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2288 struct btrfs_device *device, u64 new_size)
2289 {
2290 struct btrfs_super_block *super_copy =
2291 device->dev_root->fs_info->super_copy;
2292 u64 old_total = btrfs_super_total_bytes(super_copy);
2293 u64 diff = new_size - device->total_bytes;
2294
2295 if (!device->writeable)
2296 return -EACCES;
2297 if (new_size <= device->total_bytes ||
2298 device->is_tgtdev_for_dev_replace)
2299 return -EINVAL;
2300
2301 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2302 device->fs_devices->total_rw_bytes += diff;
2303
2304 device->total_bytes = new_size;
2305 device->disk_total_bytes = new_size;
2306 btrfs_clear_space_info_full(device->dev_root->fs_info);
2307
2308 return btrfs_update_device(trans, device);
2309 }
2310
2311 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2312 struct btrfs_device *device, u64 new_size)
2313 {
2314 int ret;
2315 lock_chunks(device->dev_root);
2316 ret = __btrfs_grow_device(trans, device, new_size);
2317 unlock_chunks(device->dev_root);
2318 return ret;
2319 }
2320
2321 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 u64 chunk_tree, u64 chunk_objectid,
2324 u64 chunk_offset)
2325 {
2326 int ret;
2327 struct btrfs_path *path;
2328 struct btrfs_key key;
2329
2330 root = root->fs_info->chunk_root;
2331 path = btrfs_alloc_path();
2332 if (!path)
2333 return -ENOMEM;
2334
2335 key.objectid = chunk_objectid;
2336 key.offset = chunk_offset;
2337 key.type = BTRFS_CHUNK_ITEM_KEY;
2338
2339 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2340 if (ret < 0)
2341 goto out;
2342 else if (ret > 0) { /* Logic error or corruption */
2343 btrfs_error(root->fs_info, -ENOENT,
2344 "Failed lookup while freeing chunk.");
2345 ret = -ENOENT;
2346 goto out;
2347 }
2348
2349 ret = btrfs_del_item(trans, root, path);
2350 if (ret < 0)
2351 btrfs_error(root->fs_info, ret,
2352 "Failed to delete chunk item.");
2353 out:
2354 btrfs_free_path(path);
2355 return ret;
2356 }
2357
2358 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2359 chunk_offset)
2360 {
2361 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2362 struct btrfs_disk_key *disk_key;
2363 struct btrfs_chunk *chunk;
2364 u8 *ptr;
2365 int ret = 0;
2366 u32 num_stripes;
2367 u32 array_size;
2368 u32 len = 0;
2369 u32 cur;
2370 struct btrfs_key key;
2371
2372 array_size = btrfs_super_sys_array_size(super_copy);
2373
2374 ptr = super_copy->sys_chunk_array;
2375 cur = 0;
2376
2377 while (cur < array_size) {
2378 disk_key = (struct btrfs_disk_key *)ptr;
2379 btrfs_disk_key_to_cpu(&key, disk_key);
2380
2381 len = sizeof(*disk_key);
2382
2383 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2384 chunk = (struct btrfs_chunk *)(ptr + len);
2385 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2386 len += btrfs_chunk_item_size(num_stripes);
2387 } else {
2388 ret = -EIO;
2389 break;
2390 }
2391 if (key.objectid == chunk_objectid &&
2392 key.offset == chunk_offset) {
2393 memmove(ptr, ptr + len, array_size - (cur + len));
2394 array_size -= len;
2395 btrfs_set_super_sys_array_size(super_copy, array_size);
2396 } else {
2397 ptr += len;
2398 cur += len;
2399 }
2400 }
2401 return ret;
2402 }
2403
2404 static int btrfs_relocate_chunk(struct btrfs_root *root,
2405 u64 chunk_tree, u64 chunk_objectid,
2406 u64 chunk_offset)
2407 {
2408 struct extent_map_tree *em_tree;
2409 struct btrfs_root *extent_root;
2410 struct btrfs_trans_handle *trans;
2411 struct extent_map *em;
2412 struct map_lookup *map;
2413 int ret;
2414 int i;
2415
2416 root = root->fs_info->chunk_root;
2417 extent_root = root->fs_info->extent_root;
2418 em_tree = &root->fs_info->mapping_tree.map_tree;
2419
2420 ret = btrfs_can_relocate(extent_root, chunk_offset);
2421 if (ret)
2422 return -ENOSPC;
2423
2424 /* step one, relocate all the extents inside this chunk */
2425 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2426 if (ret)
2427 return ret;
2428
2429 trans = btrfs_start_transaction(root, 0);
2430 if (IS_ERR(trans)) {
2431 ret = PTR_ERR(trans);
2432 btrfs_std_error(root->fs_info, ret);
2433 return ret;
2434 }
2435
2436 lock_chunks(root);
2437
2438 /*
2439 * step two, delete the device extents and the
2440 * chunk tree entries
2441 */
2442 read_lock(&em_tree->lock);
2443 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2444 read_unlock(&em_tree->lock);
2445
2446 BUG_ON(!em || em->start > chunk_offset ||
2447 em->start + em->len < chunk_offset);
2448 map = (struct map_lookup *)em->bdev;
2449
2450 for (i = 0; i < map->num_stripes; i++) {
2451 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2452 map->stripes[i].physical);
2453 BUG_ON(ret);
2454
2455 if (map->stripes[i].dev) {
2456 ret = btrfs_update_device(trans, map->stripes[i].dev);
2457 BUG_ON(ret);
2458 }
2459 }
2460 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2461 chunk_offset);
2462
2463 BUG_ON(ret);
2464
2465 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2466
2467 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2468 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2469 BUG_ON(ret);
2470 }
2471
2472 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2473 BUG_ON(ret);
2474
2475 write_lock(&em_tree->lock);
2476 remove_extent_mapping(em_tree, em);
2477 write_unlock(&em_tree->lock);
2478
2479 kfree(map);
2480 em->bdev = NULL;
2481
2482 /* once for the tree */
2483 free_extent_map(em);
2484 /* once for us */
2485 free_extent_map(em);
2486
2487 unlock_chunks(root);
2488 btrfs_end_transaction(trans, root);
2489 return 0;
2490 }
2491
2492 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2493 {
2494 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2495 struct btrfs_path *path;
2496 struct extent_buffer *leaf;
2497 struct btrfs_chunk *chunk;
2498 struct btrfs_key key;
2499 struct btrfs_key found_key;
2500 u64 chunk_tree = chunk_root->root_key.objectid;
2501 u64 chunk_type;
2502 bool retried = false;
2503 int failed = 0;
2504 int ret;
2505
2506 path = btrfs_alloc_path();
2507 if (!path)
2508 return -ENOMEM;
2509
2510 again:
2511 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2512 key.offset = (u64)-1;
2513 key.type = BTRFS_CHUNK_ITEM_KEY;
2514
2515 while (1) {
2516 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2517 if (ret < 0)
2518 goto error;
2519 BUG_ON(ret == 0); /* Corruption */
2520
2521 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2522 key.type);
2523 if (ret < 0)
2524 goto error;
2525 if (ret > 0)
2526 break;
2527
2528 leaf = path->nodes[0];
2529 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2530
2531 chunk = btrfs_item_ptr(leaf, path->slots[0],
2532 struct btrfs_chunk);
2533 chunk_type = btrfs_chunk_type(leaf, chunk);
2534 btrfs_release_path(path);
2535
2536 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2537 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2538 found_key.objectid,
2539 found_key.offset);
2540 if (ret == -ENOSPC)
2541 failed++;
2542 else if (ret)
2543 BUG();
2544 }
2545
2546 if (found_key.offset == 0)
2547 break;
2548 key.offset = found_key.offset - 1;
2549 }
2550 ret = 0;
2551 if (failed && !retried) {
2552 failed = 0;
2553 retried = true;
2554 goto again;
2555 } else if (WARN_ON(failed && retried)) {
2556 ret = -ENOSPC;
2557 }
2558 error:
2559 btrfs_free_path(path);
2560 return ret;
2561 }
2562
2563 static int insert_balance_item(struct btrfs_root *root,
2564 struct btrfs_balance_control *bctl)
2565 {
2566 struct btrfs_trans_handle *trans;
2567 struct btrfs_balance_item *item;
2568 struct btrfs_disk_balance_args disk_bargs;
2569 struct btrfs_path *path;
2570 struct extent_buffer *leaf;
2571 struct btrfs_key key;
2572 int ret, err;
2573
2574 path = btrfs_alloc_path();
2575 if (!path)
2576 return -ENOMEM;
2577
2578 trans = btrfs_start_transaction(root, 0);
2579 if (IS_ERR(trans)) {
2580 btrfs_free_path(path);
2581 return PTR_ERR(trans);
2582 }
2583
2584 key.objectid = BTRFS_BALANCE_OBJECTID;
2585 key.type = BTRFS_BALANCE_ITEM_KEY;
2586 key.offset = 0;
2587
2588 ret = btrfs_insert_empty_item(trans, root, path, &key,
2589 sizeof(*item));
2590 if (ret)
2591 goto out;
2592
2593 leaf = path->nodes[0];
2594 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2595
2596 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2597
2598 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2599 btrfs_set_balance_data(leaf, item, &disk_bargs);
2600 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2601 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2602 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2603 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2604
2605 btrfs_set_balance_flags(leaf, item, bctl->flags);
2606
2607 btrfs_mark_buffer_dirty(leaf);
2608 out:
2609 btrfs_free_path(path);
2610 err = btrfs_commit_transaction(trans, root);
2611 if (err && !ret)
2612 ret = err;
2613 return ret;
2614 }
2615
2616 static int del_balance_item(struct btrfs_root *root)
2617 {
2618 struct btrfs_trans_handle *trans;
2619 struct btrfs_path *path;
2620 struct btrfs_key key;
2621 int ret, err;
2622
2623 path = btrfs_alloc_path();
2624 if (!path)
2625 return -ENOMEM;
2626
2627 trans = btrfs_start_transaction(root, 0);
2628 if (IS_ERR(trans)) {
2629 btrfs_free_path(path);
2630 return PTR_ERR(trans);
2631 }
2632
2633 key.objectid = BTRFS_BALANCE_OBJECTID;
2634 key.type = BTRFS_BALANCE_ITEM_KEY;
2635 key.offset = 0;
2636
2637 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2638 if (ret < 0)
2639 goto out;
2640 if (ret > 0) {
2641 ret = -ENOENT;
2642 goto out;
2643 }
2644
2645 ret = btrfs_del_item(trans, root, path);
2646 out:
2647 btrfs_free_path(path);
2648 err = btrfs_commit_transaction(trans, root);
2649 if (err && !ret)
2650 ret = err;
2651 return ret;
2652 }
2653
2654 /*
2655 * This is a heuristic used to reduce the number of chunks balanced on
2656 * resume after balance was interrupted.
2657 */
2658 static void update_balance_args(struct btrfs_balance_control *bctl)
2659 {
2660 /*
2661 * Turn on soft mode for chunk types that were being converted.
2662 */
2663 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2664 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2665 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2666 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2667 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2668 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2669
2670 /*
2671 * Turn on usage filter if is not already used. The idea is
2672 * that chunks that we have already balanced should be
2673 * reasonably full. Don't do it for chunks that are being
2674 * converted - that will keep us from relocating unconverted
2675 * (albeit full) chunks.
2676 */
2677 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2678 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2679 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2680 bctl->data.usage = 90;
2681 }
2682 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2683 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2684 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2685 bctl->sys.usage = 90;
2686 }
2687 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2688 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2689 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2690 bctl->meta.usage = 90;
2691 }
2692 }
2693
2694 /*
2695 * Should be called with both balance and volume mutexes held to
2696 * serialize other volume operations (add_dev/rm_dev/resize) with
2697 * restriper. Same goes for unset_balance_control.
2698 */
2699 static void set_balance_control(struct btrfs_balance_control *bctl)
2700 {
2701 struct btrfs_fs_info *fs_info = bctl->fs_info;
2702
2703 BUG_ON(fs_info->balance_ctl);
2704
2705 spin_lock(&fs_info->balance_lock);
2706 fs_info->balance_ctl = bctl;
2707 spin_unlock(&fs_info->balance_lock);
2708 }
2709
2710 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2711 {
2712 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2713
2714 BUG_ON(!fs_info->balance_ctl);
2715
2716 spin_lock(&fs_info->balance_lock);
2717 fs_info->balance_ctl = NULL;
2718 spin_unlock(&fs_info->balance_lock);
2719
2720 kfree(bctl);
2721 }
2722
2723 /*
2724 * Balance filters. Return 1 if chunk should be filtered out
2725 * (should not be balanced).
2726 */
2727 static int chunk_profiles_filter(u64 chunk_type,
2728 struct btrfs_balance_args *bargs)
2729 {
2730 chunk_type = chunk_to_extended(chunk_type) &
2731 BTRFS_EXTENDED_PROFILE_MASK;
2732
2733 if (bargs->profiles & chunk_type)
2734 return 0;
2735
2736 return 1;
2737 }
2738
2739 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2740 struct btrfs_balance_args *bargs)
2741 {
2742 struct btrfs_block_group_cache *cache;
2743 u64 chunk_used, user_thresh;
2744 int ret = 1;
2745
2746 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2747 chunk_used = btrfs_block_group_used(&cache->item);
2748
2749 if (bargs->usage == 0)
2750 user_thresh = 1;
2751 else if (bargs->usage > 100)
2752 user_thresh = cache->key.offset;
2753 else
2754 user_thresh = div_factor_fine(cache->key.offset,
2755 bargs->usage);
2756
2757 if (chunk_used < user_thresh)
2758 ret = 0;
2759
2760 btrfs_put_block_group(cache);
2761 return ret;
2762 }
2763
2764 static int chunk_devid_filter(struct extent_buffer *leaf,
2765 struct btrfs_chunk *chunk,
2766 struct btrfs_balance_args *bargs)
2767 {
2768 struct btrfs_stripe *stripe;
2769 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2770 int i;
2771
2772 for (i = 0; i < num_stripes; i++) {
2773 stripe = btrfs_stripe_nr(chunk, i);
2774 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2775 return 0;
2776 }
2777
2778 return 1;
2779 }
2780
2781 /* [pstart, pend) */
2782 static int chunk_drange_filter(struct extent_buffer *leaf,
2783 struct btrfs_chunk *chunk,
2784 u64 chunk_offset,
2785 struct btrfs_balance_args *bargs)
2786 {
2787 struct btrfs_stripe *stripe;
2788 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2789 u64 stripe_offset;
2790 u64 stripe_length;
2791 int factor;
2792 int i;
2793
2794 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2795 return 0;
2796
2797 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2798 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2799 factor = num_stripes / 2;
2800 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2801 factor = num_stripes - 1;
2802 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2803 factor = num_stripes - 2;
2804 } else {
2805 factor = num_stripes;
2806 }
2807
2808 for (i = 0; i < num_stripes; i++) {
2809 stripe = btrfs_stripe_nr(chunk, i);
2810 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2811 continue;
2812
2813 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2814 stripe_length = btrfs_chunk_length(leaf, chunk);
2815 do_div(stripe_length, factor);
2816
2817 if (stripe_offset < bargs->pend &&
2818 stripe_offset + stripe_length > bargs->pstart)
2819 return 0;
2820 }
2821
2822 return 1;
2823 }
2824
2825 /* [vstart, vend) */
2826 static int chunk_vrange_filter(struct extent_buffer *leaf,
2827 struct btrfs_chunk *chunk,
2828 u64 chunk_offset,
2829 struct btrfs_balance_args *bargs)
2830 {
2831 if (chunk_offset < bargs->vend &&
2832 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2833 /* at least part of the chunk is inside this vrange */
2834 return 0;
2835
2836 return 1;
2837 }
2838
2839 static int chunk_soft_convert_filter(u64 chunk_type,
2840 struct btrfs_balance_args *bargs)
2841 {
2842 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2843 return 0;
2844
2845 chunk_type = chunk_to_extended(chunk_type) &
2846 BTRFS_EXTENDED_PROFILE_MASK;
2847
2848 if (bargs->target == chunk_type)
2849 return 1;
2850
2851 return 0;
2852 }
2853
2854 static int should_balance_chunk(struct btrfs_root *root,
2855 struct extent_buffer *leaf,
2856 struct btrfs_chunk *chunk, u64 chunk_offset)
2857 {
2858 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2859 struct btrfs_balance_args *bargs = NULL;
2860 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2861
2862 /* type filter */
2863 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2864 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2865 return 0;
2866 }
2867
2868 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2869 bargs = &bctl->data;
2870 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2871 bargs = &bctl->sys;
2872 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2873 bargs = &bctl->meta;
2874
2875 /* profiles filter */
2876 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2877 chunk_profiles_filter(chunk_type, bargs)) {
2878 return 0;
2879 }
2880
2881 /* usage filter */
2882 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2883 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2884 return 0;
2885 }
2886
2887 /* devid filter */
2888 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2889 chunk_devid_filter(leaf, chunk, bargs)) {
2890 return 0;
2891 }
2892
2893 /* drange filter, makes sense only with devid filter */
2894 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2895 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2896 return 0;
2897 }
2898
2899 /* vrange filter */
2900 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2901 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2902 return 0;
2903 }
2904
2905 /* soft profile changing mode */
2906 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2907 chunk_soft_convert_filter(chunk_type, bargs)) {
2908 return 0;
2909 }
2910
2911 return 1;
2912 }
2913
2914 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2915 {
2916 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2917 struct btrfs_root *chunk_root = fs_info->chunk_root;
2918 struct btrfs_root *dev_root = fs_info->dev_root;
2919 struct list_head *devices;
2920 struct btrfs_device *device;
2921 u64 old_size;
2922 u64 size_to_free;
2923 struct btrfs_chunk *chunk;
2924 struct btrfs_path *path;
2925 struct btrfs_key key;
2926 struct btrfs_key found_key;
2927 struct btrfs_trans_handle *trans;
2928 struct extent_buffer *leaf;
2929 int slot;
2930 int ret;
2931 int enospc_errors = 0;
2932 bool counting = true;
2933
2934 /* step one make some room on all the devices */
2935 devices = &fs_info->fs_devices->devices;
2936 list_for_each_entry(device, devices, dev_list) {
2937 old_size = device->total_bytes;
2938 size_to_free = div_factor(old_size, 1);
2939 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2940 if (!device->writeable ||
2941 device->total_bytes - device->bytes_used > size_to_free ||
2942 device->is_tgtdev_for_dev_replace)
2943 continue;
2944
2945 ret = btrfs_shrink_device(device, old_size - size_to_free);
2946 if (ret == -ENOSPC)
2947 break;
2948 BUG_ON(ret);
2949
2950 trans = btrfs_start_transaction(dev_root, 0);
2951 BUG_ON(IS_ERR(trans));
2952
2953 ret = btrfs_grow_device(trans, device, old_size);
2954 BUG_ON(ret);
2955
2956 btrfs_end_transaction(trans, dev_root);
2957 }
2958
2959 /* step two, relocate all the chunks */
2960 path = btrfs_alloc_path();
2961 if (!path) {
2962 ret = -ENOMEM;
2963 goto error;
2964 }
2965
2966 /* zero out stat counters */
2967 spin_lock(&fs_info->balance_lock);
2968 memset(&bctl->stat, 0, sizeof(bctl->stat));
2969 spin_unlock(&fs_info->balance_lock);
2970 again:
2971 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2972 key.offset = (u64)-1;
2973 key.type = BTRFS_CHUNK_ITEM_KEY;
2974
2975 while (1) {
2976 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2977 atomic_read(&fs_info->balance_cancel_req)) {
2978 ret = -ECANCELED;
2979 goto error;
2980 }
2981
2982 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2983 if (ret < 0)
2984 goto error;
2985
2986 /*
2987 * this shouldn't happen, it means the last relocate
2988 * failed
2989 */
2990 if (ret == 0)
2991 BUG(); /* FIXME break ? */
2992
2993 ret = btrfs_previous_item(chunk_root, path, 0,
2994 BTRFS_CHUNK_ITEM_KEY);
2995 if (ret) {
2996 ret = 0;
2997 break;
2998 }
2999
3000 leaf = path->nodes[0];
3001 slot = path->slots[0];
3002 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3003
3004 if (found_key.objectid != key.objectid)
3005 break;
3006
3007 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3008
3009 if (!counting) {
3010 spin_lock(&fs_info->balance_lock);
3011 bctl->stat.considered++;
3012 spin_unlock(&fs_info->balance_lock);
3013 }
3014
3015 ret = should_balance_chunk(chunk_root, leaf, chunk,
3016 found_key.offset);
3017 btrfs_release_path(path);
3018 if (!ret)
3019 goto loop;
3020
3021 if (counting) {
3022 spin_lock(&fs_info->balance_lock);
3023 bctl->stat.expected++;
3024 spin_unlock(&fs_info->balance_lock);
3025 goto loop;
3026 }
3027
3028 ret = btrfs_relocate_chunk(chunk_root,
3029 chunk_root->root_key.objectid,
3030 found_key.objectid,
3031 found_key.offset);
3032 if (ret && ret != -ENOSPC)
3033 goto error;
3034 if (ret == -ENOSPC) {
3035 enospc_errors++;
3036 } else {
3037 spin_lock(&fs_info->balance_lock);
3038 bctl->stat.completed++;
3039 spin_unlock(&fs_info->balance_lock);
3040 }
3041 loop:
3042 if (found_key.offset == 0)
3043 break;
3044 key.offset = found_key.offset - 1;
3045 }
3046
3047 if (counting) {
3048 btrfs_release_path(path);
3049 counting = false;
3050 goto again;
3051 }
3052 error:
3053 btrfs_free_path(path);
3054 if (enospc_errors) {
3055 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3056 enospc_errors);
3057 if (!ret)
3058 ret = -ENOSPC;
3059 }
3060
3061 return ret;
3062 }
3063
3064 /**
3065 * alloc_profile_is_valid - see if a given profile is valid and reduced
3066 * @flags: profile to validate
3067 * @extended: if true @flags is treated as an extended profile
3068 */
3069 static int alloc_profile_is_valid(u64 flags, int extended)
3070 {
3071 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3072 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3073
3074 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3075
3076 /* 1) check that all other bits are zeroed */
3077 if (flags & ~mask)
3078 return 0;
3079
3080 /* 2) see if profile is reduced */
3081 if (flags == 0)
3082 return !extended; /* "0" is valid for usual profiles */
3083
3084 /* true if exactly one bit set */
3085 return (flags & (flags - 1)) == 0;
3086 }
3087
3088 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3089 {
3090 /* cancel requested || normal exit path */
3091 return atomic_read(&fs_info->balance_cancel_req) ||
3092 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3093 atomic_read(&fs_info->balance_cancel_req) == 0);
3094 }
3095
3096 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3097 {
3098 int ret;
3099
3100 unset_balance_control(fs_info);
3101 ret = del_balance_item(fs_info->tree_root);
3102 if (ret)
3103 btrfs_std_error(fs_info, ret);
3104
3105 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3106 }
3107
3108 /*
3109 * Should be called with both balance and volume mutexes held
3110 */
3111 int btrfs_balance(struct btrfs_balance_control *bctl,
3112 struct btrfs_ioctl_balance_args *bargs)
3113 {
3114 struct btrfs_fs_info *fs_info = bctl->fs_info;
3115 u64 allowed;
3116 int mixed = 0;
3117 int ret;
3118 u64 num_devices;
3119 unsigned seq;
3120
3121 if (btrfs_fs_closing(fs_info) ||
3122 atomic_read(&fs_info->balance_pause_req) ||
3123 atomic_read(&fs_info->balance_cancel_req)) {
3124 ret = -EINVAL;
3125 goto out;
3126 }
3127
3128 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3129 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3130 mixed = 1;
3131
3132 /*
3133 * In case of mixed groups both data and meta should be picked,
3134 * and identical options should be given for both of them.
3135 */
3136 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3137 if (mixed && (bctl->flags & allowed)) {
3138 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3139 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3140 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3141 printk(KERN_ERR "btrfs: with mixed groups data and "
3142 "metadata balance options must be the same\n");
3143 ret = -EINVAL;
3144 goto out;
3145 }
3146 }
3147
3148 num_devices = fs_info->fs_devices->num_devices;
3149 btrfs_dev_replace_lock(&fs_info->dev_replace);
3150 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3151 BUG_ON(num_devices < 1);
3152 num_devices--;
3153 }
3154 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3155 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3156 if (num_devices == 1)
3157 allowed |= BTRFS_BLOCK_GROUP_DUP;
3158 else if (num_devices > 1)
3159 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3160 if (num_devices > 2)
3161 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3162 if (num_devices > 3)
3163 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3164 BTRFS_BLOCK_GROUP_RAID6);
3165 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3166 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3167 (bctl->data.target & ~allowed))) {
3168 printk(KERN_ERR "btrfs: unable to start balance with target "
3169 "data profile %llu\n",
3170 bctl->data.target);
3171 ret = -EINVAL;
3172 goto out;
3173 }
3174 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3175 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3176 (bctl->meta.target & ~allowed))) {
3177 printk(KERN_ERR "btrfs: unable to start balance with target "
3178 "metadata profile %llu\n",
3179 bctl->meta.target);
3180 ret = -EINVAL;
3181 goto out;
3182 }
3183 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3184 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3185 (bctl->sys.target & ~allowed))) {
3186 printk(KERN_ERR "btrfs: unable to start balance with target "
3187 "system profile %llu\n",
3188 bctl->sys.target);
3189 ret = -EINVAL;
3190 goto out;
3191 }
3192
3193 /* allow dup'ed data chunks only in mixed mode */
3194 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3195 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3196 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3197 ret = -EINVAL;
3198 goto out;
3199 }
3200
3201 /* allow to reduce meta or sys integrity only if force set */
3202 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3203 BTRFS_BLOCK_GROUP_RAID10 |
3204 BTRFS_BLOCK_GROUP_RAID5 |
3205 BTRFS_BLOCK_GROUP_RAID6;
3206 do {
3207 seq = read_seqbegin(&fs_info->profiles_lock);
3208
3209 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3210 (fs_info->avail_system_alloc_bits & allowed) &&
3211 !(bctl->sys.target & allowed)) ||
3212 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3213 (fs_info->avail_metadata_alloc_bits & allowed) &&
3214 !(bctl->meta.target & allowed))) {
3215 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3216 printk(KERN_INFO "btrfs: force reducing metadata "
3217 "integrity\n");
3218 } else {
3219 printk(KERN_ERR "btrfs: balance will reduce metadata "
3220 "integrity, use force if you want this\n");
3221 ret = -EINVAL;
3222 goto out;
3223 }
3224 }
3225 } while (read_seqretry(&fs_info->profiles_lock, seq));
3226
3227 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3228 int num_tolerated_disk_barrier_failures;
3229 u64 target = bctl->sys.target;
3230
3231 num_tolerated_disk_barrier_failures =
3232 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3233 if (num_tolerated_disk_barrier_failures > 0 &&
3234 (target &
3235 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3236 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3237 num_tolerated_disk_barrier_failures = 0;
3238 else if (num_tolerated_disk_barrier_failures > 1 &&
3239 (target &
3240 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3241 num_tolerated_disk_barrier_failures = 1;
3242
3243 fs_info->num_tolerated_disk_barrier_failures =
3244 num_tolerated_disk_barrier_failures;
3245 }
3246
3247 ret = insert_balance_item(fs_info->tree_root, bctl);
3248 if (ret && ret != -EEXIST)
3249 goto out;
3250
3251 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3252 BUG_ON(ret == -EEXIST);
3253 set_balance_control(bctl);
3254 } else {
3255 BUG_ON(ret != -EEXIST);
3256 spin_lock(&fs_info->balance_lock);
3257 update_balance_args(bctl);
3258 spin_unlock(&fs_info->balance_lock);
3259 }
3260
3261 atomic_inc(&fs_info->balance_running);
3262 mutex_unlock(&fs_info->balance_mutex);
3263
3264 ret = __btrfs_balance(fs_info);
3265
3266 mutex_lock(&fs_info->balance_mutex);
3267 atomic_dec(&fs_info->balance_running);
3268
3269 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3270 fs_info->num_tolerated_disk_barrier_failures =
3271 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3272 }
3273
3274 if (bargs) {
3275 memset(bargs, 0, sizeof(*bargs));
3276 update_ioctl_balance_args(fs_info, 0, bargs);
3277 }
3278
3279 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3280 balance_need_close(fs_info)) {
3281 __cancel_balance(fs_info);
3282 }
3283
3284 wake_up(&fs_info->balance_wait_q);
3285
3286 return ret;
3287 out:
3288 if (bctl->flags & BTRFS_BALANCE_RESUME)
3289 __cancel_balance(fs_info);
3290 else {
3291 kfree(bctl);
3292 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3293 }
3294 return ret;
3295 }
3296
3297 static int balance_kthread(void *data)
3298 {
3299 struct btrfs_fs_info *fs_info = data;
3300 int ret = 0;
3301
3302 mutex_lock(&fs_info->volume_mutex);
3303 mutex_lock(&fs_info->balance_mutex);
3304
3305 if (fs_info->balance_ctl) {
3306 printk(KERN_INFO "btrfs: continuing balance\n");
3307 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3308 }
3309
3310 mutex_unlock(&fs_info->balance_mutex);
3311 mutex_unlock(&fs_info->volume_mutex);
3312
3313 return ret;
3314 }
3315
3316 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3317 {
3318 struct task_struct *tsk;
3319
3320 spin_lock(&fs_info->balance_lock);
3321 if (!fs_info->balance_ctl) {
3322 spin_unlock(&fs_info->balance_lock);
3323 return 0;
3324 }
3325 spin_unlock(&fs_info->balance_lock);
3326
3327 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3328 printk(KERN_INFO "btrfs: force skipping balance\n");
3329 return 0;
3330 }
3331
3332 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3333 return PTR_ERR_OR_ZERO(tsk);
3334 }
3335
3336 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3337 {
3338 struct btrfs_balance_control *bctl;
3339 struct btrfs_balance_item *item;
3340 struct btrfs_disk_balance_args disk_bargs;
3341 struct btrfs_path *path;
3342 struct extent_buffer *leaf;
3343 struct btrfs_key key;
3344 int ret;
3345
3346 path = btrfs_alloc_path();
3347 if (!path)
3348 return -ENOMEM;
3349
3350 key.objectid = BTRFS_BALANCE_OBJECTID;
3351 key.type = BTRFS_BALANCE_ITEM_KEY;
3352 key.offset = 0;
3353
3354 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3355 if (ret < 0)
3356 goto out;
3357 if (ret > 0) { /* ret = -ENOENT; */
3358 ret = 0;
3359 goto out;
3360 }
3361
3362 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3363 if (!bctl) {
3364 ret = -ENOMEM;
3365 goto out;
3366 }
3367
3368 leaf = path->nodes[0];
3369 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3370
3371 bctl->fs_info = fs_info;
3372 bctl->flags = btrfs_balance_flags(leaf, item);
3373 bctl->flags |= BTRFS_BALANCE_RESUME;
3374
3375 btrfs_balance_data(leaf, item, &disk_bargs);
3376 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3377 btrfs_balance_meta(leaf, item, &disk_bargs);
3378 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3379 btrfs_balance_sys(leaf, item, &disk_bargs);
3380 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3381
3382 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3383
3384 mutex_lock(&fs_info->volume_mutex);
3385 mutex_lock(&fs_info->balance_mutex);
3386
3387 set_balance_control(bctl);
3388
3389 mutex_unlock(&fs_info->balance_mutex);
3390 mutex_unlock(&fs_info->volume_mutex);
3391 out:
3392 btrfs_free_path(path);
3393 return ret;
3394 }
3395
3396 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3397 {
3398 int ret = 0;
3399
3400 mutex_lock(&fs_info->balance_mutex);
3401 if (!fs_info->balance_ctl) {
3402 mutex_unlock(&fs_info->balance_mutex);
3403 return -ENOTCONN;
3404 }
3405
3406 if (atomic_read(&fs_info->balance_running)) {
3407 atomic_inc(&fs_info->balance_pause_req);
3408 mutex_unlock(&fs_info->balance_mutex);
3409
3410 wait_event(fs_info->balance_wait_q,
3411 atomic_read(&fs_info->balance_running) == 0);
3412
3413 mutex_lock(&fs_info->balance_mutex);
3414 /* we are good with balance_ctl ripped off from under us */
3415 BUG_ON(atomic_read(&fs_info->balance_running));
3416 atomic_dec(&fs_info->balance_pause_req);
3417 } else {
3418 ret = -ENOTCONN;
3419 }
3420
3421 mutex_unlock(&fs_info->balance_mutex);
3422 return ret;
3423 }
3424
3425 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3426 {
3427 if (fs_info->sb->s_flags & MS_RDONLY)
3428 return -EROFS;
3429
3430 mutex_lock(&fs_info->balance_mutex);
3431 if (!fs_info->balance_ctl) {
3432 mutex_unlock(&fs_info->balance_mutex);
3433 return -ENOTCONN;
3434 }
3435
3436 atomic_inc(&fs_info->balance_cancel_req);
3437 /*
3438 * if we are running just wait and return, balance item is
3439 * deleted in btrfs_balance in this case
3440 */
3441 if (atomic_read(&fs_info->balance_running)) {
3442 mutex_unlock(&fs_info->balance_mutex);
3443 wait_event(fs_info->balance_wait_q,
3444 atomic_read(&fs_info->balance_running) == 0);
3445 mutex_lock(&fs_info->balance_mutex);
3446 } else {
3447 /* __cancel_balance needs volume_mutex */
3448 mutex_unlock(&fs_info->balance_mutex);
3449 mutex_lock(&fs_info->volume_mutex);
3450 mutex_lock(&fs_info->balance_mutex);
3451
3452 if (fs_info->balance_ctl)
3453 __cancel_balance(fs_info);
3454
3455 mutex_unlock(&fs_info->volume_mutex);
3456 }
3457
3458 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3459 atomic_dec(&fs_info->balance_cancel_req);
3460 mutex_unlock(&fs_info->balance_mutex);
3461 return 0;
3462 }
3463
3464 static int btrfs_uuid_scan_kthread(void *data)
3465 {
3466 struct btrfs_fs_info *fs_info = data;
3467 struct btrfs_root *root = fs_info->tree_root;
3468 struct btrfs_key key;
3469 struct btrfs_key max_key;
3470 struct btrfs_path *path = NULL;
3471 int ret = 0;
3472 struct extent_buffer *eb;
3473 int slot;
3474 struct btrfs_root_item root_item;
3475 u32 item_size;
3476 struct btrfs_trans_handle *trans = NULL;
3477
3478 path = btrfs_alloc_path();
3479 if (!path) {
3480 ret = -ENOMEM;
3481 goto out;
3482 }
3483
3484 key.objectid = 0;
3485 key.type = BTRFS_ROOT_ITEM_KEY;
3486 key.offset = 0;
3487
3488 max_key.objectid = (u64)-1;
3489 max_key.type = BTRFS_ROOT_ITEM_KEY;
3490 max_key.offset = (u64)-1;
3491
3492 path->keep_locks = 1;
3493
3494 while (1) {
3495 ret = btrfs_search_forward(root, &key, path, 0);
3496 if (ret) {
3497 if (ret > 0)
3498 ret = 0;
3499 break;
3500 }
3501
3502 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3503 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3504 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3505 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3506 goto skip;
3507
3508 eb = path->nodes[0];
3509 slot = path->slots[0];
3510 item_size = btrfs_item_size_nr(eb, slot);
3511 if (item_size < sizeof(root_item))
3512 goto skip;
3513
3514 read_extent_buffer(eb, &root_item,
3515 btrfs_item_ptr_offset(eb, slot),
3516 (int)sizeof(root_item));
3517 if (btrfs_root_refs(&root_item) == 0)
3518 goto skip;
3519
3520 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3521 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3522 if (trans)
3523 goto update_tree;
3524
3525 btrfs_release_path(path);
3526 /*
3527 * 1 - subvol uuid item
3528 * 1 - received_subvol uuid item
3529 */
3530 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3531 if (IS_ERR(trans)) {
3532 ret = PTR_ERR(trans);
3533 break;
3534 }
3535 continue;
3536 } else {
3537 goto skip;
3538 }
3539 update_tree:
3540 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3541 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3542 root_item.uuid,
3543 BTRFS_UUID_KEY_SUBVOL,
3544 key.objectid);
3545 if (ret < 0) {
3546 pr_warn("btrfs: uuid_tree_add failed %d\n",
3547 ret);
3548 break;
3549 }
3550 }
3551
3552 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3553 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3554 root_item.received_uuid,
3555 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3556 key.objectid);
3557 if (ret < 0) {
3558 pr_warn("btrfs: uuid_tree_add failed %d\n",
3559 ret);
3560 break;
3561 }
3562 }
3563
3564 skip:
3565 if (trans) {
3566 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3567 trans = NULL;
3568 if (ret)
3569 break;
3570 }
3571
3572 btrfs_release_path(path);
3573 if (key.offset < (u64)-1) {
3574 key.offset++;
3575 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3576 key.offset = 0;
3577 key.type = BTRFS_ROOT_ITEM_KEY;
3578 } else if (key.objectid < (u64)-1) {
3579 key.offset = 0;
3580 key.type = BTRFS_ROOT_ITEM_KEY;
3581 key.objectid++;
3582 } else {
3583 break;
3584 }
3585 cond_resched();
3586 }
3587
3588 out:
3589 btrfs_free_path(path);
3590 if (trans && !IS_ERR(trans))
3591 btrfs_end_transaction(trans, fs_info->uuid_root);
3592 if (ret)
3593 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3594 else
3595 fs_info->update_uuid_tree_gen = 1;
3596 up(&fs_info->uuid_tree_rescan_sem);
3597 return 0;
3598 }
3599
3600 /*
3601 * Callback for btrfs_uuid_tree_iterate().
3602 * returns:
3603 * 0 check succeeded, the entry is not outdated.
3604 * < 0 if an error occured.
3605 * > 0 if the check failed, which means the caller shall remove the entry.
3606 */
3607 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3608 u8 *uuid, u8 type, u64 subid)
3609 {
3610 struct btrfs_key key;
3611 int ret = 0;
3612 struct btrfs_root *subvol_root;
3613
3614 if (type != BTRFS_UUID_KEY_SUBVOL &&
3615 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3616 goto out;
3617
3618 key.objectid = subid;
3619 key.type = BTRFS_ROOT_ITEM_KEY;
3620 key.offset = (u64)-1;
3621 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3622 if (IS_ERR(subvol_root)) {
3623 ret = PTR_ERR(subvol_root);
3624 if (ret == -ENOENT)
3625 ret = 1;
3626 goto out;
3627 }
3628
3629 switch (type) {
3630 case BTRFS_UUID_KEY_SUBVOL:
3631 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3632 ret = 1;
3633 break;
3634 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3635 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3636 BTRFS_UUID_SIZE))
3637 ret = 1;
3638 break;
3639 }
3640
3641 out:
3642 return ret;
3643 }
3644
3645 static int btrfs_uuid_rescan_kthread(void *data)
3646 {
3647 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3648 int ret;
3649
3650 /*
3651 * 1st step is to iterate through the existing UUID tree and
3652 * to delete all entries that contain outdated data.
3653 * 2nd step is to add all missing entries to the UUID tree.
3654 */
3655 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3656 if (ret < 0) {
3657 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3658 up(&fs_info->uuid_tree_rescan_sem);
3659 return ret;
3660 }
3661 return btrfs_uuid_scan_kthread(data);
3662 }
3663
3664 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3665 {
3666 struct btrfs_trans_handle *trans;
3667 struct btrfs_root *tree_root = fs_info->tree_root;
3668 struct btrfs_root *uuid_root;
3669 struct task_struct *task;
3670 int ret;
3671
3672 /*
3673 * 1 - root node
3674 * 1 - root item
3675 */
3676 trans = btrfs_start_transaction(tree_root, 2);
3677 if (IS_ERR(trans))
3678 return PTR_ERR(trans);
3679
3680 uuid_root = btrfs_create_tree(trans, fs_info,
3681 BTRFS_UUID_TREE_OBJECTID);
3682 if (IS_ERR(uuid_root)) {
3683 btrfs_abort_transaction(trans, tree_root,
3684 PTR_ERR(uuid_root));
3685 return PTR_ERR(uuid_root);
3686 }
3687
3688 fs_info->uuid_root = uuid_root;
3689
3690 ret = btrfs_commit_transaction(trans, tree_root);
3691 if (ret)
3692 return ret;
3693
3694 down(&fs_info->uuid_tree_rescan_sem);
3695 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3696 if (IS_ERR(task)) {
3697 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3698 pr_warn("btrfs: failed to start uuid_scan task\n");
3699 up(&fs_info->uuid_tree_rescan_sem);
3700 return PTR_ERR(task);
3701 }
3702
3703 return 0;
3704 }
3705
3706 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3707 {
3708 struct task_struct *task;
3709
3710 down(&fs_info->uuid_tree_rescan_sem);
3711 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3712 if (IS_ERR(task)) {
3713 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3714 pr_warn("btrfs: failed to start uuid_rescan task\n");
3715 up(&fs_info->uuid_tree_rescan_sem);
3716 return PTR_ERR(task);
3717 }
3718
3719 return 0;
3720 }
3721
3722 /*
3723 * shrinking a device means finding all of the device extents past
3724 * the new size, and then following the back refs to the chunks.
3725 * The chunk relocation code actually frees the device extent
3726 */
3727 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3728 {
3729 struct btrfs_trans_handle *trans;
3730 struct btrfs_root *root = device->dev_root;
3731 struct btrfs_dev_extent *dev_extent = NULL;
3732 struct btrfs_path *path;
3733 u64 length;
3734 u64 chunk_tree;
3735 u64 chunk_objectid;
3736 u64 chunk_offset;
3737 int ret;
3738 int slot;
3739 int failed = 0;
3740 bool retried = false;
3741 struct extent_buffer *l;
3742 struct btrfs_key key;
3743 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3744 u64 old_total = btrfs_super_total_bytes(super_copy);
3745 u64 old_size = device->total_bytes;
3746 u64 diff = device->total_bytes - new_size;
3747
3748 if (device->is_tgtdev_for_dev_replace)
3749 return -EINVAL;
3750
3751 path = btrfs_alloc_path();
3752 if (!path)
3753 return -ENOMEM;
3754
3755 path->reada = 2;
3756
3757 lock_chunks(root);
3758
3759 device->total_bytes = new_size;
3760 if (device->writeable) {
3761 device->fs_devices->total_rw_bytes -= diff;
3762 spin_lock(&root->fs_info->free_chunk_lock);
3763 root->fs_info->free_chunk_space -= diff;
3764 spin_unlock(&root->fs_info->free_chunk_lock);
3765 }
3766 unlock_chunks(root);
3767
3768 again:
3769 key.objectid = device->devid;
3770 key.offset = (u64)-1;
3771 key.type = BTRFS_DEV_EXTENT_KEY;
3772
3773 do {
3774 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3775 if (ret < 0)
3776 goto done;
3777
3778 ret = btrfs_previous_item(root, path, 0, key.type);
3779 if (ret < 0)
3780 goto done;
3781 if (ret) {
3782 ret = 0;
3783 btrfs_release_path(path);
3784 break;
3785 }
3786
3787 l = path->nodes[0];
3788 slot = path->slots[0];
3789 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3790
3791 if (key.objectid != device->devid) {
3792 btrfs_release_path(path);
3793 break;
3794 }
3795
3796 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3797 length = btrfs_dev_extent_length(l, dev_extent);
3798
3799 if (key.offset + length <= new_size) {
3800 btrfs_release_path(path);
3801 break;
3802 }
3803
3804 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3805 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3806 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3807 btrfs_release_path(path);
3808
3809 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3810 chunk_offset);
3811 if (ret && ret != -ENOSPC)
3812 goto done;
3813 if (ret == -ENOSPC)
3814 failed++;
3815 } while (key.offset-- > 0);
3816
3817 if (failed && !retried) {
3818 failed = 0;
3819 retried = true;
3820 goto again;
3821 } else if (failed && retried) {
3822 ret = -ENOSPC;
3823 lock_chunks(root);
3824
3825 device->total_bytes = old_size;
3826 if (device->writeable)
3827 device->fs_devices->total_rw_bytes += diff;
3828 spin_lock(&root->fs_info->free_chunk_lock);
3829 root->fs_info->free_chunk_space += diff;
3830 spin_unlock(&root->fs_info->free_chunk_lock);
3831 unlock_chunks(root);
3832 goto done;
3833 }
3834
3835 /* Shrinking succeeded, else we would be at "done". */
3836 trans = btrfs_start_transaction(root, 0);
3837 if (IS_ERR(trans)) {
3838 ret = PTR_ERR(trans);
3839 goto done;
3840 }
3841
3842 lock_chunks(root);
3843
3844 device->disk_total_bytes = new_size;
3845 /* Now btrfs_update_device() will change the on-disk size. */
3846 ret = btrfs_update_device(trans, device);
3847 if (ret) {
3848 unlock_chunks(root);
3849 btrfs_end_transaction(trans, root);
3850 goto done;
3851 }
3852 WARN_ON(diff > old_total);
3853 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3854 unlock_chunks(root);
3855 btrfs_end_transaction(trans, root);
3856 done:
3857 btrfs_free_path(path);
3858 return ret;
3859 }
3860
3861 static int btrfs_add_system_chunk(struct btrfs_root *root,
3862 struct btrfs_key *key,
3863 struct btrfs_chunk *chunk, int item_size)
3864 {
3865 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3866 struct btrfs_disk_key disk_key;
3867 u32 array_size;
3868 u8 *ptr;
3869
3870 array_size = btrfs_super_sys_array_size(super_copy);
3871 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3872 return -EFBIG;
3873
3874 ptr = super_copy->sys_chunk_array + array_size;
3875 btrfs_cpu_key_to_disk(&disk_key, key);
3876 memcpy(ptr, &disk_key, sizeof(disk_key));
3877 ptr += sizeof(disk_key);
3878 memcpy(ptr, chunk, item_size);
3879 item_size += sizeof(disk_key);
3880 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3881 return 0;
3882 }
3883
3884 /*
3885 * sort the devices in descending order by max_avail, total_avail
3886 */
3887 static int btrfs_cmp_device_info(const void *a, const void *b)
3888 {
3889 const struct btrfs_device_info *di_a = a;
3890 const struct btrfs_device_info *di_b = b;
3891
3892 if (di_a->max_avail > di_b->max_avail)
3893 return -1;
3894 if (di_a->max_avail < di_b->max_avail)
3895 return 1;
3896 if (di_a->total_avail > di_b->total_avail)
3897 return -1;
3898 if (di_a->total_avail < di_b->total_avail)
3899 return 1;
3900 return 0;
3901 }
3902
3903 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3904 [BTRFS_RAID_RAID10] = {
3905 .sub_stripes = 2,
3906 .dev_stripes = 1,
3907 .devs_max = 0, /* 0 == as many as possible */
3908 .devs_min = 4,
3909 .devs_increment = 2,
3910 .ncopies = 2,
3911 },
3912 [BTRFS_RAID_RAID1] = {
3913 .sub_stripes = 1,
3914 .dev_stripes = 1,
3915 .devs_max = 2,
3916 .devs_min = 2,
3917 .devs_increment = 2,
3918 .ncopies = 2,
3919 },
3920 [BTRFS_RAID_DUP] = {
3921 .sub_stripes = 1,
3922 .dev_stripes = 2,
3923 .devs_max = 1,
3924 .devs_min = 1,
3925 .devs_increment = 1,
3926 .ncopies = 2,
3927 },
3928 [BTRFS_RAID_RAID0] = {
3929 .sub_stripes = 1,
3930 .dev_stripes = 1,
3931 .devs_max = 0,
3932 .devs_min = 2,
3933 .devs_increment = 1,
3934 .ncopies = 1,
3935 },
3936 [BTRFS_RAID_SINGLE] = {
3937 .sub_stripes = 1,
3938 .dev_stripes = 1,
3939 .devs_max = 1,
3940 .devs_min = 1,
3941 .devs_increment = 1,
3942 .ncopies = 1,
3943 },
3944 [BTRFS_RAID_RAID5] = {
3945 .sub_stripes = 1,
3946 .dev_stripes = 1,
3947 .devs_max = 0,
3948 .devs_min = 2,
3949 .devs_increment = 1,
3950 .ncopies = 2,
3951 },
3952 [BTRFS_RAID_RAID6] = {
3953 .sub_stripes = 1,
3954 .dev_stripes = 1,
3955 .devs_max = 0,
3956 .devs_min = 3,
3957 .devs_increment = 1,
3958 .ncopies = 3,
3959 },
3960 };
3961
3962 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3963 {
3964 /* TODO allow them to set a preferred stripe size */
3965 return 64 * 1024;
3966 }
3967
3968 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3969 {
3970 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3971 return;
3972
3973 btrfs_set_fs_incompat(info, RAID56);
3974 }
3975
3976 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *extent_root, u64 start,
3978 u64 type)
3979 {
3980 struct btrfs_fs_info *info = extent_root->fs_info;
3981 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3982 struct list_head *cur;
3983 struct map_lookup *map = NULL;
3984 struct extent_map_tree *em_tree;
3985 struct extent_map *em;
3986 struct btrfs_device_info *devices_info = NULL;
3987 u64 total_avail;
3988 int num_stripes; /* total number of stripes to allocate */
3989 int data_stripes; /* number of stripes that count for
3990 block group size */
3991 int sub_stripes; /* sub_stripes info for map */
3992 int dev_stripes; /* stripes per dev */
3993 int devs_max; /* max devs to use */
3994 int devs_min; /* min devs needed */
3995 int devs_increment; /* ndevs has to be a multiple of this */
3996 int ncopies; /* how many copies to data has */
3997 int ret;
3998 u64 max_stripe_size;
3999 u64 max_chunk_size;
4000 u64 stripe_size;
4001 u64 num_bytes;
4002 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4003 int ndevs;
4004 int i;
4005 int j;
4006 int index;
4007
4008 BUG_ON(!alloc_profile_is_valid(type, 0));
4009
4010 if (list_empty(&fs_devices->alloc_list))
4011 return -ENOSPC;
4012
4013 index = __get_raid_index(type);
4014
4015 sub_stripes = btrfs_raid_array[index].sub_stripes;
4016 dev_stripes = btrfs_raid_array[index].dev_stripes;
4017 devs_max = btrfs_raid_array[index].devs_max;
4018 devs_min = btrfs_raid_array[index].devs_min;
4019 devs_increment = btrfs_raid_array[index].devs_increment;
4020 ncopies = btrfs_raid_array[index].ncopies;
4021
4022 if (type & BTRFS_BLOCK_GROUP_DATA) {
4023 max_stripe_size = 1024 * 1024 * 1024;
4024 max_chunk_size = 10 * max_stripe_size;
4025 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4026 /* for larger filesystems, use larger metadata chunks */
4027 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4028 max_stripe_size = 1024 * 1024 * 1024;
4029 else
4030 max_stripe_size = 256 * 1024 * 1024;
4031 max_chunk_size = max_stripe_size;
4032 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4033 max_stripe_size = 32 * 1024 * 1024;
4034 max_chunk_size = 2 * max_stripe_size;
4035 } else {
4036 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4037 type);
4038 BUG_ON(1);
4039 }
4040
4041 /* we don't want a chunk larger than 10% of writeable space */
4042 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4043 max_chunk_size);
4044
4045 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4046 GFP_NOFS);
4047 if (!devices_info)
4048 return -ENOMEM;
4049
4050 cur = fs_devices->alloc_list.next;
4051
4052 /*
4053 * in the first pass through the devices list, we gather information
4054 * about the available holes on each device.
4055 */
4056 ndevs = 0;
4057 while (cur != &fs_devices->alloc_list) {
4058 struct btrfs_device *device;
4059 u64 max_avail;
4060 u64 dev_offset;
4061
4062 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4063
4064 cur = cur->next;
4065
4066 if (!device->writeable) {
4067 WARN(1, KERN_ERR
4068 "btrfs: read-only device in alloc_list\n");
4069 continue;
4070 }
4071
4072 if (!device->in_fs_metadata ||
4073 device->is_tgtdev_for_dev_replace)
4074 continue;
4075
4076 if (device->total_bytes > device->bytes_used)
4077 total_avail = device->total_bytes - device->bytes_used;
4078 else
4079 total_avail = 0;
4080
4081 /* If there is no space on this device, skip it. */
4082 if (total_avail == 0)
4083 continue;
4084
4085 ret = find_free_dev_extent(trans, device,
4086 max_stripe_size * dev_stripes,
4087 &dev_offset, &max_avail);
4088 if (ret && ret != -ENOSPC)
4089 goto error;
4090
4091 if (ret == 0)
4092 max_avail = max_stripe_size * dev_stripes;
4093
4094 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4095 continue;
4096
4097 if (ndevs == fs_devices->rw_devices) {
4098 WARN(1, "%s: found more than %llu devices\n",
4099 __func__, fs_devices->rw_devices);
4100 break;
4101 }
4102 devices_info[ndevs].dev_offset = dev_offset;
4103 devices_info[ndevs].max_avail = max_avail;
4104 devices_info[ndevs].total_avail = total_avail;
4105 devices_info[ndevs].dev = device;
4106 ++ndevs;
4107 }
4108
4109 /*
4110 * now sort the devices by hole size / available space
4111 */
4112 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4113 btrfs_cmp_device_info, NULL);
4114
4115 /* round down to number of usable stripes */
4116 ndevs -= ndevs % devs_increment;
4117
4118 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4119 ret = -ENOSPC;
4120 goto error;
4121 }
4122
4123 if (devs_max && ndevs > devs_max)
4124 ndevs = devs_max;
4125 /*
4126 * the primary goal is to maximize the number of stripes, so use as many
4127 * devices as possible, even if the stripes are not maximum sized.
4128 */
4129 stripe_size = devices_info[ndevs-1].max_avail;
4130 num_stripes = ndevs * dev_stripes;
4131
4132 /*
4133 * this will have to be fixed for RAID1 and RAID10 over
4134 * more drives
4135 */
4136 data_stripes = num_stripes / ncopies;
4137
4138 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4139 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4140 btrfs_super_stripesize(info->super_copy));
4141 data_stripes = num_stripes - 1;
4142 }
4143 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4144 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4145 btrfs_super_stripesize(info->super_copy));
4146 data_stripes = num_stripes - 2;
4147 }
4148
4149 /*
4150 * Use the number of data stripes to figure out how big this chunk
4151 * is really going to be in terms of logical address space,
4152 * and compare that answer with the max chunk size
4153 */
4154 if (stripe_size * data_stripes > max_chunk_size) {
4155 u64 mask = (1ULL << 24) - 1;
4156 stripe_size = max_chunk_size;
4157 do_div(stripe_size, data_stripes);
4158
4159 /* bump the answer up to a 16MB boundary */
4160 stripe_size = (stripe_size + mask) & ~mask;
4161
4162 /* but don't go higher than the limits we found
4163 * while searching for free extents
4164 */
4165 if (stripe_size > devices_info[ndevs-1].max_avail)
4166 stripe_size = devices_info[ndevs-1].max_avail;
4167 }
4168
4169 do_div(stripe_size, dev_stripes);
4170
4171 /* align to BTRFS_STRIPE_LEN */
4172 do_div(stripe_size, raid_stripe_len);
4173 stripe_size *= raid_stripe_len;
4174
4175 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4176 if (!map) {
4177 ret = -ENOMEM;
4178 goto error;
4179 }
4180 map->num_stripes = num_stripes;
4181
4182 for (i = 0; i < ndevs; ++i) {
4183 for (j = 0; j < dev_stripes; ++j) {
4184 int s = i * dev_stripes + j;
4185 map->stripes[s].dev = devices_info[i].dev;
4186 map->stripes[s].physical = devices_info[i].dev_offset +
4187 j * stripe_size;
4188 }
4189 }
4190 map->sector_size = extent_root->sectorsize;
4191 map->stripe_len = raid_stripe_len;
4192 map->io_align = raid_stripe_len;
4193 map->io_width = raid_stripe_len;
4194 map->type = type;
4195 map->sub_stripes = sub_stripes;
4196
4197 num_bytes = stripe_size * data_stripes;
4198
4199 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4200
4201 em = alloc_extent_map();
4202 if (!em) {
4203 ret = -ENOMEM;
4204 goto error;
4205 }
4206 em->bdev = (struct block_device *)map;
4207 em->start = start;
4208 em->len = num_bytes;
4209 em->block_start = 0;
4210 em->block_len = em->len;
4211 em->orig_block_len = stripe_size;
4212
4213 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4214 write_lock(&em_tree->lock);
4215 ret = add_extent_mapping(em_tree, em, 0);
4216 if (!ret) {
4217 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4218 atomic_inc(&em->refs);
4219 }
4220 write_unlock(&em_tree->lock);
4221 if (ret) {
4222 free_extent_map(em);
4223 goto error;
4224 }
4225
4226 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4227 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4228 start, num_bytes);
4229 if (ret)
4230 goto error_del_extent;
4231
4232 free_extent_map(em);
4233 check_raid56_incompat_flag(extent_root->fs_info, type);
4234
4235 kfree(devices_info);
4236 return 0;
4237
4238 error_del_extent:
4239 write_lock(&em_tree->lock);
4240 remove_extent_mapping(em_tree, em);
4241 write_unlock(&em_tree->lock);
4242
4243 /* One for our allocation */
4244 free_extent_map(em);
4245 /* One for the tree reference */
4246 free_extent_map(em);
4247 error:
4248 kfree(map);
4249 kfree(devices_info);
4250 return ret;
4251 }
4252
4253 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4254 struct btrfs_root *extent_root,
4255 u64 chunk_offset, u64 chunk_size)
4256 {
4257 struct btrfs_key key;
4258 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4259 struct btrfs_device *device;
4260 struct btrfs_chunk *chunk;
4261 struct btrfs_stripe *stripe;
4262 struct extent_map_tree *em_tree;
4263 struct extent_map *em;
4264 struct map_lookup *map;
4265 size_t item_size;
4266 u64 dev_offset;
4267 u64 stripe_size;
4268 int i = 0;
4269 int ret;
4270
4271 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4272 read_lock(&em_tree->lock);
4273 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4274 read_unlock(&em_tree->lock);
4275
4276 if (!em) {
4277 btrfs_crit(extent_root->fs_info, "unable to find logical "
4278 "%Lu len %Lu", chunk_offset, chunk_size);
4279 return -EINVAL;
4280 }
4281
4282 if (em->start != chunk_offset || em->len != chunk_size) {
4283 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4284 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4285 chunk_size, em->start, em->len);
4286 free_extent_map(em);
4287 return -EINVAL;
4288 }
4289
4290 map = (struct map_lookup *)em->bdev;
4291 item_size = btrfs_chunk_item_size(map->num_stripes);
4292 stripe_size = em->orig_block_len;
4293
4294 chunk = kzalloc(item_size, GFP_NOFS);
4295 if (!chunk) {
4296 ret = -ENOMEM;
4297 goto out;
4298 }
4299
4300 for (i = 0; i < map->num_stripes; i++) {
4301 device = map->stripes[i].dev;
4302 dev_offset = map->stripes[i].physical;
4303
4304 device->bytes_used += stripe_size;
4305 ret = btrfs_update_device(trans, device);
4306 if (ret)
4307 goto out;
4308 ret = btrfs_alloc_dev_extent(trans, device,
4309 chunk_root->root_key.objectid,
4310 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4311 chunk_offset, dev_offset,
4312 stripe_size);
4313 if (ret)
4314 goto out;
4315 }
4316
4317 spin_lock(&extent_root->fs_info->free_chunk_lock);
4318 extent_root->fs_info->free_chunk_space -= (stripe_size *
4319 map->num_stripes);
4320 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4321
4322 stripe = &chunk->stripe;
4323 for (i = 0; i < map->num_stripes; i++) {
4324 device = map->stripes[i].dev;
4325 dev_offset = map->stripes[i].physical;
4326
4327 btrfs_set_stack_stripe_devid(stripe, device->devid);
4328 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4329 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4330 stripe++;
4331 }
4332
4333 btrfs_set_stack_chunk_length(chunk, chunk_size);
4334 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4335 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4336 btrfs_set_stack_chunk_type(chunk, map->type);
4337 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4338 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4339 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4340 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4341 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4342
4343 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4344 key.type = BTRFS_CHUNK_ITEM_KEY;
4345 key.offset = chunk_offset;
4346
4347 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4348 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4349 /*
4350 * TODO: Cleanup of inserted chunk root in case of
4351 * failure.
4352 */
4353 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4354 item_size);
4355 }
4356
4357 out:
4358 kfree(chunk);
4359 free_extent_map(em);
4360 return ret;
4361 }
4362
4363 /*
4364 * Chunk allocation falls into two parts. The first part does works
4365 * that make the new allocated chunk useable, but not do any operation
4366 * that modifies the chunk tree. The second part does the works that
4367 * require modifying the chunk tree. This division is important for the
4368 * bootstrap process of adding storage to a seed btrfs.
4369 */
4370 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4371 struct btrfs_root *extent_root, u64 type)
4372 {
4373 u64 chunk_offset;
4374
4375 chunk_offset = find_next_chunk(extent_root->fs_info);
4376 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4377 }
4378
4379 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4380 struct btrfs_root *root,
4381 struct btrfs_device *device)
4382 {
4383 u64 chunk_offset;
4384 u64 sys_chunk_offset;
4385 u64 alloc_profile;
4386 struct btrfs_fs_info *fs_info = root->fs_info;
4387 struct btrfs_root *extent_root = fs_info->extent_root;
4388 int ret;
4389
4390 chunk_offset = find_next_chunk(fs_info);
4391 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4392 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4393 alloc_profile);
4394 if (ret)
4395 return ret;
4396
4397 sys_chunk_offset = find_next_chunk(root->fs_info);
4398 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4399 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4400 alloc_profile);
4401 if (ret) {
4402 btrfs_abort_transaction(trans, root, ret);
4403 goto out;
4404 }
4405
4406 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4407 if (ret)
4408 btrfs_abort_transaction(trans, root, ret);
4409 out:
4410 return ret;
4411 }
4412
4413 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4414 {
4415 struct extent_map *em;
4416 struct map_lookup *map;
4417 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4418 int readonly = 0;
4419 int i;
4420
4421 read_lock(&map_tree->map_tree.lock);
4422 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4423 read_unlock(&map_tree->map_tree.lock);
4424 if (!em)
4425 return 1;
4426
4427 if (btrfs_test_opt(root, DEGRADED)) {
4428 free_extent_map(em);
4429 return 0;
4430 }
4431
4432 map = (struct map_lookup *)em->bdev;
4433 for (i = 0; i < map->num_stripes; i++) {
4434 if (!map->stripes[i].dev->writeable) {
4435 readonly = 1;
4436 break;
4437 }
4438 }
4439 free_extent_map(em);
4440 return readonly;
4441 }
4442
4443 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4444 {
4445 extent_map_tree_init(&tree->map_tree);
4446 }
4447
4448 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4449 {
4450 struct extent_map *em;
4451
4452 while (1) {
4453 write_lock(&tree->map_tree.lock);
4454 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4455 if (em)
4456 remove_extent_mapping(&tree->map_tree, em);
4457 write_unlock(&tree->map_tree.lock);
4458 if (!em)
4459 break;
4460 kfree(em->bdev);
4461 /* once for us */
4462 free_extent_map(em);
4463 /* once for the tree */
4464 free_extent_map(em);
4465 }
4466 }
4467
4468 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4469 {
4470 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4471 struct extent_map *em;
4472 struct map_lookup *map;
4473 struct extent_map_tree *em_tree = &map_tree->map_tree;
4474 int ret;
4475
4476 read_lock(&em_tree->lock);
4477 em = lookup_extent_mapping(em_tree, logical, len);
4478 read_unlock(&em_tree->lock);
4479
4480 /*
4481 * We could return errors for these cases, but that could get ugly and
4482 * we'd probably do the same thing which is just not do anything else
4483 * and exit, so return 1 so the callers don't try to use other copies.
4484 */
4485 if (!em) {
4486 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4487 logical+len);
4488 return 1;
4489 }
4490
4491 if (em->start > logical || em->start + em->len < logical) {
4492 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4493 "%Lu-%Lu\n", logical, logical+len, em->start,
4494 em->start + em->len);
4495 free_extent_map(em);
4496 return 1;
4497 }
4498
4499 map = (struct map_lookup *)em->bdev;
4500 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4501 ret = map->num_stripes;
4502 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4503 ret = map->sub_stripes;
4504 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4505 ret = 2;
4506 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4507 ret = 3;
4508 else
4509 ret = 1;
4510 free_extent_map(em);
4511
4512 btrfs_dev_replace_lock(&fs_info->dev_replace);
4513 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4514 ret++;
4515 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4516
4517 return ret;
4518 }
4519
4520 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4521 struct btrfs_mapping_tree *map_tree,
4522 u64 logical)
4523 {
4524 struct extent_map *em;
4525 struct map_lookup *map;
4526 struct extent_map_tree *em_tree = &map_tree->map_tree;
4527 unsigned long len = root->sectorsize;
4528
4529 read_lock(&em_tree->lock);
4530 em = lookup_extent_mapping(em_tree, logical, len);
4531 read_unlock(&em_tree->lock);
4532 BUG_ON(!em);
4533
4534 BUG_ON(em->start > logical || em->start + em->len < logical);
4535 map = (struct map_lookup *)em->bdev;
4536 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4537 BTRFS_BLOCK_GROUP_RAID6)) {
4538 len = map->stripe_len * nr_data_stripes(map);
4539 }
4540 free_extent_map(em);
4541 return len;
4542 }
4543
4544 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4545 u64 logical, u64 len, int mirror_num)
4546 {
4547 struct extent_map *em;
4548 struct map_lookup *map;
4549 struct extent_map_tree *em_tree = &map_tree->map_tree;
4550 int ret = 0;
4551
4552 read_lock(&em_tree->lock);
4553 em = lookup_extent_mapping(em_tree, logical, len);
4554 read_unlock(&em_tree->lock);
4555 BUG_ON(!em);
4556
4557 BUG_ON(em->start > logical || em->start + em->len < logical);
4558 map = (struct map_lookup *)em->bdev;
4559 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4560 BTRFS_BLOCK_GROUP_RAID6))
4561 ret = 1;
4562 free_extent_map(em);
4563 return ret;
4564 }
4565
4566 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4567 struct map_lookup *map, int first, int num,
4568 int optimal, int dev_replace_is_ongoing)
4569 {
4570 int i;
4571 int tolerance;
4572 struct btrfs_device *srcdev;
4573
4574 if (dev_replace_is_ongoing &&
4575 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4576 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4577 srcdev = fs_info->dev_replace.srcdev;
4578 else
4579 srcdev = NULL;
4580
4581 /*
4582 * try to avoid the drive that is the source drive for a
4583 * dev-replace procedure, only choose it if no other non-missing
4584 * mirror is available
4585 */
4586 for (tolerance = 0; tolerance < 2; tolerance++) {
4587 if (map->stripes[optimal].dev->bdev &&
4588 (tolerance || map->stripes[optimal].dev != srcdev))
4589 return optimal;
4590 for (i = first; i < first + num; i++) {
4591 if (map->stripes[i].dev->bdev &&
4592 (tolerance || map->stripes[i].dev != srcdev))
4593 return i;
4594 }
4595 }
4596
4597 /* we couldn't find one that doesn't fail. Just return something
4598 * and the io error handling code will clean up eventually
4599 */
4600 return optimal;
4601 }
4602
4603 static inline int parity_smaller(u64 a, u64 b)
4604 {
4605 return a > b;
4606 }
4607
4608 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4609 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4610 {
4611 struct btrfs_bio_stripe s;
4612 int i;
4613 u64 l;
4614 int again = 1;
4615
4616 while (again) {
4617 again = 0;
4618 for (i = 0; i < bbio->num_stripes - 1; i++) {
4619 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4620 s = bbio->stripes[i];
4621 l = raid_map[i];
4622 bbio->stripes[i] = bbio->stripes[i+1];
4623 raid_map[i] = raid_map[i+1];
4624 bbio->stripes[i+1] = s;
4625 raid_map[i+1] = l;
4626 again = 1;
4627 }
4628 }
4629 }
4630 }
4631
4632 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4633 u64 logical, u64 *length,
4634 struct btrfs_bio **bbio_ret,
4635 int mirror_num, u64 **raid_map_ret)
4636 {
4637 struct extent_map *em;
4638 struct map_lookup *map;
4639 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4640 struct extent_map_tree *em_tree = &map_tree->map_tree;
4641 u64 offset;
4642 u64 stripe_offset;
4643 u64 stripe_end_offset;
4644 u64 stripe_nr;
4645 u64 stripe_nr_orig;
4646 u64 stripe_nr_end;
4647 u64 stripe_len;
4648 u64 *raid_map = NULL;
4649 int stripe_index;
4650 int i;
4651 int ret = 0;
4652 int num_stripes;
4653 int max_errors = 0;
4654 struct btrfs_bio *bbio = NULL;
4655 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4656 int dev_replace_is_ongoing = 0;
4657 int num_alloc_stripes;
4658 int patch_the_first_stripe_for_dev_replace = 0;
4659 u64 physical_to_patch_in_first_stripe = 0;
4660 u64 raid56_full_stripe_start = (u64)-1;
4661
4662 read_lock(&em_tree->lock);
4663 em = lookup_extent_mapping(em_tree, logical, *length);
4664 read_unlock(&em_tree->lock);
4665
4666 if (!em) {
4667 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4668 logical, *length);
4669 return -EINVAL;
4670 }
4671
4672 if (em->start > logical || em->start + em->len < logical) {
4673 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4674 "found %Lu-%Lu\n", logical, em->start,
4675 em->start + em->len);
4676 free_extent_map(em);
4677 return -EINVAL;
4678 }
4679
4680 map = (struct map_lookup *)em->bdev;
4681 offset = logical - em->start;
4682
4683 stripe_len = map->stripe_len;
4684 stripe_nr = offset;
4685 /*
4686 * stripe_nr counts the total number of stripes we have to stride
4687 * to get to this block
4688 */
4689 do_div(stripe_nr, stripe_len);
4690
4691 stripe_offset = stripe_nr * stripe_len;
4692 BUG_ON(offset < stripe_offset);
4693
4694 /* stripe_offset is the offset of this block in its stripe*/
4695 stripe_offset = offset - stripe_offset;
4696
4697 /* if we're here for raid56, we need to know the stripe aligned start */
4698 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4699 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4700 raid56_full_stripe_start = offset;
4701
4702 /* allow a write of a full stripe, but make sure we don't
4703 * allow straddling of stripes
4704 */
4705 do_div(raid56_full_stripe_start, full_stripe_len);
4706 raid56_full_stripe_start *= full_stripe_len;
4707 }
4708
4709 if (rw & REQ_DISCARD) {
4710 /* we don't discard raid56 yet */
4711 if (map->type &
4712 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4713 ret = -EOPNOTSUPP;
4714 goto out;
4715 }
4716 *length = min_t(u64, em->len - offset, *length);
4717 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4718 u64 max_len;
4719 /* For writes to RAID[56], allow a full stripeset across all disks.
4720 For other RAID types and for RAID[56] reads, just allow a single
4721 stripe (on a single disk). */
4722 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4723 (rw & REQ_WRITE)) {
4724 max_len = stripe_len * nr_data_stripes(map) -
4725 (offset - raid56_full_stripe_start);
4726 } else {
4727 /* we limit the length of each bio to what fits in a stripe */
4728 max_len = stripe_len - stripe_offset;
4729 }
4730 *length = min_t(u64, em->len - offset, max_len);
4731 } else {
4732 *length = em->len - offset;
4733 }
4734
4735 /* This is for when we're called from btrfs_merge_bio_hook() and all
4736 it cares about is the length */
4737 if (!bbio_ret)
4738 goto out;
4739
4740 btrfs_dev_replace_lock(dev_replace);
4741 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4742 if (!dev_replace_is_ongoing)
4743 btrfs_dev_replace_unlock(dev_replace);
4744
4745 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4746 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4747 dev_replace->tgtdev != NULL) {
4748 /*
4749 * in dev-replace case, for repair case (that's the only
4750 * case where the mirror is selected explicitly when
4751 * calling btrfs_map_block), blocks left of the left cursor
4752 * can also be read from the target drive.
4753 * For REQ_GET_READ_MIRRORS, the target drive is added as
4754 * the last one to the array of stripes. For READ, it also
4755 * needs to be supported using the same mirror number.
4756 * If the requested block is not left of the left cursor,
4757 * EIO is returned. This can happen because btrfs_num_copies()
4758 * returns one more in the dev-replace case.
4759 */
4760 u64 tmp_length = *length;
4761 struct btrfs_bio *tmp_bbio = NULL;
4762 int tmp_num_stripes;
4763 u64 srcdev_devid = dev_replace->srcdev->devid;
4764 int index_srcdev = 0;
4765 int found = 0;
4766 u64 physical_of_found = 0;
4767
4768 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4769 logical, &tmp_length, &tmp_bbio, 0, NULL);
4770 if (ret) {
4771 WARN_ON(tmp_bbio != NULL);
4772 goto out;
4773 }
4774
4775 tmp_num_stripes = tmp_bbio->num_stripes;
4776 if (mirror_num > tmp_num_stripes) {
4777 /*
4778 * REQ_GET_READ_MIRRORS does not contain this
4779 * mirror, that means that the requested area
4780 * is not left of the left cursor
4781 */
4782 ret = -EIO;
4783 kfree(tmp_bbio);
4784 goto out;
4785 }
4786
4787 /*
4788 * process the rest of the function using the mirror_num
4789 * of the source drive. Therefore look it up first.
4790 * At the end, patch the device pointer to the one of the
4791 * target drive.
4792 */
4793 for (i = 0; i < tmp_num_stripes; i++) {
4794 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4795 /*
4796 * In case of DUP, in order to keep it
4797 * simple, only add the mirror with the
4798 * lowest physical address
4799 */
4800 if (found &&
4801 physical_of_found <=
4802 tmp_bbio->stripes[i].physical)
4803 continue;
4804 index_srcdev = i;
4805 found = 1;
4806 physical_of_found =
4807 tmp_bbio->stripes[i].physical;
4808 }
4809 }
4810
4811 if (found) {
4812 mirror_num = index_srcdev + 1;
4813 patch_the_first_stripe_for_dev_replace = 1;
4814 physical_to_patch_in_first_stripe = physical_of_found;
4815 } else {
4816 WARN_ON(1);
4817 ret = -EIO;
4818 kfree(tmp_bbio);
4819 goto out;
4820 }
4821
4822 kfree(tmp_bbio);
4823 } else if (mirror_num > map->num_stripes) {
4824 mirror_num = 0;
4825 }
4826
4827 num_stripes = 1;
4828 stripe_index = 0;
4829 stripe_nr_orig = stripe_nr;
4830 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4831 do_div(stripe_nr_end, map->stripe_len);
4832 stripe_end_offset = stripe_nr_end * map->stripe_len -
4833 (offset + *length);
4834
4835 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4836 if (rw & REQ_DISCARD)
4837 num_stripes = min_t(u64, map->num_stripes,
4838 stripe_nr_end - stripe_nr_orig);
4839 stripe_index = do_div(stripe_nr, map->num_stripes);
4840 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4841 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4842 num_stripes = map->num_stripes;
4843 else if (mirror_num)
4844 stripe_index = mirror_num - 1;
4845 else {
4846 stripe_index = find_live_mirror(fs_info, map, 0,
4847 map->num_stripes,
4848 current->pid % map->num_stripes,
4849 dev_replace_is_ongoing);
4850 mirror_num = stripe_index + 1;
4851 }
4852
4853 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4854 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4855 num_stripes = map->num_stripes;
4856 } else if (mirror_num) {
4857 stripe_index = mirror_num - 1;
4858 } else {
4859 mirror_num = 1;
4860 }
4861
4862 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4863 int factor = map->num_stripes / map->sub_stripes;
4864
4865 stripe_index = do_div(stripe_nr, factor);
4866 stripe_index *= map->sub_stripes;
4867
4868 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4869 num_stripes = map->sub_stripes;
4870 else if (rw & REQ_DISCARD)
4871 num_stripes = min_t(u64, map->sub_stripes *
4872 (stripe_nr_end - stripe_nr_orig),
4873 map->num_stripes);
4874 else if (mirror_num)
4875 stripe_index += mirror_num - 1;
4876 else {
4877 int old_stripe_index = stripe_index;
4878 stripe_index = find_live_mirror(fs_info, map,
4879 stripe_index,
4880 map->sub_stripes, stripe_index +
4881 current->pid % map->sub_stripes,
4882 dev_replace_is_ongoing);
4883 mirror_num = stripe_index - old_stripe_index + 1;
4884 }
4885
4886 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4887 BTRFS_BLOCK_GROUP_RAID6)) {
4888 u64 tmp;
4889
4890 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4891 && raid_map_ret) {
4892 int i, rot;
4893
4894 /* push stripe_nr back to the start of the full stripe */
4895 stripe_nr = raid56_full_stripe_start;
4896 do_div(stripe_nr, stripe_len);
4897
4898 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4899
4900 /* RAID[56] write or recovery. Return all stripes */
4901 num_stripes = map->num_stripes;
4902 max_errors = nr_parity_stripes(map);
4903
4904 raid_map = kmalloc_array(num_stripes, sizeof(u64),
4905 GFP_NOFS);
4906 if (!raid_map) {
4907 ret = -ENOMEM;
4908 goto out;
4909 }
4910
4911 /* Work out the disk rotation on this stripe-set */
4912 tmp = stripe_nr;
4913 rot = do_div(tmp, num_stripes);
4914
4915 /* Fill in the logical address of each stripe */
4916 tmp = stripe_nr * nr_data_stripes(map);
4917 for (i = 0; i < nr_data_stripes(map); i++)
4918 raid_map[(i+rot) % num_stripes] =
4919 em->start + (tmp + i) * map->stripe_len;
4920
4921 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4922 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4923 raid_map[(i+rot+1) % num_stripes] =
4924 RAID6_Q_STRIPE;
4925
4926 *length = map->stripe_len;
4927 stripe_index = 0;
4928 stripe_offset = 0;
4929 } else {
4930 /*
4931 * Mirror #0 or #1 means the original data block.
4932 * Mirror #2 is RAID5 parity block.
4933 * Mirror #3 is RAID6 Q block.
4934 */
4935 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4936 if (mirror_num > 1)
4937 stripe_index = nr_data_stripes(map) +
4938 mirror_num - 2;
4939
4940 /* We distribute the parity blocks across stripes */
4941 tmp = stripe_nr + stripe_index;
4942 stripe_index = do_div(tmp, map->num_stripes);
4943 }
4944 } else {
4945 /*
4946 * after this do_div call, stripe_nr is the number of stripes
4947 * on this device we have to walk to find the data, and
4948 * stripe_index is the number of our device in the stripe array
4949 */
4950 stripe_index = do_div(stripe_nr, map->num_stripes);
4951 mirror_num = stripe_index + 1;
4952 }
4953 BUG_ON(stripe_index >= map->num_stripes);
4954
4955 num_alloc_stripes = num_stripes;
4956 if (dev_replace_is_ongoing) {
4957 if (rw & (REQ_WRITE | REQ_DISCARD))
4958 num_alloc_stripes <<= 1;
4959 if (rw & REQ_GET_READ_MIRRORS)
4960 num_alloc_stripes++;
4961 }
4962 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4963 if (!bbio) {
4964 kfree(raid_map);
4965 ret = -ENOMEM;
4966 goto out;
4967 }
4968 atomic_set(&bbio->error, 0);
4969
4970 if (rw & REQ_DISCARD) {
4971 int factor = 0;
4972 int sub_stripes = 0;
4973 u64 stripes_per_dev = 0;
4974 u32 remaining_stripes = 0;
4975 u32 last_stripe = 0;
4976
4977 if (map->type &
4978 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4979 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4980 sub_stripes = 1;
4981 else
4982 sub_stripes = map->sub_stripes;
4983
4984 factor = map->num_stripes / sub_stripes;
4985 stripes_per_dev = div_u64_rem(stripe_nr_end -
4986 stripe_nr_orig,
4987 factor,
4988 &remaining_stripes);
4989 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4990 last_stripe *= sub_stripes;
4991 }
4992
4993 for (i = 0; i < num_stripes; i++) {
4994 bbio->stripes[i].physical =
4995 map->stripes[stripe_index].physical +
4996 stripe_offset + stripe_nr * map->stripe_len;
4997 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4998
4999 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5000 BTRFS_BLOCK_GROUP_RAID10)) {
5001 bbio->stripes[i].length = stripes_per_dev *
5002 map->stripe_len;
5003
5004 if (i / sub_stripes < remaining_stripes)
5005 bbio->stripes[i].length +=
5006 map->stripe_len;
5007
5008 /*
5009 * Special for the first stripe and
5010 * the last stripe:
5011 *
5012 * |-------|...|-------|
5013 * |----------|
5014 * off end_off
5015 */
5016 if (i < sub_stripes)
5017 bbio->stripes[i].length -=
5018 stripe_offset;
5019
5020 if (stripe_index >= last_stripe &&
5021 stripe_index <= (last_stripe +
5022 sub_stripes - 1))
5023 bbio->stripes[i].length -=
5024 stripe_end_offset;
5025
5026 if (i == sub_stripes - 1)
5027 stripe_offset = 0;
5028 } else
5029 bbio->stripes[i].length = *length;
5030
5031 stripe_index++;
5032 if (stripe_index == map->num_stripes) {
5033 /* This could only happen for RAID0/10 */
5034 stripe_index = 0;
5035 stripe_nr++;
5036 }
5037 }
5038 } else {
5039 for (i = 0; i < num_stripes; i++) {
5040 bbio->stripes[i].physical =
5041 map->stripes[stripe_index].physical +
5042 stripe_offset +
5043 stripe_nr * map->stripe_len;
5044 bbio->stripes[i].dev =
5045 map->stripes[stripe_index].dev;
5046 stripe_index++;
5047 }
5048 }
5049
5050 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5051 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5052 BTRFS_BLOCK_GROUP_RAID10 |
5053 BTRFS_BLOCK_GROUP_RAID5 |
5054 BTRFS_BLOCK_GROUP_DUP)) {
5055 max_errors = 1;
5056 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5057 max_errors = 2;
5058 }
5059 }
5060
5061 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5062 dev_replace->tgtdev != NULL) {
5063 int index_where_to_add;
5064 u64 srcdev_devid = dev_replace->srcdev->devid;
5065
5066 /*
5067 * duplicate the write operations while the dev replace
5068 * procedure is running. Since the copying of the old disk
5069 * to the new disk takes place at run time while the
5070 * filesystem is mounted writable, the regular write
5071 * operations to the old disk have to be duplicated to go
5072 * to the new disk as well.
5073 * Note that device->missing is handled by the caller, and
5074 * that the write to the old disk is already set up in the
5075 * stripes array.
5076 */
5077 index_where_to_add = num_stripes;
5078 for (i = 0; i < num_stripes; i++) {
5079 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5080 /* write to new disk, too */
5081 struct btrfs_bio_stripe *new =
5082 bbio->stripes + index_where_to_add;
5083 struct btrfs_bio_stripe *old =
5084 bbio->stripes + i;
5085
5086 new->physical = old->physical;
5087 new->length = old->length;
5088 new->dev = dev_replace->tgtdev;
5089 index_where_to_add++;
5090 max_errors++;
5091 }
5092 }
5093 num_stripes = index_where_to_add;
5094 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5095 dev_replace->tgtdev != NULL) {
5096 u64 srcdev_devid = dev_replace->srcdev->devid;
5097 int index_srcdev = 0;
5098 int found = 0;
5099 u64 physical_of_found = 0;
5100
5101 /*
5102 * During the dev-replace procedure, the target drive can
5103 * also be used to read data in case it is needed to repair
5104 * a corrupt block elsewhere. This is possible if the
5105 * requested area is left of the left cursor. In this area,
5106 * the target drive is a full copy of the source drive.
5107 */
5108 for (i = 0; i < num_stripes; i++) {
5109 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5110 /*
5111 * In case of DUP, in order to keep it
5112 * simple, only add the mirror with the
5113 * lowest physical address
5114 */
5115 if (found &&
5116 physical_of_found <=
5117 bbio->stripes[i].physical)
5118 continue;
5119 index_srcdev = i;
5120 found = 1;
5121 physical_of_found = bbio->stripes[i].physical;
5122 }
5123 }
5124 if (found) {
5125 u64 length = map->stripe_len;
5126
5127 if (physical_of_found + length <=
5128 dev_replace->cursor_left) {
5129 struct btrfs_bio_stripe *tgtdev_stripe =
5130 bbio->stripes + num_stripes;
5131
5132 tgtdev_stripe->physical = physical_of_found;
5133 tgtdev_stripe->length =
5134 bbio->stripes[index_srcdev].length;
5135 tgtdev_stripe->dev = dev_replace->tgtdev;
5136
5137 num_stripes++;
5138 }
5139 }
5140 }
5141
5142 *bbio_ret = bbio;
5143 bbio->num_stripes = num_stripes;
5144 bbio->max_errors = max_errors;
5145 bbio->mirror_num = mirror_num;
5146
5147 /*
5148 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5149 * mirror_num == num_stripes + 1 && dev_replace target drive is
5150 * available as a mirror
5151 */
5152 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5153 WARN_ON(num_stripes > 1);
5154 bbio->stripes[0].dev = dev_replace->tgtdev;
5155 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5156 bbio->mirror_num = map->num_stripes + 1;
5157 }
5158 if (raid_map) {
5159 sort_parity_stripes(bbio, raid_map);
5160 *raid_map_ret = raid_map;
5161 }
5162 out:
5163 if (dev_replace_is_ongoing)
5164 btrfs_dev_replace_unlock(dev_replace);
5165 free_extent_map(em);
5166 return ret;
5167 }
5168
5169 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5170 u64 logical, u64 *length,
5171 struct btrfs_bio **bbio_ret, int mirror_num)
5172 {
5173 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5174 mirror_num, NULL);
5175 }
5176
5177 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5178 u64 chunk_start, u64 physical, u64 devid,
5179 u64 **logical, int *naddrs, int *stripe_len)
5180 {
5181 struct extent_map_tree *em_tree = &map_tree->map_tree;
5182 struct extent_map *em;
5183 struct map_lookup *map;
5184 u64 *buf;
5185 u64 bytenr;
5186 u64 length;
5187 u64 stripe_nr;
5188 u64 rmap_len;
5189 int i, j, nr = 0;
5190
5191 read_lock(&em_tree->lock);
5192 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5193 read_unlock(&em_tree->lock);
5194
5195 if (!em) {
5196 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5197 chunk_start);
5198 return -EIO;
5199 }
5200
5201 if (em->start != chunk_start) {
5202 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5203 em->start, chunk_start);
5204 free_extent_map(em);
5205 return -EIO;
5206 }
5207 map = (struct map_lookup *)em->bdev;
5208
5209 length = em->len;
5210 rmap_len = map->stripe_len;
5211
5212 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5213 do_div(length, map->num_stripes / map->sub_stripes);
5214 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5215 do_div(length, map->num_stripes);
5216 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5217 BTRFS_BLOCK_GROUP_RAID6)) {
5218 do_div(length, nr_data_stripes(map));
5219 rmap_len = map->stripe_len * nr_data_stripes(map);
5220 }
5221
5222 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5223 BUG_ON(!buf); /* -ENOMEM */
5224
5225 for (i = 0; i < map->num_stripes; i++) {
5226 if (devid && map->stripes[i].dev->devid != devid)
5227 continue;
5228 if (map->stripes[i].physical > physical ||
5229 map->stripes[i].physical + length <= physical)
5230 continue;
5231
5232 stripe_nr = physical - map->stripes[i].physical;
5233 do_div(stripe_nr, map->stripe_len);
5234
5235 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5236 stripe_nr = stripe_nr * map->num_stripes + i;
5237 do_div(stripe_nr, map->sub_stripes);
5238 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5239 stripe_nr = stripe_nr * map->num_stripes + i;
5240 } /* else if RAID[56], multiply by nr_data_stripes().
5241 * Alternatively, just use rmap_len below instead of
5242 * map->stripe_len */
5243
5244 bytenr = chunk_start + stripe_nr * rmap_len;
5245 WARN_ON(nr >= map->num_stripes);
5246 for (j = 0; j < nr; j++) {
5247 if (buf[j] == bytenr)
5248 break;
5249 }
5250 if (j == nr) {
5251 WARN_ON(nr >= map->num_stripes);
5252 buf[nr++] = bytenr;
5253 }
5254 }
5255
5256 *logical = buf;
5257 *naddrs = nr;
5258 *stripe_len = rmap_len;
5259
5260 free_extent_map(em);
5261 return 0;
5262 }
5263
5264 static void btrfs_end_bio(struct bio *bio, int err)
5265 {
5266 struct btrfs_bio *bbio = bio->bi_private;
5267 int is_orig_bio = 0;
5268
5269 if (err) {
5270 atomic_inc(&bbio->error);
5271 if (err == -EIO || err == -EREMOTEIO) {
5272 unsigned int stripe_index =
5273 btrfs_io_bio(bio)->stripe_index;
5274 struct btrfs_device *dev;
5275
5276 BUG_ON(stripe_index >= bbio->num_stripes);
5277 dev = bbio->stripes[stripe_index].dev;
5278 if (dev->bdev) {
5279 if (bio->bi_rw & WRITE)
5280 btrfs_dev_stat_inc(dev,
5281 BTRFS_DEV_STAT_WRITE_ERRS);
5282 else
5283 btrfs_dev_stat_inc(dev,
5284 BTRFS_DEV_STAT_READ_ERRS);
5285 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5286 btrfs_dev_stat_inc(dev,
5287 BTRFS_DEV_STAT_FLUSH_ERRS);
5288 btrfs_dev_stat_print_on_error(dev);
5289 }
5290 }
5291 }
5292
5293 if (bio == bbio->orig_bio)
5294 is_orig_bio = 1;
5295
5296 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5297 if (!is_orig_bio) {
5298 bio_put(bio);
5299 bio = bbio->orig_bio;
5300 }
5301 bio->bi_private = bbio->private;
5302 bio->bi_end_io = bbio->end_io;
5303 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5304 /* only send an error to the higher layers if it is
5305 * beyond the tolerance of the btrfs bio
5306 */
5307 if (atomic_read(&bbio->error) > bbio->max_errors) {
5308 err = -EIO;
5309 } else {
5310 /*
5311 * this bio is actually up to date, we didn't
5312 * go over the max number of errors
5313 */
5314 set_bit(BIO_UPTODATE, &bio->bi_flags);
5315 err = 0;
5316 }
5317 kfree(bbio);
5318
5319 bio_endio(bio, err);
5320 } else if (!is_orig_bio) {
5321 bio_put(bio);
5322 }
5323 }
5324
5325 struct async_sched {
5326 struct bio *bio;
5327 int rw;
5328 struct btrfs_fs_info *info;
5329 struct btrfs_work work;
5330 };
5331
5332 /*
5333 * see run_scheduled_bios for a description of why bios are collected for
5334 * async submit.
5335 *
5336 * This will add one bio to the pending list for a device and make sure
5337 * the work struct is scheduled.
5338 */
5339 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5340 struct btrfs_device *device,
5341 int rw, struct bio *bio)
5342 {
5343 int should_queue = 1;
5344 struct btrfs_pending_bios *pending_bios;
5345
5346 if (device->missing || !device->bdev) {
5347 bio_endio(bio, -EIO);
5348 return;
5349 }
5350
5351 /* don't bother with additional async steps for reads, right now */
5352 if (!(rw & REQ_WRITE)) {
5353 bio_get(bio);
5354 btrfsic_submit_bio(rw, bio);
5355 bio_put(bio);
5356 return;
5357 }
5358
5359 /*
5360 * nr_async_bios allows us to reliably return congestion to the
5361 * higher layers. Otherwise, the async bio makes it appear we have
5362 * made progress against dirty pages when we've really just put it
5363 * on a queue for later
5364 */
5365 atomic_inc(&root->fs_info->nr_async_bios);
5366 WARN_ON(bio->bi_next);
5367 bio->bi_next = NULL;
5368 bio->bi_rw |= rw;
5369
5370 spin_lock(&device->io_lock);
5371 if (bio->bi_rw & REQ_SYNC)
5372 pending_bios = &device->pending_sync_bios;
5373 else
5374 pending_bios = &device->pending_bios;
5375
5376 if (pending_bios->tail)
5377 pending_bios->tail->bi_next = bio;
5378
5379 pending_bios->tail = bio;
5380 if (!pending_bios->head)
5381 pending_bios->head = bio;
5382 if (device->running_pending)
5383 should_queue = 0;
5384
5385 spin_unlock(&device->io_lock);
5386
5387 if (should_queue)
5388 btrfs_queue_worker(&root->fs_info->submit_workers,
5389 &device->work);
5390 }
5391
5392 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5393 sector_t sector)
5394 {
5395 struct bio_vec *prev;
5396 struct request_queue *q = bdev_get_queue(bdev);
5397 unsigned int max_sectors = queue_max_sectors(q);
5398 struct bvec_merge_data bvm = {
5399 .bi_bdev = bdev,
5400 .bi_sector = sector,
5401 .bi_rw = bio->bi_rw,
5402 };
5403
5404 if (WARN_ON(bio->bi_vcnt == 0))
5405 return 1;
5406
5407 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5408 if (bio_sectors(bio) > max_sectors)
5409 return 0;
5410
5411 if (!q->merge_bvec_fn)
5412 return 1;
5413
5414 bvm.bi_size = bio->bi_size - prev->bv_len;
5415 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5416 return 0;
5417 return 1;
5418 }
5419
5420 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5421 struct bio *bio, u64 physical, int dev_nr,
5422 int rw, int async)
5423 {
5424 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5425
5426 bio->bi_private = bbio;
5427 btrfs_io_bio(bio)->stripe_index = dev_nr;
5428 bio->bi_end_io = btrfs_end_bio;
5429 bio->bi_sector = physical >> 9;
5430 #ifdef DEBUG
5431 {
5432 struct rcu_string *name;
5433
5434 rcu_read_lock();
5435 name = rcu_dereference(dev->name);
5436 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5437 "(%s id %llu), size=%u\n", rw,
5438 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5439 name->str, dev->devid, bio->bi_size);
5440 rcu_read_unlock();
5441 }
5442 #endif
5443 bio->bi_bdev = dev->bdev;
5444 if (async)
5445 btrfs_schedule_bio(root, dev, rw, bio);
5446 else
5447 btrfsic_submit_bio(rw, bio);
5448 }
5449
5450 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5451 struct bio *first_bio, struct btrfs_device *dev,
5452 int dev_nr, int rw, int async)
5453 {
5454 struct bio_vec *bvec = first_bio->bi_io_vec;
5455 struct bio *bio;
5456 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5457 u64 physical = bbio->stripes[dev_nr].physical;
5458
5459 again:
5460 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5461 if (!bio)
5462 return -ENOMEM;
5463
5464 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5465 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5466 bvec->bv_offset) < bvec->bv_len) {
5467 u64 len = bio->bi_size;
5468
5469 atomic_inc(&bbio->stripes_pending);
5470 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5471 rw, async);
5472 physical += len;
5473 goto again;
5474 }
5475 bvec++;
5476 }
5477
5478 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5479 return 0;
5480 }
5481
5482 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5483 {
5484 atomic_inc(&bbio->error);
5485 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5486 bio->bi_private = bbio->private;
5487 bio->bi_end_io = bbio->end_io;
5488 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5489 bio->bi_sector = logical >> 9;
5490 kfree(bbio);
5491 bio_endio(bio, -EIO);
5492 }
5493 }
5494
5495 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5496 int mirror_num, int async_submit)
5497 {
5498 struct btrfs_device *dev;
5499 struct bio *first_bio = bio;
5500 u64 logical = (u64)bio->bi_sector << 9;
5501 u64 length = 0;
5502 u64 map_length;
5503 u64 *raid_map = NULL;
5504 int ret;
5505 int dev_nr = 0;
5506 int total_devs = 1;
5507 struct btrfs_bio *bbio = NULL;
5508
5509 length = bio->bi_size;
5510 map_length = length;
5511
5512 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5513 mirror_num, &raid_map);
5514 if (ret) /* -ENOMEM */
5515 return ret;
5516
5517 total_devs = bbio->num_stripes;
5518 bbio->orig_bio = first_bio;
5519 bbio->private = first_bio->bi_private;
5520 bbio->end_io = first_bio->bi_end_io;
5521 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5522
5523 if (raid_map) {
5524 /* In this case, map_length has been set to the length of
5525 a single stripe; not the whole write */
5526 if (rw & WRITE) {
5527 return raid56_parity_write(root, bio, bbio,
5528 raid_map, map_length);
5529 } else {
5530 return raid56_parity_recover(root, bio, bbio,
5531 raid_map, map_length,
5532 mirror_num);
5533 }
5534 }
5535
5536 if (map_length < length) {
5537 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5538 logical, length, map_length);
5539 BUG();
5540 }
5541
5542 while (dev_nr < total_devs) {
5543 dev = bbio->stripes[dev_nr].dev;
5544 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5545 bbio_error(bbio, first_bio, logical);
5546 dev_nr++;
5547 continue;
5548 }
5549
5550 /*
5551 * Check and see if we're ok with this bio based on it's size
5552 * and offset with the given device.
5553 */
5554 if (!bio_size_ok(dev->bdev, first_bio,
5555 bbio->stripes[dev_nr].physical >> 9)) {
5556 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5557 dev_nr, rw, async_submit);
5558 BUG_ON(ret);
5559 dev_nr++;
5560 continue;
5561 }
5562
5563 if (dev_nr < total_devs - 1) {
5564 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5565 BUG_ON(!bio); /* -ENOMEM */
5566 } else {
5567 bio = first_bio;
5568 }
5569
5570 submit_stripe_bio(root, bbio, bio,
5571 bbio->stripes[dev_nr].physical, dev_nr, rw,
5572 async_submit);
5573 dev_nr++;
5574 }
5575 return 0;
5576 }
5577
5578 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5579 u8 *uuid, u8 *fsid)
5580 {
5581 struct btrfs_device *device;
5582 struct btrfs_fs_devices *cur_devices;
5583
5584 cur_devices = fs_info->fs_devices;
5585 while (cur_devices) {
5586 if (!fsid ||
5587 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5588 device = __find_device(&cur_devices->devices,
5589 devid, uuid);
5590 if (device)
5591 return device;
5592 }
5593 cur_devices = cur_devices->seed;
5594 }
5595 return NULL;
5596 }
5597
5598 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5599 u64 devid, u8 *dev_uuid)
5600 {
5601 struct btrfs_device *device;
5602 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5603
5604 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5605 if (IS_ERR(device))
5606 return NULL;
5607
5608 list_add(&device->dev_list, &fs_devices->devices);
5609 device->fs_devices = fs_devices;
5610 fs_devices->num_devices++;
5611
5612 device->missing = 1;
5613 fs_devices->missing_devices++;
5614
5615 return device;
5616 }
5617
5618 /**
5619 * btrfs_alloc_device - allocate struct btrfs_device
5620 * @fs_info: used only for generating a new devid, can be NULL if
5621 * devid is provided (i.e. @devid != NULL).
5622 * @devid: a pointer to devid for this device. If NULL a new devid
5623 * is generated.
5624 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5625 * is generated.
5626 *
5627 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5628 * on error. Returned struct is not linked onto any lists and can be
5629 * destroyed with kfree() right away.
5630 */
5631 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5632 const u64 *devid,
5633 const u8 *uuid)
5634 {
5635 struct btrfs_device *dev;
5636 u64 tmp;
5637
5638 if (WARN_ON(!devid && !fs_info))
5639 return ERR_PTR(-EINVAL);
5640
5641 dev = __alloc_device();
5642 if (IS_ERR(dev))
5643 return dev;
5644
5645 if (devid)
5646 tmp = *devid;
5647 else {
5648 int ret;
5649
5650 ret = find_next_devid(fs_info, &tmp);
5651 if (ret) {
5652 kfree(dev);
5653 return ERR_PTR(ret);
5654 }
5655 }
5656 dev->devid = tmp;
5657
5658 if (uuid)
5659 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5660 else
5661 generate_random_uuid(dev->uuid);
5662
5663 dev->work.func = pending_bios_fn;
5664
5665 return dev;
5666 }
5667
5668 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5669 struct extent_buffer *leaf,
5670 struct btrfs_chunk *chunk)
5671 {
5672 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5673 struct map_lookup *map;
5674 struct extent_map *em;
5675 u64 logical;
5676 u64 length;
5677 u64 devid;
5678 u8 uuid[BTRFS_UUID_SIZE];
5679 int num_stripes;
5680 int ret;
5681 int i;
5682
5683 logical = key->offset;
5684 length = btrfs_chunk_length(leaf, chunk);
5685
5686 read_lock(&map_tree->map_tree.lock);
5687 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5688 read_unlock(&map_tree->map_tree.lock);
5689
5690 /* already mapped? */
5691 if (em && em->start <= logical && em->start + em->len > logical) {
5692 free_extent_map(em);
5693 return 0;
5694 } else if (em) {
5695 free_extent_map(em);
5696 }
5697
5698 em = alloc_extent_map();
5699 if (!em)
5700 return -ENOMEM;
5701 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5702 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5703 if (!map) {
5704 free_extent_map(em);
5705 return -ENOMEM;
5706 }
5707
5708 em->bdev = (struct block_device *)map;
5709 em->start = logical;
5710 em->len = length;
5711 em->orig_start = 0;
5712 em->block_start = 0;
5713 em->block_len = em->len;
5714
5715 map->num_stripes = num_stripes;
5716 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5717 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5718 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5719 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5720 map->type = btrfs_chunk_type(leaf, chunk);
5721 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5722 for (i = 0; i < num_stripes; i++) {
5723 map->stripes[i].physical =
5724 btrfs_stripe_offset_nr(leaf, chunk, i);
5725 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5726 read_extent_buffer(leaf, uuid, (unsigned long)
5727 btrfs_stripe_dev_uuid_nr(chunk, i),
5728 BTRFS_UUID_SIZE);
5729 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5730 uuid, NULL);
5731 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5732 kfree(map);
5733 free_extent_map(em);
5734 return -EIO;
5735 }
5736 if (!map->stripes[i].dev) {
5737 map->stripes[i].dev =
5738 add_missing_dev(root, devid, uuid);
5739 if (!map->stripes[i].dev) {
5740 kfree(map);
5741 free_extent_map(em);
5742 return -EIO;
5743 }
5744 }
5745 map->stripes[i].dev->in_fs_metadata = 1;
5746 }
5747
5748 write_lock(&map_tree->map_tree.lock);
5749 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5750 write_unlock(&map_tree->map_tree.lock);
5751 BUG_ON(ret); /* Tree corruption */
5752 free_extent_map(em);
5753
5754 return 0;
5755 }
5756
5757 static void fill_device_from_item(struct extent_buffer *leaf,
5758 struct btrfs_dev_item *dev_item,
5759 struct btrfs_device *device)
5760 {
5761 unsigned long ptr;
5762
5763 device->devid = btrfs_device_id(leaf, dev_item);
5764 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5765 device->total_bytes = device->disk_total_bytes;
5766 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5767 device->type = btrfs_device_type(leaf, dev_item);
5768 device->io_align = btrfs_device_io_align(leaf, dev_item);
5769 device->io_width = btrfs_device_io_width(leaf, dev_item);
5770 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5771 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5772 device->is_tgtdev_for_dev_replace = 0;
5773
5774 ptr = btrfs_device_uuid(dev_item);
5775 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5776 }
5777
5778 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5779 {
5780 struct btrfs_fs_devices *fs_devices;
5781 int ret;
5782
5783 BUG_ON(!mutex_is_locked(&uuid_mutex));
5784
5785 fs_devices = root->fs_info->fs_devices->seed;
5786 while (fs_devices) {
5787 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5788 ret = 0;
5789 goto out;
5790 }
5791 fs_devices = fs_devices->seed;
5792 }
5793
5794 fs_devices = find_fsid(fsid);
5795 if (!fs_devices) {
5796 ret = -ENOENT;
5797 goto out;
5798 }
5799
5800 fs_devices = clone_fs_devices(fs_devices);
5801 if (IS_ERR(fs_devices)) {
5802 ret = PTR_ERR(fs_devices);
5803 goto out;
5804 }
5805
5806 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5807 root->fs_info->bdev_holder);
5808 if (ret) {
5809 free_fs_devices(fs_devices);
5810 goto out;
5811 }
5812
5813 if (!fs_devices->seeding) {
5814 __btrfs_close_devices(fs_devices);
5815 free_fs_devices(fs_devices);
5816 ret = -EINVAL;
5817 goto out;
5818 }
5819
5820 fs_devices->seed = root->fs_info->fs_devices->seed;
5821 root->fs_info->fs_devices->seed = fs_devices;
5822 out:
5823 return ret;
5824 }
5825
5826 static int read_one_dev(struct btrfs_root *root,
5827 struct extent_buffer *leaf,
5828 struct btrfs_dev_item *dev_item)
5829 {
5830 struct btrfs_device *device;
5831 u64 devid;
5832 int ret;
5833 u8 fs_uuid[BTRFS_UUID_SIZE];
5834 u8 dev_uuid[BTRFS_UUID_SIZE];
5835
5836 devid = btrfs_device_id(leaf, dev_item);
5837 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5838 BTRFS_UUID_SIZE);
5839 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5840 BTRFS_UUID_SIZE);
5841
5842 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5843 ret = open_seed_devices(root, fs_uuid);
5844 if (ret && !btrfs_test_opt(root, DEGRADED))
5845 return ret;
5846 }
5847
5848 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5849 if (!device || !device->bdev) {
5850 if (!btrfs_test_opt(root, DEGRADED))
5851 return -EIO;
5852
5853 if (!device) {
5854 btrfs_warn(root->fs_info, "devid %llu missing", devid);
5855 device = add_missing_dev(root, devid, dev_uuid);
5856 if (!device)
5857 return -ENOMEM;
5858 } else if (!device->missing) {
5859 /*
5860 * this happens when a device that was properly setup
5861 * in the device info lists suddenly goes bad.
5862 * device->bdev is NULL, and so we have to set
5863 * device->missing to one here
5864 */
5865 root->fs_info->fs_devices->missing_devices++;
5866 device->missing = 1;
5867 }
5868 }
5869
5870 if (device->fs_devices != root->fs_info->fs_devices) {
5871 BUG_ON(device->writeable);
5872 if (device->generation !=
5873 btrfs_device_generation(leaf, dev_item))
5874 return -EINVAL;
5875 }
5876
5877 fill_device_from_item(leaf, dev_item, device);
5878 device->in_fs_metadata = 1;
5879 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5880 device->fs_devices->total_rw_bytes += device->total_bytes;
5881 spin_lock(&root->fs_info->free_chunk_lock);
5882 root->fs_info->free_chunk_space += device->total_bytes -
5883 device->bytes_used;
5884 spin_unlock(&root->fs_info->free_chunk_lock);
5885 }
5886 ret = 0;
5887 return ret;
5888 }
5889
5890 int btrfs_read_sys_array(struct btrfs_root *root)
5891 {
5892 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5893 struct extent_buffer *sb;
5894 struct btrfs_disk_key *disk_key;
5895 struct btrfs_chunk *chunk;
5896 u8 *ptr;
5897 unsigned long sb_ptr;
5898 int ret = 0;
5899 u32 num_stripes;
5900 u32 array_size;
5901 u32 len = 0;
5902 u32 cur;
5903 struct btrfs_key key;
5904
5905 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5906 BTRFS_SUPER_INFO_SIZE);
5907 if (!sb)
5908 return -ENOMEM;
5909 btrfs_set_buffer_uptodate(sb);
5910 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5911 /*
5912 * The sb extent buffer is artifical and just used to read the system array.
5913 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5914 * pages up-to-date when the page is larger: extent does not cover the
5915 * whole page and consequently check_page_uptodate does not find all
5916 * the page's extents up-to-date (the hole beyond sb),
5917 * write_extent_buffer then triggers a WARN_ON.
5918 *
5919 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5920 * but sb spans only this function. Add an explicit SetPageUptodate call
5921 * to silence the warning eg. on PowerPC 64.
5922 */
5923 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5924 SetPageUptodate(sb->pages[0]);
5925
5926 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5927 array_size = btrfs_super_sys_array_size(super_copy);
5928
5929 ptr = super_copy->sys_chunk_array;
5930 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5931 cur = 0;
5932
5933 while (cur < array_size) {
5934 disk_key = (struct btrfs_disk_key *)ptr;
5935 btrfs_disk_key_to_cpu(&key, disk_key);
5936
5937 len = sizeof(*disk_key); ptr += len;
5938 sb_ptr += len;
5939 cur += len;
5940
5941 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5942 chunk = (struct btrfs_chunk *)sb_ptr;
5943 ret = read_one_chunk(root, &key, sb, chunk);
5944 if (ret)
5945 break;
5946 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5947 len = btrfs_chunk_item_size(num_stripes);
5948 } else {
5949 ret = -EIO;
5950 break;
5951 }
5952 ptr += len;
5953 sb_ptr += len;
5954 cur += len;
5955 }
5956 free_extent_buffer(sb);
5957 return ret;
5958 }
5959
5960 int btrfs_read_chunk_tree(struct btrfs_root *root)
5961 {
5962 struct btrfs_path *path;
5963 struct extent_buffer *leaf;
5964 struct btrfs_key key;
5965 struct btrfs_key found_key;
5966 int ret;
5967 int slot;
5968
5969 root = root->fs_info->chunk_root;
5970
5971 path = btrfs_alloc_path();
5972 if (!path)
5973 return -ENOMEM;
5974
5975 mutex_lock(&uuid_mutex);
5976 lock_chunks(root);
5977
5978 /*
5979 * Read all device items, and then all the chunk items. All
5980 * device items are found before any chunk item (their object id
5981 * is smaller than the lowest possible object id for a chunk
5982 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5983 */
5984 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5985 key.offset = 0;
5986 key.type = 0;
5987 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5988 if (ret < 0)
5989 goto error;
5990 while (1) {
5991 leaf = path->nodes[0];
5992 slot = path->slots[0];
5993 if (slot >= btrfs_header_nritems(leaf)) {
5994 ret = btrfs_next_leaf(root, path);
5995 if (ret == 0)
5996 continue;
5997 if (ret < 0)
5998 goto error;
5999 break;
6000 }
6001 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6002 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6003 struct btrfs_dev_item *dev_item;
6004 dev_item = btrfs_item_ptr(leaf, slot,
6005 struct btrfs_dev_item);
6006 ret = read_one_dev(root, leaf, dev_item);
6007 if (ret)
6008 goto error;
6009 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6010 struct btrfs_chunk *chunk;
6011 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6012 ret = read_one_chunk(root, &found_key, leaf, chunk);
6013 if (ret)
6014 goto error;
6015 }
6016 path->slots[0]++;
6017 }
6018 ret = 0;
6019 error:
6020 unlock_chunks(root);
6021 mutex_unlock(&uuid_mutex);
6022
6023 btrfs_free_path(path);
6024 return ret;
6025 }
6026
6027 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6028 {
6029 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6030 struct btrfs_device *device;
6031
6032 mutex_lock(&fs_devices->device_list_mutex);
6033 list_for_each_entry(device, &fs_devices->devices, dev_list)
6034 device->dev_root = fs_info->dev_root;
6035 mutex_unlock(&fs_devices->device_list_mutex);
6036 }
6037
6038 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6039 {
6040 int i;
6041
6042 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6043 btrfs_dev_stat_reset(dev, i);
6044 }
6045
6046 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6047 {
6048 struct btrfs_key key;
6049 struct btrfs_key found_key;
6050 struct btrfs_root *dev_root = fs_info->dev_root;
6051 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6052 struct extent_buffer *eb;
6053 int slot;
6054 int ret = 0;
6055 struct btrfs_device *device;
6056 struct btrfs_path *path = NULL;
6057 int i;
6058
6059 path = btrfs_alloc_path();
6060 if (!path) {
6061 ret = -ENOMEM;
6062 goto out;
6063 }
6064
6065 mutex_lock(&fs_devices->device_list_mutex);
6066 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6067 int item_size;
6068 struct btrfs_dev_stats_item *ptr;
6069
6070 key.objectid = 0;
6071 key.type = BTRFS_DEV_STATS_KEY;
6072 key.offset = device->devid;
6073 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6074 if (ret) {
6075 __btrfs_reset_dev_stats(device);
6076 device->dev_stats_valid = 1;
6077 btrfs_release_path(path);
6078 continue;
6079 }
6080 slot = path->slots[0];
6081 eb = path->nodes[0];
6082 btrfs_item_key_to_cpu(eb, &found_key, slot);
6083 item_size = btrfs_item_size_nr(eb, slot);
6084
6085 ptr = btrfs_item_ptr(eb, slot,
6086 struct btrfs_dev_stats_item);
6087
6088 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6089 if (item_size >= (1 + i) * sizeof(__le64))
6090 btrfs_dev_stat_set(device, i,
6091 btrfs_dev_stats_value(eb, ptr, i));
6092 else
6093 btrfs_dev_stat_reset(device, i);
6094 }
6095
6096 device->dev_stats_valid = 1;
6097 btrfs_dev_stat_print_on_load(device);
6098 btrfs_release_path(path);
6099 }
6100 mutex_unlock(&fs_devices->device_list_mutex);
6101
6102 out:
6103 btrfs_free_path(path);
6104 return ret < 0 ? ret : 0;
6105 }
6106
6107 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6108 struct btrfs_root *dev_root,
6109 struct btrfs_device *device)
6110 {
6111 struct btrfs_path *path;
6112 struct btrfs_key key;
6113 struct extent_buffer *eb;
6114 struct btrfs_dev_stats_item *ptr;
6115 int ret;
6116 int i;
6117
6118 key.objectid = 0;
6119 key.type = BTRFS_DEV_STATS_KEY;
6120 key.offset = device->devid;
6121
6122 path = btrfs_alloc_path();
6123 BUG_ON(!path);
6124 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6125 if (ret < 0) {
6126 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6127 ret, rcu_str_deref(device->name));
6128 goto out;
6129 }
6130
6131 if (ret == 0 &&
6132 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6133 /* need to delete old one and insert a new one */
6134 ret = btrfs_del_item(trans, dev_root, path);
6135 if (ret != 0) {
6136 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6137 rcu_str_deref(device->name), ret);
6138 goto out;
6139 }
6140 ret = 1;
6141 }
6142
6143 if (ret == 1) {
6144 /* need to insert a new item */
6145 btrfs_release_path(path);
6146 ret = btrfs_insert_empty_item(trans, dev_root, path,
6147 &key, sizeof(*ptr));
6148 if (ret < 0) {
6149 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6150 rcu_str_deref(device->name), ret);
6151 goto out;
6152 }
6153 }
6154
6155 eb = path->nodes[0];
6156 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6157 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6158 btrfs_set_dev_stats_value(eb, ptr, i,
6159 btrfs_dev_stat_read(device, i));
6160 btrfs_mark_buffer_dirty(eb);
6161
6162 out:
6163 btrfs_free_path(path);
6164 return ret;
6165 }
6166
6167 /*
6168 * called from commit_transaction. Writes all changed device stats to disk.
6169 */
6170 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6171 struct btrfs_fs_info *fs_info)
6172 {
6173 struct btrfs_root *dev_root = fs_info->dev_root;
6174 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6175 struct btrfs_device *device;
6176 int ret = 0;
6177
6178 mutex_lock(&fs_devices->device_list_mutex);
6179 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6180 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6181 continue;
6182
6183 ret = update_dev_stat_item(trans, dev_root, device);
6184 if (!ret)
6185 device->dev_stats_dirty = 0;
6186 }
6187 mutex_unlock(&fs_devices->device_list_mutex);
6188
6189 return ret;
6190 }
6191
6192 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6193 {
6194 btrfs_dev_stat_inc(dev, index);
6195 btrfs_dev_stat_print_on_error(dev);
6196 }
6197
6198 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6199 {
6200 if (!dev->dev_stats_valid)
6201 return;
6202 printk_ratelimited_in_rcu(KERN_ERR
6203 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6204 rcu_str_deref(dev->name),
6205 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6206 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6207 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6208 btrfs_dev_stat_read(dev,
6209 BTRFS_DEV_STAT_CORRUPTION_ERRS),
6210 btrfs_dev_stat_read(dev,
6211 BTRFS_DEV_STAT_GENERATION_ERRS));
6212 }
6213
6214 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6215 {
6216 int i;
6217
6218 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6219 if (btrfs_dev_stat_read(dev, i) != 0)
6220 break;
6221 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6222 return; /* all values == 0, suppress message */
6223
6224 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6225 rcu_str_deref(dev->name),
6226 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6227 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6228 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6229 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6230 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6231 }
6232
6233 int btrfs_get_dev_stats(struct btrfs_root *root,
6234 struct btrfs_ioctl_get_dev_stats *stats)
6235 {
6236 struct btrfs_device *dev;
6237 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6238 int i;
6239
6240 mutex_lock(&fs_devices->device_list_mutex);
6241 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6242 mutex_unlock(&fs_devices->device_list_mutex);
6243
6244 if (!dev) {
6245 printk(KERN_WARNING
6246 "btrfs: get dev_stats failed, device not found\n");
6247 return -ENODEV;
6248 } else if (!dev->dev_stats_valid) {
6249 printk(KERN_WARNING
6250 "btrfs: get dev_stats failed, not yet valid\n");
6251 return -ENODEV;
6252 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6253 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6254 if (stats->nr_items > i)
6255 stats->values[i] =
6256 btrfs_dev_stat_read_and_reset(dev, i);
6257 else
6258 btrfs_dev_stat_reset(dev, i);
6259 }
6260 } else {
6261 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6262 if (stats->nr_items > i)
6263 stats->values[i] = btrfs_dev_stat_read(dev, i);
6264 }
6265 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6266 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6267 return 0;
6268 }
6269
6270 int btrfs_scratch_superblock(struct btrfs_device *device)
6271 {
6272 struct buffer_head *bh;
6273 struct btrfs_super_block *disk_super;
6274
6275 bh = btrfs_read_dev_super(device->bdev);
6276 if (!bh)
6277 return -EINVAL;
6278 disk_super = (struct btrfs_super_block *)bh->b_data;
6279
6280 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6281 set_buffer_dirty(bh);
6282 sync_dirty_buffer(bh);
6283 brelse(bh);
6284
6285 return 0;
6286 }
Linus' kernel tree