projects / deliverable / linux.git / blob
? search:


d38065eca7c9159c42122ca5747b4c27a222cf7a
[deliverable/linux.git] / fs / btrfs / volumes.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
31 #include "compat.h"
32 #include "ctree.h"
33 #include "extent_map.h"
34 #include "disk-io.h"
35 #include "transaction.h"
36 #include "print-tree.h"
37 #include "volumes.h"
38 #include "raid56.h"
39 #include "async-thread.h"
40 #include "check-integrity.h"
41 #include "rcu-string.h"
42 #include "math.h"
43 #include "dev-replace.h"
44
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
52
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
55
56 static void lock_chunks(struct btrfs_root *root)
57 {
58 mutex_lock(&root->fs_info->chunk_mutex);
59 }
60
61 static void unlock_chunks(struct btrfs_root *root)
62 {
63 mutex_unlock(&root->fs_info->chunk_mutex);
64 }
65
66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
67 {
68 struct btrfs_fs_devices *fs_devs;
69
70 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
71 if (!fs_devs)
72 return ERR_PTR(-ENOMEM);
73
74 mutex_init(&fs_devs->device_list_mutex);
75
76 INIT_LIST_HEAD(&fs_devs->devices);
77 INIT_LIST_HEAD(&fs_devs->alloc_list);
78 INIT_LIST_HEAD(&fs_devs->list);
79
80 return fs_devs;
81 }
82
83 /**
84 * alloc_fs_devices - allocate struct btrfs_fs_devices
85 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
86 * generated.
87 *
88 * Return: a pointer to a new &struct btrfs_fs_devices on success;
89 * ERR_PTR() on error. Returned struct is not linked onto any lists and
90 * can be destroyed with kfree() right away.
91 */
92 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
93 {
94 struct btrfs_fs_devices *fs_devs;
95
96 fs_devs = __alloc_fs_devices();
97 if (IS_ERR(fs_devs))
98 return fs_devs;
99
100 if (fsid)
101 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
102 else
103 generate_random_uuid(fs_devs->fsid);
104
105 return fs_devs;
106 }
107
108 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
109 {
110 struct btrfs_device *device;
111 WARN_ON(fs_devices->opened);
112 while (!list_empty(&fs_devices->devices)) {
113 device = list_entry(fs_devices->devices.next,
114 struct btrfs_device, dev_list);
115 list_del(&device->dev_list);
116 rcu_string_free(device->name);
117 kfree(device);
118 }
119 kfree(fs_devices);
120 }
121
122 static void btrfs_kobject_uevent(struct block_device *bdev,
123 enum kobject_action action)
124 {
125 int ret;
126
127 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
128 if (ret)
129 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
130 action,
131 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
132 &disk_to_dev(bdev->bd_disk)->kobj);
133 }
134
135 void btrfs_cleanup_fs_uuids(void)
136 {
137 struct btrfs_fs_devices *fs_devices;
138
139 while (!list_empty(&fs_uuids)) {
140 fs_devices = list_entry(fs_uuids.next,
141 struct btrfs_fs_devices, list);
142 list_del(&fs_devices->list);
143 free_fs_devices(fs_devices);
144 }
145 }
146
147 static struct btrfs_device *__alloc_device(void)
148 {
149 struct btrfs_device *dev;
150
151 dev = kzalloc(sizeof(*dev), GFP_NOFS);
152 if (!dev)
153 return ERR_PTR(-ENOMEM);
154
155 INIT_LIST_HEAD(&dev->dev_list);
156 INIT_LIST_HEAD(&dev->dev_alloc_list);
157
158 spin_lock_init(&dev->io_lock);
159
160 spin_lock_init(&dev->reada_lock);
161 atomic_set(&dev->reada_in_flight, 0);
162 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
163 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
164
165 return dev;
166 }
167
168 static noinline struct btrfs_device *__find_device(struct list_head *head,
169 u64 devid, u8 *uuid)
170 {
171 struct btrfs_device *dev;
172
173 list_for_each_entry(dev, head, dev_list) {
174 if (dev->devid == devid &&
175 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
176 return dev;
177 }
178 }
179 return NULL;
180 }
181
182 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
183 {
184 struct btrfs_fs_devices *fs_devices;
185
186 list_for_each_entry(fs_devices, &fs_uuids, list) {
187 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
188 return fs_devices;
189 }
190 return NULL;
191 }
192
193 static int
194 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
195 int flush, struct block_device **bdev,
196 struct buffer_head **bh)
197 {
198 int ret;
199
200 *bdev = blkdev_get_by_path(device_path, flags, holder);
201
202 if (IS_ERR(*bdev)) {
203 ret = PTR_ERR(*bdev);
204 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
205 goto error;
206 }
207
208 if (flush)
209 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
210 ret = set_blocksize(*bdev, 4096);
211 if (ret) {
212 blkdev_put(*bdev, flags);
213 goto error;
214 }
215 invalidate_bdev(*bdev);
216 *bh = btrfs_read_dev_super(*bdev);
217 if (!*bh) {
218 ret = -EINVAL;
219 blkdev_put(*bdev, flags);
220 goto error;
221 }
222
223 return 0;
224
225 error:
226 *bdev = NULL;
227 *bh = NULL;
228 return ret;
229 }
230
231 static void requeue_list(struct btrfs_pending_bios *pending_bios,
232 struct bio *head, struct bio *tail)
233 {
234
235 struct bio *old_head;
236
237 old_head = pending_bios->head;
238 pending_bios->head = head;
239 if (pending_bios->tail)
240 tail->bi_next = old_head;
241 else
242 pending_bios->tail = tail;
243 }
244
245 /*
246 * we try to collect pending bios for a device so we don't get a large
247 * number of procs sending bios down to the same device. This greatly
248 * improves the schedulers ability to collect and merge the bios.
249 *
250 * But, it also turns into a long list of bios to process and that is sure
251 * to eventually make the worker thread block. The solution here is to
252 * make some progress and then put this work struct back at the end of
253 * the list if the block device is congested. This way, multiple devices
254 * can make progress from a single worker thread.
255 */
256 static noinline void run_scheduled_bios(struct btrfs_device *device)
257 {
258 struct bio *pending;
259 struct backing_dev_info *bdi;
260 struct btrfs_fs_info *fs_info;
261 struct btrfs_pending_bios *pending_bios;
262 struct bio *tail;
263 struct bio *cur;
264 int again = 0;
265 unsigned long num_run;
266 unsigned long batch_run = 0;
267 unsigned long limit;
268 unsigned long last_waited = 0;
269 int force_reg = 0;
270 int sync_pending = 0;
271 struct blk_plug plug;
272
273 /*
274 * this function runs all the bios we've collected for
275 * a particular device. We don't want to wander off to
276 * another device without first sending all of these down.
277 * So, setup a plug here and finish it off before we return
278 */
279 blk_start_plug(&plug);
280
281 bdi = blk_get_backing_dev_info(device->bdev);
282 fs_info = device->dev_root->fs_info;
283 limit = btrfs_async_submit_limit(fs_info);
284 limit = limit * 2 / 3;
285
286 loop:
287 spin_lock(&device->io_lock);
288
289 loop_lock:
290 num_run = 0;
291
292 /* take all the bios off the list at once and process them
293 * later on (without the lock held). But, remember the
294 * tail and other pointers so the bios can be properly reinserted
295 * into the list if we hit congestion
296 */
297 if (!force_reg && device->pending_sync_bios.head) {
298 pending_bios = &device->pending_sync_bios;
299 force_reg = 1;
300 } else {
301 pending_bios = &device->pending_bios;
302 force_reg = 0;
303 }
304
305 pending = pending_bios->head;
306 tail = pending_bios->tail;
307 WARN_ON(pending && !tail);
308
309 /*
310 * if pending was null this time around, no bios need processing
311 * at all and we can stop. Otherwise it'll loop back up again
312 * and do an additional check so no bios are missed.
313 *
314 * device->running_pending is used to synchronize with the
315 * schedule_bio code.
316 */
317 if (device->pending_sync_bios.head == NULL &&
318 device->pending_bios.head == NULL) {
319 again = 0;
320 device->running_pending = 0;
321 } else {
322 again = 1;
323 device->running_pending = 1;
324 }
325
326 pending_bios->head = NULL;
327 pending_bios->tail = NULL;
328
329 spin_unlock(&device->io_lock);
330
331 while (pending) {
332
333 rmb();
334 /* we want to work on both lists, but do more bios on the
335 * sync list than the regular list
336 */
337 if ((num_run > 32 &&
338 pending_bios != &device->pending_sync_bios &&
339 device->pending_sync_bios.head) ||
340 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
341 device->pending_bios.head)) {
342 spin_lock(&device->io_lock);
343 requeue_list(pending_bios, pending, tail);
344 goto loop_lock;
345 }
346
347 cur = pending;
348 pending = pending->bi_next;
349 cur->bi_next = NULL;
350
351 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
352 waitqueue_active(&fs_info->async_submit_wait))
353 wake_up(&fs_info->async_submit_wait);
354
355 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
356
357 /*
358 * if we're doing the sync list, record that our
359 * plug has some sync requests on it
360 *
361 * If we're doing the regular list and there are
362 * sync requests sitting around, unplug before
363 * we add more
364 */
365 if (pending_bios == &device->pending_sync_bios) {
366 sync_pending = 1;
367 } else if (sync_pending) {
368 blk_finish_plug(&plug);
369 blk_start_plug(&plug);
370 sync_pending = 0;
371 }
372
373 btrfsic_submit_bio(cur->bi_rw, cur);
374 num_run++;
375 batch_run++;
376 if (need_resched())
377 cond_resched();
378
379 /*
380 * we made progress, there is more work to do and the bdi
381 * is now congested. Back off and let other work structs
382 * run instead
383 */
384 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
385 fs_info->fs_devices->open_devices > 1) {
386 struct io_context *ioc;
387
388 ioc = current->io_context;
389
390 /*
391 * the main goal here is that we don't want to
392 * block if we're going to be able to submit
393 * more requests without blocking.
394 *
395 * This code does two great things, it pokes into
396 * the elevator code from a filesystem _and_
397 * it makes assumptions about how batching works.
398 */
399 if (ioc && ioc->nr_batch_requests > 0 &&
400 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
401 (last_waited == 0 ||
402 ioc->last_waited == last_waited)) {
403 /*
404 * we want to go through our batch of
405 * requests and stop. So, we copy out
406 * the ioc->last_waited time and test
407 * against it before looping
408 */
409 last_waited = ioc->last_waited;
410 if (need_resched())
411 cond_resched();
412 continue;
413 }
414 spin_lock(&device->io_lock);
415 requeue_list(pending_bios, pending, tail);
416 device->running_pending = 1;
417
418 spin_unlock(&device->io_lock);
419 btrfs_requeue_work(&device->work);
420 goto done;
421 }
422 /* unplug every 64 requests just for good measure */
423 if (batch_run % 64 == 0) {
424 blk_finish_plug(&plug);
425 blk_start_plug(&plug);
426 sync_pending = 0;
427 }
428 }
429
430 cond_resched();
431 if (again)
432 goto loop;
433
434 spin_lock(&device->io_lock);
435 if (device->pending_bios.head || device->pending_sync_bios.head)
436 goto loop_lock;
437 spin_unlock(&device->io_lock);
438
439 done:
440 blk_finish_plug(&plug);
441 }
442
443 static void pending_bios_fn(struct btrfs_work *work)
444 {
445 struct btrfs_device *device;
446
447 device = container_of(work, struct btrfs_device, work);
448 run_scheduled_bios(device);
449 }
450
451 static noinline int device_list_add(const char *path,
452 struct btrfs_super_block *disk_super,
453 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
454 {
455 struct btrfs_device *device;
456 struct btrfs_fs_devices *fs_devices;
457 struct rcu_string *name;
458 u64 found_transid = btrfs_super_generation(disk_super);
459
460 fs_devices = find_fsid(disk_super->fsid);
461 if (!fs_devices) {
462 fs_devices = alloc_fs_devices(disk_super->fsid);
463 if (IS_ERR(fs_devices))
464 return PTR_ERR(fs_devices);
465
466 list_add(&fs_devices->list, &fs_uuids);
467 fs_devices->latest_devid = devid;
468 fs_devices->latest_trans = found_transid;
469
470 device = NULL;
471 } else {
472 device = __find_device(&fs_devices->devices, devid,
473 disk_super->dev_item.uuid);
474 }
475 if (!device) {
476 if (fs_devices->opened)
477 return -EBUSY;
478
479 device = btrfs_alloc_device(NULL, &devid,
480 disk_super->dev_item.uuid);
481 if (IS_ERR(device)) {
482 /* we can safely leave the fs_devices entry around */
483 return PTR_ERR(device);
484 }
485
486 name = rcu_string_strdup(path, GFP_NOFS);
487 if (!name) {
488 kfree(device);
489 return -ENOMEM;
490 }
491 rcu_assign_pointer(device->name, name);
492
493 mutex_lock(&fs_devices->device_list_mutex);
494 list_add_rcu(&device->dev_list, &fs_devices->devices);
495 mutex_unlock(&fs_devices->device_list_mutex);
496
497 device->fs_devices = fs_devices;
498 fs_devices->num_devices++;
499 } else if (!device->name || strcmp(device->name->str, path)) {
500 name = rcu_string_strdup(path, GFP_NOFS);
501 if (!name)
502 return -ENOMEM;
503 rcu_string_free(device->name);
504 rcu_assign_pointer(device->name, name);
505 if (device->missing) {
506 fs_devices->missing_devices--;
507 device->missing = 0;
508 }
509 }
510
511 if (found_transid > fs_devices->latest_trans) {
512 fs_devices->latest_devid = devid;
513 fs_devices->latest_trans = found_transid;
514 }
515 *fs_devices_ret = fs_devices;
516 return 0;
517 }
518
519 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
520 {
521 struct btrfs_fs_devices *fs_devices;
522 struct btrfs_device *device;
523 struct btrfs_device *orig_dev;
524
525 fs_devices = alloc_fs_devices(orig->fsid);
526 if (IS_ERR(fs_devices))
527 return fs_devices;
528
529 fs_devices->latest_devid = orig->latest_devid;
530 fs_devices->latest_trans = orig->latest_trans;
531 fs_devices->total_devices = orig->total_devices;
532
533 /* We have held the volume lock, it is safe to get the devices. */
534 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
535 struct rcu_string *name;
536
537 device = btrfs_alloc_device(NULL, &orig_dev->devid,
538 orig_dev->uuid);
539 if (IS_ERR(device))
540 goto error;
541
542 /*
543 * This is ok to do without rcu read locked because we hold the
544 * uuid mutex so nothing we touch in here is going to disappear.
545 */
546 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
547 if (!name) {
548 kfree(device);
549 goto error;
550 }
551 rcu_assign_pointer(device->name, name);
552
553 list_add(&device->dev_list, &fs_devices->devices);
554 device->fs_devices = fs_devices;
555 fs_devices->num_devices++;
556 }
557 return fs_devices;
558 error:
559 free_fs_devices(fs_devices);
560 return ERR_PTR(-ENOMEM);
561 }
562
563 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
564 struct btrfs_fs_devices *fs_devices, int step)
565 {
566 struct btrfs_device *device, *next;
567
568 struct block_device *latest_bdev = NULL;
569 u64 latest_devid = 0;
570 u64 latest_transid = 0;
571
572 mutex_lock(&uuid_mutex);
573 again:
574 /* This is the initialized path, it is safe to release the devices. */
575 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
576 if (device->in_fs_metadata) {
577 if (!device->is_tgtdev_for_dev_replace &&
578 (!latest_transid ||
579 device->generation > latest_transid)) {
580 latest_devid = device->devid;
581 latest_transid = device->generation;
582 latest_bdev = device->bdev;
583 }
584 continue;
585 }
586
587 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
588 /*
589 * In the first step, keep the device which has
590 * the correct fsid and the devid that is used
591 * for the dev_replace procedure.
592 * In the second step, the dev_replace state is
593 * read from the device tree and it is known
594 * whether the procedure is really active or
595 * not, which means whether this device is
596 * used or whether it should be removed.
597 */
598 if (step == 0 || device->is_tgtdev_for_dev_replace) {
599 continue;
600 }
601 }
602 if (device->bdev) {
603 blkdev_put(device->bdev, device->mode);
604 device->bdev = NULL;
605 fs_devices->open_devices--;
606 }
607 if (device->writeable) {
608 list_del_init(&device->dev_alloc_list);
609 device->writeable = 0;
610 if (!device->is_tgtdev_for_dev_replace)
611 fs_devices->rw_devices--;
612 }
613 list_del_init(&device->dev_list);
614 fs_devices->num_devices--;
615 rcu_string_free(device->name);
616 kfree(device);
617 }
618
619 if (fs_devices->seed) {
620 fs_devices = fs_devices->seed;
621 goto again;
622 }
623
624 fs_devices->latest_bdev = latest_bdev;
625 fs_devices->latest_devid = latest_devid;
626 fs_devices->latest_trans = latest_transid;
627
628 mutex_unlock(&uuid_mutex);
629 }
630
631 static void __free_device(struct work_struct *work)
632 {
633 struct btrfs_device *device;
634
635 device = container_of(work, struct btrfs_device, rcu_work);
636
637 if (device->bdev)
638 blkdev_put(device->bdev, device->mode);
639
640 rcu_string_free(device->name);
641 kfree(device);
642 }
643
644 static void free_device(struct rcu_head *head)
645 {
646 struct btrfs_device *device;
647
648 device = container_of(head, struct btrfs_device, rcu);
649
650 INIT_WORK(&device->rcu_work, __free_device);
651 schedule_work(&device->rcu_work);
652 }
653
654 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
655 {
656 struct btrfs_device *device;
657
658 if (--fs_devices->opened > 0)
659 return 0;
660
661 mutex_lock(&fs_devices->device_list_mutex);
662 list_for_each_entry(device, &fs_devices->devices, dev_list) {
663 struct btrfs_device *new_device;
664 struct rcu_string *name;
665
666 if (device->bdev)
667 fs_devices->open_devices--;
668
669 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
670 list_del_init(&device->dev_alloc_list);
671 fs_devices->rw_devices--;
672 }
673
674 if (device->can_discard)
675 fs_devices->num_can_discard--;
676
677 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
678 BUG_ON(!new_device); /* -ENOMEM */
679 memcpy(new_device, device, sizeof(*new_device));
680
681 /* Safe because we are under uuid_mutex */
682 if (device->name) {
683 name = rcu_string_strdup(device->name->str, GFP_NOFS);
684 BUG_ON(device->name && !name); /* -ENOMEM */
685 rcu_assign_pointer(new_device->name, name);
686 }
687 new_device->bdev = NULL;
688 new_device->writeable = 0;
689 new_device->in_fs_metadata = 0;
690 new_device->can_discard = 0;
691 spin_lock_init(&new_device->io_lock);
692 list_replace_rcu(&device->dev_list, &new_device->dev_list);
693
694 call_rcu(&device->rcu, free_device);
695 }
696 mutex_unlock(&fs_devices->device_list_mutex);
697
698 WARN_ON(fs_devices->open_devices);
699 WARN_ON(fs_devices->rw_devices);
700 fs_devices->opened = 0;
701 fs_devices->seeding = 0;
702
703 return 0;
704 }
705
706 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
707 {
708 struct btrfs_fs_devices *seed_devices = NULL;
709 int ret;
710
711 mutex_lock(&uuid_mutex);
712 ret = __btrfs_close_devices(fs_devices);
713 if (!fs_devices->opened) {
714 seed_devices = fs_devices->seed;
715 fs_devices->seed = NULL;
716 }
717 mutex_unlock(&uuid_mutex);
718
719 while (seed_devices) {
720 fs_devices = seed_devices;
721 seed_devices = fs_devices->seed;
722 __btrfs_close_devices(fs_devices);
723 free_fs_devices(fs_devices);
724 }
725 /*
726 * Wait for rcu kworkers under __btrfs_close_devices
727 * to finish all blkdev_puts so device is really
728 * free when umount is done.
729 */
730 rcu_barrier();
731 return ret;
732 }
733
734 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
735 fmode_t flags, void *holder)
736 {
737 struct request_queue *q;
738 struct block_device *bdev;
739 struct list_head *head = &fs_devices->devices;
740 struct btrfs_device *device;
741 struct block_device *latest_bdev = NULL;
742 struct buffer_head *bh;
743 struct btrfs_super_block *disk_super;
744 u64 latest_devid = 0;
745 u64 latest_transid = 0;
746 u64 devid;
747 int seeding = 1;
748 int ret = 0;
749
750 flags |= FMODE_EXCL;
751
752 list_for_each_entry(device, head, dev_list) {
753 if (device->bdev)
754 continue;
755 if (!device->name)
756 continue;
757
758 /* Just open everything we can; ignore failures here */
759 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
760 &bdev, &bh))
761 continue;
762
763 disk_super = (struct btrfs_super_block *)bh->b_data;
764 devid = btrfs_stack_device_id(&disk_super->dev_item);
765 if (devid != device->devid)
766 goto error_brelse;
767
768 if (memcmp(device->uuid, disk_super->dev_item.uuid,
769 BTRFS_UUID_SIZE))
770 goto error_brelse;
771
772 device->generation = btrfs_super_generation(disk_super);
773 if (!latest_transid || device->generation > latest_transid) {
774 latest_devid = devid;
775 latest_transid = device->generation;
776 latest_bdev = bdev;
777 }
778
779 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
780 device->writeable = 0;
781 } else {
782 device->writeable = !bdev_read_only(bdev);
783 seeding = 0;
784 }
785
786 q = bdev_get_queue(bdev);
787 if (blk_queue_discard(q)) {
788 device->can_discard = 1;
789 fs_devices->num_can_discard++;
790 }
791
792 device->bdev = bdev;
793 device->in_fs_metadata = 0;
794 device->mode = flags;
795
796 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
797 fs_devices->rotating = 1;
798
799 fs_devices->open_devices++;
800 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
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 "device label %s ", disk_super->label);
916 } else {
917 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
918 }
919
920 printk(KERN_CONT "devid %llu transid %llu %s\n",
921 (unsigned long long)devid, (unsigned long long)transid, path);
922
923 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
924 if (!ret && fs_devices_ret)
925 (*fs_devices_ret)->total_devices = total_devices;
926
927 error_unmap:
928 kunmap(page);
929 page_cache_release(page);
930
931 error_bdev_put:
932 blkdev_put(bdev, flags);
933 error:
934 mutex_unlock(&uuid_mutex);
935 return ret;
936 }
937
938 /* helper to account the used device space in the range */
939 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
940 u64 end, u64 *length)
941 {
942 struct btrfs_key key;
943 struct btrfs_root *root = device->dev_root;
944 struct btrfs_dev_extent *dev_extent;
945 struct btrfs_path *path;
946 u64 extent_end;
947 int ret;
948 int slot;
949 struct extent_buffer *l;
950
951 *length = 0;
952
953 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
954 return 0;
955
956 path = btrfs_alloc_path();
957 if (!path)
958 return -ENOMEM;
959 path->reada = 2;
960
961 key.objectid = device->devid;
962 key.offset = start;
963 key.type = BTRFS_DEV_EXTENT_KEY;
964
965 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
966 if (ret < 0)
967 goto out;
968 if (ret > 0) {
969 ret = btrfs_previous_item(root, path, key.objectid, key.type);
970 if (ret < 0)
971 goto out;
972 }
973
974 while (1) {
975 l = path->nodes[0];
976 slot = path->slots[0];
977 if (slot >= btrfs_header_nritems(l)) {
978 ret = btrfs_next_leaf(root, path);
979 if (ret == 0)
980 continue;
981 if (ret < 0)
982 goto out;
983
984 break;
985 }
986 btrfs_item_key_to_cpu(l, &key, slot);
987
988 if (key.objectid < device->devid)
989 goto next;
990
991 if (key.objectid > device->devid)
992 break;
993
994 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
995 goto next;
996
997 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
998 extent_end = key.offset + btrfs_dev_extent_length(l,
999 dev_extent);
1000 if (key.offset <= start && extent_end > end) {
1001 *length = end - start + 1;
1002 break;
1003 } else if (key.offset <= start && extent_end > start)
1004 *length += extent_end - start;
1005 else if (key.offset > start && extent_end <= end)
1006 *length += extent_end - key.offset;
1007 else if (key.offset > start && key.offset <= end) {
1008 *length += end - key.offset + 1;
1009 break;
1010 } else if (key.offset > end)
1011 break;
1012
1013 next:
1014 path->slots[0]++;
1015 }
1016 ret = 0;
1017 out:
1018 btrfs_free_path(path);
1019 return ret;
1020 }
1021
1022 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1023 struct btrfs_device *device,
1024 u64 *start, u64 len)
1025 {
1026 struct extent_map *em;
1027 int ret = 0;
1028
1029 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1030 struct map_lookup *map;
1031 int i;
1032
1033 map = (struct map_lookup *)em->bdev;
1034 for (i = 0; i < map->num_stripes; i++) {
1035 if (map->stripes[i].dev != device)
1036 continue;
1037 if (map->stripes[i].physical >= *start + len ||
1038 map->stripes[i].physical + em->orig_block_len <=
1039 *start)
1040 continue;
1041 *start = map->stripes[i].physical +
1042 em->orig_block_len;
1043 ret = 1;
1044 }
1045 }
1046
1047 return ret;
1048 }
1049
1050
1051 /*
1052 * find_free_dev_extent - find free space in the specified device
1053 * @device: the device which we search the free space in
1054 * @num_bytes: the size of the free space that we need
1055 * @start: store the start of the free space.
1056 * @len: the size of the free space. that we find, or the size of the max
1057 * free space if we don't find suitable free space
1058 *
1059 * this uses a pretty simple search, the expectation is that it is
1060 * called very infrequently and that a given device has a small number
1061 * of extents
1062 *
1063 * @start is used to store the start of the free space if we find. But if we
1064 * don't find suitable free space, it will be used to store the start position
1065 * of the max free space.
1066 *
1067 * @len is used to store the size of the free space that we find.
1068 * But if we don't find suitable free space, it is used to store the size of
1069 * the max free space.
1070 */
1071 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1072 struct btrfs_device *device, u64 num_bytes,
1073 u64 *start, u64 *len)
1074 {
1075 struct btrfs_key key;
1076 struct btrfs_root *root = device->dev_root;
1077 struct btrfs_dev_extent *dev_extent;
1078 struct btrfs_path *path;
1079 u64 hole_size;
1080 u64 max_hole_start;
1081 u64 max_hole_size;
1082 u64 extent_end;
1083 u64 search_start;
1084 u64 search_end = device->total_bytes;
1085 int ret;
1086 int slot;
1087 struct extent_buffer *l;
1088
1089 /* FIXME use last free of some kind */
1090
1091 /* we don't want to overwrite the superblock on the drive,
1092 * so we make sure to start at an offset of at least 1MB
1093 */
1094 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1095
1096 path = btrfs_alloc_path();
1097 if (!path)
1098 return -ENOMEM;
1099 again:
1100 max_hole_start = search_start;
1101 max_hole_size = 0;
1102 hole_size = 0;
1103
1104 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1105 ret = -ENOSPC;
1106 goto out;
1107 }
1108
1109 path->reada = 2;
1110 path->search_commit_root = 1;
1111 path->skip_locking = 1;
1112
1113 key.objectid = device->devid;
1114 key.offset = search_start;
1115 key.type = BTRFS_DEV_EXTENT_KEY;
1116
1117 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1118 if (ret < 0)
1119 goto out;
1120 if (ret > 0) {
1121 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1122 if (ret < 0)
1123 goto out;
1124 }
1125
1126 while (1) {
1127 l = path->nodes[0];
1128 slot = path->slots[0];
1129 if (slot >= btrfs_header_nritems(l)) {
1130 ret = btrfs_next_leaf(root, path);
1131 if (ret == 0)
1132 continue;
1133 if (ret < 0)
1134 goto out;
1135
1136 break;
1137 }
1138 btrfs_item_key_to_cpu(l, &key, slot);
1139
1140 if (key.objectid < device->devid)
1141 goto next;
1142
1143 if (key.objectid > device->devid)
1144 break;
1145
1146 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1147 goto next;
1148
1149 if (key.offset > search_start) {
1150 hole_size = key.offset - search_start;
1151
1152 /*
1153 * Have to check before we set max_hole_start, otherwise
1154 * we could end up sending back this offset anyway.
1155 */
1156 if (contains_pending_extent(trans, device,
1157 &search_start,
1158 hole_size))
1159 hole_size = 0;
1160
1161 if (hole_size > max_hole_size) {
1162 max_hole_start = search_start;
1163 max_hole_size = hole_size;
1164 }
1165
1166 /*
1167 * If this free space is greater than which we need,
1168 * it must be the max free space that we have found
1169 * until now, so max_hole_start must point to the start
1170 * of this free space and the length of this free space
1171 * is stored in max_hole_size. Thus, we return
1172 * max_hole_start and max_hole_size and go back to the
1173 * caller.
1174 */
1175 if (hole_size >= num_bytes) {
1176 ret = 0;
1177 goto out;
1178 }
1179 }
1180
1181 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1182 extent_end = key.offset + btrfs_dev_extent_length(l,
1183 dev_extent);
1184 if (extent_end > search_start)
1185 search_start = extent_end;
1186 next:
1187 path->slots[0]++;
1188 cond_resched();
1189 }
1190
1191 /*
1192 * At this point, search_start should be the end of
1193 * allocated dev extents, and when shrinking the device,
1194 * search_end may be smaller than search_start.
1195 */
1196 if (search_end > search_start)
1197 hole_size = search_end - search_start;
1198
1199 if (hole_size > max_hole_size) {
1200 max_hole_start = search_start;
1201 max_hole_size = hole_size;
1202 }
1203
1204 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1205 btrfs_release_path(path);
1206 goto again;
1207 }
1208
1209 /* See above. */
1210 if (hole_size < num_bytes)
1211 ret = -ENOSPC;
1212 else
1213 ret = 0;
1214
1215 out:
1216 btrfs_free_path(path);
1217 *start = max_hole_start;
1218 if (len)
1219 *len = max_hole_size;
1220 return ret;
1221 }
1222
1223 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1224 struct btrfs_device *device,
1225 u64 start)
1226 {
1227 int ret;
1228 struct btrfs_path *path;
1229 struct btrfs_root *root = device->dev_root;
1230 struct btrfs_key key;
1231 struct btrfs_key found_key;
1232 struct extent_buffer *leaf = NULL;
1233 struct btrfs_dev_extent *extent = NULL;
1234
1235 path = btrfs_alloc_path();
1236 if (!path)
1237 return -ENOMEM;
1238
1239 key.objectid = device->devid;
1240 key.offset = start;
1241 key.type = BTRFS_DEV_EXTENT_KEY;
1242 again:
1243 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1244 if (ret > 0) {
1245 ret = btrfs_previous_item(root, path, key.objectid,
1246 BTRFS_DEV_EXTENT_KEY);
1247 if (ret)
1248 goto out;
1249 leaf = path->nodes[0];
1250 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1251 extent = btrfs_item_ptr(leaf, path->slots[0],
1252 struct btrfs_dev_extent);
1253 BUG_ON(found_key.offset > start || found_key.offset +
1254 btrfs_dev_extent_length(leaf, extent) < start);
1255 key = found_key;
1256 btrfs_release_path(path);
1257 goto again;
1258 } else if (ret == 0) {
1259 leaf = path->nodes[0];
1260 extent = btrfs_item_ptr(leaf, path->slots[0],
1261 struct btrfs_dev_extent);
1262 } else {
1263 btrfs_error(root->fs_info, ret, "Slot search failed");
1264 goto out;
1265 }
1266
1267 if (device->bytes_used > 0) {
1268 u64 len = btrfs_dev_extent_length(leaf, extent);
1269 device->bytes_used -= len;
1270 spin_lock(&root->fs_info->free_chunk_lock);
1271 root->fs_info->free_chunk_space += len;
1272 spin_unlock(&root->fs_info->free_chunk_lock);
1273 }
1274 ret = btrfs_del_item(trans, root, path);
1275 if (ret) {
1276 btrfs_error(root->fs_info, ret,
1277 "Failed to remove dev extent item");
1278 }
1279 out:
1280 btrfs_free_path(path);
1281 return ret;
1282 }
1283
1284 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1285 struct btrfs_device *device,
1286 u64 chunk_tree, u64 chunk_objectid,
1287 u64 chunk_offset, u64 start, u64 num_bytes)
1288 {
1289 int ret;
1290 struct btrfs_path *path;
1291 struct btrfs_root *root = device->dev_root;
1292 struct btrfs_dev_extent *extent;
1293 struct extent_buffer *leaf;
1294 struct btrfs_key key;
1295
1296 WARN_ON(!device->in_fs_metadata);
1297 WARN_ON(device->is_tgtdev_for_dev_replace);
1298 path = btrfs_alloc_path();
1299 if (!path)
1300 return -ENOMEM;
1301
1302 key.objectid = device->devid;
1303 key.offset = start;
1304 key.type = BTRFS_DEV_EXTENT_KEY;
1305 ret = btrfs_insert_empty_item(trans, root, path, &key,
1306 sizeof(*extent));
1307 if (ret)
1308 goto out;
1309
1310 leaf = path->nodes[0];
1311 extent = btrfs_item_ptr(leaf, path->slots[0],
1312 struct btrfs_dev_extent);
1313 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1314 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1315 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1316
1317 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1318 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1319 BTRFS_UUID_SIZE);
1320
1321 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1322 btrfs_mark_buffer_dirty(leaf);
1323 out:
1324 btrfs_free_path(path);
1325 return ret;
1326 }
1327
1328 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1329 {
1330 struct extent_map_tree *em_tree;
1331 struct extent_map *em;
1332 struct rb_node *n;
1333 u64 ret = 0;
1334
1335 em_tree = &fs_info->mapping_tree.map_tree;
1336 read_lock(&em_tree->lock);
1337 n = rb_last(&em_tree->map);
1338 if (n) {
1339 em = rb_entry(n, struct extent_map, rb_node);
1340 ret = em->start + em->len;
1341 }
1342 read_unlock(&em_tree->lock);
1343
1344 return ret;
1345 }
1346
1347 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1348 u64 *devid_ret)
1349 {
1350 int ret;
1351 struct btrfs_key key;
1352 struct btrfs_key found_key;
1353 struct btrfs_path *path;
1354
1355 path = btrfs_alloc_path();
1356 if (!path)
1357 return -ENOMEM;
1358
1359 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1360 key.type = BTRFS_DEV_ITEM_KEY;
1361 key.offset = (u64)-1;
1362
1363 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1364 if (ret < 0)
1365 goto error;
1366
1367 BUG_ON(ret == 0); /* Corruption */
1368
1369 ret = btrfs_previous_item(fs_info->chunk_root, path,
1370 BTRFS_DEV_ITEMS_OBJECTID,
1371 BTRFS_DEV_ITEM_KEY);
1372 if (ret) {
1373 *devid_ret = 1;
1374 } else {
1375 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1376 path->slots[0]);
1377 *devid_ret = found_key.offset + 1;
1378 }
1379 ret = 0;
1380 error:
1381 btrfs_free_path(path);
1382 return ret;
1383 }
1384
1385 /*
1386 * the device information is stored in the chunk root
1387 * the btrfs_device struct should be fully filled in
1388 */
1389 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root,
1391 struct btrfs_device *device)
1392 {
1393 int ret;
1394 struct btrfs_path *path;
1395 struct btrfs_dev_item *dev_item;
1396 struct extent_buffer *leaf;
1397 struct btrfs_key key;
1398 unsigned long ptr;
1399
1400 root = root->fs_info->chunk_root;
1401
1402 path = btrfs_alloc_path();
1403 if (!path)
1404 return -ENOMEM;
1405
1406 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1407 key.type = BTRFS_DEV_ITEM_KEY;
1408 key.offset = device->devid;
1409
1410 ret = btrfs_insert_empty_item(trans, root, path, &key,
1411 sizeof(*dev_item));
1412 if (ret)
1413 goto out;
1414
1415 leaf = path->nodes[0];
1416 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1417
1418 btrfs_set_device_id(leaf, dev_item, device->devid);
1419 btrfs_set_device_generation(leaf, dev_item, 0);
1420 btrfs_set_device_type(leaf, dev_item, device->type);
1421 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1422 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1423 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1424 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1425 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1426 btrfs_set_device_group(leaf, dev_item, 0);
1427 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1428 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1429 btrfs_set_device_start_offset(leaf, dev_item, 0);
1430
1431 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1432 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1433 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1434 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1435 btrfs_mark_buffer_dirty(leaf);
1436
1437 ret = 0;
1438 out:
1439 btrfs_free_path(path);
1440 return ret;
1441 }
1442
1443 static int btrfs_rm_dev_item(struct btrfs_root *root,
1444 struct btrfs_device *device)
1445 {
1446 int ret;
1447 struct btrfs_path *path;
1448 struct btrfs_key key;
1449 struct btrfs_trans_handle *trans;
1450
1451 root = root->fs_info->chunk_root;
1452
1453 path = btrfs_alloc_path();
1454 if (!path)
1455 return -ENOMEM;
1456
1457 trans = btrfs_start_transaction(root, 0);
1458 if (IS_ERR(trans)) {
1459 btrfs_free_path(path);
1460 return PTR_ERR(trans);
1461 }
1462 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1463 key.type = BTRFS_DEV_ITEM_KEY;
1464 key.offset = device->devid;
1465 lock_chunks(root);
1466
1467 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1468 if (ret < 0)
1469 goto out;
1470
1471 if (ret > 0) {
1472 ret = -ENOENT;
1473 goto out;
1474 }
1475
1476 ret = btrfs_del_item(trans, root, path);
1477 if (ret)
1478 goto out;
1479 out:
1480 btrfs_free_path(path);
1481 unlock_chunks(root);
1482 btrfs_commit_transaction(trans, root);
1483 return ret;
1484 }
1485
1486 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1487 {
1488 struct btrfs_device *device;
1489 struct btrfs_device *next_device;
1490 struct block_device *bdev;
1491 struct buffer_head *bh = NULL;
1492 struct btrfs_super_block *disk_super;
1493 struct btrfs_fs_devices *cur_devices;
1494 u64 all_avail;
1495 u64 devid;
1496 u64 num_devices;
1497 u8 *dev_uuid;
1498 unsigned seq;
1499 int ret = 0;
1500 bool clear_super = false;
1501
1502 mutex_lock(&uuid_mutex);
1503
1504 do {
1505 seq = read_seqbegin(&root->fs_info->profiles_lock);
1506
1507 all_avail = root->fs_info->avail_data_alloc_bits |
1508 root->fs_info->avail_system_alloc_bits |
1509 root->fs_info->avail_metadata_alloc_bits;
1510 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1511
1512 num_devices = root->fs_info->fs_devices->num_devices;
1513 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1514 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1515 WARN_ON(num_devices < 1);
1516 num_devices--;
1517 }
1518 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1519
1520 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1521 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1522 goto out;
1523 }
1524
1525 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1526 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1527 goto out;
1528 }
1529
1530 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1531 root->fs_info->fs_devices->rw_devices <= 2) {
1532 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1533 goto out;
1534 }
1535 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1536 root->fs_info->fs_devices->rw_devices <= 3) {
1537 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1538 goto out;
1539 }
1540
1541 if (strcmp(device_path, "missing") == 0) {
1542 struct list_head *devices;
1543 struct btrfs_device *tmp;
1544
1545 device = NULL;
1546 devices = &root->fs_info->fs_devices->devices;
1547 /*
1548 * It is safe to read the devices since the volume_mutex
1549 * is held.
1550 */
1551 list_for_each_entry(tmp, devices, dev_list) {
1552 if (tmp->in_fs_metadata &&
1553 !tmp->is_tgtdev_for_dev_replace &&
1554 !tmp->bdev) {
1555 device = tmp;
1556 break;
1557 }
1558 }
1559 bdev = NULL;
1560 bh = NULL;
1561 disk_super = NULL;
1562 if (!device) {
1563 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1564 goto out;
1565 }
1566 } else {
1567 ret = btrfs_get_bdev_and_sb(device_path,
1568 FMODE_WRITE | FMODE_EXCL,
1569 root->fs_info->bdev_holder, 0,
1570 &bdev, &bh);
1571 if (ret)
1572 goto out;
1573 disk_super = (struct btrfs_super_block *)bh->b_data;
1574 devid = btrfs_stack_device_id(&disk_super->dev_item);
1575 dev_uuid = disk_super->dev_item.uuid;
1576 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1577 disk_super->fsid);
1578 if (!device) {
1579 ret = -ENOENT;
1580 goto error_brelse;
1581 }
1582 }
1583
1584 if (device->is_tgtdev_for_dev_replace) {
1585 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1586 goto error_brelse;
1587 }
1588
1589 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1590 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1591 goto error_brelse;
1592 }
1593
1594 if (device->writeable) {
1595 lock_chunks(root);
1596 list_del_init(&device->dev_alloc_list);
1597 unlock_chunks(root);
1598 root->fs_info->fs_devices->rw_devices--;
1599 clear_super = true;
1600 }
1601
1602 mutex_unlock(&uuid_mutex);
1603 ret = btrfs_shrink_device(device, 0);
1604 mutex_lock(&uuid_mutex);
1605 if (ret)
1606 goto error_undo;
1607
1608 /*
1609 * TODO: the superblock still includes this device in its num_devices
1610 * counter although write_all_supers() is not locked out. This
1611 * could give a filesystem state which requires a degraded mount.
1612 */
1613 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1614 if (ret)
1615 goto error_undo;
1616
1617 spin_lock(&root->fs_info->free_chunk_lock);
1618 root->fs_info->free_chunk_space = device->total_bytes -
1619 device->bytes_used;
1620 spin_unlock(&root->fs_info->free_chunk_lock);
1621
1622 device->in_fs_metadata = 0;
1623 btrfs_scrub_cancel_dev(root->fs_info, device);
1624
1625 /*
1626 * the device list mutex makes sure that we don't change
1627 * the device list while someone else is writing out all
1628 * the device supers.
1629 */
1630
1631 cur_devices = device->fs_devices;
1632 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1633 list_del_rcu(&device->dev_list);
1634
1635 device->fs_devices->num_devices--;
1636 device->fs_devices->total_devices--;
1637
1638 if (device->missing)
1639 root->fs_info->fs_devices->missing_devices--;
1640
1641 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1642 struct btrfs_device, dev_list);
1643 if (device->bdev == root->fs_info->sb->s_bdev)
1644 root->fs_info->sb->s_bdev = next_device->bdev;
1645 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1646 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1647
1648 if (device->bdev)
1649 device->fs_devices->open_devices--;
1650
1651 call_rcu(&device->rcu, free_device);
1652 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1653
1654 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1655 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1656
1657 if (cur_devices->open_devices == 0) {
1658 struct btrfs_fs_devices *fs_devices;
1659 fs_devices = root->fs_info->fs_devices;
1660 while (fs_devices) {
1661 if (fs_devices->seed == cur_devices)
1662 break;
1663 fs_devices = fs_devices->seed;
1664 }
1665 fs_devices->seed = cur_devices->seed;
1666 cur_devices->seed = NULL;
1667 lock_chunks(root);
1668 __btrfs_close_devices(cur_devices);
1669 unlock_chunks(root);
1670 free_fs_devices(cur_devices);
1671 }
1672
1673 root->fs_info->num_tolerated_disk_barrier_failures =
1674 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1675
1676 /*
1677 * at this point, the device is zero sized. We want to
1678 * remove it from the devices list and zero out the old super
1679 */
1680 if (clear_super && disk_super) {
1681 /* make sure this device isn't detected as part of
1682 * the FS anymore
1683 */
1684 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1685 set_buffer_dirty(bh);
1686 sync_dirty_buffer(bh);
1687 }
1688
1689 ret = 0;
1690
1691 /* Notify udev that device has changed */
1692 if (bdev)
1693 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1694
1695 error_brelse:
1696 brelse(bh);
1697 if (bdev)
1698 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1699 out:
1700 mutex_unlock(&uuid_mutex);
1701 return ret;
1702 error_undo:
1703 if (device->writeable) {
1704 lock_chunks(root);
1705 list_add(&device->dev_alloc_list,
1706 &root->fs_info->fs_devices->alloc_list);
1707 unlock_chunks(root);
1708 root->fs_info->fs_devices->rw_devices++;
1709 }
1710 goto error_brelse;
1711 }
1712
1713 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1714 struct btrfs_device *srcdev)
1715 {
1716 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1717 list_del_rcu(&srcdev->dev_list);
1718 list_del_rcu(&srcdev->dev_alloc_list);
1719 fs_info->fs_devices->num_devices--;
1720 if (srcdev->missing) {
1721 fs_info->fs_devices->missing_devices--;
1722 fs_info->fs_devices->rw_devices++;
1723 }
1724 if (srcdev->can_discard)
1725 fs_info->fs_devices->num_can_discard--;
1726 if (srcdev->bdev)
1727 fs_info->fs_devices->open_devices--;
1728
1729 call_rcu(&srcdev->rcu, free_device);
1730 }
1731
1732 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1733 struct btrfs_device *tgtdev)
1734 {
1735 struct btrfs_device *next_device;
1736
1737 WARN_ON(!tgtdev);
1738 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1739 if (tgtdev->bdev) {
1740 btrfs_scratch_superblock(tgtdev);
1741 fs_info->fs_devices->open_devices--;
1742 }
1743 fs_info->fs_devices->num_devices--;
1744 if (tgtdev->can_discard)
1745 fs_info->fs_devices->num_can_discard++;
1746
1747 next_device = list_entry(fs_info->fs_devices->devices.next,
1748 struct btrfs_device, dev_list);
1749 if (tgtdev->bdev == fs_info->sb->s_bdev)
1750 fs_info->sb->s_bdev = next_device->bdev;
1751 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1752 fs_info->fs_devices->latest_bdev = next_device->bdev;
1753 list_del_rcu(&tgtdev->dev_list);
1754
1755 call_rcu(&tgtdev->rcu, free_device);
1756
1757 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1758 }
1759
1760 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1761 struct btrfs_device **device)
1762 {
1763 int ret = 0;
1764 struct btrfs_super_block *disk_super;
1765 u64 devid;
1766 u8 *dev_uuid;
1767 struct block_device *bdev;
1768 struct buffer_head *bh;
1769
1770 *device = NULL;
1771 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1772 root->fs_info->bdev_holder, 0, &bdev, &bh);
1773 if (ret)
1774 return ret;
1775 disk_super = (struct btrfs_super_block *)bh->b_data;
1776 devid = btrfs_stack_device_id(&disk_super->dev_item);
1777 dev_uuid = disk_super->dev_item.uuid;
1778 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1779 disk_super->fsid);
1780 brelse(bh);
1781 if (!*device)
1782 ret = -ENOENT;
1783 blkdev_put(bdev, FMODE_READ);
1784 return ret;
1785 }
1786
1787 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1788 char *device_path,
1789 struct btrfs_device **device)
1790 {
1791 *device = NULL;
1792 if (strcmp(device_path, "missing") == 0) {
1793 struct list_head *devices;
1794 struct btrfs_device *tmp;
1795
1796 devices = &root->fs_info->fs_devices->devices;
1797 /*
1798 * It is safe to read the devices since the volume_mutex
1799 * is held by the caller.
1800 */
1801 list_for_each_entry(tmp, devices, dev_list) {
1802 if (tmp->in_fs_metadata && !tmp->bdev) {
1803 *device = tmp;
1804 break;
1805 }
1806 }
1807
1808 if (!*device) {
1809 pr_err("btrfs: no missing device found\n");
1810 return -ENOENT;
1811 }
1812
1813 return 0;
1814 } else {
1815 return btrfs_find_device_by_path(root, device_path, device);
1816 }
1817 }
1818
1819 /*
1820 * does all the dirty work required for changing file system's UUID.
1821 */
1822 static int btrfs_prepare_sprout(struct btrfs_root *root)
1823 {
1824 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1825 struct btrfs_fs_devices *old_devices;
1826 struct btrfs_fs_devices *seed_devices;
1827 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1828 struct btrfs_device *device;
1829 u64 super_flags;
1830
1831 BUG_ON(!mutex_is_locked(&uuid_mutex));
1832 if (!fs_devices->seeding)
1833 return -EINVAL;
1834
1835 seed_devices = __alloc_fs_devices();
1836 if (IS_ERR(seed_devices))
1837 return PTR_ERR(seed_devices);
1838
1839 old_devices = clone_fs_devices(fs_devices);
1840 if (IS_ERR(old_devices)) {
1841 kfree(seed_devices);
1842 return PTR_ERR(old_devices);
1843 }
1844
1845 list_add(&old_devices->list, &fs_uuids);
1846
1847 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1848 seed_devices->opened = 1;
1849 INIT_LIST_HEAD(&seed_devices->devices);
1850 INIT_LIST_HEAD(&seed_devices->alloc_list);
1851 mutex_init(&seed_devices->device_list_mutex);
1852
1853 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1854 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1855 synchronize_rcu);
1856 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1857
1858 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1859 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1860 device->fs_devices = seed_devices;
1861 }
1862
1863 fs_devices->seeding = 0;
1864 fs_devices->num_devices = 0;
1865 fs_devices->open_devices = 0;
1866 fs_devices->total_devices = 0;
1867 fs_devices->seed = seed_devices;
1868
1869 generate_random_uuid(fs_devices->fsid);
1870 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1871 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1872 super_flags = btrfs_super_flags(disk_super) &
1873 ~BTRFS_SUPER_FLAG_SEEDING;
1874 btrfs_set_super_flags(disk_super, super_flags);
1875
1876 return 0;
1877 }
1878
1879 /*
1880 * strore the expected generation for seed devices in device items.
1881 */
1882 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1883 struct btrfs_root *root)
1884 {
1885 struct btrfs_path *path;
1886 struct extent_buffer *leaf;
1887 struct btrfs_dev_item *dev_item;
1888 struct btrfs_device *device;
1889 struct btrfs_key key;
1890 u8 fs_uuid[BTRFS_UUID_SIZE];
1891 u8 dev_uuid[BTRFS_UUID_SIZE];
1892 u64 devid;
1893 int ret;
1894
1895 path = btrfs_alloc_path();
1896 if (!path)
1897 return -ENOMEM;
1898
1899 root = root->fs_info->chunk_root;
1900 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1901 key.offset = 0;
1902 key.type = BTRFS_DEV_ITEM_KEY;
1903
1904 while (1) {
1905 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1906 if (ret < 0)
1907 goto error;
1908
1909 leaf = path->nodes[0];
1910 next_slot:
1911 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1912 ret = btrfs_next_leaf(root, path);
1913 if (ret > 0)
1914 break;
1915 if (ret < 0)
1916 goto error;
1917 leaf = path->nodes[0];
1918 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1919 btrfs_release_path(path);
1920 continue;
1921 }
1922
1923 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1924 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1925 key.type != BTRFS_DEV_ITEM_KEY)
1926 break;
1927
1928 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1929 struct btrfs_dev_item);
1930 devid = btrfs_device_id(leaf, dev_item);
1931 read_extent_buffer(leaf, dev_uuid,
1932 (unsigned long)btrfs_device_uuid(dev_item),
1933 BTRFS_UUID_SIZE);
1934 read_extent_buffer(leaf, fs_uuid,
1935 (unsigned long)btrfs_device_fsid(dev_item),
1936 BTRFS_UUID_SIZE);
1937 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1938 fs_uuid);
1939 BUG_ON(!device); /* Logic error */
1940
1941 if (device->fs_devices->seeding) {
1942 btrfs_set_device_generation(leaf, dev_item,
1943 device->generation);
1944 btrfs_mark_buffer_dirty(leaf);
1945 }
1946
1947 path->slots[0]++;
1948 goto next_slot;
1949 }
1950 ret = 0;
1951 error:
1952 btrfs_free_path(path);
1953 return ret;
1954 }
1955
1956 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1957 {
1958 struct request_queue *q;
1959 struct btrfs_trans_handle *trans;
1960 struct btrfs_device *device;
1961 struct block_device *bdev;
1962 struct list_head *devices;
1963 struct super_block *sb = root->fs_info->sb;
1964 struct rcu_string *name;
1965 u64 total_bytes;
1966 int seeding_dev = 0;
1967 int ret = 0;
1968
1969 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1970 return -EROFS;
1971
1972 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1973 root->fs_info->bdev_holder);
1974 if (IS_ERR(bdev))
1975 return PTR_ERR(bdev);
1976
1977 if (root->fs_info->fs_devices->seeding) {
1978 seeding_dev = 1;
1979 down_write(&sb->s_umount);
1980 mutex_lock(&uuid_mutex);
1981 }
1982
1983 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1984
1985 devices = &root->fs_info->fs_devices->devices;
1986
1987 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1988 list_for_each_entry(device, devices, dev_list) {
1989 if (device->bdev == bdev) {
1990 ret = -EEXIST;
1991 mutex_unlock(
1992 &root->fs_info->fs_devices->device_list_mutex);
1993 goto error;
1994 }
1995 }
1996 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1997
1998 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
1999 if (IS_ERR(device)) {
2000 /* we can safely leave the fs_devices entry around */
2001 ret = PTR_ERR(device);
2002 goto error;
2003 }
2004
2005 name = rcu_string_strdup(device_path, GFP_NOFS);
2006 if (!name) {
2007 kfree(device);
2008 ret = -ENOMEM;
2009 goto error;
2010 }
2011 rcu_assign_pointer(device->name, name);
2012
2013 trans = btrfs_start_transaction(root, 0);
2014 if (IS_ERR(trans)) {
2015 rcu_string_free(device->name);
2016 kfree(device);
2017 ret = PTR_ERR(trans);
2018 goto error;
2019 }
2020
2021 lock_chunks(root);
2022
2023 q = bdev_get_queue(bdev);
2024 if (blk_queue_discard(q))
2025 device->can_discard = 1;
2026 device->writeable = 1;
2027 device->generation = trans->transid;
2028 device->io_width = root->sectorsize;
2029 device->io_align = root->sectorsize;
2030 device->sector_size = root->sectorsize;
2031 device->total_bytes = i_size_read(bdev->bd_inode);
2032 device->disk_total_bytes = device->total_bytes;
2033 device->dev_root = root->fs_info->dev_root;
2034 device->bdev = bdev;
2035 device->in_fs_metadata = 1;
2036 device->is_tgtdev_for_dev_replace = 0;
2037 device->mode = FMODE_EXCL;
2038 set_blocksize(device->bdev, 4096);
2039
2040 if (seeding_dev) {
2041 sb->s_flags &= ~MS_RDONLY;
2042 ret = btrfs_prepare_sprout(root);
2043 BUG_ON(ret); /* -ENOMEM */
2044 }
2045
2046 device->fs_devices = root->fs_info->fs_devices;
2047
2048 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2049 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2050 list_add(&device->dev_alloc_list,
2051 &root->fs_info->fs_devices->alloc_list);
2052 root->fs_info->fs_devices->num_devices++;
2053 root->fs_info->fs_devices->open_devices++;
2054 root->fs_info->fs_devices->rw_devices++;
2055 root->fs_info->fs_devices->total_devices++;
2056 if (device->can_discard)
2057 root->fs_info->fs_devices->num_can_discard++;
2058 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2059
2060 spin_lock(&root->fs_info->free_chunk_lock);
2061 root->fs_info->free_chunk_space += device->total_bytes;
2062 spin_unlock(&root->fs_info->free_chunk_lock);
2063
2064 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2065 root->fs_info->fs_devices->rotating = 1;
2066
2067 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2068 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2069 total_bytes + device->total_bytes);
2070
2071 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2072 btrfs_set_super_num_devices(root->fs_info->super_copy,
2073 total_bytes + 1);
2074 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2075
2076 if (seeding_dev) {
2077 ret = init_first_rw_device(trans, root, device);
2078 if (ret) {
2079 btrfs_abort_transaction(trans, root, ret);
2080 goto error_trans;
2081 }
2082 ret = btrfs_finish_sprout(trans, root);
2083 if (ret) {
2084 btrfs_abort_transaction(trans, root, ret);
2085 goto error_trans;
2086 }
2087 } else {
2088 ret = btrfs_add_device(trans, root, device);
2089 if (ret) {
2090 btrfs_abort_transaction(trans, root, ret);
2091 goto error_trans;
2092 }
2093 }
2094
2095 /*
2096 * we've got more storage, clear any full flags on the space
2097 * infos
2098 */
2099 btrfs_clear_space_info_full(root->fs_info);
2100
2101 unlock_chunks(root);
2102 root->fs_info->num_tolerated_disk_barrier_failures =
2103 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2104 ret = btrfs_commit_transaction(trans, root);
2105
2106 if (seeding_dev) {
2107 mutex_unlock(&uuid_mutex);
2108 up_write(&sb->s_umount);
2109
2110 if (ret) /* transaction commit */
2111 return ret;
2112
2113 ret = btrfs_relocate_sys_chunks(root);
2114 if (ret < 0)
2115 btrfs_error(root->fs_info, ret,
2116 "Failed to relocate sys chunks after "
2117 "device initialization. This can be fixed "
2118 "using the \"btrfs balance\" command.");
2119 trans = btrfs_attach_transaction(root);
2120 if (IS_ERR(trans)) {
2121 if (PTR_ERR(trans) == -ENOENT)
2122 return 0;
2123 return PTR_ERR(trans);
2124 }
2125 ret = btrfs_commit_transaction(trans, root);
2126 }
2127
2128 return ret;
2129
2130 error_trans:
2131 unlock_chunks(root);
2132 btrfs_end_transaction(trans, root);
2133 rcu_string_free(device->name);
2134 kfree(device);
2135 error:
2136 blkdev_put(bdev, FMODE_EXCL);
2137 if (seeding_dev) {
2138 mutex_unlock(&uuid_mutex);
2139 up_write(&sb->s_umount);
2140 }
2141 return ret;
2142 }
2143
2144 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2145 struct btrfs_device **device_out)
2146 {
2147 struct request_queue *q;
2148 struct btrfs_device *device;
2149 struct block_device *bdev;
2150 struct btrfs_fs_info *fs_info = root->fs_info;
2151 struct list_head *devices;
2152 struct rcu_string *name;
2153 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2154 int ret = 0;
2155
2156 *device_out = NULL;
2157 if (fs_info->fs_devices->seeding)
2158 return -EINVAL;
2159
2160 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2161 fs_info->bdev_holder);
2162 if (IS_ERR(bdev))
2163 return PTR_ERR(bdev);
2164
2165 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2166
2167 devices = &fs_info->fs_devices->devices;
2168 list_for_each_entry(device, devices, dev_list) {
2169 if (device->bdev == bdev) {
2170 ret = -EEXIST;
2171 goto error;
2172 }
2173 }
2174
2175 device = btrfs_alloc_device(NULL, &devid, NULL);
2176 if (IS_ERR(device)) {
2177 ret = PTR_ERR(device);
2178 goto error;
2179 }
2180
2181 name = rcu_string_strdup(device_path, GFP_NOFS);
2182 if (!name) {
2183 kfree(device);
2184 ret = -ENOMEM;
2185 goto error;
2186 }
2187 rcu_assign_pointer(device->name, name);
2188
2189 q = bdev_get_queue(bdev);
2190 if (blk_queue_discard(q))
2191 device->can_discard = 1;
2192 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2193 device->writeable = 1;
2194 device->generation = 0;
2195 device->io_width = root->sectorsize;
2196 device->io_align = root->sectorsize;
2197 device->sector_size = root->sectorsize;
2198 device->total_bytes = i_size_read(bdev->bd_inode);
2199 device->disk_total_bytes = device->total_bytes;
2200 device->dev_root = fs_info->dev_root;
2201 device->bdev = bdev;
2202 device->in_fs_metadata = 1;
2203 device->is_tgtdev_for_dev_replace = 1;
2204 device->mode = FMODE_EXCL;
2205 set_blocksize(device->bdev, 4096);
2206 device->fs_devices = fs_info->fs_devices;
2207 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2208 fs_info->fs_devices->num_devices++;
2209 fs_info->fs_devices->open_devices++;
2210 if (device->can_discard)
2211 fs_info->fs_devices->num_can_discard++;
2212 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2213
2214 *device_out = device;
2215 return ret;
2216
2217 error:
2218 blkdev_put(bdev, FMODE_EXCL);
2219 return ret;
2220 }
2221
2222 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2223 struct btrfs_device *tgtdev)
2224 {
2225 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2226 tgtdev->io_width = fs_info->dev_root->sectorsize;
2227 tgtdev->io_align = fs_info->dev_root->sectorsize;
2228 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2229 tgtdev->dev_root = fs_info->dev_root;
2230 tgtdev->in_fs_metadata = 1;
2231 }
2232
2233 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2234 struct btrfs_device *device)
2235 {
2236 int ret;
2237 struct btrfs_path *path;
2238 struct btrfs_root *root;
2239 struct btrfs_dev_item *dev_item;
2240 struct extent_buffer *leaf;
2241 struct btrfs_key key;
2242
2243 root = device->dev_root->fs_info->chunk_root;
2244
2245 path = btrfs_alloc_path();
2246 if (!path)
2247 return -ENOMEM;
2248
2249 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2250 key.type = BTRFS_DEV_ITEM_KEY;
2251 key.offset = device->devid;
2252
2253 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2254 if (ret < 0)
2255 goto out;
2256
2257 if (ret > 0) {
2258 ret = -ENOENT;
2259 goto out;
2260 }
2261
2262 leaf = path->nodes[0];
2263 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2264
2265 btrfs_set_device_id(leaf, dev_item, device->devid);
2266 btrfs_set_device_type(leaf, dev_item, device->type);
2267 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2268 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2269 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2270 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2271 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2272 btrfs_mark_buffer_dirty(leaf);
2273
2274 out:
2275 btrfs_free_path(path);
2276 return ret;
2277 }
2278
2279 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2280 struct btrfs_device *device, u64 new_size)
2281 {
2282 struct btrfs_super_block *super_copy =
2283 device->dev_root->fs_info->super_copy;
2284 u64 old_total = btrfs_super_total_bytes(super_copy);
2285 u64 diff = new_size - device->total_bytes;
2286
2287 if (!device->writeable)
2288 return -EACCES;
2289 if (new_size <= device->total_bytes ||
2290 device->is_tgtdev_for_dev_replace)
2291 return -EINVAL;
2292
2293 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2294 device->fs_devices->total_rw_bytes += diff;
2295
2296 device->total_bytes = new_size;
2297 device->disk_total_bytes = new_size;
2298 btrfs_clear_space_info_full(device->dev_root->fs_info);
2299
2300 return btrfs_update_device(trans, device);
2301 }
2302
2303 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2304 struct btrfs_device *device, u64 new_size)
2305 {
2306 int ret;
2307 lock_chunks(device->dev_root);
2308 ret = __btrfs_grow_device(trans, device, new_size);
2309 unlock_chunks(device->dev_root);
2310 return ret;
2311 }
2312
2313 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2314 struct btrfs_root *root,
2315 u64 chunk_tree, u64 chunk_objectid,
2316 u64 chunk_offset)
2317 {
2318 int ret;
2319 struct btrfs_path *path;
2320 struct btrfs_key key;
2321
2322 root = root->fs_info->chunk_root;
2323 path = btrfs_alloc_path();
2324 if (!path)
2325 return -ENOMEM;
2326
2327 key.objectid = chunk_objectid;
2328 key.offset = chunk_offset;
2329 key.type = BTRFS_CHUNK_ITEM_KEY;
2330
2331 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2332 if (ret < 0)
2333 goto out;
2334 else if (ret > 0) { /* Logic error or corruption */
2335 btrfs_error(root->fs_info, -ENOENT,
2336 "Failed lookup while freeing chunk.");
2337 ret = -ENOENT;
2338 goto out;
2339 }
2340
2341 ret = btrfs_del_item(trans, root, path);
2342 if (ret < 0)
2343 btrfs_error(root->fs_info, ret,
2344 "Failed to delete chunk item.");
2345 out:
2346 btrfs_free_path(path);
2347 return ret;
2348 }
2349
2350 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2351 chunk_offset)
2352 {
2353 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2354 struct btrfs_disk_key *disk_key;
2355 struct btrfs_chunk *chunk;
2356 u8 *ptr;
2357 int ret = 0;
2358 u32 num_stripes;
2359 u32 array_size;
2360 u32 len = 0;
2361 u32 cur;
2362 struct btrfs_key key;
2363
2364 array_size = btrfs_super_sys_array_size(super_copy);
2365
2366 ptr = super_copy->sys_chunk_array;
2367 cur = 0;
2368
2369 while (cur < array_size) {
2370 disk_key = (struct btrfs_disk_key *)ptr;
2371 btrfs_disk_key_to_cpu(&key, disk_key);
2372
2373 len = sizeof(*disk_key);
2374
2375 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2376 chunk = (struct btrfs_chunk *)(ptr + len);
2377 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2378 len += btrfs_chunk_item_size(num_stripes);
2379 } else {
2380 ret = -EIO;
2381 break;
2382 }
2383 if (key.objectid == chunk_objectid &&
2384 key.offset == chunk_offset) {
2385 memmove(ptr, ptr + len, array_size - (cur + len));
2386 array_size -= len;
2387 btrfs_set_super_sys_array_size(super_copy, array_size);
2388 } else {
2389 ptr += len;
2390 cur += len;
2391 }
2392 }
2393 return ret;
2394 }
2395
2396 static int btrfs_relocate_chunk(struct btrfs_root *root,
2397 u64 chunk_tree, u64 chunk_objectid,
2398 u64 chunk_offset)
2399 {
2400 struct extent_map_tree *em_tree;
2401 struct btrfs_root *extent_root;
2402 struct btrfs_trans_handle *trans;
2403 struct extent_map *em;
2404 struct map_lookup *map;
2405 int ret;
2406 int i;
2407
2408 root = root->fs_info->chunk_root;
2409 extent_root = root->fs_info->extent_root;
2410 em_tree = &root->fs_info->mapping_tree.map_tree;
2411
2412 ret = btrfs_can_relocate(extent_root, chunk_offset);
2413 if (ret)
2414 return -ENOSPC;
2415
2416 /* step one, relocate all the extents inside this chunk */
2417 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2418 if (ret)
2419 return ret;
2420
2421 trans = btrfs_start_transaction(root, 0);
2422 if (IS_ERR(trans)) {
2423 ret = PTR_ERR(trans);
2424 btrfs_std_error(root->fs_info, ret);
2425 return ret;
2426 }
2427
2428 lock_chunks(root);
2429
2430 /*
2431 * step two, delete the device extents and the
2432 * chunk tree entries
2433 */
2434 read_lock(&em_tree->lock);
2435 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2436 read_unlock(&em_tree->lock);
2437
2438 BUG_ON(!em || em->start > chunk_offset ||
2439 em->start + em->len < chunk_offset);
2440 map = (struct map_lookup *)em->bdev;
2441
2442 for (i = 0; i < map->num_stripes; i++) {
2443 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2444 map->stripes[i].physical);
2445 BUG_ON(ret);
2446
2447 if (map->stripes[i].dev) {
2448 ret = btrfs_update_device(trans, map->stripes[i].dev);
2449 BUG_ON(ret);
2450 }
2451 }
2452 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2453 chunk_offset);
2454
2455 BUG_ON(ret);
2456
2457 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2458
2459 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2460 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2461 BUG_ON(ret);
2462 }
2463
2464 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2465 BUG_ON(ret);
2466
2467 write_lock(&em_tree->lock);
2468 remove_extent_mapping(em_tree, em);
2469 write_unlock(&em_tree->lock);
2470
2471 kfree(map);
2472 em->bdev = NULL;
2473
2474 /* once for the tree */
2475 free_extent_map(em);
2476 /* once for us */
2477 free_extent_map(em);
2478
2479 unlock_chunks(root);
2480 btrfs_end_transaction(trans, root);
2481 return 0;
2482 }
2483
2484 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2485 {
2486 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2487 struct btrfs_path *path;
2488 struct extent_buffer *leaf;
2489 struct btrfs_chunk *chunk;
2490 struct btrfs_key key;
2491 struct btrfs_key found_key;
2492 u64 chunk_tree = chunk_root->root_key.objectid;
2493 u64 chunk_type;
2494 bool retried = false;
2495 int failed = 0;
2496 int ret;
2497
2498 path = btrfs_alloc_path();
2499 if (!path)
2500 return -ENOMEM;
2501
2502 again:
2503 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2504 key.offset = (u64)-1;
2505 key.type = BTRFS_CHUNK_ITEM_KEY;
2506
2507 while (1) {
2508 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2509 if (ret < 0)
2510 goto error;
2511 BUG_ON(ret == 0); /* Corruption */
2512
2513 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2514 key.type);
2515 if (ret < 0)
2516 goto error;
2517 if (ret > 0)
2518 break;
2519
2520 leaf = path->nodes[0];
2521 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2522
2523 chunk = btrfs_item_ptr(leaf, path->slots[0],
2524 struct btrfs_chunk);
2525 chunk_type = btrfs_chunk_type(leaf, chunk);
2526 btrfs_release_path(path);
2527
2528 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2529 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2530 found_key.objectid,
2531 found_key.offset);
2532 if (ret == -ENOSPC)
2533 failed++;
2534 else if (ret)
2535 BUG();
2536 }
2537
2538 if (found_key.offset == 0)
2539 break;
2540 key.offset = found_key.offset - 1;
2541 }
2542 ret = 0;
2543 if (failed && !retried) {
2544 failed = 0;
2545 retried = true;
2546 goto again;
2547 } else if (failed && retried) {
2548 WARN_ON(1);
2549 ret = -ENOSPC;
2550 }
2551 error:
2552 btrfs_free_path(path);
2553 return ret;
2554 }
2555
2556 static int insert_balance_item(struct btrfs_root *root,
2557 struct btrfs_balance_control *bctl)
2558 {
2559 struct btrfs_trans_handle *trans;
2560 struct btrfs_balance_item *item;
2561 struct btrfs_disk_balance_args disk_bargs;
2562 struct btrfs_path *path;
2563 struct extent_buffer *leaf;
2564 struct btrfs_key key;
2565 int ret, err;
2566
2567 path = btrfs_alloc_path();
2568 if (!path)
2569 return -ENOMEM;
2570
2571 trans = btrfs_start_transaction(root, 0);
2572 if (IS_ERR(trans)) {
2573 btrfs_free_path(path);
2574 return PTR_ERR(trans);
2575 }
2576
2577 key.objectid = BTRFS_BALANCE_OBJECTID;
2578 key.type = BTRFS_BALANCE_ITEM_KEY;
2579 key.offset = 0;
2580
2581 ret = btrfs_insert_empty_item(trans, root, path, &key,
2582 sizeof(*item));
2583 if (ret)
2584 goto out;
2585
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2588
2589 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2590
2591 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2592 btrfs_set_balance_data(leaf, item, &disk_bargs);
2593 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2594 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2595 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2596 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2597
2598 btrfs_set_balance_flags(leaf, item, bctl->flags);
2599
2600 btrfs_mark_buffer_dirty(leaf);
2601 out:
2602 btrfs_free_path(path);
2603 err = btrfs_commit_transaction(trans, root);
2604 if (err && !ret)
2605 ret = err;
2606 return ret;
2607 }
2608
2609 static int del_balance_item(struct btrfs_root *root)
2610 {
2611 struct btrfs_trans_handle *trans;
2612 struct btrfs_path *path;
2613 struct btrfs_key key;
2614 int ret, err;
2615
2616 path = btrfs_alloc_path();
2617 if (!path)
2618 return -ENOMEM;
2619
2620 trans = btrfs_start_transaction(root, 0);
2621 if (IS_ERR(trans)) {
2622 btrfs_free_path(path);
2623 return PTR_ERR(trans);
2624 }
2625
2626 key.objectid = BTRFS_BALANCE_OBJECTID;
2627 key.type = BTRFS_BALANCE_ITEM_KEY;
2628 key.offset = 0;
2629
2630 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2631 if (ret < 0)
2632 goto out;
2633 if (ret > 0) {
2634 ret = -ENOENT;
2635 goto out;
2636 }
2637
2638 ret = btrfs_del_item(trans, root, path);
2639 out:
2640 btrfs_free_path(path);
2641 err = btrfs_commit_transaction(trans, root);
2642 if (err && !ret)
2643 ret = err;
2644 return ret;
2645 }
2646
2647 /*
2648 * This is a heuristic used to reduce the number of chunks balanced on
2649 * resume after balance was interrupted.
2650 */
2651 static void update_balance_args(struct btrfs_balance_control *bctl)
2652 {
2653 /*
2654 * Turn on soft mode for chunk types that were being converted.
2655 */
2656 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2657 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2658 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2659 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2660 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2661 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2662
2663 /*
2664 * Turn on usage filter if is not already used. The idea is
2665 * that chunks that we have already balanced should be
2666 * reasonably full. Don't do it for chunks that are being
2667 * converted - that will keep us from relocating unconverted
2668 * (albeit full) chunks.
2669 */
2670 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2671 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2672 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2673 bctl->data.usage = 90;
2674 }
2675 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2676 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2677 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2678 bctl->sys.usage = 90;
2679 }
2680 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2681 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2682 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2683 bctl->meta.usage = 90;
2684 }
2685 }
2686
2687 /*
2688 * Should be called with both balance and volume mutexes held to
2689 * serialize other volume operations (add_dev/rm_dev/resize) with
2690 * restriper. Same goes for unset_balance_control.
2691 */
2692 static void set_balance_control(struct btrfs_balance_control *bctl)
2693 {
2694 struct btrfs_fs_info *fs_info = bctl->fs_info;
2695
2696 BUG_ON(fs_info->balance_ctl);
2697
2698 spin_lock(&fs_info->balance_lock);
2699 fs_info->balance_ctl = bctl;
2700 spin_unlock(&fs_info->balance_lock);
2701 }
2702
2703 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2704 {
2705 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2706
2707 BUG_ON(!fs_info->balance_ctl);
2708
2709 spin_lock(&fs_info->balance_lock);
2710 fs_info->balance_ctl = NULL;
2711 spin_unlock(&fs_info->balance_lock);
2712
2713 kfree(bctl);
2714 }
2715
2716 /*
2717 * Balance filters. Return 1 if chunk should be filtered out
2718 * (should not be balanced).
2719 */
2720 static int chunk_profiles_filter(u64 chunk_type,
2721 struct btrfs_balance_args *bargs)
2722 {
2723 chunk_type = chunk_to_extended(chunk_type) &
2724 BTRFS_EXTENDED_PROFILE_MASK;
2725
2726 if (bargs->profiles & chunk_type)
2727 return 0;
2728
2729 return 1;
2730 }
2731
2732 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2733 struct btrfs_balance_args *bargs)
2734 {
2735 struct btrfs_block_group_cache *cache;
2736 u64 chunk_used, user_thresh;
2737 int ret = 1;
2738
2739 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2740 chunk_used = btrfs_block_group_used(&cache->item);
2741
2742 if (bargs->usage == 0)
2743 user_thresh = 1;
2744 else if (bargs->usage > 100)
2745 user_thresh = cache->key.offset;
2746 else
2747 user_thresh = div_factor_fine(cache->key.offset,
2748 bargs->usage);
2749
2750 if (chunk_used < user_thresh)
2751 ret = 0;
2752
2753 btrfs_put_block_group(cache);
2754 return ret;
2755 }
2756
2757 static int chunk_devid_filter(struct extent_buffer *leaf,
2758 struct btrfs_chunk *chunk,
2759 struct btrfs_balance_args *bargs)
2760 {
2761 struct btrfs_stripe *stripe;
2762 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2763 int i;
2764
2765 for (i = 0; i < num_stripes; i++) {
2766 stripe = btrfs_stripe_nr(chunk, i);
2767 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2768 return 0;
2769 }
2770
2771 return 1;
2772 }
2773
2774 /* [pstart, pend) */
2775 static int chunk_drange_filter(struct extent_buffer *leaf,
2776 struct btrfs_chunk *chunk,
2777 u64 chunk_offset,
2778 struct btrfs_balance_args *bargs)
2779 {
2780 struct btrfs_stripe *stripe;
2781 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2782 u64 stripe_offset;
2783 u64 stripe_length;
2784 int factor;
2785 int i;
2786
2787 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2788 return 0;
2789
2790 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2791 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2792 factor = num_stripes / 2;
2793 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2794 factor = num_stripes - 1;
2795 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2796 factor = num_stripes - 2;
2797 } else {
2798 factor = num_stripes;
2799 }
2800
2801 for (i = 0; i < num_stripes; i++) {
2802 stripe = btrfs_stripe_nr(chunk, i);
2803 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2804 continue;
2805
2806 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2807 stripe_length = btrfs_chunk_length(leaf, chunk);
2808 do_div(stripe_length, factor);
2809
2810 if (stripe_offset < bargs->pend &&
2811 stripe_offset + stripe_length > bargs->pstart)
2812 return 0;
2813 }
2814
2815 return 1;
2816 }
2817
2818 /* [vstart, vend) */
2819 static int chunk_vrange_filter(struct extent_buffer *leaf,
2820 struct btrfs_chunk *chunk,
2821 u64 chunk_offset,
2822 struct btrfs_balance_args *bargs)
2823 {
2824 if (chunk_offset < bargs->vend &&
2825 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2826 /* at least part of the chunk is inside this vrange */
2827 return 0;
2828
2829 return 1;
2830 }
2831
2832 static int chunk_soft_convert_filter(u64 chunk_type,
2833 struct btrfs_balance_args *bargs)
2834 {
2835 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2836 return 0;
2837
2838 chunk_type = chunk_to_extended(chunk_type) &
2839 BTRFS_EXTENDED_PROFILE_MASK;
2840
2841 if (bargs->target == chunk_type)
2842 return 1;
2843
2844 return 0;
2845 }
2846
2847 static int should_balance_chunk(struct btrfs_root *root,
2848 struct extent_buffer *leaf,
2849 struct btrfs_chunk *chunk, u64 chunk_offset)
2850 {
2851 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2852 struct btrfs_balance_args *bargs = NULL;
2853 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2854
2855 /* type filter */
2856 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2857 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2858 return 0;
2859 }
2860
2861 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2862 bargs = &bctl->data;
2863 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2864 bargs = &bctl->sys;
2865 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2866 bargs = &bctl->meta;
2867
2868 /* profiles filter */
2869 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2870 chunk_profiles_filter(chunk_type, bargs)) {
2871 return 0;
2872 }
2873
2874 /* usage filter */
2875 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2876 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2877 return 0;
2878 }
2879
2880 /* devid filter */
2881 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2882 chunk_devid_filter(leaf, chunk, bargs)) {
2883 return 0;
2884 }
2885
2886 /* drange filter, makes sense only with devid filter */
2887 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2888 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2889 return 0;
2890 }
2891
2892 /* vrange filter */
2893 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2894 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2895 return 0;
2896 }
2897
2898 /* soft profile changing mode */
2899 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2900 chunk_soft_convert_filter(chunk_type, bargs)) {
2901 return 0;
2902 }
2903
2904 return 1;
2905 }
2906
2907 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2908 {
2909 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2910 struct btrfs_root *chunk_root = fs_info->chunk_root;
2911 struct btrfs_root *dev_root = fs_info->dev_root;
2912 struct list_head *devices;
2913 struct btrfs_device *device;
2914 u64 old_size;
2915 u64 size_to_free;
2916 struct btrfs_chunk *chunk;
2917 struct btrfs_path *path;
2918 struct btrfs_key key;
2919 struct btrfs_key found_key;
2920 struct btrfs_trans_handle *trans;
2921 struct extent_buffer *leaf;
2922 int slot;
2923 int ret;
2924 int enospc_errors = 0;
2925 bool counting = true;
2926
2927 /* step one make some room on all the devices */
2928 devices = &fs_info->fs_devices->devices;
2929 list_for_each_entry(device, devices, dev_list) {
2930 old_size = device->total_bytes;
2931 size_to_free = div_factor(old_size, 1);
2932 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2933 if (!device->writeable ||
2934 device->total_bytes - device->bytes_used > size_to_free ||
2935 device->is_tgtdev_for_dev_replace)
2936 continue;
2937
2938 ret = btrfs_shrink_device(device, old_size - size_to_free);
2939 if (ret == -ENOSPC)
2940 break;
2941 BUG_ON(ret);
2942
2943 trans = btrfs_start_transaction(dev_root, 0);
2944 BUG_ON(IS_ERR(trans));
2945
2946 ret = btrfs_grow_device(trans, device, old_size);
2947 BUG_ON(ret);
2948
2949 btrfs_end_transaction(trans, dev_root);
2950 }
2951
2952 /* step two, relocate all the chunks */
2953 path = btrfs_alloc_path();
2954 if (!path) {
2955 ret = -ENOMEM;
2956 goto error;
2957 }
2958
2959 /* zero out stat counters */
2960 spin_lock(&fs_info->balance_lock);
2961 memset(&bctl->stat, 0, sizeof(bctl->stat));
2962 spin_unlock(&fs_info->balance_lock);
2963 again:
2964 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2965 key.offset = (u64)-1;
2966 key.type = BTRFS_CHUNK_ITEM_KEY;
2967
2968 while (1) {
2969 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2970 atomic_read(&fs_info->balance_cancel_req)) {
2971 ret = -ECANCELED;
2972 goto error;
2973 }
2974
2975 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2976 if (ret < 0)
2977 goto error;
2978
2979 /*
2980 * this shouldn't happen, it means the last relocate
2981 * failed
2982 */
2983 if (ret == 0)
2984 BUG(); /* FIXME break ? */
2985
2986 ret = btrfs_previous_item(chunk_root, path, 0,
2987 BTRFS_CHUNK_ITEM_KEY);
2988 if (ret) {
2989 ret = 0;
2990 break;
2991 }
2992
2993 leaf = path->nodes[0];
2994 slot = path->slots[0];
2995 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2996
2997 if (found_key.objectid != key.objectid)
2998 break;
2999
3000 /* chunk zero is special */
3001 if (found_key.offset == 0)
3002 break;
3003
3004 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3005
3006 if (!counting) {
3007 spin_lock(&fs_info->balance_lock);
3008 bctl->stat.considered++;
3009 spin_unlock(&fs_info->balance_lock);
3010 }
3011
3012 ret = should_balance_chunk(chunk_root, leaf, chunk,
3013 found_key.offset);
3014 btrfs_release_path(path);
3015 if (!ret)
3016 goto loop;
3017
3018 if (counting) {
3019 spin_lock(&fs_info->balance_lock);
3020 bctl->stat.expected++;
3021 spin_unlock(&fs_info->balance_lock);
3022 goto loop;
3023 }
3024
3025 ret = btrfs_relocate_chunk(chunk_root,
3026 chunk_root->root_key.objectid,
3027 found_key.objectid,
3028 found_key.offset);
3029 if (ret && ret != -ENOSPC)
3030 goto error;
3031 if (ret == -ENOSPC) {
3032 enospc_errors++;
3033 } else {
3034 spin_lock(&fs_info->balance_lock);
3035 bctl->stat.completed++;
3036 spin_unlock(&fs_info->balance_lock);
3037 }
3038 loop:
3039 key.offset = found_key.offset - 1;
3040 }
3041
3042 if (counting) {
3043 btrfs_release_path(path);
3044 counting = false;
3045 goto again;
3046 }
3047 error:
3048 btrfs_free_path(path);
3049 if (enospc_errors) {
3050 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3051 enospc_errors);
3052 if (!ret)
3053 ret = -ENOSPC;
3054 }
3055
3056 return ret;
3057 }
3058
3059 /**
3060 * alloc_profile_is_valid - see if a given profile is valid and reduced
3061 * @flags: profile to validate
3062 * @extended: if true @flags is treated as an extended profile
3063 */
3064 static int alloc_profile_is_valid(u64 flags, int extended)
3065 {
3066 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3067 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3068
3069 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3070
3071 /* 1) check that all other bits are zeroed */
3072 if (flags & ~mask)
3073 return 0;
3074
3075 /* 2) see if profile is reduced */
3076 if (flags == 0)
3077 return !extended; /* "0" is valid for usual profiles */
3078
3079 /* true if exactly one bit set */
3080 return (flags & (flags - 1)) == 0;
3081 }
3082
3083 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3084 {
3085 /* cancel requested || normal exit path */
3086 return atomic_read(&fs_info->balance_cancel_req) ||
3087 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3088 atomic_read(&fs_info->balance_cancel_req) == 0);
3089 }
3090
3091 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3092 {
3093 int ret;
3094
3095 unset_balance_control(fs_info);
3096 ret = del_balance_item(fs_info->tree_root);
3097 if (ret)
3098 btrfs_std_error(fs_info, ret);
3099
3100 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3101 }
3102
3103 /*
3104 * Should be called with both balance and volume mutexes held
3105 */
3106 int btrfs_balance(struct btrfs_balance_control *bctl,
3107 struct btrfs_ioctl_balance_args *bargs)
3108 {
3109 struct btrfs_fs_info *fs_info = bctl->fs_info;
3110 u64 allowed;
3111 int mixed = 0;
3112 int ret;
3113 u64 num_devices;
3114 unsigned seq;
3115
3116 if (btrfs_fs_closing(fs_info) ||
3117 atomic_read(&fs_info->balance_pause_req) ||
3118 atomic_read(&fs_info->balance_cancel_req)) {
3119 ret = -EINVAL;
3120 goto out;
3121 }
3122
3123 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3124 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3125 mixed = 1;
3126
3127 /*
3128 * In case of mixed groups both data and meta should be picked,
3129 * and identical options should be given for both of them.
3130 */
3131 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3132 if (mixed && (bctl->flags & allowed)) {
3133 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3134 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3135 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3136 printk(KERN_ERR "btrfs: with mixed groups data and "
3137 "metadata balance options must be the same\n");
3138 ret = -EINVAL;
3139 goto out;
3140 }
3141 }
3142
3143 num_devices = fs_info->fs_devices->num_devices;
3144 btrfs_dev_replace_lock(&fs_info->dev_replace);
3145 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3146 BUG_ON(num_devices < 1);
3147 num_devices--;
3148 }
3149 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3150 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3151 if (num_devices == 1)
3152 allowed |= BTRFS_BLOCK_GROUP_DUP;
3153 else if (num_devices > 1)
3154 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3155 if (num_devices > 2)
3156 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3157 if (num_devices > 3)
3158 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3159 BTRFS_BLOCK_GROUP_RAID6);
3160 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3161 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3162 (bctl->data.target & ~allowed))) {
3163 printk(KERN_ERR "btrfs: unable to start balance with target "
3164 "data profile %llu\n",
3165 (unsigned long long)bctl->data.target);
3166 ret = -EINVAL;
3167 goto out;
3168 }
3169 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3170 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3171 (bctl->meta.target & ~allowed))) {
3172 printk(KERN_ERR "btrfs: unable to start balance with target "
3173 "metadata profile %llu\n",
3174 (unsigned long long)bctl->meta.target);
3175 ret = -EINVAL;
3176 goto out;
3177 }
3178 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3179 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3180 (bctl->sys.target & ~allowed))) {
3181 printk(KERN_ERR "btrfs: unable to start balance with target "
3182 "system profile %llu\n",
3183 (unsigned long long)bctl->sys.target);
3184 ret = -EINVAL;
3185 goto out;
3186 }
3187
3188 /* allow dup'ed data chunks only in mixed mode */
3189 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3190 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3191 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3192 ret = -EINVAL;
3193 goto out;
3194 }
3195
3196 /* allow to reduce meta or sys integrity only if force set */
3197 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3198 BTRFS_BLOCK_GROUP_RAID10 |
3199 BTRFS_BLOCK_GROUP_RAID5 |
3200 BTRFS_BLOCK_GROUP_RAID6;
3201 do {
3202 seq = read_seqbegin(&fs_info->profiles_lock);
3203
3204 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3205 (fs_info->avail_system_alloc_bits & allowed) &&
3206 !(bctl->sys.target & allowed)) ||
3207 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3208 (fs_info->avail_metadata_alloc_bits & allowed) &&
3209 !(bctl->meta.target & allowed))) {
3210 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3211 printk(KERN_INFO "btrfs: force reducing metadata "
3212 "integrity\n");
3213 } else {
3214 printk(KERN_ERR "btrfs: balance will reduce metadata "
3215 "integrity, use force if you want this\n");
3216 ret = -EINVAL;
3217 goto out;
3218 }
3219 }
3220 } while (read_seqretry(&fs_info->profiles_lock, seq));
3221
3222 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3223 int num_tolerated_disk_barrier_failures;
3224 u64 target = bctl->sys.target;
3225
3226 num_tolerated_disk_barrier_failures =
3227 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3228 if (num_tolerated_disk_barrier_failures > 0 &&
3229 (target &
3230 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3231 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3232 num_tolerated_disk_barrier_failures = 0;
3233 else if (num_tolerated_disk_barrier_failures > 1 &&
3234 (target &
3235 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3236 num_tolerated_disk_barrier_failures = 1;
3237
3238 fs_info->num_tolerated_disk_barrier_failures =
3239 num_tolerated_disk_barrier_failures;
3240 }
3241
3242 ret = insert_balance_item(fs_info->tree_root, bctl);
3243 if (ret && ret != -EEXIST)
3244 goto out;
3245
3246 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3247 BUG_ON(ret == -EEXIST);
3248 set_balance_control(bctl);
3249 } else {
3250 BUG_ON(ret != -EEXIST);
3251 spin_lock(&fs_info->balance_lock);
3252 update_balance_args(bctl);
3253 spin_unlock(&fs_info->balance_lock);
3254 }
3255
3256 atomic_inc(&fs_info->balance_running);
3257 mutex_unlock(&fs_info->balance_mutex);
3258
3259 ret = __btrfs_balance(fs_info);
3260
3261 mutex_lock(&fs_info->balance_mutex);
3262 atomic_dec(&fs_info->balance_running);
3263
3264 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3265 fs_info->num_tolerated_disk_barrier_failures =
3266 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3267 }
3268
3269 if (bargs) {
3270 memset(bargs, 0, sizeof(*bargs));
3271 update_ioctl_balance_args(fs_info, 0, bargs);
3272 }
3273
3274 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3275 balance_need_close(fs_info)) {
3276 __cancel_balance(fs_info);
3277 }
3278
3279 wake_up(&fs_info->balance_wait_q);
3280
3281 return ret;
3282 out:
3283 if (bctl->flags & BTRFS_BALANCE_RESUME)
3284 __cancel_balance(fs_info);
3285 else {
3286 kfree(bctl);
3287 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3288 }
3289 return ret;
3290 }
3291
3292 static int balance_kthread(void *data)
3293 {
3294 struct btrfs_fs_info *fs_info = data;
3295 int ret = 0;
3296
3297 mutex_lock(&fs_info->volume_mutex);
3298 mutex_lock(&fs_info->balance_mutex);
3299
3300 if (fs_info->balance_ctl) {
3301 printk(KERN_INFO "btrfs: continuing balance\n");
3302 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3303 }
3304
3305 mutex_unlock(&fs_info->balance_mutex);
3306 mutex_unlock(&fs_info->volume_mutex);
3307
3308 return ret;
3309 }
3310
3311 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3312 {
3313 struct task_struct *tsk;
3314
3315 spin_lock(&fs_info->balance_lock);
3316 if (!fs_info->balance_ctl) {
3317 spin_unlock(&fs_info->balance_lock);
3318 return 0;
3319 }
3320 spin_unlock(&fs_info->balance_lock);
3321
3322 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3323 printk(KERN_INFO "btrfs: force skipping balance\n");
3324 return 0;
3325 }
3326
3327 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3328 return PTR_RET(tsk);
3329 }
3330
3331 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3332 {
3333 struct btrfs_balance_control *bctl;
3334 struct btrfs_balance_item *item;
3335 struct btrfs_disk_balance_args disk_bargs;
3336 struct btrfs_path *path;
3337 struct extent_buffer *leaf;
3338 struct btrfs_key key;
3339 int ret;
3340
3341 path = btrfs_alloc_path();
3342 if (!path)
3343 return -ENOMEM;
3344
3345 key.objectid = BTRFS_BALANCE_OBJECTID;
3346 key.type = BTRFS_BALANCE_ITEM_KEY;
3347 key.offset = 0;
3348
3349 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3350 if (ret < 0)
3351 goto out;
3352 if (ret > 0) { /* ret = -ENOENT; */
3353 ret = 0;
3354 goto out;
3355 }
3356
3357 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3358 if (!bctl) {
3359 ret = -ENOMEM;
3360 goto out;
3361 }
3362
3363 leaf = path->nodes[0];
3364 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3365
3366 bctl->fs_info = fs_info;
3367 bctl->flags = btrfs_balance_flags(leaf, item);
3368 bctl->flags |= BTRFS_BALANCE_RESUME;
3369
3370 btrfs_balance_data(leaf, item, &disk_bargs);
3371 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3372 btrfs_balance_meta(leaf, item, &disk_bargs);
3373 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3374 btrfs_balance_sys(leaf, item, &disk_bargs);
3375 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3376
3377 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3378
3379 mutex_lock(&fs_info->volume_mutex);
3380 mutex_lock(&fs_info->balance_mutex);
3381
3382 set_balance_control(bctl);
3383
3384 mutex_unlock(&fs_info->balance_mutex);
3385 mutex_unlock(&fs_info->volume_mutex);
3386 out:
3387 btrfs_free_path(path);
3388 return ret;
3389 }
3390
3391 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3392 {
3393 int ret = 0;
3394
3395 mutex_lock(&fs_info->balance_mutex);
3396 if (!fs_info->balance_ctl) {
3397 mutex_unlock(&fs_info->balance_mutex);
3398 return -ENOTCONN;
3399 }
3400
3401 if (atomic_read(&fs_info->balance_running)) {
3402 atomic_inc(&fs_info->balance_pause_req);
3403 mutex_unlock(&fs_info->balance_mutex);
3404
3405 wait_event(fs_info->balance_wait_q,
3406 atomic_read(&fs_info->balance_running) == 0);
3407
3408 mutex_lock(&fs_info->balance_mutex);
3409 /* we are good with balance_ctl ripped off from under us */
3410 BUG_ON(atomic_read(&fs_info->balance_running));
3411 atomic_dec(&fs_info->balance_pause_req);
3412 } else {
3413 ret = -ENOTCONN;
3414 }
3415
3416 mutex_unlock(&fs_info->balance_mutex);
3417 return ret;
3418 }
3419
3420 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3421 {
3422 mutex_lock(&fs_info->balance_mutex);
3423 if (!fs_info->balance_ctl) {
3424 mutex_unlock(&fs_info->balance_mutex);
3425 return -ENOTCONN;
3426 }
3427
3428 atomic_inc(&fs_info->balance_cancel_req);
3429 /*
3430 * if we are running just wait and return, balance item is
3431 * deleted in btrfs_balance in this case
3432 */
3433 if (atomic_read(&fs_info->balance_running)) {
3434 mutex_unlock(&fs_info->balance_mutex);
3435 wait_event(fs_info->balance_wait_q,
3436 atomic_read(&fs_info->balance_running) == 0);
3437 mutex_lock(&fs_info->balance_mutex);
3438 } else {
3439 /* __cancel_balance needs volume_mutex */
3440 mutex_unlock(&fs_info->balance_mutex);
3441 mutex_lock(&fs_info->volume_mutex);
3442 mutex_lock(&fs_info->balance_mutex);
3443
3444 if (fs_info->balance_ctl)
3445 __cancel_balance(fs_info);
3446
3447 mutex_unlock(&fs_info->volume_mutex);
3448 }
3449
3450 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3451 atomic_dec(&fs_info->balance_cancel_req);
3452 mutex_unlock(&fs_info->balance_mutex);
3453 return 0;
3454 }
3455
3456 static int btrfs_uuid_scan_kthread(void *data)
3457 {
3458 struct btrfs_fs_info *fs_info = data;
3459 struct btrfs_root *root = fs_info->tree_root;
3460 struct btrfs_key key;
3461 struct btrfs_key max_key;
3462 struct btrfs_path *path = NULL;
3463 int ret = 0;
3464 struct extent_buffer *eb;
3465 int slot;
3466 struct btrfs_root_item root_item;
3467 u32 item_size;
3468 struct btrfs_trans_handle *trans;
3469
3470 path = btrfs_alloc_path();
3471 if (!path) {
3472 ret = -ENOMEM;
3473 goto out;
3474 }
3475
3476 key.objectid = 0;
3477 key.type = BTRFS_ROOT_ITEM_KEY;
3478 key.offset = 0;
3479
3480 max_key.objectid = (u64)-1;
3481 max_key.type = BTRFS_ROOT_ITEM_KEY;
3482 max_key.offset = (u64)-1;
3483
3484 path->keep_locks = 1;
3485
3486 while (1) {
3487 ret = btrfs_search_forward(root, &key, &max_key, path, 0);
3488 if (ret) {
3489 if (ret > 0)
3490 ret = 0;
3491 break;
3492 }
3493
3494 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3495 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3496 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3497 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3498 goto skip;
3499
3500 eb = path->nodes[0];
3501 slot = path->slots[0];
3502 item_size = btrfs_item_size_nr(eb, slot);
3503 if (item_size < sizeof(root_item))
3504 goto skip;
3505
3506 trans = NULL;
3507 read_extent_buffer(eb, &root_item,
3508 btrfs_item_ptr_offset(eb, slot),
3509 (int)sizeof(root_item));
3510 if (btrfs_root_refs(&root_item) == 0)
3511 goto skip;
3512 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3513 /*
3514 * 1 - subvol uuid item
3515 * 1 - received_subvol uuid item
3516 */
3517 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3518 if (IS_ERR(trans)) {
3519 ret = PTR_ERR(trans);
3520 break;
3521 }
3522 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3523 root_item.uuid,
3524 BTRFS_UUID_KEY_SUBVOL,
3525 key.objectid);
3526 if (ret < 0) {
3527 pr_warn("btrfs: uuid_tree_add failed %d\n",
3528 ret);
3529 btrfs_end_transaction(trans,
3530 fs_info->uuid_root);
3531 break;
3532 }
3533 }
3534
3535 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3536 if (!trans) {
3537 /* 1 - received_subvol uuid item */
3538 trans = btrfs_start_transaction(
3539 fs_info->uuid_root, 1);
3540 if (IS_ERR(trans)) {
3541 ret = PTR_ERR(trans);
3542 break;
3543 }
3544 }
3545 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3546 root_item.received_uuid,
3547 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3548 key.objectid);
3549 if (ret < 0) {
3550 pr_warn("btrfs: uuid_tree_add failed %d\n",
3551 ret);
3552 btrfs_end_transaction(trans,
3553 fs_info->uuid_root);
3554 break;
3555 }
3556 }
3557
3558 if (trans) {
3559 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3560 if (ret)
3561 break;
3562 }
3563
3564 skip:
3565 btrfs_release_path(path);
3566 if (key.offset < (u64)-1) {
3567 key.offset++;
3568 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3569 key.offset = 0;
3570 key.type = BTRFS_ROOT_ITEM_KEY;
3571 } else if (key.objectid < (u64)-1) {
3572 key.offset = 0;
3573 key.type = BTRFS_ROOT_ITEM_KEY;
3574 key.objectid++;
3575 } else {
3576 break;
3577 }
3578 cond_resched();
3579 }
3580
3581 out:
3582 btrfs_free_path(path);
3583 if (ret)
3584 pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret);
3585 else
3586 fs_info->update_uuid_tree_gen = 1;
3587 up(&fs_info->uuid_tree_rescan_sem);
3588 return 0;
3589 }
3590
3591 /*
3592 * Callback for btrfs_uuid_tree_iterate().
3593 * returns:
3594 * 0 check succeeded, the entry is not outdated.
3595 * < 0 if an error occured.
3596 * > 0 if the check failed, which means the caller shall remove the entry.
3597 */
3598 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3599 u8 *uuid, u8 type, u64 subid)
3600 {
3601 struct btrfs_key key;
3602 int ret = 0;
3603 struct btrfs_root *subvol_root;
3604
3605 if (type != BTRFS_UUID_KEY_SUBVOL &&
3606 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3607 goto out;
3608
3609 key.objectid = subid;
3610 key.type = BTRFS_ROOT_ITEM_KEY;
3611 key.offset = (u64)-1;
3612 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3613 if (IS_ERR(subvol_root)) {
3614 ret = PTR_ERR(subvol_root);
3615 if (ret == -ENOENT)
3616 ret = 1;
3617 goto out;
3618 }
3619
3620 switch (type) {
3621 case BTRFS_UUID_KEY_SUBVOL:
3622 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3623 ret = 1;
3624 break;
3625 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3626 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3627 BTRFS_UUID_SIZE))
3628 ret = 1;
3629 break;
3630 }
3631
3632 out:
3633 return ret;
3634 }
3635
3636 static int btrfs_uuid_rescan_kthread(void *data)
3637 {
3638 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3639 int ret;
3640
3641 /*
3642 * 1st step is to iterate through the existing UUID tree and
3643 * to delete all entries that contain outdated data.
3644 * 2nd step is to add all missing entries to the UUID tree.
3645 */
3646 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3647 if (ret < 0) {
3648 pr_warn("btrfs: iterating uuid_tree failed %d\n", ret);
3649 up(&fs_info->uuid_tree_rescan_sem);
3650 return ret;
3651 }
3652 return btrfs_uuid_scan_kthread(data);
3653 }
3654
3655 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3656 {
3657 struct btrfs_trans_handle *trans;
3658 struct btrfs_root *tree_root = fs_info->tree_root;
3659 struct btrfs_root *uuid_root;
3660 struct task_struct *task;
3661 int ret;
3662
3663 /*
3664 * 1 - root node
3665 * 1 - root item
3666 */
3667 trans = btrfs_start_transaction(tree_root, 2);
3668 if (IS_ERR(trans))
3669 return PTR_ERR(trans);
3670
3671 uuid_root = btrfs_create_tree(trans, fs_info,
3672 BTRFS_UUID_TREE_OBJECTID);
3673 if (IS_ERR(uuid_root)) {
3674 btrfs_abort_transaction(trans, tree_root,
3675 PTR_ERR(uuid_root));
3676 return PTR_ERR(uuid_root);
3677 }
3678
3679 fs_info->uuid_root = uuid_root;
3680
3681 ret = btrfs_commit_transaction(trans, tree_root);
3682 if (ret)
3683 return ret;
3684
3685 down(&fs_info->uuid_tree_rescan_sem);
3686 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3687 if (IS_ERR(task)) {
3688 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3689 pr_warn("btrfs: failed to start uuid_scan task\n");
3690 up(&fs_info->uuid_tree_rescan_sem);
3691 return PTR_ERR(task);
3692 }
3693
3694 return 0;
3695 }
3696
3697 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3698 {
3699 struct task_struct *task;
3700
3701 down(&fs_info->uuid_tree_rescan_sem);
3702 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3703 if (IS_ERR(task)) {
3704 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3705 pr_warn("btrfs: failed to start uuid_rescan task\n");
3706 up(&fs_info->uuid_tree_rescan_sem);
3707 return PTR_ERR(task);
3708 }
3709
3710 return 0;
3711 }
3712
3713 /*
3714 * shrinking a device means finding all of the device extents past
3715 * the new size, and then following the back refs to the chunks.
3716 * The chunk relocation code actually frees the device extent
3717 */
3718 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3719 {
3720 struct btrfs_trans_handle *trans;
3721 struct btrfs_root *root = device->dev_root;
3722 struct btrfs_dev_extent *dev_extent = NULL;
3723 struct btrfs_path *path;
3724 u64 length;
3725 u64 chunk_tree;
3726 u64 chunk_objectid;
3727 u64 chunk_offset;
3728 int ret;
3729 int slot;
3730 int failed = 0;
3731 bool retried = false;
3732 struct extent_buffer *l;
3733 struct btrfs_key key;
3734 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3735 u64 old_total = btrfs_super_total_bytes(super_copy);
3736 u64 old_size = device->total_bytes;
3737 u64 diff = device->total_bytes - new_size;
3738
3739 if (device->is_tgtdev_for_dev_replace)
3740 return -EINVAL;
3741
3742 path = btrfs_alloc_path();
3743 if (!path)
3744 return -ENOMEM;
3745
3746 path->reada = 2;
3747
3748 lock_chunks(root);
3749
3750 device->total_bytes = new_size;
3751 if (device->writeable) {
3752 device->fs_devices->total_rw_bytes -= diff;
3753 spin_lock(&root->fs_info->free_chunk_lock);
3754 root->fs_info->free_chunk_space -= diff;
3755 spin_unlock(&root->fs_info->free_chunk_lock);
3756 }
3757 unlock_chunks(root);
3758
3759 again:
3760 key.objectid = device->devid;
3761 key.offset = (u64)-1;
3762 key.type = BTRFS_DEV_EXTENT_KEY;
3763
3764 do {
3765 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3766 if (ret < 0)
3767 goto done;
3768
3769 ret = btrfs_previous_item(root, path, 0, key.type);
3770 if (ret < 0)
3771 goto done;
3772 if (ret) {
3773 ret = 0;
3774 btrfs_release_path(path);
3775 break;
3776 }
3777
3778 l = path->nodes[0];
3779 slot = path->slots[0];
3780 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3781
3782 if (key.objectid != device->devid) {
3783 btrfs_release_path(path);
3784 break;
3785 }
3786
3787 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3788 length = btrfs_dev_extent_length(l, dev_extent);
3789
3790 if (key.offset + length <= new_size) {
3791 btrfs_release_path(path);
3792 break;
3793 }
3794
3795 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3796 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3797 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3798 btrfs_release_path(path);
3799
3800 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3801 chunk_offset);
3802 if (ret && ret != -ENOSPC)
3803 goto done;
3804 if (ret == -ENOSPC)
3805 failed++;
3806 } while (key.offset-- > 0);
3807
3808 if (failed && !retried) {
3809 failed = 0;
3810 retried = true;
3811 goto again;
3812 } else if (failed && retried) {
3813 ret = -ENOSPC;
3814 lock_chunks(root);
3815
3816 device->total_bytes = old_size;
3817 if (device->writeable)
3818 device->fs_devices->total_rw_bytes += diff;
3819 spin_lock(&root->fs_info->free_chunk_lock);
3820 root->fs_info->free_chunk_space += diff;
3821 spin_unlock(&root->fs_info->free_chunk_lock);
3822 unlock_chunks(root);
3823 goto done;
3824 }
3825
3826 /* Shrinking succeeded, else we would be at "done". */
3827 trans = btrfs_start_transaction(root, 0);
3828 if (IS_ERR(trans)) {
3829 ret = PTR_ERR(trans);
3830 goto done;
3831 }
3832
3833 lock_chunks(root);
3834
3835 device->disk_total_bytes = new_size;
3836 /* Now btrfs_update_device() will change the on-disk size. */
3837 ret = btrfs_update_device(trans, device);
3838 if (ret) {
3839 unlock_chunks(root);
3840 btrfs_end_transaction(trans, root);
3841 goto done;
3842 }
3843 WARN_ON(diff > old_total);
3844 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3845 unlock_chunks(root);
3846 btrfs_end_transaction(trans, root);
3847 done:
3848 btrfs_free_path(path);
3849 return ret;
3850 }
3851
3852 static int btrfs_add_system_chunk(struct btrfs_root *root,
3853 struct btrfs_key *key,
3854 struct btrfs_chunk *chunk, int item_size)
3855 {
3856 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3857 struct btrfs_disk_key disk_key;
3858 u32 array_size;
3859 u8 *ptr;
3860
3861 array_size = btrfs_super_sys_array_size(super_copy);
3862 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3863 return -EFBIG;
3864
3865 ptr = super_copy->sys_chunk_array + array_size;
3866 btrfs_cpu_key_to_disk(&disk_key, key);
3867 memcpy(ptr, &disk_key, sizeof(disk_key));
3868 ptr += sizeof(disk_key);
3869 memcpy(ptr, chunk, item_size);
3870 item_size += sizeof(disk_key);
3871 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3872 return 0;
3873 }
3874
3875 /*
3876 * sort the devices in descending order by max_avail, total_avail
3877 */
3878 static int btrfs_cmp_device_info(const void *a, const void *b)
3879 {
3880 const struct btrfs_device_info *di_a = a;
3881 const struct btrfs_device_info *di_b = b;
3882
3883 if (di_a->max_avail > di_b->max_avail)
3884 return -1;
3885 if (di_a->max_avail < di_b->max_avail)
3886 return 1;
3887 if (di_a->total_avail > di_b->total_avail)
3888 return -1;
3889 if (di_a->total_avail < di_b->total_avail)
3890 return 1;
3891 return 0;
3892 }
3893
3894 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3895 [BTRFS_RAID_RAID10] = {
3896 .sub_stripes = 2,
3897 .dev_stripes = 1,
3898 .devs_max = 0, /* 0 == as many as possible */
3899 .devs_min = 4,
3900 .devs_increment = 2,
3901 .ncopies = 2,
3902 },
3903 [BTRFS_RAID_RAID1] = {
3904 .sub_stripes = 1,
3905 .dev_stripes = 1,
3906 .devs_max = 2,
3907 .devs_min = 2,
3908 .devs_increment = 2,
3909 .ncopies = 2,
3910 },
3911 [BTRFS_RAID_DUP] = {
3912 .sub_stripes = 1,
3913 .dev_stripes = 2,
3914 .devs_max = 1,
3915 .devs_min = 1,
3916 .devs_increment = 1,
3917 .ncopies = 2,
3918 },
3919 [BTRFS_RAID_RAID0] = {
3920 .sub_stripes = 1,
3921 .dev_stripes = 1,
3922 .devs_max = 0,
3923 .devs_min = 2,
3924 .devs_increment = 1,
3925 .ncopies = 1,
3926 },
3927 [BTRFS_RAID_SINGLE] = {
3928 .sub_stripes = 1,
3929 .dev_stripes = 1,
3930 .devs_max = 1,
3931 .devs_min = 1,
3932 .devs_increment = 1,
3933 .ncopies = 1,
3934 },
3935 [BTRFS_RAID_RAID5] = {
3936 .sub_stripes = 1,
3937 .dev_stripes = 1,
3938 .devs_max = 0,
3939 .devs_min = 2,
3940 .devs_increment = 1,
3941 .ncopies = 2,
3942 },
3943 [BTRFS_RAID_RAID6] = {
3944 .sub_stripes = 1,
3945 .dev_stripes = 1,
3946 .devs_max = 0,
3947 .devs_min = 3,
3948 .devs_increment = 1,
3949 .ncopies = 3,
3950 },
3951 };
3952
3953 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3954 {
3955 /* TODO allow them to set a preferred stripe size */
3956 return 64 * 1024;
3957 }
3958
3959 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3960 {
3961 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3962 return;
3963
3964 btrfs_set_fs_incompat(info, RAID56);
3965 }
3966
3967 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3968 struct btrfs_root *extent_root, u64 start,
3969 u64 type)
3970 {
3971 struct btrfs_fs_info *info = extent_root->fs_info;
3972 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3973 struct list_head *cur;
3974 struct map_lookup *map = NULL;
3975 struct extent_map_tree *em_tree;
3976 struct extent_map *em;
3977 struct btrfs_device_info *devices_info = NULL;
3978 u64 total_avail;
3979 int num_stripes; /* total number of stripes to allocate */
3980 int data_stripes; /* number of stripes that count for
3981 block group size */
3982 int sub_stripes; /* sub_stripes info for map */
3983 int dev_stripes; /* stripes per dev */
3984 int devs_max; /* max devs to use */
3985 int devs_min; /* min devs needed */
3986 int devs_increment; /* ndevs has to be a multiple of this */
3987 int ncopies; /* how many copies to data has */
3988 int ret;
3989 u64 max_stripe_size;
3990 u64 max_chunk_size;
3991 u64 stripe_size;
3992 u64 num_bytes;
3993 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3994 int ndevs;
3995 int i;
3996 int j;
3997 int index;
3998
3999 BUG_ON(!alloc_profile_is_valid(type, 0));
4000
4001 if (list_empty(&fs_devices->alloc_list))
4002 return -ENOSPC;
4003
4004 index = __get_raid_index(type);
4005
4006 sub_stripes = btrfs_raid_array[index].sub_stripes;
4007 dev_stripes = btrfs_raid_array[index].dev_stripes;
4008 devs_max = btrfs_raid_array[index].devs_max;
4009 devs_min = btrfs_raid_array[index].devs_min;
4010 devs_increment = btrfs_raid_array[index].devs_increment;
4011 ncopies = btrfs_raid_array[index].ncopies;
4012
4013 if (type & BTRFS_BLOCK_GROUP_DATA) {
4014 max_stripe_size = 1024 * 1024 * 1024;
4015 max_chunk_size = 10 * max_stripe_size;
4016 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4017 /* for larger filesystems, use larger metadata chunks */
4018 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4019 max_stripe_size = 1024 * 1024 * 1024;
4020 else
4021 max_stripe_size = 256 * 1024 * 1024;
4022 max_chunk_size = max_stripe_size;
4023 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4024 max_stripe_size = 32 * 1024 * 1024;
4025 max_chunk_size = 2 * max_stripe_size;
4026 } else {
4027 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
4028 type);
4029 BUG_ON(1);
4030 }
4031
4032 /* we don't want a chunk larger than 10% of writeable space */
4033 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4034 max_chunk_size);
4035
4036 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4037 GFP_NOFS);
4038 if (!devices_info)
4039 return -ENOMEM;
4040
4041 cur = fs_devices->alloc_list.next;
4042
4043 /*
4044 * in the first pass through the devices list, we gather information
4045 * about the available holes on each device.
4046 */
4047 ndevs = 0;
4048 while (cur != &fs_devices->alloc_list) {
4049 struct btrfs_device *device;
4050 u64 max_avail;
4051 u64 dev_offset;
4052
4053 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4054
4055 cur = cur->next;
4056
4057 if (!device->writeable) {
4058 WARN(1, KERN_ERR
4059 "btrfs: read-only device in alloc_list\n");
4060 continue;
4061 }
4062
4063 if (!device->in_fs_metadata ||
4064 device->is_tgtdev_for_dev_replace)
4065 continue;
4066
4067 if (device->total_bytes > device->bytes_used)
4068 total_avail = device->total_bytes - device->bytes_used;
4069 else
4070 total_avail = 0;
4071
4072 /* If there is no space on this device, skip it. */
4073 if (total_avail == 0)
4074 continue;
4075
4076 ret = find_free_dev_extent(trans, device,
4077 max_stripe_size * dev_stripes,
4078 &dev_offset, &max_avail);
4079 if (ret && ret != -ENOSPC)
4080 goto error;
4081
4082 if (ret == 0)
4083 max_avail = max_stripe_size * dev_stripes;
4084
4085 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4086 continue;
4087
4088 if (ndevs == fs_devices->rw_devices) {
4089 WARN(1, "%s: found more than %llu devices\n",
4090 __func__, fs_devices->rw_devices);
4091 break;
4092 }
4093 devices_info[ndevs].dev_offset = dev_offset;
4094 devices_info[ndevs].max_avail = max_avail;
4095 devices_info[ndevs].total_avail = total_avail;
4096 devices_info[ndevs].dev = device;
4097 ++ndevs;
4098 }
4099
4100 /*
4101 * now sort the devices by hole size / available space
4102 */
4103 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4104 btrfs_cmp_device_info, NULL);
4105
4106 /* round down to number of usable stripes */
4107 ndevs -= ndevs % devs_increment;
4108
4109 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4110 ret = -ENOSPC;
4111 goto error;
4112 }
4113
4114 if (devs_max && ndevs > devs_max)
4115 ndevs = devs_max;
4116 /*
4117 * the primary goal is to maximize the number of stripes, so use as many
4118 * devices as possible, even if the stripes are not maximum sized.
4119 */
4120 stripe_size = devices_info[ndevs-1].max_avail;
4121 num_stripes = ndevs * dev_stripes;
4122
4123 /*
4124 * this will have to be fixed for RAID1 and RAID10 over
4125 * more drives
4126 */
4127 data_stripes = num_stripes / ncopies;
4128
4129 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4130 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4131 btrfs_super_stripesize(info->super_copy));
4132 data_stripes = num_stripes - 1;
4133 }
4134 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4135 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4136 btrfs_super_stripesize(info->super_copy));
4137 data_stripes = num_stripes - 2;
4138 }
4139
4140 /*
4141 * Use the number of data stripes to figure out how big this chunk
4142 * is really going to be in terms of logical address space,
4143 * and compare that answer with the max chunk size
4144 */
4145 if (stripe_size * data_stripes > max_chunk_size) {
4146 u64 mask = (1ULL << 24) - 1;
4147 stripe_size = max_chunk_size;
4148 do_div(stripe_size, data_stripes);
4149
4150 /* bump the answer up to a 16MB boundary */
4151 stripe_size = (stripe_size + mask) & ~mask;
4152
4153 /* but don't go higher than the limits we found
4154 * while searching for free extents
4155 */
4156 if (stripe_size > devices_info[ndevs-1].max_avail)
4157 stripe_size = devices_info[ndevs-1].max_avail;
4158 }
4159
4160 do_div(stripe_size, dev_stripes);
4161
4162 /* align to BTRFS_STRIPE_LEN */
4163 do_div(stripe_size, raid_stripe_len);
4164 stripe_size *= raid_stripe_len;
4165
4166 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4167 if (!map) {
4168 ret = -ENOMEM;
4169 goto error;
4170 }
4171 map->num_stripes = num_stripes;
4172
4173 for (i = 0; i < ndevs; ++i) {
4174 for (j = 0; j < dev_stripes; ++j) {
4175 int s = i * dev_stripes + j;
4176 map->stripes[s].dev = devices_info[i].dev;
4177 map->stripes[s].physical = devices_info[i].dev_offset +
4178 j * stripe_size;
4179 }
4180 }
4181 map->sector_size = extent_root->sectorsize;
4182 map->stripe_len = raid_stripe_len;
4183 map->io_align = raid_stripe_len;
4184 map->io_width = raid_stripe_len;
4185 map->type = type;
4186 map->sub_stripes = sub_stripes;
4187
4188 num_bytes = stripe_size * data_stripes;
4189
4190 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4191
4192 em = alloc_extent_map();
4193 if (!em) {
4194 ret = -ENOMEM;
4195 goto error;
4196 }
4197 em->bdev = (struct block_device *)map;
4198 em->start = start;
4199 em->len = num_bytes;
4200 em->block_start = 0;
4201 em->block_len = em->len;
4202 em->orig_block_len = stripe_size;
4203
4204 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4205 write_lock(&em_tree->lock);
4206 ret = add_extent_mapping(em_tree, em, 0);
4207 if (!ret) {
4208 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4209 atomic_inc(&em->refs);
4210 }
4211 write_unlock(&em_tree->lock);
4212 if (ret) {
4213 free_extent_map(em);
4214 goto error;
4215 }
4216
4217 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4218 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4219 start, num_bytes);
4220 if (ret)
4221 goto error_del_extent;
4222
4223 free_extent_map(em);
4224 check_raid56_incompat_flag(extent_root->fs_info, type);
4225
4226 kfree(devices_info);
4227 return 0;
4228
4229 error_del_extent:
4230 write_lock(&em_tree->lock);
4231 remove_extent_mapping(em_tree, em);
4232 write_unlock(&em_tree->lock);
4233
4234 /* One for our allocation */
4235 free_extent_map(em);
4236 /* One for the tree reference */
4237 free_extent_map(em);
4238 error:
4239 kfree(map);
4240 kfree(devices_info);
4241 return ret;
4242 }
4243
4244 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4245 struct btrfs_root *extent_root,
4246 u64 chunk_offset, u64 chunk_size)
4247 {
4248 struct btrfs_key key;
4249 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4250 struct btrfs_device *device;
4251 struct btrfs_chunk *chunk;
4252 struct btrfs_stripe *stripe;
4253 struct extent_map_tree *em_tree;
4254 struct extent_map *em;
4255 struct map_lookup *map;
4256 size_t item_size;
4257 u64 dev_offset;
4258 u64 stripe_size;
4259 int i = 0;
4260 int ret;
4261
4262 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4263 read_lock(&em_tree->lock);
4264 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4265 read_unlock(&em_tree->lock);
4266
4267 if (!em) {
4268 btrfs_crit(extent_root->fs_info, "unable to find logical "
4269 "%Lu len %Lu", chunk_offset, chunk_size);
4270 return -EINVAL;
4271 }
4272
4273 if (em->start != chunk_offset || em->len != chunk_size) {
4274 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4275 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
4276 chunk_size, em->start, em->len);
4277 free_extent_map(em);
4278 return -EINVAL;
4279 }
4280
4281 map = (struct map_lookup *)em->bdev;
4282 item_size = btrfs_chunk_item_size(map->num_stripes);
4283 stripe_size = em->orig_block_len;
4284
4285 chunk = kzalloc(item_size, GFP_NOFS);
4286 if (!chunk) {
4287 ret = -ENOMEM;
4288 goto out;
4289 }
4290
4291 for (i = 0; i < map->num_stripes; i++) {
4292 device = map->stripes[i].dev;
4293 dev_offset = map->stripes[i].physical;
4294
4295 device->bytes_used += stripe_size;
4296 ret = btrfs_update_device(trans, device);
4297 if (ret)
4298 goto out;
4299 ret = btrfs_alloc_dev_extent(trans, device,
4300 chunk_root->root_key.objectid,
4301 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4302 chunk_offset, dev_offset,
4303 stripe_size);
4304 if (ret)
4305 goto out;
4306 }
4307
4308 spin_lock(&extent_root->fs_info->free_chunk_lock);
4309 extent_root->fs_info->free_chunk_space -= (stripe_size *
4310 map->num_stripes);
4311 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4312
4313 stripe = &chunk->stripe;
4314 for (i = 0; i < map->num_stripes; i++) {
4315 device = map->stripes[i].dev;
4316 dev_offset = map->stripes[i].physical;
4317
4318 btrfs_set_stack_stripe_devid(stripe, device->devid);
4319 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4320 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4321 stripe++;
4322 }
4323
4324 btrfs_set_stack_chunk_length(chunk, chunk_size);
4325 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4326 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4327 btrfs_set_stack_chunk_type(chunk, map->type);
4328 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4329 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4330 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4331 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4332 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4333
4334 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4335 key.type = BTRFS_CHUNK_ITEM_KEY;
4336 key.offset = chunk_offset;
4337
4338 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4339 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4340 /*
4341 * TODO: Cleanup of inserted chunk root in case of
4342 * failure.
4343 */
4344 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4345 item_size);
4346 }
4347
4348 out:
4349 kfree(chunk);
4350 free_extent_map(em);
4351 return ret;
4352 }
4353
4354 /*
4355 * Chunk allocation falls into two parts. The first part does works
4356 * that make the new allocated chunk useable, but not do any operation
4357 * that modifies the chunk tree. The second part does the works that
4358 * require modifying the chunk tree. This division is important for the
4359 * bootstrap process of adding storage to a seed btrfs.
4360 */
4361 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4362 struct btrfs_root *extent_root, u64 type)
4363 {
4364 u64 chunk_offset;
4365
4366 chunk_offset = find_next_chunk(extent_root->fs_info);
4367 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4368 }
4369
4370 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4371 struct btrfs_root *root,
4372 struct btrfs_device *device)
4373 {
4374 u64 chunk_offset;
4375 u64 sys_chunk_offset;
4376 u64 alloc_profile;
4377 struct btrfs_fs_info *fs_info = root->fs_info;
4378 struct btrfs_root *extent_root = fs_info->extent_root;
4379 int ret;
4380
4381 chunk_offset = find_next_chunk(fs_info);
4382 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4383 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4384 alloc_profile);
4385 if (ret)
4386 return ret;
4387
4388 sys_chunk_offset = find_next_chunk(root->fs_info);
4389 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4390 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4391 alloc_profile);
4392 if (ret) {
4393 btrfs_abort_transaction(trans, root, ret);
4394 goto out;
4395 }
4396
4397 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4398 if (ret)
4399 btrfs_abort_transaction(trans, root, ret);
4400 out:
4401 return ret;
4402 }
4403
4404 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4405 {
4406 struct extent_map *em;
4407 struct map_lookup *map;
4408 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4409 int readonly = 0;
4410 int i;
4411
4412 read_lock(&map_tree->map_tree.lock);
4413 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4414 read_unlock(&map_tree->map_tree.lock);
4415 if (!em)
4416 return 1;
4417
4418 if (btrfs_test_opt(root, DEGRADED)) {
4419 free_extent_map(em);
4420 return 0;
4421 }
4422
4423 map = (struct map_lookup *)em->bdev;
4424 for (i = 0; i < map->num_stripes; i++) {
4425 if (!map->stripes[i].dev->writeable) {
4426 readonly = 1;
4427 break;
4428 }
4429 }
4430 free_extent_map(em);
4431 return readonly;
4432 }
4433
4434 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4435 {
4436 extent_map_tree_init(&tree->map_tree);
4437 }
4438
4439 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4440 {
4441 struct extent_map *em;
4442
4443 while (1) {
4444 write_lock(&tree->map_tree.lock);
4445 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4446 if (em)
4447 remove_extent_mapping(&tree->map_tree, em);
4448 write_unlock(&tree->map_tree.lock);
4449 if (!em)
4450 break;
4451 kfree(em->bdev);
4452 /* once for us */
4453 free_extent_map(em);
4454 /* once for the tree */
4455 free_extent_map(em);
4456 }
4457 }
4458
4459 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4460 {
4461 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4462 struct extent_map *em;
4463 struct map_lookup *map;
4464 struct extent_map_tree *em_tree = &map_tree->map_tree;
4465 int ret;
4466
4467 read_lock(&em_tree->lock);
4468 em = lookup_extent_mapping(em_tree, logical, len);
4469 read_unlock(&em_tree->lock);
4470
4471 /*
4472 * We could return errors for these cases, but that could get ugly and
4473 * we'd probably do the same thing which is just not do anything else
4474 * and exit, so return 1 so the callers don't try to use other copies.
4475 */
4476 if (!em) {
4477 btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical,
4478 logical+len);
4479 return 1;
4480 }
4481
4482 if (em->start > logical || em->start + em->len < logical) {
4483 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4484 "%Lu-%Lu\n", logical, logical+len, em->start,
4485 em->start + em->len);
4486 return 1;
4487 }
4488
4489 map = (struct map_lookup *)em->bdev;
4490 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4491 ret = map->num_stripes;
4492 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4493 ret = map->sub_stripes;
4494 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4495 ret = 2;
4496 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4497 ret = 3;
4498 else
4499 ret = 1;
4500 free_extent_map(em);
4501
4502 btrfs_dev_replace_lock(&fs_info->dev_replace);
4503 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4504 ret++;
4505 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4506
4507 return ret;
4508 }
4509
4510 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4511 struct btrfs_mapping_tree *map_tree,
4512 u64 logical)
4513 {
4514 struct extent_map *em;
4515 struct map_lookup *map;
4516 struct extent_map_tree *em_tree = &map_tree->map_tree;
4517 unsigned long len = root->sectorsize;
4518
4519 read_lock(&em_tree->lock);
4520 em = lookup_extent_mapping(em_tree, logical, len);
4521 read_unlock(&em_tree->lock);
4522 BUG_ON(!em);
4523
4524 BUG_ON(em->start > logical || em->start + em->len < logical);
4525 map = (struct map_lookup *)em->bdev;
4526 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4527 BTRFS_BLOCK_GROUP_RAID6)) {
4528 len = map->stripe_len * nr_data_stripes(map);
4529 }
4530 free_extent_map(em);
4531 return len;
4532 }
4533
4534 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4535 u64 logical, u64 len, int mirror_num)
4536 {
4537 struct extent_map *em;
4538 struct map_lookup *map;
4539 struct extent_map_tree *em_tree = &map_tree->map_tree;
4540 int ret = 0;
4541
4542 read_lock(&em_tree->lock);
4543 em = lookup_extent_mapping(em_tree, logical, len);
4544 read_unlock(&em_tree->lock);
4545 BUG_ON(!em);
4546
4547 BUG_ON(em->start > logical || em->start + em->len < logical);
4548 map = (struct map_lookup *)em->bdev;
4549 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4550 BTRFS_BLOCK_GROUP_RAID6))
4551 ret = 1;
4552 free_extent_map(em);
4553 return ret;
4554 }
4555
4556 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4557 struct map_lookup *map, int first, int num,
4558 int optimal, int dev_replace_is_ongoing)
4559 {
4560 int i;
4561 int tolerance;
4562 struct btrfs_device *srcdev;
4563
4564 if (dev_replace_is_ongoing &&
4565 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4566 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4567 srcdev = fs_info->dev_replace.srcdev;
4568 else
4569 srcdev = NULL;
4570
4571 /*
4572 * try to avoid the drive that is the source drive for a
4573 * dev-replace procedure, only choose it if no other non-missing
4574 * mirror is available
4575 */
4576 for (tolerance = 0; tolerance < 2; tolerance++) {
4577 if (map->stripes[optimal].dev->bdev &&
4578 (tolerance || map->stripes[optimal].dev != srcdev))
4579 return optimal;
4580 for (i = first; i < first + num; i++) {
4581 if (map->stripes[i].dev->bdev &&
4582 (tolerance || map->stripes[i].dev != srcdev))
4583 return i;
4584 }
4585 }
4586
4587 /* we couldn't find one that doesn't fail. Just return something
4588 * and the io error handling code will clean up eventually
4589 */
4590 return optimal;
4591 }
4592
4593 static inline int parity_smaller(u64 a, u64 b)
4594 {
4595 return a > b;
4596 }
4597
4598 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4599 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4600 {
4601 struct btrfs_bio_stripe s;
4602 int i;
4603 u64 l;
4604 int again = 1;
4605
4606 while (again) {
4607 again = 0;
4608 for (i = 0; i < bbio->num_stripes - 1; i++) {
4609 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4610 s = bbio->stripes[i];
4611 l = raid_map[i];
4612 bbio->stripes[i] = bbio->stripes[i+1];
4613 raid_map[i] = raid_map[i+1];
4614 bbio->stripes[i+1] = s;
4615 raid_map[i+1] = l;
4616 again = 1;
4617 }
4618 }
4619 }
4620 }
4621
4622 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4623 u64 logical, u64 *length,
4624 struct btrfs_bio **bbio_ret,
4625 int mirror_num, u64 **raid_map_ret)
4626 {
4627 struct extent_map *em;
4628 struct map_lookup *map;
4629 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4630 struct extent_map_tree *em_tree = &map_tree->map_tree;
4631 u64 offset;
4632 u64 stripe_offset;
4633 u64 stripe_end_offset;
4634 u64 stripe_nr;
4635 u64 stripe_nr_orig;
4636 u64 stripe_nr_end;
4637 u64 stripe_len;
4638 u64 *raid_map = NULL;
4639 int stripe_index;
4640 int i;
4641 int ret = 0;
4642 int num_stripes;
4643 int max_errors = 0;
4644 struct btrfs_bio *bbio = NULL;
4645 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4646 int dev_replace_is_ongoing = 0;
4647 int num_alloc_stripes;
4648 int patch_the_first_stripe_for_dev_replace = 0;
4649 u64 physical_to_patch_in_first_stripe = 0;
4650 u64 raid56_full_stripe_start = (u64)-1;
4651
4652 read_lock(&em_tree->lock);
4653 em = lookup_extent_mapping(em_tree, logical, *length);
4654 read_unlock(&em_tree->lock);
4655
4656 if (!em) {
4657 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4658 (unsigned long long)logical,
4659 (unsigned long long)*length);
4660 return -EINVAL;
4661 }
4662
4663 if (em->start > logical || em->start + em->len < logical) {
4664 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4665 "found %Lu-%Lu\n", logical, em->start,
4666 em->start + em->len);
4667 return -EINVAL;
4668 }
4669
4670 map = (struct map_lookup *)em->bdev;
4671 offset = logical - em->start;
4672
4673 stripe_len = map->stripe_len;
4674 stripe_nr = offset;
4675 /*
4676 * stripe_nr counts the total number of stripes we have to stride
4677 * to get to this block
4678 */
4679 do_div(stripe_nr, stripe_len);
4680
4681 stripe_offset = stripe_nr * stripe_len;
4682 BUG_ON(offset < stripe_offset);
4683
4684 /* stripe_offset is the offset of this block in its stripe*/
4685 stripe_offset = offset - stripe_offset;
4686
4687 /* if we're here for raid56, we need to know the stripe aligned start */
4688 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4689 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4690 raid56_full_stripe_start = offset;
4691
4692 /* allow a write of a full stripe, but make sure we don't
4693 * allow straddling of stripes
4694 */
4695 do_div(raid56_full_stripe_start, full_stripe_len);
4696 raid56_full_stripe_start *= full_stripe_len;
4697 }
4698
4699 if (rw & REQ_DISCARD) {
4700 /* we don't discard raid56 yet */
4701 if (map->type &
4702 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4703 ret = -EOPNOTSUPP;
4704 goto out;
4705 }
4706 *length = min_t(u64, em->len - offset, *length);
4707 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4708 u64 max_len;
4709 /* For writes to RAID[56], allow a full stripeset across all disks.
4710 For other RAID types and for RAID[56] reads, just allow a single
4711 stripe (on a single disk). */
4712 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4713 (rw & REQ_WRITE)) {
4714 max_len = stripe_len * nr_data_stripes(map) -
4715 (offset - raid56_full_stripe_start);
4716 } else {
4717 /* we limit the length of each bio to what fits in a stripe */
4718 max_len = stripe_len - stripe_offset;
4719 }
4720 *length = min_t(u64, em->len - offset, max_len);
4721 } else {
4722 *length = em->len - offset;
4723 }
4724
4725 /* This is for when we're called from btrfs_merge_bio_hook() and all
4726 it cares about is the length */
4727 if (!bbio_ret)
4728 goto out;
4729
4730 btrfs_dev_replace_lock(dev_replace);
4731 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4732 if (!dev_replace_is_ongoing)
4733 btrfs_dev_replace_unlock(dev_replace);
4734
4735 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4736 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4737 dev_replace->tgtdev != NULL) {
4738 /*
4739 * in dev-replace case, for repair case (that's the only
4740 * case where the mirror is selected explicitly when
4741 * calling btrfs_map_block), blocks left of the left cursor
4742 * can also be read from the target drive.
4743 * For REQ_GET_READ_MIRRORS, the target drive is added as
4744 * the last one to the array of stripes. For READ, it also
4745 * needs to be supported using the same mirror number.
4746 * If the requested block is not left of the left cursor,
4747 * EIO is returned. This can happen because btrfs_num_copies()
4748 * returns one more in the dev-replace case.
4749 */
4750 u64 tmp_length = *length;
4751 struct btrfs_bio *tmp_bbio = NULL;
4752 int tmp_num_stripes;
4753 u64 srcdev_devid = dev_replace->srcdev->devid;
4754 int index_srcdev = 0;
4755 int found = 0;
4756 u64 physical_of_found = 0;
4757
4758 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4759 logical, &tmp_length, &tmp_bbio, 0, NULL);
4760 if (ret) {
4761 WARN_ON(tmp_bbio != NULL);
4762 goto out;
4763 }
4764
4765 tmp_num_stripes = tmp_bbio->num_stripes;
4766 if (mirror_num > tmp_num_stripes) {
4767 /*
4768 * REQ_GET_READ_MIRRORS does not contain this
4769 * mirror, that means that the requested area
4770 * is not left of the left cursor
4771 */
4772 ret = -EIO;
4773 kfree(tmp_bbio);
4774 goto out;
4775 }
4776
4777 /*
4778 * process the rest of the function using the mirror_num
4779 * of the source drive. Therefore look it up first.
4780 * At the end, patch the device pointer to the one of the
4781 * target drive.
4782 */
4783 for (i = 0; i < tmp_num_stripes; i++) {
4784 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4785 /*
4786 * In case of DUP, in order to keep it
4787 * simple, only add the mirror with the
4788 * lowest physical address
4789 */
4790 if (found &&
4791 physical_of_found <=
4792 tmp_bbio->stripes[i].physical)
4793 continue;
4794 index_srcdev = i;
4795 found = 1;
4796 physical_of_found =
4797 tmp_bbio->stripes[i].physical;
4798 }
4799 }
4800
4801 if (found) {
4802 mirror_num = index_srcdev + 1;
4803 patch_the_first_stripe_for_dev_replace = 1;
4804 physical_to_patch_in_first_stripe = physical_of_found;
4805 } else {
4806 WARN_ON(1);
4807 ret = -EIO;
4808 kfree(tmp_bbio);
4809 goto out;
4810 }
4811
4812 kfree(tmp_bbio);
4813 } else if (mirror_num > map->num_stripes) {
4814 mirror_num = 0;
4815 }
4816
4817 num_stripes = 1;
4818 stripe_index = 0;
4819 stripe_nr_orig = stripe_nr;
4820 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4821 do_div(stripe_nr_end, map->stripe_len);
4822 stripe_end_offset = stripe_nr_end * map->stripe_len -
4823 (offset + *length);
4824
4825 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4826 if (rw & REQ_DISCARD)
4827 num_stripes = min_t(u64, map->num_stripes,
4828 stripe_nr_end - stripe_nr_orig);
4829 stripe_index = do_div(stripe_nr, map->num_stripes);
4830 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4831 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4832 num_stripes = map->num_stripes;
4833 else if (mirror_num)
4834 stripe_index = mirror_num - 1;
4835 else {
4836 stripe_index = find_live_mirror(fs_info, map, 0,
4837 map->num_stripes,
4838 current->pid % map->num_stripes,
4839 dev_replace_is_ongoing);
4840 mirror_num = stripe_index + 1;
4841 }
4842
4843 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4844 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4845 num_stripes = map->num_stripes;
4846 } else if (mirror_num) {
4847 stripe_index = mirror_num - 1;
4848 } else {
4849 mirror_num = 1;
4850 }
4851
4852 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4853 int factor = map->num_stripes / map->sub_stripes;
4854
4855 stripe_index = do_div(stripe_nr, factor);
4856 stripe_index *= map->sub_stripes;
4857
4858 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4859 num_stripes = map->sub_stripes;
4860 else if (rw & REQ_DISCARD)
4861 num_stripes = min_t(u64, map->sub_stripes *
4862 (stripe_nr_end - stripe_nr_orig),
4863 map->num_stripes);
4864 else if (mirror_num)
4865 stripe_index += mirror_num - 1;
4866 else {
4867 int old_stripe_index = stripe_index;
4868 stripe_index = find_live_mirror(fs_info, map,
4869 stripe_index,
4870 map->sub_stripes, stripe_index +
4871 current->pid % map->sub_stripes,
4872 dev_replace_is_ongoing);
4873 mirror_num = stripe_index - old_stripe_index + 1;
4874 }
4875
4876 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4877 BTRFS_BLOCK_GROUP_RAID6)) {
4878 u64 tmp;
4879
4880 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4881 && raid_map_ret) {
4882 int i, rot;
4883
4884 /* push stripe_nr back to the start of the full stripe */
4885 stripe_nr = raid56_full_stripe_start;
4886 do_div(stripe_nr, stripe_len);
4887
4888 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4889
4890 /* RAID[56] write or recovery. Return all stripes */
4891 num_stripes = map->num_stripes;
4892 max_errors = nr_parity_stripes(map);
4893
4894 raid_map = kmalloc(sizeof(u64) * num_stripes,
4895 GFP_NOFS);
4896 if (!raid_map) {
4897 ret = -ENOMEM;
4898 goto out;
4899 }
4900
4901 /* Work out the disk rotation on this stripe-set */
4902 tmp = stripe_nr;
4903 rot = do_div(tmp, num_stripes);
4904
4905 /* Fill in the logical address of each stripe */
4906 tmp = stripe_nr * nr_data_stripes(map);
4907 for (i = 0; i < nr_data_stripes(map); i++)
4908 raid_map[(i+rot) % num_stripes] =
4909 em->start + (tmp + i) * map->stripe_len;
4910
4911 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4912 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4913 raid_map[(i+rot+1) % num_stripes] =
4914 RAID6_Q_STRIPE;
4915
4916 *length = map->stripe_len;
4917 stripe_index = 0;
4918 stripe_offset = 0;
4919 } else {
4920 /*
4921 * Mirror #0 or #1 means the original data block.
4922 * Mirror #2 is RAID5 parity block.
4923 * Mirror #3 is RAID6 Q block.
4924 */
4925 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4926 if (mirror_num > 1)
4927 stripe_index = nr_data_stripes(map) +
4928 mirror_num - 2;
4929
4930 /* We distribute the parity blocks across stripes */
4931 tmp = stripe_nr + stripe_index;
4932 stripe_index = do_div(tmp, map->num_stripes);
4933 }
4934 } else {
4935 /*
4936 * after this do_div call, stripe_nr is the number of stripes
4937 * on this device we have to walk to find the data, and
4938 * stripe_index is the number of our device in the stripe array
4939 */
4940 stripe_index = do_div(stripe_nr, map->num_stripes);
4941 mirror_num = stripe_index + 1;
4942 }
4943 BUG_ON(stripe_index >= map->num_stripes);
4944
4945 num_alloc_stripes = num_stripes;
4946 if (dev_replace_is_ongoing) {
4947 if (rw & (REQ_WRITE | REQ_DISCARD))
4948 num_alloc_stripes <<= 1;
4949 if (rw & REQ_GET_READ_MIRRORS)
4950 num_alloc_stripes++;
4951 }
4952 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4953 if (!bbio) {
4954 kfree(raid_map);
4955 ret = -ENOMEM;
4956 goto out;
4957 }
4958 atomic_set(&bbio->error, 0);
4959
4960 if (rw & REQ_DISCARD) {
4961 int factor = 0;
4962 int sub_stripes = 0;
4963 u64 stripes_per_dev = 0;
4964 u32 remaining_stripes = 0;
4965 u32 last_stripe = 0;
4966
4967 if (map->type &
4968 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4969 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4970 sub_stripes = 1;
4971 else
4972 sub_stripes = map->sub_stripes;
4973
4974 factor = map->num_stripes / sub_stripes;
4975 stripes_per_dev = div_u64_rem(stripe_nr_end -
4976 stripe_nr_orig,
4977 factor,
4978 &remaining_stripes);
4979 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4980 last_stripe *= sub_stripes;
4981 }
4982
4983 for (i = 0; i < num_stripes; i++) {
4984 bbio->stripes[i].physical =
4985 map->stripes[stripe_index].physical +
4986 stripe_offset + stripe_nr * map->stripe_len;
4987 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4988
4989 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4990 BTRFS_BLOCK_GROUP_RAID10)) {
4991 bbio->stripes[i].length = stripes_per_dev *
4992 map->stripe_len;
4993
4994 if (i / sub_stripes < remaining_stripes)
4995 bbio->stripes[i].length +=
4996 map->stripe_len;
4997
4998 /*
4999 * Special for the first stripe and
5000 * the last stripe:
5001 *
5002 * |-------|...|-------|
5003 * |----------|
5004 * off end_off
5005 */
5006 if (i < sub_stripes)
5007 bbio->stripes[i].length -=
5008 stripe_offset;
5009
5010 if (stripe_index >= last_stripe &&
5011 stripe_index <= (last_stripe +
5012 sub_stripes - 1))
5013 bbio->stripes[i].length -=
5014 stripe_end_offset;
5015
5016 if (i == sub_stripes - 1)
5017 stripe_offset = 0;
5018 } else
5019 bbio->stripes[i].length = *length;
5020
5021 stripe_index++;
5022 if (stripe_index == map->num_stripes) {
5023 /* This could only happen for RAID0/10 */
5024 stripe_index = 0;
5025 stripe_nr++;
5026 }
5027 }
5028 } else {
5029 for (i = 0; i < num_stripes; i++) {
5030 bbio->stripes[i].physical =
5031 map->stripes[stripe_index].physical +
5032 stripe_offset +
5033 stripe_nr * map->stripe_len;
5034 bbio->stripes[i].dev =
5035 map->stripes[stripe_index].dev;
5036 stripe_index++;
5037 }
5038 }
5039
5040 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5041 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5042 BTRFS_BLOCK_GROUP_RAID10 |
5043 BTRFS_BLOCK_GROUP_RAID5 |
5044 BTRFS_BLOCK_GROUP_DUP)) {
5045 max_errors = 1;
5046 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5047 max_errors = 2;
5048 }
5049 }
5050
5051 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5052 dev_replace->tgtdev != NULL) {
5053 int index_where_to_add;
5054 u64 srcdev_devid = dev_replace->srcdev->devid;
5055
5056 /*
5057 * duplicate the write operations while the dev replace
5058 * procedure is running. Since the copying of the old disk
5059 * to the new disk takes place at run time while the
5060 * filesystem is mounted writable, the regular write
5061 * operations to the old disk have to be duplicated to go
5062 * to the new disk as well.
5063 * Note that device->missing is handled by the caller, and
5064 * that the write to the old disk is already set up in the
5065 * stripes array.
5066 */
5067 index_where_to_add = num_stripes;
5068 for (i = 0; i < num_stripes; i++) {
5069 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5070 /* write to new disk, too */
5071 struct btrfs_bio_stripe *new =
5072 bbio->stripes + index_where_to_add;
5073 struct btrfs_bio_stripe *old =
5074 bbio->stripes + i;
5075
5076 new->physical = old->physical;
5077 new->length = old->length;
5078 new->dev = dev_replace->tgtdev;
5079 index_where_to_add++;
5080 max_errors++;
5081 }
5082 }
5083 num_stripes = index_where_to_add;
5084 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5085 dev_replace->tgtdev != NULL) {
5086 u64 srcdev_devid = dev_replace->srcdev->devid;
5087 int index_srcdev = 0;
5088 int found = 0;
5089 u64 physical_of_found = 0;
5090
5091 /*
5092 * During the dev-replace procedure, the target drive can
5093 * also be used to read data in case it is needed to repair
5094 * a corrupt block elsewhere. This is possible if the
5095 * requested area is left of the left cursor. In this area,
5096 * the target drive is a full copy of the source drive.
5097 */
5098 for (i = 0; i < num_stripes; i++) {
5099 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5100 /*
5101 * In case of DUP, in order to keep it
5102 * simple, only add the mirror with the
5103 * lowest physical address
5104 */
5105 if (found &&
5106 physical_of_found <=
5107 bbio->stripes[i].physical)
5108 continue;
5109 index_srcdev = i;
5110 found = 1;
5111 physical_of_found = bbio->stripes[i].physical;
5112 }
5113 }
5114 if (found) {
5115 u64 length = map->stripe_len;
5116
5117 if (physical_of_found + length <=
5118 dev_replace->cursor_left) {
5119 struct btrfs_bio_stripe *tgtdev_stripe =
5120 bbio->stripes + num_stripes;
5121
5122 tgtdev_stripe->physical = physical_of_found;
5123 tgtdev_stripe->length =
5124 bbio->stripes[index_srcdev].length;
5125 tgtdev_stripe->dev = dev_replace->tgtdev;
5126
5127 num_stripes++;
5128 }
5129 }
5130 }
5131
5132 *bbio_ret = bbio;
5133 bbio->num_stripes = num_stripes;
5134 bbio->max_errors = max_errors;
5135 bbio->mirror_num = mirror_num;
5136
5137 /*
5138 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5139 * mirror_num == num_stripes + 1 && dev_replace target drive is
5140 * available as a mirror
5141 */
5142 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5143 WARN_ON(num_stripes > 1);
5144 bbio->stripes[0].dev = dev_replace->tgtdev;
5145 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5146 bbio->mirror_num = map->num_stripes + 1;
5147 }
5148 if (raid_map) {
5149 sort_parity_stripes(bbio, raid_map);
5150 *raid_map_ret = raid_map;
5151 }
5152 out:
5153 if (dev_replace_is_ongoing)
5154 btrfs_dev_replace_unlock(dev_replace);
5155 free_extent_map(em);
5156 return ret;
5157 }
5158
5159 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5160 u64 logical, u64 *length,
5161 struct btrfs_bio **bbio_ret, int mirror_num)
5162 {
5163 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5164 mirror_num, NULL);
5165 }
5166
5167 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5168 u64 chunk_start, u64 physical, u64 devid,
5169 u64 **logical, int *naddrs, int *stripe_len)
5170 {
5171 struct extent_map_tree *em_tree = &map_tree->map_tree;
5172 struct extent_map *em;
5173 struct map_lookup *map;
5174 u64 *buf;
5175 u64 bytenr;
5176 u64 length;
5177 u64 stripe_nr;
5178 u64 rmap_len;
5179 int i, j, nr = 0;
5180
5181 read_lock(&em_tree->lock);
5182 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5183 read_unlock(&em_tree->lock);
5184
5185 if (!em) {
5186 printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
5187 chunk_start);
5188 return -EIO;
5189 }
5190
5191 if (em->start != chunk_start) {
5192 printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
5193 em->start, chunk_start);
5194 free_extent_map(em);
5195 return -EIO;
5196 }
5197 map = (struct map_lookup *)em->bdev;
5198
5199 length = em->len;
5200 rmap_len = map->stripe_len;
5201
5202 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5203 do_div(length, map->num_stripes / map->sub_stripes);
5204 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5205 do_div(length, map->num_stripes);
5206 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5207 BTRFS_BLOCK_GROUP_RAID6)) {
5208 do_div(length, nr_data_stripes(map));
5209 rmap_len = map->stripe_len * nr_data_stripes(map);
5210 }
5211
5212 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5213 BUG_ON(!buf); /* -ENOMEM */
5214
5215 for (i = 0; i < map->num_stripes; i++) {
5216 if (devid && map->stripes[i].dev->devid != devid)
5217 continue;
5218 if (map->stripes[i].physical > physical ||
5219 map->stripes[i].physical + length <= physical)
5220 continue;
5221
5222 stripe_nr = physical - map->stripes[i].physical;
5223 do_div(stripe_nr, map->stripe_len);
5224
5225 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5226 stripe_nr = stripe_nr * map->num_stripes + i;
5227 do_div(stripe_nr, map->sub_stripes);
5228 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5229 stripe_nr = stripe_nr * map->num_stripes + i;
5230 } /* else if RAID[56], multiply by nr_data_stripes().
5231 * Alternatively, just use rmap_len below instead of
5232 * map->stripe_len */
5233
5234 bytenr = chunk_start + stripe_nr * rmap_len;
5235 WARN_ON(nr >= map->num_stripes);
5236 for (j = 0; j < nr; j++) {
5237 if (buf[j] == bytenr)
5238 break;
5239 }
5240 if (j == nr) {
5241 WARN_ON(nr >= map->num_stripes);
5242 buf[nr++] = bytenr;
5243 }
5244 }
5245
5246 *logical = buf;
5247 *naddrs = nr;
5248 *stripe_len = rmap_len;
5249
5250 free_extent_map(em);
5251 return 0;
5252 }
5253
5254 static void btrfs_end_bio(struct bio *bio, int err)
5255 {
5256 struct btrfs_bio *bbio = bio->bi_private;
5257 int is_orig_bio = 0;
5258
5259 if (err) {
5260 atomic_inc(&bbio->error);
5261 if (err == -EIO || err == -EREMOTEIO) {
5262 unsigned int stripe_index =
5263 btrfs_io_bio(bio)->stripe_index;
5264 struct btrfs_device *dev;
5265
5266 BUG_ON(stripe_index >= bbio->num_stripes);
5267 dev = bbio->stripes[stripe_index].dev;
5268 if (dev->bdev) {
5269 if (bio->bi_rw & WRITE)
5270 btrfs_dev_stat_inc(dev,
5271 BTRFS_DEV_STAT_WRITE_ERRS);
5272 else
5273 btrfs_dev_stat_inc(dev,
5274 BTRFS_DEV_STAT_READ_ERRS);
5275 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5276 btrfs_dev_stat_inc(dev,
5277 BTRFS_DEV_STAT_FLUSH_ERRS);
5278 btrfs_dev_stat_print_on_error(dev);
5279 }
5280 }
5281 }
5282
5283 if (bio == bbio->orig_bio)
5284 is_orig_bio = 1;
5285
5286 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5287 if (!is_orig_bio) {
5288 bio_put(bio);
5289 bio = bbio->orig_bio;
5290 }
5291 bio->bi_private = bbio->private;
5292 bio->bi_end_io = bbio->end_io;
5293 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5294 /* only send an error to the higher layers if it is
5295 * beyond the tolerance of the btrfs bio
5296 */
5297 if (atomic_read(&bbio->error) > bbio->max_errors) {
5298 err = -EIO;
5299 } else {
5300 /*
5301 * this bio is actually up to date, we didn't
5302 * go over the max number of errors
5303 */
5304 set_bit(BIO_UPTODATE, &bio->bi_flags);
5305 err = 0;
5306 }
5307 kfree(bbio);
5308
5309 bio_endio(bio, err);
5310 } else if (!is_orig_bio) {
5311 bio_put(bio);
5312 }
5313 }
5314
5315 struct async_sched {
5316 struct bio *bio;
5317 int rw;
5318 struct btrfs_fs_info *info;
5319 struct btrfs_work work;
5320 };
5321
5322 /*
5323 * see run_scheduled_bios for a description of why bios are collected for
5324 * async submit.
5325 *
5326 * This will add one bio to the pending list for a device and make sure
5327 * the work struct is scheduled.
5328 */
5329 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5330 struct btrfs_device *device,
5331 int rw, struct bio *bio)
5332 {
5333 int should_queue = 1;
5334 struct btrfs_pending_bios *pending_bios;
5335
5336 if (device->missing || !device->bdev) {
5337 bio_endio(bio, -EIO);
5338 return;
5339 }
5340
5341 /* don't bother with additional async steps for reads, right now */
5342 if (!(rw & REQ_WRITE)) {
5343 bio_get(bio);
5344 btrfsic_submit_bio(rw, bio);
5345 bio_put(bio);
5346 return;
5347 }
5348
5349 /*
5350 * nr_async_bios allows us to reliably return congestion to the
5351 * higher layers. Otherwise, the async bio makes it appear we have
5352 * made progress against dirty pages when we've really just put it
5353 * on a queue for later
5354 */
5355 atomic_inc(&root->fs_info->nr_async_bios);
5356 WARN_ON(bio->bi_next);
5357 bio->bi_next = NULL;
5358 bio->bi_rw |= rw;
5359
5360 spin_lock(&device->io_lock);
5361 if (bio->bi_rw & REQ_SYNC)
5362 pending_bios = &device->pending_sync_bios;
5363 else
5364 pending_bios = &device->pending_bios;
5365
5366 if (pending_bios->tail)
5367 pending_bios->tail->bi_next = bio;
5368
5369 pending_bios->tail = bio;
5370 if (!pending_bios->head)
5371 pending_bios->head = bio;
5372 if (device->running_pending)
5373 should_queue = 0;
5374
5375 spin_unlock(&device->io_lock);
5376
5377 if (should_queue)
5378 btrfs_queue_worker(&root->fs_info->submit_workers,
5379 &device->work);
5380 }
5381
5382 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5383 sector_t sector)
5384 {
5385 struct bio_vec *prev;
5386 struct request_queue *q = bdev_get_queue(bdev);
5387 unsigned short max_sectors = queue_max_sectors(q);
5388 struct bvec_merge_data bvm = {
5389 .bi_bdev = bdev,
5390 .bi_sector = sector,
5391 .bi_rw = bio->bi_rw,
5392 };
5393
5394 if (bio->bi_vcnt == 0) {
5395 WARN_ON(1);
5396 return 1;
5397 }
5398
5399 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5400 if (bio_sectors(bio) > max_sectors)
5401 return 0;
5402
5403 if (!q->merge_bvec_fn)
5404 return 1;
5405
5406 bvm.bi_size = bio->bi_size - prev->bv_len;
5407 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5408 return 0;
5409 return 1;
5410 }
5411
5412 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5413 struct bio *bio, u64 physical, int dev_nr,
5414 int rw, int async)
5415 {
5416 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5417
5418 bio->bi_private = bbio;
5419 btrfs_io_bio(bio)->stripe_index = dev_nr;
5420 bio->bi_end_io = btrfs_end_bio;
5421 bio->bi_sector = physical >> 9;
5422 #ifdef DEBUG
5423 {
5424 struct rcu_string *name;
5425
5426 rcu_read_lock();
5427 name = rcu_dereference(dev->name);
5428 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5429 "(%s id %llu), size=%u\n", rw,
5430 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5431 name->str, dev->devid, bio->bi_size);
5432 rcu_read_unlock();
5433 }
5434 #endif
5435 bio->bi_bdev = dev->bdev;
5436 if (async)
5437 btrfs_schedule_bio(root, dev, rw, bio);
5438 else
5439 btrfsic_submit_bio(rw, bio);
5440 }
5441
5442 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5443 struct bio *first_bio, struct btrfs_device *dev,
5444 int dev_nr, int rw, int async)
5445 {
5446 struct bio_vec *bvec = first_bio->bi_io_vec;
5447 struct bio *bio;
5448 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5449 u64 physical = bbio->stripes[dev_nr].physical;
5450
5451 again:
5452 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5453 if (!bio)
5454 return -ENOMEM;
5455
5456 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5457 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5458 bvec->bv_offset) < bvec->bv_len) {
5459 u64 len = bio->bi_size;
5460
5461 atomic_inc(&bbio->stripes_pending);
5462 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5463 rw, async);
5464 physical += len;
5465 goto again;
5466 }
5467 bvec++;
5468 }
5469
5470 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5471 return 0;
5472 }
5473
5474 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5475 {
5476 atomic_inc(&bbio->error);
5477 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5478 bio->bi_private = bbio->private;
5479 bio->bi_end_io = bbio->end_io;
5480 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5481 bio->bi_sector = logical >> 9;
5482 kfree(bbio);
5483 bio_endio(bio, -EIO);
5484 }
5485 }
5486
5487 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5488 int mirror_num, int async_submit)
5489 {
5490 struct btrfs_device *dev;
5491 struct bio *first_bio = bio;
5492 u64 logical = (u64)bio->bi_sector << 9;
5493 u64 length = 0;
5494 u64 map_length;
5495 u64 *raid_map = NULL;
5496 int ret;
5497 int dev_nr = 0;
5498 int total_devs = 1;
5499 struct btrfs_bio *bbio = NULL;
5500
5501 length = bio->bi_size;
5502 map_length = length;
5503
5504 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5505 mirror_num, &raid_map);
5506 if (ret) /* -ENOMEM */
5507 return ret;
5508
5509 total_devs = bbio->num_stripes;
5510 bbio->orig_bio = first_bio;
5511 bbio->private = first_bio->bi_private;
5512 bbio->end_io = first_bio->bi_end_io;
5513 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5514
5515 if (raid_map) {
5516 /* In this case, map_length has been set to the length of
5517 a single stripe; not the whole write */
5518 if (rw & WRITE) {
5519 return raid56_parity_write(root, bio, bbio,
5520 raid_map, map_length);
5521 } else {
5522 return raid56_parity_recover(root, bio, bbio,
5523 raid_map, map_length,
5524 mirror_num);
5525 }
5526 }
5527
5528 if (map_length < length) {
5529 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5530 (unsigned long long)logical,
5531 (unsigned long long)length,
5532 (unsigned long long)map_length);
5533 BUG();
5534 }
5535
5536 while (dev_nr < total_devs) {
5537 dev = bbio->stripes[dev_nr].dev;
5538 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5539 bbio_error(bbio, first_bio, logical);
5540 dev_nr++;
5541 continue;
5542 }
5543
5544 /*
5545 * Check and see if we're ok with this bio based on it's size
5546 * and offset with the given device.
5547 */
5548 if (!bio_size_ok(dev->bdev, first_bio,
5549 bbio->stripes[dev_nr].physical >> 9)) {
5550 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5551 dev_nr, rw, async_submit);
5552 BUG_ON(ret);
5553 dev_nr++;
5554 continue;
5555 }
5556
5557 if (dev_nr < total_devs - 1) {
5558 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5559 BUG_ON(!bio); /* -ENOMEM */
5560 } else {
5561 bio = first_bio;
5562 }
5563
5564 submit_stripe_bio(root, bbio, bio,
5565 bbio->stripes[dev_nr].physical, dev_nr, rw,
5566 async_submit);
5567 dev_nr++;
5568 }
5569 return 0;
5570 }
5571
5572 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5573 u8 *uuid, u8 *fsid)
5574 {
5575 struct btrfs_device *device;
5576 struct btrfs_fs_devices *cur_devices;
5577
5578 cur_devices = fs_info->fs_devices;
5579 while (cur_devices) {
5580 if (!fsid ||
5581 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5582 device = __find_device(&cur_devices->devices,
5583 devid, uuid);
5584 if (device)
5585 return device;
5586 }
5587 cur_devices = cur_devices->seed;
5588 }
5589 return NULL;
5590 }
5591
5592 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5593 u64 devid, u8 *dev_uuid)
5594 {
5595 struct btrfs_device *device;
5596 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5597
5598 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5599 if (IS_ERR(device))
5600 return NULL;
5601
5602 list_add(&device->dev_list, &fs_devices->devices);
5603 device->fs_devices = fs_devices;
5604 fs_devices->num_devices++;
5605
5606 device->missing = 1;
5607 fs_devices->missing_devices++;
5608
5609 return device;
5610 }
5611
5612 /**
5613 * btrfs_alloc_device - allocate struct btrfs_device
5614 * @fs_info: used only for generating a new devid, can be NULL if
5615 * devid is provided (i.e. @devid != NULL).
5616 * @devid: a pointer to devid for this device. If NULL a new devid
5617 * is generated.
5618 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5619 * is generated.
5620 *
5621 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5622 * on error. Returned struct is not linked onto any lists and can be
5623 * destroyed with kfree() right away.
5624 */
5625 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5626 const u64 *devid,
5627 const u8 *uuid)
5628 {
5629 struct btrfs_device *dev;
5630 u64 tmp;
5631
5632 if (!devid && !fs_info) {
5633 WARN_ON(1);
5634 return ERR_PTR(-EINVAL);
5635 }
5636
5637 dev = __alloc_device();
5638 if (IS_ERR(dev))
5639 return dev;
5640
5641 if (devid)
5642 tmp = *devid;
5643 else {
5644 int ret;
5645
5646 ret = find_next_devid(fs_info, &tmp);
5647 if (ret) {
5648 kfree(dev);
5649 return ERR_PTR(ret);
5650 }
5651 }
5652 dev->devid = tmp;
5653
5654 if (uuid)
5655 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5656 else
5657 generate_random_uuid(dev->uuid);
5658
5659 dev->work.func = pending_bios_fn;
5660
5661 return dev;
5662 }
5663
5664 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5665 struct extent_buffer *leaf,
5666 struct btrfs_chunk *chunk)
5667 {
5668 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5669 struct map_lookup *map;
5670 struct extent_map *em;
5671 u64 logical;
5672 u64 length;
5673 u64 devid;
5674 u8 uuid[BTRFS_UUID_SIZE];
5675 int num_stripes;
5676 int ret;
5677 int i;
5678
5679 logical = key->offset;
5680 length = btrfs_chunk_length(leaf, chunk);
5681
5682 read_lock(&map_tree->map_tree.lock);
5683 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5684 read_unlock(&map_tree->map_tree.lock);
5685
5686 /* already mapped? */
5687 if (em && em->start <= logical && em->start + em->len > logical) {
5688 free_extent_map(em);
5689 return 0;
5690 } else if (em) {
5691 free_extent_map(em);
5692 }
5693
5694 em = alloc_extent_map();
5695 if (!em)
5696 return -ENOMEM;
5697 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5698 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5699 if (!map) {
5700 free_extent_map(em);
5701 return -ENOMEM;
5702 }
5703
5704 em->bdev = (struct block_device *)map;
5705 em->start = logical;
5706 em->len = length;
5707 em->orig_start = 0;
5708 em->block_start = 0;
5709 em->block_len = em->len;
5710
5711 map->num_stripes = num_stripes;
5712 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5713 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5714 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5715 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5716 map->type = btrfs_chunk_type(leaf, chunk);
5717 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5718 for (i = 0; i < num_stripes; i++) {
5719 map->stripes[i].physical =
5720 btrfs_stripe_offset_nr(leaf, chunk, i);
5721 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5722 read_extent_buffer(leaf, uuid, (unsigned long)
5723 btrfs_stripe_dev_uuid_nr(chunk, i),
5724 BTRFS_UUID_SIZE);
5725 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5726 uuid, NULL);
5727 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5728 kfree(map);
5729 free_extent_map(em);
5730 return -EIO;
5731 }
5732 if (!map->stripes[i].dev) {
5733 map->stripes[i].dev =
5734 add_missing_dev(root, devid, uuid);
5735 if (!map->stripes[i].dev) {
5736 kfree(map);
5737 free_extent_map(em);
5738 return -EIO;
5739 }
5740 }
5741 map->stripes[i].dev->in_fs_metadata = 1;
5742 }
5743
5744 write_lock(&map_tree->map_tree.lock);
5745 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5746 write_unlock(&map_tree->map_tree.lock);
5747 BUG_ON(ret); /* Tree corruption */
5748 free_extent_map(em);
5749
5750 return 0;
5751 }
5752
5753 static void fill_device_from_item(struct extent_buffer *leaf,
5754 struct btrfs_dev_item *dev_item,
5755 struct btrfs_device *device)
5756 {
5757 unsigned long ptr;
5758
5759 device->devid = btrfs_device_id(leaf, dev_item);
5760 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5761 device->total_bytes = device->disk_total_bytes;
5762 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5763 device->type = btrfs_device_type(leaf, dev_item);
5764 device->io_align = btrfs_device_io_align(leaf, dev_item);
5765 device->io_width = btrfs_device_io_width(leaf, dev_item);
5766 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5767 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5768 device->is_tgtdev_for_dev_replace = 0;
5769
5770 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5771 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5772 }
5773
5774 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5775 {
5776 struct btrfs_fs_devices *fs_devices;
5777 int ret;
5778
5779 BUG_ON(!mutex_is_locked(&uuid_mutex));
5780
5781 fs_devices = root->fs_info->fs_devices->seed;
5782 while (fs_devices) {
5783 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5784 ret = 0;
5785 goto out;
5786 }
5787 fs_devices = fs_devices->seed;
5788 }
5789
5790 fs_devices = find_fsid(fsid);
5791 if (!fs_devices) {
5792 ret = -ENOENT;
5793 goto out;
5794 }
5795
5796 fs_devices = clone_fs_devices(fs_devices);
5797 if (IS_ERR(fs_devices)) {
5798 ret = PTR_ERR(fs_devices);
5799 goto out;
5800 }
5801
5802 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5803 root->fs_info->bdev_holder);
5804 if (ret) {
5805 free_fs_devices(fs_devices);
5806 goto out;
5807 }
5808
5809 if (!fs_devices->seeding) {
5810 __btrfs_close_devices(fs_devices);
5811 free_fs_devices(fs_devices);
5812 ret = -EINVAL;
5813 goto out;
5814 }
5815
5816 fs_devices->seed = root->fs_info->fs_devices->seed;
5817 root->fs_info->fs_devices->seed = fs_devices;
5818 out:
5819 return ret;
5820 }
5821
5822 static int read_one_dev(struct btrfs_root *root,
5823 struct extent_buffer *leaf,
5824 struct btrfs_dev_item *dev_item)
5825 {
5826 struct btrfs_device *device;
5827 u64 devid;
5828 int ret;
5829 u8 fs_uuid[BTRFS_UUID_SIZE];
5830 u8 dev_uuid[BTRFS_UUID_SIZE];
5831
5832 devid = btrfs_device_id(leaf, dev_item);
5833 read_extent_buffer(leaf, dev_uuid,
5834 (unsigned long)btrfs_device_uuid(dev_item),
5835 BTRFS_UUID_SIZE);
5836 read_extent_buffer(leaf, fs_uuid,
5837 (unsigned long)btrfs_device_fsid(dev_item),
5838 BTRFS_UUID_SIZE);
5839
5840 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5841 ret = open_seed_devices(root, fs_uuid);
5842 if (ret && !btrfs_test_opt(root, DEGRADED))
5843 return ret;
5844 }
5845
5846 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5847 if (!device || !device->bdev) {
5848 if (!btrfs_test_opt(root, DEGRADED))
5849 return -EIO;
5850
5851 if (!device) {
5852 btrfs_warn(root->fs_info, "devid %llu missing",
5853 (unsigned long long)devid);
5854 device = add_missing_dev(root, devid, dev_uuid);
5855 if (!device)
5856 return -ENOMEM;
5857 } else if (!device->missing) {
5858 /*
5859 * this happens when a device that was properly setup
5860 * in the device info lists suddenly goes bad.
5861 * device->bdev is NULL, and so we have to set
5862 * device->missing to one here
5863 */
5864 root->fs_info->fs_devices->missing_devices++;
5865 device->missing = 1;
5866 }
5867 }
5868
5869 if (device->fs_devices != root->fs_info->fs_devices) {
5870 BUG_ON(device->writeable);
5871 if (device->generation !=
5872 btrfs_device_generation(leaf, dev_item))
5873 return -EINVAL;
5874 }
5875
5876 fill_device_from_item(leaf, dev_item, device);
5877 device->in_fs_metadata = 1;
5878 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5879 device->fs_devices->total_rw_bytes += device->total_bytes;
5880 spin_lock(&root->fs_info->free_chunk_lock);
5881 root->fs_info->free_chunk_space += device->total_bytes -
5882 device->bytes_used;
5883 spin_unlock(&root->fs_info->free_chunk_lock);
5884 }
5885 ret = 0;
5886 return ret;
5887 }
5888
5889 int btrfs_read_sys_array(struct btrfs_root *root)
5890 {
5891 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5892 struct extent_buffer *sb;
5893 struct btrfs_disk_key *disk_key;
5894 struct btrfs_chunk *chunk;
5895 u8 *ptr;
5896 unsigned long sb_ptr;
5897 int ret = 0;
5898 u32 num_stripes;
5899 u32 array_size;
5900 u32 len = 0;
5901 u32 cur;
5902 struct btrfs_key key;
5903
5904 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5905 BTRFS_SUPER_INFO_SIZE);
5906 if (!sb)
5907 return -ENOMEM;
5908 btrfs_set_buffer_uptodate(sb);
5909 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5910 /*
5911 * The sb extent buffer is artifical and just used to read the system array.
5912 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5913 * pages up-to-date when the page is larger: extent does not cover the
5914 * whole page and consequently check_page_uptodate does not find all
5915 * the page's extents up-to-date (the hole beyond sb),
5916 * write_extent_buffer then triggers a WARN_ON.
5917 *
5918 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5919 * but sb spans only this function. Add an explicit SetPageUptodate call
5920 * to silence the warning eg. on PowerPC 64.
5921 */
5922 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5923 SetPageUptodate(sb->pages[0]);
5924
5925 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5926 array_size = btrfs_super_sys_array_size(super_copy);
5927
5928 ptr = super_copy->sys_chunk_array;
5929 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5930 cur = 0;
5931
5932 while (cur < array_size) {
5933 disk_key = (struct btrfs_disk_key *)ptr;
5934 btrfs_disk_key_to_cpu(&key, disk_key);
5935
5936 len = sizeof(*disk_key); ptr += len;
5937 sb_ptr += len;
5938 cur += len;
5939
5940 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5941 chunk = (struct btrfs_chunk *)sb_ptr;
5942 ret = read_one_chunk(root, &key, sb, chunk);
5943 if (ret)
5944 break;
5945 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5946 len = btrfs_chunk_item_size(num_stripes);
5947 } else {
5948 ret = -EIO;
5949 break;
5950 }
5951 ptr += len;
5952 sb_ptr += len;
5953 cur += len;
5954 }
5955 free_extent_buffer(sb);
5956 return ret;
5957 }
5958
5959 int btrfs_read_chunk_tree(struct btrfs_root *root)
5960 {
5961 struct btrfs_path *path;
5962 struct extent_buffer *leaf;
5963 struct btrfs_key key;
5964 struct btrfs_key found_key;
5965 int ret;
5966 int slot;
5967
5968 root = root->fs_info->chunk_root;
5969
5970 path = btrfs_alloc_path();
5971 if (!path)
5972 return -ENOMEM;
5973
5974 mutex_lock(&uuid_mutex);
5975 lock_chunks(root);
5976
5977 /*
5978 * Read all device items, and then all the chunk items. All
5979 * device items are found before any chunk item (their object id
5980 * is smaller than the lowest possible object id for a chunk
5981 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
5982 */
5983 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5984 key.offset = 0;
5985 key.type = 0;
5986 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5987 if (ret < 0)
5988 goto error;
5989 while (1) {
5990 leaf = path->nodes[0];
5991 slot = path->slots[0];
5992 if (slot >= btrfs_header_nritems(leaf)) {
5993 ret = btrfs_next_leaf(root, path);
5994 if (ret == 0)
5995 continue;
5996 if (ret < 0)
5997 goto error;
5998 break;
5999 }
6000 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6001 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6002 struct btrfs_dev_item *dev_item;
6003 dev_item = btrfs_item_ptr(leaf, slot,
6004 struct btrfs_dev_item);
6005 ret = read_one_dev(root, leaf, dev_item);
6006 if (ret)
6007 goto error;
6008 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6009 struct btrfs_chunk *chunk;
6010 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6011 ret = read_one_chunk(root, &found_key, leaf, chunk);
6012 if (ret)
6013 goto error;
6014 }
6015 path->slots[0]++;
6016 }
6017 ret = 0;
6018 error:
6019 unlock_chunks(root);
6020 mutex_unlock(&uuid_mutex);
6021
6022 btrfs_free_path(path);
6023 return ret;
6024 }
6025
6026 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6027 {
6028 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6029 struct btrfs_device *device;
6030
6031 mutex_lock(&fs_devices->device_list_mutex);
6032 list_for_each_entry(device, &fs_devices->devices, dev_list)
6033 device->dev_root = fs_info->dev_root;
6034 mutex_unlock(&fs_devices->device_list_mutex);
6035 }
6036
6037 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6038 {
6039 int i;
6040
6041 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6042 btrfs_dev_stat_reset(dev, i);
6043 }
6044
6045 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6046 {
6047 struct btrfs_key key;
6048 struct btrfs_key found_key;
6049 struct btrfs_root *dev_root = fs_info->dev_root;
6050 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6051 struct extent_buffer *eb;
6052 int slot;
6053 int ret = 0;
6054 struct btrfs_device *device;
6055 struct btrfs_path *path = NULL;
6056 int i;
6057
6058 path = btrfs_alloc_path();
6059 if (!path) {
6060 ret = -ENOMEM;
6061 goto out;
6062 }
6063
6064 mutex_lock(&fs_devices->device_list_mutex);
6065 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6066 int item_size;
6067 struct btrfs_dev_stats_item *ptr;
6068
6069 key.objectid = 0;
6070 key.type = BTRFS_DEV_STATS_KEY;
6071 key.offset = device->devid;
6072 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6073 if (ret) {
6074 __btrfs_reset_dev_stats(device);
6075 device->dev_stats_valid = 1;
6076 btrfs_release_path(path);
6077 continue;
6078 }
6079 slot = path->slots[0];
6080 eb = path->nodes[0];
6081 btrfs_item_key_to_cpu(eb, &found_key, slot);
6082 item_size = btrfs_item_size_nr(eb, slot);
6083
6084 ptr = btrfs_item_ptr(eb, slot,
6085 struct btrfs_dev_stats_item);
6086
6087 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6088 if (item_size >= (1 + i) * sizeof(__le64))
6089 btrfs_dev_stat_set(device, i,
6090 btrfs_dev_stats_value(eb, ptr, i));
6091 else
6092 btrfs_dev_stat_reset(device, i);
6093 }
6094
6095 device->dev_stats_valid = 1;
6096 btrfs_dev_stat_print_on_load(device);
6097 btrfs_release_path(path);
6098 }
6099 mutex_unlock(&fs_devices->device_list_mutex);
6100
6101 out:
6102 btrfs_free_path(path);
6103 return ret < 0 ? ret : 0;
6104 }
6105
6106 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6107 struct btrfs_root *dev_root,
6108 struct btrfs_device *device)
6109 {
6110 struct btrfs_path *path;
6111 struct btrfs_key key;
6112 struct extent_buffer *eb;
6113 struct btrfs_dev_stats_item *ptr;
6114 int ret;
6115 int i;
6116
6117 key.objectid = 0;
6118 key.type = BTRFS_DEV_STATS_KEY;
6119 key.offset = device->devid;
6120
6121 path = btrfs_alloc_path();
6122 BUG_ON(!path);
6123 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6124 if (ret < 0) {
6125 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
6126 ret, rcu_str_deref(device->name));
6127 goto out;
6128 }
6129
6130 if (ret == 0 &&
6131 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6132 /* need to delete old one and insert a new one */
6133 ret = btrfs_del_item(trans, dev_root, path);
6134 if (ret != 0) {
6135 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
6136 rcu_str_deref(device->name), ret);
6137 goto out;
6138 }
6139 ret = 1;
6140 }
6141
6142 if (ret == 1) {
6143 /* need to insert a new item */
6144 btrfs_release_path(path);
6145 ret = btrfs_insert_empty_item(trans, dev_root, path,
6146 &key, sizeof(*ptr));
6147 if (ret < 0) {
6148 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
6149 rcu_str_deref(device->name), ret);
6150 goto out;
6151 }
6152 }
6153
6154 eb = path->nodes[0];
6155 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6156 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6157 btrfs_set_dev_stats_value(eb, ptr, i,
6158 btrfs_dev_stat_read(device, i));
6159 btrfs_mark_buffer_dirty(eb);
6160
6161 out:
6162 btrfs_free_path(path);
6163 return ret;
6164 }
6165
6166 /*
6167 * called from commit_transaction. Writes all changed device stats to disk.
6168 */
6169 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6170 struct btrfs_fs_info *fs_info)
6171 {
6172 struct btrfs_root *dev_root = fs_info->dev_root;
6173 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6174 struct btrfs_device *device;
6175 int ret = 0;
6176
6177 mutex_lock(&fs_devices->device_list_mutex);
6178 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6179 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6180 continue;
6181
6182 ret = update_dev_stat_item(trans, dev_root, device);
6183 if (!ret)
6184 device->dev_stats_dirty = 0;
6185 }
6186 mutex_unlock(&fs_devices->device_list_mutex);
6187
6188 return ret;
6189 }
6190
6191 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6192 {
6193 btrfs_dev_stat_inc(dev, index);
6194 btrfs_dev_stat_print_on_error(dev);
6195 }
6196
6197 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6198 {
6199 if (!dev->dev_stats_valid)
6200 return;
6201 printk_ratelimited_in_rcu(KERN_ERR
6202 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6203 rcu_str_deref(dev->name),
6204 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6205 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6206 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6207 btrfs_dev_stat_read(dev,
6208 BTRFS_DEV_STAT_CORRUPTION_ERRS),
6209 btrfs_dev_stat_read(dev,
6210 BTRFS_DEV_STAT_GENERATION_ERRS));
6211 }
6212
6213 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6214 {
6215 int i;
6216
6217 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6218 if (btrfs_dev_stat_read(dev, i) != 0)
6219 break;
6220 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6221 return; /* all values == 0, suppress message */
6222
6223 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6224 rcu_str_deref(dev->name),
6225 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6226 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6227 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6228 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6229 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6230 }
6231
6232 int btrfs_get_dev_stats(struct btrfs_root *root,
6233 struct btrfs_ioctl_get_dev_stats *stats)
6234 {
6235 struct btrfs_device *dev;
6236 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6237 int i;
6238
6239 mutex_lock(&fs_devices->device_list_mutex);
6240 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6241 mutex_unlock(&fs_devices->device_list_mutex);
6242
6243 if (!dev) {
6244 printk(KERN_WARNING
6245 "btrfs: get dev_stats failed, device not found\n");
6246 return -ENODEV;
6247 } else if (!dev->dev_stats_valid) {
6248 printk(KERN_WARNING
6249 "btrfs: get dev_stats failed, not yet valid\n");
6250 return -ENODEV;
6251 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6252 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6253 if (stats->nr_items > i)
6254 stats->values[i] =
6255 btrfs_dev_stat_read_and_reset(dev, i);
6256 else
6257 btrfs_dev_stat_reset(dev, i);
6258 }
6259 } else {
6260 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6261 if (stats->nr_items > i)
6262 stats->values[i] = btrfs_dev_stat_read(dev, i);
6263 }
6264 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6265 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6266 return 0;
6267 }
6268
6269 int btrfs_scratch_superblock(struct btrfs_device *device)
6270 {
6271 struct buffer_head *bh;
6272 struct btrfs_super_block *disk_super;
6273
6274 bh = btrfs_read_dev_super(device->bdev);
6275 if (!bh)
6276 return -EINVAL;
6277 disk_super = (struct btrfs_super_block *)bh->b_data;
6278
6279 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6280 set_buffer_dirty(bh);
6281 sync_dirty_buffer(bh);
6282 brelse(bh);
6283
6284 return 0;
6285 }
Linux kernel - Deliverables
This page took 0.202044 seconds and 4 git commands to generate.