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


5349e17d88639d0f1d861fe3ebb72ca3cd62ec6a
[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 "compat.h"
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
38 #include "math.h"
39 #include "dev-replace.h"
40
41 static int init_first_rw_device(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 struct btrfs_device *device);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
45 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
47
48 static DEFINE_MUTEX(uuid_mutex);
49 static LIST_HEAD(fs_uuids);
50
51 static void lock_chunks(struct btrfs_root *root)
52 {
53 mutex_lock(&root->fs_info->chunk_mutex);
54 }
55
56 static void unlock_chunks(struct btrfs_root *root)
57 {
58 mutex_unlock(&root->fs_info->chunk_mutex);
59 }
60
61 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62 {
63 struct btrfs_device *device;
64 WARN_ON(fs_devices->opened);
65 while (!list_empty(&fs_devices->devices)) {
66 device = list_entry(fs_devices->devices.next,
67 struct btrfs_device, dev_list);
68 list_del(&device->dev_list);
69 rcu_string_free(device->name);
70 kfree(device);
71 }
72 kfree(fs_devices);
73 }
74
75 static void btrfs_kobject_uevent(struct block_device *bdev,
76 enum kobject_action action)
77 {
78 int ret;
79
80 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
81 if (ret)
82 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
83 action,
84 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
85 &disk_to_dev(bdev->bd_disk)->kobj);
86 }
87
88 void btrfs_cleanup_fs_uuids(void)
89 {
90 struct btrfs_fs_devices *fs_devices;
91
92 while (!list_empty(&fs_uuids)) {
93 fs_devices = list_entry(fs_uuids.next,
94 struct btrfs_fs_devices, list);
95 list_del(&fs_devices->list);
96 free_fs_devices(fs_devices);
97 }
98 }
99
100 static noinline struct btrfs_device *__find_device(struct list_head *head,
101 u64 devid, u8 *uuid)
102 {
103 struct btrfs_device *dev;
104
105 list_for_each_entry(dev, head, dev_list) {
106 if (dev->devid == devid &&
107 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
108 return dev;
109 }
110 }
111 return NULL;
112 }
113
114 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 {
116 struct btrfs_fs_devices *fs_devices;
117
118 list_for_each_entry(fs_devices, &fs_uuids, list) {
119 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
120 return fs_devices;
121 }
122 return NULL;
123 }
124
125 static int
126 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
127 int flush, struct block_device **bdev,
128 struct buffer_head **bh)
129 {
130 int ret;
131
132 *bdev = blkdev_get_by_path(device_path, flags, holder);
133
134 if (IS_ERR(*bdev)) {
135 ret = PTR_ERR(*bdev);
136 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
137 goto error;
138 }
139
140 if (flush)
141 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
142 ret = set_blocksize(*bdev, 4096);
143 if (ret) {
144 blkdev_put(*bdev, flags);
145 goto error;
146 }
147 invalidate_bdev(*bdev);
148 *bh = btrfs_read_dev_super(*bdev);
149 if (!*bh) {
150 ret = -EINVAL;
151 blkdev_put(*bdev, flags);
152 goto error;
153 }
154
155 return 0;
156
157 error:
158 *bdev = NULL;
159 *bh = NULL;
160 return ret;
161 }
162
163 static void requeue_list(struct btrfs_pending_bios *pending_bios,
164 struct bio *head, struct bio *tail)
165 {
166
167 struct bio *old_head;
168
169 old_head = pending_bios->head;
170 pending_bios->head = head;
171 if (pending_bios->tail)
172 tail->bi_next = old_head;
173 else
174 pending_bios->tail = tail;
175 }
176
177 /*
178 * we try to collect pending bios for a device so we don't get a large
179 * number of procs sending bios down to the same device. This greatly
180 * improves the schedulers ability to collect and merge the bios.
181 *
182 * But, it also turns into a long list of bios to process and that is sure
183 * to eventually make the worker thread block. The solution here is to
184 * make some progress and then put this work struct back at the end of
185 * the list if the block device is congested. This way, multiple devices
186 * can make progress from a single worker thread.
187 */
188 static noinline void run_scheduled_bios(struct btrfs_device *device)
189 {
190 struct bio *pending;
191 struct backing_dev_info *bdi;
192 struct btrfs_fs_info *fs_info;
193 struct btrfs_pending_bios *pending_bios;
194 struct bio *tail;
195 struct bio *cur;
196 int again = 0;
197 unsigned long num_run;
198 unsigned long batch_run = 0;
199 unsigned long limit;
200 unsigned long last_waited = 0;
201 int force_reg = 0;
202 int sync_pending = 0;
203 struct blk_plug plug;
204
205 /*
206 * this function runs all the bios we've collected for
207 * a particular device. We don't want to wander off to
208 * another device without first sending all of these down.
209 * So, setup a plug here and finish it off before we return
210 */
211 blk_start_plug(&plug);
212
213 bdi = blk_get_backing_dev_info(device->bdev);
214 fs_info = device->dev_root->fs_info;
215 limit = btrfs_async_submit_limit(fs_info);
216 limit = limit * 2 / 3;
217
218 loop:
219 spin_lock(&device->io_lock);
220
221 loop_lock:
222 num_run = 0;
223
224 /* take all the bios off the list at once and process them
225 * later on (without the lock held). But, remember the
226 * tail and other pointers so the bios can be properly reinserted
227 * into the list if we hit congestion
228 */
229 if (!force_reg && device->pending_sync_bios.head) {
230 pending_bios = &device->pending_sync_bios;
231 force_reg = 1;
232 } else {
233 pending_bios = &device->pending_bios;
234 force_reg = 0;
235 }
236
237 pending = pending_bios->head;
238 tail = pending_bios->tail;
239 WARN_ON(pending && !tail);
240
241 /*
242 * if pending was null this time around, no bios need processing
243 * at all and we can stop. Otherwise it'll loop back up again
244 * and do an additional check so no bios are missed.
245 *
246 * device->running_pending is used to synchronize with the
247 * schedule_bio code.
248 */
249 if (device->pending_sync_bios.head == NULL &&
250 device->pending_bios.head == NULL) {
251 again = 0;
252 device->running_pending = 0;
253 } else {
254 again = 1;
255 device->running_pending = 1;
256 }
257
258 pending_bios->head = NULL;
259 pending_bios->tail = NULL;
260
261 spin_unlock(&device->io_lock);
262
263 while (pending) {
264
265 rmb();
266 /* we want to work on both lists, but do more bios on the
267 * sync list than the regular list
268 */
269 if ((num_run > 32 &&
270 pending_bios != &device->pending_sync_bios &&
271 device->pending_sync_bios.head) ||
272 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
273 device->pending_bios.head)) {
274 spin_lock(&device->io_lock);
275 requeue_list(pending_bios, pending, tail);
276 goto loop_lock;
277 }
278
279 cur = pending;
280 pending = pending->bi_next;
281 cur->bi_next = NULL;
282
283 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
284 waitqueue_active(&fs_info->async_submit_wait))
285 wake_up(&fs_info->async_submit_wait);
286
287 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
288
289 /*
290 * if we're doing the sync list, record that our
291 * plug has some sync requests on it
292 *
293 * If we're doing the regular list and there are
294 * sync requests sitting around, unplug before
295 * we add more
296 */
297 if (pending_bios == &device->pending_sync_bios) {
298 sync_pending = 1;
299 } else if (sync_pending) {
300 blk_finish_plug(&plug);
301 blk_start_plug(&plug);
302 sync_pending = 0;
303 }
304
305 btrfsic_submit_bio(cur->bi_rw, cur);
306 num_run++;
307 batch_run++;
308 if (need_resched())
309 cond_resched();
310
311 /*
312 * we made progress, there is more work to do and the bdi
313 * is now congested. Back off and let other work structs
314 * run instead
315 */
316 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
317 fs_info->fs_devices->open_devices > 1) {
318 struct io_context *ioc;
319
320 ioc = current->io_context;
321
322 /*
323 * the main goal here is that we don't want to
324 * block if we're going to be able to submit
325 * more requests without blocking.
326 *
327 * This code does two great things, it pokes into
328 * the elevator code from a filesystem _and_
329 * it makes assumptions about how batching works.
330 */
331 if (ioc && ioc->nr_batch_requests > 0 &&
332 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
333 (last_waited == 0 ||
334 ioc->last_waited == last_waited)) {
335 /*
336 * we want to go through our batch of
337 * requests and stop. So, we copy out
338 * the ioc->last_waited time and test
339 * against it before looping
340 */
341 last_waited = ioc->last_waited;
342 if (need_resched())
343 cond_resched();
344 continue;
345 }
346 spin_lock(&device->io_lock);
347 requeue_list(pending_bios, pending, tail);
348 device->running_pending = 1;
349
350 spin_unlock(&device->io_lock);
351 btrfs_requeue_work(&device->work);
352 goto done;
353 }
354 /* unplug every 64 requests just for good measure */
355 if (batch_run % 64 == 0) {
356 blk_finish_plug(&plug);
357 blk_start_plug(&plug);
358 sync_pending = 0;
359 }
360 }
361
362 cond_resched();
363 if (again)
364 goto loop;
365
366 spin_lock(&device->io_lock);
367 if (device->pending_bios.head || device->pending_sync_bios.head)
368 goto loop_lock;
369 spin_unlock(&device->io_lock);
370
371 done:
372 blk_finish_plug(&plug);
373 }
374
375 static void pending_bios_fn(struct btrfs_work *work)
376 {
377 struct btrfs_device *device;
378
379 device = container_of(work, struct btrfs_device, work);
380 run_scheduled_bios(device);
381 }
382
383 static noinline int device_list_add(const char *path,
384 struct btrfs_super_block *disk_super,
385 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
386 {
387 struct btrfs_device *device;
388 struct btrfs_fs_devices *fs_devices;
389 struct rcu_string *name;
390 u64 found_transid = btrfs_super_generation(disk_super);
391
392 fs_devices = find_fsid(disk_super->fsid);
393 if (!fs_devices) {
394 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
395 if (!fs_devices)
396 return -ENOMEM;
397 INIT_LIST_HEAD(&fs_devices->devices);
398 INIT_LIST_HEAD(&fs_devices->alloc_list);
399 list_add(&fs_devices->list, &fs_uuids);
400 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
401 fs_devices->latest_devid = devid;
402 fs_devices->latest_trans = found_transid;
403 mutex_init(&fs_devices->device_list_mutex);
404 device = NULL;
405 } else {
406 device = __find_device(&fs_devices->devices, devid,
407 disk_super->dev_item.uuid);
408 }
409 if (!device) {
410 if (fs_devices->opened)
411 return -EBUSY;
412
413 device = kzalloc(sizeof(*device), GFP_NOFS);
414 if (!device) {
415 /* we can safely leave the fs_devices entry around */
416 return -ENOMEM;
417 }
418 device->devid = devid;
419 device->dev_stats_valid = 0;
420 device->work.func = pending_bios_fn;
421 memcpy(device->uuid, disk_super->dev_item.uuid,
422 BTRFS_UUID_SIZE);
423 spin_lock_init(&device->io_lock);
424
425 name = rcu_string_strdup(path, GFP_NOFS);
426 if (!name) {
427 kfree(device);
428 return -ENOMEM;
429 }
430 rcu_assign_pointer(device->name, name);
431 INIT_LIST_HEAD(&device->dev_alloc_list);
432
433 /* init readahead state */
434 spin_lock_init(&device->reada_lock);
435 device->reada_curr_zone = NULL;
436 atomic_set(&device->reada_in_flight, 0);
437 device->reada_next = 0;
438 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
439 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
440
441 mutex_lock(&fs_devices->device_list_mutex);
442 list_add_rcu(&device->dev_list, &fs_devices->devices);
443 mutex_unlock(&fs_devices->device_list_mutex);
444
445 device->fs_devices = fs_devices;
446 fs_devices->num_devices++;
447 } else if (!device->name || strcmp(device->name->str, path)) {
448 name = rcu_string_strdup(path, GFP_NOFS);
449 if (!name)
450 return -ENOMEM;
451 rcu_string_free(device->name);
452 rcu_assign_pointer(device->name, name);
453 if (device->missing) {
454 fs_devices->missing_devices--;
455 device->missing = 0;
456 }
457 }
458
459 if (found_transid > fs_devices->latest_trans) {
460 fs_devices->latest_devid = devid;
461 fs_devices->latest_trans = found_transid;
462 }
463 *fs_devices_ret = fs_devices;
464 return 0;
465 }
466
467 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
468 {
469 struct btrfs_fs_devices *fs_devices;
470 struct btrfs_device *device;
471 struct btrfs_device *orig_dev;
472
473 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
474 if (!fs_devices)
475 return ERR_PTR(-ENOMEM);
476
477 INIT_LIST_HEAD(&fs_devices->devices);
478 INIT_LIST_HEAD(&fs_devices->alloc_list);
479 INIT_LIST_HEAD(&fs_devices->list);
480 mutex_init(&fs_devices->device_list_mutex);
481 fs_devices->latest_devid = orig->latest_devid;
482 fs_devices->latest_trans = orig->latest_trans;
483 fs_devices->total_devices = orig->total_devices;
484 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
485
486 /* We have held the volume lock, it is safe to get the devices. */
487 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
488 struct rcu_string *name;
489
490 device = kzalloc(sizeof(*device), GFP_NOFS);
491 if (!device)
492 goto error;
493
494 /*
495 * This is ok to do without rcu read locked because we hold the
496 * uuid mutex so nothing we touch in here is going to disappear.
497 */
498 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
499 if (!name) {
500 kfree(device);
501 goto error;
502 }
503 rcu_assign_pointer(device->name, name);
504
505 device->devid = orig_dev->devid;
506 device->work.func = pending_bios_fn;
507 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
508 spin_lock_init(&device->io_lock);
509 INIT_LIST_HEAD(&device->dev_list);
510 INIT_LIST_HEAD(&device->dev_alloc_list);
511
512 list_add(&device->dev_list, &fs_devices->devices);
513 device->fs_devices = fs_devices;
514 fs_devices->num_devices++;
515 }
516 return fs_devices;
517 error:
518 free_fs_devices(fs_devices);
519 return ERR_PTR(-ENOMEM);
520 }
521
522 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
523 struct btrfs_fs_devices *fs_devices, int step)
524 {
525 struct btrfs_device *device, *next;
526
527 struct block_device *latest_bdev = NULL;
528 u64 latest_devid = 0;
529 u64 latest_transid = 0;
530
531 mutex_lock(&uuid_mutex);
532 again:
533 /* This is the initialized path, it is safe to release the devices. */
534 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
535 if (device->in_fs_metadata) {
536 if (!device->is_tgtdev_for_dev_replace &&
537 (!latest_transid ||
538 device->generation > latest_transid)) {
539 latest_devid = device->devid;
540 latest_transid = device->generation;
541 latest_bdev = device->bdev;
542 }
543 continue;
544 }
545
546 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
547 /*
548 * In the first step, keep the device which has
549 * the correct fsid and the devid that is used
550 * for the dev_replace procedure.
551 * In the second step, the dev_replace state is
552 * read from the device tree and it is known
553 * whether the procedure is really active or
554 * not, which means whether this device is
555 * used or whether it should be removed.
556 */
557 if (step == 0 || device->is_tgtdev_for_dev_replace) {
558 continue;
559 }
560 }
561 if (device->bdev) {
562 blkdev_put(device->bdev, device->mode);
563 device->bdev = NULL;
564 fs_devices->open_devices--;
565 }
566 if (device->writeable) {
567 list_del_init(&device->dev_alloc_list);
568 device->writeable = 0;
569 if (!device->is_tgtdev_for_dev_replace)
570 fs_devices->rw_devices--;
571 }
572 list_del_init(&device->dev_list);
573 fs_devices->num_devices--;
574 rcu_string_free(device->name);
575 kfree(device);
576 }
577
578 if (fs_devices->seed) {
579 fs_devices = fs_devices->seed;
580 goto again;
581 }
582
583 fs_devices->latest_bdev = latest_bdev;
584 fs_devices->latest_devid = latest_devid;
585 fs_devices->latest_trans = latest_transid;
586
587 mutex_unlock(&uuid_mutex);
588 }
589
590 static void __free_device(struct work_struct *work)
591 {
592 struct btrfs_device *device;
593
594 device = container_of(work, struct btrfs_device, rcu_work);
595
596 if (device->bdev)
597 blkdev_put(device->bdev, device->mode);
598
599 rcu_string_free(device->name);
600 kfree(device);
601 }
602
603 static void free_device(struct rcu_head *head)
604 {
605 struct btrfs_device *device;
606
607 device = container_of(head, struct btrfs_device, rcu);
608
609 INIT_WORK(&device->rcu_work, __free_device);
610 schedule_work(&device->rcu_work);
611 }
612
613 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
614 {
615 struct btrfs_device *device;
616
617 if (--fs_devices->opened > 0)
618 return 0;
619
620 mutex_lock(&fs_devices->device_list_mutex);
621 list_for_each_entry(device, &fs_devices->devices, dev_list) {
622 struct btrfs_device *new_device;
623 struct rcu_string *name;
624
625 if (device->bdev)
626 fs_devices->open_devices--;
627
628 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
629 list_del_init(&device->dev_alloc_list);
630 fs_devices->rw_devices--;
631 }
632
633 if (device->can_discard)
634 fs_devices->num_can_discard--;
635
636 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
637 BUG_ON(!new_device); /* -ENOMEM */
638 memcpy(new_device, device, sizeof(*new_device));
639
640 /* Safe because we are under uuid_mutex */
641 if (device->name) {
642 name = rcu_string_strdup(device->name->str, GFP_NOFS);
643 BUG_ON(device->name && !name); /* -ENOMEM */
644 rcu_assign_pointer(new_device->name, name);
645 }
646 new_device->bdev = NULL;
647 new_device->writeable = 0;
648 new_device->in_fs_metadata = 0;
649 new_device->can_discard = 0;
650 list_replace_rcu(&device->dev_list, &new_device->dev_list);
651
652 call_rcu(&device->rcu, free_device);
653 }
654 mutex_unlock(&fs_devices->device_list_mutex);
655
656 WARN_ON(fs_devices->open_devices);
657 WARN_ON(fs_devices->rw_devices);
658 fs_devices->opened = 0;
659 fs_devices->seeding = 0;
660
661 return 0;
662 }
663
664 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
665 {
666 struct btrfs_fs_devices *seed_devices = NULL;
667 int ret;
668
669 mutex_lock(&uuid_mutex);
670 ret = __btrfs_close_devices(fs_devices);
671 if (!fs_devices->opened) {
672 seed_devices = fs_devices->seed;
673 fs_devices->seed = NULL;
674 }
675 mutex_unlock(&uuid_mutex);
676
677 while (seed_devices) {
678 fs_devices = seed_devices;
679 seed_devices = fs_devices->seed;
680 __btrfs_close_devices(fs_devices);
681 free_fs_devices(fs_devices);
682 }
683 return ret;
684 }
685
686 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
687 fmode_t flags, void *holder)
688 {
689 struct request_queue *q;
690 struct block_device *bdev;
691 struct list_head *head = &fs_devices->devices;
692 struct btrfs_device *device;
693 struct block_device *latest_bdev = NULL;
694 struct buffer_head *bh;
695 struct btrfs_super_block *disk_super;
696 u64 latest_devid = 0;
697 u64 latest_transid = 0;
698 u64 devid;
699 int seeding = 1;
700 int ret = 0;
701
702 flags |= FMODE_EXCL;
703
704 list_for_each_entry(device, head, dev_list) {
705 if (device->bdev)
706 continue;
707 if (!device->name)
708 continue;
709
710 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
711 &bdev, &bh);
712 if (ret)
713 continue;
714
715 disk_super = (struct btrfs_super_block *)bh->b_data;
716 devid = btrfs_stack_device_id(&disk_super->dev_item);
717 if (devid != device->devid)
718 goto error_brelse;
719
720 if (memcmp(device->uuid, disk_super->dev_item.uuid,
721 BTRFS_UUID_SIZE))
722 goto error_brelse;
723
724 device->generation = btrfs_super_generation(disk_super);
725 if (!latest_transid || device->generation > latest_transid) {
726 latest_devid = devid;
727 latest_transid = device->generation;
728 latest_bdev = bdev;
729 }
730
731 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
732 device->writeable = 0;
733 } else {
734 device->writeable = !bdev_read_only(bdev);
735 seeding = 0;
736 }
737
738 q = bdev_get_queue(bdev);
739 if (blk_queue_discard(q)) {
740 device->can_discard = 1;
741 fs_devices->num_can_discard++;
742 }
743
744 device->bdev = bdev;
745 device->in_fs_metadata = 0;
746 device->mode = flags;
747
748 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
749 fs_devices->rotating = 1;
750
751 fs_devices->open_devices++;
752 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
753 fs_devices->rw_devices++;
754 list_add(&device->dev_alloc_list,
755 &fs_devices->alloc_list);
756 }
757 brelse(bh);
758 continue;
759
760 error_brelse:
761 brelse(bh);
762 blkdev_put(bdev, flags);
763 continue;
764 }
765 if (fs_devices->open_devices == 0) {
766 ret = -EINVAL;
767 goto out;
768 }
769 fs_devices->seeding = seeding;
770 fs_devices->opened = 1;
771 fs_devices->latest_bdev = latest_bdev;
772 fs_devices->latest_devid = latest_devid;
773 fs_devices->latest_trans = latest_transid;
774 fs_devices->total_rw_bytes = 0;
775 out:
776 return ret;
777 }
778
779 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
780 fmode_t flags, void *holder)
781 {
782 int ret;
783
784 mutex_lock(&uuid_mutex);
785 if (fs_devices->opened) {
786 fs_devices->opened++;
787 ret = 0;
788 } else {
789 ret = __btrfs_open_devices(fs_devices, flags, holder);
790 }
791 mutex_unlock(&uuid_mutex);
792 return ret;
793 }
794
795 /*
796 * Look for a btrfs signature on a device. This may be called out of the mount path
797 * and we are not allowed to call set_blocksize during the scan. The superblock
798 * is read via pagecache
799 */
800 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
801 struct btrfs_fs_devices **fs_devices_ret)
802 {
803 struct btrfs_super_block *disk_super;
804 struct block_device *bdev;
805 struct page *page;
806 void *p;
807 int ret = -EINVAL;
808 u64 devid;
809 u64 transid;
810 u64 total_devices;
811 u64 bytenr;
812 pgoff_t index;
813
814 /*
815 * we would like to check all the supers, but that would make
816 * a btrfs mount succeed after a mkfs from a different FS.
817 * So, we need to add a special mount option to scan for
818 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
819 */
820 bytenr = btrfs_sb_offset(0);
821 flags |= FMODE_EXCL;
822 mutex_lock(&uuid_mutex);
823
824 bdev = blkdev_get_by_path(path, flags, holder);
825
826 if (IS_ERR(bdev)) {
827 ret = PTR_ERR(bdev);
828 printk(KERN_INFO "btrfs: open %s failed\n", path);
829 goto error;
830 }
831
832 /* make sure our super fits in the device */
833 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
834 goto error_bdev_put;
835
836 /* make sure our super fits in the page */
837 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
838 goto error_bdev_put;
839
840 /* make sure our super doesn't straddle pages on disk */
841 index = bytenr >> PAGE_CACHE_SHIFT;
842 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
843 goto error_bdev_put;
844
845 /* pull in the page with our super */
846 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
847 index, GFP_NOFS);
848
849 if (IS_ERR_OR_NULL(page))
850 goto error_bdev_put;
851
852 p = kmap(page);
853
854 /* align our pointer to the offset of the super block */
855 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
856
857 if (btrfs_super_bytenr(disk_super) != bytenr ||
858 strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
859 sizeof(disk_super->magic)))
860 goto error_unmap;
861
862 devid = btrfs_stack_device_id(&disk_super->dev_item);
863 transid = btrfs_super_generation(disk_super);
864 total_devices = btrfs_super_num_devices(disk_super);
865
866 if (disk_super->label[0]) {
867 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
868 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
869 printk(KERN_INFO "device label %s ", disk_super->label);
870 } else {
871 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
872 }
873
874 printk(KERN_CONT "devid %llu transid %llu %s\n",
875 (unsigned long long)devid, (unsigned long long)transid, path);
876
877 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
878 if (!ret && fs_devices_ret)
879 (*fs_devices_ret)->total_devices = total_devices;
880
881 error_unmap:
882 kunmap(page);
883 page_cache_release(page);
884
885 error_bdev_put:
886 blkdev_put(bdev, flags);
887 error:
888 mutex_unlock(&uuid_mutex);
889 return ret;
890 }
891
892 /* helper to account the used device space in the range */
893 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
894 u64 end, u64 *length)
895 {
896 struct btrfs_key key;
897 struct btrfs_root *root = device->dev_root;
898 struct btrfs_dev_extent *dev_extent;
899 struct btrfs_path *path;
900 u64 extent_end;
901 int ret;
902 int slot;
903 struct extent_buffer *l;
904
905 *length = 0;
906
907 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
908 return 0;
909
910 path = btrfs_alloc_path();
911 if (!path)
912 return -ENOMEM;
913 path->reada = 2;
914
915 key.objectid = device->devid;
916 key.offset = start;
917 key.type = BTRFS_DEV_EXTENT_KEY;
918
919 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
920 if (ret < 0)
921 goto out;
922 if (ret > 0) {
923 ret = btrfs_previous_item(root, path, key.objectid, key.type);
924 if (ret < 0)
925 goto out;
926 }
927
928 while (1) {
929 l = path->nodes[0];
930 slot = path->slots[0];
931 if (slot >= btrfs_header_nritems(l)) {
932 ret = btrfs_next_leaf(root, path);
933 if (ret == 0)
934 continue;
935 if (ret < 0)
936 goto out;
937
938 break;
939 }
940 btrfs_item_key_to_cpu(l, &key, slot);
941
942 if (key.objectid < device->devid)
943 goto next;
944
945 if (key.objectid > device->devid)
946 break;
947
948 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
949 goto next;
950
951 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
952 extent_end = key.offset + btrfs_dev_extent_length(l,
953 dev_extent);
954 if (key.offset <= start && extent_end > end) {
955 *length = end - start + 1;
956 break;
957 } else if (key.offset <= start && extent_end > start)
958 *length += extent_end - start;
959 else if (key.offset > start && extent_end <= end)
960 *length += extent_end - key.offset;
961 else if (key.offset > start && key.offset <= end) {
962 *length += end - key.offset + 1;
963 break;
964 } else if (key.offset > end)
965 break;
966
967 next:
968 path->slots[0]++;
969 }
970 ret = 0;
971 out:
972 btrfs_free_path(path);
973 return ret;
974 }
975
976 /*
977 * find_free_dev_extent - find free space in the specified device
978 * @device: the device which we search the free space in
979 * @num_bytes: the size of the free space that we need
980 * @start: store the start of the free space.
981 * @len: the size of the free space. that we find, or the size of the max
982 * free space if we don't find suitable free space
983 *
984 * this uses a pretty simple search, the expectation is that it is
985 * called very infrequently and that a given device has a small number
986 * of extents
987 *
988 * @start is used to store the start of the free space if we find. But if we
989 * don't find suitable free space, it will be used to store the start position
990 * of the max free space.
991 *
992 * @len is used to store the size of the free space that we find.
993 * But if we don't find suitable free space, it is used to store the size of
994 * the max free space.
995 */
996 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
997 u64 *start, u64 *len)
998 {
999 struct btrfs_key key;
1000 struct btrfs_root *root = device->dev_root;
1001 struct btrfs_dev_extent *dev_extent;
1002 struct btrfs_path *path;
1003 u64 hole_size;
1004 u64 max_hole_start;
1005 u64 max_hole_size;
1006 u64 extent_end;
1007 u64 search_start;
1008 u64 search_end = device->total_bytes;
1009 int ret;
1010 int slot;
1011 struct extent_buffer *l;
1012
1013 /* FIXME use last free of some kind */
1014
1015 /* we don't want to overwrite the superblock on the drive,
1016 * so we make sure to start at an offset of at least 1MB
1017 */
1018 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1019
1020 max_hole_start = search_start;
1021 max_hole_size = 0;
1022 hole_size = 0;
1023
1024 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1025 ret = -ENOSPC;
1026 goto error;
1027 }
1028
1029 path = btrfs_alloc_path();
1030 if (!path) {
1031 ret = -ENOMEM;
1032 goto error;
1033 }
1034 path->reada = 2;
1035
1036 key.objectid = device->devid;
1037 key.offset = search_start;
1038 key.type = BTRFS_DEV_EXTENT_KEY;
1039
1040 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1041 if (ret < 0)
1042 goto out;
1043 if (ret > 0) {
1044 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1045 if (ret < 0)
1046 goto out;
1047 }
1048
1049 while (1) {
1050 l = path->nodes[0];
1051 slot = path->slots[0];
1052 if (slot >= btrfs_header_nritems(l)) {
1053 ret = btrfs_next_leaf(root, path);
1054 if (ret == 0)
1055 continue;
1056 if (ret < 0)
1057 goto out;
1058
1059 break;
1060 }
1061 btrfs_item_key_to_cpu(l, &key, slot);
1062
1063 if (key.objectid < device->devid)
1064 goto next;
1065
1066 if (key.objectid > device->devid)
1067 break;
1068
1069 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1070 goto next;
1071
1072 if (key.offset > search_start) {
1073 hole_size = key.offset - search_start;
1074
1075 if (hole_size > max_hole_size) {
1076 max_hole_start = search_start;
1077 max_hole_size = hole_size;
1078 }
1079
1080 /*
1081 * If this free space is greater than which we need,
1082 * it must be the max free space that we have found
1083 * until now, so max_hole_start must point to the start
1084 * of this free space and the length of this free space
1085 * is stored in max_hole_size. Thus, we return
1086 * max_hole_start and max_hole_size and go back to the
1087 * caller.
1088 */
1089 if (hole_size >= num_bytes) {
1090 ret = 0;
1091 goto out;
1092 }
1093 }
1094
1095 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1096 extent_end = key.offset + btrfs_dev_extent_length(l,
1097 dev_extent);
1098 if (extent_end > search_start)
1099 search_start = extent_end;
1100 next:
1101 path->slots[0]++;
1102 cond_resched();
1103 }
1104
1105 /*
1106 * At this point, search_start should be the end of
1107 * allocated dev extents, and when shrinking the device,
1108 * search_end may be smaller than search_start.
1109 */
1110 if (search_end > search_start)
1111 hole_size = search_end - search_start;
1112
1113 if (hole_size > max_hole_size) {
1114 max_hole_start = search_start;
1115 max_hole_size = hole_size;
1116 }
1117
1118 /* See above. */
1119 if (hole_size < num_bytes)
1120 ret = -ENOSPC;
1121 else
1122 ret = 0;
1123
1124 out:
1125 btrfs_free_path(path);
1126 error:
1127 *start = max_hole_start;
1128 if (len)
1129 *len = max_hole_size;
1130 return ret;
1131 }
1132
1133 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1134 struct btrfs_device *device,
1135 u64 start)
1136 {
1137 int ret;
1138 struct btrfs_path *path;
1139 struct btrfs_root *root = device->dev_root;
1140 struct btrfs_key key;
1141 struct btrfs_key found_key;
1142 struct extent_buffer *leaf = NULL;
1143 struct btrfs_dev_extent *extent = NULL;
1144
1145 path = btrfs_alloc_path();
1146 if (!path)
1147 return -ENOMEM;
1148
1149 key.objectid = device->devid;
1150 key.offset = start;
1151 key.type = BTRFS_DEV_EXTENT_KEY;
1152 again:
1153 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1154 if (ret > 0) {
1155 ret = btrfs_previous_item(root, path, key.objectid,
1156 BTRFS_DEV_EXTENT_KEY);
1157 if (ret)
1158 goto out;
1159 leaf = path->nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1161 extent = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_dev_extent);
1163 BUG_ON(found_key.offset > start || found_key.offset +
1164 btrfs_dev_extent_length(leaf, extent) < start);
1165 key = found_key;
1166 btrfs_release_path(path);
1167 goto again;
1168 } else if (ret == 0) {
1169 leaf = path->nodes[0];
1170 extent = btrfs_item_ptr(leaf, path->slots[0],
1171 struct btrfs_dev_extent);
1172 } else {
1173 btrfs_error(root->fs_info, ret, "Slot search failed");
1174 goto out;
1175 }
1176
1177 if (device->bytes_used > 0) {
1178 u64 len = btrfs_dev_extent_length(leaf, extent);
1179 device->bytes_used -= len;
1180 spin_lock(&root->fs_info->free_chunk_lock);
1181 root->fs_info->free_chunk_space += len;
1182 spin_unlock(&root->fs_info->free_chunk_lock);
1183 }
1184 ret = btrfs_del_item(trans, root, path);
1185 if (ret) {
1186 btrfs_error(root->fs_info, ret,
1187 "Failed to remove dev extent item");
1188 }
1189 out:
1190 btrfs_free_path(path);
1191 return ret;
1192 }
1193
1194 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1195 struct btrfs_device *device,
1196 u64 chunk_tree, u64 chunk_objectid,
1197 u64 chunk_offset, u64 start, u64 num_bytes)
1198 {
1199 int ret;
1200 struct btrfs_path *path;
1201 struct btrfs_root *root = device->dev_root;
1202 struct btrfs_dev_extent *extent;
1203 struct extent_buffer *leaf;
1204 struct btrfs_key key;
1205
1206 WARN_ON(!device->in_fs_metadata);
1207 WARN_ON(device->is_tgtdev_for_dev_replace);
1208 path = btrfs_alloc_path();
1209 if (!path)
1210 return -ENOMEM;
1211
1212 key.objectid = device->devid;
1213 key.offset = start;
1214 key.type = BTRFS_DEV_EXTENT_KEY;
1215 ret = btrfs_insert_empty_item(trans, root, path, &key,
1216 sizeof(*extent));
1217 if (ret)
1218 goto out;
1219
1220 leaf = path->nodes[0];
1221 extent = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_dev_extent);
1223 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1224 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1225 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1226
1227 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1228 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1229 BTRFS_UUID_SIZE);
1230
1231 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1232 btrfs_mark_buffer_dirty(leaf);
1233 out:
1234 btrfs_free_path(path);
1235 return ret;
1236 }
1237
1238 static noinline int find_next_chunk(struct btrfs_root *root,
1239 u64 objectid, u64 *offset)
1240 {
1241 struct btrfs_path *path;
1242 int ret;
1243 struct btrfs_key key;
1244 struct btrfs_chunk *chunk;
1245 struct btrfs_key found_key;
1246
1247 path = btrfs_alloc_path();
1248 if (!path)
1249 return -ENOMEM;
1250
1251 key.objectid = objectid;
1252 key.offset = (u64)-1;
1253 key.type = BTRFS_CHUNK_ITEM_KEY;
1254
1255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1256 if (ret < 0)
1257 goto error;
1258
1259 BUG_ON(ret == 0); /* Corruption */
1260
1261 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1262 if (ret) {
1263 *offset = 0;
1264 } else {
1265 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1266 path->slots[0]);
1267 if (found_key.objectid != objectid)
1268 *offset = 0;
1269 else {
1270 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1271 struct btrfs_chunk);
1272 *offset = found_key.offset +
1273 btrfs_chunk_length(path->nodes[0], chunk);
1274 }
1275 }
1276 ret = 0;
1277 error:
1278 btrfs_free_path(path);
1279 return ret;
1280 }
1281
1282 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1283 {
1284 int ret;
1285 struct btrfs_key key;
1286 struct btrfs_key found_key;
1287 struct btrfs_path *path;
1288
1289 root = root->fs_info->chunk_root;
1290
1291 path = btrfs_alloc_path();
1292 if (!path)
1293 return -ENOMEM;
1294
1295 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1296 key.type = BTRFS_DEV_ITEM_KEY;
1297 key.offset = (u64)-1;
1298
1299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1300 if (ret < 0)
1301 goto error;
1302
1303 BUG_ON(ret == 0); /* Corruption */
1304
1305 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1306 BTRFS_DEV_ITEM_KEY);
1307 if (ret) {
1308 *objectid = 1;
1309 } else {
1310 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1311 path->slots[0]);
1312 *objectid = found_key.offset + 1;
1313 }
1314 ret = 0;
1315 error:
1316 btrfs_free_path(path);
1317 return ret;
1318 }
1319
1320 /*
1321 * the device information is stored in the chunk root
1322 * the btrfs_device struct should be fully filled in
1323 */
1324 int btrfs_add_device(struct btrfs_trans_handle *trans,
1325 struct btrfs_root *root,
1326 struct btrfs_device *device)
1327 {
1328 int ret;
1329 struct btrfs_path *path;
1330 struct btrfs_dev_item *dev_item;
1331 struct extent_buffer *leaf;
1332 struct btrfs_key key;
1333 unsigned long ptr;
1334
1335 root = root->fs_info->chunk_root;
1336
1337 path = btrfs_alloc_path();
1338 if (!path)
1339 return -ENOMEM;
1340
1341 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1342 key.type = BTRFS_DEV_ITEM_KEY;
1343 key.offset = device->devid;
1344
1345 ret = btrfs_insert_empty_item(trans, root, path, &key,
1346 sizeof(*dev_item));
1347 if (ret)
1348 goto out;
1349
1350 leaf = path->nodes[0];
1351 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1352
1353 btrfs_set_device_id(leaf, dev_item, device->devid);
1354 btrfs_set_device_generation(leaf, dev_item, 0);
1355 btrfs_set_device_type(leaf, dev_item, device->type);
1356 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1357 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1358 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1359 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1360 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1361 btrfs_set_device_group(leaf, dev_item, 0);
1362 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1363 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1364 btrfs_set_device_start_offset(leaf, dev_item, 0);
1365
1366 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1367 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1368 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1369 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1370 btrfs_mark_buffer_dirty(leaf);
1371
1372 ret = 0;
1373 out:
1374 btrfs_free_path(path);
1375 return ret;
1376 }
1377
1378 static int btrfs_rm_dev_item(struct btrfs_root *root,
1379 struct btrfs_device *device)
1380 {
1381 int ret;
1382 struct btrfs_path *path;
1383 struct btrfs_key key;
1384 struct btrfs_trans_handle *trans;
1385
1386 root = root->fs_info->chunk_root;
1387
1388 path = btrfs_alloc_path();
1389 if (!path)
1390 return -ENOMEM;
1391
1392 trans = btrfs_start_transaction(root, 0);
1393 if (IS_ERR(trans)) {
1394 btrfs_free_path(path);
1395 return PTR_ERR(trans);
1396 }
1397 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1398 key.type = BTRFS_DEV_ITEM_KEY;
1399 key.offset = device->devid;
1400 lock_chunks(root);
1401
1402 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 if (ret < 0)
1404 goto out;
1405
1406 if (ret > 0) {
1407 ret = -ENOENT;
1408 goto out;
1409 }
1410
1411 ret = btrfs_del_item(trans, root, path);
1412 if (ret)
1413 goto out;
1414 out:
1415 btrfs_free_path(path);
1416 unlock_chunks(root);
1417 btrfs_commit_transaction(trans, root);
1418 return ret;
1419 }
1420
1421 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1422 {
1423 struct btrfs_device *device;
1424 struct btrfs_device *next_device;
1425 struct block_device *bdev;
1426 struct buffer_head *bh = NULL;
1427 struct btrfs_super_block *disk_super;
1428 struct btrfs_fs_devices *cur_devices;
1429 u64 all_avail;
1430 u64 devid;
1431 u64 num_devices;
1432 u8 *dev_uuid;
1433 int ret = 0;
1434 bool clear_super = false;
1435
1436 mutex_lock(&uuid_mutex);
1437
1438 all_avail = root->fs_info->avail_data_alloc_bits |
1439 root->fs_info->avail_system_alloc_bits |
1440 root->fs_info->avail_metadata_alloc_bits;
1441
1442 num_devices = root->fs_info->fs_devices->num_devices;
1443 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1444 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1445 WARN_ON(num_devices < 1);
1446 num_devices--;
1447 }
1448 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1449
1450 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1451 printk(KERN_ERR "btrfs: unable to go below four devices "
1452 "on raid10\n");
1453 ret = -EINVAL;
1454 goto out;
1455 }
1456
1457 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1458 printk(KERN_ERR "btrfs: unable to go below two "
1459 "devices on raid1\n");
1460 ret = -EINVAL;
1461 goto out;
1462 }
1463
1464 if (strcmp(device_path, "missing") == 0) {
1465 struct list_head *devices;
1466 struct btrfs_device *tmp;
1467
1468 device = NULL;
1469 devices = &root->fs_info->fs_devices->devices;
1470 /*
1471 * It is safe to read the devices since the volume_mutex
1472 * is held.
1473 */
1474 list_for_each_entry(tmp, devices, dev_list) {
1475 if (tmp->in_fs_metadata &&
1476 !tmp->is_tgtdev_for_dev_replace &&
1477 !tmp->bdev) {
1478 device = tmp;
1479 break;
1480 }
1481 }
1482 bdev = NULL;
1483 bh = NULL;
1484 disk_super = NULL;
1485 if (!device) {
1486 printk(KERN_ERR "btrfs: no missing devices found to "
1487 "remove\n");
1488 goto out;
1489 }
1490 } else {
1491 ret = btrfs_get_bdev_and_sb(device_path,
1492 FMODE_WRITE | FMODE_EXCL,
1493 root->fs_info->bdev_holder, 0,
1494 &bdev, &bh);
1495 if (ret)
1496 goto out;
1497 disk_super = (struct btrfs_super_block *)bh->b_data;
1498 devid = btrfs_stack_device_id(&disk_super->dev_item);
1499 dev_uuid = disk_super->dev_item.uuid;
1500 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1501 disk_super->fsid);
1502 if (!device) {
1503 ret = -ENOENT;
1504 goto error_brelse;
1505 }
1506 }
1507
1508 if (device->is_tgtdev_for_dev_replace) {
1509 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1510 ret = -EINVAL;
1511 goto error_brelse;
1512 }
1513
1514 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1515 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1516 "device\n");
1517 ret = -EINVAL;
1518 goto error_brelse;
1519 }
1520
1521 if (device->writeable) {
1522 lock_chunks(root);
1523 list_del_init(&device->dev_alloc_list);
1524 unlock_chunks(root);
1525 root->fs_info->fs_devices->rw_devices--;
1526 clear_super = true;
1527 }
1528
1529 ret = btrfs_shrink_device(device, 0);
1530 if (ret)
1531 goto error_undo;
1532
1533 /*
1534 * TODO: the superblock still includes this device in its num_devices
1535 * counter although write_all_supers() is not locked out. This
1536 * could give a filesystem state which requires a degraded mount.
1537 */
1538 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1539 if (ret)
1540 goto error_undo;
1541
1542 spin_lock(&root->fs_info->free_chunk_lock);
1543 root->fs_info->free_chunk_space = device->total_bytes -
1544 device->bytes_used;
1545 spin_unlock(&root->fs_info->free_chunk_lock);
1546
1547 device->in_fs_metadata = 0;
1548 btrfs_scrub_cancel_dev(root->fs_info, device);
1549
1550 /*
1551 * the device list mutex makes sure that we don't change
1552 * the device list while someone else is writing out all
1553 * the device supers.
1554 */
1555
1556 cur_devices = device->fs_devices;
1557 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1558 list_del_rcu(&device->dev_list);
1559
1560 device->fs_devices->num_devices--;
1561 device->fs_devices->total_devices--;
1562
1563 if (device->missing)
1564 root->fs_info->fs_devices->missing_devices--;
1565
1566 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1567 struct btrfs_device, dev_list);
1568 if (device->bdev == root->fs_info->sb->s_bdev)
1569 root->fs_info->sb->s_bdev = next_device->bdev;
1570 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1571 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1572
1573 if (device->bdev)
1574 device->fs_devices->open_devices--;
1575
1576 call_rcu(&device->rcu, free_device);
1577 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1578
1579 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1580 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1581
1582 if (cur_devices->open_devices == 0) {
1583 struct btrfs_fs_devices *fs_devices;
1584 fs_devices = root->fs_info->fs_devices;
1585 while (fs_devices) {
1586 if (fs_devices->seed == cur_devices)
1587 break;
1588 fs_devices = fs_devices->seed;
1589 }
1590 fs_devices->seed = cur_devices->seed;
1591 cur_devices->seed = NULL;
1592 lock_chunks(root);
1593 __btrfs_close_devices(cur_devices);
1594 unlock_chunks(root);
1595 free_fs_devices(cur_devices);
1596 }
1597
1598 root->fs_info->num_tolerated_disk_barrier_failures =
1599 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1600
1601 /*
1602 * at this point, the device is zero sized. We want to
1603 * remove it from the devices list and zero out the old super
1604 */
1605 if (clear_super && disk_super) {
1606 /* make sure this device isn't detected as part of
1607 * the FS anymore
1608 */
1609 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1610 set_buffer_dirty(bh);
1611 sync_dirty_buffer(bh);
1612 }
1613
1614 ret = 0;
1615
1616 /* Notify udev that device has changed */
1617 if (bdev)
1618 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1619
1620 error_brelse:
1621 brelse(bh);
1622 if (bdev)
1623 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1624 out:
1625 mutex_unlock(&uuid_mutex);
1626 return ret;
1627 error_undo:
1628 if (device->writeable) {
1629 lock_chunks(root);
1630 list_add(&device->dev_alloc_list,
1631 &root->fs_info->fs_devices->alloc_list);
1632 unlock_chunks(root);
1633 root->fs_info->fs_devices->rw_devices++;
1634 }
1635 goto error_brelse;
1636 }
1637
1638 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1639 struct btrfs_device *srcdev)
1640 {
1641 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1642 list_del_rcu(&srcdev->dev_list);
1643 list_del_rcu(&srcdev->dev_alloc_list);
1644 fs_info->fs_devices->num_devices--;
1645 if (srcdev->missing) {
1646 fs_info->fs_devices->missing_devices--;
1647 fs_info->fs_devices->rw_devices++;
1648 }
1649 if (srcdev->can_discard)
1650 fs_info->fs_devices->num_can_discard--;
1651 if (srcdev->bdev)
1652 fs_info->fs_devices->open_devices--;
1653
1654 call_rcu(&srcdev->rcu, free_device);
1655 }
1656
1657 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1658 struct btrfs_device *tgtdev)
1659 {
1660 struct btrfs_device *next_device;
1661
1662 WARN_ON(!tgtdev);
1663 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1664 if (tgtdev->bdev) {
1665 btrfs_scratch_superblock(tgtdev);
1666 fs_info->fs_devices->open_devices--;
1667 }
1668 fs_info->fs_devices->num_devices--;
1669 if (tgtdev->can_discard)
1670 fs_info->fs_devices->num_can_discard++;
1671
1672 next_device = list_entry(fs_info->fs_devices->devices.next,
1673 struct btrfs_device, dev_list);
1674 if (tgtdev->bdev == fs_info->sb->s_bdev)
1675 fs_info->sb->s_bdev = next_device->bdev;
1676 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1677 fs_info->fs_devices->latest_bdev = next_device->bdev;
1678 list_del_rcu(&tgtdev->dev_list);
1679
1680 call_rcu(&tgtdev->rcu, free_device);
1681
1682 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1683 }
1684
1685 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1686 struct btrfs_device **device)
1687 {
1688 int ret = 0;
1689 struct btrfs_super_block *disk_super;
1690 u64 devid;
1691 u8 *dev_uuid;
1692 struct block_device *bdev;
1693 struct buffer_head *bh;
1694
1695 *device = NULL;
1696 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1697 root->fs_info->bdev_holder, 0, &bdev, &bh);
1698 if (ret)
1699 return ret;
1700 disk_super = (struct btrfs_super_block *)bh->b_data;
1701 devid = btrfs_stack_device_id(&disk_super->dev_item);
1702 dev_uuid = disk_super->dev_item.uuid;
1703 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1704 disk_super->fsid);
1705 brelse(bh);
1706 if (!*device)
1707 ret = -ENOENT;
1708 blkdev_put(bdev, FMODE_READ);
1709 return ret;
1710 }
1711
1712 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1713 char *device_path,
1714 struct btrfs_device **device)
1715 {
1716 *device = NULL;
1717 if (strcmp(device_path, "missing") == 0) {
1718 struct list_head *devices;
1719 struct btrfs_device *tmp;
1720
1721 devices = &root->fs_info->fs_devices->devices;
1722 /*
1723 * It is safe to read the devices since the volume_mutex
1724 * is held by the caller.
1725 */
1726 list_for_each_entry(tmp, devices, dev_list) {
1727 if (tmp->in_fs_metadata && !tmp->bdev) {
1728 *device = tmp;
1729 break;
1730 }
1731 }
1732
1733 if (!*device) {
1734 pr_err("btrfs: no missing device found\n");
1735 return -ENOENT;
1736 }
1737
1738 return 0;
1739 } else {
1740 return btrfs_find_device_by_path(root, device_path, device);
1741 }
1742 }
1743
1744 /*
1745 * does all the dirty work required for changing file system's UUID.
1746 */
1747 static int btrfs_prepare_sprout(struct btrfs_root *root)
1748 {
1749 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1750 struct btrfs_fs_devices *old_devices;
1751 struct btrfs_fs_devices *seed_devices;
1752 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1753 struct btrfs_device *device;
1754 u64 super_flags;
1755
1756 BUG_ON(!mutex_is_locked(&uuid_mutex));
1757 if (!fs_devices->seeding)
1758 return -EINVAL;
1759
1760 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1761 if (!seed_devices)
1762 return -ENOMEM;
1763
1764 old_devices = clone_fs_devices(fs_devices);
1765 if (IS_ERR(old_devices)) {
1766 kfree(seed_devices);
1767 return PTR_ERR(old_devices);
1768 }
1769
1770 list_add(&old_devices->list, &fs_uuids);
1771
1772 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1773 seed_devices->opened = 1;
1774 INIT_LIST_HEAD(&seed_devices->devices);
1775 INIT_LIST_HEAD(&seed_devices->alloc_list);
1776 mutex_init(&seed_devices->device_list_mutex);
1777
1778 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1779 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1780 synchronize_rcu);
1781 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1782
1783 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1784 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1785 device->fs_devices = seed_devices;
1786 }
1787
1788 fs_devices->seeding = 0;
1789 fs_devices->num_devices = 0;
1790 fs_devices->open_devices = 0;
1791 fs_devices->total_devices = 0;
1792 fs_devices->seed = seed_devices;
1793
1794 generate_random_uuid(fs_devices->fsid);
1795 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1796 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1797 super_flags = btrfs_super_flags(disk_super) &
1798 ~BTRFS_SUPER_FLAG_SEEDING;
1799 btrfs_set_super_flags(disk_super, super_flags);
1800
1801 return 0;
1802 }
1803
1804 /*
1805 * strore the expected generation for seed devices in device items.
1806 */
1807 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *root)
1809 {
1810 struct btrfs_path *path;
1811 struct extent_buffer *leaf;
1812 struct btrfs_dev_item *dev_item;
1813 struct btrfs_device *device;
1814 struct btrfs_key key;
1815 u8 fs_uuid[BTRFS_UUID_SIZE];
1816 u8 dev_uuid[BTRFS_UUID_SIZE];
1817 u64 devid;
1818 int ret;
1819
1820 path = btrfs_alloc_path();
1821 if (!path)
1822 return -ENOMEM;
1823
1824 root = root->fs_info->chunk_root;
1825 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1826 key.offset = 0;
1827 key.type = BTRFS_DEV_ITEM_KEY;
1828
1829 while (1) {
1830 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1831 if (ret < 0)
1832 goto error;
1833
1834 leaf = path->nodes[0];
1835 next_slot:
1836 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1837 ret = btrfs_next_leaf(root, path);
1838 if (ret > 0)
1839 break;
1840 if (ret < 0)
1841 goto error;
1842 leaf = path->nodes[0];
1843 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1844 btrfs_release_path(path);
1845 continue;
1846 }
1847
1848 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1849 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1850 key.type != BTRFS_DEV_ITEM_KEY)
1851 break;
1852
1853 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1854 struct btrfs_dev_item);
1855 devid = btrfs_device_id(leaf, dev_item);
1856 read_extent_buffer(leaf, dev_uuid,
1857 (unsigned long)btrfs_device_uuid(dev_item),
1858 BTRFS_UUID_SIZE);
1859 read_extent_buffer(leaf, fs_uuid,
1860 (unsigned long)btrfs_device_fsid(dev_item),
1861 BTRFS_UUID_SIZE);
1862 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1863 fs_uuid);
1864 BUG_ON(!device); /* Logic error */
1865
1866 if (device->fs_devices->seeding) {
1867 btrfs_set_device_generation(leaf, dev_item,
1868 device->generation);
1869 btrfs_mark_buffer_dirty(leaf);
1870 }
1871
1872 path->slots[0]++;
1873 goto next_slot;
1874 }
1875 ret = 0;
1876 error:
1877 btrfs_free_path(path);
1878 return ret;
1879 }
1880
1881 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1882 {
1883 struct request_queue *q;
1884 struct btrfs_trans_handle *trans;
1885 struct btrfs_device *device;
1886 struct block_device *bdev;
1887 struct list_head *devices;
1888 struct super_block *sb = root->fs_info->sb;
1889 struct rcu_string *name;
1890 u64 total_bytes;
1891 int seeding_dev = 0;
1892 int ret = 0;
1893
1894 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1895 return -EROFS;
1896
1897 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1898 root->fs_info->bdev_holder);
1899 if (IS_ERR(bdev))
1900 return PTR_ERR(bdev);
1901
1902 if (root->fs_info->fs_devices->seeding) {
1903 seeding_dev = 1;
1904 down_write(&sb->s_umount);
1905 mutex_lock(&uuid_mutex);
1906 }
1907
1908 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1909
1910 devices = &root->fs_info->fs_devices->devices;
1911
1912 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1913 list_for_each_entry(device, devices, dev_list) {
1914 if (device->bdev == bdev) {
1915 ret = -EEXIST;
1916 mutex_unlock(
1917 &root->fs_info->fs_devices->device_list_mutex);
1918 goto error;
1919 }
1920 }
1921 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1922
1923 device = kzalloc(sizeof(*device), GFP_NOFS);
1924 if (!device) {
1925 /* we can safely leave the fs_devices entry around */
1926 ret = -ENOMEM;
1927 goto error;
1928 }
1929
1930 name = rcu_string_strdup(device_path, GFP_NOFS);
1931 if (!name) {
1932 kfree(device);
1933 ret = -ENOMEM;
1934 goto error;
1935 }
1936 rcu_assign_pointer(device->name, name);
1937
1938 ret = find_next_devid(root, &device->devid);
1939 if (ret) {
1940 rcu_string_free(device->name);
1941 kfree(device);
1942 goto error;
1943 }
1944
1945 trans = btrfs_start_transaction(root, 0);
1946 if (IS_ERR(trans)) {
1947 rcu_string_free(device->name);
1948 kfree(device);
1949 ret = PTR_ERR(trans);
1950 goto error;
1951 }
1952
1953 lock_chunks(root);
1954
1955 q = bdev_get_queue(bdev);
1956 if (blk_queue_discard(q))
1957 device->can_discard = 1;
1958 device->writeable = 1;
1959 device->work.func = pending_bios_fn;
1960 generate_random_uuid(device->uuid);
1961 spin_lock_init(&device->io_lock);
1962 device->generation = trans->transid;
1963 device->io_width = root->sectorsize;
1964 device->io_align = root->sectorsize;
1965 device->sector_size = root->sectorsize;
1966 device->total_bytes = i_size_read(bdev->bd_inode);
1967 device->disk_total_bytes = device->total_bytes;
1968 device->dev_root = root->fs_info->dev_root;
1969 device->bdev = bdev;
1970 device->in_fs_metadata = 1;
1971 device->is_tgtdev_for_dev_replace = 0;
1972 device->mode = FMODE_EXCL;
1973 set_blocksize(device->bdev, 4096);
1974
1975 if (seeding_dev) {
1976 sb->s_flags &= ~MS_RDONLY;
1977 ret = btrfs_prepare_sprout(root);
1978 BUG_ON(ret); /* -ENOMEM */
1979 }
1980
1981 device->fs_devices = root->fs_info->fs_devices;
1982
1983 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1984 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1985 list_add(&device->dev_alloc_list,
1986 &root->fs_info->fs_devices->alloc_list);
1987 root->fs_info->fs_devices->num_devices++;
1988 root->fs_info->fs_devices->open_devices++;
1989 root->fs_info->fs_devices->rw_devices++;
1990 root->fs_info->fs_devices->total_devices++;
1991 if (device->can_discard)
1992 root->fs_info->fs_devices->num_can_discard++;
1993 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1994
1995 spin_lock(&root->fs_info->free_chunk_lock);
1996 root->fs_info->free_chunk_space += device->total_bytes;
1997 spin_unlock(&root->fs_info->free_chunk_lock);
1998
1999 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2000 root->fs_info->fs_devices->rotating = 1;
2001
2002 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2003 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2004 total_bytes + device->total_bytes);
2005
2006 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2007 btrfs_set_super_num_devices(root->fs_info->super_copy,
2008 total_bytes + 1);
2009 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2010
2011 if (seeding_dev) {
2012 ret = init_first_rw_device(trans, root, device);
2013 if (ret) {
2014 btrfs_abort_transaction(trans, root, ret);
2015 goto error_trans;
2016 }
2017 ret = btrfs_finish_sprout(trans, root);
2018 if (ret) {
2019 btrfs_abort_transaction(trans, root, ret);
2020 goto error_trans;
2021 }
2022 } else {
2023 ret = btrfs_add_device(trans, root, device);
2024 if (ret) {
2025 btrfs_abort_transaction(trans, root, ret);
2026 goto error_trans;
2027 }
2028 }
2029
2030 /*
2031 * we've got more storage, clear any full flags on the space
2032 * infos
2033 */
2034 btrfs_clear_space_info_full(root->fs_info);
2035
2036 unlock_chunks(root);
2037 root->fs_info->num_tolerated_disk_barrier_failures =
2038 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2039 ret = btrfs_commit_transaction(trans, root);
2040
2041 if (seeding_dev) {
2042 mutex_unlock(&uuid_mutex);
2043 up_write(&sb->s_umount);
2044
2045 if (ret) /* transaction commit */
2046 return ret;
2047
2048 ret = btrfs_relocate_sys_chunks(root);
2049 if (ret < 0)
2050 btrfs_error(root->fs_info, ret,
2051 "Failed to relocate sys chunks after "
2052 "device initialization. This can be fixed "
2053 "using the \"btrfs balance\" command.");
2054 trans = btrfs_attach_transaction(root);
2055 if (IS_ERR(trans)) {
2056 if (PTR_ERR(trans) == -ENOENT)
2057 return 0;
2058 return PTR_ERR(trans);
2059 }
2060 ret = btrfs_commit_transaction(trans, root);
2061 }
2062
2063 return ret;
2064
2065 error_trans:
2066 unlock_chunks(root);
2067 btrfs_end_transaction(trans, root);
2068 rcu_string_free(device->name);
2069 kfree(device);
2070 error:
2071 blkdev_put(bdev, FMODE_EXCL);
2072 if (seeding_dev) {
2073 mutex_unlock(&uuid_mutex);
2074 up_write(&sb->s_umount);
2075 }
2076 return ret;
2077 }
2078
2079 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2080 struct btrfs_device **device_out)
2081 {
2082 struct request_queue *q;
2083 struct btrfs_device *device;
2084 struct block_device *bdev;
2085 struct btrfs_fs_info *fs_info = root->fs_info;
2086 struct list_head *devices;
2087 struct rcu_string *name;
2088 int ret = 0;
2089
2090 *device_out = NULL;
2091 if (fs_info->fs_devices->seeding)
2092 return -EINVAL;
2093
2094 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2095 fs_info->bdev_holder);
2096 if (IS_ERR(bdev))
2097 return PTR_ERR(bdev);
2098
2099 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2100
2101 devices = &fs_info->fs_devices->devices;
2102 list_for_each_entry(device, devices, dev_list) {
2103 if (device->bdev == bdev) {
2104 ret = -EEXIST;
2105 goto error;
2106 }
2107 }
2108
2109 device = kzalloc(sizeof(*device), GFP_NOFS);
2110 if (!device) {
2111 ret = -ENOMEM;
2112 goto error;
2113 }
2114
2115 name = rcu_string_strdup(device_path, GFP_NOFS);
2116 if (!name) {
2117 kfree(device);
2118 ret = -ENOMEM;
2119 goto error;
2120 }
2121 rcu_assign_pointer(device->name, name);
2122
2123 q = bdev_get_queue(bdev);
2124 if (blk_queue_discard(q))
2125 device->can_discard = 1;
2126 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2127 device->writeable = 1;
2128 device->work.func = pending_bios_fn;
2129 generate_random_uuid(device->uuid);
2130 device->devid = BTRFS_DEV_REPLACE_DEVID;
2131 spin_lock_init(&device->io_lock);
2132 device->generation = 0;
2133 device->io_width = root->sectorsize;
2134 device->io_align = root->sectorsize;
2135 device->sector_size = root->sectorsize;
2136 device->total_bytes = i_size_read(bdev->bd_inode);
2137 device->disk_total_bytes = device->total_bytes;
2138 device->dev_root = fs_info->dev_root;
2139 device->bdev = bdev;
2140 device->in_fs_metadata = 1;
2141 device->is_tgtdev_for_dev_replace = 1;
2142 device->mode = FMODE_EXCL;
2143 set_blocksize(device->bdev, 4096);
2144 device->fs_devices = fs_info->fs_devices;
2145 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2146 fs_info->fs_devices->num_devices++;
2147 fs_info->fs_devices->open_devices++;
2148 if (device->can_discard)
2149 fs_info->fs_devices->num_can_discard++;
2150 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2151
2152 *device_out = device;
2153 return ret;
2154
2155 error:
2156 blkdev_put(bdev, FMODE_EXCL);
2157 return ret;
2158 }
2159
2160 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2161 struct btrfs_device *tgtdev)
2162 {
2163 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2164 tgtdev->io_width = fs_info->dev_root->sectorsize;
2165 tgtdev->io_align = fs_info->dev_root->sectorsize;
2166 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2167 tgtdev->dev_root = fs_info->dev_root;
2168 tgtdev->in_fs_metadata = 1;
2169 }
2170
2171 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2172 struct btrfs_device *device)
2173 {
2174 int ret;
2175 struct btrfs_path *path;
2176 struct btrfs_root *root;
2177 struct btrfs_dev_item *dev_item;
2178 struct extent_buffer *leaf;
2179 struct btrfs_key key;
2180
2181 root = device->dev_root->fs_info->chunk_root;
2182
2183 path = btrfs_alloc_path();
2184 if (!path)
2185 return -ENOMEM;
2186
2187 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2188 key.type = BTRFS_DEV_ITEM_KEY;
2189 key.offset = device->devid;
2190
2191 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2192 if (ret < 0)
2193 goto out;
2194
2195 if (ret > 0) {
2196 ret = -ENOENT;
2197 goto out;
2198 }
2199
2200 leaf = path->nodes[0];
2201 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2202
2203 btrfs_set_device_id(leaf, dev_item, device->devid);
2204 btrfs_set_device_type(leaf, dev_item, device->type);
2205 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2206 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2207 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2208 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2209 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2210 btrfs_mark_buffer_dirty(leaf);
2211
2212 out:
2213 btrfs_free_path(path);
2214 return ret;
2215 }
2216
2217 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2218 struct btrfs_device *device, u64 new_size)
2219 {
2220 struct btrfs_super_block *super_copy =
2221 device->dev_root->fs_info->super_copy;
2222 u64 old_total = btrfs_super_total_bytes(super_copy);
2223 u64 diff = new_size - device->total_bytes;
2224
2225 if (!device->writeable)
2226 return -EACCES;
2227 if (new_size <= device->total_bytes ||
2228 device->is_tgtdev_for_dev_replace)
2229 return -EINVAL;
2230
2231 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2232 device->fs_devices->total_rw_bytes += diff;
2233
2234 device->total_bytes = new_size;
2235 device->disk_total_bytes = new_size;
2236 btrfs_clear_space_info_full(device->dev_root->fs_info);
2237
2238 return btrfs_update_device(trans, device);
2239 }
2240
2241 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2242 struct btrfs_device *device, u64 new_size)
2243 {
2244 int ret;
2245 lock_chunks(device->dev_root);
2246 ret = __btrfs_grow_device(trans, device, new_size);
2247 unlock_chunks(device->dev_root);
2248 return ret;
2249 }
2250
2251 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2252 struct btrfs_root *root,
2253 u64 chunk_tree, u64 chunk_objectid,
2254 u64 chunk_offset)
2255 {
2256 int ret;
2257 struct btrfs_path *path;
2258 struct btrfs_key key;
2259
2260 root = root->fs_info->chunk_root;
2261 path = btrfs_alloc_path();
2262 if (!path)
2263 return -ENOMEM;
2264
2265 key.objectid = chunk_objectid;
2266 key.offset = chunk_offset;
2267 key.type = BTRFS_CHUNK_ITEM_KEY;
2268
2269 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2270 if (ret < 0)
2271 goto out;
2272 else if (ret > 0) { /* Logic error or corruption */
2273 btrfs_error(root->fs_info, -ENOENT,
2274 "Failed lookup while freeing chunk.");
2275 ret = -ENOENT;
2276 goto out;
2277 }
2278
2279 ret = btrfs_del_item(trans, root, path);
2280 if (ret < 0)
2281 btrfs_error(root->fs_info, ret,
2282 "Failed to delete chunk item.");
2283 out:
2284 btrfs_free_path(path);
2285 return ret;
2286 }
2287
2288 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2289 chunk_offset)
2290 {
2291 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2292 struct btrfs_disk_key *disk_key;
2293 struct btrfs_chunk *chunk;
2294 u8 *ptr;
2295 int ret = 0;
2296 u32 num_stripes;
2297 u32 array_size;
2298 u32 len = 0;
2299 u32 cur;
2300 struct btrfs_key key;
2301
2302 array_size = btrfs_super_sys_array_size(super_copy);
2303
2304 ptr = super_copy->sys_chunk_array;
2305 cur = 0;
2306
2307 while (cur < array_size) {
2308 disk_key = (struct btrfs_disk_key *)ptr;
2309 btrfs_disk_key_to_cpu(&key, disk_key);
2310
2311 len = sizeof(*disk_key);
2312
2313 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2314 chunk = (struct btrfs_chunk *)(ptr + len);
2315 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2316 len += btrfs_chunk_item_size(num_stripes);
2317 } else {
2318 ret = -EIO;
2319 break;
2320 }
2321 if (key.objectid == chunk_objectid &&
2322 key.offset == chunk_offset) {
2323 memmove(ptr, ptr + len, array_size - (cur + len));
2324 array_size -= len;
2325 btrfs_set_super_sys_array_size(super_copy, array_size);
2326 } else {
2327 ptr += len;
2328 cur += len;
2329 }
2330 }
2331 return ret;
2332 }
2333
2334 static int btrfs_relocate_chunk(struct btrfs_root *root,
2335 u64 chunk_tree, u64 chunk_objectid,
2336 u64 chunk_offset)
2337 {
2338 struct extent_map_tree *em_tree;
2339 struct btrfs_root *extent_root;
2340 struct btrfs_trans_handle *trans;
2341 struct extent_map *em;
2342 struct map_lookup *map;
2343 int ret;
2344 int i;
2345
2346 root = root->fs_info->chunk_root;
2347 extent_root = root->fs_info->extent_root;
2348 em_tree = &root->fs_info->mapping_tree.map_tree;
2349
2350 ret = btrfs_can_relocate(extent_root, chunk_offset);
2351 if (ret)
2352 return -ENOSPC;
2353
2354 /* step one, relocate all the extents inside this chunk */
2355 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2356 if (ret)
2357 return ret;
2358
2359 trans = btrfs_start_transaction(root, 0);
2360 BUG_ON(IS_ERR(trans));
2361
2362 lock_chunks(root);
2363
2364 /*
2365 * step two, delete the device extents and the
2366 * chunk tree entries
2367 */
2368 read_lock(&em_tree->lock);
2369 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2370 read_unlock(&em_tree->lock);
2371
2372 BUG_ON(!em || em->start > chunk_offset ||
2373 em->start + em->len < chunk_offset);
2374 map = (struct map_lookup *)em->bdev;
2375
2376 for (i = 0; i < map->num_stripes; i++) {
2377 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2378 map->stripes[i].physical);
2379 BUG_ON(ret);
2380
2381 if (map->stripes[i].dev) {
2382 ret = btrfs_update_device(trans, map->stripes[i].dev);
2383 BUG_ON(ret);
2384 }
2385 }
2386 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2387 chunk_offset);
2388
2389 BUG_ON(ret);
2390
2391 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2392
2393 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2394 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2395 BUG_ON(ret);
2396 }
2397
2398 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2399 BUG_ON(ret);
2400
2401 write_lock(&em_tree->lock);
2402 remove_extent_mapping(em_tree, em);
2403 write_unlock(&em_tree->lock);
2404
2405 kfree(map);
2406 em->bdev = NULL;
2407
2408 /* once for the tree */
2409 free_extent_map(em);
2410 /* once for us */
2411 free_extent_map(em);
2412
2413 unlock_chunks(root);
2414 btrfs_end_transaction(trans, root);
2415 return 0;
2416 }
2417
2418 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2419 {
2420 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2421 struct btrfs_path *path;
2422 struct extent_buffer *leaf;
2423 struct btrfs_chunk *chunk;
2424 struct btrfs_key key;
2425 struct btrfs_key found_key;
2426 u64 chunk_tree = chunk_root->root_key.objectid;
2427 u64 chunk_type;
2428 bool retried = false;
2429 int failed = 0;
2430 int ret;
2431
2432 path = btrfs_alloc_path();
2433 if (!path)
2434 return -ENOMEM;
2435
2436 again:
2437 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2438 key.offset = (u64)-1;
2439 key.type = BTRFS_CHUNK_ITEM_KEY;
2440
2441 while (1) {
2442 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2443 if (ret < 0)
2444 goto error;
2445 BUG_ON(ret == 0); /* Corruption */
2446
2447 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2448 key.type);
2449 if (ret < 0)
2450 goto error;
2451 if (ret > 0)
2452 break;
2453
2454 leaf = path->nodes[0];
2455 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2456
2457 chunk = btrfs_item_ptr(leaf, path->slots[0],
2458 struct btrfs_chunk);
2459 chunk_type = btrfs_chunk_type(leaf, chunk);
2460 btrfs_release_path(path);
2461
2462 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2463 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2464 found_key.objectid,
2465 found_key.offset);
2466 if (ret == -ENOSPC)
2467 failed++;
2468 else if (ret)
2469 BUG();
2470 }
2471
2472 if (found_key.offset == 0)
2473 break;
2474 key.offset = found_key.offset - 1;
2475 }
2476 ret = 0;
2477 if (failed && !retried) {
2478 failed = 0;
2479 retried = true;
2480 goto again;
2481 } else if (failed && retried) {
2482 WARN_ON(1);
2483 ret = -ENOSPC;
2484 }
2485 error:
2486 btrfs_free_path(path);
2487 return ret;
2488 }
2489
2490 static int insert_balance_item(struct btrfs_root *root,
2491 struct btrfs_balance_control *bctl)
2492 {
2493 struct btrfs_trans_handle *trans;
2494 struct btrfs_balance_item *item;
2495 struct btrfs_disk_balance_args disk_bargs;
2496 struct btrfs_path *path;
2497 struct extent_buffer *leaf;
2498 struct btrfs_key key;
2499 int ret, err;
2500
2501 path = btrfs_alloc_path();
2502 if (!path)
2503 return -ENOMEM;
2504
2505 trans = btrfs_start_transaction(root, 0);
2506 if (IS_ERR(trans)) {
2507 btrfs_free_path(path);
2508 return PTR_ERR(trans);
2509 }
2510
2511 key.objectid = BTRFS_BALANCE_OBJECTID;
2512 key.type = BTRFS_BALANCE_ITEM_KEY;
2513 key.offset = 0;
2514
2515 ret = btrfs_insert_empty_item(trans, root, path, &key,
2516 sizeof(*item));
2517 if (ret)
2518 goto out;
2519
2520 leaf = path->nodes[0];
2521 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2522
2523 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2524
2525 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2526 btrfs_set_balance_data(leaf, item, &disk_bargs);
2527 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2528 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2529 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2530 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2531
2532 btrfs_set_balance_flags(leaf, item, bctl->flags);
2533
2534 btrfs_mark_buffer_dirty(leaf);
2535 out:
2536 btrfs_free_path(path);
2537 err = btrfs_commit_transaction(trans, root);
2538 if (err && !ret)
2539 ret = err;
2540 return ret;
2541 }
2542
2543 static int del_balance_item(struct btrfs_root *root)
2544 {
2545 struct btrfs_trans_handle *trans;
2546 struct btrfs_path *path;
2547 struct btrfs_key key;
2548 int ret, err;
2549
2550 path = btrfs_alloc_path();
2551 if (!path)
2552 return -ENOMEM;
2553
2554 trans = btrfs_start_transaction(root, 0);
2555 if (IS_ERR(trans)) {
2556 btrfs_free_path(path);
2557 return PTR_ERR(trans);
2558 }
2559
2560 key.objectid = BTRFS_BALANCE_OBJECTID;
2561 key.type = BTRFS_BALANCE_ITEM_KEY;
2562 key.offset = 0;
2563
2564 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2565 if (ret < 0)
2566 goto out;
2567 if (ret > 0) {
2568 ret = -ENOENT;
2569 goto out;
2570 }
2571
2572 ret = btrfs_del_item(trans, root, path);
2573 out:
2574 btrfs_free_path(path);
2575 err = btrfs_commit_transaction(trans, root);
2576 if (err && !ret)
2577 ret = err;
2578 return ret;
2579 }
2580
2581 /*
2582 * This is a heuristic used to reduce the number of chunks balanced on
2583 * resume after balance was interrupted.
2584 */
2585 static void update_balance_args(struct btrfs_balance_control *bctl)
2586 {
2587 /*
2588 * Turn on soft mode for chunk types that were being converted.
2589 */
2590 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2591 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2592 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2593 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2594 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2595 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2596
2597 /*
2598 * Turn on usage filter if is not already used. The idea is
2599 * that chunks that we have already balanced should be
2600 * reasonably full. Don't do it for chunks that are being
2601 * converted - that will keep us from relocating unconverted
2602 * (albeit full) chunks.
2603 */
2604 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2605 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2606 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2607 bctl->data.usage = 90;
2608 }
2609 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2610 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2611 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2612 bctl->sys.usage = 90;
2613 }
2614 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2615 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2616 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2617 bctl->meta.usage = 90;
2618 }
2619 }
2620
2621 /*
2622 * Should be called with both balance and volume mutexes held to
2623 * serialize other volume operations (add_dev/rm_dev/resize) with
2624 * restriper. Same goes for unset_balance_control.
2625 */
2626 static void set_balance_control(struct btrfs_balance_control *bctl)
2627 {
2628 struct btrfs_fs_info *fs_info = bctl->fs_info;
2629
2630 BUG_ON(fs_info->balance_ctl);
2631
2632 spin_lock(&fs_info->balance_lock);
2633 fs_info->balance_ctl = bctl;
2634 spin_unlock(&fs_info->balance_lock);
2635 }
2636
2637 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2638 {
2639 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2640
2641 BUG_ON(!fs_info->balance_ctl);
2642
2643 spin_lock(&fs_info->balance_lock);
2644 fs_info->balance_ctl = NULL;
2645 spin_unlock(&fs_info->balance_lock);
2646
2647 kfree(bctl);
2648 }
2649
2650 /*
2651 * Balance filters. Return 1 if chunk should be filtered out
2652 * (should not be balanced).
2653 */
2654 static int chunk_profiles_filter(u64 chunk_type,
2655 struct btrfs_balance_args *bargs)
2656 {
2657 chunk_type = chunk_to_extended(chunk_type) &
2658 BTRFS_EXTENDED_PROFILE_MASK;
2659
2660 if (bargs->profiles & chunk_type)
2661 return 0;
2662
2663 return 1;
2664 }
2665
2666 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2667 struct btrfs_balance_args *bargs)
2668 {
2669 struct btrfs_block_group_cache *cache;
2670 u64 chunk_used, user_thresh;
2671 int ret = 1;
2672
2673 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2674 chunk_used = btrfs_block_group_used(&cache->item);
2675
2676 if (bargs->usage == 0)
2677 user_thresh = 0;
2678 else if (bargs->usage > 100)
2679 user_thresh = cache->key.offset;
2680 else
2681 user_thresh = div_factor_fine(cache->key.offset,
2682 bargs->usage);
2683
2684 if (chunk_used < user_thresh)
2685 ret = 0;
2686
2687 btrfs_put_block_group(cache);
2688 return ret;
2689 }
2690
2691 static int chunk_devid_filter(struct extent_buffer *leaf,
2692 struct btrfs_chunk *chunk,
2693 struct btrfs_balance_args *bargs)
2694 {
2695 struct btrfs_stripe *stripe;
2696 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2697 int i;
2698
2699 for (i = 0; i < num_stripes; i++) {
2700 stripe = btrfs_stripe_nr(chunk, i);
2701 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2702 return 0;
2703 }
2704
2705 return 1;
2706 }
2707
2708 /* [pstart, pend) */
2709 static int chunk_drange_filter(struct extent_buffer *leaf,
2710 struct btrfs_chunk *chunk,
2711 u64 chunk_offset,
2712 struct btrfs_balance_args *bargs)
2713 {
2714 struct btrfs_stripe *stripe;
2715 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2716 u64 stripe_offset;
2717 u64 stripe_length;
2718 int factor;
2719 int i;
2720
2721 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2722 return 0;
2723
2724 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2725 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2726 factor = 2;
2727 else
2728 factor = 1;
2729 factor = num_stripes / factor;
2730
2731 for (i = 0; i < num_stripes; i++) {
2732 stripe = btrfs_stripe_nr(chunk, i);
2733 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2734 continue;
2735
2736 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2737 stripe_length = btrfs_chunk_length(leaf, chunk);
2738 do_div(stripe_length, factor);
2739
2740 if (stripe_offset < bargs->pend &&
2741 stripe_offset + stripe_length > bargs->pstart)
2742 return 0;
2743 }
2744
2745 return 1;
2746 }
2747
2748 /* [vstart, vend) */
2749 static int chunk_vrange_filter(struct extent_buffer *leaf,
2750 struct btrfs_chunk *chunk,
2751 u64 chunk_offset,
2752 struct btrfs_balance_args *bargs)
2753 {
2754 if (chunk_offset < bargs->vend &&
2755 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2756 /* at least part of the chunk is inside this vrange */
2757 return 0;
2758
2759 return 1;
2760 }
2761
2762 static int chunk_soft_convert_filter(u64 chunk_type,
2763 struct btrfs_balance_args *bargs)
2764 {
2765 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2766 return 0;
2767
2768 chunk_type = chunk_to_extended(chunk_type) &
2769 BTRFS_EXTENDED_PROFILE_MASK;
2770
2771 if (bargs->target == chunk_type)
2772 return 1;
2773
2774 return 0;
2775 }
2776
2777 static int should_balance_chunk(struct btrfs_root *root,
2778 struct extent_buffer *leaf,
2779 struct btrfs_chunk *chunk, u64 chunk_offset)
2780 {
2781 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2782 struct btrfs_balance_args *bargs = NULL;
2783 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2784
2785 /* type filter */
2786 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2787 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2788 return 0;
2789 }
2790
2791 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2792 bargs = &bctl->data;
2793 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2794 bargs = &bctl->sys;
2795 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2796 bargs = &bctl->meta;
2797
2798 /* profiles filter */
2799 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2800 chunk_profiles_filter(chunk_type, bargs)) {
2801 return 0;
2802 }
2803
2804 /* usage filter */
2805 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2806 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2807 return 0;
2808 }
2809
2810 /* devid filter */
2811 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2812 chunk_devid_filter(leaf, chunk, bargs)) {
2813 return 0;
2814 }
2815
2816 /* drange filter, makes sense only with devid filter */
2817 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2818 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2819 return 0;
2820 }
2821
2822 /* vrange filter */
2823 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2824 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2825 return 0;
2826 }
2827
2828 /* soft profile changing mode */
2829 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2830 chunk_soft_convert_filter(chunk_type, bargs)) {
2831 return 0;
2832 }
2833
2834 return 1;
2835 }
2836
2837 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2838 {
2839 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2840 struct btrfs_root *chunk_root = fs_info->chunk_root;
2841 struct btrfs_root *dev_root = fs_info->dev_root;
2842 struct list_head *devices;
2843 struct btrfs_device *device;
2844 u64 old_size;
2845 u64 size_to_free;
2846 struct btrfs_chunk *chunk;
2847 struct btrfs_path *path;
2848 struct btrfs_key key;
2849 struct btrfs_key found_key;
2850 struct btrfs_trans_handle *trans;
2851 struct extent_buffer *leaf;
2852 int slot;
2853 int ret;
2854 int enospc_errors = 0;
2855 bool counting = true;
2856
2857 /* step one make some room on all the devices */
2858 devices = &fs_info->fs_devices->devices;
2859 list_for_each_entry(device, devices, dev_list) {
2860 old_size = device->total_bytes;
2861 size_to_free = div_factor(old_size, 1);
2862 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2863 if (!device->writeable ||
2864 device->total_bytes - device->bytes_used > size_to_free ||
2865 device->is_tgtdev_for_dev_replace)
2866 continue;
2867
2868 ret = btrfs_shrink_device(device, old_size - size_to_free);
2869 if (ret == -ENOSPC)
2870 break;
2871 BUG_ON(ret);
2872
2873 trans = btrfs_start_transaction(dev_root, 0);
2874 BUG_ON(IS_ERR(trans));
2875
2876 ret = btrfs_grow_device(trans, device, old_size);
2877 BUG_ON(ret);
2878
2879 btrfs_end_transaction(trans, dev_root);
2880 }
2881
2882 /* step two, relocate all the chunks */
2883 path = btrfs_alloc_path();
2884 if (!path) {
2885 ret = -ENOMEM;
2886 goto error;
2887 }
2888
2889 /* zero out stat counters */
2890 spin_lock(&fs_info->balance_lock);
2891 memset(&bctl->stat, 0, sizeof(bctl->stat));
2892 spin_unlock(&fs_info->balance_lock);
2893 again:
2894 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2895 key.offset = (u64)-1;
2896 key.type = BTRFS_CHUNK_ITEM_KEY;
2897
2898 while (1) {
2899 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2900 atomic_read(&fs_info->balance_cancel_req)) {
2901 ret = -ECANCELED;
2902 goto error;
2903 }
2904
2905 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2906 if (ret < 0)
2907 goto error;
2908
2909 /*
2910 * this shouldn't happen, it means the last relocate
2911 * failed
2912 */
2913 if (ret == 0)
2914 BUG(); /* FIXME break ? */
2915
2916 ret = btrfs_previous_item(chunk_root, path, 0,
2917 BTRFS_CHUNK_ITEM_KEY);
2918 if (ret) {
2919 ret = 0;
2920 break;
2921 }
2922
2923 leaf = path->nodes[0];
2924 slot = path->slots[0];
2925 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2926
2927 if (found_key.objectid != key.objectid)
2928 break;
2929
2930 /* chunk zero is special */
2931 if (found_key.offset == 0)
2932 break;
2933
2934 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2935
2936 if (!counting) {
2937 spin_lock(&fs_info->balance_lock);
2938 bctl->stat.considered++;
2939 spin_unlock(&fs_info->balance_lock);
2940 }
2941
2942 ret = should_balance_chunk(chunk_root, leaf, chunk,
2943 found_key.offset);
2944 btrfs_release_path(path);
2945 if (!ret)
2946 goto loop;
2947
2948 if (counting) {
2949 spin_lock(&fs_info->balance_lock);
2950 bctl->stat.expected++;
2951 spin_unlock(&fs_info->balance_lock);
2952 goto loop;
2953 }
2954
2955 ret = btrfs_relocate_chunk(chunk_root,
2956 chunk_root->root_key.objectid,
2957 found_key.objectid,
2958 found_key.offset);
2959 if (ret && ret != -ENOSPC)
2960 goto error;
2961 if (ret == -ENOSPC) {
2962 enospc_errors++;
2963 } else {
2964 spin_lock(&fs_info->balance_lock);
2965 bctl->stat.completed++;
2966 spin_unlock(&fs_info->balance_lock);
2967 }
2968 loop:
2969 key.offset = found_key.offset - 1;
2970 }
2971
2972 if (counting) {
2973 btrfs_release_path(path);
2974 counting = false;
2975 goto again;
2976 }
2977 error:
2978 btrfs_free_path(path);
2979 if (enospc_errors) {
2980 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2981 enospc_errors);
2982 if (!ret)
2983 ret = -ENOSPC;
2984 }
2985
2986 return ret;
2987 }
2988
2989 /**
2990 * alloc_profile_is_valid - see if a given profile is valid and reduced
2991 * @flags: profile to validate
2992 * @extended: if true @flags is treated as an extended profile
2993 */
2994 static int alloc_profile_is_valid(u64 flags, int extended)
2995 {
2996 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2997 BTRFS_BLOCK_GROUP_PROFILE_MASK);
2998
2999 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3000
3001 /* 1) check that all other bits are zeroed */
3002 if (flags & ~mask)
3003 return 0;
3004
3005 /* 2) see if profile is reduced */
3006 if (flags == 0)
3007 return !extended; /* "0" is valid for usual profiles */
3008
3009 /* true if exactly one bit set */
3010 return (flags & (flags - 1)) == 0;
3011 }
3012
3013 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3014 {
3015 /* cancel requested || normal exit path */
3016 return atomic_read(&fs_info->balance_cancel_req) ||
3017 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3018 atomic_read(&fs_info->balance_cancel_req) == 0);
3019 }
3020
3021 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3022 {
3023 int ret;
3024
3025 unset_balance_control(fs_info);
3026 ret = del_balance_item(fs_info->tree_root);
3027 BUG_ON(ret);
3028
3029 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3030 }
3031
3032 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3033 struct btrfs_ioctl_balance_args *bargs);
3034
3035 /*
3036 * Should be called with both balance and volume mutexes held
3037 */
3038 int btrfs_balance(struct btrfs_balance_control *bctl,
3039 struct btrfs_ioctl_balance_args *bargs)
3040 {
3041 struct btrfs_fs_info *fs_info = bctl->fs_info;
3042 u64 allowed;
3043 int mixed = 0;
3044 int ret;
3045 u64 num_devices;
3046
3047 if (btrfs_fs_closing(fs_info) ||
3048 atomic_read(&fs_info->balance_pause_req) ||
3049 atomic_read(&fs_info->balance_cancel_req)) {
3050 ret = -EINVAL;
3051 goto out;
3052 }
3053
3054 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3055 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3056 mixed = 1;
3057
3058 /*
3059 * In case of mixed groups both data and meta should be picked,
3060 * and identical options should be given for both of them.
3061 */
3062 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3063 if (mixed && (bctl->flags & allowed)) {
3064 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3065 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3066 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3067 printk(KERN_ERR "btrfs: with mixed groups data and "
3068 "metadata balance options must be the same\n");
3069 ret = -EINVAL;
3070 goto out;
3071 }
3072 }
3073
3074 num_devices = fs_info->fs_devices->num_devices;
3075 btrfs_dev_replace_lock(&fs_info->dev_replace);
3076 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3077 BUG_ON(num_devices < 1);
3078 num_devices--;
3079 }
3080 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3081 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3082 if (num_devices == 1)
3083 allowed |= BTRFS_BLOCK_GROUP_DUP;
3084 else if (num_devices < 4)
3085 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3086 else
3087 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3088 BTRFS_BLOCK_GROUP_RAID10);
3089
3090 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3091 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3092 (bctl->data.target & ~allowed))) {
3093 printk(KERN_ERR "btrfs: unable to start balance with target "
3094 "data profile %llu\n",
3095 (unsigned long long)bctl->data.target);
3096 ret = -EINVAL;
3097 goto out;
3098 }
3099 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3100 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3101 (bctl->meta.target & ~allowed))) {
3102 printk(KERN_ERR "btrfs: unable to start balance with target "
3103 "metadata profile %llu\n",
3104 (unsigned long long)bctl->meta.target);
3105 ret = -EINVAL;
3106 goto out;
3107 }
3108 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3109 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3110 (bctl->sys.target & ~allowed))) {
3111 printk(KERN_ERR "btrfs: unable to start balance with target "
3112 "system profile %llu\n",
3113 (unsigned long long)bctl->sys.target);
3114 ret = -EINVAL;
3115 goto out;
3116 }
3117
3118 /* allow dup'ed data chunks only in mixed mode */
3119 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3120 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3121 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3122 ret = -EINVAL;
3123 goto out;
3124 }
3125
3126 /* allow to reduce meta or sys integrity only if force set */
3127 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3128 BTRFS_BLOCK_GROUP_RAID10;
3129 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3130 (fs_info->avail_system_alloc_bits & allowed) &&
3131 !(bctl->sys.target & allowed)) ||
3132 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3133 (fs_info->avail_metadata_alloc_bits & allowed) &&
3134 !(bctl->meta.target & allowed))) {
3135 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3136 printk(KERN_INFO "btrfs: force reducing metadata "
3137 "integrity\n");
3138 } else {
3139 printk(KERN_ERR "btrfs: balance will reduce metadata "
3140 "integrity, use force if you want this\n");
3141 ret = -EINVAL;
3142 goto out;
3143 }
3144 }
3145
3146 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3147 int num_tolerated_disk_barrier_failures;
3148 u64 target = bctl->sys.target;
3149
3150 num_tolerated_disk_barrier_failures =
3151 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3152 if (num_tolerated_disk_barrier_failures > 0 &&
3153 (target &
3154 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3155 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3156 num_tolerated_disk_barrier_failures = 0;
3157 else if (num_tolerated_disk_barrier_failures > 1 &&
3158 (target &
3159 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3160 num_tolerated_disk_barrier_failures = 1;
3161
3162 fs_info->num_tolerated_disk_barrier_failures =
3163 num_tolerated_disk_barrier_failures;
3164 }
3165
3166 ret = insert_balance_item(fs_info->tree_root, bctl);
3167 if (ret && ret != -EEXIST)
3168 goto out;
3169
3170 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3171 BUG_ON(ret == -EEXIST);
3172 set_balance_control(bctl);
3173 } else {
3174 BUG_ON(ret != -EEXIST);
3175 spin_lock(&fs_info->balance_lock);
3176 update_balance_args(bctl);
3177 spin_unlock(&fs_info->balance_lock);
3178 }
3179
3180 atomic_inc(&fs_info->balance_running);
3181 mutex_unlock(&fs_info->balance_mutex);
3182
3183 ret = __btrfs_balance(fs_info);
3184
3185 mutex_lock(&fs_info->balance_mutex);
3186 atomic_dec(&fs_info->balance_running);
3187
3188 if (bargs) {
3189 memset(bargs, 0, sizeof(*bargs));
3190 update_ioctl_balance_args(fs_info, 0, bargs);
3191 }
3192
3193 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3194 balance_need_close(fs_info)) {
3195 __cancel_balance(fs_info);
3196 }
3197
3198 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3199 fs_info->num_tolerated_disk_barrier_failures =
3200 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3201 }
3202
3203 wake_up(&fs_info->balance_wait_q);
3204
3205 return ret;
3206 out:
3207 if (bctl->flags & BTRFS_BALANCE_RESUME)
3208 __cancel_balance(fs_info);
3209 else {
3210 kfree(bctl);
3211 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3212 }
3213 return ret;
3214 }
3215
3216 static int balance_kthread(void *data)
3217 {
3218 struct btrfs_fs_info *fs_info = data;
3219 int ret = 0;
3220
3221 mutex_lock(&fs_info->volume_mutex);
3222 mutex_lock(&fs_info->balance_mutex);
3223
3224 if (fs_info->balance_ctl) {
3225 printk(KERN_INFO "btrfs: continuing balance\n");
3226 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3227 }
3228
3229 mutex_unlock(&fs_info->balance_mutex);
3230 mutex_unlock(&fs_info->volume_mutex);
3231
3232 return ret;
3233 }
3234
3235 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3236 {
3237 struct task_struct *tsk;
3238
3239 spin_lock(&fs_info->balance_lock);
3240 if (!fs_info->balance_ctl) {
3241 spin_unlock(&fs_info->balance_lock);
3242 return 0;
3243 }
3244 spin_unlock(&fs_info->balance_lock);
3245
3246 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3247 printk(KERN_INFO "btrfs: force skipping balance\n");
3248 return 0;
3249 }
3250
3251 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3252 if (IS_ERR(tsk))
3253 return PTR_ERR(tsk);
3254
3255 return 0;
3256 }
3257
3258 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3259 {
3260 struct btrfs_balance_control *bctl;
3261 struct btrfs_balance_item *item;
3262 struct btrfs_disk_balance_args disk_bargs;
3263 struct btrfs_path *path;
3264 struct extent_buffer *leaf;
3265 struct btrfs_key key;
3266 int ret;
3267
3268 path = btrfs_alloc_path();
3269 if (!path)
3270 return -ENOMEM;
3271
3272 key.objectid = BTRFS_BALANCE_OBJECTID;
3273 key.type = BTRFS_BALANCE_ITEM_KEY;
3274 key.offset = 0;
3275
3276 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3277 if (ret < 0)
3278 goto out;
3279 if (ret > 0) { /* ret = -ENOENT; */
3280 ret = 0;
3281 goto out;
3282 }
3283
3284 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3285 if (!bctl) {
3286 ret = -ENOMEM;
3287 goto out;
3288 }
3289
3290 leaf = path->nodes[0];
3291 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3292
3293 bctl->fs_info = fs_info;
3294 bctl->flags = btrfs_balance_flags(leaf, item);
3295 bctl->flags |= BTRFS_BALANCE_RESUME;
3296
3297 btrfs_balance_data(leaf, item, &disk_bargs);
3298 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3299 btrfs_balance_meta(leaf, item, &disk_bargs);
3300 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3301 btrfs_balance_sys(leaf, item, &disk_bargs);
3302 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3303
3304 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3305
3306 mutex_lock(&fs_info->volume_mutex);
3307 mutex_lock(&fs_info->balance_mutex);
3308
3309 set_balance_control(bctl);
3310
3311 mutex_unlock(&fs_info->balance_mutex);
3312 mutex_unlock(&fs_info->volume_mutex);
3313 out:
3314 btrfs_free_path(path);
3315 return ret;
3316 }
3317
3318 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3319 {
3320 int ret = 0;
3321
3322 mutex_lock(&fs_info->balance_mutex);
3323 if (!fs_info->balance_ctl) {
3324 mutex_unlock(&fs_info->balance_mutex);
3325 return -ENOTCONN;
3326 }
3327
3328 if (atomic_read(&fs_info->balance_running)) {
3329 atomic_inc(&fs_info->balance_pause_req);
3330 mutex_unlock(&fs_info->balance_mutex);
3331
3332 wait_event(fs_info->balance_wait_q,
3333 atomic_read(&fs_info->balance_running) == 0);
3334
3335 mutex_lock(&fs_info->balance_mutex);
3336 /* we are good with balance_ctl ripped off from under us */
3337 BUG_ON(atomic_read(&fs_info->balance_running));
3338 atomic_dec(&fs_info->balance_pause_req);
3339 } else {
3340 ret = -ENOTCONN;
3341 }
3342
3343 mutex_unlock(&fs_info->balance_mutex);
3344 return ret;
3345 }
3346
3347 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3348 {
3349 mutex_lock(&fs_info->balance_mutex);
3350 if (!fs_info->balance_ctl) {
3351 mutex_unlock(&fs_info->balance_mutex);
3352 return -ENOTCONN;
3353 }
3354
3355 atomic_inc(&fs_info->balance_cancel_req);
3356 /*
3357 * if we are running just wait and return, balance item is
3358 * deleted in btrfs_balance in this case
3359 */
3360 if (atomic_read(&fs_info->balance_running)) {
3361 mutex_unlock(&fs_info->balance_mutex);
3362 wait_event(fs_info->balance_wait_q,
3363 atomic_read(&fs_info->balance_running) == 0);
3364 mutex_lock(&fs_info->balance_mutex);
3365 } else {
3366 /* __cancel_balance needs volume_mutex */
3367 mutex_unlock(&fs_info->balance_mutex);
3368 mutex_lock(&fs_info->volume_mutex);
3369 mutex_lock(&fs_info->balance_mutex);
3370
3371 if (fs_info->balance_ctl)
3372 __cancel_balance(fs_info);
3373
3374 mutex_unlock(&fs_info->volume_mutex);
3375 }
3376
3377 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3378 atomic_dec(&fs_info->balance_cancel_req);
3379 mutex_unlock(&fs_info->balance_mutex);
3380 return 0;
3381 }
3382
3383 /*
3384 * shrinking a device means finding all of the device extents past
3385 * the new size, and then following the back refs to the chunks.
3386 * The chunk relocation code actually frees the device extent
3387 */
3388 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3389 {
3390 struct btrfs_trans_handle *trans;
3391 struct btrfs_root *root = device->dev_root;
3392 struct btrfs_dev_extent *dev_extent = NULL;
3393 struct btrfs_path *path;
3394 u64 length;
3395 u64 chunk_tree;
3396 u64 chunk_objectid;
3397 u64 chunk_offset;
3398 int ret;
3399 int slot;
3400 int failed = 0;
3401 bool retried = false;
3402 struct extent_buffer *l;
3403 struct btrfs_key key;
3404 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3405 u64 old_total = btrfs_super_total_bytes(super_copy);
3406 u64 old_size = device->total_bytes;
3407 u64 diff = device->total_bytes - new_size;
3408
3409 if (device->is_tgtdev_for_dev_replace)
3410 return -EINVAL;
3411
3412 path = btrfs_alloc_path();
3413 if (!path)
3414 return -ENOMEM;
3415
3416 path->reada = 2;
3417
3418 lock_chunks(root);
3419
3420 device->total_bytes = new_size;
3421 if (device->writeable) {
3422 device->fs_devices->total_rw_bytes -= diff;
3423 spin_lock(&root->fs_info->free_chunk_lock);
3424 root->fs_info->free_chunk_space -= diff;
3425 spin_unlock(&root->fs_info->free_chunk_lock);
3426 }
3427 unlock_chunks(root);
3428
3429 again:
3430 key.objectid = device->devid;
3431 key.offset = (u64)-1;
3432 key.type = BTRFS_DEV_EXTENT_KEY;
3433
3434 do {
3435 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3436 if (ret < 0)
3437 goto done;
3438
3439 ret = btrfs_previous_item(root, path, 0, key.type);
3440 if (ret < 0)
3441 goto done;
3442 if (ret) {
3443 ret = 0;
3444 btrfs_release_path(path);
3445 break;
3446 }
3447
3448 l = path->nodes[0];
3449 slot = path->slots[0];
3450 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3451
3452 if (key.objectid != device->devid) {
3453 btrfs_release_path(path);
3454 break;
3455 }
3456
3457 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3458 length = btrfs_dev_extent_length(l, dev_extent);
3459
3460 if (key.offset + length <= new_size) {
3461 btrfs_release_path(path);
3462 break;
3463 }
3464
3465 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3466 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3467 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3468 btrfs_release_path(path);
3469
3470 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3471 chunk_offset);
3472 if (ret && ret != -ENOSPC)
3473 goto done;
3474 if (ret == -ENOSPC)
3475 failed++;
3476 } while (key.offset-- > 0);
3477
3478 if (failed && !retried) {
3479 failed = 0;
3480 retried = true;
3481 goto again;
3482 } else if (failed && retried) {
3483 ret = -ENOSPC;
3484 lock_chunks(root);
3485
3486 device->total_bytes = old_size;
3487 if (device->writeable)
3488 device->fs_devices->total_rw_bytes += diff;
3489 spin_lock(&root->fs_info->free_chunk_lock);
3490 root->fs_info->free_chunk_space += diff;
3491 spin_unlock(&root->fs_info->free_chunk_lock);
3492 unlock_chunks(root);
3493 goto done;
3494 }
3495
3496 /* Shrinking succeeded, else we would be at "done". */
3497 trans = btrfs_start_transaction(root, 0);
3498 if (IS_ERR(trans)) {
3499 ret = PTR_ERR(trans);
3500 goto done;
3501 }
3502
3503 lock_chunks(root);
3504
3505 device->disk_total_bytes = new_size;
3506 /* Now btrfs_update_device() will change the on-disk size. */
3507 ret = btrfs_update_device(trans, device);
3508 if (ret) {
3509 unlock_chunks(root);
3510 btrfs_end_transaction(trans, root);
3511 goto done;
3512 }
3513 WARN_ON(diff > old_total);
3514 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3515 unlock_chunks(root);
3516 btrfs_end_transaction(trans, root);
3517 done:
3518 btrfs_free_path(path);
3519 return ret;
3520 }
3521
3522 static int btrfs_add_system_chunk(struct btrfs_root *root,
3523 struct btrfs_key *key,
3524 struct btrfs_chunk *chunk, int item_size)
3525 {
3526 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3527 struct btrfs_disk_key disk_key;
3528 u32 array_size;
3529 u8 *ptr;
3530
3531 array_size = btrfs_super_sys_array_size(super_copy);
3532 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3533 return -EFBIG;
3534
3535 ptr = super_copy->sys_chunk_array + array_size;
3536 btrfs_cpu_key_to_disk(&disk_key, key);
3537 memcpy(ptr, &disk_key, sizeof(disk_key));
3538 ptr += sizeof(disk_key);
3539 memcpy(ptr, chunk, item_size);
3540 item_size += sizeof(disk_key);
3541 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3542 return 0;
3543 }
3544
3545 /*
3546 * sort the devices in descending order by max_avail, total_avail
3547 */
3548 static int btrfs_cmp_device_info(const void *a, const void *b)
3549 {
3550 const struct btrfs_device_info *di_a = a;
3551 const struct btrfs_device_info *di_b = b;
3552
3553 if (di_a->max_avail > di_b->max_avail)
3554 return -1;
3555 if (di_a->max_avail < di_b->max_avail)
3556 return 1;
3557 if (di_a->total_avail > di_b->total_avail)
3558 return -1;
3559 if (di_a->total_avail < di_b->total_avail)
3560 return 1;
3561 return 0;
3562 }
3563
3564 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3565 { 2, 1, 0, 4, 2, 2 /* raid10 */ },
3566 { 1, 1, 2, 2, 2, 2 /* raid1 */ },
3567 { 1, 2, 1, 1, 1, 2 /* dup */ },
3568 { 1, 1, 0, 2, 1, 1 /* raid0 */ },
3569 { 1, 1, 1, 1, 1, 1 /* single */ },
3570 };
3571
3572 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3573 struct btrfs_root *extent_root,
3574 struct map_lookup **map_ret,
3575 u64 *num_bytes_out, u64 *stripe_size_out,
3576 u64 start, u64 type)
3577 {
3578 struct btrfs_fs_info *info = extent_root->fs_info;
3579 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3580 struct list_head *cur;
3581 struct map_lookup *map = NULL;
3582 struct extent_map_tree *em_tree;
3583 struct extent_map *em;
3584 struct btrfs_device_info *devices_info = NULL;
3585 u64 total_avail;
3586 int num_stripes; /* total number of stripes to allocate */
3587 int sub_stripes; /* sub_stripes info for map */
3588 int dev_stripes; /* stripes per dev */
3589 int devs_max; /* max devs to use */
3590 int devs_min; /* min devs needed */
3591 int devs_increment; /* ndevs has to be a multiple of this */
3592 int ncopies; /* how many copies to data has */
3593 int ret;
3594 u64 max_stripe_size;
3595 u64 max_chunk_size;
3596 u64 stripe_size;
3597 u64 num_bytes;
3598 int ndevs;
3599 int i;
3600 int j;
3601 int index;
3602
3603 BUG_ON(!alloc_profile_is_valid(type, 0));
3604
3605 if (list_empty(&fs_devices->alloc_list))
3606 return -ENOSPC;
3607
3608 index = __get_raid_index(type);
3609
3610 sub_stripes = btrfs_raid_array[index].sub_stripes;
3611 dev_stripes = btrfs_raid_array[index].dev_stripes;
3612 devs_max = btrfs_raid_array[index].devs_max;
3613 devs_min = btrfs_raid_array[index].devs_min;
3614 devs_increment = btrfs_raid_array[index].devs_increment;
3615 ncopies = btrfs_raid_array[index].ncopies;
3616
3617 if (type & BTRFS_BLOCK_GROUP_DATA) {
3618 max_stripe_size = 1024 * 1024 * 1024;
3619 max_chunk_size = 10 * max_stripe_size;
3620 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3621 /* for larger filesystems, use larger metadata chunks */
3622 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3623 max_stripe_size = 1024 * 1024 * 1024;
3624 else
3625 max_stripe_size = 256 * 1024 * 1024;
3626 max_chunk_size = max_stripe_size;
3627 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3628 max_stripe_size = 32 * 1024 * 1024;
3629 max_chunk_size = 2 * max_stripe_size;
3630 } else {
3631 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3632 type);
3633 BUG_ON(1);
3634 }
3635
3636 /* we don't want a chunk larger than 10% of writeable space */
3637 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3638 max_chunk_size);
3639
3640 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3641 GFP_NOFS);
3642 if (!devices_info)
3643 return -ENOMEM;
3644
3645 cur = fs_devices->alloc_list.next;
3646
3647 /*
3648 * in the first pass through the devices list, we gather information
3649 * about the available holes on each device.
3650 */
3651 ndevs = 0;
3652 while (cur != &fs_devices->alloc_list) {
3653 struct btrfs_device *device;
3654 u64 max_avail;
3655 u64 dev_offset;
3656
3657 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3658
3659 cur = cur->next;
3660
3661 if (!device->writeable) {
3662 WARN(1, KERN_ERR
3663 "btrfs: read-only device in alloc_list\n");
3664 continue;
3665 }
3666
3667 if (!device->in_fs_metadata ||
3668 device->is_tgtdev_for_dev_replace)
3669 continue;
3670
3671 if (device->total_bytes > device->bytes_used)
3672 total_avail = device->total_bytes - device->bytes_used;
3673 else
3674 total_avail = 0;
3675
3676 /* If there is no space on this device, skip it. */
3677 if (total_avail == 0)
3678 continue;
3679
3680 ret = find_free_dev_extent(device,
3681 max_stripe_size * dev_stripes,
3682 &dev_offset, &max_avail);
3683 if (ret && ret != -ENOSPC)
3684 goto error;
3685
3686 if (ret == 0)
3687 max_avail = max_stripe_size * dev_stripes;
3688
3689 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3690 continue;
3691
3692 devices_info[ndevs].dev_offset = dev_offset;
3693 devices_info[ndevs].max_avail = max_avail;
3694 devices_info[ndevs].total_avail = total_avail;
3695 devices_info[ndevs].dev = device;
3696 ++ndevs;
3697 WARN_ON(ndevs > fs_devices->rw_devices);
3698 }
3699
3700 /*
3701 * now sort the devices by hole size / available space
3702 */
3703 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3704 btrfs_cmp_device_info, NULL);
3705
3706 /* round down to number of usable stripes */
3707 ndevs -= ndevs % devs_increment;
3708
3709 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3710 ret = -ENOSPC;
3711 goto error;
3712 }
3713
3714 if (devs_max && ndevs > devs_max)
3715 ndevs = devs_max;
3716 /*
3717 * the primary goal is to maximize the number of stripes, so use as many
3718 * devices as possible, even if the stripes are not maximum sized.
3719 */
3720 stripe_size = devices_info[ndevs-1].max_avail;
3721 num_stripes = ndevs * dev_stripes;
3722
3723 if (stripe_size * ndevs > max_chunk_size * ncopies) {
3724 stripe_size = max_chunk_size * ncopies;
3725 do_div(stripe_size, ndevs);
3726 }
3727
3728 do_div(stripe_size, dev_stripes);
3729
3730 /* align to BTRFS_STRIPE_LEN */
3731 do_div(stripe_size, BTRFS_STRIPE_LEN);
3732 stripe_size *= BTRFS_STRIPE_LEN;
3733
3734 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3735 if (!map) {
3736 ret = -ENOMEM;
3737 goto error;
3738 }
3739 map->num_stripes = num_stripes;
3740
3741 for (i = 0; i < ndevs; ++i) {
3742 for (j = 0; j < dev_stripes; ++j) {
3743 int s = i * dev_stripes + j;
3744 map->stripes[s].dev = devices_info[i].dev;
3745 map->stripes[s].physical = devices_info[i].dev_offset +
3746 j * stripe_size;
3747 }
3748 }
3749 map->sector_size = extent_root->sectorsize;
3750 map->stripe_len = BTRFS_STRIPE_LEN;
3751 map->io_align = BTRFS_STRIPE_LEN;
3752 map->io_width = BTRFS_STRIPE_LEN;
3753 map->type = type;
3754 map->sub_stripes = sub_stripes;
3755
3756 *map_ret = map;
3757 num_bytes = stripe_size * (num_stripes / ncopies);
3758
3759 *stripe_size_out = stripe_size;
3760 *num_bytes_out = num_bytes;
3761
3762 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3763
3764 em = alloc_extent_map();
3765 if (!em) {
3766 ret = -ENOMEM;
3767 goto error;
3768 }
3769 em->bdev = (struct block_device *)map;
3770 em->start = start;
3771 em->len = num_bytes;
3772 em->block_start = 0;
3773 em->block_len = em->len;
3774
3775 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3776 write_lock(&em_tree->lock);
3777 ret = add_extent_mapping(em_tree, em);
3778 write_unlock(&em_tree->lock);
3779 free_extent_map(em);
3780 if (ret)
3781 goto error;
3782
3783 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3784 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3785 start, num_bytes);
3786 if (ret)
3787 goto error;
3788
3789 for (i = 0; i < map->num_stripes; ++i) {
3790 struct btrfs_device *device;
3791 u64 dev_offset;
3792
3793 device = map->stripes[i].dev;
3794 dev_offset = map->stripes[i].physical;
3795
3796 ret = btrfs_alloc_dev_extent(trans, device,
3797 info->chunk_root->root_key.objectid,
3798 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3799 start, dev_offset, stripe_size);
3800 if (ret) {
3801 btrfs_abort_transaction(trans, extent_root, ret);
3802 goto error;
3803 }
3804 }
3805
3806 kfree(devices_info);
3807 return 0;
3808
3809 error:
3810 kfree(map);
3811 kfree(devices_info);
3812 return ret;
3813 }
3814
3815 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *extent_root,
3817 struct map_lookup *map, u64 chunk_offset,
3818 u64 chunk_size, u64 stripe_size)
3819 {
3820 u64 dev_offset;
3821 struct btrfs_key key;
3822 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3823 struct btrfs_device *device;
3824 struct btrfs_chunk *chunk;
3825 struct btrfs_stripe *stripe;
3826 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3827 int index = 0;
3828 int ret;
3829
3830 chunk = kzalloc(item_size, GFP_NOFS);
3831 if (!chunk)
3832 return -ENOMEM;
3833
3834 index = 0;
3835 while (index < map->num_stripes) {
3836 device = map->stripes[index].dev;
3837 device->bytes_used += stripe_size;
3838 ret = btrfs_update_device(trans, device);
3839 if (ret)
3840 goto out_free;
3841 index++;
3842 }
3843
3844 spin_lock(&extent_root->fs_info->free_chunk_lock);
3845 extent_root->fs_info->free_chunk_space -= (stripe_size *
3846 map->num_stripes);
3847 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3848
3849 index = 0;
3850 stripe = &chunk->stripe;
3851 while (index < map->num_stripes) {
3852 device = map->stripes[index].dev;
3853 dev_offset = map->stripes[index].physical;
3854
3855 btrfs_set_stack_stripe_devid(stripe, device->devid);
3856 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3857 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3858 stripe++;
3859 index++;
3860 }
3861
3862 btrfs_set_stack_chunk_length(chunk, chunk_size);
3863 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3864 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3865 btrfs_set_stack_chunk_type(chunk, map->type);
3866 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3867 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3868 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3869 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3870 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3871
3872 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3873 key.type = BTRFS_CHUNK_ITEM_KEY;
3874 key.offset = chunk_offset;
3875
3876 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3877
3878 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3879 /*
3880 * TODO: Cleanup of inserted chunk root in case of
3881 * failure.
3882 */
3883 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3884 item_size);
3885 }
3886
3887 out_free:
3888 kfree(chunk);
3889 return ret;
3890 }
3891
3892 /*
3893 * Chunk allocation falls into two parts. The first part does works
3894 * that make the new allocated chunk useable, but not do any operation
3895 * that modifies the chunk tree. The second part does the works that
3896 * require modifying the chunk tree. This division is important for the
3897 * bootstrap process of adding storage to a seed btrfs.
3898 */
3899 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3900 struct btrfs_root *extent_root, u64 type)
3901 {
3902 u64 chunk_offset;
3903 u64 chunk_size;
3904 u64 stripe_size;
3905 struct map_lookup *map;
3906 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3907 int ret;
3908
3909 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3910 &chunk_offset);
3911 if (ret)
3912 return ret;
3913
3914 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3915 &stripe_size, chunk_offset, type);
3916 if (ret)
3917 return ret;
3918
3919 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3920 chunk_size, stripe_size);
3921 if (ret)
3922 return ret;
3923 return 0;
3924 }
3925
3926 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3927 struct btrfs_root *root,
3928 struct btrfs_device *device)
3929 {
3930 u64 chunk_offset;
3931 u64 sys_chunk_offset;
3932 u64 chunk_size;
3933 u64 sys_chunk_size;
3934 u64 stripe_size;
3935 u64 sys_stripe_size;
3936 u64 alloc_profile;
3937 struct map_lookup *map;
3938 struct map_lookup *sys_map;
3939 struct btrfs_fs_info *fs_info = root->fs_info;
3940 struct btrfs_root *extent_root = fs_info->extent_root;
3941 int ret;
3942
3943 ret = find_next_chunk(fs_info->chunk_root,
3944 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3945 if (ret)
3946 return ret;
3947
3948 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3949 fs_info->avail_metadata_alloc_bits;
3950 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3951
3952 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3953 &stripe_size, chunk_offset, alloc_profile);
3954 if (ret)
3955 return ret;
3956
3957 sys_chunk_offset = chunk_offset + chunk_size;
3958
3959 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3960 fs_info->avail_system_alloc_bits;
3961 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3962
3963 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3964 &sys_chunk_size, &sys_stripe_size,
3965 sys_chunk_offset, alloc_profile);
3966 if (ret) {
3967 btrfs_abort_transaction(trans, root, ret);
3968 goto out;
3969 }
3970
3971 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3972 if (ret) {
3973 btrfs_abort_transaction(trans, root, ret);
3974 goto out;
3975 }
3976
3977 /*
3978 * Modifying chunk tree needs allocating new blocks from both
3979 * system block group and metadata block group. So we only can
3980 * do operations require modifying the chunk tree after both
3981 * block groups were created.
3982 */
3983 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3984 chunk_size, stripe_size);
3985 if (ret) {
3986 btrfs_abort_transaction(trans, root, ret);
3987 goto out;
3988 }
3989
3990 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3991 sys_chunk_offset, sys_chunk_size,
3992 sys_stripe_size);
3993 if (ret)
3994 btrfs_abort_transaction(trans, root, ret);
3995
3996 out:
3997
3998 return ret;
3999 }
4000
4001 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4002 {
4003 struct extent_map *em;
4004 struct map_lookup *map;
4005 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4006 int readonly = 0;
4007 int i;
4008
4009 read_lock(&map_tree->map_tree.lock);
4010 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4011 read_unlock(&map_tree->map_tree.lock);
4012 if (!em)
4013 return 1;
4014
4015 if (btrfs_test_opt(root, DEGRADED)) {
4016 free_extent_map(em);
4017 return 0;
4018 }
4019
4020 map = (struct map_lookup *)em->bdev;
4021 for (i = 0; i < map->num_stripes; i++) {
4022 if (!map->stripes[i].dev->writeable) {
4023 readonly = 1;
4024 break;
4025 }
4026 }
4027 free_extent_map(em);
4028 return readonly;
4029 }
4030
4031 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4032 {
4033 extent_map_tree_init(&tree->map_tree);
4034 }
4035
4036 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4037 {
4038 struct extent_map *em;
4039
4040 while (1) {
4041 write_lock(&tree->map_tree.lock);
4042 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4043 if (em)
4044 remove_extent_mapping(&tree->map_tree, em);
4045 write_unlock(&tree->map_tree.lock);
4046 if (!em)
4047 break;
4048 kfree(em->bdev);
4049 /* once for us */
4050 free_extent_map(em);
4051 /* once for the tree */
4052 free_extent_map(em);
4053 }
4054 }
4055
4056 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4057 {
4058 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4059 struct extent_map *em;
4060 struct map_lookup *map;
4061 struct extent_map_tree *em_tree = &map_tree->map_tree;
4062 int ret;
4063
4064 read_lock(&em_tree->lock);
4065 em = lookup_extent_mapping(em_tree, logical, len);
4066 read_unlock(&em_tree->lock);
4067 BUG_ON(!em);
4068
4069 BUG_ON(em->start > logical || em->start + em->len < logical);
4070 map = (struct map_lookup *)em->bdev;
4071 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4072 ret = map->num_stripes;
4073 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4074 ret = map->sub_stripes;
4075 else
4076 ret = 1;
4077 free_extent_map(em);
4078
4079 btrfs_dev_replace_lock(&fs_info->dev_replace);
4080 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4081 ret++;
4082 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4083
4084 return ret;
4085 }
4086
4087 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4088 struct map_lookup *map, int first, int num,
4089 int optimal, int dev_replace_is_ongoing)
4090 {
4091 int i;
4092 int tolerance;
4093 struct btrfs_device *srcdev;
4094
4095 if (dev_replace_is_ongoing &&
4096 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4097 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4098 srcdev = fs_info->dev_replace.srcdev;
4099 else
4100 srcdev = NULL;
4101
4102 /*
4103 * try to avoid the drive that is the source drive for a
4104 * dev-replace procedure, only choose it if no other non-missing
4105 * mirror is available
4106 */
4107 for (tolerance = 0; tolerance < 2; tolerance++) {
4108 if (map->stripes[optimal].dev->bdev &&
4109 (tolerance || map->stripes[optimal].dev != srcdev))
4110 return optimal;
4111 for (i = first; i < first + num; i++) {
4112 if (map->stripes[i].dev->bdev &&
4113 (tolerance || map->stripes[i].dev != srcdev))
4114 return i;
4115 }
4116 }
4117
4118 /* we couldn't find one that doesn't fail. Just return something
4119 * and the io error handling code will clean up eventually
4120 */
4121 return optimal;
4122 }
4123
4124 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4125 u64 logical, u64 *length,
4126 struct btrfs_bio **bbio_ret,
4127 int mirror_num)
4128 {
4129 struct extent_map *em;
4130 struct map_lookup *map;
4131 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4132 struct extent_map_tree *em_tree = &map_tree->map_tree;
4133 u64 offset;
4134 u64 stripe_offset;
4135 u64 stripe_end_offset;
4136 u64 stripe_nr;
4137 u64 stripe_nr_orig;
4138 u64 stripe_nr_end;
4139 int stripe_index;
4140 int i;
4141 int ret = 0;
4142 int num_stripes;
4143 int max_errors = 0;
4144 struct btrfs_bio *bbio = NULL;
4145 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4146 int dev_replace_is_ongoing = 0;
4147 int num_alloc_stripes;
4148 int patch_the_first_stripe_for_dev_replace = 0;
4149 u64 physical_to_patch_in_first_stripe = 0;
4150
4151 read_lock(&em_tree->lock);
4152 em = lookup_extent_mapping(em_tree, logical, *length);
4153 read_unlock(&em_tree->lock);
4154
4155 if (!em) {
4156 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4157 (unsigned long long)logical,
4158 (unsigned long long)*length);
4159 BUG();
4160 }
4161
4162 BUG_ON(em->start > logical || em->start + em->len < logical);
4163 map = (struct map_lookup *)em->bdev;
4164 offset = logical - em->start;
4165
4166 stripe_nr = offset;
4167 /*
4168 * stripe_nr counts the total number of stripes we have to stride
4169 * to get to this block
4170 */
4171 do_div(stripe_nr, map->stripe_len);
4172
4173 stripe_offset = stripe_nr * map->stripe_len;
4174 BUG_ON(offset < stripe_offset);
4175
4176 /* stripe_offset is the offset of this block in its stripe*/
4177 stripe_offset = offset - stripe_offset;
4178
4179 if (rw & REQ_DISCARD)
4180 *length = min_t(u64, em->len - offset, *length);
4181 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4182 /* we limit the length of each bio to what fits in a stripe */
4183 *length = min_t(u64, em->len - offset,
4184 map->stripe_len - stripe_offset);
4185 } else {
4186 *length = em->len - offset;
4187 }
4188
4189 if (!bbio_ret)
4190 goto out;
4191
4192 btrfs_dev_replace_lock(dev_replace);
4193 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4194 if (!dev_replace_is_ongoing)
4195 btrfs_dev_replace_unlock(dev_replace);
4196
4197 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4198 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4199 dev_replace->tgtdev != NULL) {
4200 /*
4201 * in dev-replace case, for repair case (that's the only
4202 * case where the mirror is selected explicitly when
4203 * calling btrfs_map_block), blocks left of the left cursor
4204 * can also be read from the target drive.
4205 * For REQ_GET_READ_MIRRORS, the target drive is added as
4206 * the last one to the array of stripes. For READ, it also
4207 * needs to be supported using the same mirror number.
4208 * If the requested block is not left of the left cursor,
4209 * EIO is returned. This can happen because btrfs_num_copies()
4210 * returns one more in the dev-replace case.
4211 */
4212 u64 tmp_length = *length;
4213 struct btrfs_bio *tmp_bbio = NULL;
4214 int tmp_num_stripes;
4215 u64 srcdev_devid = dev_replace->srcdev->devid;
4216 int index_srcdev = 0;
4217 int found = 0;
4218 u64 physical_of_found = 0;
4219
4220 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4221 logical, &tmp_length, &tmp_bbio, 0);
4222 if (ret) {
4223 WARN_ON(tmp_bbio != NULL);
4224 goto out;
4225 }
4226
4227 tmp_num_stripes = tmp_bbio->num_stripes;
4228 if (mirror_num > tmp_num_stripes) {
4229 /*
4230 * REQ_GET_READ_MIRRORS does not contain this
4231 * mirror, that means that the requested area
4232 * is not left of the left cursor
4233 */
4234 ret = -EIO;
4235 kfree(tmp_bbio);
4236 goto out;
4237 }
4238
4239 /*
4240 * process the rest of the function using the mirror_num
4241 * of the source drive. Therefore look it up first.
4242 * At the end, patch the device pointer to the one of the
4243 * target drive.
4244 */
4245 for (i = 0; i < tmp_num_stripes; i++) {
4246 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4247 /*
4248 * In case of DUP, in order to keep it
4249 * simple, only add the mirror with the
4250 * lowest physical address
4251 */
4252 if (found &&
4253 physical_of_found <=
4254 tmp_bbio->stripes[i].physical)
4255 continue;
4256 index_srcdev = i;
4257 found = 1;
4258 physical_of_found =
4259 tmp_bbio->stripes[i].physical;
4260 }
4261 }
4262
4263 if (found) {
4264 mirror_num = index_srcdev + 1;
4265 patch_the_first_stripe_for_dev_replace = 1;
4266 physical_to_patch_in_first_stripe = physical_of_found;
4267 } else {
4268 WARN_ON(1);
4269 ret = -EIO;
4270 kfree(tmp_bbio);
4271 goto out;
4272 }
4273
4274 kfree(tmp_bbio);
4275 } else if (mirror_num > map->num_stripes) {
4276 mirror_num = 0;
4277 }
4278
4279 num_stripes = 1;
4280 stripe_index = 0;
4281 stripe_nr_orig = stripe_nr;
4282 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4283 (~(map->stripe_len - 1));
4284 do_div(stripe_nr_end, map->stripe_len);
4285 stripe_end_offset = stripe_nr_end * map->stripe_len -
4286 (offset + *length);
4287 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4288 if (rw & REQ_DISCARD)
4289 num_stripes = min_t(u64, map->num_stripes,
4290 stripe_nr_end - stripe_nr_orig);
4291 stripe_index = do_div(stripe_nr, map->num_stripes);
4292 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4293 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4294 num_stripes = map->num_stripes;
4295 else if (mirror_num)
4296 stripe_index = mirror_num - 1;
4297 else {
4298 stripe_index = find_live_mirror(fs_info, map, 0,
4299 map->num_stripes,
4300 current->pid % map->num_stripes,
4301 dev_replace_is_ongoing);
4302 mirror_num = stripe_index + 1;
4303 }
4304
4305 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4306 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4307 num_stripes = map->num_stripes;
4308 } else if (mirror_num) {
4309 stripe_index = mirror_num - 1;
4310 } else {
4311 mirror_num = 1;
4312 }
4313
4314 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4315 int factor = map->num_stripes / map->sub_stripes;
4316
4317 stripe_index = do_div(stripe_nr, factor);
4318 stripe_index *= map->sub_stripes;
4319
4320 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4321 num_stripes = map->sub_stripes;
4322 else if (rw & REQ_DISCARD)
4323 num_stripes = min_t(u64, map->sub_stripes *
4324 (stripe_nr_end - stripe_nr_orig),
4325 map->num_stripes);
4326 else if (mirror_num)
4327 stripe_index += mirror_num - 1;
4328 else {
4329 int old_stripe_index = stripe_index;
4330 stripe_index = find_live_mirror(fs_info, map,
4331 stripe_index,
4332 map->sub_stripes, stripe_index +
4333 current->pid % map->sub_stripes,
4334 dev_replace_is_ongoing);
4335 mirror_num = stripe_index - old_stripe_index + 1;
4336 }
4337 } else {
4338 /*
4339 * after this do_div call, stripe_nr is the number of stripes
4340 * on this device we have to walk to find the data, and
4341 * stripe_index is the number of our device in the stripe array
4342 */
4343 stripe_index = do_div(stripe_nr, map->num_stripes);
4344 mirror_num = stripe_index + 1;
4345 }
4346 BUG_ON(stripe_index >= map->num_stripes);
4347
4348 num_alloc_stripes = num_stripes;
4349 if (dev_replace_is_ongoing) {
4350 if (rw & (REQ_WRITE | REQ_DISCARD))
4351 num_alloc_stripes <<= 1;
4352 if (rw & REQ_GET_READ_MIRRORS)
4353 num_alloc_stripes++;
4354 }
4355 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4356 if (!bbio) {
4357 ret = -ENOMEM;
4358 goto out;
4359 }
4360 atomic_set(&bbio->error, 0);
4361
4362 if (rw & REQ_DISCARD) {
4363 int factor = 0;
4364 int sub_stripes = 0;
4365 u64 stripes_per_dev = 0;
4366 u32 remaining_stripes = 0;
4367 u32 last_stripe = 0;
4368
4369 if (map->type &
4370 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4371 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4372 sub_stripes = 1;
4373 else
4374 sub_stripes = map->sub_stripes;
4375
4376 factor = map->num_stripes / sub_stripes;
4377 stripes_per_dev = div_u64_rem(stripe_nr_end -
4378 stripe_nr_orig,
4379 factor,
4380 &remaining_stripes);
4381 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4382 last_stripe *= sub_stripes;
4383 }
4384
4385 for (i = 0; i < num_stripes; i++) {
4386 bbio->stripes[i].physical =
4387 map->stripes[stripe_index].physical +
4388 stripe_offset + stripe_nr * map->stripe_len;
4389 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4390
4391 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4392 BTRFS_BLOCK_GROUP_RAID10)) {
4393 bbio->stripes[i].length = stripes_per_dev *
4394 map->stripe_len;
4395
4396 if (i / sub_stripes < remaining_stripes)
4397 bbio->stripes[i].length +=
4398 map->stripe_len;
4399
4400 /*
4401 * Special for the first stripe and
4402 * the last stripe:
4403 *
4404 * |-------|...|-------|
4405 * |----------|
4406 * off end_off
4407 */
4408 if (i < sub_stripes)
4409 bbio->stripes[i].length -=
4410 stripe_offset;
4411
4412 if (stripe_index >= last_stripe &&
4413 stripe_index <= (last_stripe +
4414 sub_stripes - 1))
4415 bbio->stripes[i].length -=
4416 stripe_end_offset;
4417
4418 if (i == sub_stripes - 1)
4419 stripe_offset = 0;
4420 } else
4421 bbio->stripes[i].length = *length;
4422
4423 stripe_index++;
4424 if (stripe_index == map->num_stripes) {
4425 /* This could only happen for RAID0/10 */
4426 stripe_index = 0;
4427 stripe_nr++;
4428 }
4429 }
4430 } else {
4431 for (i = 0; i < num_stripes; i++) {
4432 bbio->stripes[i].physical =
4433 map->stripes[stripe_index].physical +
4434 stripe_offset +
4435 stripe_nr * map->stripe_len;
4436 bbio->stripes[i].dev =
4437 map->stripes[stripe_index].dev;
4438 stripe_index++;
4439 }
4440 }
4441
4442 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4443 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4444 BTRFS_BLOCK_GROUP_RAID10 |
4445 BTRFS_BLOCK_GROUP_DUP)) {
4446 max_errors = 1;
4447 }
4448 }
4449
4450 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4451 dev_replace->tgtdev != NULL) {
4452 int index_where_to_add;
4453 u64 srcdev_devid = dev_replace->srcdev->devid;
4454
4455 /*
4456 * duplicate the write operations while the dev replace
4457 * procedure is running. Since the copying of the old disk
4458 * to the new disk takes place at run time while the
4459 * filesystem is mounted writable, the regular write
4460 * operations to the old disk have to be duplicated to go
4461 * to the new disk as well.
4462 * Note that device->missing is handled by the caller, and
4463 * that the write to the old disk is already set up in the
4464 * stripes array.
4465 */
4466 index_where_to_add = num_stripes;
4467 for (i = 0; i < num_stripes; i++) {
4468 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4469 /* write to new disk, too */
4470 struct btrfs_bio_stripe *new =
4471 bbio->stripes + index_where_to_add;
4472 struct btrfs_bio_stripe *old =
4473 bbio->stripes + i;
4474
4475 new->physical = old->physical;
4476 new->length = old->length;
4477 new->dev = dev_replace->tgtdev;
4478 index_where_to_add++;
4479 max_errors++;
4480 }
4481 }
4482 num_stripes = index_where_to_add;
4483 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4484 dev_replace->tgtdev != NULL) {
4485 u64 srcdev_devid = dev_replace->srcdev->devid;
4486 int index_srcdev = 0;
4487 int found = 0;
4488 u64 physical_of_found = 0;
4489
4490 /*
4491 * During the dev-replace procedure, the target drive can
4492 * also be used to read data in case it is needed to repair
4493 * a corrupt block elsewhere. This is possible if the
4494 * requested area is left of the left cursor. In this area,
4495 * the target drive is a full copy of the source drive.
4496 */
4497 for (i = 0; i < num_stripes; i++) {
4498 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4499 /*
4500 * In case of DUP, in order to keep it
4501 * simple, only add the mirror with the
4502 * lowest physical address
4503 */
4504 if (found &&
4505 physical_of_found <=
4506 bbio->stripes[i].physical)
4507 continue;
4508 index_srcdev = i;
4509 found = 1;
4510 physical_of_found = bbio->stripes[i].physical;
4511 }
4512 }
4513 if (found) {
4514 u64 length = map->stripe_len;
4515
4516 if (physical_of_found + length <=
4517 dev_replace->cursor_left) {
4518 struct btrfs_bio_stripe *tgtdev_stripe =
4519 bbio->stripes + num_stripes;
4520
4521 tgtdev_stripe->physical = physical_of_found;
4522 tgtdev_stripe->length =
4523 bbio->stripes[index_srcdev].length;
4524 tgtdev_stripe->dev = dev_replace->tgtdev;
4525
4526 num_stripes++;
4527 }
4528 }
4529 }
4530
4531 *bbio_ret = bbio;
4532 bbio->num_stripes = num_stripes;
4533 bbio->max_errors = max_errors;
4534 bbio->mirror_num = mirror_num;
4535
4536 /*
4537 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4538 * mirror_num == num_stripes + 1 && dev_replace target drive is
4539 * available as a mirror
4540 */
4541 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4542 WARN_ON(num_stripes > 1);
4543 bbio->stripes[0].dev = dev_replace->tgtdev;
4544 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4545 bbio->mirror_num = map->num_stripes + 1;
4546 }
4547 out:
4548 if (dev_replace_is_ongoing)
4549 btrfs_dev_replace_unlock(dev_replace);
4550 free_extent_map(em);
4551 return ret;
4552 }
4553
4554 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4555 u64 logical, u64 *length,
4556 struct btrfs_bio **bbio_ret, int mirror_num)
4557 {
4558 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4559 mirror_num);
4560 }
4561
4562 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4563 u64 chunk_start, u64 physical, u64 devid,
4564 u64 **logical, int *naddrs, int *stripe_len)
4565 {
4566 struct extent_map_tree *em_tree = &map_tree->map_tree;
4567 struct extent_map *em;
4568 struct map_lookup *map;
4569 u64 *buf;
4570 u64 bytenr;
4571 u64 length;
4572 u64 stripe_nr;
4573 int i, j, nr = 0;
4574
4575 read_lock(&em_tree->lock);
4576 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4577 read_unlock(&em_tree->lock);
4578
4579 BUG_ON(!em || em->start != chunk_start);
4580 map = (struct map_lookup *)em->bdev;
4581
4582 length = em->len;
4583 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4584 do_div(length, map->num_stripes / map->sub_stripes);
4585 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4586 do_div(length, map->num_stripes);
4587
4588 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4589 BUG_ON(!buf); /* -ENOMEM */
4590
4591 for (i = 0; i < map->num_stripes; i++) {
4592 if (devid && map->stripes[i].dev->devid != devid)
4593 continue;
4594 if (map->stripes[i].physical > physical ||
4595 map->stripes[i].physical + length <= physical)
4596 continue;
4597
4598 stripe_nr = physical - map->stripes[i].physical;
4599 do_div(stripe_nr, map->stripe_len);
4600
4601 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4602 stripe_nr = stripe_nr * map->num_stripes + i;
4603 do_div(stripe_nr, map->sub_stripes);
4604 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4605 stripe_nr = stripe_nr * map->num_stripes + i;
4606 }
4607 bytenr = chunk_start + stripe_nr * map->stripe_len;
4608 WARN_ON(nr >= map->num_stripes);
4609 for (j = 0; j < nr; j++) {
4610 if (buf[j] == bytenr)
4611 break;
4612 }
4613 if (j == nr) {
4614 WARN_ON(nr >= map->num_stripes);
4615 buf[nr++] = bytenr;
4616 }
4617 }
4618
4619 *logical = buf;
4620 *naddrs = nr;
4621 *stripe_len = map->stripe_len;
4622
4623 free_extent_map(em);
4624 return 0;
4625 }
4626
4627 static void *merge_stripe_index_into_bio_private(void *bi_private,
4628 unsigned int stripe_index)
4629 {
4630 /*
4631 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4632 * at most 1.
4633 * The alternative solution (instead of stealing bits from the
4634 * pointer) would be to allocate an intermediate structure
4635 * that contains the old private pointer plus the stripe_index.
4636 */
4637 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4638 BUG_ON(stripe_index > 3);
4639 return (void *)(((uintptr_t)bi_private) | stripe_index);
4640 }
4641
4642 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4643 {
4644 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4645 }
4646
4647 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4648 {
4649 return (unsigned int)((uintptr_t)bi_private) & 3;
4650 }
4651
4652 static void btrfs_end_bio(struct bio *bio, int err)
4653 {
4654 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4655 int is_orig_bio = 0;
4656
4657 if (err) {
4658 atomic_inc(&bbio->error);
4659 if (err == -EIO || err == -EREMOTEIO) {
4660 unsigned int stripe_index =
4661 extract_stripe_index_from_bio_private(
4662 bio->bi_private);
4663 struct btrfs_device *dev;
4664
4665 BUG_ON(stripe_index >= bbio->num_stripes);
4666 dev = bbio->stripes[stripe_index].dev;
4667 if (dev->bdev) {
4668 if (bio->bi_rw & WRITE)
4669 btrfs_dev_stat_inc(dev,
4670 BTRFS_DEV_STAT_WRITE_ERRS);
4671 else
4672 btrfs_dev_stat_inc(dev,
4673 BTRFS_DEV_STAT_READ_ERRS);
4674 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4675 btrfs_dev_stat_inc(dev,
4676 BTRFS_DEV_STAT_FLUSH_ERRS);
4677 btrfs_dev_stat_print_on_error(dev);
4678 }
4679 }
4680 }
4681
4682 if (bio == bbio->orig_bio)
4683 is_orig_bio = 1;
4684
4685 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4686 if (!is_orig_bio) {
4687 bio_put(bio);
4688 bio = bbio->orig_bio;
4689 }
4690 bio->bi_private = bbio->private;
4691 bio->bi_end_io = bbio->end_io;
4692 bio->bi_bdev = (struct block_device *)
4693 (unsigned long)bbio->mirror_num;
4694 /* only send an error to the higher layers if it is
4695 * beyond the tolerance of the multi-bio
4696 */
4697 if (atomic_read(&bbio->error) > bbio->max_errors) {
4698 err = -EIO;
4699 } else {
4700 /*
4701 * this bio is actually up to date, we didn't
4702 * go over the max number of errors
4703 */
4704 set_bit(BIO_UPTODATE, &bio->bi_flags);
4705 err = 0;
4706 }
4707 kfree(bbio);
4708
4709 bio_endio(bio, err);
4710 } else if (!is_orig_bio) {
4711 bio_put(bio);
4712 }
4713 }
4714
4715 struct async_sched {
4716 struct bio *bio;
4717 int rw;
4718 struct btrfs_fs_info *info;
4719 struct btrfs_work work;
4720 };
4721
4722 /*
4723 * see run_scheduled_bios for a description of why bios are collected for
4724 * async submit.
4725 *
4726 * This will add one bio to the pending list for a device and make sure
4727 * the work struct is scheduled.
4728 */
4729 static noinline void schedule_bio(struct btrfs_root *root,
4730 struct btrfs_device *device,
4731 int rw, struct bio *bio)
4732 {
4733 int should_queue = 1;
4734 struct btrfs_pending_bios *pending_bios;
4735
4736 /* don't bother with additional async steps for reads, right now */
4737 if (!(rw & REQ_WRITE)) {
4738 bio_get(bio);
4739 btrfsic_submit_bio(rw, bio);
4740 bio_put(bio);
4741 return;
4742 }
4743
4744 /*
4745 * nr_async_bios allows us to reliably return congestion to the
4746 * higher layers. Otherwise, the async bio makes it appear we have
4747 * made progress against dirty pages when we've really just put it
4748 * on a queue for later
4749 */
4750 atomic_inc(&root->fs_info->nr_async_bios);
4751 WARN_ON(bio->bi_next);
4752 bio->bi_next = NULL;
4753 bio->bi_rw |= rw;
4754
4755 spin_lock(&device->io_lock);
4756 if (bio->bi_rw & REQ_SYNC)
4757 pending_bios = &device->pending_sync_bios;
4758 else
4759 pending_bios = &device->pending_bios;
4760
4761 if (pending_bios->tail)
4762 pending_bios->tail->bi_next = bio;
4763
4764 pending_bios->tail = bio;
4765 if (!pending_bios->head)
4766 pending_bios->head = bio;
4767 if (device->running_pending)
4768 should_queue = 0;
4769
4770 spin_unlock(&device->io_lock);
4771
4772 if (should_queue)
4773 btrfs_queue_worker(&root->fs_info->submit_workers,
4774 &device->work);
4775 }
4776
4777 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4778 sector_t sector)
4779 {
4780 struct bio_vec *prev;
4781 struct request_queue *q = bdev_get_queue(bdev);
4782 unsigned short max_sectors = queue_max_sectors(q);
4783 struct bvec_merge_data bvm = {
4784 .bi_bdev = bdev,
4785 .bi_sector = sector,
4786 .bi_rw = bio->bi_rw,
4787 };
4788
4789 if (bio->bi_vcnt == 0) {
4790 WARN_ON(1);
4791 return 1;
4792 }
4793
4794 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4795 if ((bio->bi_size >> 9) > max_sectors)
4796 return 0;
4797
4798 if (!q->merge_bvec_fn)
4799 return 1;
4800
4801 bvm.bi_size = bio->bi_size - prev->bv_len;
4802 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4803 return 0;
4804 return 1;
4805 }
4806
4807 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4808 struct bio *bio, u64 physical, int dev_nr,
4809 int rw, int async)
4810 {
4811 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4812
4813 bio->bi_private = bbio;
4814 bio->bi_private = merge_stripe_index_into_bio_private(
4815 bio->bi_private, (unsigned int)dev_nr);
4816 bio->bi_end_io = btrfs_end_bio;
4817 bio->bi_sector = physical >> 9;
4818 #ifdef DEBUG
4819 {
4820 struct rcu_string *name;
4821
4822 rcu_read_lock();
4823 name = rcu_dereference(dev->name);
4824 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4825 "(%s id %llu), size=%u\n", rw,
4826 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4827 name->str, dev->devid, bio->bi_size);
4828 rcu_read_unlock();
4829 }
4830 #endif
4831 bio->bi_bdev = dev->bdev;
4832 if (async)
4833 schedule_bio(root, dev, rw, bio);
4834 else
4835 btrfsic_submit_bio(rw, bio);
4836 }
4837
4838 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4839 struct bio *first_bio, struct btrfs_device *dev,
4840 int dev_nr, int rw, int async)
4841 {
4842 struct bio_vec *bvec = first_bio->bi_io_vec;
4843 struct bio *bio;
4844 int nr_vecs = bio_get_nr_vecs(dev->bdev);
4845 u64 physical = bbio->stripes[dev_nr].physical;
4846
4847 again:
4848 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4849 if (!bio)
4850 return -ENOMEM;
4851
4852 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4853 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4854 bvec->bv_offset) < bvec->bv_len) {
4855 u64 len = bio->bi_size;
4856
4857 atomic_inc(&bbio->stripes_pending);
4858 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4859 rw, async);
4860 physical += len;
4861 goto again;
4862 }
4863 bvec++;
4864 }
4865
4866 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4867 return 0;
4868 }
4869
4870 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4871 {
4872 atomic_inc(&bbio->error);
4873 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4874 bio->bi_private = bbio->private;
4875 bio->bi_end_io = bbio->end_io;
4876 bio->bi_bdev = (struct block_device *)
4877 (unsigned long)bbio->mirror_num;
4878 bio->bi_sector = logical >> 9;
4879 kfree(bbio);
4880 bio_endio(bio, -EIO);
4881 }
4882 }
4883
4884 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4885 int mirror_num, int async_submit)
4886 {
4887 struct btrfs_device *dev;
4888 struct bio *first_bio = bio;
4889 u64 logical = (u64)bio->bi_sector << 9;
4890 u64 length = 0;
4891 u64 map_length;
4892 int ret;
4893 int dev_nr = 0;
4894 int total_devs = 1;
4895 struct btrfs_bio *bbio = NULL;
4896
4897 length = bio->bi_size;
4898 map_length = length;
4899
4900 ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4901 mirror_num);
4902 if (ret)
4903 return ret;
4904
4905 total_devs = bbio->num_stripes;
4906 if (map_length < length) {
4907 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4908 "len %llu\n", (unsigned long long)logical,
4909 (unsigned long long)length,
4910 (unsigned long long)map_length);
4911 BUG();
4912 }
4913
4914 bbio->orig_bio = first_bio;
4915 bbio->private = first_bio->bi_private;
4916 bbio->end_io = first_bio->bi_end_io;
4917 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4918
4919 while (dev_nr < total_devs) {
4920 dev = bbio->stripes[dev_nr].dev;
4921 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4922 bbio_error(bbio, first_bio, logical);
4923 dev_nr++;
4924 continue;
4925 }
4926
4927 /*
4928 * Check and see if we're ok with this bio based on it's size
4929 * and offset with the given device.
4930 */
4931 if (!bio_size_ok(dev->bdev, first_bio,
4932 bbio->stripes[dev_nr].physical >> 9)) {
4933 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4934 dev_nr, rw, async_submit);
4935 BUG_ON(ret);
4936 dev_nr++;
4937 continue;
4938 }
4939
4940 if (dev_nr < total_devs - 1) {
4941 bio = bio_clone(first_bio, GFP_NOFS);
4942 BUG_ON(!bio); /* -ENOMEM */
4943 } else {
4944 bio = first_bio;
4945 }
4946
4947 submit_stripe_bio(root, bbio, bio,
4948 bbio->stripes[dev_nr].physical, dev_nr, rw,
4949 async_submit);
4950 dev_nr++;
4951 }
4952 return 0;
4953 }
4954
4955 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4956 u8 *uuid, u8 *fsid)
4957 {
4958 struct btrfs_device *device;
4959 struct btrfs_fs_devices *cur_devices;
4960
4961 cur_devices = fs_info->fs_devices;
4962 while (cur_devices) {
4963 if (!fsid ||
4964 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4965 device = __find_device(&cur_devices->devices,
4966 devid, uuid);
4967 if (device)
4968 return device;
4969 }
4970 cur_devices = cur_devices->seed;
4971 }
4972 return NULL;
4973 }
4974
4975 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4976 u64 devid, u8 *dev_uuid)
4977 {
4978 struct btrfs_device *device;
4979 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4980
4981 device = kzalloc(sizeof(*device), GFP_NOFS);
4982 if (!device)
4983 return NULL;
4984 list_add(&device->dev_list,
4985 &fs_devices->devices);
4986 device->dev_root = root->fs_info->dev_root;
4987 device->devid = devid;
4988 device->work.func = pending_bios_fn;
4989 device->fs_devices = fs_devices;
4990 device->missing = 1;
4991 fs_devices->num_devices++;
4992 fs_devices->missing_devices++;
4993 spin_lock_init(&device->io_lock);
4994 INIT_LIST_HEAD(&device->dev_alloc_list);
4995 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4996 return device;
4997 }
4998
4999 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5000 struct extent_buffer *leaf,
5001 struct btrfs_chunk *chunk)
5002 {
5003 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5004 struct map_lookup *map;
5005 struct extent_map *em;
5006 u64 logical;
5007 u64 length;
5008 u64 devid;
5009 u8 uuid[BTRFS_UUID_SIZE];
5010 int num_stripes;
5011 int ret;
5012 int i;
5013
5014 logical = key->offset;
5015 length = btrfs_chunk_length(leaf, chunk);
5016
5017 read_lock(&map_tree->map_tree.lock);
5018 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5019 read_unlock(&map_tree->map_tree.lock);
5020
5021 /* already mapped? */
5022 if (em && em->start <= logical && em->start + em->len > logical) {
5023 free_extent_map(em);
5024 return 0;
5025 } else if (em) {
5026 free_extent_map(em);
5027 }
5028
5029 em = alloc_extent_map();
5030 if (!em)
5031 return -ENOMEM;
5032 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5033 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5034 if (!map) {
5035 free_extent_map(em);
5036 return -ENOMEM;
5037 }
5038
5039 em->bdev = (struct block_device *)map;
5040 em->start = logical;
5041 em->len = length;
5042 em->orig_start = 0;
5043 em->block_start = 0;
5044 em->block_len = em->len;
5045
5046 map->num_stripes = num_stripes;
5047 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5048 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5049 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5050 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5051 map->type = btrfs_chunk_type(leaf, chunk);
5052 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5053 for (i = 0; i < num_stripes; i++) {
5054 map->stripes[i].physical =
5055 btrfs_stripe_offset_nr(leaf, chunk, i);
5056 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5057 read_extent_buffer(leaf, uuid, (unsigned long)
5058 btrfs_stripe_dev_uuid_nr(chunk, i),
5059 BTRFS_UUID_SIZE);
5060 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5061 uuid, NULL);
5062 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5063 kfree(map);
5064 free_extent_map(em);
5065 return -EIO;
5066 }
5067 if (!map->stripes[i].dev) {
5068 map->stripes[i].dev =
5069 add_missing_dev(root, devid, uuid);
5070 if (!map->stripes[i].dev) {
5071 kfree(map);
5072 free_extent_map(em);
5073 return -EIO;
5074 }
5075 }
5076 map->stripes[i].dev->in_fs_metadata = 1;
5077 }
5078
5079 write_lock(&map_tree->map_tree.lock);
5080 ret = add_extent_mapping(&map_tree->map_tree, em);
5081 write_unlock(&map_tree->map_tree.lock);
5082 BUG_ON(ret); /* Tree corruption */
5083 free_extent_map(em);
5084
5085 return 0;
5086 }
5087
5088 static void fill_device_from_item(struct extent_buffer *leaf,
5089 struct btrfs_dev_item *dev_item,
5090 struct btrfs_device *device)
5091 {
5092 unsigned long ptr;
5093
5094 device->devid = btrfs_device_id(leaf, dev_item);
5095 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5096 device->total_bytes = device->disk_total_bytes;
5097 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5098 device->type = btrfs_device_type(leaf, dev_item);
5099 device->io_align = btrfs_device_io_align(leaf, dev_item);
5100 device->io_width = btrfs_device_io_width(leaf, dev_item);
5101 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5102 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5103 device->is_tgtdev_for_dev_replace = 0;
5104
5105 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5106 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5107 }
5108
5109 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5110 {
5111 struct btrfs_fs_devices *fs_devices;
5112 int ret;
5113
5114 BUG_ON(!mutex_is_locked(&uuid_mutex));
5115
5116 fs_devices = root->fs_info->fs_devices->seed;
5117 while (fs_devices) {
5118 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5119 ret = 0;
5120 goto out;
5121 }
5122 fs_devices = fs_devices->seed;
5123 }
5124
5125 fs_devices = find_fsid(fsid);
5126 if (!fs_devices) {
5127 ret = -ENOENT;
5128 goto out;
5129 }
5130
5131 fs_devices = clone_fs_devices(fs_devices);
5132 if (IS_ERR(fs_devices)) {
5133 ret = PTR_ERR(fs_devices);
5134 goto out;
5135 }
5136
5137 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5138 root->fs_info->bdev_holder);
5139 if (ret) {
5140 free_fs_devices(fs_devices);
5141 goto out;
5142 }
5143
5144 if (!fs_devices->seeding) {
5145 __btrfs_close_devices(fs_devices);
5146 free_fs_devices(fs_devices);
5147 ret = -EINVAL;
5148 goto out;
5149 }
5150
5151 fs_devices->seed = root->fs_info->fs_devices->seed;
5152 root->fs_info->fs_devices->seed = fs_devices;
5153 out:
5154 return ret;
5155 }
5156
5157 static int read_one_dev(struct btrfs_root *root,
5158 struct extent_buffer *leaf,
5159 struct btrfs_dev_item *dev_item)
5160 {
5161 struct btrfs_device *device;
5162 u64 devid;
5163 int ret;
5164 u8 fs_uuid[BTRFS_UUID_SIZE];
5165 u8 dev_uuid[BTRFS_UUID_SIZE];
5166
5167 devid = btrfs_device_id(leaf, dev_item);
5168 read_extent_buffer(leaf, dev_uuid,
5169 (unsigned long)btrfs_device_uuid(dev_item),
5170 BTRFS_UUID_SIZE);
5171 read_extent_buffer(leaf, fs_uuid,
5172 (unsigned long)btrfs_device_fsid(dev_item),
5173 BTRFS_UUID_SIZE);
5174
5175 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5176 ret = open_seed_devices(root, fs_uuid);
5177 if (ret && !btrfs_test_opt(root, DEGRADED))
5178 return ret;
5179 }
5180
5181 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5182 if (!device || !device->bdev) {
5183 if (!btrfs_test_opt(root, DEGRADED))
5184 return -EIO;
5185
5186 if (!device) {
5187 printk(KERN_WARNING "warning devid %llu missing\n",
5188 (unsigned long long)devid);
5189 device = add_missing_dev(root, devid, dev_uuid);
5190 if (!device)
5191 return -ENOMEM;
5192 } else if (!device->missing) {
5193 /*
5194 * this happens when a device that was properly setup
5195 * in the device info lists suddenly goes bad.
5196 * device->bdev is NULL, and so we have to set
5197 * device->missing to one here
5198 */
5199 root->fs_info->fs_devices->missing_devices++;
5200 device->missing = 1;
5201 }
5202 }
5203
5204 if (device->fs_devices != root->fs_info->fs_devices) {
5205 BUG_ON(device->writeable);
5206 if (device->generation !=
5207 btrfs_device_generation(leaf, dev_item))
5208 return -EINVAL;
5209 }
5210
5211 fill_device_from_item(leaf, dev_item, device);
5212 device->dev_root = root->fs_info->dev_root;
5213 device->in_fs_metadata = 1;
5214 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5215 device->fs_devices->total_rw_bytes += device->total_bytes;
5216 spin_lock(&root->fs_info->free_chunk_lock);
5217 root->fs_info->free_chunk_space += device->total_bytes -
5218 device->bytes_used;
5219 spin_unlock(&root->fs_info->free_chunk_lock);
5220 }
5221 ret = 0;
5222 return ret;
5223 }
5224
5225 int btrfs_read_sys_array(struct btrfs_root *root)
5226 {
5227 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5228 struct extent_buffer *sb;
5229 struct btrfs_disk_key *disk_key;
5230 struct btrfs_chunk *chunk;
5231 u8 *ptr;
5232 unsigned long sb_ptr;
5233 int ret = 0;
5234 u32 num_stripes;
5235 u32 array_size;
5236 u32 len = 0;
5237 u32 cur;
5238 struct btrfs_key key;
5239
5240 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5241 BTRFS_SUPER_INFO_SIZE);
5242 if (!sb)
5243 return -ENOMEM;
5244 btrfs_set_buffer_uptodate(sb);
5245 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5246 /*
5247 * The sb extent buffer is artifical and just used to read the system array.
5248 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5249 * pages up-to-date when the page is larger: extent does not cover the
5250 * whole page and consequently check_page_uptodate does not find all
5251 * the page's extents up-to-date (the hole beyond sb),
5252 * write_extent_buffer then triggers a WARN_ON.
5253 *
5254 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5255 * but sb spans only this function. Add an explicit SetPageUptodate call
5256 * to silence the warning eg. on PowerPC 64.
5257 */
5258 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5259 SetPageUptodate(sb->pages[0]);
5260
5261 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5262 array_size = btrfs_super_sys_array_size(super_copy);
5263
5264 ptr = super_copy->sys_chunk_array;
5265 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5266 cur = 0;
5267
5268 while (cur < array_size) {
5269 disk_key = (struct btrfs_disk_key *)ptr;
5270 btrfs_disk_key_to_cpu(&key, disk_key);
5271
5272 len = sizeof(*disk_key); ptr += len;
5273 sb_ptr += len;
5274 cur += len;
5275
5276 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5277 chunk = (struct btrfs_chunk *)sb_ptr;
5278 ret = read_one_chunk(root, &key, sb, chunk);
5279 if (ret)
5280 break;
5281 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5282 len = btrfs_chunk_item_size(num_stripes);
5283 } else {
5284 ret = -EIO;
5285 break;
5286 }
5287 ptr += len;
5288 sb_ptr += len;
5289 cur += len;
5290 }
5291 free_extent_buffer(sb);
5292 return ret;
5293 }
5294
5295 int btrfs_read_chunk_tree(struct btrfs_root *root)
5296 {
5297 struct btrfs_path *path;
5298 struct extent_buffer *leaf;
5299 struct btrfs_key key;
5300 struct btrfs_key found_key;
5301 int ret;
5302 int slot;
5303
5304 root = root->fs_info->chunk_root;
5305
5306 path = btrfs_alloc_path();
5307 if (!path)
5308 return -ENOMEM;
5309
5310 mutex_lock(&uuid_mutex);
5311 lock_chunks(root);
5312
5313 /* first we search for all of the device items, and then we
5314 * read in all of the chunk items. This way we can create chunk
5315 * mappings that reference all of the devices that are afound
5316 */
5317 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5318 key.offset = 0;
5319 key.type = 0;
5320 again:
5321 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5322 if (ret < 0)
5323 goto error;
5324 while (1) {
5325 leaf = path->nodes[0];
5326 slot = path->slots[0];
5327 if (slot >= btrfs_header_nritems(leaf)) {
5328 ret = btrfs_next_leaf(root, path);
5329 if (ret == 0)
5330 continue;
5331 if (ret < 0)
5332 goto error;
5333 break;
5334 }
5335 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5336 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5337 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5338 break;
5339 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5340 struct btrfs_dev_item *dev_item;
5341 dev_item = btrfs_item_ptr(leaf, slot,
5342 struct btrfs_dev_item);
5343 ret = read_one_dev(root, leaf, dev_item);
5344 if (ret)
5345 goto error;
5346 }
5347 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5348 struct btrfs_chunk *chunk;
5349 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5350 ret = read_one_chunk(root, &found_key, leaf, chunk);
5351 if (ret)
5352 goto error;
5353 }
5354 path->slots[0]++;
5355 }
5356 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5357 key.objectid = 0;
5358 btrfs_release_path(path);
5359 goto again;
5360 }
5361 ret = 0;
5362 error:
5363 unlock_chunks(root);
5364 mutex_unlock(&uuid_mutex);
5365
5366 btrfs_free_path(path);
5367 return ret;
5368 }
5369
5370 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5371 {
5372 int i;
5373
5374 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5375 btrfs_dev_stat_reset(dev, i);
5376 }
5377
5378 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5379 {
5380 struct btrfs_key key;
5381 struct btrfs_key found_key;
5382 struct btrfs_root *dev_root = fs_info->dev_root;
5383 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5384 struct extent_buffer *eb;
5385 int slot;
5386 int ret = 0;
5387 struct btrfs_device *device;
5388 struct btrfs_path *path = NULL;
5389 int i;
5390
5391 path = btrfs_alloc_path();
5392 if (!path) {
5393 ret = -ENOMEM;
5394 goto out;
5395 }
5396
5397 mutex_lock(&fs_devices->device_list_mutex);
5398 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5399 int item_size;
5400 struct btrfs_dev_stats_item *ptr;
5401
5402 key.objectid = 0;
5403 key.type = BTRFS_DEV_STATS_KEY;
5404 key.offset = device->devid;
5405 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5406 if (ret) {
5407 __btrfs_reset_dev_stats(device);
5408 device->dev_stats_valid = 1;
5409 btrfs_release_path(path);
5410 continue;
5411 }
5412 slot = path->slots[0];
5413 eb = path->nodes[0];
5414 btrfs_item_key_to_cpu(eb, &found_key, slot);
5415 item_size = btrfs_item_size_nr(eb, slot);
5416
5417 ptr = btrfs_item_ptr(eb, slot,
5418 struct btrfs_dev_stats_item);
5419
5420 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5421 if (item_size >= (1 + i) * sizeof(__le64))
5422 btrfs_dev_stat_set(device, i,
5423 btrfs_dev_stats_value(eb, ptr, i));
5424 else
5425 btrfs_dev_stat_reset(device, i);
5426 }
5427
5428 device->dev_stats_valid = 1;
5429 btrfs_dev_stat_print_on_load(device);
5430 btrfs_release_path(path);
5431 }
5432 mutex_unlock(&fs_devices->device_list_mutex);
5433
5434 out:
5435 btrfs_free_path(path);
5436 return ret < 0 ? ret : 0;
5437 }
5438
5439 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5440 struct btrfs_root *dev_root,
5441 struct btrfs_device *device)
5442 {
5443 struct btrfs_path *path;
5444 struct btrfs_key key;
5445 struct extent_buffer *eb;
5446 struct btrfs_dev_stats_item *ptr;
5447 int ret;
5448 int i;
5449
5450 key.objectid = 0;
5451 key.type = BTRFS_DEV_STATS_KEY;
5452 key.offset = device->devid;
5453
5454 path = btrfs_alloc_path();
5455 BUG_ON(!path);
5456 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5457 if (ret < 0) {
5458 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5459 ret, rcu_str_deref(device->name));
5460 goto out;
5461 }
5462
5463 if (ret == 0 &&
5464 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5465 /* need to delete old one and insert a new one */
5466 ret = btrfs_del_item(trans, dev_root, path);
5467 if (ret != 0) {
5468 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5469 rcu_str_deref(device->name), ret);
5470 goto out;
5471 }
5472 ret = 1;
5473 }
5474
5475 if (ret == 1) {
5476 /* need to insert a new item */
5477 btrfs_release_path(path);
5478 ret = btrfs_insert_empty_item(trans, dev_root, path,
5479 &key, sizeof(*ptr));
5480 if (ret < 0) {
5481 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5482 rcu_str_deref(device->name), ret);
5483 goto out;
5484 }
5485 }
5486
5487 eb = path->nodes[0];
5488 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5489 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5490 btrfs_set_dev_stats_value(eb, ptr, i,
5491 btrfs_dev_stat_read(device, i));
5492 btrfs_mark_buffer_dirty(eb);
5493
5494 out:
5495 btrfs_free_path(path);
5496 return ret;
5497 }
5498
5499 /*
5500 * called from commit_transaction. Writes all changed device stats to disk.
5501 */
5502 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5503 struct btrfs_fs_info *fs_info)
5504 {
5505 struct btrfs_root *dev_root = fs_info->dev_root;
5506 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5507 struct btrfs_device *device;
5508 int ret = 0;
5509
5510 mutex_lock(&fs_devices->device_list_mutex);
5511 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5512 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5513 continue;
5514
5515 ret = update_dev_stat_item(trans, dev_root, device);
5516 if (!ret)
5517 device->dev_stats_dirty = 0;
5518 }
5519 mutex_unlock(&fs_devices->device_list_mutex);
5520
5521 return ret;
5522 }
5523
5524 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5525 {
5526 btrfs_dev_stat_inc(dev, index);
5527 btrfs_dev_stat_print_on_error(dev);
5528 }
5529
5530 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5531 {
5532 if (!dev->dev_stats_valid)
5533 return;
5534 printk_ratelimited_in_rcu(KERN_ERR
5535 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5536 rcu_str_deref(dev->name),
5537 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5538 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5539 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5540 btrfs_dev_stat_read(dev,
5541 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5542 btrfs_dev_stat_read(dev,
5543 BTRFS_DEV_STAT_GENERATION_ERRS));
5544 }
5545
5546 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5547 {
5548 int i;
5549
5550 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5551 if (btrfs_dev_stat_read(dev, i) != 0)
5552 break;
5553 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5554 return; /* all values == 0, suppress message */
5555
5556 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5557 rcu_str_deref(dev->name),
5558 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5559 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5560 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5561 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5562 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5563 }
5564
5565 int btrfs_get_dev_stats(struct btrfs_root *root,
5566 struct btrfs_ioctl_get_dev_stats *stats)
5567 {
5568 struct btrfs_device *dev;
5569 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5570 int i;
5571
5572 mutex_lock(&fs_devices->device_list_mutex);
5573 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5574 mutex_unlock(&fs_devices->device_list_mutex);
5575
5576 if (!dev) {
5577 printk(KERN_WARNING
5578 "btrfs: get dev_stats failed, device not found\n");
5579 return -ENODEV;
5580 } else if (!dev->dev_stats_valid) {
5581 printk(KERN_WARNING
5582 "btrfs: get dev_stats failed, not yet valid\n");
5583 return -ENODEV;
5584 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5585 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5586 if (stats->nr_items > i)
5587 stats->values[i] =
5588 btrfs_dev_stat_read_and_reset(dev, i);
5589 else
5590 btrfs_dev_stat_reset(dev, i);
5591 }
5592 } else {
5593 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5594 if (stats->nr_items > i)
5595 stats->values[i] = btrfs_dev_stat_read(dev, i);
5596 }
5597 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5598 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5599 return 0;
5600 }
5601
5602 int btrfs_scratch_superblock(struct btrfs_device *device)
5603 {
5604 struct buffer_head *bh;
5605 struct btrfs_super_block *disk_super;
5606
5607 bh = btrfs_read_dev_super(device->bdev);
5608 if (!bh)
5609 return -EINVAL;
5610 disk_super = (struct btrfs_super_block *)bh->b_data;
5611
5612 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5613 set_buffer_dirty(bh);
5614 sync_dirty_buffer(bh);
5615 brelse(bh);
5616
5617 return 0;
5618 }
Linux kernel - Deliverables
This page took 0.158998 seconds and 5 git commands to generate.