projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: define BTRFS_MAGIC as a u64 value
[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 disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
859 goto error_unmap;
860
861 devid = btrfs_stack_device_id(&disk_super->dev_item);
862 transid = btrfs_super_generation(disk_super);
863 total_devices = btrfs_super_num_devices(disk_super);
864
865 if (disk_super->label[0]) {
866 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
867 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
868 printk(KERN_INFO "device label %s ", disk_super->label);
869 } else {
870 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
871 }
872
873 printk(KERN_CONT "devid %llu transid %llu %s\n",
874 (unsigned long long)devid, (unsigned long long)transid, path);
875
876 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
877 if (!ret && fs_devices_ret)
878 (*fs_devices_ret)->total_devices = total_devices;
879
880 error_unmap:
881 kunmap(page);
882 page_cache_release(page);
883
884 error_bdev_put:
885 blkdev_put(bdev, flags);
886 error:
887 mutex_unlock(&uuid_mutex);
888 return ret;
889 }
890
891 /* helper to account the used device space in the range */
892 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
893 u64 end, u64 *length)
894 {
895 struct btrfs_key key;
896 struct btrfs_root *root = device->dev_root;
897 struct btrfs_dev_extent *dev_extent;
898 struct btrfs_path *path;
899 u64 extent_end;
900 int ret;
901 int slot;
902 struct extent_buffer *l;
903
904 *length = 0;
905
906 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
907 return 0;
908
909 path = btrfs_alloc_path();
910 if (!path)
911 return -ENOMEM;
912 path->reada = 2;
913
914 key.objectid = device->devid;
915 key.offset = start;
916 key.type = BTRFS_DEV_EXTENT_KEY;
917
918 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
919 if (ret < 0)
920 goto out;
921 if (ret > 0) {
922 ret = btrfs_previous_item(root, path, key.objectid, key.type);
923 if (ret < 0)
924 goto out;
925 }
926
927 while (1) {
928 l = path->nodes[0];
929 slot = path->slots[0];
930 if (slot >= btrfs_header_nritems(l)) {
931 ret = btrfs_next_leaf(root, path);
932 if (ret == 0)
933 continue;
934 if (ret < 0)
935 goto out;
936
937 break;
938 }
939 btrfs_item_key_to_cpu(l, &key, slot);
940
941 if (key.objectid < device->devid)
942 goto next;
943
944 if (key.objectid > device->devid)
945 break;
946
947 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
948 goto next;
949
950 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
951 extent_end = key.offset + btrfs_dev_extent_length(l,
952 dev_extent);
953 if (key.offset <= start && extent_end > end) {
954 *length = end - start + 1;
955 break;
956 } else if (key.offset <= start && extent_end > start)
957 *length += extent_end - start;
958 else if (key.offset > start && extent_end <= end)
959 *length += extent_end - key.offset;
960 else if (key.offset > start && key.offset <= end) {
961 *length += end - key.offset + 1;
962 break;
963 } else if (key.offset > end)
964 break;
965
966 next:
967 path->slots[0]++;
968 }
969 ret = 0;
970 out:
971 btrfs_free_path(path);
972 return ret;
973 }
974
975 /*
976 * find_free_dev_extent - find free space in the specified device
977 * @device: the device which we search the free space in
978 * @num_bytes: the size of the free space that we need
979 * @start: store the start of the free space.
980 * @len: the size of the free space. that we find, or the size of the max
981 * free space if we don't find suitable free space
982 *
983 * this uses a pretty simple search, the expectation is that it is
984 * called very infrequently and that a given device has a small number
985 * of extents
986 *
987 * @start is used to store the start of the free space if we find. But if we
988 * don't find suitable free space, it will be used to store the start position
989 * of the max free space.
990 *
991 * @len is used to store the size of the free space that we find.
992 * But if we don't find suitable free space, it is used to store the size of
993 * the max free space.
994 */
995 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
996 u64 *start, u64 *len)
997 {
998 struct btrfs_key key;
999 struct btrfs_root *root = device->dev_root;
1000 struct btrfs_dev_extent *dev_extent;
1001 struct btrfs_path *path;
1002 u64 hole_size;
1003 u64 max_hole_start;
1004 u64 max_hole_size;
1005 u64 extent_end;
1006 u64 search_start;
1007 u64 search_end = device->total_bytes;
1008 int ret;
1009 int slot;
1010 struct extent_buffer *l;
1011
1012 /* FIXME use last free of some kind */
1013
1014 /* we don't want to overwrite the superblock on the drive,
1015 * so we make sure to start at an offset of at least 1MB
1016 */
1017 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1018
1019 max_hole_start = search_start;
1020 max_hole_size = 0;
1021 hole_size = 0;
1022
1023 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1024 ret = -ENOSPC;
1025 goto error;
1026 }
1027
1028 path = btrfs_alloc_path();
1029 if (!path) {
1030 ret = -ENOMEM;
1031 goto error;
1032 }
1033 path->reada = 2;
1034
1035 key.objectid = device->devid;
1036 key.offset = search_start;
1037 key.type = BTRFS_DEV_EXTENT_KEY;
1038
1039 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1040 if (ret < 0)
1041 goto out;
1042 if (ret > 0) {
1043 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1044 if (ret < 0)
1045 goto out;
1046 }
1047
1048 while (1) {
1049 l = path->nodes[0];
1050 slot = path->slots[0];
1051 if (slot >= btrfs_header_nritems(l)) {
1052 ret = btrfs_next_leaf(root, path);
1053 if (ret == 0)
1054 continue;
1055 if (ret < 0)
1056 goto out;
1057
1058 break;
1059 }
1060 btrfs_item_key_to_cpu(l, &key, slot);
1061
1062 if (key.objectid < device->devid)
1063 goto next;
1064
1065 if (key.objectid > device->devid)
1066 break;
1067
1068 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1069 goto next;
1070
1071 if (key.offset > search_start) {
1072 hole_size = key.offset - search_start;
1073
1074 if (hole_size > max_hole_size) {
1075 max_hole_start = search_start;
1076 max_hole_size = hole_size;
1077 }
1078
1079 /*
1080 * If this free space is greater than which we need,
1081 * it must be the max free space that we have found
1082 * until now, so max_hole_start must point to the start
1083 * of this free space and the length of this free space
1084 * is stored in max_hole_size. Thus, we return
1085 * max_hole_start and max_hole_size and go back to the
1086 * caller.
1087 */
1088 if (hole_size >= num_bytes) {
1089 ret = 0;
1090 goto out;
1091 }
1092 }
1093
1094 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1095 extent_end = key.offset + btrfs_dev_extent_length(l,
1096 dev_extent);
1097 if (extent_end > search_start)
1098 search_start = extent_end;
1099 next:
1100 path->slots[0]++;
1101 cond_resched();
1102 }
1103
1104 /*
1105 * At this point, search_start should be the end of
1106 * allocated dev extents, and when shrinking the device,
1107 * search_end may be smaller than search_start.
1108 */
1109 if (search_end > search_start)
1110 hole_size = search_end - search_start;
1111
1112 if (hole_size > max_hole_size) {
1113 max_hole_start = search_start;
1114 max_hole_size = hole_size;
1115 }
1116
1117 /* See above. */
1118 if (hole_size < num_bytes)
1119 ret = -ENOSPC;
1120 else
1121 ret = 0;
1122
1123 out:
1124 btrfs_free_path(path);
1125 error:
1126 *start = max_hole_start;
1127 if (len)
1128 *len = max_hole_size;
1129 return ret;
1130 }
1131
1132 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1133 struct btrfs_device *device,
1134 u64 start)
1135 {
1136 int ret;
1137 struct btrfs_path *path;
1138 struct btrfs_root *root = device->dev_root;
1139 struct btrfs_key key;
1140 struct btrfs_key found_key;
1141 struct extent_buffer *leaf = NULL;
1142 struct btrfs_dev_extent *extent = NULL;
1143
1144 path = btrfs_alloc_path();
1145 if (!path)
1146 return -ENOMEM;
1147
1148 key.objectid = device->devid;
1149 key.offset = start;
1150 key.type = BTRFS_DEV_EXTENT_KEY;
1151 again:
1152 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1153 if (ret > 0) {
1154 ret = btrfs_previous_item(root, path, key.objectid,
1155 BTRFS_DEV_EXTENT_KEY);
1156 if (ret)
1157 goto out;
1158 leaf = path->nodes[0];
1159 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1160 extent = btrfs_item_ptr(leaf, path->slots[0],
1161 struct btrfs_dev_extent);
1162 BUG_ON(found_key.offset > start || found_key.offset +
1163 btrfs_dev_extent_length(leaf, extent) < start);
1164 key = found_key;
1165 btrfs_release_path(path);
1166 goto again;
1167 } else if (ret == 0) {
1168 leaf = path->nodes[0];
1169 extent = btrfs_item_ptr(leaf, path->slots[0],
1170 struct btrfs_dev_extent);
1171 } else {
1172 btrfs_error(root->fs_info, ret, "Slot search failed");
1173 goto out;
1174 }
1175
1176 if (device->bytes_used > 0) {
1177 u64 len = btrfs_dev_extent_length(leaf, extent);
1178 device->bytes_used -= len;
1179 spin_lock(&root->fs_info->free_chunk_lock);
1180 root->fs_info->free_chunk_space += len;
1181 spin_unlock(&root->fs_info->free_chunk_lock);
1182 }
1183 ret = btrfs_del_item(trans, root, path);
1184 if (ret) {
1185 btrfs_error(root->fs_info, ret,
1186 "Failed to remove dev extent item");
1187 }
1188 out:
1189 btrfs_free_path(path);
1190 return ret;
1191 }
1192
1193 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1194 struct btrfs_device *device,
1195 u64 chunk_tree, u64 chunk_objectid,
1196 u64 chunk_offset, u64 start, u64 num_bytes)
1197 {
1198 int ret;
1199 struct btrfs_path *path;
1200 struct btrfs_root *root = device->dev_root;
1201 struct btrfs_dev_extent *extent;
1202 struct extent_buffer *leaf;
1203 struct btrfs_key key;
1204
1205 WARN_ON(!device->in_fs_metadata);
1206 WARN_ON(device->is_tgtdev_for_dev_replace);
1207 path = btrfs_alloc_path();
1208 if (!path)
1209 return -ENOMEM;
1210
1211 key.objectid = device->devid;
1212 key.offset = start;
1213 key.type = BTRFS_DEV_EXTENT_KEY;
1214 ret = btrfs_insert_empty_item(trans, root, path, &key,
1215 sizeof(*extent));
1216 if (ret)
1217 goto out;
1218
1219 leaf = path->nodes[0];
1220 extent = btrfs_item_ptr(leaf, path->slots[0],
1221 struct btrfs_dev_extent);
1222 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1223 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1224 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1225
1226 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1227 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1228 BTRFS_UUID_SIZE);
1229
1230 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1231 btrfs_mark_buffer_dirty(leaf);
1232 out:
1233 btrfs_free_path(path);
1234 return ret;
1235 }
1236
1237 static noinline int find_next_chunk(struct btrfs_root *root,
1238 u64 objectid, u64 *offset)
1239 {
1240 struct btrfs_path *path;
1241 int ret;
1242 struct btrfs_key key;
1243 struct btrfs_chunk *chunk;
1244 struct btrfs_key found_key;
1245
1246 path = btrfs_alloc_path();
1247 if (!path)
1248 return -ENOMEM;
1249
1250 key.objectid = objectid;
1251 key.offset = (u64)-1;
1252 key.type = BTRFS_CHUNK_ITEM_KEY;
1253
1254 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1255 if (ret < 0)
1256 goto error;
1257
1258 BUG_ON(ret == 0); /* Corruption */
1259
1260 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1261 if (ret) {
1262 *offset = 0;
1263 } else {
1264 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1265 path->slots[0]);
1266 if (found_key.objectid != objectid)
1267 *offset = 0;
1268 else {
1269 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1270 struct btrfs_chunk);
1271 *offset = found_key.offset +
1272 btrfs_chunk_length(path->nodes[0], chunk);
1273 }
1274 }
1275 ret = 0;
1276 error:
1277 btrfs_free_path(path);
1278 return ret;
1279 }
1280
1281 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1282 {
1283 int ret;
1284 struct btrfs_key key;
1285 struct btrfs_key found_key;
1286 struct btrfs_path *path;
1287
1288 root = root->fs_info->chunk_root;
1289
1290 path = btrfs_alloc_path();
1291 if (!path)
1292 return -ENOMEM;
1293
1294 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1295 key.type = BTRFS_DEV_ITEM_KEY;
1296 key.offset = (u64)-1;
1297
1298 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1299 if (ret < 0)
1300 goto error;
1301
1302 BUG_ON(ret == 0); /* Corruption */
1303
1304 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1305 BTRFS_DEV_ITEM_KEY);
1306 if (ret) {
1307 *objectid = 1;
1308 } else {
1309 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1310 path->slots[0]);
1311 *objectid = found_key.offset + 1;
1312 }
1313 ret = 0;
1314 error:
1315 btrfs_free_path(path);
1316 return ret;
1317 }
1318
1319 /*
1320 * the device information is stored in the chunk root
1321 * the btrfs_device struct should be fully filled in
1322 */
1323 int btrfs_add_device(struct btrfs_trans_handle *trans,
1324 struct btrfs_root *root,
1325 struct btrfs_device *device)
1326 {
1327 int ret;
1328 struct btrfs_path *path;
1329 struct btrfs_dev_item *dev_item;
1330 struct extent_buffer *leaf;
1331 struct btrfs_key key;
1332 unsigned long ptr;
1333
1334 root = root->fs_info->chunk_root;
1335
1336 path = btrfs_alloc_path();
1337 if (!path)
1338 return -ENOMEM;
1339
1340 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1341 key.type = BTRFS_DEV_ITEM_KEY;
1342 key.offset = device->devid;
1343
1344 ret = btrfs_insert_empty_item(trans, root, path, &key,
1345 sizeof(*dev_item));
1346 if (ret)
1347 goto out;
1348
1349 leaf = path->nodes[0];
1350 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1351
1352 btrfs_set_device_id(leaf, dev_item, device->devid);
1353 btrfs_set_device_generation(leaf, dev_item, 0);
1354 btrfs_set_device_type(leaf, dev_item, device->type);
1355 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1356 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1357 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1358 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1359 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1360 btrfs_set_device_group(leaf, dev_item, 0);
1361 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1362 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1363 btrfs_set_device_start_offset(leaf, dev_item, 0);
1364
1365 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1366 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1367 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1368 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1369 btrfs_mark_buffer_dirty(leaf);
1370
1371 ret = 0;
1372 out:
1373 btrfs_free_path(path);
1374 return ret;
1375 }
1376
1377 static int btrfs_rm_dev_item(struct btrfs_root *root,
1378 struct btrfs_device *device)
1379 {
1380 int ret;
1381 struct btrfs_path *path;
1382 struct btrfs_key key;
1383 struct btrfs_trans_handle *trans;
1384
1385 root = root->fs_info->chunk_root;
1386
1387 path = btrfs_alloc_path();
1388 if (!path)
1389 return -ENOMEM;
1390
1391 trans = btrfs_start_transaction(root, 0);
1392 if (IS_ERR(trans)) {
1393 btrfs_free_path(path);
1394 return PTR_ERR(trans);
1395 }
1396 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1397 key.type = BTRFS_DEV_ITEM_KEY;
1398 key.offset = device->devid;
1399 lock_chunks(root);
1400
1401 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1402 if (ret < 0)
1403 goto out;
1404
1405 if (ret > 0) {
1406 ret = -ENOENT;
1407 goto out;
1408 }
1409
1410 ret = btrfs_del_item(trans, root, path);
1411 if (ret)
1412 goto out;
1413 out:
1414 btrfs_free_path(path);
1415 unlock_chunks(root);
1416 btrfs_commit_transaction(trans, root);
1417 return ret;
1418 }
1419
1420 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1421 {
1422 struct btrfs_device *device;
1423 struct btrfs_device *next_device;
1424 struct block_device *bdev;
1425 struct buffer_head *bh = NULL;
1426 struct btrfs_super_block *disk_super;
1427 struct btrfs_fs_devices *cur_devices;
1428 u64 all_avail;
1429 u64 devid;
1430 u64 num_devices;
1431 u8 *dev_uuid;
1432 unsigned seq;
1433 int ret = 0;
1434 bool clear_super = false;
1435
1436 mutex_lock(&uuid_mutex);
1437
1438 do {
1439 seq = read_seqbegin(&root->fs_info->profiles_lock);
1440
1441 all_avail = root->fs_info->avail_data_alloc_bits |
1442 root->fs_info->avail_system_alloc_bits |
1443 root->fs_info->avail_metadata_alloc_bits;
1444 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1445
1446 num_devices = root->fs_info->fs_devices->num_devices;
1447 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1448 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1449 WARN_ON(num_devices < 1);
1450 num_devices--;
1451 }
1452 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1453
1454 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1455 printk(KERN_ERR "btrfs: unable to go below four devices "
1456 "on raid10\n");
1457 ret = -EINVAL;
1458 goto out;
1459 }
1460
1461 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1462 printk(KERN_ERR "btrfs: unable to go below two "
1463 "devices on raid1\n");
1464 ret = -EINVAL;
1465 goto out;
1466 }
1467
1468 if (strcmp(device_path, "missing") == 0) {
1469 struct list_head *devices;
1470 struct btrfs_device *tmp;
1471
1472 device = NULL;
1473 devices = &root->fs_info->fs_devices->devices;
1474 /*
1475 * It is safe to read the devices since the volume_mutex
1476 * is held.
1477 */
1478 list_for_each_entry(tmp, devices, dev_list) {
1479 if (tmp->in_fs_metadata &&
1480 !tmp->is_tgtdev_for_dev_replace &&
1481 !tmp->bdev) {
1482 device = tmp;
1483 break;
1484 }
1485 }
1486 bdev = NULL;
1487 bh = NULL;
1488 disk_super = NULL;
1489 if (!device) {
1490 printk(KERN_ERR "btrfs: no missing devices found to "
1491 "remove\n");
1492 goto out;
1493 }
1494 } else {
1495 ret = btrfs_get_bdev_and_sb(device_path,
1496 FMODE_WRITE | FMODE_EXCL,
1497 root->fs_info->bdev_holder, 0,
1498 &bdev, &bh);
1499 if (ret)
1500 goto out;
1501 disk_super = (struct btrfs_super_block *)bh->b_data;
1502 devid = btrfs_stack_device_id(&disk_super->dev_item);
1503 dev_uuid = disk_super->dev_item.uuid;
1504 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1505 disk_super->fsid);
1506 if (!device) {
1507 ret = -ENOENT;
1508 goto error_brelse;
1509 }
1510 }
1511
1512 if (device->is_tgtdev_for_dev_replace) {
1513 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1514 ret = -EINVAL;
1515 goto error_brelse;
1516 }
1517
1518 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1519 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1520 "device\n");
1521 ret = -EINVAL;
1522 goto error_brelse;
1523 }
1524
1525 if (device->writeable) {
1526 lock_chunks(root);
1527 list_del_init(&device->dev_alloc_list);
1528 unlock_chunks(root);
1529 root->fs_info->fs_devices->rw_devices--;
1530 clear_super = true;
1531 }
1532
1533 ret = btrfs_shrink_device(device, 0);
1534 if (ret)
1535 goto error_undo;
1536
1537 /*
1538 * TODO: the superblock still includes this device in its num_devices
1539 * counter although write_all_supers() is not locked out. This
1540 * could give a filesystem state which requires a degraded mount.
1541 */
1542 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1543 if (ret)
1544 goto error_undo;
1545
1546 spin_lock(&root->fs_info->free_chunk_lock);
1547 root->fs_info->free_chunk_space = device->total_bytes -
1548 device->bytes_used;
1549 spin_unlock(&root->fs_info->free_chunk_lock);
1550
1551 device->in_fs_metadata = 0;
1552 btrfs_scrub_cancel_dev(root->fs_info, device);
1553
1554 /*
1555 * the device list mutex makes sure that we don't change
1556 * the device list while someone else is writing out all
1557 * the device supers.
1558 */
1559
1560 cur_devices = device->fs_devices;
1561 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1562 list_del_rcu(&device->dev_list);
1563
1564 device->fs_devices->num_devices--;
1565 device->fs_devices->total_devices--;
1566
1567 if (device->missing)
1568 root->fs_info->fs_devices->missing_devices--;
1569
1570 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1571 struct btrfs_device, dev_list);
1572 if (device->bdev == root->fs_info->sb->s_bdev)
1573 root->fs_info->sb->s_bdev = next_device->bdev;
1574 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1575 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1576
1577 if (device->bdev)
1578 device->fs_devices->open_devices--;
1579
1580 call_rcu(&device->rcu, free_device);
1581 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1582
1583 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1584 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1585
1586 if (cur_devices->open_devices == 0) {
1587 struct btrfs_fs_devices *fs_devices;
1588 fs_devices = root->fs_info->fs_devices;
1589 while (fs_devices) {
1590 if (fs_devices->seed == cur_devices)
1591 break;
1592 fs_devices = fs_devices->seed;
1593 }
1594 fs_devices->seed = cur_devices->seed;
1595 cur_devices->seed = NULL;
1596 lock_chunks(root);
1597 __btrfs_close_devices(cur_devices);
1598 unlock_chunks(root);
1599 free_fs_devices(cur_devices);
1600 }
1601
1602 root->fs_info->num_tolerated_disk_barrier_failures =
1603 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1604
1605 /*
1606 * at this point, the device is zero sized. We want to
1607 * remove it from the devices list and zero out the old super
1608 */
1609 if (clear_super && disk_super) {
1610 /* make sure this device isn't detected as part of
1611 * the FS anymore
1612 */
1613 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1614 set_buffer_dirty(bh);
1615 sync_dirty_buffer(bh);
1616 }
1617
1618 ret = 0;
1619
1620 /* Notify udev that device has changed */
1621 if (bdev)
1622 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1623
1624 error_brelse:
1625 brelse(bh);
1626 if (bdev)
1627 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1628 out:
1629 mutex_unlock(&uuid_mutex);
1630 return ret;
1631 error_undo:
1632 if (device->writeable) {
1633 lock_chunks(root);
1634 list_add(&device->dev_alloc_list,
1635 &root->fs_info->fs_devices->alloc_list);
1636 unlock_chunks(root);
1637 root->fs_info->fs_devices->rw_devices++;
1638 }
1639 goto error_brelse;
1640 }
1641
1642 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1643 struct btrfs_device *srcdev)
1644 {
1645 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1646 list_del_rcu(&srcdev->dev_list);
1647 list_del_rcu(&srcdev->dev_alloc_list);
1648 fs_info->fs_devices->num_devices--;
1649 if (srcdev->missing) {
1650 fs_info->fs_devices->missing_devices--;
1651 fs_info->fs_devices->rw_devices++;
1652 }
1653 if (srcdev->can_discard)
1654 fs_info->fs_devices->num_can_discard--;
1655 if (srcdev->bdev)
1656 fs_info->fs_devices->open_devices--;
1657
1658 call_rcu(&srcdev->rcu, free_device);
1659 }
1660
1661 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1662 struct btrfs_device *tgtdev)
1663 {
1664 struct btrfs_device *next_device;
1665
1666 WARN_ON(!tgtdev);
1667 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1668 if (tgtdev->bdev) {
1669 btrfs_scratch_superblock(tgtdev);
1670 fs_info->fs_devices->open_devices--;
1671 }
1672 fs_info->fs_devices->num_devices--;
1673 if (tgtdev->can_discard)
1674 fs_info->fs_devices->num_can_discard++;
1675
1676 next_device = list_entry(fs_info->fs_devices->devices.next,
1677 struct btrfs_device, dev_list);
1678 if (tgtdev->bdev == fs_info->sb->s_bdev)
1679 fs_info->sb->s_bdev = next_device->bdev;
1680 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1681 fs_info->fs_devices->latest_bdev = next_device->bdev;
1682 list_del_rcu(&tgtdev->dev_list);
1683
1684 call_rcu(&tgtdev->rcu, free_device);
1685
1686 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1687 }
1688
1689 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1690 struct btrfs_device **device)
1691 {
1692 int ret = 0;
1693 struct btrfs_super_block *disk_super;
1694 u64 devid;
1695 u8 *dev_uuid;
1696 struct block_device *bdev;
1697 struct buffer_head *bh;
1698
1699 *device = NULL;
1700 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1701 root->fs_info->bdev_holder, 0, &bdev, &bh);
1702 if (ret)
1703 return ret;
1704 disk_super = (struct btrfs_super_block *)bh->b_data;
1705 devid = btrfs_stack_device_id(&disk_super->dev_item);
1706 dev_uuid = disk_super->dev_item.uuid;
1707 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1708 disk_super->fsid);
1709 brelse(bh);
1710 if (!*device)
1711 ret = -ENOENT;
1712 blkdev_put(bdev, FMODE_READ);
1713 return ret;
1714 }
1715
1716 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1717 char *device_path,
1718 struct btrfs_device **device)
1719 {
1720 *device = NULL;
1721 if (strcmp(device_path, "missing") == 0) {
1722 struct list_head *devices;
1723 struct btrfs_device *tmp;
1724
1725 devices = &root->fs_info->fs_devices->devices;
1726 /*
1727 * It is safe to read the devices since the volume_mutex
1728 * is held by the caller.
1729 */
1730 list_for_each_entry(tmp, devices, dev_list) {
1731 if (tmp->in_fs_metadata && !tmp->bdev) {
1732 *device = tmp;
1733 break;
1734 }
1735 }
1736
1737 if (!*device) {
1738 pr_err("btrfs: no missing device found\n");
1739 return -ENOENT;
1740 }
1741
1742 return 0;
1743 } else {
1744 return btrfs_find_device_by_path(root, device_path, device);
1745 }
1746 }
1747
1748 /*
1749 * does all the dirty work required for changing file system's UUID.
1750 */
1751 static int btrfs_prepare_sprout(struct btrfs_root *root)
1752 {
1753 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1754 struct btrfs_fs_devices *old_devices;
1755 struct btrfs_fs_devices *seed_devices;
1756 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1757 struct btrfs_device *device;
1758 u64 super_flags;
1759
1760 BUG_ON(!mutex_is_locked(&uuid_mutex));
1761 if (!fs_devices->seeding)
1762 return -EINVAL;
1763
1764 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1765 if (!seed_devices)
1766 return -ENOMEM;
1767
1768 old_devices = clone_fs_devices(fs_devices);
1769 if (IS_ERR(old_devices)) {
1770 kfree(seed_devices);
1771 return PTR_ERR(old_devices);
1772 }
1773
1774 list_add(&old_devices->list, &fs_uuids);
1775
1776 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1777 seed_devices->opened = 1;
1778 INIT_LIST_HEAD(&seed_devices->devices);
1779 INIT_LIST_HEAD(&seed_devices->alloc_list);
1780 mutex_init(&seed_devices->device_list_mutex);
1781
1782 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1783 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1784 synchronize_rcu);
1785 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1786
1787 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1788 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1789 device->fs_devices = seed_devices;
1790 }
1791
1792 fs_devices->seeding = 0;
1793 fs_devices->num_devices = 0;
1794 fs_devices->open_devices = 0;
1795 fs_devices->total_devices = 0;
1796 fs_devices->seed = seed_devices;
1797
1798 generate_random_uuid(fs_devices->fsid);
1799 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1800 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1801 super_flags = btrfs_super_flags(disk_super) &
1802 ~BTRFS_SUPER_FLAG_SEEDING;
1803 btrfs_set_super_flags(disk_super, super_flags);
1804
1805 return 0;
1806 }
1807
1808 /*
1809 * strore the expected generation for seed devices in device items.
1810 */
1811 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1812 struct btrfs_root *root)
1813 {
1814 struct btrfs_path *path;
1815 struct extent_buffer *leaf;
1816 struct btrfs_dev_item *dev_item;
1817 struct btrfs_device *device;
1818 struct btrfs_key key;
1819 u8 fs_uuid[BTRFS_UUID_SIZE];
1820 u8 dev_uuid[BTRFS_UUID_SIZE];
1821 u64 devid;
1822 int ret;
1823
1824 path = btrfs_alloc_path();
1825 if (!path)
1826 return -ENOMEM;
1827
1828 root = root->fs_info->chunk_root;
1829 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1830 key.offset = 0;
1831 key.type = BTRFS_DEV_ITEM_KEY;
1832
1833 while (1) {
1834 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1835 if (ret < 0)
1836 goto error;
1837
1838 leaf = path->nodes[0];
1839 next_slot:
1840 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1841 ret = btrfs_next_leaf(root, path);
1842 if (ret > 0)
1843 break;
1844 if (ret < 0)
1845 goto error;
1846 leaf = path->nodes[0];
1847 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1848 btrfs_release_path(path);
1849 continue;
1850 }
1851
1852 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1853 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1854 key.type != BTRFS_DEV_ITEM_KEY)
1855 break;
1856
1857 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1858 struct btrfs_dev_item);
1859 devid = btrfs_device_id(leaf, dev_item);
1860 read_extent_buffer(leaf, dev_uuid,
1861 (unsigned long)btrfs_device_uuid(dev_item),
1862 BTRFS_UUID_SIZE);
1863 read_extent_buffer(leaf, fs_uuid,
1864 (unsigned long)btrfs_device_fsid(dev_item),
1865 BTRFS_UUID_SIZE);
1866 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1867 fs_uuid);
1868 BUG_ON(!device); /* Logic error */
1869
1870 if (device->fs_devices->seeding) {
1871 btrfs_set_device_generation(leaf, dev_item,
1872 device->generation);
1873 btrfs_mark_buffer_dirty(leaf);
1874 }
1875
1876 path->slots[0]++;
1877 goto next_slot;
1878 }
1879 ret = 0;
1880 error:
1881 btrfs_free_path(path);
1882 return ret;
1883 }
1884
1885 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1886 {
1887 struct request_queue *q;
1888 struct btrfs_trans_handle *trans;
1889 struct btrfs_device *device;
1890 struct block_device *bdev;
1891 struct list_head *devices;
1892 struct super_block *sb = root->fs_info->sb;
1893 struct rcu_string *name;
1894 u64 total_bytes;
1895 int seeding_dev = 0;
1896 int ret = 0;
1897
1898 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1899 return -EROFS;
1900
1901 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1902 root->fs_info->bdev_holder);
1903 if (IS_ERR(bdev))
1904 return PTR_ERR(bdev);
1905
1906 if (root->fs_info->fs_devices->seeding) {
1907 seeding_dev = 1;
1908 down_write(&sb->s_umount);
1909 mutex_lock(&uuid_mutex);
1910 }
1911
1912 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1913
1914 devices = &root->fs_info->fs_devices->devices;
1915
1916 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1917 list_for_each_entry(device, devices, dev_list) {
1918 if (device->bdev == bdev) {
1919 ret = -EEXIST;
1920 mutex_unlock(
1921 &root->fs_info->fs_devices->device_list_mutex);
1922 goto error;
1923 }
1924 }
1925 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1926
1927 device = kzalloc(sizeof(*device), GFP_NOFS);
1928 if (!device) {
1929 /* we can safely leave the fs_devices entry around */
1930 ret = -ENOMEM;
1931 goto error;
1932 }
1933
1934 name = rcu_string_strdup(device_path, GFP_NOFS);
1935 if (!name) {
1936 kfree(device);
1937 ret = -ENOMEM;
1938 goto error;
1939 }
1940 rcu_assign_pointer(device->name, name);
1941
1942 ret = find_next_devid(root, &device->devid);
1943 if (ret) {
1944 rcu_string_free(device->name);
1945 kfree(device);
1946 goto error;
1947 }
1948
1949 trans = btrfs_start_transaction(root, 0);
1950 if (IS_ERR(trans)) {
1951 rcu_string_free(device->name);
1952 kfree(device);
1953 ret = PTR_ERR(trans);
1954 goto error;
1955 }
1956
1957 lock_chunks(root);
1958
1959 q = bdev_get_queue(bdev);
1960 if (blk_queue_discard(q))
1961 device->can_discard = 1;
1962 device->writeable = 1;
1963 device->work.func = pending_bios_fn;
1964 generate_random_uuid(device->uuid);
1965 spin_lock_init(&device->io_lock);
1966 device->generation = trans->transid;
1967 device->io_width = root->sectorsize;
1968 device->io_align = root->sectorsize;
1969 device->sector_size = root->sectorsize;
1970 device->total_bytes = i_size_read(bdev->bd_inode);
1971 device->disk_total_bytes = device->total_bytes;
1972 device->dev_root = root->fs_info->dev_root;
1973 device->bdev = bdev;
1974 device->in_fs_metadata = 1;
1975 device->is_tgtdev_for_dev_replace = 0;
1976 device->mode = FMODE_EXCL;
1977 set_blocksize(device->bdev, 4096);
1978
1979 if (seeding_dev) {
1980 sb->s_flags &= ~MS_RDONLY;
1981 ret = btrfs_prepare_sprout(root);
1982 BUG_ON(ret); /* -ENOMEM */
1983 }
1984
1985 device->fs_devices = root->fs_info->fs_devices;
1986
1987 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1988 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1989 list_add(&device->dev_alloc_list,
1990 &root->fs_info->fs_devices->alloc_list);
1991 root->fs_info->fs_devices->num_devices++;
1992 root->fs_info->fs_devices->open_devices++;
1993 root->fs_info->fs_devices->rw_devices++;
1994 root->fs_info->fs_devices->total_devices++;
1995 if (device->can_discard)
1996 root->fs_info->fs_devices->num_can_discard++;
1997 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1998
1999 spin_lock(&root->fs_info->free_chunk_lock);
2000 root->fs_info->free_chunk_space += device->total_bytes;
2001 spin_unlock(&root->fs_info->free_chunk_lock);
2002
2003 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2004 root->fs_info->fs_devices->rotating = 1;
2005
2006 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2007 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2008 total_bytes + device->total_bytes);
2009
2010 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2011 btrfs_set_super_num_devices(root->fs_info->super_copy,
2012 total_bytes + 1);
2013 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2014
2015 if (seeding_dev) {
2016 ret = init_first_rw_device(trans, root, device);
2017 if (ret) {
2018 btrfs_abort_transaction(trans, root, ret);
2019 goto error_trans;
2020 }
2021 ret = btrfs_finish_sprout(trans, root);
2022 if (ret) {
2023 btrfs_abort_transaction(trans, root, ret);
2024 goto error_trans;
2025 }
2026 } else {
2027 ret = btrfs_add_device(trans, root, device);
2028 if (ret) {
2029 btrfs_abort_transaction(trans, root, ret);
2030 goto error_trans;
2031 }
2032 }
2033
2034 /*
2035 * we've got more storage, clear any full flags on the space
2036 * infos
2037 */
2038 btrfs_clear_space_info_full(root->fs_info);
2039
2040 unlock_chunks(root);
2041 root->fs_info->num_tolerated_disk_barrier_failures =
2042 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2043 ret = btrfs_commit_transaction(trans, root);
2044
2045 if (seeding_dev) {
2046 mutex_unlock(&uuid_mutex);
2047 up_write(&sb->s_umount);
2048
2049 if (ret) /* transaction commit */
2050 return ret;
2051
2052 ret = btrfs_relocate_sys_chunks(root);
2053 if (ret < 0)
2054 btrfs_error(root->fs_info, ret,
2055 "Failed to relocate sys chunks after "
2056 "device initialization. This can be fixed "
2057 "using the \"btrfs balance\" command.");
2058 trans = btrfs_attach_transaction(root);
2059 if (IS_ERR(trans)) {
2060 if (PTR_ERR(trans) == -ENOENT)
2061 return 0;
2062 return PTR_ERR(trans);
2063 }
2064 ret = btrfs_commit_transaction(trans, root);
2065 }
2066
2067 return ret;
2068
2069 error_trans:
2070 unlock_chunks(root);
2071 btrfs_end_transaction(trans, root);
2072 rcu_string_free(device->name);
2073 kfree(device);
2074 error:
2075 blkdev_put(bdev, FMODE_EXCL);
2076 if (seeding_dev) {
2077 mutex_unlock(&uuid_mutex);
2078 up_write(&sb->s_umount);
2079 }
2080 return ret;
2081 }
2082
2083 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2084 struct btrfs_device **device_out)
2085 {
2086 struct request_queue *q;
2087 struct btrfs_device *device;
2088 struct block_device *bdev;
2089 struct btrfs_fs_info *fs_info = root->fs_info;
2090 struct list_head *devices;
2091 struct rcu_string *name;
2092 int ret = 0;
2093
2094 *device_out = NULL;
2095 if (fs_info->fs_devices->seeding)
2096 return -EINVAL;
2097
2098 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2099 fs_info->bdev_holder);
2100 if (IS_ERR(bdev))
2101 return PTR_ERR(bdev);
2102
2103 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2104
2105 devices = &fs_info->fs_devices->devices;
2106 list_for_each_entry(device, devices, dev_list) {
2107 if (device->bdev == bdev) {
2108 ret = -EEXIST;
2109 goto error;
2110 }
2111 }
2112
2113 device = kzalloc(sizeof(*device), GFP_NOFS);
2114 if (!device) {
2115 ret = -ENOMEM;
2116 goto error;
2117 }
2118
2119 name = rcu_string_strdup(device_path, GFP_NOFS);
2120 if (!name) {
2121 kfree(device);
2122 ret = -ENOMEM;
2123 goto error;
2124 }
2125 rcu_assign_pointer(device->name, name);
2126
2127 q = bdev_get_queue(bdev);
2128 if (blk_queue_discard(q))
2129 device->can_discard = 1;
2130 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2131 device->writeable = 1;
2132 device->work.func = pending_bios_fn;
2133 generate_random_uuid(device->uuid);
2134 device->devid = BTRFS_DEV_REPLACE_DEVID;
2135 spin_lock_init(&device->io_lock);
2136 device->generation = 0;
2137 device->io_width = root->sectorsize;
2138 device->io_align = root->sectorsize;
2139 device->sector_size = root->sectorsize;
2140 device->total_bytes = i_size_read(bdev->bd_inode);
2141 device->disk_total_bytes = device->total_bytes;
2142 device->dev_root = fs_info->dev_root;
2143 device->bdev = bdev;
2144 device->in_fs_metadata = 1;
2145 device->is_tgtdev_for_dev_replace = 1;
2146 device->mode = FMODE_EXCL;
2147 set_blocksize(device->bdev, 4096);
2148 device->fs_devices = fs_info->fs_devices;
2149 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2150 fs_info->fs_devices->num_devices++;
2151 fs_info->fs_devices->open_devices++;
2152 if (device->can_discard)
2153 fs_info->fs_devices->num_can_discard++;
2154 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2155
2156 *device_out = device;
2157 return ret;
2158
2159 error:
2160 blkdev_put(bdev, FMODE_EXCL);
2161 return ret;
2162 }
2163
2164 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2165 struct btrfs_device *tgtdev)
2166 {
2167 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2168 tgtdev->io_width = fs_info->dev_root->sectorsize;
2169 tgtdev->io_align = fs_info->dev_root->sectorsize;
2170 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2171 tgtdev->dev_root = fs_info->dev_root;
2172 tgtdev->in_fs_metadata = 1;
2173 }
2174
2175 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2176 struct btrfs_device *device)
2177 {
2178 int ret;
2179 struct btrfs_path *path;
2180 struct btrfs_root *root;
2181 struct btrfs_dev_item *dev_item;
2182 struct extent_buffer *leaf;
2183 struct btrfs_key key;
2184
2185 root = device->dev_root->fs_info->chunk_root;
2186
2187 path = btrfs_alloc_path();
2188 if (!path)
2189 return -ENOMEM;
2190
2191 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2192 key.type = BTRFS_DEV_ITEM_KEY;
2193 key.offset = device->devid;
2194
2195 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2196 if (ret < 0)
2197 goto out;
2198
2199 if (ret > 0) {
2200 ret = -ENOENT;
2201 goto out;
2202 }
2203
2204 leaf = path->nodes[0];
2205 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2206
2207 btrfs_set_device_id(leaf, dev_item, device->devid);
2208 btrfs_set_device_type(leaf, dev_item, device->type);
2209 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2210 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2211 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2212 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2213 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2214 btrfs_mark_buffer_dirty(leaf);
2215
2216 out:
2217 btrfs_free_path(path);
2218 return ret;
2219 }
2220
2221 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2222 struct btrfs_device *device, u64 new_size)
2223 {
2224 struct btrfs_super_block *super_copy =
2225 device->dev_root->fs_info->super_copy;
2226 u64 old_total = btrfs_super_total_bytes(super_copy);
2227 u64 diff = new_size - device->total_bytes;
2228
2229 if (!device->writeable)
2230 return -EACCES;
2231 if (new_size <= device->total_bytes ||
2232 device->is_tgtdev_for_dev_replace)
2233 return -EINVAL;
2234
2235 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2236 device->fs_devices->total_rw_bytes += diff;
2237
2238 device->total_bytes = new_size;
2239 device->disk_total_bytes = new_size;
2240 btrfs_clear_space_info_full(device->dev_root->fs_info);
2241
2242 return btrfs_update_device(trans, device);
2243 }
2244
2245 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2246 struct btrfs_device *device, u64 new_size)
2247 {
2248 int ret;
2249 lock_chunks(device->dev_root);
2250 ret = __btrfs_grow_device(trans, device, new_size);
2251 unlock_chunks(device->dev_root);
2252 return ret;
2253 }
2254
2255 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2256 struct btrfs_root *root,
2257 u64 chunk_tree, u64 chunk_objectid,
2258 u64 chunk_offset)
2259 {
2260 int ret;
2261 struct btrfs_path *path;
2262 struct btrfs_key key;
2263
2264 root = root->fs_info->chunk_root;
2265 path = btrfs_alloc_path();
2266 if (!path)
2267 return -ENOMEM;
2268
2269 key.objectid = chunk_objectid;
2270 key.offset = chunk_offset;
2271 key.type = BTRFS_CHUNK_ITEM_KEY;
2272
2273 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2274 if (ret < 0)
2275 goto out;
2276 else if (ret > 0) { /* Logic error or corruption */
2277 btrfs_error(root->fs_info, -ENOENT,
2278 "Failed lookup while freeing chunk.");
2279 ret = -ENOENT;
2280 goto out;
2281 }
2282
2283 ret = btrfs_del_item(trans, root, path);
2284 if (ret < 0)
2285 btrfs_error(root->fs_info, ret,
2286 "Failed to delete chunk item.");
2287 out:
2288 btrfs_free_path(path);
2289 return ret;
2290 }
2291
2292 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2293 chunk_offset)
2294 {
2295 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2296 struct btrfs_disk_key *disk_key;
2297 struct btrfs_chunk *chunk;
2298 u8 *ptr;
2299 int ret = 0;
2300 u32 num_stripes;
2301 u32 array_size;
2302 u32 len = 0;
2303 u32 cur;
2304 struct btrfs_key key;
2305
2306 array_size = btrfs_super_sys_array_size(super_copy);
2307
2308 ptr = super_copy->sys_chunk_array;
2309 cur = 0;
2310
2311 while (cur < array_size) {
2312 disk_key = (struct btrfs_disk_key *)ptr;
2313 btrfs_disk_key_to_cpu(&key, disk_key);
2314
2315 len = sizeof(*disk_key);
2316
2317 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2318 chunk = (struct btrfs_chunk *)(ptr + len);
2319 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2320 len += btrfs_chunk_item_size(num_stripes);
2321 } else {
2322 ret = -EIO;
2323 break;
2324 }
2325 if (key.objectid == chunk_objectid &&
2326 key.offset == chunk_offset) {
2327 memmove(ptr, ptr + len, array_size - (cur + len));
2328 array_size -= len;
2329 btrfs_set_super_sys_array_size(super_copy, array_size);
2330 } else {
2331 ptr += len;
2332 cur += len;
2333 }
2334 }
2335 return ret;
2336 }
2337
2338 static int btrfs_relocate_chunk(struct btrfs_root *root,
2339 u64 chunk_tree, u64 chunk_objectid,
2340 u64 chunk_offset)
2341 {
2342 struct extent_map_tree *em_tree;
2343 struct btrfs_root *extent_root;
2344 struct btrfs_trans_handle *trans;
2345 struct extent_map *em;
2346 struct map_lookup *map;
2347 int ret;
2348 int i;
2349
2350 root = root->fs_info->chunk_root;
2351 extent_root = root->fs_info->extent_root;
2352 em_tree = &root->fs_info->mapping_tree.map_tree;
2353
2354 ret = btrfs_can_relocate(extent_root, chunk_offset);
2355 if (ret)
2356 return -ENOSPC;
2357
2358 /* step one, relocate all the extents inside this chunk */
2359 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2360 if (ret)
2361 return ret;
2362
2363 trans = btrfs_start_transaction(root, 0);
2364 BUG_ON(IS_ERR(trans));
2365
2366 lock_chunks(root);
2367
2368 /*
2369 * step two, delete the device extents and the
2370 * chunk tree entries
2371 */
2372 read_lock(&em_tree->lock);
2373 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2374 read_unlock(&em_tree->lock);
2375
2376 BUG_ON(!em || em->start > chunk_offset ||
2377 em->start + em->len < chunk_offset);
2378 map = (struct map_lookup *)em->bdev;
2379
2380 for (i = 0; i < map->num_stripes; i++) {
2381 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2382 map->stripes[i].physical);
2383 BUG_ON(ret);
2384
2385 if (map->stripes[i].dev) {
2386 ret = btrfs_update_device(trans, map->stripes[i].dev);
2387 BUG_ON(ret);
2388 }
2389 }
2390 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2391 chunk_offset);
2392
2393 BUG_ON(ret);
2394
2395 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2396
2397 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2398 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2399 BUG_ON(ret);
2400 }
2401
2402 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2403 BUG_ON(ret);
2404
2405 write_lock(&em_tree->lock);
2406 remove_extent_mapping(em_tree, em);
2407 write_unlock(&em_tree->lock);
2408
2409 kfree(map);
2410 em->bdev = NULL;
2411
2412 /* once for the tree */
2413 free_extent_map(em);
2414 /* once for us */
2415 free_extent_map(em);
2416
2417 unlock_chunks(root);
2418 btrfs_end_transaction(trans, root);
2419 return 0;
2420 }
2421
2422 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2423 {
2424 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2425 struct btrfs_path *path;
2426 struct extent_buffer *leaf;
2427 struct btrfs_chunk *chunk;
2428 struct btrfs_key key;
2429 struct btrfs_key found_key;
2430 u64 chunk_tree = chunk_root->root_key.objectid;
2431 u64 chunk_type;
2432 bool retried = false;
2433 int failed = 0;
2434 int ret;
2435
2436 path = btrfs_alloc_path();
2437 if (!path)
2438 return -ENOMEM;
2439
2440 again:
2441 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2442 key.offset = (u64)-1;
2443 key.type = BTRFS_CHUNK_ITEM_KEY;
2444
2445 while (1) {
2446 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2447 if (ret < 0)
2448 goto error;
2449 BUG_ON(ret == 0); /* Corruption */
2450
2451 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2452 key.type);
2453 if (ret < 0)
2454 goto error;
2455 if (ret > 0)
2456 break;
2457
2458 leaf = path->nodes[0];
2459 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2460
2461 chunk = btrfs_item_ptr(leaf, path->slots[0],
2462 struct btrfs_chunk);
2463 chunk_type = btrfs_chunk_type(leaf, chunk);
2464 btrfs_release_path(path);
2465
2466 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2467 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2468 found_key.objectid,
2469 found_key.offset);
2470 if (ret == -ENOSPC)
2471 failed++;
2472 else if (ret)
2473 BUG();
2474 }
2475
2476 if (found_key.offset == 0)
2477 break;
2478 key.offset = found_key.offset - 1;
2479 }
2480 ret = 0;
2481 if (failed && !retried) {
2482 failed = 0;
2483 retried = true;
2484 goto again;
2485 } else if (failed && retried) {
2486 WARN_ON(1);
2487 ret = -ENOSPC;
2488 }
2489 error:
2490 btrfs_free_path(path);
2491 return ret;
2492 }
2493
2494 static int insert_balance_item(struct btrfs_root *root,
2495 struct btrfs_balance_control *bctl)
2496 {
2497 struct btrfs_trans_handle *trans;
2498 struct btrfs_balance_item *item;
2499 struct btrfs_disk_balance_args disk_bargs;
2500 struct btrfs_path *path;
2501 struct extent_buffer *leaf;
2502 struct btrfs_key key;
2503 int ret, err;
2504
2505 path = btrfs_alloc_path();
2506 if (!path)
2507 return -ENOMEM;
2508
2509 trans = btrfs_start_transaction(root, 0);
2510 if (IS_ERR(trans)) {
2511 btrfs_free_path(path);
2512 return PTR_ERR(trans);
2513 }
2514
2515 key.objectid = BTRFS_BALANCE_OBJECTID;
2516 key.type = BTRFS_BALANCE_ITEM_KEY;
2517 key.offset = 0;
2518
2519 ret = btrfs_insert_empty_item(trans, root, path, &key,
2520 sizeof(*item));
2521 if (ret)
2522 goto out;
2523
2524 leaf = path->nodes[0];
2525 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2526
2527 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2528
2529 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2530 btrfs_set_balance_data(leaf, item, &disk_bargs);
2531 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2532 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2533 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2534 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2535
2536 btrfs_set_balance_flags(leaf, item, bctl->flags);
2537
2538 btrfs_mark_buffer_dirty(leaf);
2539 out:
2540 btrfs_free_path(path);
2541 err = btrfs_commit_transaction(trans, root);
2542 if (err && !ret)
2543 ret = err;
2544 return ret;
2545 }
2546
2547 static int del_balance_item(struct btrfs_root *root)
2548 {
2549 struct btrfs_trans_handle *trans;
2550 struct btrfs_path *path;
2551 struct btrfs_key key;
2552 int ret, err;
2553
2554 path = btrfs_alloc_path();
2555 if (!path)
2556 return -ENOMEM;
2557
2558 trans = btrfs_start_transaction(root, 0);
2559 if (IS_ERR(trans)) {
2560 btrfs_free_path(path);
2561 return PTR_ERR(trans);
2562 }
2563
2564 key.objectid = BTRFS_BALANCE_OBJECTID;
2565 key.type = BTRFS_BALANCE_ITEM_KEY;
2566 key.offset = 0;
2567
2568 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2569 if (ret < 0)
2570 goto out;
2571 if (ret > 0) {
2572 ret = -ENOENT;
2573 goto out;
2574 }
2575
2576 ret = btrfs_del_item(trans, root, path);
2577 out:
2578 btrfs_free_path(path);
2579 err = btrfs_commit_transaction(trans, root);
2580 if (err && !ret)
2581 ret = err;
2582 return ret;
2583 }
2584
2585 /*
2586 * This is a heuristic used to reduce the number of chunks balanced on
2587 * resume after balance was interrupted.
2588 */
2589 static void update_balance_args(struct btrfs_balance_control *bctl)
2590 {
2591 /*
2592 * Turn on soft mode for chunk types that were being converted.
2593 */
2594 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2595 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2596 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2597 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2598 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2599 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2600
2601 /*
2602 * Turn on usage filter if is not already used. The idea is
2603 * that chunks that we have already balanced should be
2604 * reasonably full. Don't do it for chunks that are being
2605 * converted - that will keep us from relocating unconverted
2606 * (albeit full) chunks.
2607 */
2608 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2609 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2610 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2611 bctl->data.usage = 90;
2612 }
2613 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2614 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2615 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2616 bctl->sys.usage = 90;
2617 }
2618 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2619 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2620 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2621 bctl->meta.usage = 90;
2622 }
2623 }
2624
2625 /*
2626 * Should be called with both balance and volume mutexes held to
2627 * serialize other volume operations (add_dev/rm_dev/resize) with
2628 * restriper. Same goes for unset_balance_control.
2629 */
2630 static void set_balance_control(struct btrfs_balance_control *bctl)
2631 {
2632 struct btrfs_fs_info *fs_info = bctl->fs_info;
2633
2634 BUG_ON(fs_info->balance_ctl);
2635
2636 spin_lock(&fs_info->balance_lock);
2637 fs_info->balance_ctl = bctl;
2638 spin_unlock(&fs_info->balance_lock);
2639 }
2640
2641 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2642 {
2643 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2644
2645 BUG_ON(!fs_info->balance_ctl);
2646
2647 spin_lock(&fs_info->balance_lock);
2648 fs_info->balance_ctl = NULL;
2649 spin_unlock(&fs_info->balance_lock);
2650
2651 kfree(bctl);
2652 }
2653
2654 /*
2655 * Balance filters. Return 1 if chunk should be filtered out
2656 * (should not be balanced).
2657 */
2658 static int chunk_profiles_filter(u64 chunk_type,
2659 struct btrfs_balance_args *bargs)
2660 {
2661 chunk_type = chunk_to_extended(chunk_type) &
2662 BTRFS_EXTENDED_PROFILE_MASK;
2663
2664 if (bargs->profiles & chunk_type)
2665 return 0;
2666
2667 return 1;
2668 }
2669
2670 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2671 struct btrfs_balance_args *bargs)
2672 {
2673 struct btrfs_block_group_cache *cache;
2674 u64 chunk_used, user_thresh;
2675 int ret = 1;
2676
2677 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2678 chunk_used = btrfs_block_group_used(&cache->item);
2679
2680 if (bargs->usage == 0)
2681 user_thresh = 1;
2682 else if (bargs->usage > 100)
2683 user_thresh = cache->key.offset;
2684 else
2685 user_thresh = div_factor_fine(cache->key.offset,
2686 bargs->usage);
2687
2688 if (chunk_used < user_thresh)
2689 ret = 0;
2690
2691 btrfs_put_block_group(cache);
2692 return ret;
2693 }
2694
2695 static int chunk_devid_filter(struct extent_buffer *leaf,
2696 struct btrfs_chunk *chunk,
2697 struct btrfs_balance_args *bargs)
2698 {
2699 struct btrfs_stripe *stripe;
2700 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2701 int i;
2702
2703 for (i = 0; i < num_stripes; i++) {
2704 stripe = btrfs_stripe_nr(chunk, i);
2705 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2706 return 0;
2707 }
2708
2709 return 1;
2710 }
2711
2712 /* [pstart, pend) */
2713 static int chunk_drange_filter(struct extent_buffer *leaf,
2714 struct btrfs_chunk *chunk,
2715 u64 chunk_offset,
2716 struct btrfs_balance_args *bargs)
2717 {
2718 struct btrfs_stripe *stripe;
2719 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2720 u64 stripe_offset;
2721 u64 stripe_length;
2722 int factor;
2723 int i;
2724
2725 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2726 return 0;
2727
2728 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2729 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2730 factor = 2;
2731 else
2732 factor = 1;
2733 factor = num_stripes / factor;
2734
2735 for (i = 0; i < num_stripes; i++) {
2736 stripe = btrfs_stripe_nr(chunk, i);
2737 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2738 continue;
2739
2740 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2741 stripe_length = btrfs_chunk_length(leaf, chunk);
2742 do_div(stripe_length, factor);
2743
2744 if (stripe_offset < bargs->pend &&
2745 stripe_offset + stripe_length > bargs->pstart)
2746 return 0;
2747 }
2748
2749 return 1;
2750 }
2751
2752 /* [vstart, vend) */
2753 static int chunk_vrange_filter(struct extent_buffer *leaf,
2754 struct btrfs_chunk *chunk,
2755 u64 chunk_offset,
2756 struct btrfs_balance_args *bargs)
2757 {
2758 if (chunk_offset < bargs->vend &&
2759 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2760 /* at least part of the chunk is inside this vrange */
2761 return 0;
2762
2763 return 1;
2764 }
2765
2766 static int chunk_soft_convert_filter(u64 chunk_type,
2767 struct btrfs_balance_args *bargs)
2768 {
2769 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2770 return 0;
2771
2772 chunk_type = chunk_to_extended(chunk_type) &
2773 BTRFS_EXTENDED_PROFILE_MASK;
2774
2775 if (bargs->target == chunk_type)
2776 return 1;
2777
2778 return 0;
2779 }
2780
2781 static int should_balance_chunk(struct btrfs_root *root,
2782 struct extent_buffer *leaf,
2783 struct btrfs_chunk *chunk, u64 chunk_offset)
2784 {
2785 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2786 struct btrfs_balance_args *bargs = NULL;
2787 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2788
2789 /* type filter */
2790 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2791 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2792 return 0;
2793 }
2794
2795 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2796 bargs = &bctl->data;
2797 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2798 bargs = &bctl->sys;
2799 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2800 bargs = &bctl->meta;
2801
2802 /* profiles filter */
2803 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2804 chunk_profiles_filter(chunk_type, bargs)) {
2805 return 0;
2806 }
2807
2808 /* usage filter */
2809 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2810 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2811 return 0;
2812 }
2813
2814 /* devid filter */
2815 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2816 chunk_devid_filter(leaf, chunk, bargs)) {
2817 return 0;
2818 }
2819
2820 /* drange filter, makes sense only with devid filter */
2821 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2822 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2823 return 0;
2824 }
2825
2826 /* vrange filter */
2827 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2828 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2829 return 0;
2830 }
2831
2832 /* soft profile changing mode */
2833 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2834 chunk_soft_convert_filter(chunk_type, bargs)) {
2835 return 0;
2836 }
2837
2838 return 1;
2839 }
2840
2841 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2842 {
2843 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2844 struct btrfs_root *chunk_root = fs_info->chunk_root;
2845 struct btrfs_root *dev_root = fs_info->dev_root;
2846 struct list_head *devices;
2847 struct btrfs_device *device;
2848 u64 old_size;
2849 u64 size_to_free;
2850 struct btrfs_chunk *chunk;
2851 struct btrfs_path *path;
2852 struct btrfs_key key;
2853 struct btrfs_key found_key;
2854 struct btrfs_trans_handle *trans;
2855 struct extent_buffer *leaf;
2856 int slot;
2857 int ret;
2858 int enospc_errors = 0;
2859 bool counting = true;
2860
2861 /* step one make some room on all the devices */
2862 devices = &fs_info->fs_devices->devices;
2863 list_for_each_entry(device, devices, dev_list) {
2864 old_size = device->total_bytes;
2865 size_to_free = div_factor(old_size, 1);
2866 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2867 if (!device->writeable ||
2868 device->total_bytes - device->bytes_used > size_to_free ||
2869 device->is_tgtdev_for_dev_replace)
2870 continue;
2871
2872 ret = btrfs_shrink_device(device, old_size - size_to_free);
2873 if (ret == -ENOSPC)
2874 break;
2875 BUG_ON(ret);
2876
2877 trans = btrfs_start_transaction(dev_root, 0);
2878 BUG_ON(IS_ERR(trans));
2879
2880 ret = btrfs_grow_device(trans, device, old_size);
2881 BUG_ON(ret);
2882
2883 btrfs_end_transaction(trans, dev_root);
2884 }
2885
2886 /* step two, relocate all the chunks */
2887 path = btrfs_alloc_path();
2888 if (!path) {
2889 ret = -ENOMEM;
2890 goto error;
2891 }
2892
2893 /* zero out stat counters */
2894 spin_lock(&fs_info->balance_lock);
2895 memset(&bctl->stat, 0, sizeof(bctl->stat));
2896 spin_unlock(&fs_info->balance_lock);
2897 again:
2898 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2899 key.offset = (u64)-1;
2900 key.type = BTRFS_CHUNK_ITEM_KEY;
2901
2902 while (1) {
2903 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2904 atomic_read(&fs_info->balance_cancel_req)) {
2905 ret = -ECANCELED;
2906 goto error;
2907 }
2908
2909 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2910 if (ret < 0)
2911 goto error;
2912
2913 /*
2914 * this shouldn't happen, it means the last relocate
2915 * failed
2916 */
2917 if (ret == 0)
2918 BUG(); /* FIXME break ? */
2919
2920 ret = btrfs_previous_item(chunk_root, path, 0,
2921 BTRFS_CHUNK_ITEM_KEY);
2922 if (ret) {
2923 ret = 0;
2924 break;
2925 }
2926
2927 leaf = path->nodes[0];
2928 slot = path->slots[0];
2929 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2930
2931 if (found_key.objectid != key.objectid)
2932 break;
2933
2934 /* chunk zero is special */
2935 if (found_key.offset == 0)
2936 break;
2937
2938 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2939
2940 if (!counting) {
2941 spin_lock(&fs_info->balance_lock);
2942 bctl->stat.considered++;
2943 spin_unlock(&fs_info->balance_lock);
2944 }
2945
2946 ret = should_balance_chunk(chunk_root, leaf, chunk,
2947 found_key.offset);
2948 btrfs_release_path(path);
2949 if (!ret)
2950 goto loop;
2951
2952 if (counting) {
2953 spin_lock(&fs_info->balance_lock);
2954 bctl->stat.expected++;
2955 spin_unlock(&fs_info->balance_lock);
2956 goto loop;
2957 }
2958
2959 ret = btrfs_relocate_chunk(chunk_root,
2960 chunk_root->root_key.objectid,
2961 found_key.objectid,
2962 found_key.offset);
2963 if (ret && ret != -ENOSPC)
2964 goto error;
2965 if (ret == -ENOSPC) {
2966 enospc_errors++;
2967 } else {
2968 spin_lock(&fs_info->balance_lock);
2969 bctl->stat.completed++;
2970 spin_unlock(&fs_info->balance_lock);
2971 }
2972 loop:
2973 key.offset = found_key.offset - 1;
2974 }
2975
2976 if (counting) {
2977 btrfs_release_path(path);
2978 counting = false;
2979 goto again;
2980 }
2981 error:
2982 btrfs_free_path(path);
2983 if (enospc_errors) {
2984 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2985 enospc_errors);
2986 if (!ret)
2987 ret = -ENOSPC;
2988 }
2989
2990 return ret;
2991 }
2992
2993 /**
2994 * alloc_profile_is_valid - see if a given profile is valid and reduced
2995 * @flags: profile to validate
2996 * @extended: if true @flags is treated as an extended profile
2997 */
2998 static int alloc_profile_is_valid(u64 flags, int extended)
2999 {
3000 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3001 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3002
3003 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3004
3005 /* 1) check that all other bits are zeroed */
3006 if (flags & ~mask)
3007 return 0;
3008
3009 /* 2) see if profile is reduced */
3010 if (flags == 0)
3011 return !extended; /* "0" is valid for usual profiles */
3012
3013 /* true if exactly one bit set */
3014 return (flags & (flags - 1)) == 0;
3015 }
3016
3017 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3018 {
3019 /* cancel requested || normal exit path */
3020 return atomic_read(&fs_info->balance_cancel_req) ||
3021 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3022 atomic_read(&fs_info->balance_cancel_req) == 0);
3023 }
3024
3025 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3026 {
3027 int ret;
3028
3029 unset_balance_control(fs_info);
3030 ret = del_balance_item(fs_info->tree_root);
3031 BUG_ON(ret);
3032
3033 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3034 }
3035
3036 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3037 struct btrfs_ioctl_balance_args *bargs);
3038
3039 /*
3040 * Should be called with both balance and volume mutexes held
3041 */
3042 int btrfs_balance(struct btrfs_balance_control *bctl,
3043 struct btrfs_ioctl_balance_args *bargs)
3044 {
3045 struct btrfs_fs_info *fs_info = bctl->fs_info;
3046 u64 allowed;
3047 int mixed = 0;
3048 int ret;
3049 u64 num_devices;
3050 unsigned seq;
3051
3052 if (btrfs_fs_closing(fs_info) ||
3053 atomic_read(&fs_info->balance_pause_req) ||
3054 atomic_read(&fs_info->balance_cancel_req)) {
3055 ret = -EINVAL;
3056 goto out;
3057 }
3058
3059 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3060 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3061 mixed = 1;
3062
3063 /*
3064 * In case of mixed groups both data and meta should be picked,
3065 * and identical options should be given for both of them.
3066 */
3067 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3068 if (mixed && (bctl->flags & allowed)) {
3069 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3070 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3071 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3072 printk(KERN_ERR "btrfs: with mixed groups data and "
3073 "metadata balance options must be the same\n");
3074 ret = -EINVAL;
3075 goto out;
3076 }
3077 }
3078
3079 num_devices = fs_info->fs_devices->num_devices;
3080 btrfs_dev_replace_lock(&fs_info->dev_replace);
3081 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3082 BUG_ON(num_devices < 1);
3083 num_devices--;
3084 }
3085 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3086 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3087 if (num_devices == 1)
3088 allowed |= BTRFS_BLOCK_GROUP_DUP;
3089 else if (num_devices < 4)
3090 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3091 else
3092 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3093 BTRFS_BLOCK_GROUP_RAID10);
3094
3095 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3096 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3097 (bctl->data.target & ~allowed))) {
3098 printk(KERN_ERR "btrfs: unable to start balance with target "
3099 "data profile %llu\n",
3100 (unsigned long long)bctl->data.target);
3101 ret = -EINVAL;
3102 goto out;
3103 }
3104 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3105 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3106 (bctl->meta.target & ~allowed))) {
3107 printk(KERN_ERR "btrfs: unable to start balance with target "
3108 "metadata profile %llu\n",
3109 (unsigned long long)bctl->meta.target);
3110 ret = -EINVAL;
3111 goto out;
3112 }
3113 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3114 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3115 (bctl->sys.target & ~allowed))) {
3116 printk(KERN_ERR "btrfs: unable to start balance with target "
3117 "system profile %llu\n",
3118 (unsigned long long)bctl->sys.target);
3119 ret = -EINVAL;
3120 goto out;
3121 }
3122
3123 /* allow dup'ed data chunks only in mixed mode */
3124 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3125 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3126 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3127 ret = -EINVAL;
3128 goto out;
3129 }
3130
3131 /* allow to reduce meta or sys integrity only if force set */
3132 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3133 BTRFS_BLOCK_GROUP_RAID10;
3134 do {
3135 seq = read_seqbegin(&fs_info->profiles_lock);
3136
3137 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3138 (fs_info->avail_system_alloc_bits & allowed) &&
3139 !(bctl->sys.target & allowed)) ||
3140 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3141 (fs_info->avail_metadata_alloc_bits & allowed) &&
3142 !(bctl->meta.target & allowed))) {
3143 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3144 printk(KERN_INFO "btrfs: force reducing metadata "
3145 "integrity\n");
3146 } else {
3147 printk(KERN_ERR "btrfs: balance will reduce metadata "
3148 "integrity, use force if you want this\n");
3149 ret = -EINVAL;
3150 goto out;
3151 }
3152 }
3153 } while (read_seqretry(&fs_info->profiles_lock, seq));
3154
3155 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3156 int num_tolerated_disk_barrier_failures;
3157 u64 target = bctl->sys.target;
3158
3159 num_tolerated_disk_barrier_failures =
3160 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3161 if (num_tolerated_disk_barrier_failures > 0 &&
3162 (target &
3163 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3164 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3165 num_tolerated_disk_barrier_failures = 0;
3166 else if (num_tolerated_disk_barrier_failures > 1 &&
3167 (target &
3168 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3169 num_tolerated_disk_barrier_failures = 1;
3170
3171 fs_info->num_tolerated_disk_barrier_failures =
3172 num_tolerated_disk_barrier_failures;
3173 }
3174
3175 ret = insert_balance_item(fs_info->tree_root, bctl);
3176 if (ret && ret != -EEXIST)
3177 goto out;
3178
3179 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3180 BUG_ON(ret == -EEXIST);
3181 set_balance_control(bctl);
3182 } else {
3183 BUG_ON(ret != -EEXIST);
3184 spin_lock(&fs_info->balance_lock);
3185 update_balance_args(bctl);
3186 spin_unlock(&fs_info->balance_lock);
3187 }
3188
3189 atomic_inc(&fs_info->balance_running);
3190 mutex_unlock(&fs_info->balance_mutex);
3191
3192 ret = __btrfs_balance(fs_info);
3193
3194 mutex_lock(&fs_info->balance_mutex);
3195 atomic_dec(&fs_info->balance_running);
3196
3197 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3198 fs_info->num_tolerated_disk_barrier_failures =
3199 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3200 }
3201
3202 if (bargs) {
3203 memset(bargs, 0, sizeof(*bargs));
3204 update_ioctl_balance_args(fs_info, 0, bargs);
3205 }
3206
3207 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3208 balance_need_close(fs_info)) {
3209 __cancel_balance(fs_info);
3210 }
3211
3212 wake_up(&fs_info->balance_wait_q);
3213
3214 return ret;
3215 out:
3216 if (bctl->flags & BTRFS_BALANCE_RESUME)
3217 __cancel_balance(fs_info);
3218 else {
3219 kfree(bctl);
3220 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3221 }
3222 return ret;
3223 }
3224
3225 static int balance_kthread(void *data)
3226 {
3227 struct btrfs_fs_info *fs_info = data;
3228 int ret = 0;
3229
3230 mutex_lock(&fs_info->volume_mutex);
3231 mutex_lock(&fs_info->balance_mutex);
3232
3233 if (fs_info->balance_ctl) {
3234 printk(KERN_INFO "btrfs: continuing balance\n");
3235 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3236 }
3237
3238 mutex_unlock(&fs_info->balance_mutex);
3239 mutex_unlock(&fs_info->volume_mutex);
3240
3241 return ret;
3242 }
3243
3244 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3245 {
3246 struct task_struct *tsk;
3247
3248 spin_lock(&fs_info->balance_lock);
3249 if (!fs_info->balance_ctl) {
3250 spin_unlock(&fs_info->balance_lock);
3251 return 0;
3252 }
3253 spin_unlock(&fs_info->balance_lock);
3254
3255 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3256 printk(KERN_INFO "btrfs: force skipping balance\n");
3257 return 0;
3258 }
3259
3260 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3261 if (IS_ERR(tsk))
3262 return PTR_ERR(tsk);
3263
3264 return 0;
3265 }
3266
3267 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3268 {
3269 struct btrfs_balance_control *bctl;
3270 struct btrfs_balance_item *item;
3271 struct btrfs_disk_balance_args disk_bargs;
3272 struct btrfs_path *path;
3273 struct extent_buffer *leaf;
3274 struct btrfs_key key;
3275 int ret;
3276
3277 path = btrfs_alloc_path();
3278 if (!path)
3279 return -ENOMEM;
3280
3281 key.objectid = BTRFS_BALANCE_OBJECTID;
3282 key.type = BTRFS_BALANCE_ITEM_KEY;
3283 key.offset = 0;
3284
3285 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3286 if (ret < 0)
3287 goto out;
3288 if (ret > 0) { /* ret = -ENOENT; */
3289 ret = 0;
3290 goto out;
3291 }
3292
3293 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3294 if (!bctl) {
3295 ret = -ENOMEM;
3296 goto out;
3297 }
3298
3299 leaf = path->nodes[0];
3300 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3301
3302 bctl->fs_info = fs_info;
3303 bctl->flags = btrfs_balance_flags(leaf, item);
3304 bctl->flags |= BTRFS_BALANCE_RESUME;
3305
3306 btrfs_balance_data(leaf, item, &disk_bargs);
3307 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3308 btrfs_balance_meta(leaf, item, &disk_bargs);
3309 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3310 btrfs_balance_sys(leaf, item, &disk_bargs);
3311 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3312
3313 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3314
3315 mutex_lock(&fs_info->volume_mutex);
3316 mutex_lock(&fs_info->balance_mutex);
3317
3318 set_balance_control(bctl);
3319
3320 mutex_unlock(&fs_info->balance_mutex);
3321 mutex_unlock(&fs_info->volume_mutex);
3322 out:
3323 btrfs_free_path(path);
3324 return ret;
3325 }
3326
3327 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3328 {
3329 int ret = 0;
3330
3331 mutex_lock(&fs_info->balance_mutex);
3332 if (!fs_info->balance_ctl) {
3333 mutex_unlock(&fs_info->balance_mutex);
3334 return -ENOTCONN;
3335 }
3336
3337 if (atomic_read(&fs_info->balance_running)) {
3338 atomic_inc(&fs_info->balance_pause_req);
3339 mutex_unlock(&fs_info->balance_mutex);
3340
3341 wait_event(fs_info->balance_wait_q,
3342 atomic_read(&fs_info->balance_running) == 0);
3343
3344 mutex_lock(&fs_info->balance_mutex);
3345 /* we are good with balance_ctl ripped off from under us */
3346 BUG_ON(atomic_read(&fs_info->balance_running));
3347 atomic_dec(&fs_info->balance_pause_req);
3348 } else {
3349 ret = -ENOTCONN;
3350 }
3351
3352 mutex_unlock(&fs_info->balance_mutex);
3353 return ret;
3354 }
3355
3356 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3357 {
3358 mutex_lock(&fs_info->balance_mutex);
3359 if (!fs_info->balance_ctl) {
3360 mutex_unlock(&fs_info->balance_mutex);
3361 return -ENOTCONN;
3362 }
3363
3364 atomic_inc(&fs_info->balance_cancel_req);
3365 /*
3366 * if we are running just wait and return, balance item is
3367 * deleted in btrfs_balance in this case
3368 */
3369 if (atomic_read(&fs_info->balance_running)) {
3370 mutex_unlock(&fs_info->balance_mutex);
3371 wait_event(fs_info->balance_wait_q,
3372 atomic_read(&fs_info->balance_running) == 0);
3373 mutex_lock(&fs_info->balance_mutex);
3374 } else {
3375 /* __cancel_balance needs volume_mutex */
3376 mutex_unlock(&fs_info->balance_mutex);
3377 mutex_lock(&fs_info->volume_mutex);
3378 mutex_lock(&fs_info->balance_mutex);
3379
3380 if (fs_info->balance_ctl)
3381 __cancel_balance(fs_info);
3382
3383 mutex_unlock(&fs_info->volume_mutex);
3384 }
3385
3386 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3387 atomic_dec(&fs_info->balance_cancel_req);
3388 mutex_unlock(&fs_info->balance_mutex);
3389 return 0;
3390 }
3391
3392 /*
3393 * shrinking a device means finding all of the device extents past
3394 * the new size, and then following the back refs to the chunks.
3395 * The chunk relocation code actually frees the device extent
3396 */
3397 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3398 {
3399 struct btrfs_trans_handle *trans;
3400 struct btrfs_root *root = device->dev_root;
3401 struct btrfs_dev_extent *dev_extent = NULL;
3402 struct btrfs_path *path;
3403 u64 length;
3404 u64 chunk_tree;
3405 u64 chunk_objectid;
3406 u64 chunk_offset;
3407 int ret;
3408 int slot;
3409 int failed = 0;
3410 bool retried = false;
3411 struct extent_buffer *l;
3412 struct btrfs_key key;
3413 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3414 u64 old_total = btrfs_super_total_bytes(super_copy);
3415 u64 old_size = device->total_bytes;
3416 u64 diff = device->total_bytes - new_size;
3417
3418 if (device->is_tgtdev_for_dev_replace)
3419 return -EINVAL;
3420
3421 path = btrfs_alloc_path();
3422 if (!path)
3423 return -ENOMEM;
3424
3425 path->reada = 2;
3426
3427 lock_chunks(root);
3428
3429 device->total_bytes = new_size;
3430 if (device->writeable) {
3431 device->fs_devices->total_rw_bytes -= diff;
3432 spin_lock(&root->fs_info->free_chunk_lock);
3433 root->fs_info->free_chunk_space -= diff;
3434 spin_unlock(&root->fs_info->free_chunk_lock);
3435 }
3436 unlock_chunks(root);
3437
3438 again:
3439 key.objectid = device->devid;
3440 key.offset = (u64)-1;
3441 key.type = BTRFS_DEV_EXTENT_KEY;
3442
3443 do {
3444 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3445 if (ret < 0)
3446 goto done;
3447
3448 ret = btrfs_previous_item(root, path, 0, key.type);
3449 if (ret < 0)
3450 goto done;
3451 if (ret) {
3452 ret = 0;
3453 btrfs_release_path(path);
3454 break;
3455 }
3456
3457 l = path->nodes[0];
3458 slot = path->slots[0];
3459 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3460
3461 if (key.objectid != device->devid) {
3462 btrfs_release_path(path);
3463 break;
3464 }
3465
3466 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3467 length = btrfs_dev_extent_length(l, dev_extent);
3468
3469 if (key.offset + length <= new_size) {
3470 btrfs_release_path(path);
3471 break;
3472 }
3473
3474 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3475 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3476 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3477 btrfs_release_path(path);
3478
3479 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3480 chunk_offset);
3481 if (ret && ret != -ENOSPC)
3482 goto done;
3483 if (ret == -ENOSPC)
3484 failed++;
3485 } while (key.offset-- > 0);
3486
3487 if (failed && !retried) {
3488 failed = 0;
3489 retried = true;
3490 goto again;
3491 } else if (failed && retried) {
3492 ret = -ENOSPC;
3493 lock_chunks(root);
3494
3495 device->total_bytes = old_size;
3496 if (device->writeable)
3497 device->fs_devices->total_rw_bytes += diff;
3498 spin_lock(&root->fs_info->free_chunk_lock);
3499 root->fs_info->free_chunk_space += diff;
3500 spin_unlock(&root->fs_info->free_chunk_lock);
3501 unlock_chunks(root);
3502 goto done;
3503 }
3504
3505 /* Shrinking succeeded, else we would be at "done". */
3506 trans = btrfs_start_transaction(root, 0);
3507 if (IS_ERR(trans)) {
3508 ret = PTR_ERR(trans);
3509 goto done;
3510 }
3511
3512 lock_chunks(root);
3513
3514 device->disk_total_bytes = new_size;
3515 /* Now btrfs_update_device() will change the on-disk size. */
3516 ret = btrfs_update_device(trans, device);
3517 if (ret) {
3518 unlock_chunks(root);
3519 btrfs_end_transaction(trans, root);
3520 goto done;
3521 }
3522 WARN_ON(diff > old_total);
3523 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3524 unlock_chunks(root);
3525 btrfs_end_transaction(trans, root);
3526 done:
3527 btrfs_free_path(path);
3528 return ret;
3529 }
3530
3531 static int btrfs_add_system_chunk(struct btrfs_root *root,
3532 struct btrfs_key *key,
3533 struct btrfs_chunk *chunk, int item_size)
3534 {
3535 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3536 struct btrfs_disk_key disk_key;
3537 u32 array_size;
3538 u8 *ptr;
3539
3540 array_size = btrfs_super_sys_array_size(super_copy);
3541 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3542 return -EFBIG;
3543
3544 ptr = super_copy->sys_chunk_array + array_size;
3545 btrfs_cpu_key_to_disk(&disk_key, key);
3546 memcpy(ptr, &disk_key, sizeof(disk_key));
3547 ptr += sizeof(disk_key);
3548 memcpy(ptr, chunk, item_size);
3549 item_size += sizeof(disk_key);
3550 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3551 return 0;
3552 }
3553
3554 /*
3555 * sort the devices in descending order by max_avail, total_avail
3556 */
3557 static int btrfs_cmp_device_info(const void *a, const void *b)
3558 {
3559 const struct btrfs_device_info *di_a = a;
3560 const struct btrfs_device_info *di_b = b;
3561
3562 if (di_a->max_avail > di_b->max_avail)
3563 return -1;
3564 if (di_a->max_avail < di_b->max_avail)
3565 return 1;
3566 if (di_a->total_avail > di_b->total_avail)
3567 return -1;
3568 if (di_a->total_avail < di_b->total_avail)
3569 return 1;
3570 return 0;
3571 }
3572
3573 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3574 [BTRFS_RAID_RAID10] = {
3575 .sub_stripes = 2,
3576 .dev_stripes = 1,
3577 .devs_max = 0, /* 0 == as many as possible */
3578 .devs_min = 4,
3579 .devs_increment = 2,
3580 .ncopies = 2,
3581 },
3582 [BTRFS_RAID_RAID1] = {
3583 .sub_stripes = 1,
3584 .dev_stripes = 1,
3585 .devs_max = 2,
3586 .devs_min = 2,
3587 .devs_increment = 2,
3588 .ncopies = 2,
3589 },
3590 [BTRFS_RAID_DUP] = {
3591 .sub_stripes = 1,
3592 .dev_stripes = 2,
3593 .devs_max = 1,
3594 .devs_min = 1,
3595 .devs_increment = 1,
3596 .ncopies = 2,
3597 },
3598 [BTRFS_RAID_RAID0] = {
3599 .sub_stripes = 1,
3600 .dev_stripes = 1,
3601 .devs_max = 0,
3602 .devs_min = 2,
3603 .devs_increment = 1,
3604 .ncopies = 1,
3605 },
3606 [BTRFS_RAID_SINGLE] = {
3607 .sub_stripes = 1,
3608 .dev_stripes = 1,
3609 .devs_max = 1,
3610 .devs_min = 1,
3611 .devs_increment = 1,
3612 .ncopies = 1,
3613 },
3614 };
3615
3616 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *extent_root,
3618 struct map_lookup **map_ret,
3619 u64 *num_bytes_out, u64 *stripe_size_out,
3620 u64 start, u64 type)
3621 {
3622 struct btrfs_fs_info *info = extent_root->fs_info;
3623 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3624 struct list_head *cur;
3625 struct map_lookup *map = NULL;
3626 struct extent_map_tree *em_tree;
3627 struct extent_map *em;
3628 struct btrfs_device_info *devices_info = NULL;
3629 u64 total_avail;
3630 int num_stripes; /* total number of stripes to allocate */
3631 int sub_stripes; /* sub_stripes info for map */
3632 int dev_stripes; /* stripes per dev */
3633 int devs_max; /* max devs to use */
3634 int devs_min; /* min devs needed */
3635 int devs_increment; /* ndevs has to be a multiple of this */
3636 int ncopies; /* how many copies to data has */
3637 int ret;
3638 u64 max_stripe_size;
3639 u64 max_chunk_size;
3640 u64 stripe_size;
3641 u64 num_bytes;
3642 int ndevs;
3643 int i;
3644 int j;
3645 int index;
3646
3647 BUG_ON(!alloc_profile_is_valid(type, 0));
3648
3649 if (list_empty(&fs_devices->alloc_list))
3650 return -ENOSPC;
3651
3652 index = __get_raid_index(type);
3653
3654 sub_stripes = btrfs_raid_array[index].sub_stripes;
3655 dev_stripes = btrfs_raid_array[index].dev_stripes;
3656 devs_max = btrfs_raid_array[index].devs_max;
3657 devs_min = btrfs_raid_array[index].devs_min;
3658 devs_increment = btrfs_raid_array[index].devs_increment;
3659 ncopies = btrfs_raid_array[index].ncopies;
3660
3661 if (type & BTRFS_BLOCK_GROUP_DATA) {
3662 max_stripe_size = 1024 * 1024 * 1024;
3663 max_chunk_size = 10 * max_stripe_size;
3664 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3665 /* for larger filesystems, use larger metadata chunks */
3666 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3667 max_stripe_size = 1024 * 1024 * 1024;
3668 else
3669 max_stripe_size = 256 * 1024 * 1024;
3670 max_chunk_size = max_stripe_size;
3671 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3672 max_stripe_size = 32 * 1024 * 1024;
3673 max_chunk_size = 2 * max_stripe_size;
3674 } else {
3675 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3676 type);
3677 BUG_ON(1);
3678 }
3679
3680 /* we don't want a chunk larger than 10% of writeable space */
3681 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3682 max_chunk_size);
3683
3684 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3685 GFP_NOFS);
3686 if (!devices_info)
3687 return -ENOMEM;
3688
3689 cur = fs_devices->alloc_list.next;
3690
3691 /*
3692 * in the first pass through the devices list, we gather information
3693 * about the available holes on each device.
3694 */
3695 ndevs = 0;
3696 while (cur != &fs_devices->alloc_list) {
3697 struct btrfs_device *device;
3698 u64 max_avail;
3699 u64 dev_offset;
3700
3701 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3702
3703 cur = cur->next;
3704
3705 if (!device->writeable) {
3706 WARN(1, KERN_ERR
3707 "btrfs: read-only device in alloc_list\n");
3708 continue;
3709 }
3710
3711 if (!device->in_fs_metadata ||
3712 device->is_tgtdev_for_dev_replace)
3713 continue;
3714
3715 if (device->total_bytes > device->bytes_used)
3716 total_avail = device->total_bytes - device->bytes_used;
3717 else
3718 total_avail = 0;
3719
3720 /* If there is no space on this device, skip it. */
3721 if (total_avail == 0)
3722 continue;
3723
3724 ret = find_free_dev_extent(device,
3725 max_stripe_size * dev_stripes,
3726 &dev_offset, &max_avail);
3727 if (ret && ret != -ENOSPC)
3728 goto error;
3729
3730 if (ret == 0)
3731 max_avail = max_stripe_size * dev_stripes;
3732
3733 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3734 continue;
3735
3736 if (ndevs == fs_devices->rw_devices) {
3737 WARN(1, "%s: found more than %llu devices\n",
3738 __func__, fs_devices->rw_devices);
3739 break;
3740 }
3741 devices_info[ndevs].dev_offset = dev_offset;
3742 devices_info[ndevs].max_avail = max_avail;
3743 devices_info[ndevs].total_avail = total_avail;
3744 devices_info[ndevs].dev = device;
3745 ++ndevs;
3746 }
3747
3748 /*
3749 * now sort the devices by hole size / available space
3750 */
3751 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3752 btrfs_cmp_device_info, NULL);
3753
3754 /* round down to number of usable stripes */
3755 ndevs -= ndevs % devs_increment;
3756
3757 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3758 ret = -ENOSPC;
3759 goto error;
3760 }
3761
3762 if (devs_max && ndevs > devs_max)
3763 ndevs = devs_max;
3764 /*
3765 * the primary goal is to maximize the number of stripes, so use as many
3766 * devices as possible, even if the stripes are not maximum sized.
3767 */
3768 stripe_size = devices_info[ndevs-1].max_avail;
3769 num_stripes = ndevs * dev_stripes;
3770
3771 if (stripe_size * ndevs > max_chunk_size * ncopies) {
3772 stripe_size = max_chunk_size * ncopies;
3773 do_div(stripe_size, ndevs);
3774 }
3775
3776 do_div(stripe_size, dev_stripes);
3777
3778 /* align to BTRFS_STRIPE_LEN */
3779 do_div(stripe_size, BTRFS_STRIPE_LEN);
3780 stripe_size *= BTRFS_STRIPE_LEN;
3781
3782 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3783 if (!map) {
3784 ret = -ENOMEM;
3785 goto error;
3786 }
3787 map->num_stripes = num_stripes;
3788
3789 for (i = 0; i < ndevs; ++i) {
3790 for (j = 0; j < dev_stripes; ++j) {
3791 int s = i * dev_stripes + j;
3792 map->stripes[s].dev = devices_info[i].dev;
3793 map->stripes[s].physical = devices_info[i].dev_offset +
3794 j * stripe_size;
3795 }
3796 }
3797 map->sector_size = extent_root->sectorsize;
3798 map->stripe_len = BTRFS_STRIPE_LEN;
3799 map->io_align = BTRFS_STRIPE_LEN;
3800 map->io_width = BTRFS_STRIPE_LEN;
3801 map->type = type;
3802 map->sub_stripes = sub_stripes;
3803
3804 *map_ret = map;
3805 num_bytes = stripe_size * (num_stripes / ncopies);
3806
3807 *stripe_size_out = stripe_size;
3808 *num_bytes_out = num_bytes;
3809
3810 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3811
3812 em = alloc_extent_map();
3813 if (!em) {
3814 ret = -ENOMEM;
3815 goto error;
3816 }
3817 em->bdev = (struct block_device *)map;
3818 em->start = start;
3819 em->len = num_bytes;
3820 em->block_start = 0;
3821 em->block_len = em->len;
3822
3823 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3824 write_lock(&em_tree->lock);
3825 ret = add_extent_mapping(em_tree, em);
3826 write_unlock(&em_tree->lock);
3827 if (ret) {
3828 free_extent_map(em);
3829 goto error;
3830 }
3831
3832 for (i = 0; i < map->num_stripes; ++i) {
3833 struct btrfs_device *device;
3834 u64 dev_offset;
3835
3836 device = map->stripes[i].dev;
3837 dev_offset = map->stripes[i].physical;
3838
3839 ret = btrfs_alloc_dev_extent(trans, device,
3840 info->chunk_root->root_key.objectid,
3841 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3842 start, dev_offset, stripe_size);
3843 if (ret)
3844 goto error_dev_extent;
3845 }
3846
3847 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3848 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3849 start, num_bytes);
3850 if (ret) {
3851 i = map->num_stripes - 1;
3852 goto error_dev_extent;
3853 }
3854
3855 free_extent_map(em);
3856 kfree(devices_info);
3857 return 0;
3858
3859 error_dev_extent:
3860 for (; i >= 0; i--) {
3861 struct btrfs_device *device;
3862 int err;
3863
3864 device = map->stripes[i].dev;
3865 err = btrfs_free_dev_extent(trans, device, start);
3866 if (err) {
3867 btrfs_abort_transaction(trans, extent_root, err);
3868 break;
3869 }
3870 }
3871 write_lock(&em_tree->lock);
3872 remove_extent_mapping(em_tree, em);
3873 write_unlock(&em_tree->lock);
3874
3875 /* One for our allocation */
3876 free_extent_map(em);
3877 /* One for the tree reference */
3878 free_extent_map(em);
3879 error:
3880 kfree(map);
3881 kfree(devices_info);
3882 return ret;
3883 }
3884
3885 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3886 struct btrfs_root *extent_root,
3887 struct map_lookup *map, u64 chunk_offset,
3888 u64 chunk_size, u64 stripe_size)
3889 {
3890 u64 dev_offset;
3891 struct btrfs_key key;
3892 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3893 struct btrfs_device *device;
3894 struct btrfs_chunk *chunk;
3895 struct btrfs_stripe *stripe;
3896 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3897 int index = 0;
3898 int ret;
3899
3900 chunk = kzalloc(item_size, GFP_NOFS);
3901 if (!chunk)
3902 return -ENOMEM;
3903
3904 index = 0;
3905 while (index < map->num_stripes) {
3906 device = map->stripes[index].dev;
3907 device->bytes_used += stripe_size;
3908 ret = btrfs_update_device(trans, device);
3909 if (ret)
3910 goto out_free;
3911 index++;
3912 }
3913
3914 spin_lock(&extent_root->fs_info->free_chunk_lock);
3915 extent_root->fs_info->free_chunk_space -= (stripe_size *
3916 map->num_stripes);
3917 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3918
3919 index = 0;
3920 stripe = &chunk->stripe;
3921 while (index < map->num_stripes) {
3922 device = map->stripes[index].dev;
3923 dev_offset = map->stripes[index].physical;
3924
3925 btrfs_set_stack_stripe_devid(stripe, device->devid);
3926 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3927 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3928 stripe++;
3929 index++;
3930 }
3931
3932 btrfs_set_stack_chunk_length(chunk, chunk_size);
3933 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3934 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3935 btrfs_set_stack_chunk_type(chunk, map->type);
3936 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3937 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3938 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3939 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3940 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3941
3942 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3943 key.type = BTRFS_CHUNK_ITEM_KEY;
3944 key.offset = chunk_offset;
3945
3946 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3947
3948 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3949 /*
3950 * TODO: Cleanup of inserted chunk root in case of
3951 * failure.
3952 */
3953 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3954 item_size);
3955 }
3956
3957 out_free:
3958 kfree(chunk);
3959 return ret;
3960 }
3961
3962 /*
3963 * Chunk allocation falls into two parts. The first part does works
3964 * that make the new allocated chunk useable, but not do any operation
3965 * that modifies the chunk tree. The second part does the works that
3966 * require modifying the chunk tree. This division is important for the
3967 * bootstrap process of adding storage to a seed btrfs.
3968 */
3969 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3970 struct btrfs_root *extent_root, u64 type)
3971 {
3972 u64 chunk_offset;
3973 u64 chunk_size;
3974 u64 stripe_size;
3975 struct map_lookup *map;
3976 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3977 int ret;
3978
3979 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3980 &chunk_offset);
3981 if (ret)
3982 return ret;
3983
3984 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3985 &stripe_size, chunk_offset, type);
3986 if (ret)
3987 return ret;
3988
3989 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3990 chunk_size, stripe_size);
3991 if (ret)
3992 return ret;
3993 return 0;
3994 }
3995
3996 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3997 struct btrfs_root *root,
3998 struct btrfs_device *device)
3999 {
4000 u64 chunk_offset;
4001 u64 sys_chunk_offset;
4002 u64 chunk_size;
4003 u64 sys_chunk_size;
4004 u64 stripe_size;
4005 u64 sys_stripe_size;
4006 u64 alloc_profile;
4007 struct map_lookup *map;
4008 struct map_lookup *sys_map;
4009 struct btrfs_fs_info *fs_info = root->fs_info;
4010 struct btrfs_root *extent_root = fs_info->extent_root;
4011 int ret;
4012
4013 ret = find_next_chunk(fs_info->chunk_root,
4014 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4015 if (ret)
4016 return ret;
4017
4018 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4019 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4020 &stripe_size, chunk_offset, alloc_profile);
4021 if (ret)
4022 return ret;
4023
4024 sys_chunk_offset = chunk_offset + chunk_size;
4025
4026 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4027 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4028 &sys_chunk_size, &sys_stripe_size,
4029 sys_chunk_offset, alloc_profile);
4030 if (ret) {
4031 btrfs_abort_transaction(trans, root, ret);
4032 goto out;
4033 }
4034
4035 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4036 if (ret) {
4037 btrfs_abort_transaction(trans, root, ret);
4038 goto out;
4039 }
4040
4041 /*
4042 * Modifying chunk tree needs allocating new blocks from both
4043 * system block group and metadata block group. So we only can
4044 * do operations require modifying the chunk tree after both
4045 * block groups were created.
4046 */
4047 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4048 chunk_size, stripe_size);
4049 if (ret) {
4050 btrfs_abort_transaction(trans, root, ret);
4051 goto out;
4052 }
4053
4054 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4055 sys_chunk_offset, sys_chunk_size,
4056 sys_stripe_size);
4057 if (ret)
4058 btrfs_abort_transaction(trans, root, ret);
4059
4060 out:
4061
4062 return ret;
4063 }
4064
4065 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4066 {
4067 struct extent_map *em;
4068 struct map_lookup *map;
4069 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4070 int readonly = 0;
4071 int i;
4072
4073 read_lock(&map_tree->map_tree.lock);
4074 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4075 read_unlock(&map_tree->map_tree.lock);
4076 if (!em)
4077 return 1;
4078
4079 if (btrfs_test_opt(root, DEGRADED)) {
4080 free_extent_map(em);
4081 return 0;
4082 }
4083
4084 map = (struct map_lookup *)em->bdev;
4085 for (i = 0; i < map->num_stripes; i++) {
4086 if (!map->stripes[i].dev->writeable) {
4087 readonly = 1;
4088 break;
4089 }
4090 }
4091 free_extent_map(em);
4092 return readonly;
4093 }
4094
4095 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4096 {
4097 extent_map_tree_init(&tree->map_tree);
4098 }
4099
4100 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4101 {
4102 struct extent_map *em;
4103
4104 while (1) {
4105 write_lock(&tree->map_tree.lock);
4106 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4107 if (em)
4108 remove_extent_mapping(&tree->map_tree, em);
4109 write_unlock(&tree->map_tree.lock);
4110 if (!em)
4111 break;
4112 kfree(em->bdev);
4113 /* once for us */
4114 free_extent_map(em);
4115 /* once for the tree */
4116 free_extent_map(em);
4117 }
4118 }
4119
4120 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4121 {
4122 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4123 struct extent_map *em;
4124 struct map_lookup *map;
4125 struct extent_map_tree *em_tree = &map_tree->map_tree;
4126 int ret;
4127
4128 read_lock(&em_tree->lock);
4129 em = lookup_extent_mapping(em_tree, logical, len);
4130 read_unlock(&em_tree->lock);
4131 BUG_ON(!em);
4132
4133 BUG_ON(em->start > logical || em->start + em->len < logical);
4134 map = (struct map_lookup *)em->bdev;
4135 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4136 ret = map->num_stripes;
4137 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4138 ret = map->sub_stripes;
4139 else
4140 ret = 1;
4141 free_extent_map(em);
4142
4143 btrfs_dev_replace_lock(&fs_info->dev_replace);
4144 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4145 ret++;
4146 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4147
4148 return ret;
4149 }
4150
4151 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4152 struct map_lookup *map, int first, int num,
4153 int optimal, int dev_replace_is_ongoing)
4154 {
4155 int i;
4156 int tolerance;
4157 struct btrfs_device *srcdev;
4158
4159 if (dev_replace_is_ongoing &&
4160 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4161 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4162 srcdev = fs_info->dev_replace.srcdev;
4163 else
4164 srcdev = NULL;
4165
4166 /*
4167 * try to avoid the drive that is the source drive for a
4168 * dev-replace procedure, only choose it if no other non-missing
4169 * mirror is available
4170 */
4171 for (tolerance = 0; tolerance < 2; tolerance++) {
4172 if (map->stripes[optimal].dev->bdev &&
4173 (tolerance || map->stripes[optimal].dev != srcdev))
4174 return optimal;
4175 for (i = first; i < first + num; i++) {
4176 if (map->stripes[i].dev->bdev &&
4177 (tolerance || map->stripes[i].dev != srcdev))
4178 return i;
4179 }
4180 }
4181
4182 /* we couldn't find one that doesn't fail. Just return something
4183 * and the io error handling code will clean up eventually
4184 */
4185 return optimal;
4186 }
4187
4188 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4189 u64 logical, u64 *length,
4190 struct btrfs_bio **bbio_ret,
4191 int mirror_num)
4192 {
4193 struct extent_map *em;
4194 struct map_lookup *map;
4195 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4196 struct extent_map_tree *em_tree = &map_tree->map_tree;
4197 u64 offset;
4198 u64 stripe_offset;
4199 u64 stripe_end_offset;
4200 u64 stripe_nr;
4201 u64 stripe_nr_orig;
4202 u64 stripe_nr_end;
4203 int stripe_index;
4204 int i;
4205 int ret = 0;
4206 int num_stripes;
4207 int max_errors = 0;
4208 struct btrfs_bio *bbio = NULL;
4209 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4210 int dev_replace_is_ongoing = 0;
4211 int num_alloc_stripes;
4212 int patch_the_first_stripe_for_dev_replace = 0;
4213 u64 physical_to_patch_in_first_stripe = 0;
4214
4215 read_lock(&em_tree->lock);
4216 em = lookup_extent_mapping(em_tree, logical, *length);
4217 read_unlock(&em_tree->lock);
4218
4219 if (!em) {
4220 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4221 (unsigned long long)logical,
4222 (unsigned long long)*length);
4223 BUG();
4224 }
4225
4226 BUG_ON(em->start > logical || em->start + em->len < logical);
4227 map = (struct map_lookup *)em->bdev;
4228 offset = logical - em->start;
4229
4230 stripe_nr = offset;
4231 /*
4232 * stripe_nr counts the total number of stripes we have to stride
4233 * to get to this block
4234 */
4235 do_div(stripe_nr, map->stripe_len);
4236
4237 stripe_offset = stripe_nr * map->stripe_len;
4238 BUG_ON(offset < stripe_offset);
4239
4240 /* stripe_offset is the offset of this block in its stripe*/
4241 stripe_offset = offset - stripe_offset;
4242
4243 if (rw & REQ_DISCARD)
4244 *length = min_t(u64, em->len - offset, *length);
4245 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4246 /* we limit the length of each bio to what fits in a stripe */
4247 *length = min_t(u64, em->len - offset,
4248 map->stripe_len - stripe_offset);
4249 } else {
4250 *length = em->len - offset;
4251 }
4252
4253 if (!bbio_ret)
4254 goto out;
4255
4256 btrfs_dev_replace_lock(dev_replace);
4257 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4258 if (!dev_replace_is_ongoing)
4259 btrfs_dev_replace_unlock(dev_replace);
4260
4261 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4262 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4263 dev_replace->tgtdev != NULL) {
4264 /*
4265 * in dev-replace case, for repair case (that's the only
4266 * case where the mirror is selected explicitly when
4267 * calling btrfs_map_block), blocks left of the left cursor
4268 * can also be read from the target drive.
4269 * For REQ_GET_READ_MIRRORS, the target drive is added as
4270 * the last one to the array of stripes. For READ, it also
4271 * needs to be supported using the same mirror number.
4272 * If the requested block is not left of the left cursor,
4273 * EIO is returned. This can happen because btrfs_num_copies()
4274 * returns one more in the dev-replace case.
4275 */
4276 u64 tmp_length = *length;
4277 struct btrfs_bio *tmp_bbio = NULL;
4278 int tmp_num_stripes;
4279 u64 srcdev_devid = dev_replace->srcdev->devid;
4280 int index_srcdev = 0;
4281 int found = 0;
4282 u64 physical_of_found = 0;
4283
4284 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4285 logical, &tmp_length, &tmp_bbio, 0);
4286 if (ret) {
4287 WARN_ON(tmp_bbio != NULL);
4288 goto out;
4289 }
4290
4291 tmp_num_stripes = tmp_bbio->num_stripes;
4292 if (mirror_num > tmp_num_stripes) {
4293 /*
4294 * REQ_GET_READ_MIRRORS does not contain this
4295 * mirror, that means that the requested area
4296 * is not left of the left cursor
4297 */
4298 ret = -EIO;
4299 kfree(tmp_bbio);
4300 goto out;
4301 }
4302
4303 /*
4304 * process the rest of the function using the mirror_num
4305 * of the source drive. Therefore look it up first.
4306 * At the end, patch the device pointer to the one of the
4307 * target drive.
4308 */
4309 for (i = 0; i < tmp_num_stripes; i++) {
4310 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4311 /*
4312 * In case of DUP, in order to keep it
4313 * simple, only add the mirror with the
4314 * lowest physical address
4315 */
4316 if (found &&
4317 physical_of_found <=
4318 tmp_bbio->stripes[i].physical)
4319 continue;
4320 index_srcdev = i;
4321 found = 1;
4322 physical_of_found =
4323 tmp_bbio->stripes[i].physical;
4324 }
4325 }
4326
4327 if (found) {
4328 mirror_num = index_srcdev + 1;
4329 patch_the_first_stripe_for_dev_replace = 1;
4330 physical_to_patch_in_first_stripe = physical_of_found;
4331 } else {
4332 WARN_ON(1);
4333 ret = -EIO;
4334 kfree(tmp_bbio);
4335 goto out;
4336 }
4337
4338 kfree(tmp_bbio);
4339 } else if (mirror_num > map->num_stripes) {
4340 mirror_num = 0;
4341 }
4342
4343 num_stripes = 1;
4344 stripe_index = 0;
4345 stripe_nr_orig = stripe_nr;
4346 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4347 (~(map->stripe_len - 1));
4348 do_div(stripe_nr_end, map->stripe_len);
4349 stripe_end_offset = stripe_nr_end * map->stripe_len -
4350 (offset + *length);
4351 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4352 if (rw & REQ_DISCARD)
4353 num_stripes = min_t(u64, map->num_stripes,
4354 stripe_nr_end - stripe_nr_orig);
4355 stripe_index = do_div(stripe_nr, map->num_stripes);
4356 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4357 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4358 num_stripes = map->num_stripes;
4359 else if (mirror_num)
4360 stripe_index = mirror_num - 1;
4361 else {
4362 stripe_index = find_live_mirror(fs_info, map, 0,
4363 map->num_stripes,
4364 current->pid % map->num_stripes,
4365 dev_replace_is_ongoing);
4366 mirror_num = stripe_index + 1;
4367 }
4368
4369 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4370 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4371 num_stripes = map->num_stripes;
4372 } else if (mirror_num) {
4373 stripe_index = mirror_num - 1;
4374 } else {
4375 mirror_num = 1;
4376 }
4377
4378 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4379 int factor = map->num_stripes / map->sub_stripes;
4380
4381 stripe_index = do_div(stripe_nr, factor);
4382 stripe_index *= map->sub_stripes;
4383
4384 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4385 num_stripes = map->sub_stripes;
4386 else if (rw & REQ_DISCARD)
4387 num_stripes = min_t(u64, map->sub_stripes *
4388 (stripe_nr_end - stripe_nr_orig),
4389 map->num_stripes);
4390 else if (mirror_num)
4391 stripe_index += mirror_num - 1;
4392 else {
4393 int old_stripe_index = stripe_index;
4394 stripe_index = find_live_mirror(fs_info, map,
4395 stripe_index,
4396 map->sub_stripes, stripe_index +
4397 current->pid % map->sub_stripes,
4398 dev_replace_is_ongoing);
4399 mirror_num = stripe_index - old_stripe_index + 1;
4400 }
4401 } else {
4402 /*
4403 * after this do_div call, stripe_nr is the number of stripes
4404 * on this device we have to walk to find the data, and
4405 * stripe_index is the number of our device in the stripe array
4406 */
4407 stripe_index = do_div(stripe_nr, map->num_stripes);
4408 mirror_num = stripe_index + 1;
4409 }
4410 BUG_ON(stripe_index >= map->num_stripes);
4411
4412 num_alloc_stripes = num_stripes;
4413 if (dev_replace_is_ongoing) {
4414 if (rw & (REQ_WRITE | REQ_DISCARD))
4415 num_alloc_stripes <<= 1;
4416 if (rw & REQ_GET_READ_MIRRORS)
4417 num_alloc_stripes++;
4418 }
4419 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4420 if (!bbio) {
4421 ret = -ENOMEM;
4422 goto out;
4423 }
4424 atomic_set(&bbio->error, 0);
4425
4426 if (rw & REQ_DISCARD) {
4427 int factor = 0;
4428 int sub_stripes = 0;
4429 u64 stripes_per_dev = 0;
4430 u32 remaining_stripes = 0;
4431 u32 last_stripe = 0;
4432
4433 if (map->type &
4434 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4435 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4436 sub_stripes = 1;
4437 else
4438 sub_stripes = map->sub_stripes;
4439
4440 factor = map->num_stripes / sub_stripes;
4441 stripes_per_dev = div_u64_rem(stripe_nr_end -
4442 stripe_nr_orig,
4443 factor,
4444 &remaining_stripes);
4445 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4446 last_stripe *= sub_stripes;
4447 }
4448
4449 for (i = 0; i < num_stripes; i++) {
4450 bbio->stripes[i].physical =
4451 map->stripes[stripe_index].physical +
4452 stripe_offset + stripe_nr * map->stripe_len;
4453 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4454
4455 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4456 BTRFS_BLOCK_GROUP_RAID10)) {
4457 bbio->stripes[i].length = stripes_per_dev *
4458 map->stripe_len;
4459
4460 if (i / sub_stripes < remaining_stripes)
4461 bbio->stripes[i].length +=
4462 map->stripe_len;
4463
4464 /*
4465 * Special for the first stripe and
4466 * the last stripe:
4467 *
4468 * |-------|...|-------|
4469 * |----------|
4470 * off end_off
4471 */
4472 if (i < sub_stripes)
4473 bbio->stripes[i].length -=
4474 stripe_offset;
4475
4476 if (stripe_index >= last_stripe &&
4477 stripe_index <= (last_stripe +
4478 sub_stripes - 1))
4479 bbio->stripes[i].length -=
4480 stripe_end_offset;
4481
4482 if (i == sub_stripes - 1)
4483 stripe_offset = 0;
4484 } else
4485 bbio->stripes[i].length = *length;
4486
4487 stripe_index++;
4488 if (stripe_index == map->num_stripes) {
4489 /* This could only happen for RAID0/10 */
4490 stripe_index = 0;
4491 stripe_nr++;
4492 }
4493 }
4494 } else {
4495 for (i = 0; i < num_stripes; i++) {
4496 bbio->stripes[i].physical =
4497 map->stripes[stripe_index].physical +
4498 stripe_offset +
4499 stripe_nr * map->stripe_len;
4500 bbio->stripes[i].dev =
4501 map->stripes[stripe_index].dev;
4502 stripe_index++;
4503 }
4504 }
4505
4506 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4507 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4508 BTRFS_BLOCK_GROUP_RAID10 |
4509 BTRFS_BLOCK_GROUP_DUP)) {
4510 max_errors = 1;
4511 }
4512 }
4513
4514 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4515 dev_replace->tgtdev != NULL) {
4516 int index_where_to_add;
4517 u64 srcdev_devid = dev_replace->srcdev->devid;
4518
4519 /*
4520 * duplicate the write operations while the dev replace
4521 * procedure is running. Since the copying of the old disk
4522 * to the new disk takes place at run time while the
4523 * filesystem is mounted writable, the regular write
4524 * operations to the old disk have to be duplicated to go
4525 * to the new disk as well.
4526 * Note that device->missing is handled by the caller, and
4527 * that the write to the old disk is already set up in the
4528 * stripes array.
4529 */
4530 index_where_to_add = num_stripes;
4531 for (i = 0; i < num_stripes; i++) {
4532 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4533 /* write to new disk, too */
4534 struct btrfs_bio_stripe *new =
4535 bbio->stripes + index_where_to_add;
4536 struct btrfs_bio_stripe *old =
4537 bbio->stripes + i;
4538
4539 new->physical = old->physical;
4540 new->length = old->length;
4541 new->dev = dev_replace->tgtdev;
4542 index_where_to_add++;
4543 max_errors++;
4544 }
4545 }
4546 num_stripes = index_where_to_add;
4547 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4548 dev_replace->tgtdev != NULL) {
4549 u64 srcdev_devid = dev_replace->srcdev->devid;
4550 int index_srcdev = 0;
4551 int found = 0;
4552 u64 physical_of_found = 0;
4553
4554 /*
4555 * During the dev-replace procedure, the target drive can
4556 * also be used to read data in case it is needed to repair
4557 * a corrupt block elsewhere. This is possible if the
4558 * requested area is left of the left cursor. In this area,
4559 * the target drive is a full copy of the source drive.
4560 */
4561 for (i = 0; i < num_stripes; i++) {
4562 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4563 /*
4564 * In case of DUP, in order to keep it
4565 * simple, only add the mirror with the
4566 * lowest physical address
4567 */
4568 if (found &&
4569 physical_of_found <=
4570 bbio->stripes[i].physical)
4571 continue;
4572 index_srcdev = i;
4573 found = 1;
4574 physical_of_found = bbio->stripes[i].physical;
4575 }
4576 }
4577 if (found) {
4578 u64 length = map->stripe_len;
4579
4580 if (physical_of_found + length <=
4581 dev_replace->cursor_left) {
4582 struct btrfs_bio_stripe *tgtdev_stripe =
4583 bbio->stripes + num_stripes;
4584
4585 tgtdev_stripe->physical = physical_of_found;
4586 tgtdev_stripe->length =
4587 bbio->stripes[index_srcdev].length;
4588 tgtdev_stripe->dev = dev_replace->tgtdev;
4589
4590 num_stripes++;
4591 }
4592 }
4593 }
4594
4595 *bbio_ret = bbio;
4596 bbio->num_stripes = num_stripes;
4597 bbio->max_errors = max_errors;
4598 bbio->mirror_num = mirror_num;
4599
4600 /*
4601 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4602 * mirror_num == num_stripes + 1 && dev_replace target drive is
4603 * available as a mirror
4604 */
4605 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4606 WARN_ON(num_stripes > 1);
4607 bbio->stripes[0].dev = dev_replace->tgtdev;
4608 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4609 bbio->mirror_num = map->num_stripes + 1;
4610 }
4611 out:
4612 if (dev_replace_is_ongoing)
4613 btrfs_dev_replace_unlock(dev_replace);
4614 free_extent_map(em);
4615 return ret;
4616 }
4617
4618 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4619 u64 logical, u64 *length,
4620 struct btrfs_bio **bbio_ret, int mirror_num)
4621 {
4622 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4623 mirror_num);
4624 }
4625
4626 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4627 u64 chunk_start, u64 physical, u64 devid,
4628 u64 **logical, int *naddrs, int *stripe_len)
4629 {
4630 struct extent_map_tree *em_tree = &map_tree->map_tree;
4631 struct extent_map *em;
4632 struct map_lookup *map;
4633 u64 *buf;
4634 u64 bytenr;
4635 u64 length;
4636 u64 stripe_nr;
4637 int i, j, nr = 0;
4638
4639 read_lock(&em_tree->lock);
4640 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4641 read_unlock(&em_tree->lock);
4642
4643 BUG_ON(!em || em->start != chunk_start);
4644 map = (struct map_lookup *)em->bdev;
4645
4646 length = em->len;
4647 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4648 do_div(length, map->num_stripes / map->sub_stripes);
4649 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4650 do_div(length, map->num_stripes);
4651
4652 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4653 BUG_ON(!buf); /* -ENOMEM */
4654
4655 for (i = 0; i < map->num_stripes; i++) {
4656 if (devid && map->stripes[i].dev->devid != devid)
4657 continue;
4658 if (map->stripes[i].physical > physical ||
4659 map->stripes[i].physical + length <= physical)
4660 continue;
4661
4662 stripe_nr = physical - map->stripes[i].physical;
4663 do_div(stripe_nr, map->stripe_len);
4664
4665 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4666 stripe_nr = stripe_nr * map->num_stripes + i;
4667 do_div(stripe_nr, map->sub_stripes);
4668 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4669 stripe_nr = stripe_nr * map->num_stripes + i;
4670 }
4671 bytenr = chunk_start + stripe_nr * map->stripe_len;
4672 WARN_ON(nr >= map->num_stripes);
4673 for (j = 0; j < nr; j++) {
4674 if (buf[j] == bytenr)
4675 break;
4676 }
4677 if (j == nr) {
4678 WARN_ON(nr >= map->num_stripes);
4679 buf[nr++] = bytenr;
4680 }
4681 }
4682
4683 *logical = buf;
4684 *naddrs = nr;
4685 *stripe_len = map->stripe_len;
4686
4687 free_extent_map(em);
4688 return 0;
4689 }
4690
4691 static void *merge_stripe_index_into_bio_private(void *bi_private,
4692 unsigned int stripe_index)
4693 {
4694 /*
4695 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4696 * at most 1.
4697 * The alternative solution (instead of stealing bits from the
4698 * pointer) would be to allocate an intermediate structure
4699 * that contains the old private pointer plus the stripe_index.
4700 */
4701 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4702 BUG_ON(stripe_index > 3);
4703 return (void *)(((uintptr_t)bi_private) | stripe_index);
4704 }
4705
4706 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4707 {
4708 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4709 }
4710
4711 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4712 {
4713 return (unsigned int)((uintptr_t)bi_private) & 3;
4714 }
4715
4716 static void btrfs_end_bio(struct bio *bio, int err)
4717 {
4718 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4719 int is_orig_bio = 0;
4720
4721 if (err) {
4722 atomic_inc(&bbio->error);
4723 if (err == -EIO || err == -EREMOTEIO) {
4724 unsigned int stripe_index =
4725 extract_stripe_index_from_bio_private(
4726 bio->bi_private);
4727 struct btrfs_device *dev;
4728
4729 BUG_ON(stripe_index >= bbio->num_stripes);
4730 dev = bbio->stripes[stripe_index].dev;
4731 if (dev->bdev) {
4732 if (bio->bi_rw & WRITE)
4733 btrfs_dev_stat_inc(dev,
4734 BTRFS_DEV_STAT_WRITE_ERRS);
4735 else
4736 btrfs_dev_stat_inc(dev,
4737 BTRFS_DEV_STAT_READ_ERRS);
4738 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4739 btrfs_dev_stat_inc(dev,
4740 BTRFS_DEV_STAT_FLUSH_ERRS);
4741 btrfs_dev_stat_print_on_error(dev);
4742 }
4743 }
4744 }
4745
4746 if (bio == bbio->orig_bio)
4747 is_orig_bio = 1;
4748
4749 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4750 if (!is_orig_bio) {
4751 bio_put(bio);
4752 bio = bbio->orig_bio;
4753 }
4754 bio->bi_private = bbio->private;
4755 bio->bi_end_io = bbio->end_io;
4756 bio->bi_bdev = (struct block_device *)
4757 (unsigned long)bbio->mirror_num;
4758 /* only send an error to the higher layers if it is
4759 * beyond the tolerance of the multi-bio
4760 */
4761 if (atomic_read(&bbio->error) > bbio->max_errors) {
4762 err = -EIO;
4763 } else {
4764 /*
4765 * this bio is actually up to date, we didn't
4766 * go over the max number of errors
4767 */
4768 set_bit(BIO_UPTODATE, &bio->bi_flags);
4769 err = 0;
4770 }
4771 kfree(bbio);
4772
4773 bio_endio(bio, err);
4774 } else if (!is_orig_bio) {
4775 bio_put(bio);
4776 }
4777 }
4778
4779 struct async_sched {
4780 struct bio *bio;
4781 int rw;
4782 struct btrfs_fs_info *info;
4783 struct btrfs_work work;
4784 };
4785
4786 /*
4787 * see run_scheduled_bios for a description of why bios are collected for
4788 * async submit.
4789 *
4790 * This will add one bio to the pending list for a device and make sure
4791 * the work struct is scheduled.
4792 */
4793 static noinline void schedule_bio(struct btrfs_root *root,
4794 struct btrfs_device *device,
4795 int rw, struct bio *bio)
4796 {
4797 int should_queue = 1;
4798 struct btrfs_pending_bios *pending_bios;
4799
4800 /* don't bother with additional async steps for reads, right now */
4801 if (!(rw & REQ_WRITE)) {
4802 bio_get(bio);
4803 btrfsic_submit_bio(rw, bio);
4804 bio_put(bio);
4805 return;
4806 }
4807
4808 /*
4809 * nr_async_bios allows us to reliably return congestion to the
4810 * higher layers. Otherwise, the async bio makes it appear we have
4811 * made progress against dirty pages when we've really just put it
4812 * on a queue for later
4813 */
4814 atomic_inc(&root->fs_info->nr_async_bios);
4815 WARN_ON(bio->bi_next);
4816 bio->bi_next = NULL;
4817 bio->bi_rw |= rw;
4818
4819 spin_lock(&device->io_lock);
4820 if (bio->bi_rw & REQ_SYNC)
4821 pending_bios = &device->pending_sync_bios;
4822 else
4823 pending_bios = &device->pending_bios;
4824
4825 if (pending_bios->tail)
4826 pending_bios->tail->bi_next = bio;
4827
4828 pending_bios->tail = bio;
4829 if (!pending_bios->head)
4830 pending_bios->head = bio;
4831 if (device->running_pending)
4832 should_queue = 0;
4833
4834 spin_unlock(&device->io_lock);
4835
4836 if (should_queue)
4837 btrfs_queue_worker(&root->fs_info->submit_workers,
4838 &device->work);
4839 }
4840
4841 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4842 sector_t sector)
4843 {
4844 struct bio_vec *prev;
4845 struct request_queue *q = bdev_get_queue(bdev);
4846 unsigned short max_sectors = queue_max_sectors(q);
4847 struct bvec_merge_data bvm = {
4848 .bi_bdev = bdev,
4849 .bi_sector = sector,
4850 .bi_rw = bio->bi_rw,
4851 };
4852
4853 if (bio->bi_vcnt == 0) {
4854 WARN_ON(1);
4855 return 1;
4856 }
4857
4858 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4859 if ((bio->bi_size >> 9) > max_sectors)
4860 return 0;
4861
4862 if (!q->merge_bvec_fn)
4863 return 1;
4864
4865 bvm.bi_size = bio->bi_size - prev->bv_len;
4866 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4867 return 0;
4868 return 1;
4869 }
4870
4871 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4872 struct bio *bio, u64 physical, int dev_nr,
4873 int rw, int async)
4874 {
4875 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4876
4877 bio->bi_private = bbio;
4878 bio->bi_private = merge_stripe_index_into_bio_private(
4879 bio->bi_private, (unsigned int)dev_nr);
4880 bio->bi_end_io = btrfs_end_bio;
4881 bio->bi_sector = physical >> 9;
4882 #ifdef DEBUG
4883 {
4884 struct rcu_string *name;
4885
4886 rcu_read_lock();
4887 name = rcu_dereference(dev->name);
4888 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4889 "(%s id %llu), size=%u\n", rw,
4890 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4891 name->str, dev->devid, bio->bi_size);
4892 rcu_read_unlock();
4893 }
4894 #endif
4895 bio->bi_bdev = dev->bdev;
4896 if (async)
4897 schedule_bio(root, dev, rw, bio);
4898 else
4899 btrfsic_submit_bio(rw, bio);
4900 }
4901
4902 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4903 struct bio *first_bio, struct btrfs_device *dev,
4904 int dev_nr, int rw, int async)
4905 {
4906 struct bio_vec *bvec = first_bio->bi_io_vec;
4907 struct bio *bio;
4908 int nr_vecs = bio_get_nr_vecs(dev->bdev);
4909 u64 physical = bbio->stripes[dev_nr].physical;
4910
4911 again:
4912 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4913 if (!bio)
4914 return -ENOMEM;
4915
4916 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4917 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4918 bvec->bv_offset) < bvec->bv_len) {
4919 u64 len = bio->bi_size;
4920
4921 atomic_inc(&bbio->stripes_pending);
4922 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4923 rw, async);
4924 physical += len;
4925 goto again;
4926 }
4927 bvec++;
4928 }
4929
4930 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4931 return 0;
4932 }
4933
4934 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4935 {
4936 atomic_inc(&bbio->error);
4937 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4938 bio->bi_private = bbio->private;
4939 bio->bi_end_io = bbio->end_io;
4940 bio->bi_bdev = (struct block_device *)
4941 (unsigned long)bbio->mirror_num;
4942 bio->bi_sector = logical >> 9;
4943 kfree(bbio);
4944 bio_endio(bio, -EIO);
4945 }
4946 }
4947
4948 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4949 int mirror_num, int async_submit)
4950 {
4951 struct btrfs_device *dev;
4952 struct bio *first_bio = bio;
4953 u64 logical = (u64)bio->bi_sector << 9;
4954 u64 length = 0;
4955 u64 map_length;
4956 int ret;
4957 int dev_nr = 0;
4958 int total_devs = 1;
4959 struct btrfs_bio *bbio = NULL;
4960
4961 length = bio->bi_size;
4962 map_length = length;
4963
4964 ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4965 mirror_num);
4966 if (ret)
4967 return ret;
4968
4969 total_devs = bbio->num_stripes;
4970 if (map_length < length) {
4971 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4972 "len %llu\n", (unsigned long long)logical,
4973 (unsigned long long)length,
4974 (unsigned long long)map_length);
4975 BUG();
4976 }
4977
4978 bbio->orig_bio = first_bio;
4979 bbio->private = first_bio->bi_private;
4980 bbio->end_io = first_bio->bi_end_io;
4981 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4982
4983 while (dev_nr < total_devs) {
4984 dev = bbio->stripes[dev_nr].dev;
4985 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4986 bbio_error(bbio, first_bio, logical);
4987 dev_nr++;
4988 continue;
4989 }
4990
4991 /*
4992 * Check and see if we're ok with this bio based on it's size
4993 * and offset with the given device.
4994 */
4995 if (!bio_size_ok(dev->bdev, first_bio,
4996 bbio->stripes[dev_nr].physical >> 9)) {
4997 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4998 dev_nr, rw, async_submit);
4999 BUG_ON(ret);
5000 dev_nr++;
5001 continue;
5002 }
5003
5004 if (dev_nr < total_devs - 1) {
5005 bio = bio_clone(first_bio, GFP_NOFS);
5006 BUG_ON(!bio); /* -ENOMEM */
5007 } else {
5008 bio = first_bio;
5009 }
5010
5011 submit_stripe_bio(root, bbio, bio,
5012 bbio->stripes[dev_nr].physical, dev_nr, rw,
5013 async_submit);
5014 dev_nr++;
5015 }
5016 return 0;
5017 }
5018
5019 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5020 u8 *uuid, u8 *fsid)
5021 {
5022 struct btrfs_device *device;
5023 struct btrfs_fs_devices *cur_devices;
5024
5025 cur_devices = fs_info->fs_devices;
5026 while (cur_devices) {
5027 if (!fsid ||
5028 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5029 device = __find_device(&cur_devices->devices,
5030 devid, uuid);
5031 if (device)
5032 return device;
5033 }
5034 cur_devices = cur_devices->seed;
5035 }
5036 return NULL;
5037 }
5038
5039 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5040 u64 devid, u8 *dev_uuid)
5041 {
5042 struct btrfs_device *device;
5043 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5044
5045 device = kzalloc(sizeof(*device), GFP_NOFS);
5046 if (!device)
5047 return NULL;
5048 list_add(&device->dev_list,
5049 &fs_devices->devices);
5050 device->dev_root = root->fs_info->dev_root;
5051 device->devid = devid;
5052 device->work.func = pending_bios_fn;
5053 device->fs_devices = fs_devices;
5054 device->missing = 1;
5055 fs_devices->num_devices++;
5056 fs_devices->missing_devices++;
5057 spin_lock_init(&device->io_lock);
5058 INIT_LIST_HEAD(&device->dev_alloc_list);
5059 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5060 return device;
5061 }
5062
5063 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5064 struct extent_buffer *leaf,
5065 struct btrfs_chunk *chunk)
5066 {
5067 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5068 struct map_lookup *map;
5069 struct extent_map *em;
5070 u64 logical;
5071 u64 length;
5072 u64 devid;
5073 u8 uuid[BTRFS_UUID_SIZE];
5074 int num_stripes;
5075 int ret;
5076 int i;
5077
5078 logical = key->offset;
5079 length = btrfs_chunk_length(leaf, chunk);
5080
5081 read_lock(&map_tree->map_tree.lock);
5082 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5083 read_unlock(&map_tree->map_tree.lock);
5084
5085 /* already mapped? */
5086 if (em && em->start <= logical && em->start + em->len > logical) {
5087 free_extent_map(em);
5088 return 0;
5089 } else if (em) {
5090 free_extent_map(em);
5091 }
5092
5093 em = alloc_extent_map();
5094 if (!em)
5095 return -ENOMEM;
5096 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5097 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5098 if (!map) {
5099 free_extent_map(em);
5100 return -ENOMEM;
5101 }
5102
5103 em->bdev = (struct block_device *)map;
5104 em->start = logical;
5105 em->len = length;
5106 em->orig_start = 0;
5107 em->block_start = 0;
5108 em->block_len = em->len;
5109
5110 map->num_stripes = num_stripes;
5111 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5112 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5113 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5114 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5115 map->type = btrfs_chunk_type(leaf, chunk);
5116 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5117 for (i = 0; i < num_stripes; i++) {
5118 map->stripes[i].physical =
5119 btrfs_stripe_offset_nr(leaf, chunk, i);
5120 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5121 read_extent_buffer(leaf, uuid, (unsigned long)
5122 btrfs_stripe_dev_uuid_nr(chunk, i),
5123 BTRFS_UUID_SIZE);
5124 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5125 uuid, NULL);
5126 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5127 kfree(map);
5128 free_extent_map(em);
5129 return -EIO;
5130 }
5131 if (!map->stripes[i].dev) {
5132 map->stripes[i].dev =
5133 add_missing_dev(root, devid, uuid);
5134 if (!map->stripes[i].dev) {
5135 kfree(map);
5136 free_extent_map(em);
5137 return -EIO;
5138 }
5139 }
5140 map->stripes[i].dev->in_fs_metadata = 1;
5141 }
5142
5143 write_lock(&map_tree->map_tree.lock);
5144 ret = add_extent_mapping(&map_tree->map_tree, em);
5145 write_unlock(&map_tree->map_tree.lock);
5146 BUG_ON(ret); /* Tree corruption */
5147 free_extent_map(em);
5148
5149 return 0;
5150 }
5151
5152 static void fill_device_from_item(struct extent_buffer *leaf,
5153 struct btrfs_dev_item *dev_item,
5154 struct btrfs_device *device)
5155 {
5156 unsigned long ptr;
5157
5158 device->devid = btrfs_device_id(leaf, dev_item);
5159 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5160 device->total_bytes = device->disk_total_bytes;
5161 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5162 device->type = btrfs_device_type(leaf, dev_item);
5163 device->io_align = btrfs_device_io_align(leaf, dev_item);
5164 device->io_width = btrfs_device_io_width(leaf, dev_item);
5165 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5166 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5167 device->is_tgtdev_for_dev_replace = 0;
5168
5169 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5170 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5171 }
5172
5173 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5174 {
5175 struct btrfs_fs_devices *fs_devices;
5176 int ret;
5177
5178 BUG_ON(!mutex_is_locked(&uuid_mutex));
5179
5180 fs_devices = root->fs_info->fs_devices->seed;
5181 while (fs_devices) {
5182 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5183 ret = 0;
5184 goto out;
5185 }
5186 fs_devices = fs_devices->seed;
5187 }
5188
5189 fs_devices = find_fsid(fsid);
5190 if (!fs_devices) {
5191 ret = -ENOENT;
5192 goto out;
5193 }
5194
5195 fs_devices = clone_fs_devices(fs_devices);
5196 if (IS_ERR(fs_devices)) {
5197 ret = PTR_ERR(fs_devices);
5198 goto out;
5199 }
5200
5201 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5202 root->fs_info->bdev_holder);
5203 if (ret) {
5204 free_fs_devices(fs_devices);
5205 goto out;
5206 }
5207
5208 if (!fs_devices->seeding) {
5209 __btrfs_close_devices(fs_devices);
5210 free_fs_devices(fs_devices);
5211 ret = -EINVAL;
5212 goto out;
5213 }
5214
5215 fs_devices->seed = root->fs_info->fs_devices->seed;
5216 root->fs_info->fs_devices->seed = fs_devices;
5217 out:
5218 return ret;
5219 }
5220
5221 static int read_one_dev(struct btrfs_root *root,
5222 struct extent_buffer *leaf,
5223 struct btrfs_dev_item *dev_item)
5224 {
5225 struct btrfs_device *device;
5226 u64 devid;
5227 int ret;
5228 u8 fs_uuid[BTRFS_UUID_SIZE];
5229 u8 dev_uuid[BTRFS_UUID_SIZE];
5230
5231 devid = btrfs_device_id(leaf, dev_item);
5232 read_extent_buffer(leaf, dev_uuid,
5233 (unsigned long)btrfs_device_uuid(dev_item),
5234 BTRFS_UUID_SIZE);
5235 read_extent_buffer(leaf, fs_uuid,
5236 (unsigned long)btrfs_device_fsid(dev_item),
5237 BTRFS_UUID_SIZE);
5238
5239 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5240 ret = open_seed_devices(root, fs_uuid);
5241 if (ret && !btrfs_test_opt(root, DEGRADED))
5242 return ret;
5243 }
5244
5245 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5246 if (!device || !device->bdev) {
5247 if (!btrfs_test_opt(root, DEGRADED))
5248 return -EIO;
5249
5250 if (!device) {
5251 printk(KERN_WARNING "warning devid %llu missing\n",
5252 (unsigned long long)devid);
5253 device = add_missing_dev(root, devid, dev_uuid);
5254 if (!device)
5255 return -ENOMEM;
5256 } else if (!device->missing) {
5257 /*
5258 * this happens when a device that was properly setup
5259 * in the device info lists suddenly goes bad.
5260 * device->bdev is NULL, and so we have to set
5261 * device->missing to one here
5262 */
5263 root->fs_info->fs_devices->missing_devices++;
5264 device->missing = 1;
5265 }
5266 }
5267
5268 if (device->fs_devices != root->fs_info->fs_devices) {
5269 BUG_ON(device->writeable);
5270 if (device->generation !=
5271 btrfs_device_generation(leaf, dev_item))
5272 return -EINVAL;
5273 }
5274
5275 fill_device_from_item(leaf, dev_item, device);
5276 device->dev_root = root->fs_info->dev_root;
5277 device->in_fs_metadata = 1;
5278 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5279 device->fs_devices->total_rw_bytes += device->total_bytes;
5280 spin_lock(&root->fs_info->free_chunk_lock);
5281 root->fs_info->free_chunk_space += device->total_bytes -
5282 device->bytes_used;
5283 spin_unlock(&root->fs_info->free_chunk_lock);
5284 }
5285 ret = 0;
5286 return ret;
5287 }
5288
5289 int btrfs_read_sys_array(struct btrfs_root *root)
5290 {
5291 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5292 struct extent_buffer *sb;
5293 struct btrfs_disk_key *disk_key;
5294 struct btrfs_chunk *chunk;
5295 u8 *ptr;
5296 unsigned long sb_ptr;
5297 int ret = 0;
5298 u32 num_stripes;
5299 u32 array_size;
5300 u32 len = 0;
5301 u32 cur;
5302 struct btrfs_key key;
5303
5304 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5305 BTRFS_SUPER_INFO_SIZE);
5306 if (!sb)
5307 return -ENOMEM;
5308 btrfs_set_buffer_uptodate(sb);
5309 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5310 /*
5311 * The sb extent buffer is artifical and just used to read the system array.
5312 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5313 * pages up-to-date when the page is larger: extent does not cover the
5314 * whole page and consequently check_page_uptodate does not find all
5315 * the page's extents up-to-date (the hole beyond sb),
5316 * write_extent_buffer then triggers a WARN_ON.
5317 *
5318 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5319 * but sb spans only this function. Add an explicit SetPageUptodate call
5320 * to silence the warning eg. on PowerPC 64.
5321 */
5322 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5323 SetPageUptodate(sb->pages[0]);
5324
5325 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5326 array_size = btrfs_super_sys_array_size(super_copy);
5327
5328 ptr = super_copy->sys_chunk_array;
5329 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5330 cur = 0;
5331
5332 while (cur < array_size) {
5333 disk_key = (struct btrfs_disk_key *)ptr;
5334 btrfs_disk_key_to_cpu(&key, disk_key);
5335
5336 len = sizeof(*disk_key); ptr += len;
5337 sb_ptr += len;
5338 cur += len;
5339
5340 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5341 chunk = (struct btrfs_chunk *)sb_ptr;
5342 ret = read_one_chunk(root, &key, sb, chunk);
5343 if (ret)
5344 break;
5345 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5346 len = btrfs_chunk_item_size(num_stripes);
5347 } else {
5348 ret = -EIO;
5349 break;
5350 }
5351 ptr += len;
5352 sb_ptr += len;
5353 cur += len;
5354 }
5355 free_extent_buffer(sb);
5356 return ret;
5357 }
5358
5359 int btrfs_read_chunk_tree(struct btrfs_root *root)
5360 {
5361 struct btrfs_path *path;
5362 struct extent_buffer *leaf;
5363 struct btrfs_key key;
5364 struct btrfs_key found_key;
5365 int ret;
5366 int slot;
5367
5368 root = root->fs_info->chunk_root;
5369
5370 path = btrfs_alloc_path();
5371 if (!path)
5372 return -ENOMEM;
5373
5374 mutex_lock(&uuid_mutex);
5375 lock_chunks(root);
5376
5377 /* first we search for all of the device items, and then we
5378 * read in all of the chunk items. This way we can create chunk
5379 * mappings that reference all of the devices that are afound
5380 */
5381 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5382 key.offset = 0;
5383 key.type = 0;
5384 again:
5385 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5386 if (ret < 0)
5387 goto error;
5388 while (1) {
5389 leaf = path->nodes[0];
5390 slot = path->slots[0];
5391 if (slot >= btrfs_header_nritems(leaf)) {
5392 ret = btrfs_next_leaf(root, path);
5393 if (ret == 0)
5394 continue;
5395 if (ret < 0)
5396 goto error;
5397 break;
5398 }
5399 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5400 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5401 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5402 break;
5403 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5404 struct btrfs_dev_item *dev_item;
5405 dev_item = btrfs_item_ptr(leaf, slot,
5406 struct btrfs_dev_item);
5407 ret = read_one_dev(root, leaf, dev_item);
5408 if (ret)
5409 goto error;
5410 }
5411 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5412 struct btrfs_chunk *chunk;
5413 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5414 ret = read_one_chunk(root, &found_key, leaf, chunk);
5415 if (ret)
5416 goto error;
5417 }
5418 path->slots[0]++;
5419 }
5420 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5421 key.objectid = 0;
5422 btrfs_release_path(path);
5423 goto again;
5424 }
5425 ret = 0;
5426 error:
5427 unlock_chunks(root);
5428 mutex_unlock(&uuid_mutex);
5429
5430 btrfs_free_path(path);
5431 return ret;
5432 }
5433
5434 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5435 {
5436 int i;
5437
5438 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5439 btrfs_dev_stat_reset(dev, i);
5440 }
5441
5442 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5443 {
5444 struct btrfs_key key;
5445 struct btrfs_key found_key;
5446 struct btrfs_root *dev_root = fs_info->dev_root;
5447 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5448 struct extent_buffer *eb;
5449 int slot;
5450 int ret = 0;
5451 struct btrfs_device *device;
5452 struct btrfs_path *path = NULL;
5453 int i;
5454
5455 path = btrfs_alloc_path();
5456 if (!path) {
5457 ret = -ENOMEM;
5458 goto out;
5459 }
5460
5461 mutex_lock(&fs_devices->device_list_mutex);
5462 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5463 int item_size;
5464 struct btrfs_dev_stats_item *ptr;
5465
5466 key.objectid = 0;
5467 key.type = BTRFS_DEV_STATS_KEY;
5468 key.offset = device->devid;
5469 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5470 if (ret) {
5471 __btrfs_reset_dev_stats(device);
5472 device->dev_stats_valid = 1;
5473 btrfs_release_path(path);
5474 continue;
5475 }
5476 slot = path->slots[0];
5477 eb = path->nodes[0];
5478 btrfs_item_key_to_cpu(eb, &found_key, slot);
5479 item_size = btrfs_item_size_nr(eb, slot);
5480
5481 ptr = btrfs_item_ptr(eb, slot,
5482 struct btrfs_dev_stats_item);
5483
5484 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5485 if (item_size >= (1 + i) * sizeof(__le64))
5486 btrfs_dev_stat_set(device, i,
5487 btrfs_dev_stats_value(eb, ptr, i));
5488 else
5489 btrfs_dev_stat_reset(device, i);
5490 }
5491
5492 device->dev_stats_valid = 1;
5493 btrfs_dev_stat_print_on_load(device);
5494 btrfs_release_path(path);
5495 }
5496 mutex_unlock(&fs_devices->device_list_mutex);
5497
5498 out:
5499 btrfs_free_path(path);
5500 return ret < 0 ? ret : 0;
5501 }
5502
5503 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5504 struct btrfs_root *dev_root,
5505 struct btrfs_device *device)
5506 {
5507 struct btrfs_path *path;
5508 struct btrfs_key key;
5509 struct extent_buffer *eb;
5510 struct btrfs_dev_stats_item *ptr;
5511 int ret;
5512 int i;
5513
5514 key.objectid = 0;
5515 key.type = BTRFS_DEV_STATS_KEY;
5516 key.offset = device->devid;
5517
5518 path = btrfs_alloc_path();
5519 BUG_ON(!path);
5520 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5521 if (ret < 0) {
5522 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5523 ret, rcu_str_deref(device->name));
5524 goto out;
5525 }
5526
5527 if (ret == 0 &&
5528 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5529 /* need to delete old one and insert a new one */
5530 ret = btrfs_del_item(trans, dev_root, path);
5531 if (ret != 0) {
5532 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5533 rcu_str_deref(device->name), ret);
5534 goto out;
5535 }
5536 ret = 1;
5537 }
5538
5539 if (ret == 1) {
5540 /* need to insert a new item */
5541 btrfs_release_path(path);
5542 ret = btrfs_insert_empty_item(trans, dev_root, path,
5543 &key, sizeof(*ptr));
5544 if (ret < 0) {
5545 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5546 rcu_str_deref(device->name), ret);
5547 goto out;
5548 }
5549 }
5550
5551 eb = path->nodes[0];
5552 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5553 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5554 btrfs_set_dev_stats_value(eb, ptr, i,
5555 btrfs_dev_stat_read(device, i));
5556 btrfs_mark_buffer_dirty(eb);
5557
5558 out:
5559 btrfs_free_path(path);
5560 return ret;
5561 }
5562
5563 /*
5564 * called from commit_transaction. Writes all changed device stats to disk.
5565 */
5566 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5567 struct btrfs_fs_info *fs_info)
5568 {
5569 struct btrfs_root *dev_root = fs_info->dev_root;
5570 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5571 struct btrfs_device *device;
5572 int ret = 0;
5573
5574 mutex_lock(&fs_devices->device_list_mutex);
5575 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5576 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5577 continue;
5578
5579 ret = update_dev_stat_item(trans, dev_root, device);
5580 if (!ret)
5581 device->dev_stats_dirty = 0;
5582 }
5583 mutex_unlock(&fs_devices->device_list_mutex);
5584
5585 return ret;
5586 }
5587
5588 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5589 {
5590 btrfs_dev_stat_inc(dev, index);
5591 btrfs_dev_stat_print_on_error(dev);
5592 }
5593
5594 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5595 {
5596 if (!dev->dev_stats_valid)
5597 return;
5598 printk_ratelimited_in_rcu(KERN_ERR
5599 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5600 rcu_str_deref(dev->name),
5601 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5602 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5603 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5604 btrfs_dev_stat_read(dev,
5605 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5606 btrfs_dev_stat_read(dev,
5607 BTRFS_DEV_STAT_GENERATION_ERRS));
5608 }
5609
5610 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5611 {
5612 int i;
5613
5614 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5615 if (btrfs_dev_stat_read(dev, i) != 0)
5616 break;
5617 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5618 return; /* all values == 0, suppress message */
5619
5620 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5621 rcu_str_deref(dev->name),
5622 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5623 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5624 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5625 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5626 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5627 }
5628
5629 int btrfs_get_dev_stats(struct btrfs_root *root,
5630 struct btrfs_ioctl_get_dev_stats *stats)
5631 {
5632 struct btrfs_device *dev;
5633 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5634 int i;
5635
5636 mutex_lock(&fs_devices->device_list_mutex);
5637 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5638 mutex_unlock(&fs_devices->device_list_mutex);
5639
5640 if (!dev) {
5641 printk(KERN_WARNING
5642 "btrfs: get dev_stats failed, device not found\n");
5643 return -ENODEV;
5644 } else if (!dev->dev_stats_valid) {
5645 printk(KERN_WARNING
5646 "btrfs: get dev_stats failed, not yet valid\n");
5647 return -ENODEV;
5648 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5649 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5650 if (stats->nr_items > i)
5651 stats->values[i] =
5652 btrfs_dev_stat_read_and_reset(dev, i);
5653 else
5654 btrfs_dev_stat_reset(dev, i);
5655 }
5656 } else {
5657 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5658 if (stats->nr_items > i)
5659 stats->values[i] = btrfs_dev_stat_read(dev, i);
5660 }
5661 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5662 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5663 return 0;
5664 }
5665
5666 int btrfs_scratch_superblock(struct btrfs_device *device)
5667 {
5668 struct buffer_head *bh;
5669 struct btrfs_super_block *disk_super;
5670
5671 bh = btrfs_read_dev_super(device->bdev);
5672 if (!bh)
5673 return -EINVAL;
5674 disk_super = (struct btrfs_super_block *)bh->b_data;
5675
5676 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5677 set_buffer_dirty(bh);
5678 sync_dirty_buffer(bh);
5679 brelse(bh);
5680
5681 return 0;
5682 }
Linux kernel - Deliverables
This page took 0.143256 seconds and 6 git commands to generate.