?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'akpm' (fixes from Andrew)
[linux-2.6/cjktty.git] / fs / btrfs / volumes.c
blob5989a92236f7f1b578ec3ccf88078bb69c1093d3
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <asm/div64.h>
30 #include "compat.h"
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 struct btrfs_root *root,
46 struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
49 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
50
51 static DEFINE_MUTEX(uuid_mutex);
52 static LIST_HEAD(fs_uuids);
53
54 static void lock_chunks(struct btrfs_root *root)
55 {
56 mutex_lock(&root->fs_info->chunk_mutex);
57 }
58
59 static void unlock_chunks(struct btrfs_root *root)
60 {
61 mutex_unlock(&root->fs_info->chunk_mutex);
62 }
63
64 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
65 {
66 struct btrfs_device *device;
67 WARN_ON(fs_devices->opened);
68 while (!list_empty(&fs_devices->devices)) {
69 device = list_entry(fs_devices->devices.next,
70 struct btrfs_device, dev_list);
71 list_del(&device->dev_list);
72 rcu_string_free(device->name);
73 kfree(device);
74 }
75 kfree(fs_devices);
76 }
77
78 static void btrfs_kobject_uevent(struct block_device *bdev,
79 enum kobject_action action)
80 {
81 int ret;
82
83 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
84 if (ret)
85 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
86 action,
87 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
88 &disk_to_dev(bdev->bd_disk)->kobj);
89 }
90
91 void btrfs_cleanup_fs_uuids(void)
92 {
93 struct btrfs_fs_devices *fs_devices;
94
95 while (!list_empty(&fs_uuids)) {
96 fs_devices = list_entry(fs_uuids.next,
97 struct btrfs_fs_devices, list);
98 list_del(&fs_devices->list);
99 free_fs_devices(fs_devices);
100 }
101 }
102
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
104 u64 devid, u8 *uuid)
105 {
106 struct btrfs_device *dev;
107
108 list_for_each_entry(dev, head, dev_list) {
109 if (dev->devid == devid &&
110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
111 return dev;
112 }
113 }
114 return NULL;
115 }
116
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
118 {
119 struct btrfs_fs_devices *fs_devices;
120
121 list_for_each_entry(fs_devices, &fs_uuids, list) {
122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
123 return fs_devices;
124 }
125 return NULL;
126 }
127
128 static int
129 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
130 int flush, struct block_device **bdev,
131 struct buffer_head **bh)
132 {
133 int ret;
134
135 *bdev = blkdev_get_by_path(device_path, flags, holder);
136
137 if (IS_ERR(*bdev)) {
138 ret = PTR_ERR(*bdev);
139 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
140 goto error;
141 }
142
143 if (flush)
144 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
145 ret = set_blocksize(*bdev, 4096);
146 if (ret) {
147 blkdev_put(*bdev, flags);
148 goto error;
149 }
150 invalidate_bdev(*bdev);
151 *bh = btrfs_read_dev_super(*bdev);
152 if (!*bh) {
153 ret = -EINVAL;
154 blkdev_put(*bdev, flags);
155 goto error;
156 }
157
158 return 0;
159
160 error:
161 *bdev = NULL;
162 *bh = NULL;
163 return ret;
164 }
165
166 static void requeue_list(struct btrfs_pending_bios *pending_bios,
167 struct bio *head, struct bio *tail)
168 {
169
170 struct bio *old_head;
171
172 old_head = pending_bios->head;
173 pending_bios->head = head;
174 if (pending_bios->tail)
175 tail->bi_next = old_head;
176 else
177 pending_bios->tail = tail;
178 }
179
180 /*
181 * we try to collect pending bios for a device so we don't get a large
182 * number of procs sending bios down to the same device. This greatly
183 * improves the schedulers ability to collect and merge the bios.
184 *
185 * But, it also turns into a long list of bios to process and that is sure
186 * to eventually make the worker thread block. The solution here is to
187 * make some progress and then put this work struct back at the end of
188 * the list if the block device is congested. This way, multiple devices
189 * can make progress from a single worker thread.
190 */
191 static noinline void run_scheduled_bios(struct btrfs_device *device)
192 {
193 struct bio *pending;
194 struct backing_dev_info *bdi;
195 struct btrfs_fs_info *fs_info;
196 struct btrfs_pending_bios *pending_bios;
197 struct bio *tail;
198 struct bio *cur;
199 int again = 0;
200 unsigned long num_run;
201 unsigned long batch_run = 0;
202 unsigned long limit;
203 unsigned long last_waited = 0;
204 int force_reg = 0;
205 int sync_pending = 0;
206 struct blk_plug plug;
207
208 /*
209 * this function runs all the bios we've collected for
210 * a particular device. We don't want to wander off to
211 * another device without first sending all of these down.
212 * So, setup a plug here and finish it off before we return
213 */
214 blk_start_plug(&plug);
215
216 bdi = blk_get_backing_dev_info(device->bdev);
217 fs_info = device->dev_root->fs_info;
218 limit = btrfs_async_submit_limit(fs_info);
219 limit = limit * 2 / 3;
220
221 loop:
222 spin_lock(&device->io_lock);
223
224 loop_lock:
225 num_run = 0;
226
227 /* take all the bios off the list at once and process them
228 * later on (without the lock held). But, remember the
229 * tail and other pointers so the bios can be properly reinserted
230 * into the list if we hit congestion
231 */
232 if (!force_reg && device->pending_sync_bios.head) {
233 pending_bios = &device->pending_sync_bios;
234 force_reg = 1;
235 } else {
236 pending_bios = &device->pending_bios;
237 force_reg = 0;
238 }
239
240 pending = pending_bios->head;
241 tail = pending_bios->tail;
242 WARN_ON(pending && !tail);
243
244 /*
245 * if pending was null this time around, no bios need processing
246 * at all and we can stop. Otherwise it'll loop back up again
247 * and do an additional check so no bios are missed.
248 *
249 * device->running_pending is used to synchronize with the
250 * schedule_bio code.
251 */
252 if (device->pending_sync_bios.head == NULL &&
253 device->pending_bios.head == NULL) {
254 again = 0;
255 device->running_pending = 0;
256 } else {
257 again = 1;
258 device->running_pending = 1;
259 }
260
261 pending_bios->head = NULL;
262 pending_bios->tail = NULL;
263
264 spin_unlock(&device->io_lock);
265
266 while (pending) {
267
268 rmb();
269 /* we want to work on both lists, but do more bios on the
270 * sync list than the regular list
271 */
272 if ((num_run > 32 &&
273 pending_bios != &device->pending_sync_bios &&
274 device->pending_sync_bios.head) ||
275 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
276 device->pending_bios.head)) {
277 spin_lock(&device->io_lock);
278 requeue_list(pending_bios, pending, tail);
279 goto loop_lock;
280 }
281
282 cur = pending;
283 pending = pending->bi_next;
284 cur->bi_next = NULL;
285
286 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
287 waitqueue_active(&fs_info->async_submit_wait))
288 wake_up(&fs_info->async_submit_wait);
289
290 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
291
292 /*
293 * if we're doing the sync list, record that our
294 * plug has some sync requests on it
295 *
296 * If we're doing the regular list and there are
297 * sync requests sitting around, unplug before
298 * we add more
299 */
300 if (pending_bios == &device->pending_sync_bios) {
301 sync_pending = 1;
302 } else if (sync_pending) {
303 blk_finish_plug(&plug);
304 blk_start_plug(&plug);
305 sync_pending = 0;
306 }
307
308 btrfsic_submit_bio(cur->bi_rw, cur);
309 num_run++;
310 batch_run++;
311 if (need_resched())
312 cond_resched();
313
314 /*
315 * we made progress, there is more work to do and the bdi
316 * is now congested. Back off and let other work structs
317 * run instead
318 */
319 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
320 fs_info->fs_devices->open_devices > 1) {
321 struct io_context *ioc;
322
323 ioc = current->io_context;
324
325 /*
326 * the main goal here is that we don't want to
327 * block if we're going to be able to submit
328 * more requests without blocking.
329 *
330 * This code does two great things, it pokes into
331 * the elevator code from a filesystem _and_
332 * it makes assumptions about how batching works.
333 */
334 if (ioc && ioc->nr_batch_requests > 0 &&
335 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
336 (last_waited == 0 ||
337 ioc->last_waited == last_waited)) {
338 /*
339 * we want to go through our batch of
340 * requests and stop. So, we copy out
341 * the ioc->last_waited time and test
342 * against it before looping
343 */
344 last_waited = ioc->last_waited;
345 if (need_resched())
346 cond_resched();
347 continue;
348 }
349 spin_lock(&device->io_lock);
350 requeue_list(pending_bios, pending, tail);
351 device->running_pending = 1;
352
353 spin_unlock(&device->io_lock);
354 btrfs_requeue_work(&device->work);
355 goto done;
356 }
357 /* unplug every 64 requests just for good measure */
358 if (batch_run % 64 == 0) {
359 blk_finish_plug(&plug);
360 blk_start_plug(&plug);
361 sync_pending = 0;
362 }
363 }
364
365 cond_resched();
366 if (again)
367 goto loop;
368
369 spin_lock(&device->io_lock);
370 if (device->pending_bios.head || device->pending_sync_bios.head)
371 goto loop_lock;
372 spin_unlock(&device->io_lock);
373
374 done:
375 blk_finish_plug(&plug);
376 }
377
378 static void pending_bios_fn(struct btrfs_work *work)
379 {
380 struct btrfs_device *device;
381
382 device = container_of(work, struct btrfs_device, work);
383 run_scheduled_bios(device);
384 }
385
386 static noinline int device_list_add(const char *path,
387 struct btrfs_super_block *disk_super,
388 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
389 {
390 struct btrfs_device *device;
391 struct btrfs_fs_devices *fs_devices;
392 struct rcu_string *name;
393 u64 found_transid = btrfs_super_generation(disk_super);
394
395 fs_devices = find_fsid(disk_super->fsid);
396 if (!fs_devices) {
397 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
398 if (!fs_devices)
399 return -ENOMEM;
400 INIT_LIST_HEAD(&fs_devices->devices);
401 INIT_LIST_HEAD(&fs_devices->alloc_list);
402 list_add(&fs_devices->list, &fs_uuids);
403 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
404 fs_devices->latest_devid = devid;
405 fs_devices->latest_trans = found_transid;
406 mutex_init(&fs_devices->device_list_mutex);
407 device = NULL;
408 } else {
409 device = __find_device(&fs_devices->devices, devid,
410 disk_super->dev_item.uuid);
411 }
412 if (!device) {
413 if (fs_devices->opened)
414 return -EBUSY;
415
416 device = kzalloc(sizeof(*device), GFP_NOFS);
417 if (!device) {
418 /* we can safely leave the fs_devices entry around */
419 return -ENOMEM;
420 }
421 device->devid = devid;
422 device->dev_stats_valid = 0;
423 device->work.func = pending_bios_fn;
424 memcpy(device->uuid, disk_super->dev_item.uuid,
425 BTRFS_UUID_SIZE);
426 spin_lock_init(&device->io_lock);
427
428 name = rcu_string_strdup(path, GFP_NOFS);
429 if (!name) {
430 kfree(device);
431 return -ENOMEM;
432 }
433 rcu_assign_pointer(device->name, name);
434 INIT_LIST_HEAD(&device->dev_alloc_list);
435
436 /* init readahead state */
437 spin_lock_init(&device->reada_lock);
438 device->reada_curr_zone = NULL;
439 atomic_set(&device->reada_in_flight, 0);
440 device->reada_next = 0;
441 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
442 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
443
444 mutex_lock(&fs_devices->device_list_mutex);
445 list_add_rcu(&device->dev_list, &fs_devices->devices);
446 mutex_unlock(&fs_devices->device_list_mutex);
447
448 device->fs_devices = fs_devices;
449 fs_devices->num_devices++;
450 } else if (!device->name || strcmp(device->name->str, path)) {
451 name = rcu_string_strdup(path, GFP_NOFS);
452 if (!name)
453 return -ENOMEM;
454 rcu_string_free(device->name);
455 rcu_assign_pointer(device->name, name);
456 if (device->missing) {
457 fs_devices->missing_devices--;
458 device->missing = 0;
459 }
460 }
461
462 if (found_transid > fs_devices->latest_trans) {
463 fs_devices->latest_devid = devid;
464 fs_devices->latest_trans = found_transid;
465 }
466 *fs_devices_ret = fs_devices;
467 return 0;
468 }
469
470 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
471 {
472 struct btrfs_fs_devices *fs_devices;
473 struct btrfs_device *device;
474 struct btrfs_device *orig_dev;
475
476 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
477 if (!fs_devices)
478 return ERR_PTR(-ENOMEM);
479
480 INIT_LIST_HEAD(&fs_devices->devices);
481 INIT_LIST_HEAD(&fs_devices->alloc_list);
482 INIT_LIST_HEAD(&fs_devices->list);
483 mutex_init(&fs_devices->device_list_mutex);
484 fs_devices->latest_devid = orig->latest_devid;
485 fs_devices->latest_trans = orig->latest_trans;
486 fs_devices->total_devices = orig->total_devices;
487 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
488
489 /* We have held the volume lock, it is safe to get the devices. */
490 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
491 struct rcu_string *name;
492
493 device = kzalloc(sizeof(*device), GFP_NOFS);
494 if (!device)
495 goto error;
496
497 /*
498 * This is ok to do without rcu read locked because we hold the
499 * uuid mutex so nothing we touch in here is going to disappear.
500 */
501 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
502 if (!name) {
503 kfree(device);
504 goto error;
505 }
506 rcu_assign_pointer(device->name, name);
507
508 device->devid = orig_dev->devid;
509 device->work.func = pending_bios_fn;
510 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
511 spin_lock_init(&device->io_lock);
512 INIT_LIST_HEAD(&device->dev_list);
513 INIT_LIST_HEAD(&device->dev_alloc_list);
514
515 list_add(&device->dev_list, &fs_devices->devices);
516 device->fs_devices = fs_devices;
517 fs_devices->num_devices++;
518 }
519 return fs_devices;
520 error:
521 free_fs_devices(fs_devices);
522 return ERR_PTR(-ENOMEM);
523 }
524
525 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
526 struct btrfs_fs_devices *fs_devices, int step)
527 {
528 struct btrfs_device *device, *next;
529
530 struct block_device *latest_bdev = NULL;
531 u64 latest_devid = 0;
532 u64 latest_transid = 0;
533
534 mutex_lock(&uuid_mutex);
535 again:
536 /* This is the initialized path, it is safe to release the devices. */
537 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
538 if (device->in_fs_metadata) {
539 if (!device->is_tgtdev_for_dev_replace &&
540 (!latest_transid ||
541 device->generation > latest_transid)) {
542 latest_devid = device->devid;
543 latest_transid = device->generation;
544 latest_bdev = device->bdev;
545 }
546 continue;
547 }
548
549 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
550 /*
551 * In the first step, keep the device which has
552 * the correct fsid and the devid that is used
553 * for the dev_replace procedure.
554 * In the second step, the dev_replace state is
555 * read from the device tree and it is known
556 * whether the procedure is really active or
557 * not, which means whether this device is
558 * used or whether it should be removed.
559 */
560 if (step == 0 || device->is_tgtdev_for_dev_replace) {
561 continue;
562 }
563 }
564 if (device->bdev) {
565 blkdev_put(device->bdev, device->mode);
566 device->bdev = NULL;
567 fs_devices->open_devices--;
568 }
569 if (device->writeable) {
570 list_del_init(&device->dev_alloc_list);
571 device->writeable = 0;
572 if (!device->is_tgtdev_for_dev_replace)
573 fs_devices->rw_devices--;
574 }
575 list_del_init(&device->dev_list);
576 fs_devices->num_devices--;
577 rcu_string_free(device->name);
578 kfree(device);
579 }
580
581 if (fs_devices->seed) {
582 fs_devices = fs_devices->seed;
583 goto again;
584 }
585
586 fs_devices->latest_bdev = latest_bdev;
587 fs_devices->latest_devid = latest_devid;
588 fs_devices->latest_trans = latest_transid;
589
590 mutex_unlock(&uuid_mutex);
591 }
592
593 static void __free_device(struct work_struct *work)
594 {
595 struct btrfs_device *device;
596
597 device = container_of(work, struct btrfs_device, rcu_work);
598
599 if (device->bdev)
600 blkdev_put(device->bdev, device->mode);
601
602 rcu_string_free(device->name);
603 kfree(device);
604 }
605
606 static void free_device(struct rcu_head *head)
607 {
608 struct btrfs_device *device;
609
610 device = container_of(head, struct btrfs_device, rcu);
611
612 INIT_WORK(&device->rcu_work, __free_device);
613 schedule_work(&device->rcu_work);
614 }
615
616 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
617 {
618 struct btrfs_device *device;
619
620 if (--fs_devices->opened > 0)
621 return 0;
622
623 mutex_lock(&fs_devices->device_list_mutex);
624 list_for_each_entry(device, &fs_devices->devices, dev_list) {
625 struct btrfs_device *new_device;
626 struct rcu_string *name;
627
628 if (device->bdev)
629 fs_devices->open_devices--;
630
631 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
632 list_del_init(&device->dev_alloc_list);
633 fs_devices->rw_devices--;
634 }
635
636 if (device->can_discard)
637 fs_devices->num_can_discard--;
638
639 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
640 BUG_ON(!new_device); /* -ENOMEM */
641 memcpy(new_device, device, sizeof(*new_device));
642
643 /* Safe because we are under uuid_mutex */
644 if (device->name) {
645 name = rcu_string_strdup(device->name->str, GFP_NOFS);
646 BUG_ON(device->name && !name); /* -ENOMEM */
647 rcu_assign_pointer(new_device->name, name);
648 }
649 new_device->bdev = NULL;
650 new_device->writeable = 0;
651 new_device->in_fs_metadata = 0;
652 new_device->can_discard = 0;
653 spin_lock_init(&new_device->io_lock);
654 list_replace_rcu(&device->dev_list, &new_device->dev_list);
655
656 call_rcu(&device->rcu, free_device);
657 }
658 mutex_unlock(&fs_devices->device_list_mutex);
659
660 WARN_ON(fs_devices->open_devices);
661 WARN_ON(fs_devices->rw_devices);
662 fs_devices->opened = 0;
663 fs_devices->seeding = 0;
664
665 return 0;
666 }
667
668 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
669 {
670 struct btrfs_fs_devices *seed_devices = NULL;
671 int ret;
672
673 mutex_lock(&uuid_mutex);
674 ret = __btrfs_close_devices(fs_devices);
675 if (!fs_devices->opened) {
676 seed_devices = fs_devices->seed;
677 fs_devices->seed = NULL;
678 }
679 mutex_unlock(&uuid_mutex);
680
681 while (seed_devices) {
682 fs_devices = seed_devices;
683 seed_devices = fs_devices->seed;
684 __btrfs_close_devices(fs_devices);
685 free_fs_devices(fs_devices);
686 }
687 /*
688 * Wait for rcu kworkers under __btrfs_close_devices
689 * to finish all blkdev_puts so device is really
690 * free when umount is done.
691 */
692 rcu_barrier();
693 return ret;
694 }
695
696 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
697 fmode_t flags, void *holder)
698 {
699 struct request_queue *q;
700 struct block_device *bdev;
701 struct list_head *head = &fs_devices->devices;
702 struct btrfs_device *device;
703 struct block_device *latest_bdev = NULL;
704 struct buffer_head *bh;
705 struct btrfs_super_block *disk_super;
706 u64 latest_devid = 0;
707 u64 latest_transid = 0;
708 u64 devid;
709 int seeding = 1;
710 int ret = 0;
711
712 flags |= FMODE_EXCL;
713
714 list_for_each_entry(device, head, dev_list) {
715 if (device->bdev)
716 continue;
717 if (!device->name)
718 continue;
719
720 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
721 &bdev, &bh);
722 if (ret)
723 continue;
724
725 disk_super = (struct btrfs_super_block *)bh->b_data;
726 devid = btrfs_stack_device_id(&disk_super->dev_item);
727 if (devid != device->devid)
728 goto error_brelse;
729
730 if (memcmp(device->uuid, disk_super->dev_item.uuid,
731 BTRFS_UUID_SIZE))
732 goto error_brelse;
733
734 device->generation = btrfs_super_generation(disk_super);
735 if (!latest_transid || device->generation > latest_transid) {
736 latest_devid = devid;
737 latest_transid = device->generation;
738 latest_bdev = bdev;
739 }
740
741 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
742 device->writeable = 0;
743 } else {
744 device->writeable = !bdev_read_only(bdev);
745 seeding = 0;
746 }
747
748 q = bdev_get_queue(bdev);
749 if (blk_queue_discard(q)) {
750 device->can_discard = 1;
751 fs_devices->num_can_discard++;
752 }
753
754 device->bdev = bdev;
755 device->in_fs_metadata = 0;
756 device->mode = flags;
757
758 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
759 fs_devices->rotating = 1;
760
761 fs_devices->open_devices++;
762 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
763 fs_devices->rw_devices++;
764 list_add(&device->dev_alloc_list,
765 &fs_devices->alloc_list);
766 }
767 brelse(bh);
768 continue;
769
770 error_brelse:
771 brelse(bh);
772 blkdev_put(bdev, flags);
773 continue;
774 }
775 if (fs_devices->open_devices == 0) {
776 ret = -EINVAL;
777 goto out;
778 }
779 fs_devices->seeding = seeding;
780 fs_devices->opened = 1;
781 fs_devices->latest_bdev = latest_bdev;
782 fs_devices->latest_devid = latest_devid;
783 fs_devices->latest_trans = latest_transid;
784 fs_devices->total_rw_bytes = 0;
785 out:
786 return ret;
787 }
788
789 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
790 fmode_t flags, void *holder)
791 {
792 int ret;
793
794 mutex_lock(&uuid_mutex);
795 if (fs_devices->opened) {
796 fs_devices->opened++;
797 ret = 0;
798 } else {
799 ret = __btrfs_open_devices(fs_devices, flags, holder);
800 }
801 mutex_unlock(&uuid_mutex);
802 return ret;
803 }
804
805 /*
806 * Look for a btrfs signature on a device. This may be called out of the mount path
807 * and we are not allowed to call set_blocksize during the scan. The superblock
808 * is read via pagecache
809 */
810 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
811 struct btrfs_fs_devices **fs_devices_ret)
812 {
813 struct btrfs_super_block *disk_super;
814 struct block_device *bdev;
815 struct page *page;
816 void *p;
817 int ret = -EINVAL;
818 u64 devid;
819 u64 transid;
820 u64 total_devices;
821 u64 bytenr;
822 pgoff_t index;
823
824 /*
825 * we would like to check all the supers, but that would make
826 * a btrfs mount succeed after a mkfs from a different FS.
827 * So, we need to add a special mount option to scan for
828 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
829 */
830 bytenr = btrfs_sb_offset(0);
831 flags |= FMODE_EXCL;
832 mutex_lock(&uuid_mutex);
833
834 bdev = blkdev_get_by_path(path, flags, holder);
835
836 if (IS_ERR(bdev)) {
837 ret = PTR_ERR(bdev);
838 goto error;
839 }
840
841 /* make sure our super fits in the device */
842 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
843 goto error_bdev_put;
844
845 /* make sure our super fits in the page */
846 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
847 goto error_bdev_put;
848
849 /* make sure our super doesn't straddle pages on disk */
850 index = bytenr >> PAGE_CACHE_SHIFT;
851 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
852 goto error_bdev_put;
853
854 /* pull in the page with our super */
855 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
856 index, GFP_NOFS);
857
858 if (IS_ERR_OR_NULL(page))
859 goto error_bdev_put;
860
861 p = kmap(page);
862
863 /* align our pointer to the offset of the super block */
864 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
865
866 if (btrfs_super_bytenr(disk_super) != bytenr ||
867 disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
868 goto error_unmap;
869
870 devid = btrfs_stack_device_id(&disk_super->dev_item);
871 transid = btrfs_super_generation(disk_super);
872 total_devices = btrfs_super_num_devices(disk_super);
873
874 if (disk_super->label[0]) {
875 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
876 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
877 printk(KERN_INFO "device label %s ", disk_super->label);
878 } else {
879 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
880 }
881
882 printk(KERN_CONT "devid %llu transid %llu %s\n",
883 (unsigned long long)devid, (unsigned long long)transid, path);
884
885 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
886 if (!ret && fs_devices_ret)
887 (*fs_devices_ret)->total_devices = total_devices;
888
889 error_unmap:
890 kunmap(page);
891 page_cache_release(page);
892
893 error_bdev_put:
894 blkdev_put(bdev, flags);
895 error:
896 mutex_unlock(&uuid_mutex);
897 return ret;
898 }
899
900 /* helper to account the used device space in the range */
901 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
902 u64 end, u64 *length)
903 {
904 struct btrfs_key key;
905 struct btrfs_root *root = device->dev_root;
906 struct btrfs_dev_extent *dev_extent;
907 struct btrfs_path *path;
908 u64 extent_end;
909 int ret;
910 int slot;
911 struct extent_buffer *l;
912
913 *length = 0;
914
915 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
916 return 0;
917
918 path = btrfs_alloc_path();
919 if (!path)
920 return -ENOMEM;
921 path->reada = 2;
922
923 key.objectid = device->devid;
924 key.offset = start;
925 key.type = BTRFS_DEV_EXTENT_KEY;
926
927 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
928 if (ret < 0)
929 goto out;
930 if (ret > 0) {
931 ret = btrfs_previous_item(root, path, key.objectid, key.type);
932 if (ret < 0)
933 goto out;
934 }
935
936 while (1) {
937 l = path->nodes[0];
938 slot = path->slots[0];
939 if (slot >= btrfs_header_nritems(l)) {
940 ret = btrfs_next_leaf(root, path);
941 if (ret == 0)
942 continue;
943 if (ret < 0)
944 goto out;
945
946 break;
947 }
948 btrfs_item_key_to_cpu(l, &key, slot);
949
950 if (key.objectid < device->devid)
951 goto next;
952
953 if (key.objectid > device->devid)
954 break;
955
956 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
957 goto next;
958
959 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
960 extent_end = key.offset + btrfs_dev_extent_length(l,
961 dev_extent);
962 if (key.offset <= start && extent_end > end) {
963 *length = end - start + 1;
964 break;
965 } else if (key.offset <= start && extent_end > start)
966 *length += extent_end - start;
967 else if (key.offset > start && extent_end <= end)
968 *length += extent_end - key.offset;
969 else if (key.offset > start && key.offset <= end) {
970 *length += end - key.offset + 1;
971 break;
972 } else if (key.offset > end)
973 break;
974
975 next:
976 path->slots[0]++;
977 }
978 ret = 0;
979 out:
980 btrfs_free_path(path);
981 return ret;
982 }
983
984 /*
985 * find_free_dev_extent - find free space in the specified device
986 * @device: the device which we search the free space in
987 * @num_bytes: the size of the free space that we need
988 * @start: store the start of the free space.
989 * @len: the size of the free space. that we find, or the size of the max
990 * free space if we don't find suitable free space
991 *
992 * this uses a pretty simple search, the expectation is that it is
993 * called very infrequently and that a given device has a small number
994 * of extents
995 *
996 * @start is used to store the start of the free space if we find. But if we
997 * don't find suitable free space, it will be used to store the start position
998 * of the max free space.
999 *
1000 * @len is used to store the size of the free space that we find.
1001 * But if we don't find suitable free space, it is used to store the size of
1002 * the max free space.
1003 */
1004 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1005 u64 *start, u64 *len)
1006 {
1007 struct btrfs_key key;
1008 struct btrfs_root *root = device->dev_root;
1009 struct btrfs_dev_extent *dev_extent;
1010 struct btrfs_path *path;
1011 u64 hole_size;
1012 u64 max_hole_start;
1013 u64 max_hole_size;
1014 u64 extent_end;
1015 u64 search_start;
1016 u64 search_end = device->total_bytes;
1017 int ret;
1018 int slot;
1019 struct extent_buffer *l;
1020
1021 /* FIXME use last free of some kind */
1022
1023 /* we don't want to overwrite the superblock on the drive,
1024 * so we make sure to start at an offset of at least 1MB
1025 */
1026 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1027
1028 max_hole_start = search_start;
1029 max_hole_size = 0;
1030 hole_size = 0;
1031
1032 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1033 ret = -ENOSPC;
1034 goto error;
1035 }
1036
1037 path = btrfs_alloc_path();
1038 if (!path) {
1039 ret = -ENOMEM;
1040 goto error;
1041 }
1042 path->reada = 2;
1043
1044 key.objectid = device->devid;
1045 key.offset = search_start;
1046 key.type = BTRFS_DEV_EXTENT_KEY;
1047
1048 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1049 if (ret < 0)
1050 goto out;
1051 if (ret > 0) {
1052 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1053 if (ret < 0)
1054 goto out;
1055 }
1056
1057 while (1) {
1058 l = path->nodes[0];
1059 slot = path->slots[0];
1060 if (slot >= btrfs_header_nritems(l)) {
1061 ret = btrfs_next_leaf(root, path);
1062 if (ret == 0)
1063 continue;
1064 if (ret < 0)
1065 goto out;
1066
1067 break;
1068 }
1069 btrfs_item_key_to_cpu(l, &key, slot);
1070
1071 if (key.objectid < device->devid)
1072 goto next;
1073
1074 if (key.objectid > device->devid)
1075 break;
1076
1077 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1078 goto next;
1079
1080 if (key.offset > search_start) {
1081 hole_size = key.offset - search_start;
1082
1083 if (hole_size > max_hole_size) {
1084 max_hole_start = search_start;
1085 max_hole_size = hole_size;
1086 }
1087
1088 /*
1089 * If this free space is greater than which we need,
1090 * it must be the max free space that we have found
1091 * until now, so max_hole_start must point to the start
1092 * of this free space and the length of this free space
1093 * is stored in max_hole_size. Thus, we return
1094 * max_hole_start and max_hole_size and go back to the
1095 * caller.
1096 */
1097 if (hole_size >= num_bytes) {
1098 ret = 0;
1099 goto out;
1100 }
1101 }
1102
1103 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1104 extent_end = key.offset + btrfs_dev_extent_length(l,
1105 dev_extent);
1106 if (extent_end > search_start)
1107 search_start = extent_end;
1108 next:
1109 path->slots[0]++;
1110 cond_resched();
1111 }
1112
1113 /*
1114 * At this point, search_start should be the end of
1115 * allocated dev extents, and when shrinking the device,
1116 * search_end may be smaller than search_start.
1117 */
1118 if (search_end > search_start)
1119 hole_size = search_end - search_start;
1120
1121 if (hole_size > max_hole_size) {
1122 max_hole_start = search_start;
1123 max_hole_size = hole_size;
1124 }
1125
1126 /* See above. */
1127 if (hole_size < num_bytes)
1128 ret = -ENOSPC;
1129 else
1130 ret = 0;
1131
1132 out:
1133 btrfs_free_path(path);
1134 error:
1135 *start = max_hole_start;
1136 if (len)
1137 *len = max_hole_size;
1138 return ret;
1139 }
1140
1141 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1142 struct btrfs_device *device,
1143 u64 start)
1144 {
1145 int ret;
1146 struct btrfs_path *path;
1147 struct btrfs_root *root = device->dev_root;
1148 struct btrfs_key key;
1149 struct btrfs_key found_key;
1150 struct extent_buffer *leaf = NULL;
1151 struct btrfs_dev_extent *extent = NULL;
1152
1153 path = btrfs_alloc_path();
1154 if (!path)
1155 return -ENOMEM;
1156
1157 key.objectid = device->devid;
1158 key.offset = start;
1159 key.type = BTRFS_DEV_EXTENT_KEY;
1160 again:
1161 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1162 if (ret > 0) {
1163 ret = btrfs_previous_item(root, path, key.objectid,
1164 BTRFS_DEV_EXTENT_KEY);
1165 if (ret)
1166 goto out;
1167 leaf = path->nodes[0];
1168 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1169 extent = btrfs_item_ptr(leaf, path->slots[0],
1170 struct btrfs_dev_extent);
1171 BUG_ON(found_key.offset > start || found_key.offset +
1172 btrfs_dev_extent_length(leaf, extent) < start);
1173 key = found_key;
1174 btrfs_release_path(path);
1175 goto again;
1176 } else if (ret == 0) {
1177 leaf = path->nodes[0];
1178 extent = btrfs_item_ptr(leaf, path->slots[0],
1179 struct btrfs_dev_extent);
1180 } else {
1181 btrfs_error(root->fs_info, ret, "Slot search failed");
1182 goto out;
1183 }
1184
1185 if (device->bytes_used > 0) {
1186 u64 len = btrfs_dev_extent_length(leaf, extent);
1187 device->bytes_used -= len;
1188 spin_lock(&root->fs_info->free_chunk_lock);
1189 root->fs_info->free_chunk_space += len;
1190 spin_unlock(&root->fs_info->free_chunk_lock);
1191 }
1192 ret = btrfs_del_item(trans, root, path);
1193 if (ret) {
1194 btrfs_error(root->fs_info, ret,
1195 "Failed to remove dev extent item");
1196 }
1197 out:
1198 btrfs_free_path(path);
1199 return ret;
1200 }
1201
1202 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1203 struct btrfs_device *device,
1204 u64 chunk_tree, u64 chunk_objectid,
1205 u64 chunk_offset, u64 start, u64 num_bytes)
1206 {
1207 int ret;
1208 struct btrfs_path *path;
1209 struct btrfs_root *root = device->dev_root;
1210 struct btrfs_dev_extent *extent;
1211 struct extent_buffer *leaf;
1212 struct btrfs_key key;
1213
1214 WARN_ON(!device->in_fs_metadata);
1215 WARN_ON(device->is_tgtdev_for_dev_replace);
1216 path = btrfs_alloc_path();
1217 if (!path)
1218 return -ENOMEM;
1219
1220 key.objectid = device->devid;
1221 key.offset = start;
1222 key.type = BTRFS_DEV_EXTENT_KEY;
1223 ret = btrfs_insert_empty_item(trans, root, path, &key,
1224 sizeof(*extent));
1225 if (ret)
1226 goto out;
1227
1228 leaf = path->nodes[0];
1229 extent = btrfs_item_ptr(leaf, path->slots[0],
1230 struct btrfs_dev_extent);
1231 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1232 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1233 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1234
1235 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1236 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1237 BTRFS_UUID_SIZE);
1238
1239 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1240 btrfs_mark_buffer_dirty(leaf);
1241 out:
1242 btrfs_free_path(path);
1243 return ret;
1244 }
1245
1246 static noinline int find_next_chunk(struct btrfs_root *root,
1247 u64 objectid, u64 *offset)
1248 {
1249 struct btrfs_path *path;
1250 int ret;
1251 struct btrfs_key key;
1252 struct btrfs_chunk *chunk;
1253 struct btrfs_key found_key;
1254
1255 path = btrfs_alloc_path();
1256 if (!path)
1257 return -ENOMEM;
1258
1259 key.objectid = objectid;
1260 key.offset = (u64)-1;
1261 key.type = BTRFS_CHUNK_ITEM_KEY;
1262
1263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1264 if (ret < 0)
1265 goto error;
1266
1267 BUG_ON(ret == 0); /* Corruption */
1268
1269 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1270 if (ret) {
1271 *offset = 0;
1272 } else {
1273 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1274 path->slots[0]);
1275 if (found_key.objectid != objectid)
1276 *offset = 0;
1277 else {
1278 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1279 struct btrfs_chunk);
1280 *offset = found_key.offset +
1281 btrfs_chunk_length(path->nodes[0], chunk);
1282 }
1283 }
1284 ret = 0;
1285 error:
1286 btrfs_free_path(path);
1287 return ret;
1288 }
1289
1290 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1291 {
1292 int ret;
1293 struct btrfs_key key;
1294 struct btrfs_key found_key;
1295 struct btrfs_path *path;
1296
1297 root = root->fs_info->chunk_root;
1298
1299 path = btrfs_alloc_path();
1300 if (!path)
1301 return -ENOMEM;
1302
1303 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1304 key.type = BTRFS_DEV_ITEM_KEY;
1305 key.offset = (u64)-1;
1306
1307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1308 if (ret < 0)
1309 goto error;
1310
1311 BUG_ON(ret == 0); /* Corruption */
1312
1313 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1314 BTRFS_DEV_ITEM_KEY);
1315 if (ret) {
1316 *objectid = 1;
1317 } else {
1318 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1319 path->slots[0]);
1320 *objectid = found_key.offset + 1;
1321 }
1322 ret = 0;
1323 error:
1324 btrfs_free_path(path);
1325 return ret;
1326 }
1327
1328 /*
1329 * the device information is stored in the chunk root
1330 * the btrfs_device struct should be fully filled in
1331 */
1332 int btrfs_add_device(struct btrfs_trans_handle *trans,
1333 struct btrfs_root *root,
1334 struct btrfs_device *device)
1335 {
1336 int ret;
1337 struct btrfs_path *path;
1338 struct btrfs_dev_item *dev_item;
1339 struct extent_buffer *leaf;
1340 struct btrfs_key key;
1341 unsigned long ptr;
1342
1343 root = root->fs_info->chunk_root;
1344
1345 path = btrfs_alloc_path();
1346 if (!path)
1347 return -ENOMEM;
1348
1349 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1350 key.type = BTRFS_DEV_ITEM_KEY;
1351 key.offset = device->devid;
1352
1353 ret = btrfs_insert_empty_item(trans, root, path, &key,
1354 sizeof(*dev_item));
1355 if (ret)
1356 goto out;
1357
1358 leaf = path->nodes[0];
1359 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1360
1361 btrfs_set_device_id(leaf, dev_item, device->devid);
1362 btrfs_set_device_generation(leaf, dev_item, 0);
1363 btrfs_set_device_type(leaf, dev_item, device->type);
1364 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1365 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1366 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1367 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1368 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1369 btrfs_set_device_group(leaf, dev_item, 0);
1370 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1371 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1372 btrfs_set_device_start_offset(leaf, dev_item, 0);
1373
1374 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1375 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1376 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1377 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1378 btrfs_mark_buffer_dirty(leaf);
1379
1380 ret = 0;
1381 out:
1382 btrfs_free_path(path);
1383 return ret;
1384 }
1385
1386 static int btrfs_rm_dev_item(struct btrfs_root *root,
1387 struct btrfs_device *device)
1388 {
1389 int ret;
1390 struct btrfs_path *path;
1391 struct btrfs_key key;
1392 struct btrfs_trans_handle *trans;
1393
1394 root = root->fs_info->chunk_root;
1395
1396 path = btrfs_alloc_path();
1397 if (!path)
1398 return -ENOMEM;
1399
1400 trans = btrfs_start_transaction(root, 0);
1401 if (IS_ERR(trans)) {
1402 btrfs_free_path(path);
1403 return PTR_ERR(trans);
1404 }
1405 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1406 key.type = BTRFS_DEV_ITEM_KEY;
1407 key.offset = device->devid;
1408 lock_chunks(root);
1409
1410 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1411 if (ret < 0)
1412 goto out;
1413
1414 if (ret > 0) {
1415 ret = -ENOENT;
1416 goto out;
1417 }
1418
1419 ret = btrfs_del_item(trans, root, path);
1420 if (ret)
1421 goto out;
1422 out:
1423 btrfs_free_path(path);
1424 unlock_chunks(root);
1425 btrfs_commit_transaction(trans, root);
1426 return ret;
1427 }
1428
1429 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1430 {
1431 struct btrfs_device *device;
1432 struct btrfs_device *next_device;
1433 struct block_device *bdev;
1434 struct buffer_head *bh = NULL;
1435 struct btrfs_super_block *disk_super;
1436 struct btrfs_fs_devices *cur_devices;
1437 u64 all_avail;
1438 u64 devid;
1439 u64 num_devices;
1440 u8 *dev_uuid;
1441 unsigned seq;
1442 int ret = 0;
1443 bool clear_super = false;
1444
1445 mutex_lock(&uuid_mutex);
1446
1447 do {
1448 seq = read_seqbegin(&root->fs_info->profiles_lock);
1449
1450 all_avail = root->fs_info->avail_data_alloc_bits |
1451 root->fs_info->avail_system_alloc_bits |
1452 root->fs_info->avail_metadata_alloc_bits;
1453 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1454
1455 num_devices = root->fs_info->fs_devices->num_devices;
1456 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1457 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1458 WARN_ON(num_devices < 1);
1459 num_devices--;
1460 }
1461 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1462
1463 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1464 printk(KERN_ERR "btrfs: unable to go below four devices "
1465 "on raid10\n");
1466 ret = -EINVAL;
1467 goto out;
1468 }
1469
1470 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1471 printk(KERN_ERR "btrfs: unable to go below two "
1472 "devices on raid1\n");
1473 ret = -EINVAL;
1474 goto out;
1475 }
1476
1477 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1478 root->fs_info->fs_devices->rw_devices <= 2) {
1479 printk(KERN_ERR "btrfs: unable to go below two "
1480 "devices on raid5\n");
1481 ret = -EINVAL;
1482 goto out;
1483 }
1484 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1485 root->fs_info->fs_devices->rw_devices <= 3) {
1486 printk(KERN_ERR "btrfs: unable to go below three "
1487 "devices on raid6\n");
1488 ret = -EINVAL;
1489 goto out;
1490 }
1491
1492 if (strcmp(device_path, "missing") == 0) {
1493 struct list_head *devices;
1494 struct btrfs_device *tmp;
1495
1496 device = NULL;
1497 devices = &root->fs_info->fs_devices->devices;
1498 /*
1499 * It is safe to read the devices since the volume_mutex
1500 * is held.
1501 */
1502 list_for_each_entry(tmp, devices, dev_list) {
1503 if (tmp->in_fs_metadata &&
1504 !tmp->is_tgtdev_for_dev_replace &&
1505 !tmp->bdev) {
1506 device = tmp;
1507 break;
1508 }
1509 }
1510 bdev = NULL;
1511 bh = NULL;
1512 disk_super = NULL;
1513 if (!device) {
1514 printk(KERN_ERR "btrfs: no missing devices found to "
1515 "remove\n");
1516 goto out;
1517 }
1518 } else {
1519 ret = btrfs_get_bdev_and_sb(device_path,
1520 FMODE_WRITE | FMODE_EXCL,
1521 root->fs_info->bdev_holder, 0,
1522 &bdev, &bh);
1523 if (ret)
1524 goto out;
1525 disk_super = (struct btrfs_super_block *)bh->b_data;
1526 devid = btrfs_stack_device_id(&disk_super->dev_item);
1527 dev_uuid = disk_super->dev_item.uuid;
1528 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1529 disk_super->fsid);
1530 if (!device) {
1531 ret = -ENOENT;
1532 goto error_brelse;
1533 }
1534 }
1535
1536 if (device->is_tgtdev_for_dev_replace) {
1537 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1538 ret = -EINVAL;
1539 goto error_brelse;
1540 }
1541
1542 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1543 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1544 "device\n");
1545 ret = -EINVAL;
1546 goto error_brelse;
1547 }
1548
1549 if (device->writeable) {
1550 lock_chunks(root);
1551 list_del_init(&device->dev_alloc_list);
1552 unlock_chunks(root);
1553 root->fs_info->fs_devices->rw_devices--;
1554 clear_super = true;
1555 }
1556
1557 ret = btrfs_shrink_device(device, 0);
1558 if (ret)
1559 goto error_undo;
1560
1561 /*
1562 * TODO: the superblock still includes this device in its num_devices
1563 * counter although write_all_supers() is not locked out. This
1564 * could give a filesystem state which requires a degraded mount.
1565 */
1566 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1567 if (ret)
1568 goto error_undo;
1569
1570 spin_lock(&root->fs_info->free_chunk_lock);
1571 root->fs_info->free_chunk_space = device->total_bytes -
1572 device->bytes_used;
1573 spin_unlock(&root->fs_info->free_chunk_lock);
1574
1575 device->in_fs_metadata = 0;
1576 btrfs_scrub_cancel_dev(root->fs_info, device);
1577
1578 /*
1579 * the device list mutex makes sure that we don't change
1580 * the device list while someone else is writing out all
1581 * the device supers.
1582 */
1583
1584 cur_devices = device->fs_devices;
1585 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1586 list_del_rcu(&device->dev_list);
1587
1588 device->fs_devices->num_devices--;
1589 device->fs_devices->total_devices--;
1590
1591 if (device->missing)
1592 root->fs_info->fs_devices->missing_devices--;
1593
1594 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1595 struct btrfs_device, dev_list);
1596 if (device->bdev == root->fs_info->sb->s_bdev)
1597 root->fs_info->sb->s_bdev = next_device->bdev;
1598 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1599 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1600
1601 if (device->bdev)
1602 device->fs_devices->open_devices--;
1603
1604 call_rcu(&device->rcu, free_device);
1605 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1606
1607 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1608 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1609
1610 if (cur_devices->open_devices == 0) {
1611 struct btrfs_fs_devices *fs_devices;
1612 fs_devices = root->fs_info->fs_devices;
1613 while (fs_devices) {
1614 if (fs_devices->seed == cur_devices)
1615 break;
1616 fs_devices = fs_devices->seed;
1617 }
1618 fs_devices->seed = cur_devices->seed;
1619 cur_devices->seed = NULL;
1620 lock_chunks(root);
1621 __btrfs_close_devices(cur_devices);
1622 unlock_chunks(root);
1623 free_fs_devices(cur_devices);
1624 }
1625
1626 root->fs_info->num_tolerated_disk_barrier_failures =
1627 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1628
1629 /*
1630 * at this point, the device is zero sized. We want to
1631 * remove it from the devices list and zero out the old super
1632 */
1633 if (clear_super && disk_super) {
1634 /* make sure this device isn't detected as part of
1635 * the FS anymore
1636 */
1637 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1638 set_buffer_dirty(bh);
1639 sync_dirty_buffer(bh);
1640 }
1641
1642 ret = 0;
1643
1644 /* Notify udev that device has changed */
1645 if (bdev)
1646 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1647
1648 error_brelse:
1649 brelse(bh);
1650 if (bdev)
1651 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1652 out:
1653 mutex_unlock(&uuid_mutex);
1654 return ret;
1655 error_undo:
1656 if (device->writeable) {
1657 lock_chunks(root);
1658 list_add(&device->dev_alloc_list,
1659 &root->fs_info->fs_devices->alloc_list);
1660 unlock_chunks(root);
1661 root->fs_info->fs_devices->rw_devices++;
1662 }
1663 goto error_brelse;
1664 }
1665
1666 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1667 struct btrfs_device *srcdev)
1668 {
1669 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1670 list_del_rcu(&srcdev->dev_list);
1671 list_del_rcu(&srcdev->dev_alloc_list);
1672 fs_info->fs_devices->num_devices--;
1673 if (srcdev->missing) {
1674 fs_info->fs_devices->missing_devices--;
1675 fs_info->fs_devices->rw_devices++;
1676 }
1677 if (srcdev->can_discard)
1678 fs_info->fs_devices->num_can_discard--;
1679 if (srcdev->bdev)
1680 fs_info->fs_devices->open_devices--;
1681
1682 call_rcu(&srcdev->rcu, free_device);
1683 }
1684
1685 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1686 struct btrfs_device *tgtdev)
1687 {
1688 struct btrfs_device *next_device;
1689
1690 WARN_ON(!tgtdev);
1691 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1692 if (tgtdev->bdev) {
1693 btrfs_scratch_superblock(tgtdev);
1694 fs_info->fs_devices->open_devices--;
1695 }
1696 fs_info->fs_devices->num_devices--;
1697 if (tgtdev->can_discard)
1698 fs_info->fs_devices->num_can_discard++;
1699
1700 next_device = list_entry(fs_info->fs_devices->devices.next,
1701 struct btrfs_device, dev_list);
1702 if (tgtdev->bdev == fs_info->sb->s_bdev)
1703 fs_info->sb->s_bdev = next_device->bdev;
1704 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1705 fs_info->fs_devices->latest_bdev = next_device->bdev;
1706 list_del_rcu(&tgtdev->dev_list);
1707
1708 call_rcu(&tgtdev->rcu, free_device);
1709
1710 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1711 }
1712
1713 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1714 struct btrfs_device **device)
1715 {
1716 int ret = 0;
1717 struct btrfs_super_block *disk_super;
1718 u64 devid;
1719 u8 *dev_uuid;
1720 struct block_device *bdev;
1721 struct buffer_head *bh;
1722
1723 *device = NULL;
1724 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1725 root->fs_info->bdev_holder, 0, &bdev, &bh);
1726 if (ret)
1727 return ret;
1728 disk_super = (struct btrfs_super_block *)bh->b_data;
1729 devid = btrfs_stack_device_id(&disk_super->dev_item);
1730 dev_uuid = disk_super->dev_item.uuid;
1731 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1732 disk_super->fsid);
1733 brelse(bh);
1734 if (!*device)
1735 ret = -ENOENT;
1736 blkdev_put(bdev, FMODE_READ);
1737 return ret;
1738 }
1739
1740 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1741 char *device_path,
1742 struct btrfs_device **device)
1743 {
1744 *device = NULL;
1745 if (strcmp(device_path, "missing") == 0) {
1746 struct list_head *devices;
1747 struct btrfs_device *tmp;
1748
1749 devices = &root->fs_info->fs_devices->devices;
1750 /*
1751 * It is safe to read the devices since the volume_mutex
1752 * is held by the caller.
1753 */
1754 list_for_each_entry(tmp, devices, dev_list) {
1755 if (tmp->in_fs_metadata && !tmp->bdev) {
1756 *device = tmp;
1757 break;
1758 }
1759 }
1760
1761 if (!*device) {
1762 pr_err("btrfs: no missing device found\n");
1763 return -ENOENT;
1764 }
1765
1766 return 0;
1767 } else {
1768 return btrfs_find_device_by_path(root, device_path, device);
1769 }
1770 }
1771
1772 /*
1773 * does all the dirty work required for changing file system's UUID.
1774 */
1775 static int btrfs_prepare_sprout(struct btrfs_root *root)
1776 {
1777 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1778 struct btrfs_fs_devices *old_devices;
1779 struct btrfs_fs_devices *seed_devices;
1780 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1781 struct btrfs_device *device;
1782 u64 super_flags;
1783
1784 BUG_ON(!mutex_is_locked(&uuid_mutex));
1785 if (!fs_devices->seeding)
1786 return -EINVAL;
1787
1788 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1789 if (!seed_devices)
1790 return -ENOMEM;
1791
1792 old_devices = clone_fs_devices(fs_devices);
1793 if (IS_ERR(old_devices)) {
1794 kfree(seed_devices);
1795 return PTR_ERR(old_devices);
1796 }
1797
1798 list_add(&old_devices->list, &fs_uuids);
1799
1800 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1801 seed_devices->opened = 1;
1802 INIT_LIST_HEAD(&seed_devices->devices);
1803 INIT_LIST_HEAD(&seed_devices->alloc_list);
1804 mutex_init(&seed_devices->device_list_mutex);
1805
1806 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1807 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1808 synchronize_rcu);
1809 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1810
1811 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1812 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1813 device->fs_devices = seed_devices;
1814 }
1815
1816 fs_devices->seeding = 0;
1817 fs_devices->num_devices = 0;
1818 fs_devices->open_devices = 0;
1819 fs_devices->total_devices = 0;
1820 fs_devices->seed = seed_devices;
1821
1822 generate_random_uuid(fs_devices->fsid);
1823 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1824 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1825 super_flags = btrfs_super_flags(disk_super) &
1826 ~BTRFS_SUPER_FLAG_SEEDING;
1827 btrfs_set_super_flags(disk_super, super_flags);
1828
1829 return 0;
1830 }
1831
1832 /*
1833 * strore the expected generation for seed devices in device items.
1834 */
1835 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1836 struct btrfs_root *root)
1837 {
1838 struct btrfs_path *path;
1839 struct extent_buffer *leaf;
1840 struct btrfs_dev_item *dev_item;
1841 struct btrfs_device *device;
1842 struct btrfs_key key;
1843 u8 fs_uuid[BTRFS_UUID_SIZE];
1844 u8 dev_uuid[BTRFS_UUID_SIZE];
1845 u64 devid;
1846 int ret;
1847
1848 path = btrfs_alloc_path();
1849 if (!path)
1850 return -ENOMEM;
1851
1852 root = root->fs_info->chunk_root;
1853 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1854 key.offset = 0;
1855 key.type = BTRFS_DEV_ITEM_KEY;
1856
1857 while (1) {
1858 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1859 if (ret < 0)
1860 goto error;
1861
1862 leaf = path->nodes[0];
1863 next_slot:
1864 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1865 ret = btrfs_next_leaf(root, path);
1866 if (ret > 0)
1867 break;
1868 if (ret < 0)
1869 goto error;
1870 leaf = path->nodes[0];
1871 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1872 btrfs_release_path(path);
1873 continue;
1874 }
1875
1876 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1877 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1878 key.type != BTRFS_DEV_ITEM_KEY)
1879 break;
1880
1881 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1882 struct btrfs_dev_item);
1883 devid = btrfs_device_id(leaf, dev_item);
1884 read_extent_buffer(leaf, dev_uuid,
1885 (unsigned long)btrfs_device_uuid(dev_item),
1886 BTRFS_UUID_SIZE);
1887 read_extent_buffer(leaf, fs_uuid,
1888 (unsigned long)btrfs_device_fsid(dev_item),
1889 BTRFS_UUID_SIZE);
1890 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1891 fs_uuid);
1892 BUG_ON(!device); /* Logic error */
1893
1894 if (device->fs_devices->seeding) {
1895 btrfs_set_device_generation(leaf, dev_item,
1896 device->generation);
1897 btrfs_mark_buffer_dirty(leaf);
1898 }
1899
1900 path->slots[0]++;
1901 goto next_slot;
1902 }
1903 ret = 0;
1904 error:
1905 btrfs_free_path(path);
1906 return ret;
1907 }
1908
1909 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1910 {
1911 struct request_queue *q;
1912 struct btrfs_trans_handle *trans;
1913 struct btrfs_device *device;
1914 struct block_device *bdev;
1915 struct list_head *devices;
1916 struct super_block *sb = root->fs_info->sb;
1917 struct rcu_string *name;
1918 u64 total_bytes;
1919 int seeding_dev = 0;
1920 int ret = 0;
1921
1922 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1923 return -EROFS;
1924
1925 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1926 root->fs_info->bdev_holder);
1927 if (IS_ERR(bdev))
1928 return PTR_ERR(bdev);
1929
1930 if (root->fs_info->fs_devices->seeding) {
1931 seeding_dev = 1;
1932 down_write(&sb->s_umount);
1933 mutex_lock(&uuid_mutex);
1934 }
1935
1936 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1937
1938 devices = &root->fs_info->fs_devices->devices;
1939
1940 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1941 list_for_each_entry(device, devices, dev_list) {
1942 if (device->bdev == bdev) {
1943 ret = -EEXIST;
1944 mutex_unlock(
1945 &root->fs_info->fs_devices->device_list_mutex);
1946 goto error;
1947 }
1948 }
1949 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1950
1951 device = kzalloc(sizeof(*device), GFP_NOFS);
1952 if (!device) {
1953 /* we can safely leave the fs_devices entry around */
1954 ret = -ENOMEM;
1955 goto error;
1956 }
1957
1958 name = rcu_string_strdup(device_path, GFP_NOFS);
1959 if (!name) {
1960 kfree(device);
1961 ret = -ENOMEM;
1962 goto error;
1963 }
1964 rcu_assign_pointer(device->name, name);
1965
1966 ret = find_next_devid(root, &device->devid);
1967 if (ret) {
1968 rcu_string_free(device->name);
1969 kfree(device);
1970 goto error;
1971 }
1972
1973 trans = btrfs_start_transaction(root, 0);
1974 if (IS_ERR(trans)) {
1975 rcu_string_free(device->name);
1976 kfree(device);
1977 ret = PTR_ERR(trans);
1978 goto error;
1979 }
1980
1981 lock_chunks(root);
1982
1983 q = bdev_get_queue(bdev);
1984 if (blk_queue_discard(q))
1985 device->can_discard = 1;
1986 device->writeable = 1;
1987 device->work.func = pending_bios_fn;
1988 generate_random_uuid(device->uuid);
1989 spin_lock_init(&device->io_lock);
1990 device->generation = trans->transid;
1991 device->io_width = root->sectorsize;
1992 device->io_align = root->sectorsize;
1993 device->sector_size = root->sectorsize;
1994 device->total_bytes = i_size_read(bdev->bd_inode);
1995 device->disk_total_bytes = device->total_bytes;
1996 device->dev_root = root->fs_info->dev_root;
1997 device->bdev = bdev;
1998 device->in_fs_metadata = 1;
1999 device->is_tgtdev_for_dev_replace = 0;
2000 device->mode = FMODE_EXCL;
2001 set_blocksize(device->bdev, 4096);
2002
2003 if (seeding_dev) {
2004 sb->s_flags &= ~MS_RDONLY;
2005 ret = btrfs_prepare_sprout(root);
2006 BUG_ON(ret); /* -ENOMEM */
2007 }
2008
2009 device->fs_devices = root->fs_info->fs_devices;
2010
2011 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2012 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2013 list_add(&device->dev_alloc_list,
2014 &root->fs_info->fs_devices->alloc_list);
2015 root->fs_info->fs_devices->num_devices++;
2016 root->fs_info->fs_devices->open_devices++;
2017 root->fs_info->fs_devices->rw_devices++;
2018 root->fs_info->fs_devices->total_devices++;
2019 if (device->can_discard)
2020 root->fs_info->fs_devices->num_can_discard++;
2021 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2022
2023 spin_lock(&root->fs_info->free_chunk_lock);
2024 root->fs_info->free_chunk_space += device->total_bytes;
2025 spin_unlock(&root->fs_info->free_chunk_lock);
2026
2027 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2028 root->fs_info->fs_devices->rotating = 1;
2029
2030 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2031 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2032 total_bytes + device->total_bytes);
2033
2034 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2035 btrfs_set_super_num_devices(root->fs_info->super_copy,
2036 total_bytes + 1);
2037 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2038
2039 if (seeding_dev) {
2040 ret = init_first_rw_device(trans, root, device);
2041 if (ret) {
2042 btrfs_abort_transaction(trans, root, ret);
2043 goto error_trans;
2044 }
2045 ret = btrfs_finish_sprout(trans, root);
2046 if (ret) {
2047 btrfs_abort_transaction(trans, root, ret);
2048 goto error_trans;
2049 }
2050 } else {
2051 ret = btrfs_add_device(trans, root, device);
2052 if (ret) {
2053 btrfs_abort_transaction(trans, root, ret);
2054 goto error_trans;
2055 }
2056 }
2057
2058 /*
2059 * we've got more storage, clear any full flags on the space
2060 * infos
2061 */
2062 btrfs_clear_space_info_full(root->fs_info);
2063
2064 unlock_chunks(root);
2065 root->fs_info->num_tolerated_disk_barrier_failures =
2066 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2067 ret = btrfs_commit_transaction(trans, root);
2068
2069 if (seeding_dev) {
2070 mutex_unlock(&uuid_mutex);
2071 up_write(&sb->s_umount);
2072
2073 if (ret) /* transaction commit */
2074 return ret;
2075
2076 ret = btrfs_relocate_sys_chunks(root);
2077 if (ret < 0)
2078 btrfs_error(root->fs_info, ret,
2079 "Failed to relocate sys chunks after "
2080 "device initialization. This can be fixed "
2081 "using the \"btrfs balance\" command.");
2082 trans = btrfs_attach_transaction(root);
2083 if (IS_ERR(trans)) {
2084 if (PTR_ERR(trans) == -ENOENT)
2085 return 0;
2086 return PTR_ERR(trans);
2087 }
2088 ret = btrfs_commit_transaction(trans, root);
2089 }
2090
2091 return ret;
2092
2093 error_trans:
2094 unlock_chunks(root);
2095 btrfs_end_transaction(trans, root);
2096 rcu_string_free(device->name);
2097 kfree(device);
2098 error:
2099 blkdev_put(bdev, FMODE_EXCL);
2100 if (seeding_dev) {
2101 mutex_unlock(&uuid_mutex);
2102 up_write(&sb->s_umount);
2103 }
2104 return ret;
2105 }
2106
2107 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2108 struct btrfs_device **device_out)
2109 {
2110 struct request_queue *q;
2111 struct btrfs_device *device;
2112 struct block_device *bdev;
2113 struct btrfs_fs_info *fs_info = root->fs_info;
2114 struct list_head *devices;
2115 struct rcu_string *name;
2116 int ret = 0;
2117
2118 *device_out = NULL;
2119 if (fs_info->fs_devices->seeding)
2120 return -EINVAL;
2121
2122 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2123 fs_info->bdev_holder);
2124 if (IS_ERR(bdev))
2125 return PTR_ERR(bdev);
2126
2127 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2128
2129 devices = &fs_info->fs_devices->devices;
2130 list_for_each_entry(device, devices, dev_list) {
2131 if (device->bdev == bdev) {
2132 ret = -EEXIST;
2133 goto error;
2134 }
2135 }
2136
2137 device = kzalloc(sizeof(*device), GFP_NOFS);
2138 if (!device) {
2139 ret = -ENOMEM;
2140 goto error;
2141 }
2142
2143 name = rcu_string_strdup(device_path, GFP_NOFS);
2144 if (!name) {
2145 kfree(device);
2146 ret = -ENOMEM;
2147 goto error;
2148 }
2149 rcu_assign_pointer(device->name, name);
2150
2151 q = bdev_get_queue(bdev);
2152 if (blk_queue_discard(q))
2153 device->can_discard = 1;
2154 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2155 device->writeable = 1;
2156 device->work.func = pending_bios_fn;
2157 generate_random_uuid(device->uuid);
2158 device->devid = BTRFS_DEV_REPLACE_DEVID;
2159 spin_lock_init(&device->io_lock);
2160 device->generation = 0;
2161 device->io_width = root->sectorsize;
2162 device->io_align = root->sectorsize;
2163 device->sector_size = root->sectorsize;
2164 device->total_bytes = i_size_read(bdev->bd_inode);
2165 device->disk_total_bytes = device->total_bytes;
2166 device->dev_root = fs_info->dev_root;
2167 device->bdev = bdev;
2168 device->in_fs_metadata = 1;
2169 device->is_tgtdev_for_dev_replace = 1;
2170 device->mode = FMODE_EXCL;
2171 set_blocksize(device->bdev, 4096);
2172 device->fs_devices = fs_info->fs_devices;
2173 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2174 fs_info->fs_devices->num_devices++;
2175 fs_info->fs_devices->open_devices++;
2176 if (device->can_discard)
2177 fs_info->fs_devices->num_can_discard++;
2178 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2179
2180 *device_out = device;
2181 return ret;
2182
2183 error:
2184 blkdev_put(bdev, FMODE_EXCL);
2185 return ret;
2186 }
2187
2188 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2189 struct btrfs_device *tgtdev)
2190 {
2191 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2192 tgtdev->io_width = fs_info->dev_root->sectorsize;
2193 tgtdev->io_align = fs_info->dev_root->sectorsize;
2194 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2195 tgtdev->dev_root = fs_info->dev_root;
2196 tgtdev->in_fs_metadata = 1;
2197 }
2198
2199 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2200 struct btrfs_device *device)
2201 {
2202 int ret;
2203 struct btrfs_path *path;
2204 struct btrfs_root *root;
2205 struct btrfs_dev_item *dev_item;
2206 struct extent_buffer *leaf;
2207 struct btrfs_key key;
2208
2209 root = device->dev_root->fs_info->chunk_root;
2210
2211 path = btrfs_alloc_path();
2212 if (!path)
2213 return -ENOMEM;
2214
2215 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2216 key.type = BTRFS_DEV_ITEM_KEY;
2217 key.offset = device->devid;
2218
2219 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2220 if (ret < 0)
2221 goto out;
2222
2223 if (ret > 0) {
2224 ret = -ENOENT;
2225 goto out;
2226 }
2227
2228 leaf = path->nodes[0];
2229 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2230
2231 btrfs_set_device_id(leaf, dev_item, device->devid);
2232 btrfs_set_device_type(leaf, dev_item, device->type);
2233 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2234 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2235 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2236 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2237 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2238 btrfs_mark_buffer_dirty(leaf);
2239
2240 out:
2241 btrfs_free_path(path);
2242 return ret;
2243 }
2244
2245 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2246 struct btrfs_device *device, u64 new_size)
2247 {
2248 struct btrfs_super_block *super_copy =
2249 device->dev_root->fs_info->super_copy;
2250 u64 old_total = btrfs_super_total_bytes(super_copy);
2251 u64 diff = new_size - device->total_bytes;
2252
2253 if (!device->writeable)
2254 return -EACCES;
2255 if (new_size <= device->total_bytes ||
2256 device->is_tgtdev_for_dev_replace)
2257 return -EINVAL;
2258
2259 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2260 device->fs_devices->total_rw_bytes += diff;
2261
2262 device->total_bytes = new_size;
2263 device->disk_total_bytes = new_size;
2264 btrfs_clear_space_info_full(device->dev_root->fs_info);
2265
2266 return btrfs_update_device(trans, device);
2267 }
2268
2269 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2270 struct btrfs_device *device, u64 new_size)
2271 {
2272 int ret;
2273 lock_chunks(device->dev_root);
2274 ret = __btrfs_grow_device(trans, device, new_size);
2275 unlock_chunks(device->dev_root);
2276 return ret;
2277 }
2278
2279 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2280 struct btrfs_root *root,
2281 u64 chunk_tree, u64 chunk_objectid,
2282 u64 chunk_offset)
2283 {
2284 int ret;
2285 struct btrfs_path *path;
2286 struct btrfs_key key;
2287
2288 root = root->fs_info->chunk_root;
2289 path = btrfs_alloc_path();
2290 if (!path)
2291 return -ENOMEM;
2292
2293 key.objectid = chunk_objectid;
2294 key.offset = chunk_offset;
2295 key.type = BTRFS_CHUNK_ITEM_KEY;
2296
2297 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2298 if (ret < 0)
2299 goto out;
2300 else if (ret > 0) { /* Logic error or corruption */
2301 btrfs_error(root->fs_info, -ENOENT,
2302 "Failed lookup while freeing chunk.");
2303 ret = -ENOENT;
2304 goto out;
2305 }
2306
2307 ret = btrfs_del_item(trans, root, path);
2308 if (ret < 0)
2309 btrfs_error(root->fs_info, ret,
2310 "Failed to delete chunk item.");
2311 out:
2312 btrfs_free_path(path);
2313 return ret;
2314 }
2315
2316 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2317 chunk_offset)
2318 {
2319 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2320 struct btrfs_disk_key *disk_key;
2321 struct btrfs_chunk *chunk;
2322 u8 *ptr;
2323 int ret = 0;
2324 u32 num_stripes;
2325 u32 array_size;
2326 u32 len = 0;
2327 u32 cur;
2328 struct btrfs_key key;
2329
2330 array_size = btrfs_super_sys_array_size(super_copy);
2331
2332 ptr = super_copy->sys_chunk_array;
2333 cur = 0;
2334
2335 while (cur < array_size) {
2336 disk_key = (struct btrfs_disk_key *)ptr;
2337 btrfs_disk_key_to_cpu(&key, disk_key);
2338
2339 len = sizeof(*disk_key);
2340
2341 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2342 chunk = (struct btrfs_chunk *)(ptr + len);
2343 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2344 len += btrfs_chunk_item_size(num_stripes);
2345 } else {
2346 ret = -EIO;
2347 break;
2348 }
2349 if (key.objectid == chunk_objectid &&
2350 key.offset == chunk_offset) {
2351 memmove(ptr, ptr + len, array_size - (cur + len));
2352 array_size -= len;
2353 btrfs_set_super_sys_array_size(super_copy, array_size);
2354 } else {
2355 ptr += len;
2356 cur += len;
2357 }
2358 }
2359 return ret;
2360 }
2361
2362 static int btrfs_relocate_chunk(struct btrfs_root *root,
2363 u64 chunk_tree, u64 chunk_objectid,
2364 u64 chunk_offset)
2365 {
2366 struct extent_map_tree *em_tree;
2367 struct btrfs_root *extent_root;
2368 struct btrfs_trans_handle *trans;
2369 struct extent_map *em;
2370 struct map_lookup *map;
2371 int ret;
2372 int i;
2373
2374 root = root->fs_info->chunk_root;
2375 extent_root = root->fs_info->extent_root;
2376 em_tree = &root->fs_info->mapping_tree.map_tree;
2377
2378 ret = btrfs_can_relocate(extent_root, chunk_offset);
2379 if (ret)
2380 return -ENOSPC;
2381
2382 /* step one, relocate all the extents inside this chunk */
2383 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2384 if (ret)
2385 return ret;
2386
2387 trans = btrfs_start_transaction(root, 0);
2388 if (IS_ERR(trans)) {
2389 ret = PTR_ERR(trans);
2390 btrfs_std_error(root->fs_info, ret);
2391 return ret;
2392 }
2393
2394 lock_chunks(root);
2395
2396 /*
2397 * step two, delete the device extents and the
2398 * chunk tree entries
2399 */
2400 read_lock(&em_tree->lock);
2401 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2402 read_unlock(&em_tree->lock);
2403
2404 BUG_ON(!em || em->start > chunk_offset ||
2405 em->start + em->len < chunk_offset);
2406 map = (struct map_lookup *)em->bdev;
2407
2408 for (i = 0; i < map->num_stripes; i++) {
2409 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2410 map->stripes[i].physical);
2411 BUG_ON(ret);
2412
2413 if (map->stripes[i].dev) {
2414 ret = btrfs_update_device(trans, map->stripes[i].dev);
2415 BUG_ON(ret);
2416 }
2417 }
2418 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2419 chunk_offset);
2420
2421 BUG_ON(ret);
2422
2423 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2424
2425 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2426 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2427 BUG_ON(ret);
2428 }
2429
2430 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2431 BUG_ON(ret);
2432
2433 write_lock(&em_tree->lock);
2434 remove_extent_mapping(em_tree, em);
2435 write_unlock(&em_tree->lock);
2436
2437 kfree(map);
2438 em->bdev = NULL;
2439
2440 /* once for the tree */
2441 free_extent_map(em);
2442 /* once for us */
2443 free_extent_map(em);
2444
2445 unlock_chunks(root);
2446 btrfs_end_transaction(trans, root);
2447 return 0;
2448 }
2449
2450 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2451 {
2452 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2453 struct btrfs_path *path;
2454 struct extent_buffer *leaf;
2455 struct btrfs_chunk *chunk;
2456 struct btrfs_key key;
2457 struct btrfs_key found_key;
2458 u64 chunk_tree = chunk_root->root_key.objectid;
2459 u64 chunk_type;
2460 bool retried = false;
2461 int failed = 0;
2462 int ret;
2463
2464 path = btrfs_alloc_path();
2465 if (!path)
2466 return -ENOMEM;
2467
2468 again:
2469 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2470 key.offset = (u64)-1;
2471 key.type = BTRFS_CHUNK_ITEM_KEY;
2472
2473 while (1) {
2474 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2475 if (ret < 0)
2476 goto error;
2477 BUG_ON(ret == 0); /* Corruption */
2478
2479 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2480 key.type);
2481 if (ret < 0)
2482 goto error;
2483 if (ret > 0)
2484 break;
2485
2486 leaf = path->nodes[0];
2487 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2488
2489 chunk = btrfs_item_ptr(leaf, path->slots[0],
2490 struct btrfs_chunk);
2491 chunk_type = btrfs_chunk_type(leaf, chunk);
2492 btrfs_release_path(path);
2493
2494 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2495 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2496 found_key.objectid,
2497 found_key.offset);
2498 if (ret == -ENOSPC)
2499 failed++;
2500 else if (ret)
2501 BUG();
2502 }
2503
2504 if (found_key.offset == 0)
2505 break;
2506 key.offset = found_key.offset - 1;
2507 }
2508 ret = 0;
2509 if (failed && !retried) {
2510 failed = 0;
2511 retried = true;
2512 goto again;
2513 } else if (failed && retried) {
2514 WARN_ON(1);
2515 ret = -ENOSPC;
2516 }
2517 error:
2518 btrfs_free_path(path);
2519 return ret;
2520 }
2521
2522 static int insert_balance_item(struct btrfs_root *root,
2523 struct btrfs_balance_control *bctl)
2524 {
2525 struct btrfs_trans_handle *trans;
2526 struct btrfs_balance_item *item;
2527 struct btrfs_disk_balance_args disk_bargs;
2528 struct btrfs_path *path;
2529 struct extent_buffer *leaf;
2530 struct btrfs_key key;
2531 int ret, err;
2532
2533 path = btrfs_alloc_path();
2534 if (!path)
2535 return -ENOMEM;
2536
2537 trans = btrfs_start_transaction(root, 0);
2538 if (IS_ERR(trans)) {
2539 btrfs_free_path(path);
2540 return PTR_ERR(trans);
2541 }
2542
2543 key.objectid = BTRFS_BALANCE_OBJECTID;
2544 key.type = BTRFS_BALANCE_ITEM_KEY;
2545 key.offset = 0;
2546
2547 ret = btrfs_insert_empty_item(trans, root, path, &key,
2548 sizeof(*item));
2549 if (ret)
2550 goto out;
2551
2552 leaf = path->nodes[0];
2553 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2554
2555 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2556
2557 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2558 btrfs_set_balance_data(leaf, item, &disk_bargs);
2559 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2560 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2561 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2562 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2563
2564 btrfs_set_balance_flags(leaf, item, bctl->flags);
2565
2566 btrfs_mark_buffer_dirty(leaf);
2567 out:
2568 btrfs_free_path(path);
2569 err = btrfs_commit_transaction(trans, root);
2570 if (err && !ret)
2571 ret = err;
2572 return ret;
2573 }
2574
2575 static int del_balance_item(struct btrfs_root *root)
2576 {
2577 struct btrfs_trans_handle *trans;
2578 struct btrfs_path *path;
2579 struct btrfs_key key;
2580 int ret, err;
2581
2582 path = btrfs_alloc_path();
2583 if (!path)
2584 return -ENOMEM;
2585
2586 trans = btrfs_start_transaction(root, 0);
2587 if (IS_ERR(trans)) {
2588 btrfs_free_path(path);
2589 return PTR_ERR(trans);
2590 }
2591
2592 key.objectid = BTRFS_BALANCE_OBJECTID;
2593 key.type = BTRFS_BALANCE_ITEM_KEY;
2594 key.offset = 0;
2595
2596 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2597 if (ret < 0)
2598 goto out;
2599 if (ret > 0) {
2600 ret = -ENOENT;
2601 goto out;
2602 }
2603
2604 ret = btrfs_del_item(trans, root, path);
2605 out:
2606 btrfs_free_path(path);
2607 err = btrfs_commit_transaction(trans, root);
2608 if (err && !ret)
2609 ret = err;
2610 return ret;
2611 }
2612
2613 /*
2614 * This is a heuristic used to reduce the number of chunks balanced on
2615 * resume after balance was interrupted.
2616 */
2617 static void update_balance_args(struct btrfs_balance_control *bctl)
2618 {
2619 /*
2620 * Turn on soft mode for chunk types that were being converted.
2621 */
2622 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2623 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2624 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2625 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2626 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2627 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2628
2629 /*
2630 * Turn on usage filter if is not already used. The idea is
2631 * that chunks that we have already balanced should be
2632 * reasonably full. Don't do it for chunks that are being
2633 * converted - that will keep us from relocating unconverted
2634 * (albeit full) chunks.
2635 */
2636 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2637 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2638 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2639 bctl->data.usage = 90;
2640 }
2641 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2642 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2643 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2644 bctl->sys.usage = 90;
2645 }
2646 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2647 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2648 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2649 bctl->meta.usage = 90;
2650 }
2651 }
2652
2653 /*
2654 * Should be called with both balance and volume mutexes held to
2655 * serialize other volume operations (add_dev/rm_dev/resize) with
2656 * restriper. Same goes for unset_balance_control.
2657 */
2658 static void set_balance_control(struct btrfs_balance_control *bctl)
2659 {
2660 struct btrfs_fs_info *fs_info = bctl->fs_info;
2661
2662 BUG_ON(fs_info->balance_ctl);
2663
2664 spin_lock(&fs_info->balance_lock);
2665 fs_info->balance_ctl = bctl;
2666 spin_unlock(&fs_info->balance_lock);
2667 }
2668
2669 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2670 {
2671 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2672
2673 BUG_ON(!fs_info->balance_ctl);
2674
2675 spin_lock(&fs_info->balance_lock);
2676 fs_info->balance_ctl = NULL;
2677 spin_unlock(&fs_info->balance_lock);
2678
2679 kfree(bctl);
2680 }
2681
2682 /*
2683 * Balance filters. Return 1 if chunk should be filtered out
2684 * (should not be balanced).
2685 */
2686 static int chunk_profiles_filter(u64 chunk_type,
2687 struct btrfs_balance_args *bargs)
2688 {
2689 chunk_type = chunk_to_extended(chunk_type) &
2690 BTRFS_EXTENDED_PROFILE_MASK;
2691
2692 if (bargs->profiles & chunk_type)
2693 return 0;
2694
2695 return 1;
2696 }
2697
2698 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2699 struct btrfs_balance_args *bargs)
2700 {
2701 struct btrfs_block_group_cache *cache;
2702 u64 chunk_used, user_thresh;
2703 int ret = 1;
2704
2705 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2706 chunk_used = btrfs_block_group_used(&cache->item);
2707
2708 if (bargs->usage == 0)
2709 user_thresh = 1;
2710 else if (bargs->usage > 100)
2711 user_thresh = cache->key.offset;
2712 else
2713 user_thresh = div_factor_fine(cache->key.offset,
2714 bargs->usage);
2715
2716 if (chunk_used < user_thresh)
2717 ret = 0;
2718
2719 btrfs_put_block_group(cache);
2720 return ret;
2721 }
2722
2723 static int chunk_devid_filter(struct extent_buffer *leaf,
2724 struct btrfs_chunk *chunk,
2725 struct btrfs_balance_args *bargs)
2726 {
2727 struct btrfs_stripe *stripe;
2728 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2729 int i;
2730
2731 for (i = 0; i < num_stripes; i++) {
2732 stripe = btrfs_stripe_nr(chunk, i);
2733 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2734 return 0;
2735 }
2736
2737 return 1;
2738 }
2739
2740 /* [pstart, pend) */
2741 static int chunk_drange_filter(struct extent_buffer *leaf,
2742 struct btrfs_chunk *chunk,
2743 u64 chunk_offset,
2744 struct btrfs_balance_args *bargs)
2745 {
2746 struct btrfs_stripe *stripe;
2747 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2748 u64 stripe_offset;
2749 u64 stripe_length;
2750 int factor;
2751 int i;
2752
2753 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2754 return 0;
2755
2756 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2757 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2758 factor = num_stripes / 2;
2759 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2760 factor = num_stripes - 1;
2761 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2762 factor = num_stripes - 2;
2763 } else {
2764 factor = num_stripes;
2765 }
2766
2767 for (i = 0; i < num_stripes; i++) {
2768 stripe = btrfs_stripe_nr(chunk, i);
2769 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2770 continue;
2771
2772 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2773 stripe_length = btrfs_chunk_length(leaf, chunk);
2774 do_div(stripe_length, factor);
2775
2776 if (stripe_offset < bargs->pend &&
2777 stripe_offset + stripe_length > bargs->pstart)
2778 return 0;
2779 }
2780
2781 return 1;
2782 }
2783
2784 /* [vstart, vend) */
2785 static int chunk_vrange_filter(struct extent_buffer *leaf,
2786 struct btrfs_chunk *chunk,
2787 u64 chunk_offset,
2788 struct btrfs_balance_args *bargs)
2789 {
2790 if (chunk_offset < bargs->vend &&
2791 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2792 /* at least part of the chunk is inside this vrange */
2793 return 0;
2794
2795 return 1;
2796 }
2797
2798 static int chunk_soft_convert_filter(u64 chunk_type,
2799 struct btrfs_balance_args *bargs)
2800 {
2801 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2802 return 0;
2803
2804 chunk_type = chunk_to_extended(chunk_type) &
2805 BTRFS_EXTENDED_PROFILE_MASK;
2806
2807 if (bargs->target == chunk_type)
2808 return 1;
2809
2810 return 0;
2811 }
2812
2813 static int should_balance_chunk(struct btrfs_root *root,
2814 struct extent_buffer *leaf,
2815 struct btrfs_chunk *chunk, u64 chunk_offset)
2816 {
2817 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2818 struct btrfs_balance_args *bargs = NULL;
2819 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2820
2821 /* type filter */
2822 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2823 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2824 return 0;
2825 }
2826
2827 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2828 bargs = &bctl->data;
2829 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2830 bargs = &bctl->sys;
2831 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2832 bargs = &bctl->meta;
2833
2834 /* profiles filter */
2835 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2836 chunk_profiles_filter(chunk_type, bargs)) {
2837 return 0;
2838 }
2839
2840 /* usage filter */
2841 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2842 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2843 return 0;
2844 }
2845
2846 /* devid filter */
2847 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2848 chunk_devid_filter(leaf, chunk, bargs)) {
2849 return 0;
2850 }
2851
2852 /* drange filter, makes sense only with devid filter */
2853 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2854 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2855 return 0;
2856 }
2857
2858 /* vrange filter */
2859 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2860 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2861 return 0;
2862 }
2863
2864 /* soft profile changing mode */
2865 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2866 chunk_soft_convert_filter(chunk_type, bargs)) {
2867 return 0;
2868 }
2869
2870 return 1;
2871 }
2872
2873 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2874 {
2875 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2876 struct btrfs_root *chunk_root = fs_info->chunk_root;
2877 struct btrfs_root *dev_root = fs_info->dev_root;
2878 struct list_head *devices;
2879 struct btrfs_device *device;
2880 u64 old_size;
2881 u64 size_to_free;
2882 struct btrfs_chunk *chunk;
2883 struct btrfs_path *path;
2884 struct btrfs_key key;
2885 struct btrfs_key found_key;
2886 struct btrfs_trans_handle *trans;
2887 struct extent_buffer *leaf;
2888 int slot;
2889 int ret;
2890 int enospc_errors = 0;
2891 bool counting = true;
2892
2893 /* step one make some room on all the devices */
2894 devices = &fs_info->fs_devices->devices;
2895 list_for_each_entry(device, devices, dev_list) {
2896 old_size = device->total_bytes;
2897 size_to_free = div_factor(old_size, 1);
2898 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2899 if (!device->writeable ||
2900 device->total_bytes - device->bytes_used > size_to_free ||
2901 device->is_tgtdev_for_dev_replace)
2902 continue;
2903
2904 ret = btrfs_shrink_device(device, old_size - size_to_free);
2905 if (ret == -ENOSPC)
2906 break;
2907 BUG_ON(ret);
2908
2909 trans = btrfs_start_transaction(dev_root, 0);
2910 BUG_ON(IS_ERR(trans));
2911
2912 ret = btrfs_grow_device(trans, device, old_size);
2913 BUG_ON(ret);
2914
2915 btrfs_end_transaction(trans, dev_root);
2916 }
2917
2918 /* step two, relocate all the chunks */
2919 path = btrfs_alloc_path();
2920 if (!path) {
2921 ret = -ENOMEM;
2922 goto error;
2923 }
2924
2925 /* zero out stat counters */
2926 spin_lock(&fs_info->balance_lock);
2927 memset(&bctl->stat, 0, sizeof(bctl->stat));
2928 spin_unlock(&fs_info->balance_lock);
2929 again:
2930 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2931 key.offset = (u64)-1;
2932 key.type = BTRFS_CHUNK_ITEM_KEY;
2933
2934 while (1) {
2935 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2936 atomic_read(&fs_info->balance_cancel_req)) {
2937 ret = -ECANCELED;
2938 goto error;
2939 }
2940
2941 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2942 if (ret < 0)
2943 goto error;
2944
2945 /*
2946 * this shouldn't happen, it means the last relocate
2947 * failed
2948 */
2949 if (ret == 0)
2950 BUG(); /* FIXME break ? */
2951
2952 ret = btrfs_previous_item(chunk_root, path, 0,
2953 BTRFS_CHUNK_ITEM_KEY);
2954 if (ret) {
2955 ret = 0;
2956 break;
2957 }
2958
2959 leaf = path->nodes[0];
2960 slot = path->slots[0];
2961 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2962
2963 if (found_key.objectid != key.objectid)
2964 break;
2965
2966 /* chunk zero is special */
2967 if (found_key.offset == 0)
2968 break;
2969
2970 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2971
2972 if (!counting) {
2973 spin_lock(&fs_info->balance_lock);
2974 bctl->stat.considered++;
2975 spin_unlock(&fs_info->balance_lock);
2976 }
2977
2978 ret = should_balance_chunk(chunk_root, leaf, chunk,
2979 found_key.offset);
2980 btrfs_release_path(path);
2981 if (!ret)
2982 goto loop;
2983
2984 if (counting) {
2985 spin_lock(&fs_info->balance_lock);
2986 bctl->stat.expected++;
2987 spin_unlock(&fs_info->balance_lock);
2988 goto loop;
2989 }
2990
2991 ret = btrfs_relocate_chunk(chunk_root,
2992 chunk_root->root_key.objectid,
2993 found_key.objectid,
2994 found_key.offset);
2995 if (ret && ret != -ENOSPC)
2996 goto error;
2997 if (ret == -ENOSPC) {
2998 enospc_errors++;
2999 } else {
3000 spin_lock(&fs_info->balance_lock);
3001 bctl->stat.completed++;
3002 spin_unlock(&fs_info->balance_lock);
3003 }
3004 loop:
3005 key.offset = found_key.offset - 1;
3006 }
3007
3008 if (counting) {
3009 btrfs_release_path(path);
3010 counting = false;
3011 goto again;
3012 }
3013 error:
3014 btrfs_free_path(path);
3015 if (enospc_errors) {
3016 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3017 enospc_errors);
3018 if (!ret)
3019 ret = -ENOSPC;
3020 }
3021
3022 return ret;
3023 }
3024
3025 /**
3026 * alloc_profile_is_valid - see if a given profile is valid and reduced
3027 * @flags: profile to validate
3028 * @extended: if true @flags is treated as an extended profile
3029 */
3030 static int alloc_profile_is_valid(u64 flags, int extended)
3031 {
3032 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3033 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3034
3035 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3036
3037 /* 1) check that all other bits are zeroed */
3038 if (flags & ~mask)
3039 return 0;
3040
3041 /* 2) see if profile is reduced */
3042 if (flags == 0)
3043 return !extended; /* "0" is valid for usual profiles */
3044
3045 /* true if exactly one bit set */
3046 return (flags & (flags - 1)) == 0;
3047 }
3048
3049 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3050 {
3051 /* cancel requested || normal exit path */
3052 return atomic_read(&fs_info->balance_cancel_req) ||
3053 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3054 atomic_read(&fs_info->balance_cancel_req) == 0);
3055 }
3056
3057 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3058 {
3059 int ret;
3060
3061 unset_balance_control(fs_info);
3062 ret = del_balance_item(fs_info->tree_root);
3063 if (ret)
3064 btrfs_std_error(fs_info, ret);
3065
3066 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3067 }
3068
3069 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3070 struct btrfs_ioctl_balance_args *bargs);
3071
3072 /*
3073 * Should be called with both balance and volume mutexes held
3074 */
3075 int btrfs_balance(struct btrfs_balance_control *bctl,
3076 struct btrfs_ioctl_balance_args *bargs)
3077 {
3078 struct btrfs_fs_info *fs_info = bctl->fs_info;
3079 u64 allowed;
3080 int mixed = 0;
3081 int ret;
3082 u64 num_devices;
3083 unsigned seq;
3084
3085 if (btrfs_fs_closing(fs_info) ||
3086 atomic_read(&fs_info->balance_pause_req) ||
3087 atomic_read(&fs_info->balance_cancel_req)) {
3088 ret = -EINVAL;
3089 goto out;
3090 }
3091
3092 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3093 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3094 mixed = 1;
3095
3096 /*
3097 * In case of mixed groups both data and meta should be picked,
3098 * and identical options should be given for both of them.
3099 */
3100 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3101 if (mixed && (bctl->flags & allowed)) {
3102 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3103 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3104 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3105 printk(KERN_ERR "btrfs: with mixed groups data and "
3106 "metadata balance options must be the same\n");
3107 ret = -EINVAL;
3108 goto out;
3109 }
3110 }
3111
3112 num_devices = fs_info->fs_devices->num_devices;
3113 btrfs_dev_replace_lock(&fs_info->dev_replace);
3114 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3115 BUG_ON(num_devices < 1);
3116 num_devices--;
3117 }
3118 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3119 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3120 if (num_devices == 1)
3121 allowed |= BTRFS_BLOCK_GROUP_DUP;
3122 else if (num_devices < 4)
3123 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3124 else
3125 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3126 BTRFS_BLOCK_GROUP_RAID10 |
3127 BTRFS_BLOCK_GROUP_RAID5 |
3128 BTRFS_BLOCK_GROUP_RAID6);
3129
3130 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3131 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3132 (bctl->data.target & ~allowed))) {
3133 printk(KERN_ERR "btrfs: unable to start balance with target "
3134 "data profile %llu\n",
3135 (unsigned long long)bctl->data.target);
3136 ret = -EINVAL;
3137 goto out;
3138 }
3139 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3140 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3141 (bctl->meta.target & ~allowed))) {
3142 printk(KERN_ERR "btrfs: unable to start balance with target "
3143 "metadata profile %llu\n",
3144 (unsigned long long)bctl->meta.target);
3145 ret = -EINVAL;
3146 goto out;
3147 }
3148 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3149 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3150 (bctl->sys.target & ~allowed))) {
3151 printk(KERN_ERR "btrfs: unable to start balance with target "
3152 "system profile %llu\n",
3153 (unsigned long long)bctl->sys.target);
3154 ret = -EINVAL;
3155 goto out;
3156 }
3157
3158 /* allow dup'ed data chunks only in mixed mode */
3159 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3160 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3161 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3162 ret = -EINVAL;
3163 goto out;
3164 }
3165
3166 /* allow to reduce meta or sys integrity only if force set */
3167 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3168 BTRFS_BLOCK_GROUP_RAID10 |
3169 BTRFS_BLOCK_GROUP_RAID5 |
3170 BTRFS_BLOCK_GROUP_RAID6;
3171 do {
3172 seq = read_seqbegin(&fs_info->profiles_lock);
3173
3174 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3175 (fs_info->avail_system_alloc_bits & allowed) &&
3176 !(bctl->sys.target & allowed)) ||
3177 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3178 (fs_info->avail_metadata_alloc_bits & allowed) &&
3179 !(bctl->meta.target & allowed))) {
3180 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3181 printk(KERN_INFO "btrfs: force reducing metadata "
3182 "integrity\n");
3183 } else {
3184 printk(KERN_ERR "btrfs: balance will reduce metadata "
3185 "integrity, use force if you want this\n");
3186 ret = -EINVAL;
3187 goto out;
3188 }
3189 }
3190 } while (read_seqretry(&fs_info->profiles_lock, seq));
3191
3192 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3193 int num_tolerated_disk_barrier_failures;
3194 u64 target = bctl->sys.target;
3195
3196 num_tolerated_disk_barrier_failures =
3197 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3198 if (num_tolerated_disk_barrier_failures > 0 &&
3199 (target &
3200 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3201 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3202 num_tolerated_disk_barrier_failures = 0;
3203 else if (num_tolerated_disk_barrier_failures > 1 &&
3204 (target &
3205 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3206 num_tolerated_disk_barrier_failures = 1;
3207
3208 fs_info->num_tolerated_disk_barrier_failures =
3209 num_tolerated_disk_barrier_failures;
3210 }
3211
3212 ret = insert_balance_item(fs_info->tree_root, bctl);
3213 if (ret && ret != -EEXIST)
3214 goto out;
3215
3216 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3217 BUG_ON(ret == -EEXIST);
3218 set_balance_control(bctl);
3219 } else {
3220 BUG_ON(ret != -EEXIST);
3221 spin_lock(&fs_info->balance_lock);
3222 update_balance_args(bctl);
3223 spin_unlock(&fs_info->balance_lock);
3224 }
3225
3226 atomic_inc(&fs_info->balance_running);
3227 mutex_unlock(&fs_info->balance_mutex);
3228
3229 ret = __btrfs_balance(fs_info);
3230
3231 mutex_lock(&fs_info->balance_mutex);
3232 atomic_dec(&fs_info->balance_running);
3233
3234 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3235 fs_info->num_tolerated_disk_barrier_failures =
3236 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3237 }
3238
3239 if (bargs) {
3240 memset(bargs, 0, sizeof(*bargs));
3241 update_ioctl_balance_args(fs_info, 0, bargs);
3242 }
3243
3244 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3245 balance_need_close(fs_info)) {
3246 __cancel_balance(fs_info);
3247 }
3248
3249 wake_up(&fs_info->balance_wait_q);
3250
3251 return ret;
3252 out:
3253 if (bctl->flags & BTRFS_BALANCE_RESUME)
3254 __cancel_balance(fs_info);
3255 else {
3256 kfree(bctl);
3257 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3258 }
3259 return ret;
3260 }
3261
3262 static int balance_kthread(void *data)
3263 {
3264 struct btrfs_fs_info *fs_info = data;
3265 int ret = 0;
3266
3267 mutex_lock(&fs_info->volume_mutex);
3268 mutex_lock(&fs_info->balance_mutex);
3269
3270 if (fs_info->balance_ctl) {
3271 printk(KERN_INFO "btrfs: continuing balance\n");
3272 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3273 }
3274
3275 mutex_unlock(&fs_info->balance_mutex);
3276 mutex_unlock(&fs_info->volume_mutex);
3277
3278 return ret;
3279 }
3280
3281 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3282 {
3283 struct task_struct *tsk;
3284
3285 spin_lock(&fs_info->balance_lock);
3286 if (!fs_info->balance_ctl) {
3287 spin_unlock(&fs_info->balance_lock);
3288 return 0;
3289 }
3290 spin_unlock(&fs_info->balance_lock);
3291
3292 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3293 printk(KERN_INFO "btrfs: force skipping balance\n");
3294 return 0;
3295 }
3296
3297 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3298 if (IS_ERR(tsk))
3299 return PTR_ERR(tsk);
3300
3301 return 0;
3302 }
3303
3304 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3305 {
3306 struct btrfs_balance_control *bctl;
3307 struct btrfs_balance_item *item;
3308 struct btrfs_disk_balance_args disk_bargs;
3309 struct btrfs_path *path;
3310 struct extent_buffer *leaf;
3311 struct btrfs_key key;
3312 int ret;
3313
3314 path = btrfs_alloc_path();
3315 if (!path)
3316 return -ENOMEM;
3317
3318 key.objectid = BTRFS_BALANCE_OBJECTID;
3319 key.type = BTRFS_BALANCE_ITEM_KEY;
3320 key.offset = 0;
3321
3322 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3323 if (ret < 0)
3324 goto out;
3325 if (ret > 0) { /* ret = -ENOENT; */
3326 ret = 0;
3327 goto out;
3328 }
3329
3330 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3331 if (!bctl) {
3332 ret = -ENOMEM;
3333 goto out;
3334 }
3335
3336 leaf = path->nodes[0];
3337 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3338
3339 bctl->fs_info = fs_info;
3340 bctl->flags = btrfs_balance_flags(leaf, item);
3341 bctl->flags |= BTRFS_BALANCE_RESUME;
3342
3343 btrfs_balance_data(leaf, item, &disk_bargs);
3344 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3345 btrfs_balance_meta(leaf, item, &disk_bargs);
3346 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3347 btrfs_balance_sys(leaf, item, &disk_bargs);
3348 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3349
3350 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3351
3352 mutex_lock(&fs_info->volume_mutex);
3353 mutex_lock(&fs_info->balance_mutex);
3354
3355 set_balance_control(bctl);
3356
3357 mutex_unlock(&fs_info->balance_mutex);
3358 mutex_unlock(&fs_info->volume_mutex);
3359 out:
3360 btrfs_free_path(path);
3361 return ret;
3362 }
3363
3364 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3365 {
3366 int ret = 0;
3367
3368 mutex_lock(&fs_info->balance_mutex);
3369 if (!fs_info->balance_ctl) {
3370 mutex_unlock(&fs_info->balance_mutex);
3371 return -ENOTCONN;
3372 }
3373
3374 if (atomic_read(&fs_info->balance_running)) {
3375 atomic_inc(&fs_info->balance_pause_req);
3376 mutex_unlock(&fs_info->balance_mutex);
3377
3378 wait_event(fs_info->balance_wait_q,
3379 atomic_read(&fs_info->balance_running) == 0);
3380
3381 mutex_lock(&fs_info->balance_mutex);
3382 /* we are good with balance_ctl ripped off from under us */
3383 BUG_ON(atomic_read(&fs_info->balance_running));
3384 atomic_dec(&fs_info->balance_pause_req);
3385 } else {
3386 ret = -ENOTCONN;
3387 }
3388
3389 mutex_unlock(&fs_info->balance_mutex);
3390 return ret;
3391 }
3392
3393 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3394 {
3395 mutex_lock(&fs_info->balance_mutex);
3396 if (!fs_info->balance_ctl) {
3397 mutex_unlock(&fs_info->balance_mutex);
3398 return -ENOTCONN;
3399 }
3400
3401 atomic_inc(&fs_info->balance_cancel_req);
3402 /*
3403 * if we are running just wait and return, balance item is
3404 * deleted in btrfs_balance in this case
3405 */
3406 if (atomic_read(&fs_info->balance_running)) {
3407 mutex_unlock(&fs_info->balance_mutex);
3408 wait_event(fs_info->balance_wait_q,
3409 atomic_read(&fs_info->balance_running) == 0);
3410 mutex_lock(&fs_info->balance_mutex);
3411 } else {
3412 /* __cancel_balance needs volume_mutex */
3413 mutex_unlock(&fs_info->balance_mutex);
3414 mutex_lock(&fs_info->volume_mutex);
3415 mutex_lock(&fs_info->balance_mutex);
3416
3417 if (fs_info->balance_ctl)
3418 __cancel_balance(fs_info);
3419
3420 mutex_unlock(&fs_info->volume_mutex);
3421 }
3422
3423 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3424 atomic_dec(&fs_info->balance_cancel_req);
3425 mutex_unlock(&fs_info->balance_mutex);
3426 return 0;
3427 }
3428
3429 /*
3430 * shrinking a device means finding all of the device extents past
3431 * the new size, and then following the back refs to the chunks.
3432 * The chunk relocation code actually frees the device extent
3433 */
3434 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3435 {
3436 struct btrfs_trans_handle *trans;
3437 struct btrfs_root *root = device->dev_root;
3438 struct btrfs_dev_extent *dev_extent = NULL;
3439 struct btrfs_path *path;
3440 u64 length;
3441 u64 chunk_tree;
3442 u64 chunk_objectid;
3443 u64 chunk_offset;
3444 int ret;
3445 int slot;
3446 int failed = 0;
3447 bool retried = false;
3448 struct extent_buffer *l;
3449 struct btrfs_key key;
3450 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3451 u64 old_total = btrfs_super_total_bytes(super_copy);
3452 u64 old_size = device->total_bytes;
3453 u64 diff = device->total_bytes - new_size;
3454
3455 if (device->is_tgtdev_for_dev_replace)
3456 return -EINVAL;
3457
3458 path = btrfs_alloc_path();
3459 if (!path)
3460 return -ENOMEM;
3461
3462 path->reada = 2;
3463
3464 lock_chunks(root);
3465
3466 device->total_bytes = new_size;
3467 if (device->writeable) {
3468 device->fs_devices->total_rw_bytes -= diff;
3469 spin_lock(&root->fs_info->free_chunk_lock);
3470 root->fs_info->free_chunk_space -= diff;
3471 spin_unlock(&root->fs_info->free_chunk_lock);
3472 }
3473 unlock_chunks(root);
3474
3475 again:
3476 key.objectid = device->devid;
3477 key.offset = (u64)-1;
3478 key.type = BTRFS_DEV_EXTENT_KEY;
3479
3480 do {
3481 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3482 if (ret < 0)
3483 goto done;
3484
3485 ret = btrfs_previous_item(root, path, 0, key.type);
3486 if (ret < 0)
3487 goto done;
3488 if (ret) {
3489 ret = 0;
3490 btrfs_release_path(path);
3491 break;
3492 }
3493
3494 l = path->nodes[0];
3495 slot = path->slots[0];
3496 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3497
3498 if (key.objectid != device->devid) {
3499 btrfs_release_path(path);
3500 break;
3501 }
3502
3503 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3504 length = btrfs_dev_extent_length(l, dev_extent);
3505
3506 if (key.offset + length <= new_size) {
3507 btrfs_release_path(path);
3508 break;
3509 }
3510
3511 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3512 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3513 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3514 btrfs_release_path(path);
3515
3516 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3517 chunk_offset);
3518 if (ret && ret != -ENOSPC)
3519 goto done;
3520 if (ret == -ENOSPC)
3521 failed++;
3522 } while (key.offset-- > 0);
3523
3524 if (failed && !retried) {
3525 failed = 0;
3526 retried = true;
3527 goto again;
3528 } else if (failed && retried) {
3529 ret = -ENOSPC;
3530 lock_chunks(root);
3531
3532 device->total_bytes = old_size;
3533 if (device->writeable)
3534 device->fs_devices->total_rw_bytes += diff;
3535 spin_lock(&root->fs_info->free_chunk_lock);
3536 root->fs_info->free_chunk_space += diff;
3537 spin_unlock(&root->fs_info->free_chunk_lock);
3538 unlock_chunks(root);
3539 goto done;
3540 }
3541
3542 /* Shrinking succeeded, else we would be at "done". */
3543 trans = btrfs_start_transaction(root, 0);
3544 if (IS_ERR(trans)) {
3545 ret = PTR_ERR(trans);
3546 goto done;
3547 }
3548
3549 lock_chunks(root);
3550
3551 device->disk_total_bytes = new_size;
3552 /* Now btrfs_update_device() will change the on-disk size. */
3553 ret = btrfs_update_device(trans, device);
3554 if (ret) {
3555 unlock_chunks(root);
3556 btrfs_end_transaction(trans, root);
3557 goto done;
3558 }
3559 WARN_ON(diff > old_total);
3560 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3561 unlock_chunks(root);
3562 btrfs_end_transaction(trans, root);
3563 done:
3564 btrfs_free_path(path);
3565 return ret;
3566 }
3567
3568 static int btrfs_add_system_chunk(struct btrfs_root *root,
3569 struct btrfs_key *key,
3570 struct btrfs_chunk *chunk, int item_size)
3571 {
3572 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3573 struct btrfs_disk_key disk_key;
3574 u32 array_size;
3575 u8 *ptr;
3576
3577 array_size = btrfs_super_sys_array_size(super_copy);
3578 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3579 return -EFBIG;
3580
3581 ptr = super_copy->sys_chunk_array + array_size;
3582 btrfs_cpu_key_to_disk(&disk_key, key);
3583 memcpy(ptr, &disk_key, sizeof(disk_key));
3584 ptr += sizeof(disk_key);
3585 memcpy(ptr, chunk, item_size);
3586 item_size += sizeof(disk_key);
3587 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3588 return 0;
3589 }
3590
3591 /*
3592 * sort the devices in descending order by max_avail, total_avail
3593 */
3594 static int btrfs_cmp_device_info(const void *a, const void *b)
3595 {
3596 const struct btrfs_device_info *di_a = a;
3597 const struct btrfs_device_info *di_b = b;
3598
3599 if (di_a->max_avail > di_b->max_avail)
3600 return -1;
3601 if (di_a->max_avail < di_b->max_avail)
3602 return 1;
3603 if (di_a->total_avail > di_b->total_avail)
3604 return -1;
3605 if (di_a->total_avail < di_b->total_avail)
3606 return 1;
3607 return 0;
3608 }
3609
3610 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3611 [BTRFS_RAID_RAID10] = {
3612 .sub_stripes = 2,
3613 .dev_stripes = 1,
3614 .devs_max = 0, /* 0 == as many as possible */
3615 .devs_min = 4,
3616 .devs_increment = 2,
3617 .ncopies = 2,
3618 },
3619 [BTRFS_RAID_RAID1] = {
3620 .sub_stripes = 1,
3621 .dev_stripes = 1,
3622 .devs_max = 2,
3623 .devs_min = 2,
3624 .devs_increment = 2,
3625 .ncopies = 2,
3626 },
3627 [BTRFS_RAID_DUP] = {
3628 .sub_stripes = 1,
3629 .dev_stripes = 2,
3630 .devs_max = 1,
3631 .devs_min = 1,
3632 .devs_increment = 1,
3633 .ncopies = 2,
3634 },
3635 [BTRFS_RAID_RAID0] = {
3636 .sub_stripes = 1,
3637 .dev_stripes = 1,
3638 .devs_max = 0,
3639 .devs_min = 2,
3640 .devs_increment = 1,
3641 .ncopies = 1,
3642 },
3643 [BTRFS_RAID_SINGLE] = {
3644 .sub_stripes = 1,
3645 .dev_stripes = 1,
3646 .devs_max = 1,
3647 .devs_min = 1,
3648 .devs_increment = 1,
3649 .ncopies = 1,
3650 },
3651 [BTRFS_RAID_RAID5] = {
3652 .sub_stripes = 1,
3653 .dev_stripes = 1,
3654 .devs_max = 0,
3655 .devs_min = 2,
3656 .devs_increment = 1,
3657 .ncopies = 2,
3658 },
3659 [BTRFS_RAID_RAID6] = {
3660 .sub_stripes = 1,
3661 .dev_stripes = 1,
3662 .devs_max = 0,
3663 .devs_min = 3,
3664 .devs_increment = 1,
3665 .ncopies = 3,
3666 },
3667 };
3668
3669 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3670 {
3671 /* TODO allow them to set a preferred stripe size */
3672 return 64 * 1024;
3673 }
3674
3675 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3676 {
3677 u64 features;
3678
3679 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3680 return;
3681
3682 features = btrfs_super_incompat_flags(info->super_copy);
3683 if (features & BTRFS_FEATURE_INCOMPAT_RAID56)
3684 return;
3685
3686 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
3687 btrfs_set_super_incompat_flags(info->super_copy, features);
3688 printk(KERN_INFO "btrfs: setting RAID5/6 feature flag\n");
3689 }
3690
3691 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3692 struct btrfs_root *extent_root,
3693 struct map_lookup **map_ret,
3694 u64 *num_bytes_out, u64 *stripe_size_out,
3695 u64 start, u64 type)
3696 {
3697 struct btrfs_fs_info *info = extent_root->fs_info;
3698 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3699 struct list_head *cur;
3700 struct map_lookup *map = NULL;
3701 struct extent_map_tree *em_tree;
3702 struct extent_map *em;
3703 struct btrfs_device_info *devices_info = NULL;
3704 u64 total_avail;
3705 int num_stripes; /* total number of stripes to allocate */
3706 int data_stripes; /* number of stripes that count for
3707 block group size */
3708 int sub_stripes; /* sub_stripes info for map */
3709 int dev_stripes; /* stripes per dev */
3710 int devs_max; /* max devs to use */
3711 int devs_min; /* min devs needed */
3712 int devs_increment; /* ndevs has to be a multiple of this */
3713 int ncopies; /* how many copies to data has */
3714 int ret;
3715 u64 max_stripe_size;
3716 u64 max_chunk_size;
3717 u64 stripe_size;
3718 u64 num_bytes;
3719 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3720 int ndevs;
3721 int i;
3722 int j;
3723 int index;
3724
3725 BUG_ON(!alloc_profile_is_valid(type, 0));
3726
3727 if (list_empty(&fs_devices->alloc_list))
3728 return -ENOSPC;
3729
3730 index = __get_raid_index(type);
3731
3732 sub_stripes = btrfs_raid_array[index].sub_stripes;
3733 dev_stripes = btrfs_raid_array[index].dev_stripes;
3734 devs_max = btrfs_raid_array[index].devs_max;
3735 devs_min = btrfs_raid_array[index].devs_min;
3736 devs_increment = btrfs_raid_array[index].devs_increment;
3737 ncopies = btrfs_raid_array[index].ncopies;
3738
3739 if (type & BTRFS_BLOCK_GROUP_DATA) {
3740 max_stripe_size = 1024 * 1024 * 1024;
3741 max_chunk_size = 10 * max_stripe_size;
3742 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3743 /* for larger filesystems, use larger metadata chunks */
3744 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3745 max_stripe_size = 1024 * 1024 * 1024;
3746 else
3747 max_stripe_size = 256 * 1024 * 1024;
3748 max_chunk_size = max_stripe_size;
3749 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3750 max_stripe_size = 32 * 1024 * 1024;
3751 max_chunk_size = 2 * max_stripe_size;
3752 } else {
3753 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3754 type);
3755 BUG_ON(1);
3756 }
3757
3758 /* we don't want a chunk larger than 10% of writeable space */
3759 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3760 max_chunk_size);
3761
3762 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3763 GFP_NOFS);
3764 if (!devices_info)
3765 return -ENOMEM;
3766
3767 cur = fs_devices->alloc_list.next;
3768
3769 /*
3770 * in the first pass through the devices list, we gather information
3771 * about the available holes on each device.
3772 */
3773 ndevs = 0;
3774 while (cur != &fs_devices->alloc_list) {
3775 struct btrfs_device *device;
3776 u64 max_avail;
3777 u64 dev_offset;
3778
3779 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3780
3781 cur = cur->next;
3782
3783 if (!device->writeable) {
3784 WARN(1, KERN_ERR
3785 "btrfs: read-only device in alloc_list\n");
3786 continue;
3787 }
3788
3789 if (!device->in_fs_metadata ||
3790 device->is_tgtdev_for_dev_replace)
3791 continue;
3792
3793 if (device->total_bytes > device->bytes_used)
3794 total_avail = device->total_bytes - device->bytes_used;
3795 else
3796 total_avail = 0;
3797
3798 /* If there is no space on this device, skip it. */
3799 if (total_avail == 0)
3800 continue;
3801
3802 ret = find_free_dev_extent(device,
3803 max_stripe_size * dev_stripes,
3804 &dev_offset, &max_avail);
3805 if (ret && ret != -ENOSPC)
3806 goto error;
3807
3808 if (ret == 0)
3809 max_avail = max_stripe_size * dev_stripes;
3810
3811 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3812 continue;
3813
3814 if (ndevs == fs_devices->rw_devices) {
3815 WARN(1, "%s: found more than %llu devices\n",
3816 __func__, fs_devices->rw_devices);
3817 break;
3818 }
3819 devices_info[ndevs].dev_offset = dev_offset;
3820 devices_info[ndevs].max_avail = max_avail;
3821 devices_info[ndevs].total_avail = total_avail;
3822 devices_info[ndevs].dev = device;
3823 ++ndevs;
3824 }
3825
3826 /*
3827 * now sort the devices by hole size / available space
3828 */
3829 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3830 btrfs_cmp_device_info, NULL);
3831
3832 /* round down to number of usable stripes */
3833 ndevs -= ndevs % devs_increment;
3834
3835 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3836 ret = -ENOSPC;
3837 goto error;
3838 }
3839
3840 if (devs_max && ndevs > devs_max)
3841 ndevs = devs_max;
3842 /*
3843 * the primary goal is to maximize the number of stripes, so use as many
3844 * devices as possible, even if the stripes are not maximum sized.
3845 */
3846 stripe_size = devices_info[ndevs-1].max_avail;
3847 num_stripes = ndevs * dev_stripes;
3848
3849 /*
3850 * this will have to be fixed for RAID1 and RAID10 over
3851 * more drives
3852 */
3853 data_stripes = num_stripes / ncopies;
3854
3855 if (type & BTRFS_BLOCK_GROUP_RAID5) {
3856 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3857 btrfs_super_stripesize(info->super_copy));
3858 data_stripes = num_stripes - 1;
3859 }
3860 if (type & BTRFS_BLOCK_GROUP_RAID6) {
3861 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3862 btrfs_super_stripesize(info->super_copy));
3863 data_stripes = num_stripes - 2;
3864 }
3865
3866 /*
3867 * Use the number of data stripes to figure out how big this chunk
3868 * is really going to be in terms of logical address space,
3869 * and compare that answer with the max chunk size
3870 */
3871 if (stripe_size * data_stripes > max_chunk_size) {
3872 u64 mask = (1ULL << 24) - 1;
3873 stripe_size = max_chunk_size;
3874 do_div(stripe_size, data_stripes);
3875
3876 /* bump the answer up to a 16MB boundary */
3877 stripe_size = (stripe_size + mask) & ~mask;
3878
3879 /* but don't go higher than the limits we found
3880 * while searching for free extents
3881 */
3882 if (stripe_size > devices_info[ndevs-1].max_avail)
3883 stripe_size = devices_info[ndevs-1].max_avail;
3884 }
3885
3886 do_div(stripe_size, dev_stripes);
3887
3888 /* align to BTRFS_STRIPE_LEN */
3889 do_div(stripe_size, raid_stripe_len);
3890 stripe_size *= raid_stripe_len;
3891
3892 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3893 if (!map) {
3894 ret = -ENOMEM;
3895 goto error;
3896 }
3897 map->num_stripes = num_stripes;
3898
3899 for (i = 0; i < ndevs; ++i) {
3900 for (j = 0; j < dev_stripes; ++j) {
3901 int s = i * dev_stripes + j;
3902 map->stripes[s].dev = devices_info[i].dev;
3903 map->stripes[s].physical = devices_info[i].dev_offset +
3904 j * stripe_size;
3905 }
3906 }
3907 map->sector_size = extent_root->sectorsize;
3908 map->stripe_len = raid_stripe_len;
3909 map->io_align = raid_stripe_len;
3910 map->io_width = raid_stripe_len;
3911 map->type = type;
3912 map->sub_stripes = sub_stripes;
3913
3914 *map_ret = map;
3915 num_bytes = stripe_size * data_stripes;
3916
3917 *stripe_size_out = stripe_size;
3918 *num_bytes_out = num_bytes;
3919
3920 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3921
3922 em = alloc_extent_map();
3923 if (!em) {
3924 ret = -ENOMEM;
3925 goto error;
3926 }
3927 em->bdev = (struct block_device *)map;
3928 em->start = start;
3929 em->len = num_bytes;
3930 em->block_start = 0;
3931 em->block_len = em->len;
3932
3933 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3934 write_lock(&em_tree->lock);
3935 ret = add_extent_mapping(em_tree, em);
3936 write_unlock(&em_tree->lock);
3937 if (ret) {
3938 free_extent_map(em);
3939 goto error;
3940 }
3941
3942 for (i = 0; i < map->num_stripes; ++i) {
3943 struct btrfs_device *device;
3944 u64 dev_offset;
3945
3946 device = map->stripes[i].dev;
3947 dev_offset = map->stripes[i].physical;
3948
3949 ret = btrfs_alloc_dev_extent(trans, device,
3950 info->chunk_root->root_key.objectid,
3951 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3952 start, dev_offset, stripe_size);
3953 if (ret)
3954 goto error_dev_extent;
3955 }
3956
3957 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3958 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3959 start, num_bytes);
3960 if (ret) {
3961 i = map->num_stripes - 1;
3962 goto error_dev_extent;
3963 }
3964
3965 free_extent_map(em);
3966 check_raid56_incompat_flag(extent_root->fs_info, type);
3967
3968 kfree(devices_info);
3969 return 0;
3970
3971 error_dev_extent:
3972 for (; i >= 0; i--) {
3973 struct btrfs_device *device;
3974 int err;
3975
3976 device = map->stripes[i].dev;
3977 err = btrfs_free_dev_extent(trans, device, start);
3978 if (err) {
3979 btrfs_abort_transaction(trans, extent_root, err);
3980 break;
3981 }
3982 }
3983 write_lock(&em_tree->lock);
3984 remove_extent_mapping(em_tree, em);
3985 write_unlock(&em_tree->lock);
3986
3987 /* One for our allocation */
3988 free_extent_map(em);
3989 /* One for the tree reference */
3990 free_extent_map(em);
3991 error:
3992 kfree(map);
3993 kfree(devices_info);
3994 return ret;
3995 }
3996
3997 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3998 struct btrfs_root *extent_root,
3999 struct map_lookup *map, u64 chunk_offset,
4000 u64 chunk_size, u64 stripe_size)
4001 {
4002 u64 dev_offset;
4003 struct btrfs_key key;
4004 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4005 struct btrfs_device *device;
4006 struct btrfs_chunk *chunk;
4007 struct btrfs_stripe *stripe;
4008 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
4009 int index = 0;
4010 int ret;
4011
4012 chunk = kzalloc(item_size, GFP_NOFS);
4013 if (!chunk)
4014 return -ENOMEM;
4015
4016 index = 0;
4017 while (index < map->num_stripes) {
4018 device = map->stripes[index].dev;
4019 device->bytes_used += stripe_size;
4020 ret = btrfs_update_device(trans, device);
4021 if (ret)
4022 goto out_free;
4023 index++;
4024 }
4025
4026 spin_lock(&extent_root->fs_info->free_chunk_lock);
4027 extent_root->fs_info->free_chunk_space -= (stripe_size *
4028 map->num_stripes);
4029 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4030
4031 index = 0;
4032 stripe = &chunk->stripe;
4033 while (index < map->num_stripes) {
4034 device = map->stripes[index].dev;
4035 dev_offset = map->stripes[index].physical;
4036
4037 btrfs_set_stack_stripe_devid(stripe, device->devid);
4038 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4039 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4040 stripe++;
4041 index++;
4042 }
4043
4044 btrfs_set_stack_chunk_length(chunk, chunk_size);
4045 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4046 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4047 btrfs_set_stack_chunk_type(chunk, map->type);
4048 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4049 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4050 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4051 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4052 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4053
4054 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4055 key.type = BTRFS_CHUNK_ITEM_KEY;
4056 key.offset = chunk_offset;
4057
4058 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4059
4060 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4061 /*
4062 * TODO: Cleanup of inserted chunk root in case of
4063 * failure.
4064 */
4065 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4066 item_size);
4067 }
4068
4069 out_free:
4070 kfree(chunk);
4071 return ret;
4072 }
4073
4074 /*
4075 * Chunk allocation falls into two parts. The first part does works
4076 * that make the new allocated chunk useable, but not do any operation
4077 * that modifies the chunk tree. The second part does the works that
4078 * require modifying the chunk tree. This division is important for the
4079 * bootstrap process of adding storage to a seed btrfs.
4080 */
4081 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4082 struct btrfs_root *extent_root, u64 type)
4083 {
4084 u64 chunk_offset;
4085 u64 chunk_size;
4086 u64 stripe_size;
4087 struct map_lookup *map;
4088 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4089 int ret;
4090
4091 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4092 &chunk_offset);
4093 if (ret)
4094 return ret;
4095
4096 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4097 &stripe_size, chunk_offset, type);
4098 if (ret)
4099 return ret;
4100
4101 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4102 chunk_size, stripe_size);
4103 if (ret)
4104 return ret;
4105 return 0;
4106 }
4107
4108 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4109 struct btrfs_root *root,
4110 struct btrfs_device *device)
4111 {
4112 u64 chunk_offset;
4113 u64 sys_chunk_offset;
4114 u64 chunk_size;
4115 u64 sys_chunk_size;
4116 u64 stripe_size;
4117 u64 sys_stripe_size;
4118 u64 alloc_profile;
4119 struct map_lookup *map;
4120 struct map_lookup *sys_map;
4121 struct btrfs_fs_info *fs_info = root->fs_info;
4122 struct btrfs_root *extent_root = fs_info->extent_root;
4123 int ret;
4124
4125 ret = find_next_chunk(fs_info->chunk_root,
4126 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4127 if (ret)
4128 return ret;
4129
4130 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4131 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4132 &stripe_size, chunk_offset, alloc_profile);
4133 if (ret)
4134 return ret;
4135
4136 sys_chunk_offset = chunk_offset + chunk_size;
4137
4138 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4139 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4140 &sys_chunk_size, &sys_stripe_size,
4141 sys_chunk_offset, alloc_profile);
4142 if (ret) {
4143 btrfs_abort_transaction(trans, root, ret);
4144 goto out;
4145 }
4146
4147 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4148 if (ret) {
4149 btrfs_abort_transaction(trans, root, ret);
4150 goto out;
4151 }
4152
4153 /*
4154 * Modifying chunk tree needs allocating new blocks from both
4155 * system block group and metadata block group. So we only can
4156 * do operations require modifying the chunk tree after both
4157 * block groups were created.
4158 */
4159 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4160 chunk_size, stripe_size);
4161 if (ret) {
4162 btrfs_abort_transaction(trans, root, ret);
4163 goto out;
4164 }
4165
4166 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4167 sys_chunk_offset, sys_chunk_size,
4168 sys_stripe_size);
4169 if (ret)
4170 btrfs_abort_transaction(trans, root, ret);
4171
4172 out:
4173
4174 return ret;
4175 }
4176
4177 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4178 {
4179 struct extent_map *em;
4180 struct map_lookup *map;
4181 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4182 int readonly = 0;
4183 int i;
4184
4185 read_lock(&map_tree->map_tree.lock);
4186 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4187 read_unlock(&map_tree->map_tree.lock);
4188 if (!em)
4189 return 1;
4190
4191 if (btrfs_test_opt(root, DEGRADED)) {
4192 free_extent_map(em);
4193 return 0;
4194 }
4195
4196 map = (struct map_lookup *)em->bdev;
4197 for (i = 0; i < map->num_stripes; i++) {
4198 if (!map->stripes[i].dev->writeable) {
4199 readonly = 1;
4200 break;
4201 }
4202 }
4203 free_extent_map(em);
4204 return readonly;
4205 }
4206
4207 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4208 {
4209 extent_map_tree_init(&tree->map_tree);
4210 }
4211
4212 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4213 {
4214 struct extent_map *em;
4215
4216 while (1) {
4217 write_lock(&tree->map_tree.lock);
4218 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4219 if (em)
4220 remove_extent_mapping(&tree->map_tree, em);
4221 write_unlock(&tree->map_tree.lock);
4222 if (!em)
4223 break;
4224 kfree(em->bdev);
4225 /* once for us */
4226 free_extent_map(em);
4227 /* once for the tree */
4228 free_extent_map(em);
4229 }
4230 }
4231
4232 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4233 {
4234 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4235 struct extent_map *em;
4236 struct map_lookup *map;
4237 struct extent_map_tree *em_tree = &map_tree->map_tree;
4238 int ret;
4239
4240 read_lock(&em_tree->lock);
4241 em = lookup_extent_mapping(em_tree, logical, len);
4242 read_unlock(&em_tree->lock);
4243 BUG_ON(!em);
4244
4245 BUG_ON(em->start > logical || em->start + em->len < logical);
4246 map = (struct map_lookup *)em->bdev;
4247 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4248 ret = map->num_stripes;
4249 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4250 ret = map->sub_stripes;
4251 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4252 ret = 2;
4253 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4254 ret = 3;
4255 else
4256 ret = 1;
4257 free_extent_map(em);
4258
4259 btrfs_dev_replace_lock(&fs_info->dev_replace);
4260 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4261 ret++;
4262 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4263
4264 return ret;
4265 }
4266
4267 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4268 struct btrfs_mapping_tree *map_tree,
4269 u64 logical)
4270 {
4271 struct extent_map *em;
4272 struct map_lookup *map;
4273 struct extent_map_tree *em_tree = &map_tree->map_tree;
4274 unsigned long len = root->sectorsize;
4275
4276 read_lock(&em_tree->lock);
4277 em = lookup_extent_mapping(em_tree, logical, len);
4278 read_unlock(&em_tree->lock);
4279 BUG_ON(!em);
4280
4281 BUG_ON(em->start > logical || em->start + em->len < logical);
4282 map = (struct map_lookup *)em->bdev;
4283 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4284 BTRFS_BLOCK_GROUP_RAID6)) {
4285 len = map->stripe_len * nr_data_stripes(map);
4286 }
4287 free_extent_map(em);
4288 return len;
4289 }
4290
4291 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4292 u64 logical, u64 len, int mirror_num)
4293 {
4294 struct extent_map *em;
4295 struct map_lookup *map;
4296 struct extent_map_tree *em_tree = &map_tree->map_tree;
4297 int ret = 0;
4298
4299 read_lock(&em_tree->lock);
4300 em = lookup_extent_mapping(em_tree, logical, len);
4301 read_unlock(&em_tree->lock);
4302 BUG_ON(!em);
4303
4304 BUG_ON(em->start > logical || em->start + em->len < logical);
4305 map = (struct map_lookup *)em->bdev;
4306 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4307 BTRFS_BLOCK_GROUP_RAID6))
4308 ret = 1;
4309 free_extent_map(em);
4310 return ret;
4311 }
4312
4313 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4314 struct map_lookup *map, int first, int num,
4315 int optimal, int dev_replace_is_ongoing)
4316 {
4317 int i;
4318 int tolerance;
4319 struct btrfs_device *srcdev;
4320
4321 if (dev_replace_is_ongoing &&
4322 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4323 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4324 srcdev = fs_info->dev_replace.srcdev;
4325 else
4326 srcdev = NULL;
4327
4328 /*
4329 * try to avoid the drive that is the source drive for a
4330 * dev-replace procedure, only choose it if no other non-missing
4331 * mirror is available
4332 */
4333 for (tolerance = 0; tolerance < 2; tolerance++) {
4334 if (map->stripes[optimal].dev->bdev &&
4335 (tolerance || map->stripes[optimal].dev != srcdev))
4336 return optimal;
4337 for (i = first; i < first + num; i++) {
4338 if (map->stripes[i].dev->bdev &&
4339 (tolerance || map->stripes[i].dev != srcdev))
4340 return i;
4341 }
4342 }
4343
4344 /* we couldn't find one that doesn't fail. Just return something
4345 * and the io error handling code will clean up eventually
4346 */
4347 return optimal;
4348 }
4349
4350 static inline int parity_smaller(u64 a, u64 b)
4351 {
4352 return a > b;
4353 }
4354
4355 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4356 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4357 {
4358 struct btrfs_bio_stripe s;
4359 int i;
4360 u64 l;
4361 int again = 1;
4362
4363 while (again) {
4364 again = 0;
4365 for (i = 0; i < bbio->num_stripes - 1; i++) {
4366 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4367 s = bbio->stripes[i];
4368 l = raid_map[i];
4369 bbio->stripes[i] = bbio->stripes[i+1];
4370 raid_map[i] = raid_map[i+1];
4371 bbio->stripes[i+1] = s;
4372 raid_map[i+1] = l;
4373 again = 1;
4374 }
4375 }
4376 }
4377 }
4378
4379 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4380 u64 logical, u64 *length,
4381 struct btrfs_bio **bbio_ret,
4382 int mirror_num, u64 **raid_map_ret)
4383 {
4384 struct extent_map *em;
4385 struct map_lookup *map;
4386 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4387 struct extent_map_tree *em_tree = &map_tree->map_tree;
4388 u64 offset;
4389 u64 stripe_offset;
4390 u64 stripe_end_offset;
4391 u64 stripe_nr;
4392 u64 stripe_nr_orig;
4393 u64 stripe_nr_end;
4394 u64 stripe_len;
4395 u64 *raid_map = NULL;
4396 int stripe_index;
4397 int i;
4398 int ret = 0;
4399 int num_stripes;
4400 int max_errors = 0;
4401 struct btrfs_bio *bbio = NULL;
4402 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4403 int dev_replace_is_ongoing = 0;
4404 int num_alloc_stripes;
4405 int patch_the_first_stripe_for_dev_replace = 0;
4406 u64 physical_to_patch_in_first_stripe = 0;
4407 u64 raid56_full_stripe_start = (u64)-1;
4408
4409 read_lock(&em_tree->lock);
4410 em = lookup_extent_mapping(em_tree, logical, *length);
4411 read_unlock(&em_tree->lock);
4412
4413 if (!em) {
4414 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4415 (unsigned long long)logical,
4416 (unsigned long long)*length);
4417 BUG();
4418 }
4419
4420 BUG_ON(em->start > logical || em->start + em->len < logical);
4421 map = (struct map_lookup *)em->bdev;
4422 offset = logical - em->start;
4423
4424 if (mirror_num > map->num_stripes)
4425 mirror_num = 0;
4426
4427 stripe_len = map->stripe_len;
4428 stripe_nr = offset;
4429 /*
4430 * stripe_nr counts the total number of stripes we have to stride
4431 * to get to this block
4432 */
4433 do_div(stripe_nr, stripe_len);
4434
4435 stripe_offset = stripe_nr * stripe_len;
4436 BUG_ON(offset < stripe_offset);
4437
4438 /* stripe_offset is the offset of this block in its stripe*/
4439 stripe_offset = offset - stripe_offset;
4440
4441 /* if we're here for raid56, we need to know the stripe aligned start */
4442 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4443 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4444 raid56_full_stripe_start = offset;
4445
4446 /* allow a write of a full stripe, but make sure we don't
4447 * allow straddling of stripes
4448 */
4449 do_div(raid56_full_stripe_start, full_stripe_len);
4450 raid56_full_stripe_start *= full_stripe_len;
4451 }
4452
4453 if (rw & REQ_DISCARD) {
4454 /* we don't discard raid56 yet */
4455 if (map->type &
4456 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4457 ret = -EOPNOTSUPP;
4458 goto out;
4459 }
4460 *length = min_t(u64, em->len - offset, *length);
4461 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4462 u64 max_len;
4463 /* For writes to RAID[56], allow a full stripeset across all disks.
4464 For other RAID types and for RAID[56] reads, just allow a single
4465 stripe (on a single disk). */
4466 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4467 (rw & REQ_WRITE)) {
4468 max_len = stripe_len * nr_data_stripes(map) -
4469 (offset - raid56_full_stripe_start);
4470 } else {
4471 /* we limit the length of each bio to what fits in a stripe */
4472 max_len = stripe_len - stripe_offset;
4473 }
4474 *length = min_t(u64, em->len - offset, max_len);
4475 } else {
4476 *length = em->len - offset;
4477 }
4478
4479 /* This is for when we're called from btrfs_merge_bio_hook() and all
4480 it cares about is the length */
4481 if (!bbio_ret)
4482 goto out;
4483
4484 btrfs_dev_replace_lock(dev_replace);
4485 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4486 if (!dev_replace_is_ongoing)
4487 btrfs_dev_replace_unlock(dev_replace);
4488
4489 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4490 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4491 dev_replace->tgtdev != NULL) {
4492 /*
4493 * in dev-replace case, for repair case (that's the only
4494 * case where the mirror is selected explicitly when
4495 * calling btrfs_map_block), blocks left of the left cursor
4496 * can also be read from the target drive.
4497 * For REQ_GET_READ_MIRRORS, the target drive is added as
4498 * the last one to the array of stripes. For READ, it also
4499 * needs to be supported using the same mirror number.
4500 * If the requested block is not left of the left cursor,
4501 * EIO is returned. This can happen because btrfs_num_copies()
4502 * returns one more in the dev-replace case.
4503 */
4504 u64 tmp_length = *length;
4505 struct btrfs_bio *tmp_bbio = NULL;
4506 int tmp_num_stripes;
4507 u64 srcdev_devid = dev_replace->srcdev->devid;
4508 int index_srcdev = 0;
4509 int found = 0;
4510 u64 physical_of_found = 0;
4511
4512 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4513 logical, &tmp_length, &tmp_bbio, 0, NULL);
4514 if (ret) {
4515 WARN_ON(tmp_bbio != NULL);
4516 goto out;
4517 }
4518
4519 tmp_num_stripes = tmp_bbio->num_stripes;
4520 if (mirror_num > tmp_num_stripes) {
4521 /*
4522 * REQ_GET_READ_MIRRORS does not contain this
4523 * mirror, that means that the requested area
4524 * is not left of the left cursor
4525 */
4526 ret = -EIO;
4527 kfree(tmp_bbio);
4528 goto out;
4529 }
4530
4531 /*
4532 * process the rest of the function using the mirror_num
4533 * of the source drive. Therefore look it up first.
4534 * At the end, patch the device pointer to the one of the
4535 * target drive.
4536 */
4537 for (i = 0; i < tmp_num_stripes; i++) {
4538 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4539 /*
4540 * In case of DUP, in order to keep it
4541 * simple, only add the mirror with the
4542 * lowest physical address
4543 */
4544 if (found &&
4545 physical_of_found <=
4546 tmp_bbio->stripes[i].physical)
4547 continue;
4548 index_srcdev = i;
4549 found = 1;
4550 physical_of_found =
4551 tmp_bbio->stripes[i].physical;
4552 }
4553 }
4554
4555 if (found) {
4556 mirror_num = index_srcdev + 1;
4557 patch_the_first_stripe_for_dev_replace = 1;
4558 physical_to_patch_in_first_stripe = physical_of_found;
4559 } else {
4560 WARN_ON(1);
4561 ret = -EIO;
4562 kfree(tmp_bbio);
4563 goto out;
4564 }
4565
4566 kfree(tmp_bbio);
4567 } else if (mirror_num > map->num_stripes) {
4568 mirror_num = 0;
4569 }
4570
4571 num_stripes = 1;
4572 stripe_index = 0;
4573 stripe_nr_orig = stripe_nr;
4574 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4575 do_div(stripe_nr_end, map->stripe_len);
4576 stripe_end_offset = stripe_nr_end * map->stripe_len -
4577 (offset + *length);
4578
4579 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4580 if (rw & REQ_DISCARD)
4581 num_stripes = min_t(u64, map->num_stripes,
4582 stripe_nr_end - stripe_nr_orig);
4583 stripe_index = do_div(stripe_nr, map->num_stripes);
4584 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4585 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4586 num_stripes = map->num_stripes;
4587 else if (mirror_num)
4588 stripe_index = mirror_num - 1;
4589 else {
4590 stripe_index = find_live_mirror(fs_info, map, 0,
4591 map->num_stripes,
4592 current->pid % map->num_stripes,
4593 dev_replace_is_ongoing);
4594 mirror_num = stripe_index + 1;
4595 }
4596
4597 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4598 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4599 num_stripes = map->num_stripes;
4600 } else if (mirror_num) {
4601 stripe_index = mirror_num - 1;
4602 } else {
4603 mirror_num = 1;
4604 }
4605
4606 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4607 int factor = map->num_stripes / map->sub_stripes;
4608
4609 stripe_index = do_div(stripe_nr, factor);
4610 stripe_index *= map->sub_stripes;
4611
4612 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4613 num_stripes = map->sub_stripes;
4614 else if (rw & REQ_DISCARD)
4615 num_stripes = min_t(u64, map->sub_stripes *
4616 (stripe_nr_end - stripe_nr_orig),
4617 map->num_stripes);
4618 else if (mirror_num)
4619 stripe_index += mirror_num - 1;
4620 else {
4621 int old_stripe_index = stripe_index;
4622 stripe_index = find_live_mirror(fs_info, map,
4623 stripe_index,
4624 map->sub_stripes, stripe_index +
4625 current->pid % map->sub_stripes,
4626 dev_replace_is_ongoing);
4627 mirror_num = stripe_index - old_stripe_index + 1;
4628 }
4629
4630 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4631 BTRFS_BLOCK_GROUP_RAID6)) {
4632 u64 tmp;
4633
4634 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4635 && raid_map_ret) {
4636 int i, rot;
4637
4638 /* push stripe_nr back to the start of the full stripe */
4639 stripe_nr = raid56_full_stripe_start;
4640 do_div(stripe_nr, stripe_len);
4641
4642 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4643
4644 /* RAID[56] write or recovery. Return all stripes */
4645 num_stripes = map->num_stripes;
4646 max_errors = nr_parity_stripes(map);
4647
4648 raid_map = kmalloc(sizeof(u64) * num_stripes,
4649 GFP_NOFS);
4650 if (!raid_map) {
4651 ret = -ENOMEM;
4652 goto out;
4653 }
4654
4655 /* Work out the disk rotation on this stripe-set */
4656 tmp = stripe_nr;
4657 rot = do_div(tmp, num_stripes);
4658
4659 /* Fill in the logical address of each stripe */
4660 tmp = stripe_nr * nr_data_stripes(map);
4661 for (i = 0; i < nr_data_stripes(map); i++)
4662 raid_map[(i+rot) % num_stripes] =
4663 em->start + (tmp + i) * map->stripe_len;
4664
4665 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4666 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4667 raid_map[(i+rot+1) % num_stripes] =
4668 RAID6_Q_STRIPE;
4669
4670 *length = map->stripe_len;
4671 stripe_index = 0;
4672 stripe_offset = 0;
4673 } else {
4674 /*
4675 * Mirror #0 or #1 means the original data block.
4676 * Mirror #2 is RAID5 parity block.
4677 * Mirror #3 is RAID6 Q block.
4678 */
4679 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4680 if (mirror_num > 1)
4681 stripe_index = nr_data_stripes(map) +
4682 mirror_num - 2;
4683
4684 /* We distribute the parity blocks across stripes */
4685 tmp = stripe_nr + stripe_index;
4686 stripe_index = do_div(tmp, map->num_stripes);
4687 }
4688 } else {
4689 /*
4690 * after this do_div call, stripe_nr is the number of stripes
4691 * on this device we have to walk to find the data, and
4692 * stripe_index is the number of our device in the stripe array
4693 */
4694 stripe_index = do_div(stripe_nr, map->num_stripes);
4695 mirror_num = stripe_index + 1;
4696 }
4697 BUG_ON(stripe_index >= map->num_stripes);
4698
4699 num_alloc_stripes = num_stripes;
4700 if (dev_replace_is_ongoing) {
4701 if (rw & (REQ_WRITE | REQ_DISCARD))
4702 num_alloc_stripes <<= 1;
4703 if (rw & REQ_GET_READ_MIRRORS)
4704 num_alloc_stripes++;
4705 }
4706 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4707 if (!bbio) {
4708 ret = -ENOMEM;
4709 goto out;
4710 }
4711 atomic_set(&bbio->error, 0);
4712
4713 if (rw & REQ_DISCARD) {
4714 int factor = 0;
4715 int sub_stripes = 0;
4716 u64 stripes_per_dev = 0;
4717 u32 remaining_stripes = 0;
4718 u32 last_stripe = 0;
4719
4720 if (map->type &
4721 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4722 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4723 sub_stripes = 1;
4724 else
4725 sub_stripes = map->sub_stripes;
4726
4727 factor = map->num_stripes / sub_stripes;
4728 stripes_per_dev = div_u64_rem(stripe_nr_end -
4729 stripe_nr_orig,
4730 factor,
4731 &remaining_stripes);
4732 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4733 last_stripe *= sub_stripes;
4734 }
4735
4736 for (i = 0; i < num_stripes; i++) {
4737 bbio->stripes[i].physical =
4738 map->stripes[stripe_index].physical +
4739 stripe_offset + stripe_nr * map->stripe_len;
4740 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4741
4742 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4743 BTRFS_BLOCK_GROUP_RAID10)) {
4744 bbio->stripes[i].length = stripes_per_dev *
4745 map->stripe_len;
4746
4747 if (i / sub_stripes < remaining_stripes)
4748 bbio->stripes[i].length +=
4749 map->stripe_len;
4750
4751 /*
4752 * Special for the first stripe and
4753 * the last stripe:
4754 *
4755 * |-------|...|-------|
4756 * |----------|
4757 * off end_off
4758 */
4759 if (i < sub_stripes)
4760 bbio->stripes[i].length -=
4761 stripe_offset;
4762
4763 if (stripe_index >= last_stripe &&
4764 stripe_index <= (last_stripe +
4765 sub_stripes - 1))
4766 bbio->stripes[i].length -=
4767 stripe_end_offset;
4768
4769 if (i == sub_stripes - 1)
4770 stripe_offset = 0;
4771 } else
4772 bbio->stripes[i].length = *length;
4773
4774 stripe_index++;
4775 if (stripe_index == map->num_stripes) {
4776 /* This could only happen for RAID0/10 */
4777 stripe_index = 0;
4778 stripe_nr++;
4779 }
4780 }
4781 } else {
4782 for (i = 0; i < num_stripes; i++) {
4783 bbio->stripes[i].physical =
4784 map->stripes[stripe_index].physical +
4785 stripe_offset +
4786 stripe_nr * map->stripe_len;
4787 bbio->stripes[i].dev =
4788 map->stripes[stripe_index].dev;
4789 stripe_index++;
4790 }
4791 }
4792
4793 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4794 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4795 BTRFS_BLOCK_GROUP_RAID10 |
4796 BTRFS_BLOCK_GROUP_RAID5 |
4797 BTRFS_BLOCK_GROUP_DUP)) {
4798 max_errors = 1;
4799 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4800 max_errors = 2;
4801 }
4802 }
4803
4804 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4805 dev_replace->tgtdev != NULL) {
4806 int index_where_to_add;
4807 u64 srcdev_devid = dev_replace->srcdev->devid;
4808
4809 /*
4810 * duplicate the write operations while the dev replace
4811 * procedure is running. Since the copying of the old disk
4812 * to the new disk takes place at run time while the
4813 * filesystem is mounted writable, the regular write
4814 * operations to the old disk have to be duplicated to go
4815 * to the new disk as well.
4816 * Note that device->missing is handled by the caller, and
4817 * that the write to the old disk is already set up in the
4818 * stripes array.
4819 */
4820 index_where_to_add = num_stripes;
4821 for (i = 0; i < num_stripes; i++) {
4822 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4823 /* write to new disk, too */
4824 struct btrfs_bio_stripe *new =
4825 bbio->stripes + index_where_to_add;
4826 struct btrfs_bio_stripe *old =
4827 bbio->stripes + i;
4828
4829 new->physical = old->physical;
4830 new->length = old->length;
4831 new->dev = dev_replace->tgtdev;
4832 index_where_to_add++;
4833 max_errors++;
4834 }
4835 }
4836 num_stripes = index_where_to_add;
4837 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4838 dev_replace->tgtdev != NULL) {
4839 u64 srcdev_devid = dev_replace->srcdev->devid;
4840 int index_srcdev = 0;
4841 int found = 0;
4842 u64 physical_of_found = 0;
4843
4844 /*
4845 * During the dev-replace procedure, the target drive can
4846 * also be used to read data in case it is needed to repair
4847 * a corrupt block elsewhere. This is possible if the
4848 * requested area is left of the left cursor. In this area,
4849 * the target drive is a full copy of the source drive.
4850 */
4851 for (i = 0; i < num_stripes; i++) {
4852 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4853 /*
4854 * In case of DUP, in order to keep it
4855 * simple, only add the mirror with the
4856 * lowest physical address
4857 */
4858 if (found &&
4859 physical_of_found <=
4860 bbio->stripes[i].physical)
4861 continue;
4862 index_srcdev = i;
4863 found = 1;
4864 physical_of_found = bbio->stripes[i].physical;
4865 }
4866 }
4867 if (found) {
4868 u64 length = map->stripe_len;
4869
4870 if (physical_of_found + length <=
4871 dev_replace->cursor_left) {
4872 struct btrfs_bio_stripe *tgtdev_stripe =
4873 bbio->stripes + num_stripes;
4874
4875 tgtdev_stripe->physical = physical_of_found;
4876 tgtdev_stripe->length =
4877 bbio->stripes[index_srcdev].length;
4878 tgtdev_stripe->dev = dev_replace->tgtdev;
4879
4880 num_stripes++;
4881 }
4882 }
4883 }
4884
4885 *bbio_ret = bbio;
4886 bbio->num_stripes = num_stripes;
4887 bbio->max_errors = max_errors;
4888 bbio->mirror_num = mirror_num;
4889
4890 /*
4891 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4892 * mirror_num == num_stripes + 1 && dev_replace target drive is
4893 * available as a mirror
4894 */
4895 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4896 WARN_ON(num_stripes > 1);
4897 bbio->stripes[0].dev = dev_replace->tgtdev;
4898 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4899 bbio->mirror_num = map->num_stripes + 1;
4900 }
4901 if (raid_map) {
4902 sort_parity_stripes(bbio, raid_map);
4903 *raid_map_ret = raid_map;
4904 }
4905 out:
4906 if (dev_replace_is_ongoing)
4907 btrfs_dev_replace_unlock(dev_replace);
4908 free_extent_map(em);
4909 return ret;
4910 }
4911
4912 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4913 u64 logical, u64 *length,
4914 struct btrfs_bio **bbio_ret, int mirror_num)
4915 {
4916 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4917 mirror_num, NULL);
4918 }
4919
4920 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4921 u64 chunk_start, u64 physical, u64 devid,
4922 u64 **logical, int *naddrs, int *stripe_len)
4923 {
4924 struct extent_map_tree *em_tree = &map_tree->map_tree;
4925 struct extent_map *em;
4926 struct map_lookup *map;
4927 u64 *buf;
4928 u64 bytenr;
4929 u64 length;
4930 u64 stripe_nr;
4931 u64 rmap_len;
4932 int i, j, nr = 0;
4933
4934 read_lock(&em_tree->lock);
4935 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4936 read_unlock(&em_tree->lock);
4937
4938 BUG_ON(!em || em->start != chunk_start);
4939 map = (struct map_lookup *)em->bdev;
4940
4941 length = em->len;
4942 rmap_len = map->stripe_len;
4943
4944 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4945 do_div(length, map->num_stripes / map->sub_stripes);
4946 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4947 do_div(length, map->num_stripes);
4948 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4949 BTRFS_BLOCK_GROUP_RAID6)) {
4950 do_div(length, nr_data_stripes(map));
4951 rmap_len = map->stripe_len * nr_data_stripes(map);
4952 }
4953
4954 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4955 BUG_ON(!buf); /* -ENOMEM */
4956
4957 for (i = 0; i < map->num_stripes; i++) {
4958 if (devid && map->stripes[i].dev->devid != devid)
4959 continue;
4960 if (map->stripes[i].physical > physical ||
4961 map->stripes[i].physical + length <= physical)
4962 continue;
4963
4964 stripe_nr = physical - map->stripes[i].physical;
4965 do_div(stripe_nr, map->stripe_len);
4966
4967 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4968 stripe_nr = stripe_nr * map->num_stripes + i;
4969 do_div(stripe_nr, map->sub_stripes);
4970 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4971 stripe_nr = stripe_nr * map->num_stripes + i;
4972 } /* else if RAID[56], multiply by nr_data_stripes().
4973 * Alternatively, just use rmap_len below instead of
4974 * map->stripe_len */
4975
4976 bytenr = chunk_start + stripe_nr * rmap_len;
4977 WARN_ON(nr >= map->num_stripes);
4978 for (j = 0; j < nr; j++) {
4979 if (buf[j] == bytenr)
4980 break;
4981 }
4982 if (j == nr) {
4983 WARN_ON(nr >= map->num_stripes);
4984 buf[nr++] = bytenr;
4985 }
4986 }
4987
4988 *logical = buf;
4989 *naddrs = nr;
4990 *stripe_len = rmap_len;
4991
4992 free_extent_map(em);
4993 return 0;
4994 }
4995
4996 static void *merge_stripe_index_into_bio_private(void *bi_private,
4997 unsigned int stripe_index)
4998 {
4999 /*
5000 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
5001 * at most 1.
5002 * The alternative solution (instead of stealing bits from the
5003 * pointer) would be to allocate an intermediate structure
5004 * that contains the old private pointer plus the stripe_index.
5005 */
5006 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
5007 BUG_ON(stripe_index > 3);
5008 return (void *)(((uintptr_t)bi_private) | stripe_index);
5009 }
5010
5011 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
5012 {
5013 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
5014 }
5015
5016 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
5017 {
5018 return (unsigned int)((uintptr_t)bi_private) & 3;
5019 }
5020
5021 static void btrfs_end_bio(struct bio *bio, int err)
5022 {
5023 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
5024 int is_orig_bio = 0;
5025
5026 if (err) {
5027 atomic_inc(&bbio->error);
5028 if (err == -EIO || err == -EREMOTEIO) {
5029 unsigned int stripe_index =
5030 extract_stripe_index_from_bio_private(
5031 bio->bi_private);
5032 struct btrfs_device *dev;
5033
5034 BUG_ON(stripe_index >= bbio->num_stripes);
5035 dev = bbio->stripes[stripe_index].dev;
5036 if (dev->bdev) {
5037 if (bio->bi_rw & WRITE)
5038 btrfs_dev_stat_inc(dev,
5039 BTRFS_DEV_STAT_WRITE_ERRS);
5040 else
5041 btrfs_dev_stat_inc(dev,
5042 BTRFS_DEV_STAT_READ_ERRS);
5043 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5044 btrfs_dev_stat_inc(dev,
5045 BTRFS_DEV_STAT_FLUSH_ERRS);
5046 btrfs_dev_stat_print_on_error(dev);
5047 }
5048 }
5049 }
5050
5051 if (bio == bbio->orig_bio)
5052 is_orig_bio = 1;
5053
5054 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5055 if (!is_orig_bio) {
5056 bio_put(bio);
5057 bio = bbio->orig_bio;
5058 }
5059 bio->bi_private = bbio->private;
5060 bio->bi_end_io = bbio->end_io;
5061 bio->bi_bdev = (struct block_device *)
5062 (unsigned long)bbio->mirror_num;
5063 /* only send an error to the higher layers if it is
5064 * beyond the tolerance of the btrfs bio
5065 */
5066 if (atomic_read(&bbio->error) > bbio->max_errors) {
5067 err = -EIO;
5068 } else {
5069 /*
5070 * this bio is actually up to date, we didn't
5071 * go over the max number of errors
5072 */
5073 set_bit(BIO_UPTODATE, &bio->bi_flags);
5074 err = 0;
5075 }
5076 kfree(bbio);
5077
5078 bio_endio(bio, err);
5079 } else if (!is_orig_bio) {
5080 bio_put(bio);
5081 }
5082 }
5083
5084 struct async_sched {
5085 struct bio *bio;
5086 int rw;
5087 struct btrfs_fs_info *info;
5088 struct btrfs_work work;
5089 };
5090
5091 /*
5092 * see run_scheduled_bios for a description of why bios are collected for
5093 * async submit.
5094 *
5095 * This will add one bio to the pending list for a device and make sure
5096 * the work struct is scheduled.
5097 */
5098 noinline void btrfs_schedule_bio(struct btrfs_root *root,
5099 struct btrfs_device *device,
5100 int rw, struct bio *bio)
5101 {
5102 int should_queue = 1;
5103 struct btrfs_pending_bios *pending_bios;
5104
5105 if (device->missing || !device->bdev) {
5106 bio_endio(bio, -EIO);
5107 return;
5108 }
5109
5110 /* don't bother with additional async steps for reads, right now */
5111 if (!(rw & REQ_WRITE)) {
5112 bio_get(bio);
5113 btrfsic_submit_bio(rw, bio);
5114 bio_put(bio);
5115 return;
5116 }
5117
5118 /*
5119 * nr_async_bios allows us to reliably return congestion to the
5120 * higher layers. Otherwise, the async bio makes it appear we have
5121 * made progress against dirty pages when we've really just put it
5122 * on a queue for later
5123 */
5124 atomic_inc(&root->fs_info->nr_async_bios);
5125 WARN_ON(bio->bi_next);
5126 bio->bi_next = NULL;
5127 bio->bi_rw |= rw;
5128
5129 spin_lock(&device->io_lock);
5130 if (bio->bi_rw & REQ_SYNC)
5131 pending_bios = &device->pending_sync_bios;
5132 else
5133 pending_bios = &device->pending_bios;
5134
5135 if (pending_bios->tail)
5136 pending_bios->tail->bi_next = bio;
5137
5138 pending_bios->tail = bio;
5139 if (!pending_bios->head)
5140 pending_bios->head = bio;
5141 if (device->running_pending)
5142 should_queue = 0;
5143
5144 spin_unlock(&device->io_lock);
5145
5146 if (should_queue)
5147 btrfs_queue_worker(&root->fs_info->submit_workers,
5148 &device->work);
5149 }
5150
5151 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5152 sector_t sector)
5153 {
5154 struct bio_vec *prev;
5155 struct request_queue *q = bdev_get_queue(bdev);
5156 unsigned short max_sectors = queue_max_sectors(q);
5157 struct bvec_merge_data bvm = {
5158 .bi_bdev = bdev,
5159 .bi_sector = sector,
5160 .bi_rw = bio->bi_rw,
5161 };
5162
5163 if (bio->bi_vcnt == 0) {
5164 WARN_ON(1);
5165 return 1;
5166 }
5167
5168 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5169 if ((bio->bi_size >> 9) > max_sectors)
5170 return 0;
5171
5172 if (!q->merge_bvec_fn)
5173 return 1;
5174
5175 bvm.bi_size = bio->bi_size - prev->bv_len;
5176 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5177 return 0;
5178 return 1;
5179 }
5180
5181 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5182 struct bio *bio, u64 physical, int dev_nr,
5183 int rw, int async)
5184 {
5185 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5186
5187 bio->bi_private = bbio;
5188 bio->bi_private = merge_stripe_index_into_bio_private(
5189 bio->bi_private, (unsigned int)dev_nr);
5190 bio->bi_end_io = btrfs_end_bio;
5191 bio->bi_sector = physical >> 9;
5192 #ifdef DEBUG
5193 {
5194 struct rcu_string *name;
5195
5196 rcu_read_lock();
5197 name = rcu_dereference(dev->name);
5198 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5199 "(%s id %llu), size=%u\n", rw,
5200 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5201 name->str, dev->devid, bio->bi_size);
5202 rcu_read_unlock();
5203 }
5204 #endif
5205 bio->bi_bdev = dev->bdev;
5206 if (async)
5207 btrfs_schedule_bio(root, dev, rw, bio);
5208 else
5209 btrfsic_submit_bio(rw, bio);
5210 }
5211
5212 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5213 struct bio *first_bio, struct btrfs_device *dev,
5214 int dev_nr, int rw, int async)
5215 {
5216 struct bio_vec *bvec = first_bio->bi_io_vec;
5217 struct bio *bio;
5218 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5219 u64 physical = bbio->stripes[dev_nr].physical;
5220
5221 again:
5222 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5223 if (!bio)
5224 return -ENOMEM;
5225
5226 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5227 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5228 bvec->bv_offset) < bvec->bv_len) {
5229 u64 len = bio->bi_size;
5230
5231 atomic_inc(&bbio->stripes_pending);
5232 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5233 rw, async);
5234 physical += len;
5235 goto again;
5236 }
5237 bvec++;
5238 }
5239
5240 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5241 return 0;
5242 }
5243
5244 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5245 {
5246 atomic_inc(&bbio->error);
5247 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5248 bio->bi_private = bbio->private;
5249 bio->bi_end_io = bbio->end_io;
5250 bio->bi_bdev = (struct block_device *)
5251 (unsigned long)bbio->mirror_num;
5252 bio->bi_sector = logical >> 9;
5253 kfree(bbio);
5254 bio_endio(bio, -EIO);
5255 }
5256 }
5257
5258 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5259 int mirror_num, int async_submit)
5260 {
5261 struct btrfs_device *dev;
5262 struct bio *first_bio = bio;
5263 u64 logical = (u64)bio->bi_sector << 9;
5264 u64 length = 0;
5265 u64 map_length;
5266 u64 *raid_map = NULL;
5267 int ret;
5268 int dev_nr = 0;
5269 int total_devs = 1;
5270 struct btrfs_bio *bbio = NULL;
5271
5272 length = bio->bi_size;
5273 map_length = length;
5274
5275 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5276 mirror_num, &raid_map);
5277 if (ret) /* -ENOMEM */
5278 return ret;
5279
5280 total_devs = bbio->num_stripes;
5281 bbio->orig_bio = first_bio;
5282 bbio->private = first_bio->bi_private;
5283 bbio->end_io = first_bio->bi_end_io;
5284 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5285
5286 if (raid_map) {
5287 /* In this case, map_length has been set to the length of
5288 a single stripe; not the whole write */
5289 if (rw & WRITE) {
5290 return raid56_parity_write(root, bio, bbio,
5291 raid_map, map_length);
5292 } else {
5293 return raid56_parity_recover(root, bio, bbio,
5294 raid_map, map_length,
5295 mirror_num);
5296 }
5297 }
5298
5299 if (map_length < length) {
5300 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
5301 "len %llu\n", (unsigned long long)logical,
5302 (unsigned long long)length,
5303 (unsigned long long)map_length);
5304 BUG();
5305 }
5306
5307 while (dev_nr < total_devs) {
5308 dev = bbio->stripes[dev_nr].dev;
5309 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5310 bbio_error(bbio, first_bio, logical);
5311 dev_nr++;
5312 continue;
5313 }
5314
5315 /*
5316 * Check and see if we're ok with this bio based on it's size
5317 * and offset with the given device.
5318 */
5319 if (!bio_size_ok(dev->bdev, first_bio,
5320 bbio->stripes[dev_nr].physical >> 9)) {
5321 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5322 dev_nr, rw, async_submit);
5323 BUG_ON(ret);
5324 dev_nr++;
5325 continue;
5326 }
5327
5328 if (dev_nr < total_devs - 1) {
5329 bio = bio_clone(first_bio, GFP_NOFS);
5330 BUG_ON(!bio); /* -ENOMEM */
5331 } else {
5332 bio = first_bio;
5333 }
5334
5335 submit_stripe_bio(root, bbio, bio,
5336 bbio->stripes[dev_nr].physical, dev_nr, rw,
5337 async_submit);
5338 dev_nr++;
5339 }
5340 return 0;
5341 }
5342
5343 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5344 u8 *uuid, u8 *fsid)
5345 {
5346 struct btrfs_device *device;
5347 struct btrfs_fs_devices *cur_devices;
5348
5349 cur_devices = fs_info->fs_devices;
5350 while (cur_devices) {
5351 if (!fsid ||
5352 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5353 device = __find_device(&cur_devices->devices,
5354 devid, uuid);
5355 if (device)
5356 return device;
5357 }
5358 cur_devices = cur_devices->seed;
5359 }
5360 return NULL;
5361 }
5362
5363 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5364 u64 devid, u8 *dev_uuid)
5365 {
5366 struct btrfs_device *device;
5367 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5368
5369 device = kzalloc(sizeof(*device), GFP_NOFS);
5370 if (!device)
5371 return NULL;
5372 list_add(&device->dev_list,
5373 &fs_devices->devices);
5374 device->dev_root = root->fs_info->dev_root;
5375 device->devid = devid;
5376 device->work.func = pending_bios_fn;
5377 device->fs_devices = fs_devices;
5378 device->missing = 1;
5379 fs_devices->num_devices++;
5380 fs_devices->missing_devices++;
5381 spin_lock_init(&device->io_lock);
5382 INIT_LIST_HEAD(&device->dev_alloc_list);
5383 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5384 return device;
5385 }
5386
5387 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5388 struct extent_buffer *leaf,
5389 struct btrfs_chunk *chunk)
5390 {
5391 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5392 struct map_lookup *map;
5393 struct extent_map *em;
5394 u64 logical;
5395 u64 length;
5396 u64 devid;
5397 u8 uuid[BTRFS_UUID_SIZE];
5398 int num_stripes;
5399 int ret;
5400 int i;
5401
5402 logical = key->offset;
5403 length = btrfs_chunk_length(leaf, chunk);
5404
5405 read_lock(&map_tree->map_tree.lock);
5406 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5407 read_unlock(&map_tree->map_tree.lock);
5408
5409 /* already mapped? */
5410 if (em && em->start <= logical && em->start + em->len > logical) {
5411 free_extent_map(em);
5412 return 0;
5413 } else if (em) {
5414 free_extent_map(em);
5415 }
5416
5417 em = alloc_extent_map();
5418 if (!em)
5419 return -ENOMEM;
5420 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5421 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5422 if (!map) {
5423 free_extent_map(em);
5424 return -ENOMEM;
5425 }
5426
5427 em->bdev = (struct block_device *)map;
5428 em->start = logical;
5429 em->len = length;
5430 em->orig_start = 0;
5431 em->block_start = 0;
5432 em->block_len = em->len;
5433
5434 map->num_stripes = num_stripes;
5435 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5436 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5437 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5438 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5439 map->type = btrfs_chunk_type(leaf, chunk);
5440 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5441 for (i = 0; i < num_stripes; i++) {
5442 map->stripes[i].physical =
5443 btrfs_stripe_offset_nr(leaf, chunk, i);
5444 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5445 read_extent_buffer(leaf, uuid, (unsigned long)
5446 btrfs_stripe_dev_uuid_nr(chunk, i),
5447 BTRFS_UUID_SIZE);
5448 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5449 uuid, NULL);
5450 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5451 kfree(map);
5452 free_extent_map(em);
5453 return -EIO;
5454 }
5455 if (!map->stripes[i].dev) {
5456 map->stripes[i].dev =
5457 add_missing_dev(root, devid, uuid);
5458 if (!map->stripes[i].dev) {
5459 kfree(map);
5460 free_extent_map(em);
5461 return -EIO;
5462 }
5463 }
5464 map->stripes[i].dev->in_fs_metadata = 1;
5465 }
5466
5467 write_lock(&map_tree->map_tree.lock);
5468 ret = add_extent_mapping(&map_tree->map_tree, em);
5469 write_unlock(&map_tree->map_tree.lock);
5470 BUG_ON(ret); /* Tree corruption */
5471 free_extent_map(em);
5472
5473 return 0;
5474 }
5475
5476 static void fill_device_from_item(struct extent_buffer *leaf,
5477 struct btrfs_dev_item *dev_item,
5478 struct btrfs_device *device)
5479 {
5480 unsigned long ptr;
5481
5482 device->devid = btrfs_device_id(leaf, dev_item);
5483 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5484 device->total_bytes = device->disk_total_bytes;
5485 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5486 device->type = btrfs_device_type(leaf, dev_item);
5487 device->io_align = btrfs_device_io_align(leaf, dev_item);
5488 device->io_width = btrfs_device_io_width(leaf, dev_item);
5489 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5490 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5491 device->is_tgtdev_for_dev_replace = 0;
5492
5493 ptr = (unsigned long)btrfs_device_uuid(dev_item);
5494 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5495 }
5496
5497 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5498 {
5499 struct btrfs_fs_devices *fs_devices;
5500 int ret;
5501
5502 BUG_ON(!mutex_is_locked(&uuid_mutex));
5503
5504 fs_devices = root->fs_info->fs_devices->seed;
5505 while (fs_devices) {
5506 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5507 ret = 0;
5508 goto out;
5509 }
5510 fs_devices = fs_devices->seed;
5511 }
5512
5513 fs_devices = find_fsid(fsid);
5514 if (!fs_devices) {
5515 ret = -ENOENT;
5516 goto out;
5517 }
5518
5519 fs_devices = clone_fs_devices(fs_devices);
5520 if (IS_ERR(fs_devices)) {
5521 ret = PTR_ERR(fs_devices);
5522 goto out;
5523 }
5524
5525 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5526 root->fs_info->bdev_holder);
5527 if (ret) {
5528 free_fs_devices(fs_devices);
5529 goto out;
5530 }
5531
5532 if (!fs_devices->seeding) {
5533 __btrfs_close_devices(fs_devices);
5534 free_fs_devices(fs_devices);
5535 ret = -EINVAL;
5536 goto out;
5537 }
5538
5539 fs_devices->seed = root->fs_info->fs_devices->seed;
5540 root->fs_info->fs_devices->seed = fs_devices;
5541 out:
5542 return ret;
5543 }
5544
5545 static int read_one_dev(struct btrfs_root *root,
5546 struct extent_buffer *leaf,
5547 struct btrfs_dev_item *dev_item)
5548 {
5549 struct btrfs_device *device;
5550 u64 devid;
5551 int ret;
5552 u8 fs_uuid[BTRFS_UUID_SIZE];
5553 u8 dev_uuid[BTRFS_UUID_SIZE];
5554
5555 devid = btrfs_device_id(leaf, dev_item);
5556 read_extent_buffer(leaf, dev_uuid,
5557 (unsigned long)btrfs_device_uuid(dev_item),
5558 BTRFS_UUID_SIZE);
5559 read_extent_buffer(leaf, fs_uuid,
5560 (unsigned long)btrfs_device_fsid(dev_item),
5561 BTRFS_UUID_SIZE);
5562
5563 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5564 ret = open_seed_devices(root, fs_uuid);
5565 if (ret && !btrfs_test_opt(root, DEGRADED))
5566 return ret;
5567 }
5568
5569 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5570 if (!device || !device->bdev) {
5571 if (!btrfs_test_opt(root, DEGRADED))
5572 return -EIO;
5573
5574 if (!device) {
5575 printk(KERN_WARNING "warning devid %llu missing\n",
5576 (unsigned long long)devid);
5577 device = add_missing_dev(root, devid, dev_uuid);
5578 if (!device)
5579 return -ENOMEM;
5580 } else if (!device->missing) {
5581 /*
5582 * this happens when a device that was properly setup
5583 * in the device info lists suddenly goes bad.
5584 * device->bdev is NULL, and so we have to set
5585 * device->missing to one here
5586 */
5587 root->fs_info->fs_devices->missing_devices++;
5588 device->missing = 1;
5589 }
5590 }
5591
5592 if (device->fs_devices != root->fs_info->fs_devices) {
5593 BUG_ON(device->writeable);
5594 if (device->generation !=
5595 btrfs_device_generation(leaf, dev_item))
5596 return -EINVAL;
5597 }
5598
5599 fill_device_from_item(leaf, dev_item, device);
5600 device->dev_root = root->fs_info->dev_root;
5601 device->in_fs_metadata = 1;
5602 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5603 device->fs_devices->total_rw_bytes += device->total_bytes;
5604 spin_lock(&root->fs_info->free_chunk_lock);
5605 root->fs_info->free_chunk_space += device->total_bytes -
5606 device->bytes_used;
5607 spin_unlock(&root->fs_info->free_chunk_lock);
5608 }
5609 ret = 0;
5610 return ret;
5611 }
5612
5613 int btrfs_read_sys_array(struct btrfs_root *root)
5614 {
5615 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5616 struct extent_buffer *sb;
5617 struct btrfs_disk_key *disk_key;
5618 struct btrfs_chunk *chunk;
5619 u8 *ptr;
5620 unsigned long sb_ptr;
5621 int ret = 0;
5622 u32 num_stripes;
5623 u32 array_size;
5624 u32 len = 0;
5625 u32 cur;
5626 struct btrfs_key key;
5627
5628 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5629 BTRFS_SUPER_INFO_SIZE);
5630 if (!sb)
5631 return -ENOMEM;
5632 btrfs_set_buffer_uptodate(sb);
5633 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5634 /*
5635 * The sb extent buffer is artifical and just used to read the system array.
5636 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5637 * pages up-to-date when the page is larger: extent does not cover the
5638 * whole page and consequently check_page_uptodate does not find all
5639 * the page's extents up-to-date (the hole beyond sb),
5640 * write_extent_buffer then triggers a WARN_ON.
5641 *
5642 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5643 * but sb spans only this function. Add an explicit SetPageUptodate call
5644 * to silence the warning eg. on PowerPC 64.
5645 */
5646 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5647 SetPageUptodate(sb->pages[0]);
5648
5649 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5650 array_size = btrfs_super_sys_array_size(super_copy);
5651
5652 ptr = super_copy->sys_chunk_array;
5653 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5654 cur = 0;
5655
5656 while (cur < array_size) {
5657 disk_key = (struct btrfs_disk_key *)ptr;
5658 btrfs_disk_key_to_cpu(&key, disk_key);
5659
5660 len = sizeof(*disk_key); ptr += len;
5661 sb_ptr += len;
5662 cur += len;
5663
5664 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5665 chunk = (struct btrfs_chunk *)sb_ptr;
5666 ret = read_one_chunk(root, &key, sb, chunk);
5667 if (ret)
5668 break;
5669 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5670 len = btrfs_chunk_item_size(num_stripes);
5671 } else {
5672 ret = -EIO;
5673 break;
5674 }
5675 ptr += len;
5676 sb_ptr += len;
5677 cur += len;
5678 }
5679 free_extent_buffer(sb);
5680 return ret;
5681 }
5682
5683 int btrfs_read_chunk_tree(struct btrfs_root *root)
5684 {
5685 struct btrfs_path *path;
5686 struct extent_buffer *leaf;
5687 struct btrfs_key key;
5688 struct btrfs_key found_key;
5689 int ret;
5690 int slot;
5691
5692 root = root->fs_info->chunk_root;
5693
5694 path = btrfs_alloc_path();
5695 if (!path)
5696 return -ENOMEM;
5697
5698 mutex_lock(&uuid_mutex);
5699 lock_chunks(root);
5700
5701 /* first we search for all of the device items, and then we
5702 * read in all of the chunk items. This way we can create chunk
5703 * mappings that reference all of the devices that are afound
5704 */
5705 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5706 key.offset = 0;
5707 key.type = 0;
5708 again:
5709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5710 if (ret < 0)
5711 goto error;
5712 while (1) {
5713 leaf = path->nodes[0];
5714 slot = path->slots[0];
5715 if (slot >= btrfs_header_nritems(leaf)) {
5716 ret = btrfs_next_leaf(root, path);
5717 if (ret == 0)
5718 continue;
5719 if (ret < 0)
5720 goto error;
5721 break;
5722 }
5723 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5724 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5725 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5726 break;
5727 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5728 struct btrfs_dev_item *dev_item;
5729 dev_item = btrfs_item_ptr(leaf, slot,
5730 struct btrfs_dev_item);
5731 ret = read_one_dev(root, leaf, dev_item);
5732 if (ret)
5733 goto error;
5734 }
5735 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5736 struct btrfs_chunk *chunk;
5737 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5738 ret = read_one_chunk(root, &found_key, leaf, chunk);
5739 if (ret)
5740 goto error;
5741 }
5742 path->slots[0]++;
5743 }
5744 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5745 key.objectid = 0;
5746 btrfs_release_path(path);
5747 goto again;
5748 }
5749 ret = 0;
5750 error:
5751 unlock_chunks(root);
5752 mutex_unlock(&uuid_mutex);
5753
5754 btrfs_free_path(path);
5755 return ret;
5756 }
5757
5758 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5759 {
5760 int i;
5761
5762 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5763 btrfs_dev_stat_reset(dev, i);
5764 }
5765
5766 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5767 {
5768 struct btrfs_key key;
5769 struct btrfs_key found_key;
5770 struct btrfs_root *dev_root = fs_info->dev_root;
5771 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5772 struct extent_buffer *eb;
5773 int slot;
5774 int ret = 0;
5775 struct btrfs_device *device;
5776 struct btrfs_path *path = NULL;
5777 int i;
5778
5779 path = btrfs_alloc_path();
5780 if (!path) {
5781 ret = -ENOMEM;
5782 goto out;
5783 }
5784
5785 mutex_lock(&fs_devices->device_list_mutex);
5786 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5787 int item_size;
5788 struct btrfs_dev_stats_item *ptr;
5789
5790 key.objectid = 0;
5791 key.type = BTRFS_DEV_STATS_KEY;
5792 key.offset = device->devid;
5793 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5794 if (ret) {
5795 __btrfs_reset_dev_stats(device);
5796 device->dev_stats_valid = 1;
5797 btrfs_release_path(path);
5798 continue;
5799 }
5800 slot = path->slots[0];
5801 eb = path->nodes[0];
5802 btrfs_item_key_to_cpu(eb, &found_key, slot);
5803 item_size = btrfs_item_size_nr(eb, slot);
5804
5805 ptr = btrfs_item_ptr(eb, slot,
5806 struct btrfs_dev_stats_item);
5807
5808 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5809 if (item_size >= (1 + i) * sizeof(__le64))
5810 btrfs_dev_stat_set(device, i,
5811 btrfs_dev_stats_value(eb, ptr, i));
5812 else
5813 btrfs_dev_stat_reset(device, i);
5814 }
5815
5816 device->dev_stats_valid = 1;
5817 btrfs_dev_stat_print_on_load(device);
5818 btrfs_release_path(path);
5819 }
5820 mutex_unlock(&fs_devices->device_list_mutex);
5821
5822 out:
5823 btrfs_free_path(path);
5824 return ret < 0 ? ret : 0;
5825 }
5826
5827 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5828 struct btrfs_root *dev_root,
5829 struct btrfs_device *device)
5830 {
5831 struct btrfs_path *path;
5832 struct btrfs_key key;
5833 struct extent_buffer *eb;
5834 struct btrfs_dev_stats_item *ptr;
5835 int ret;
5836 int i;
5837
5838 key.objectid = 0;
5839 key.type = BTRFS_DEV_STATS_KEY;
5840 key.offset = device->devid;
5841
5842 path = btrfs_alloc_path();
5843 BUG_ON(!path);
5844 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5845 if (ret < 0) {
5846 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5847 ret, rcu_str_deref(device->name));
5848 goto out;
5849 }
5850
5851 if (ret == 0 &&
5852 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5853 /* need to delete old one and insert a new one */
5854 ret = btrfs_del_item(trans, dev_root, path);
5855 if (ret != 0) {
5856 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5857 rcu_str_deref(device->name), ret);
5858 goto out;
5859 }
5860 ret = 1;
5861 }
5862
5863 if (ret == 1) {
5864 /* need to insert a new item */
5865 btrfs_release_path(path);
5866 ret = btrfs_insert_empty_item(trans, dev_root, path,
5867 &key, sizeof(*ptr));
5868 if (ret < 0) {
5869 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5870 rcu_str_deref(device->name), ret);
5871 goto out;
5872 }
5873 }
5874
5875 eb = path->nodes[0];
5876 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5877 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5878 btrfs_set_dev_stats_value(eb, ptr, i,
5879 btrfs_dev_stat_read(device, i));
5880 btrfs_mark_buffer_dirty(eb);
5881
5882 out:
5883 btrfs_free_path(path);
5884 return ret;
5885 }
5886
5887 /*
5888 * called from commit_transaction. Writes all changed device stats to disk.
5889 */
5890 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5891 struct btrfs_fs_info *fs_info)
5892 {
5893 struct btrfs_root *dev_root = fs_info->dev_root;
5894 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5895 struct btrfs_device *device;
5896 int ret = 0;
5897
5898 mutex_lock(&fs_devices->device_list_mutex);
5899 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5900 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5901 continue;
5902
5903 ret = update_dev_stat_item(trans, dev_root, device);
5904 if (!ret)
5905 device->dev_stats_dirty = 0;
5906 }
5907 mutex_unlock(&fs_devices->device_list_mutex);
5908
5909 return ret;
5910 }
5911
5912 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5913 {
5914 btrfs_dev_stat_inc(dev, index);
5915 btrfs_dev_stat_print_on_error(dev);
5916 }
5917
5918 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5919 {
5920 if (!dev->dev_stats_valid)
5921 return;
5922 printk_ratelimited_in_rcu(KERN_ERR
5923 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5924 rcu_str_deref(dev->name),
5925 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5926 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5927 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5928 btrfs_dev_stat_read(dev,
5929 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5930 btrfs_dev_stat_read(dev,
5931 BTRFS_DEV_STAT_GENERATION_ERRS));
5932 }
5933
5934 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5935 {
5936 int i;
5937
5938 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5939 if (btrfs_dev_stat_read(dev, i) != 0)
5940 break;
5941 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5942 return; /* all values == 0, suppress message */
5943
5944 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5945 rcu_str_deref(dev->name),
5946 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5947 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5948 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5949 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5950 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5951 }
5952
5953 int btrfs_get_dev_stats(struct btrfs_root *root,
5954 struct btrfs_ioctl_get_dev_stats *stats)
5955 {
5956 struct btrfs_device *dev;
5957 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5958 int i;
5959
5960 mutex_lock(&fs_devices->device_list_mutex);
5961 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5962 mutex_unlock(&fs_devices->device_list_mutex);
5963
5964 if (!dev) {
5965 printk(KERN_WARNING
5966 "btrfs: get dev_stats failed, device not found\n");
5967 return -ENODEV;
5968 } else if (!dev->dev_stats_valid) {
5969 printk(KERN_WARNING
5970 "btrfs: get dev_stats failed, not yet valid\n");
5971 return -ENODEV;
5972 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5973 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5974 if (stats->nr_items > i)
5975 stats->values[i] =
5976 btrfs_dev_stat_read_and_reset(dev, i);
5977 else
5978 btrfs_dev_stat_reset(dev, i);
5979 }
5980 } else {
5981 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5982 if (stats->nr_items > i)
5983 stats->values[i] = btrfs_dev_stat_read(dev, i);
5984 }
5985 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5986 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5987 return 0;
5988 }
5989
5990 int btrfs_scratch_superblock(struct btrfs_device *device)
5991 {
5992 struct buffer_head *bh;
5993 struct btrfs_super_block *disk_super;
5994
5995 bh = btrfs_read_dev_super(device->bdev);
5996 if (!bh)
5997 return -EINVAL;
5998 disk_super = (struct btrfs_super_block *)bh->b_data;
5999
6000 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6001 set_buffer_dirty(bh);
6002 sync_dirty_buffer(bh);
6003 brelse(bh);
6004
6005 return 0;
6006 }
CJK support for tty framebuffer vt