aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/volumes.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/volumes.c')
-rw-r--r--fs/btrfs/volumes.c3218
1 files changed, 3218 insertions, 0 deletions
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
new file mode 100644
index 000000000000..b187b537888e
--- /dev/null
+++ b/fs/btrfs/volumes.c
@@ -0,0 +1,3218 @@
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/buffer_head.h>
21#include <linux/blkdev.h>
22#include <linux/random.h>
23#include <linux/version.h>
24#include <asm/div64.h>
25#include "compat.h"
26#include "ctree.h"
27#include "extent_map.h"
28#include "disk-io.h"
29#include "transaction.h"
30#include "print-tree.h"
31#include "volumes.h"
32#include "async-thread.h"
33
34struct map_lookup {
35 u64 type;
36 int io_align;
37 int io_width;
38 int stripe_len;
39 int sector_size;
40 int num_stripes;
41 int sub_stripes;
42 struct btrfs_bio_stripe stripes[];
43};
44
45static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49
50#define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
52
53static DEFINE_MUTEX(uuid_mutex);
54static LIST_HEAD(fs_uuids);
55
56void btrfs_lock_volumes(void)
57{
58 mutex_lock(&uuid_mutex);
59}
60
61void btrfs_unlock_volumes(void)
62{
63 mutex_unlock(&uuid_mutex);
64}
65
66static void lock_chunks(struct btrfs_root *root)
67{
68 mutex_lock(&root->fs_info->chunk_mutex);
69}
70
71static void unlock_chunks(struct btrfs_root *root)
72{
73 mutex_unlock(&root->fs_info->chunk_mutex);
74}
75
76static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
77{
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
84 kfree(device->name);
85 kfree(device);
86 }
87 kfree(fs_devices);
88}
89
90int btrfs_cleanup_fs_uuids(void)
91{
92 struct btrfs_fs_devices *fs_devices;
93
94 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
98 free_fs_devices(fs_devices);
99 }
100 return 0;
101}
102
103static noinline struct btrfs_device *__find_device(struct list_head *head,
104 u64 devid, u8 *uuid)
105{
106 struct btrfs_device *dev;
107 struct list_head *cur;
108
109 list_for_each(cur, head) {
110 dev = list_entry(cur, struct btrfs_device, dev_list);
111 if (dev->devid == devid &&
112 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
113 return dev;
114 }
115 }
116 return NULL;
117}
118
119static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
120{
121 struct list_head *cur;
122 struct btrfs_fs_devices *fs_devices;
123
124 list_for_each(cur, &fs_uuids) {
125 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
126 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
127 return fs_devices;
128 }
129 return NULL;
130}
131
132/*
133 * we try to collect pending bios for a device so we don't get a large
134 * number of procs sending bios down to the same device. This greatly
135 * improves the schedulers ability to collect and merge the bios.
136 *
137 * But, it also turns into a long list of bios to process and that is sure
138 * to eventually make the worker thread block. The solution here is to
139 * make some progress and then put this work struct back at the end of
140 * the list if the block device is congested. This way, multiple devices
141 * can make progress from a single worker thread.
142 */
143static noinline int run_scheduled_bios(struct btrfs_device *device)
144{
145 struct bio *pending;
146 struct backing_dev_info *bdi;
147 struct btrfs_fs_info *fs_info;
148 struct bio *tail;
149 struct bio *cur;
150 int again = 0;
151 unsigned long num_run = 0;
152 unsigned long limit;
153
154 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
155 fs_info = device->dev_root->fs_info;
156 limit = btrfs_async_submit_limit(fs_info);
157 limit = limit * 2 / 3;
158
159loop:
160 spin_lock(&device->io_lock);
161
162 /* take all the bios off the list at once and process them
163 * later on (without the lock held). But, remember the
164 * tail and other pointers so the bios can be properly reinserted
165 * into the list if we hit congestion
166 */
167 pending = device->pending_bios;
168 tail = device->pending_bio_tail;
169 WARN_ON(pending && !tail);
170 device->pending_bios = NULL;
171 device->pending_bio_tail = NULL;
172
173 /*
174 * if pending was null this time around, no bios need processing
175 * at all and we can stop. Otherwise it'll loop back up again
176 * and do an additional check so no bios are missed.
177 *
178 * device->running_pending is used to synchronize with the
179 * schedule_bio code.
180 */
181 if (pending) {
182 again = 1;
183 device->running_pending = 1;
184 } else {
185 again = 0;
186 device->running_pending = 0;
187 }
188 spin_unlock(&device->io_lock);
189
190 while (pending) {
191 cur = pending;
192 pending = pending->bi_next;
193 cur->bi_next = NULL;
194 atomic_dec(&fs_info->nr_async_bios);
195
196 if (atomic_read(&fs_info->nr_async_bios) < limit &&
197 waitqueue_active(&fs_info->async_submit_wait))
198 wake_up(&fs_info->async_submit_wait);
199
200 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
201 bio_get(cur);
202 submit_bio(cur->bi_rw, cur);
203 bio_put(cur);
204 num_run++;
205
206 /*
207 * we made progress, there is more work to do and the bdi
208 * is now congested. Back off and let other work structs
209 * run instead
210 */
211 if (pending && bdi_write_congested(bdi) &&
212 fs_info->fs_devices->open_devices > 1) {
213 struct bio *old_head;
214
215 spin_lock(&device->io_lock);
216
217 old_head = device->pending_bios;
218 device->pending_bios = pending;
219 if (device->pending_bio_tail)
220 tail->bi_next = old_head;
221 else
222 device->pending_bio_tail = tail;
223
224 spin_unlock(&device->io_lock);
225 btrfs_requeue_work(&device->work);
226 goto done;
227 }
228 }
229 if (again)
230 goto loop;
231done:
232 return 0;
233}
234
235static void pending_bios_fn(struct btrfs_work *work)
236{
237 struct btrfs_device *device;
238
239 device = container_of(work, struct btrfs_device, work);
240 run_scheduled_bios(device);
241}
242
243static noinline int device_list_add(const char *path,
244 struct btrfs_super_block *disk_super,
245 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
246{
247 struct btrfs_device *device;
248 struct btrfs_fs_devices *fs_devices;
249 u64 found_transid = btrfs_super_generation(disk_super);
250
251 fs_devices = find_fsid(disk_super->fsid);
252 if (!fs_devices) {
253 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
254 if (!fs_devices)
255 return -ENOMEM;
256 INIT_LIST_HEAD(&fs_devices->devices);
257 INIT_LIST_HEAD(&fs_devices->alloc_list);
258 list_add(&fs_devices->list, &fs_uuids);
259 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
260 fs_devices->latest_devid = devid;
261 fs_devices->latest_trans = found_transid;
262 device = NULL;
263 } else {
264 device = __find_device(&fs_devices->devices, devid,
265 disk_super->dev_item.uuid);
266 }
267 if (!device) {
268 if (fs_devices->opened)
269 return -EBUSY;
270
271 device = kzalloc(sizeof(*device), GFP_NOFS);
272 if (!device) {
273 /* we can safely leave the fs_devices entry around */
274 return -ENOMEM;
275 }
276 device->devid = devid;
277 device->work.func = pending_bios_fn;
278 memcpy(device->uuid, disk_super->dev_item.uuid,
279 BTRFS_UUID_SIZE);
280 device->barriers = 1;
281 spin_lock_init(&device->io_lock);
282 device->name = kstrdup(path, GFP_NOFS);
283 if (!device->name) {
284 kfree(device);
285 return -ENOMEM;
286 }
287 INIT_LIST_HEAD(&device->dev_alloc_list);
288 list_add(&device->dev_list, &fs_devices->devices);
289 device->fs_devices = fs_devices;
290 fs_devices->num_devices++;
291 }
292
293 if (found_transid > fs_devices->latest_trans) {
294 fs_devices->latest_devid = devid;
295 fs_devices->latest_trans = found_transid;
296 }
297 *fs_devices_ret = fs_devices;
298 return 0;
299}
300
301static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
302{
303 struct btrfs_fs_devices *fs_devices;
304 struct btrfs_device *device;
305 struct btrfs_device *orig_dev;
306
307 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
308 if (!fs_devices)
309 return ERR_PTR(-ENOMEM);
310
311 INIT_LIST_HEAD(&fs_devices->devices);
312 INIT_LIST_HEAD(&fs_devices->alloc_list);
313 INIT_LIST_HEAD(&fs_devices->list);
314 fs_devices->latest_devid = orig->latest_devid;
315 fs_devices->latest_trans = orig->latest_trans;
316 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
317
318 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
319 device = kzalloc(sizeof(*device), GFP_NOFS);
320 if (!device)
321 goto error;
322
323 device->name = kstrdup(orig_dev->name, GFP_NOFS);
324 if (!device->name)
325 goto error;
326
327 device->devid = orig_dev->devid;
328 device->work.func = pending_bios_fn;
329 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
330 device->barriers = 1;
331 spin_lock_init(&device->io_lock);
332 INIT_LIST_HEAD(&device->dev_list);
333 INIT_LIST_HEAD(&device->dev_alloc_list);
334
335 list_add(&device->dev_list, &fs_devices->devices);
336 device->fs_devices = fs_devices;
337 fs_devices->num_devices++;
338 }
339 return fs_devices;
340error:
341 free_fs_devices(fs_devices);
342 return ERR_PTR(-ENOMEM);
343}
344
345int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
346{
347 struct list_head *tmp;
348 struct list_head *cur;
349 struct btrfs_device *device;
350
351 mutex_lock(&uuid_mutex);
352again:
353 list_for_each_safe(cur, tmp, &fs_devices->devices) {
354 device = list_entry(cur, struct btrfs_device, dev_list);
355 if (device->in_fs_metadata)
356 continue;
357
358 if (device->bdev) {
359 close_bdev_exclusive(device->bdev, device->mode);
360 device->bdev = NULL;
361 fs_devices->open_devices--;
362 }
363 if (device->writeable) {
364 list_del_init(&device->dev_alloc_list);
365 device->writeable = 0;
366 fs_devices->rw_devices--;
367 }
368 list_del_init(&device->dev_list);
369 fs_devices->num_devices--;
370 kfree(device->name);
371 kfree(device);
372 }
373
374 if (fs_devices->seed) {
375 fs_devices = fs_devices->seed;
376 goto again;
377 }
378
379 mutex_unlock(&uuid_mutex);
380 return 0;
381}
382
383static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
384{
385 struct list_head *cur;
386 struct btrfs_device *device;
387
388 if (--fs_devices->opened > 0)
389 return 0;
390
391 list_for_each(cur, &fs_devices->devices) {
392 device = list_entry(cur, struct btrfs_device, dev_list);
393 if (device->bdev) {
394 close_bdev_exclusive(device->bdev, device->mode);
395 fs_devices->open_devices--;
396 }
397 if (device->writeable) {
398 list_del_init(&device->dev_alloc_list);
399 fs_devices->rw_devices--;
400 }
401
402 device->bdev = NULL;
403 device->writeable = 0;
404 device->in_fs_metadata = 0;
405 }
406 WARN_ON(fs_devices->open_devices);
407 WARN_ON(fs_devices->rw_devices);
408 fs_devices->opened = 0;
409 fs_devices->seeding = 0;
410
411 return 0;
412}
413
414int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
415{
416 struct btrfs_fs_devices *seed_devices = NULL;
417 int ret;
418
419 mutex_lock(&uuid_mutex);
420 ret = __btrfs_close_devices(fs_devices);
421 if (!fs_devices->opened) {
422 seed_devices = fs_devices->seed;
423 fs_devices->seed = NULL;
424 }
425 mutex_unlock(&uuid_mutex);
426
427 while (seed_devices) {
428 fs_devices = seed_devices;
429 seed_devices = fs_devices->seed;
430 __btrfs_close_devices(fs_devices);
431 free_fs_devices(fs_devices);
432 }
433 return ret;
434}
435
436static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
437 fmode_t flags, void *holder)
438{
439 struct block_device *bdev;
440 struct list_head *head = &fs_devices->devices;
441 struct list_head *cur;
442 struct btrfs_device *device;
443 struct block_device *latest_bdev = NULL;
444 struct buffer_head *bh;
445 struct btrfs_super_block *disk_super;
446 u64 latest_devid = 0;
447 u64 latest_transid = 0;
448 u64 devid;
449 int seeding = 1;
450 int ret = 0;
451
452 list_for_each(cur, head) {
453 device = list_entry(cur, struct btrfs_device, dev_list);
454 if (device->bdev)
455 continue;
456 if (!device->name)
457 continue;
458
459 bdev = open_bdev_exclusive(device->name, flags, holder);
460 if (IS_ERR(bdev)) {
461 printk(KERN_INFO "open %s failed\n", device->name);
462 goto error;
463 }
464 set_blocksize(bdev, 4096);
465
466 bh = btrfs_read_dev_super(bdev);
467 if (!bh)
468 goto error_close;
469
470 disk_super = (struct btrfs_super_block *)bh->b_data;
471 devid = le64_to_cpu(disk_super->dev_item.devid);
472 if (devid != device->devid)
473 goto error_brelse;
474
475 if (memcmp(device->uuid, disk_super->dev_item.uuid,
476 BTRFS_UUID_SIZE))
477 goto error_brelse;
478
479 device->generation = btrfs_super_generation(disk_super);
480 if (!latest_transid || device->generation > latest_transid) {
481 latest_devid = devid;
482 latest_transid = device->generation;
483 latest_bdev = bdev;
484 }
485
486 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
487 device->writeable = 0;
488 } else {
489 device->writeable = !bdev_read_only(bdev);
490 seeding = 0;
491 }
492
493 device->bdev = bdev;
494 device->in_fs_metadata = 0;
495 device->mode = flags;
496
497 fs_devices->open_devices++;
498 if (device->writeable) {
499 fs_devices->rw_devices++;
500 list_add(&device->dev_alloc_list,
501 &fs_devices->alloc_list);
502 }
503 continue;
504
505error_brelse:
506 brelse(bh);
507error_close:
508 close_bdev_exclusive(bdev, FMODE_READ);
509error:
510 continue;
511 }
512 if (fs_devices->open_devices == 0) {
513 ret = -EIO;
514 goto out;
515 }
516 fs_devices->seeding = seeding;
517 fs_devices->opened = 1;
518 fs_devices->latest_bdev = latest_bdev;
519 fs_devices->latest_devid = latest_devid;
520 fs_devices->latest_trans = latest_transid;
521 fs_devices->total_rw_bytes = 0;
522out:
523 return ret;
524}
525
526int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
527 fmode_t flags, void *holder)
528{
529 int ret;
530
531 mutex_lock(&uuid_mutex);
532 if (fs_devices->opened) {
533 fs_devices->opened++;
534 ret = 0;
535 } else {
536 ret = __btrfs_open_devices(fs_devices, flags, holder);
537 }
538 mutex_unlock(&uuid_mutex);
539 return ret;
540}
541
542int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
543 struct btrfs_fs_devices **fs_devices_ret)
544{
545 struct btrfs_super_block *disk_super;
546 struct block_device *bdev;
547 struct buffer_head *bh;
548 int ret;
549 u64 devid;
550 u64 transid;
551
552 mutex_lock(&uuid_mutex);
553
554 bdev = open_bdev_exclusive(path, flags, holder);
555
556 if (IS_ERR(bdev)) {
557 ret = PTR_ERR(bdev);
558 goto error;
559 }
560
561 ret = set_blocksize(bdev, 4096);
562 if (ret)
563 goto error_close;
564 bh = btrfs_read_dev_super(bdev);
565 if (!bh) {
566 ret = -EIO;
567 goto error_close;
568 }
569 disk_super = (struct btrfs_super_block *)bh->b_data;
570 devid = le64_to_cpu(disk_super->dev_item.devid);
571 transid = btrfs_super_generation(disk_super);
572 if (disk_super->label[0])
573 printk(KERN_INFO "device label %s ", disk_super->label);
574 else {
575 /* FIXME, make a readl uuid parser */
576 printk(KERN_INFO "device fsid %llx-%llx ",
577 *(unsigned long long *)disk_super->fsid,
578 *(unsigned long long *)(disk_super->fsid + 8));
579 }
580 printk(KERN_INFO "devid %llu transid %llu %s\n",
581 (unsigned long long)devid, (unsigned long long)transid, path);
582 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
583
584 brelse(bh);
585error_close:
586 close_bdev_exclusive(bdev, flags);
587error:
588 mutex_unlock(&uuid_mutex);
589 return ret;
590}
591
592/*
593 * this uses a pretty simple search, the expectation is that it is
594 * called very infrequently and that a given device has a small number
595 * of extents
596 */
597static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
598 struct btrfs_device *device,
599 u64 num_bytes, u64 *start)
600{
601 struct btrfs_key key;
602 struct btrfs_root *root = device->dev_root;
603 struct btrfs_dev_extent *dev_extent = NULL;
604 struct btrfs_path *path;
605 u64 hole_size = 0;
606 u64 last_byte = 0;
607 u64 search_start = 0;
608 u64 search_end = device->total_bytes;
609 int ret;
610 int slot = 0;
611 int start_found;
612 struct extent_buffer *l;
613
614 path = btrfs_alloc_path();
615 if (!path)
616 return -ENOMEM;
617 path->reada = 2;
618 start_found = 0;
619
620 /* FIXME use last free of some kind */
621
622 /* we don't want to overwrite the superblock on the drive,
623 * so we make sure to start at an offset of at least 1MB
624 */
625 search_start = max((u64)1024 * 1024, search_start);
626
627 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
628 search_start = max(root->fs_info->alloc_start, search_start);
629
630 key.objectid = device->devid;
631 key.offset = search_start;
632 key.type = BTRFS_DEV_EXTENT_KEY;
633 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
634 if (ret < 0)
635 goto error;
636 ret = btrfs_previous_item(root, path, 0, key.type);
637 if (ret < 0)
638 goto error;
639 l = path->nodes[0];
640 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
641 while (1) {
642 l = path->nodes[0];
643 slot = path->slots[0];
644 if (slot >= btrfs_header_nritems(l)) {
645 ret = btrfs_next_leaf(root, path);
646 if (ret == 0)
647 continue;
648 if (ret < 0)
649 goto error;
650no_more_items:
651 if (!start_found) {
652 if (search_start >= search_end) {
653 ret = -ENOSPC;
654 goto error;
655 }
656 *start = search_start;
657 start_found = 1;
658 goto check_pending;
659 }
660 *start = last_byte > search_start ?
661 last_byte : search_start;
662 if (search_end <= *start) {
663 ret = -ENOSPC;
664 goto error;
665 }
666 goto check_pending;
667 }
668 btrfs_item_key_to_cpu(l, &key, slot);
669
670 if (key.objectid < device->devid)
671 goto next;
672
673 if (key.objectid > device->devid)
674 goto no_more_items;
675
676 if (key.offset >= search_start && key.offset > last_byte &&
677 start_found) {
678 if (last_byte < search_start)
679 last_byte = search_start;
680 hole_size = key.offset - last_byte;
681 if (key.offset > last_byte &&
682 hole_size >= num_bytes) {
683 *start = last_byte;
684 goto check_pending;
685 }
686 }
687 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
688 goto next;
689
690 start_found = 1;
691 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
692 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
693next:
694 path->slots[0]++;
695 cond_resched();
696 }
697check_pending:
698 /* we have to make sure we didn't find an extent that has already
699 * been allocated by the map tree or the original allocation
700 */
701 BUG_ON(*start < search_start);
702
703 if (*start + num_bytes > search_end) {
704 ret = -ENOSPC;
705 goto error;
706 }
707 /* check for pending inserts here */
708 ret = 0;
709
710error:
711 btrfs_free_path(path);
712 return ret;
713}
714
715static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
716 struct btrfs_device *device,
717 u64 start)
718{
719 int ret;
720 struct btrfs_path *path;
721 struct btrfs_root *root = device->dev_root;
722 struct btrfs_key key;
723 struct btrfs_key found_key;
724 struct extent_buffer *leaf = NULL;
725 struct btrfs_dev_extent *extent = NULL;
726
727 path = btrfs_alloc_path();
728 if (!path)
729 return -ENOMEM;
730
731 key.objectid = device->devid;
732 key.offset = start;
733 key.type = BTRFS_DEV_EXTENT_KEY;
734
735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
736 if (ret > 0) {
737 ret = btrfs_previous_item(root, path, key.objectid,
738 BTRFS_DEV_EXTENT_KEY);
739 BUG_ON(ret);
740 leaf = path->nodes[0];
741 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
742 extent = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_dev_extent);
744 BUG_ON(found_key.offset > start || found_key.offset +
745 btrfs_dev_extent_length(leaf, extent) < start);
746 ret = 0;
747 } else if (ret == 0) {
748 leaf = path->nodes[0];
749 extent = btrfs_item_ptr(leaf, path->slots[0],
750 struct btrfs_dev_extent);
751 }
752 BUG_ON(ret);
753
754 if (device->bytes_used > 0)
755 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
756 ret = btrfs_del_item(trans, root, path);
757 BUG_ON(ret);
758
759 btrfs_free_path(path);
760 return ret;
761}
762
763int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
764 struct btrfs_device *device,
765 u64 chunk_tree, u64 chunk_objectid,
766 u64 chunk_offset, u64 start, u64 num_bytes)
767{
768 int ret;
769 struct btrfs_path *path;
770 struct btrfs_root *root = device->dev_root;
771 struct btrfs_dev_extent *extent;
772 struct extent_buffer *leaf;
773 struct btrfs_key key;
774
775 WARN_ON(!device->in_fs_metadata);
776 path = btrfs_alloc_path();
777 if (!path)
778 return -ENOMEM;
779
780 key.objectid = device->devid;
781 key.offset = start;
782 key.type = BTRFS_DEV_EXTENT_KEY;
783 ret = btrfs_insert_empty_item(trans, root, path, &key,
784 sizeof(*extent));
785 BUG_ON(ret);
786
787 leaf = path->nodes[0];
788 extent = btrfs_item_ptr(leaf, path->slots[0],
789 struct btrfs_dev_extent);
790 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
791 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
792 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
793
794 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
795 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
796 BTRFS_UUID_SIZE);
797
798 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
799 btrfs_mark_buffer_dirty(leaf);
800 btrfs_free_path(path);
801 return ret;
802}
803
804static noinline int find_next_chunk(struct btrfs_root *root,
805 u64 objectid, u64 *offset)
806{
807 struct btrfs_path *path;
808 int ret;
809 struct btrfs_key key;
810 struct btrfs_chunk *chunk;
811 struct btrfs_key found_key;
812
813 path = btrfs_alloc_path();
814 BUG_ON(!path);
815
816 key.objectid = objectid;
817 key.offset = (u64)-1;
818 key.type = BTRFS_CHUNK_ITEM_KEY;
819
820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
821 if (ret < 0)
822 goto error;
823
824 BUG_ON(ret == 0);
825
826 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
827 if (ret) {
828 *offset = 0;
829 } else {
830 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
831 path->slots[0]);
832 if (found_key.objectid != objectid)
833 *offset = 0;
834 else {
835 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
836 struct btrfs_chunk);
837 *offset = found_key.offset +
838 btrfs_chunk_length(path->nodes[0], chunk);
839 }
840 }
841 ret = 0;
842error:
843 btrfs_free_path(path);
844 return ret;
845}
846
847static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
848{
849 int ret;
850 struct btrfs_key key;
851 struct btrfs_key found_key;
852 struct btrfs_path *path;
853
854 root = root->fs_info->chunk_root;
855
856 path = btrfs_alloc_path();
857 if (!path)
858 return -ENOMEM;
859
860 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
861 key.type = BTRFS_DEV_ITEM_KEY;
862 key.offset = (u64)-1;
863
864 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
865 if (ret < 0)
866 goto error;
867
868 BUG_ON(ret == 0);
869
870 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
871 BTRFS_DEV_ITEM_KEY);
872 if (ret) {
873 *objectid = 1;
874 } else {
875 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
876 path->slots[0]);
877 *objectid = found_key.offset + 1;
878 }
879 ret = 0;
880error:
881 btrfs_free_path(path);
882 return ret;
883}
884
885/*
886 * the device information is stored in the chunk root
887 * the btrfs_device struct should be fully filled in
888 */
889int btrfs_add_device(struct btrfs_trans_handle *trans,
890 struct btrfs_root *root,
891 struct btrfs_device *device)
892{
893 int ret;
894 struct btrfs_path *path;
895 struct btrfs_dev_item *dev_item;
896 struct extent_buffer *leaf;
897 struct btrfs_key key;
898 unsigned long ptr;
899
900 root = root->fs_info->chunk_root;
901
902 path = btrfs_alloc_path();
903 if (!path)
904 return -ENOMEM;
905
906 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
907 key.type = BTRFS_DEV_ITEM_KEY;
908 key.offset = device->devid;
909
910 ret = btrfs_insert_empty_item(trans, root, path, &key,
911 sizeof(*dev_item));
912 if (ret)
913 goto out;
914
915 leaf = path->nodes[0];
916 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
917
918 btrfs_set_device_id(leaf, dev_item, device->devid);
919 btrfs_set_device_generation(leaf, dev_item, 0);
920 btrfs_set_device_type(leaf, dev_item, device->type);
921 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
922 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
923 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
924 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
925 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
926 btrfs_set_device_group(leaf, dev_item, 0);
927 btrfs_set_device_seek_speed(leaf, dev_item, 0);
928 btrfs_set_device_bandwidth(leaf, dev_item, 0);
929 btrfs_set_device_start_offset(leaf, dev_item, 0);
930
931 ptr = (unsigned long)btrfs_device_uuid(dev_item);
932 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
933 ptr = (unsigned long)btrfs_device_fsid(dev_item);
934 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
935 btrfs_mark_buffer_dirty(leaf);
936
937 ret = 0;
938out:
939 btrfs_free_path(path);
940 return ret;
941}
942
943static int btrfs_rm_dev_item(struct btrfs_root *root,
944 struct btrfs_device *device)
945{
946 int ret;
947 struct btrfs_path *path;
948 struct btrfs_key key;
949 struct btrfs_trans_handle *trans;
950
951 root = root->fs_info->chunk_root;
952
953 path = btrfs_alloc_path();
954 if (!path)
955 return -ENOMEM;
956
957 trans = btrfs_start_transaction(root, 1);
958 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
959 key.type = BTRFS_DEV_ITEM_KEY;
960 key.offset = device->devid;
961 lock_chunks(root);
962
963 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
964 if (ret < 0)
965 goto out;
966
967 if (ret > 0) {
968 ret = -ENOENT;
969 goto out;
970 }
971
972 ret = btrfs_del_item(trans, root, path);
973 if (ret)
974 goto out;
975out:
976 btrfs_free_path(path);
977 unlock_chunks(root);
978 btrfs_commit_transaction(trans, root);
979 return ret;
980}
981
982int btrfs_rm_device(struct btrfs_root *root, char *device_path)
983{
984 struct btrfs_device *device;
985 struct btrfs_device *next_device;
986 struct block_device *bdev;
987 struct buffer_head *bh = NULL;
988 struct btrfs_super_block *disk_super;
989 u64 all_avail;
990 u64 devid;
991 u64 num_devices;
992 u8 *dev_uuid;
993 int ret = 0;
994
995 mutex_lock(&uuid_mutex);
996 mutex_lock(&root->fs_info->volume_mutex);
997
998 all_avail = root->fs_info->avail_data_alloc_bits |
999 root->fs_info->avail_system_alloc_bits |
1000 root->fs_info->avail_metadata_alloc_bits;
1001
1002 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1003 root->fs_info->fs_devices->rw_devices <= 4) {
1004 printk(KERN_ERR "btrfs: unable to go below four devices "
1005 "on raid10\n");
1006 ret = -EINVAL;
1007 goto out;
1008 }
1009
1010 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1011 root->fs_info->fs_devices->rw_devices <= 2) {
1012 printk(KERN_ERR "btrfs: unable to go below two "
1013 "devices on raid1\n");
1014 ret = -EINVAL;
1015 goto out;
1016 }
1017
1018 if (strcmp(device_path, "missing") == 0) {
1019 struct list_head *cur;
1020 struct list_head *devices;
1021 struct btrfs_device *tmp;
1022
1023 device = NULL;
1024 devices = &root->fs_info->fs_devices->devices;
1025 list_for_each(cur, devices) {
1026 tmp = list_entry(cur, struct btrfs_device, dev_list);
1027 if (tmp->in_fs_metadata && !tmp->bdev) {
1028 device = tmp;
1029 break;
1030 }
1031 }
1032 bdev = NULL;
1033 bh = NULL;
1034 disk_super = NULL;
1035 if (!device) {
1036 printk(KERN_ERR "btrfs: no missing devices found to "
1037 "remove\n");
1038 goto out;
1039 }
1040 } else {
1041 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1042 root->fs_info->bdev_holder);
1043 if (IS_ERR(bdev)) {
1044 ret = PTR_ERR(bdev);
1045 goto out;
1046 }
1047
1048 set_blocksize(bdev, 4096);
1049 bh = btrfs_read_dev_super(bdev);
1050 if (!bh) {
1051 ret = -EIO;
1052 goto error_close;
1053 }
1054 disk_super = (struct btrfs_super_block *)bh->b_data;
1055 devid = le64_to_cpu(disk_super->dev_item.devid);
1056 dev_uuid = disk_super->dev_item.uuid;
1057 device = btrfs_find_device(root, devid, dev_uuid,
1058 disk_super->fsid);
1059 if (!device) {
1060 ret = -ENOENT;
1061 goto error_brelse;
1062 }
1063 }
1064
1065 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1066 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1067 "device\n");
1068 ret = -EINVAL;
1069 goto error_brelse;
1070 }
1071
1072 if (device->writeable) {
1073 list_del_init(&device->dev_alloc_list);
1074 root->fs_info->fs_devices->rw_devices--;
1075 }
1076
1077 ret = btrfs_shrink_device(device, 0);
1078 if (ret)
1079 goto error_brelse;
1080
1081 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1082 if (ret)
1083 goto error_brelse;
1084
1085 device->in_fs_metadata = 0;
1086 list_del_init(&device->dev_list);
1087 device->fs_devices->num_devices--;
1088
1089 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1090 struct btrfs_device, dev_list);
1091 if (device->bdev == root->fs_info->sb->s_bdev)
1092 root->fs_info->sb->s_bdev = next_device->bdev;
1093 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1094 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1095
1096 if (device->bdev) {
1097 close_bdev_exclusive(device->bdev, device->mode);
1098 device->bdev = NULL;
1099 device->fs_devices->open_devices--;
1100 }
1101
1102 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1103 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1104
1105 if (device->fs_devices->open_devices == 0) {
1106 struct btrfs_fs_devices *fs_devices;
1107 fs_devices = root->fs_info->fs_devices;
1108 while (fs_devices) {
1109 if (fs_devices->seed == device->fs_devices)
1110 break;
1111 fs_devices = fs_devices->seed;
1112 }
1113 fs_devices->seed = device->fs_devices->seed;
1114 device->fs_devices->seed = NULL;
1115 __btrfs_close_devices(device->fs_devices);
1116 free_fs_devices(device->fs_devices);
1117 }
1118
1119 /*
1120 * at this point, the device is zero sized. We want to
1121 * remove it from the devices list and zero out the old super
1122 */
1123 if (device->writeable) {
1124 /* make sure this device isn't detected as part of
1125 * the FS anymore
1126 */
1127 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1128 set_buffer_dirty(bh);
1129 sync_dirty_buffer(bh);
1130 }
1131
1132 kfree(device->name);
1133 kfree(device);
1134 ret = 0;
1135
1136error_brelse:
1137 brelse(bh);
1138error_close:
1139 if (bdev)
1140 close_bdev_exclusive(bdev, FMODE_READ);
1141out:
1142 mutex_unlock(&root->fs_info->volume_mutex);
1143 mutex_unlock(&uuid_mutex);
1144 return ret;
1145}
1146
1147/*
1148 * does all the dirty work required for changing file system's UUID.
1149 */
1150static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1151 struct btrfs_root *root)
1152{
1153 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1154 struct btrfs_fs_devices *old_devices;
1155 struct btrfs_fs_devices *seed_devices;
1156 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1157 struct btrfs_device *device;
1158 u64 super_flags;
1159
1160 BUG_ON(!mutex_is_locked(&uuid_mutex));
1161 if (!fs_devices->seeding)
1162 return -EINVAL;
1163
1164 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1165 if (!seed_devices)
1166 return -ENOMEM;
1167
1168 old_devices = clone_fs_devices(fs_devices);
1169 if (IS_ERR(old_devices)) {
1170 kfree(seed_devices);
1171 return PTR_ERR(old_devices);
1172 }
1173
1174 list_add(&old_devices->list, &fs_uuids);
1175
1176 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1177 seed_devices->opened = 1;
1178 INIT_LIST_HEAD(&seed_devices->devices);
1179 INIT_LIST_HEAD(&seed_devices->alloc_list);
1180 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1181 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1182 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1183 device->fs_devices = seed_devices;
1184 }
1185
1186 fs_devices->seeding = 0;
1187 fs_devices->num_devices = 0;
1188 fs_devices->open_devices = 0;
1189 fs_devices->seed = seed_devices;
1190
1191 generate_random_uuid(fs_devices->fsid);
1192 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1193 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1194 super_flags = btrfs_super_flags(disk_super) &
1195 ~BTRFS_SUPER_FLAG_SEEDING;
1196 btrfs_set_super_flags(disk_super, super_flags);
1197
1198 return 0;
1199}
1200
1201/*
1202 * strore the expected generation for seed devices in device items.
1203 */
1204static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1205 struct btrfs_root *root)
1206{
1207 struct btrfs_path *path;
1208 struct extent_buffer *leaf;
1209 struct btrfs_dev_item *dev_item;
1210 struct btrfs_device *device;
1211 struct btrfs_key key;
1212 u8 fs_uuid[BTRFS_UUID_SIZE];
1213 u8 dev_uuid[BTRFS_UUID_SIZE];
1214 u64 devid;
1215 int ret;
1216
1217 path = btrfs_alloc_path();
1218 if (!path)
1219 return -ENOMEM;
1220
1221 root = root->fs_info->chunk_root;
1222 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1223 key.offset = 0;
1224 key.type = BTRFS_DEV_ITEM_KEY;
1225
1226 while (1) {
1227 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1228 if (ret < 0)
1229 goto error;
1230
1231 leaf = path->nodes[0];
1232next_slot:
1233 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1234 ret = btrfs_next_leaf(root, path);
1235 if (ret > 0)
1236 break;
1237 if (ret < 0)
1238 goto error;
1239 leaf = path->nodes[0];
1240 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1241 btrfs_release_path(root, path);
1242 continue;
1243 }
1244
1245 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1246 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1247 key.type != BTRFS_DEV_ITEM_KEY)
1248 break;
1249
1250 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1251 struct btrfs_dev_item);
1252 devid = btrfs_device_id(leaf, dev_item);
1253 read_extent_buffer(leaf, dev_uuid,
1254 (unsigned long)btrfs_device_uuid(dev_item),
1255 BTRFS_UUID_SIZE);
1256 read_extent_buffer(leaf, fs_uuid,
1257 (unsigned long)btrfs_device_fsid(dev_item),
1258 BTRFS_UUID_SIZE);
1259 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1260 BUG_ON(!device);
1261
1262 if (device->fs_devices->seeding) {
1263 btrfs_set_device_generation(leaf, dev_item,
1264 device->generation);
1265 btrfs_mark_buffer_dirty(leaf);
1266 }
1267
1268 path->slots[0]++;
1269 goto next_slot;
1270 }
1271 ret = 0;
1272error:
1273 btrfs_free_path(path);
1274 return ret;
1275}
1276
1277int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1278{
1279 struct btrfs_trans_handle *trans;
1280 struct btrfs_device *device;
1281 struct block_device *bdev;
1282 struct list_head *cur;
1283 struct list_head *devices;
1284 struct super_block *sb = root->fs_info->sb;
1285 u64 total_bytes;
1286 int seeding_dev = 0;
1287 int ret = 0;
1288
1289 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1290 return -EINVAL;
1291
1292 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1293 if (!bdev)
1294 return -EIO;
1295
1296 if (root->fs_info->fs_devices->seeding) {
1297 seeding_dev = 1;
1298 down_write(&sb->s_umount);
1299 mutex_lock(&uuid_mutex);
1300 }
1301
1302 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1303 mutex_lock(&root->fs_info->volume_mutex);
1304
1305 devices = &root->fs_info->fs_devices->devices;
1306 list_for_each(cur, devices) {
1307 device = list_entry(cur, struct btrfs_device, dev_list);
1308 if (device->bdev == bdev) {
1309 ret = -EEXIST;
1310 goto error;
1311 }
1312 }
1313
1314 device = kzalloc(sizeof(*device), GFP_NOFS);
1315 if (!device) {
1316 /* we can safely leave the fs_devices entry around */
1317 ret = -ENOMEM;
1318 goto error;
1319 }
1320
1321 device->name = kstrdup(device_path, GFP_NOFS);
1322 if (!device->name) {
1323 kfree(device);
1324 ret = -ENOMEM;
1325 goto error;
1326 }
1327
1328 ret = find_next_devid(root, &device->devid);
1329 if (ret) {
1330 kfree(device);
1331 goto error;
1332 }
1333
1334 trans = btrfs_start_transaction(root, 1);
1335 lock_chunks(root);
1336
1337 device->barriers = 1;
1338 device->writeable = 1;
1339 device->work.func = pending_bios_fn;
1340 generate_random_uuid(device->uuid);
1341 spin_lock_init(&device->io_lock);
1342 device->generation = trans->transid;
1343 device->io_width = root->sectorsize;
1344 device->io_align = root->sectorsize;
1345 device->sector_size = root->sectorsize;
1346 device->total_bytes = i_size_read(bdev->bd_inode);
1347 device->dev_root = root->fs_info->dev_root;
1348 device->bdev = bdev;
1349 device->in_fs_metadata = 1;
1350 device->mode = 0;
1351 set_blocksize(device->bdev, 4096);
1352
1353 if (seeding_dev) {
1354 sb->s_flags &= ~MS_RDONLY;
1355 ret = btrfs_prepare_sprout(trans, root);
1356 BUG_ON(ret);
1357 }
1358
1359 device->fs_devices = root->fs_info->fs_devices;
1360 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1361 list_add(&device->dev_alloc_list,
1362 &root->fs_info->fs_devices->alloc_list);
1363 root->fs_info->fs_devices->num_devices++;
1364 root->fs_info->fs_devices->open_devices++;
1365 root->fs_info->fs_devices->rw_devices++;
1366 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1367
1368 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1369 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1370 total_bytes + device->total_bytes);
1371
1372 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1373 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1374 total_bytes + 1);
1375
1376 if (seeding_dev) {
1377 ret = init_first_rw_device(trans, root, device);
1378 BUG_ON(ret);
1379 ret = btrfs_finish_sprout(trans, root);
1380 BUG_ON(ret);
1381 } else {
1382 ret = btrfs_add_device(trans, root, device);
1383 }
1384
1385 unlock_chunks(root);
1386 btrfs_commit_transaction(trans, root);
1387
1388 if (seeding_dev) {
1389 mutex_unlock(&uuid_mutex);
1390 up_write(&sb->s_umount);
1391
1392 ret = btrfs_relocate_sys_chunks(root);
1393 BUG_ON(ret);
1394 }
1395out:
1396 mutex_unlock(&root->fs_info->volume_mutex);
1397 return ret;
1398error:
1399 close_bdev_exclusive(bdev, 0);
1400 if (seeding_dev) {
1401 mutex_unlock(&uuid_mutex);
1402 up_write(&sb->s_umount);
1403 }
1404 goto out;
1405}
1406
1407static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1408 struct btrfs_device *device)
1409{
1410 int ret;
1411 struct btrfs_path *path;
1412 struct btrfs_root *root;
1413 struct btrfs_dev_item *dev_item;
1414 struct extent_buffer *leaf;
1415 struct btrfs_key key;
1416
1417 root = device->dev_root->fs_info->chunk_root;
1418
1419 path = btrfs_alloc_path();
1420 if (!path)
1421 return -ENOMEM;
1422
1423 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1424 key.type = BTRFS_DEV_ITEM_KEY;
1425 key.offset = device->devid;
1426
1427 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1428 if (ret < 0)
1429 goto out;
1430
1431 if (ret > 0) {
1432 ret = -ENOENT;
1433 goto out;
1434 }
1435
1436 leaf = path->nodes[0];
1437 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1438
1439 btrfs_set_device_id(leaf, dev_item, device->devid);
1440 btrfs_set_device_type(leaf, dev_item, device->type);
1441 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1442 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1443 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1444 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1445 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1446 btrfs_mark_buffer_dirty(leaf);
1447
1448out:
1449 btrfs_free_path(path);
1450 return ret;
1451}
1452
1453static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1454 struct btrfs_device *device, u64 new_size)
1455{
1456 struct btrfs_super_block *super_copy =
1457 &device->dev_root->fs_info->super_copy;
1458 u64 old_total = btrfs_super_total_bytes(super_copy);
1459 u64 diff = new_size - device->total_bytes;
1460
1461 if (!device->writeable)
1462 return -EACCES;
1463 if (new_size <= device->total_bytes)
1464 return -EINVAL;
1465
1466 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1467 device->fs_devices->total_rw_bytes += diff;
1468
1469 device->total_bytes = new_size;
1470 return btrfs_update_device(trans, device);
1471}
1472
1473int btrfs_grow_device(struct btrfs_trans_handle *trans,
1474 struct btrfs_device *device, u64 new_size)
1475{
1476 int ret;
1477 lock_chunks(device->dev_root);
1478 ret = __btrfs_grow_device(trans, device, new_size);
1479 unlock_chunks(device->dev_root);
1480 return ret;
1481}
1482
1483static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1484 struct btrfs_root *root,
1485 u64 chunk_tree, u64 chunk_objectid,
1486 u64 chunk_offset)
1487{
1488 int ret;
1489 struct btrfs_path *path;
1490 struct btrfs_key key;
1491
1492 root = root->fs_info->chunk_root;
1493 path = btrfs_alloc_path();
1494 if (!path)
1495 return -ENOMEM;
1496
1497 key.objectid = chunk_objectid;
1498 key.offset = chunk_offset;
1499 key.type = BTRFS_CHUNK_ITEM_KEY;
1500
1501 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1502 BUG_ON(ret);
1503
1504 ret = btrfs_del_item(trans, root, path);
1505 BUG_ON(ret);
1506
1507 btrfs_free_path(path);
1508 return 0;
1509}
1510
1511static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1512 chunk_offset)
1513{
1514 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1515 struct btrfs_disk_key *disk_key;
1516 struct btrfs_chunk *chunk;
1517 u8 *ptr;
1518 int ret = 0;
1519 u32 num_stripes;
1520 u32 array_size;
1521 u32 len = 0;
1522 u32 cur;
1523 struct btrfs_key key;
1524
1525 array_size = btrfs_super_sys_array_size(super_copy);
1526
1527 ptr = super_copy->sys_chunk_array;
1528 cur = 0;
1529
1530 while (cur < array_size) {
1531 disk_key = (struct btrfs_disk_key *)ptr;
1532 btrfs_disk_key_to_cpu(&key, disk_key);
1533
1534 len = sizeof(*disk_key);
1535
1536 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1537 chunk = (struct btrfs_chunk *)(ptr + len);
1538 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1539 len += btrfs_chunk_item_size(num_stripes);
1540 } else {
1541 ret = -EIO;
1542 break;
1543 }
1544 if (key.objectid == chunk_objectid &&
1545 key.offset == chunk_offset) {
1546 memmove(ptr, ptr + len, array_size - (cur + len));
1547 array_size -= len;
1548 btrfs_set_super_sys_array_size(super_copy, array_size);
1549 } else {
1550 ptr += len;
1551 cur += len;
1552 }
1553 }
1554 return ret;
1555}
1556
1557static int btrfs_relocate_chunk(struct btrfs_root *root,
1558 u64 chunk_tree, u64 chunk_objectid,
1559 u64 chunk_offset)
1560{
1561 struct extent_map_tree *em_tree;
1562 struct btrfs_root *extent_root;
1563 struct btrfs_trans_handle *trans;
1564 struct extent_map *em;
1565 struct map_lookup *map;
1566 int ret;
1567 int i;
1568
1569 printk(KERN_INFO "btrfs relocating chunk %llu\n",
1570 (unsigned long long)chunk_offset);
1571 root = root->fs_info->chunk_root;
1572 extent_root = root->fs_info->extent_root;
1573 em_tree = &root->fs_info->mapping_tree.map_tree;
1574
1575 /* step one, relocate all the extents inside this chunk */
1576 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1577 BUG_ON(ret);
1578
1579 trans = btrfs_start_transaction(root, 1);
1580 BUG_ON(!trans);
1581
1582 lock_chunks(root);
1583
1584 /*
1585 * step two, delete the device extents and the
1586 * chunk tree entries
1587 */
1588 spin_lock(&em_tree->lock);
1589 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1590 spin_unlock(&em_tree->lock);
1591
1592 BUG_ON(em->start > chunk_offset ||
1593 em->start + em->len < chunk_offset);
1594 map = (struct map_lookup *)em->bdev;
1595
1596 for (i = 0; i < map->num_stripes; i++) {
1597 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1598 map->stripes[i].physical);
1599 BUG_ON(ret);
1600
1601 if (map->stripes[i].dev) {
1602 ret = btrfs_update_device(trans, map->stripes[i].dev);
1603 BUG_ON(ret);
1604 }
1605 }
1606 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1607 chunk_offset);
1608
1609 BUG_ON(ret);
1610
1611 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1612 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1613 BUG_ON(ret);
1614 }
1615
1616 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1617 BUG_ON(ret);
1618
1619 spin_lock(&em_tree->lock);
1620 remove_extent_mapping(em_tree, em);
1621 spin_unlock(&em_tree->lock);
1622
1623 kfree(map);
1624 em->bdev = NULL;
1625
1626 /* once for the tree */
1627 free_extent_map(em);
1628 /* once for us */
1629 free_extent_map(em);
1630
1631 unlock_chunks(root);
1632 btrfs_end_transaction(trans, root);
1633 return 0;
1634}
1635
1636static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1637{
1638 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1639 struct btrfs_path *path;
1640 struct extent_buffer *leaf;
1641 struct btrfs_chunk *chunk;
1642 struct btrfs_key key;
1643 struct btrfs_key found_key;
1644 u64 chunk_tree = chunk_root->root_key.objectid;
1645 u64 chunk_type;
1646 int ret;
1647
1648 path = btrfs_alloc_path();
1649 if (!path)
1650 return -ENOMEM;
1651
1652 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1653 key.offset = (u64)-1;
1654 key.type = BTRFS_CHUNK_ITEM_KEY;
1655
1656 while (1) {
1657 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1658 if (ret < 0)
1659 goto error;
1660 BUG_ON(ret == 0);
1661
1662 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1663 key.type);
1664 if (ret < 0)
1665 goto error;
1666 if (ret > 0)
1667 break;
1668
1669 leaf = path->nodes[0];
1670 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1671
1672 chunk = btrfs_item_ptr(leaf, path->slots[0],
1673 struct btrfs_chunk);
1674 chunk_type = btrfs_chunk_type(leaf, chunk);
1675 btrfs_release_path(chunk_root, path);
1676
1677 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1678 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1679 found_key.objectid,
1680 found_key.offset);
1681 BUG_ON(ret);
1682 }
1683
1684 if (found_key.offset == 0)
1685 break;
1686 key.offset = found_key.offset - 1;
1687 }
1688 ret = 0;
1689error:
1690 btrfs_free_path(path);
1691 return ret;
1692}
1693
1694static u64 div_factor(u64 num, int factor)
1695{
1696 if (factor == 10)
1697 return num;
1698 num *= factor;
1699 do_div(num, 10);
1700 return num;
1701}
1702
1703int btrfs_balance(struct btrfs_root *dev_root)
1704{
1705 int ret;
1706 struct list_head *cur;
1707 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1708 struct btrfs_device *device;
1709 u64 old_size;
1710 u64 size_to_free;
1711 struct btrfs_path *path;
1712 struct btrfs_key key;
1713 struct btrfs_chunk *chunk;
1714 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1715 struct btrfs_trans_handle *trans;
1716 struct btrfs_key found_key;
1717
1718 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1719 return -EROFS;
1720
1721 mutex_lock(&dev_root->fs_info->volume_mutex);
1722 dev_root = dev_root->fs_info->dev_root;
1723
1724 /* step one make some room on all the devices */
1725 list_for_each(cur, devices) {
1726 device = list_entry(cur, struct btrfs_device, dev_list);
1727 old_size = device->total_bytes;
1728 size_to_free = div_factor(old_size, 1);
1729 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1730 if (!device->writeable ||
1731 device->total_bytes - device->bytes_used > size_to_free)
1732 continue;
1733
1734 ret = btrfs_shrink_device(device, old_size - size_to_free);
1735 BUG_ON(ret);
1736
1737 trans = btrfs_start_transaction(dev_root, 1);
1738 BUG_ON(!trans);
1739
1740 ret = btrfs_grow_device(trans, device, old_size);
1741 BUG_ON(ret);
1742
1743 btrfs_end_transaction(trans, dev_root);
1744 }
1745
1746 /* step two, relocate all the chunks */
1747 path = btrfs_alloc_path();
1748 BUG_ON(!path);
1749
1750 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1751 key.offset = (u64)-1;
1752 key.type = BTRFS_CHUNK_ITEM_KEY;
1753
1754 while (1) {
1755 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1756 if (ret < 0)
1757 goto error;
1758
1759 /*
1760 * this shouldn't happen, it means the last relocate
1761 * failed
1762 */
1763 if (ret == 0)
1764 break;
1765
1766 ret = btrfs_previous_item(chunk_root, path, 0,
1767 BTRFS_CHUNK_ITEM_KEY);
1768 if (ret)
1769 break;
1770
1771 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1772 path->slots[0]);
1773 if (found_key.objectid != key.objectid)
1774 break;
1775
1776 chunk = btrfs_item_ptr(path->nodes[0],
1777 path->slots[0],
1778 struct btrfs_chunk);
1779 key.offset = found_key.offset;
1780 /* chunk zero is special */
1781 if (key.offset == 0)
1782 break;
1783
1784 btrfs_release_path(chunk_root, path);
1785 ret = btrfs_relocate_chunk(chunk_root,
1786 chunk_root->root_key.objectid,
1787 found_key.objectid,
1788 found_key.offset);
1789 BUG_ON(ret);
1790 }
1791 ret = 0;
1792error:
1793 btrfs_free_path(path);
1794 mutex_unlock(&dev_root->fs_info->volume_mutex);
1795 return ret;
1796}
1797
1798/*
1799 * shrinking a device means finding all of the device extents past
1800 * the new size, and then following the back refs to the chunks.
1801 * The chunk relocation code actually frees the device extent
1802 */
1803int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1804{
1805 struct btrfs_trans_handle *trans;
1806 struct btrfs_root *root = device->dev_root;
1807 struct btrfs_dev_extent *dev_extent = NULL;
1808 struct btrfs_path *path;
1809 u64 length;
1810 u64 chunk_tree;
1811 u64 chunk_objectid;
1812 u64 chunk_offset;
1813 int ret;
1814 int slot;
1815 struct extent_buffer *l;
1816 struct btrfs_key key;
1817 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1818 u64 old_total = btrfs_super_total_bytes(super_copy);
1819 u64 diff = device->total_bytes - new_size;
1820
1821 if (new_size >= device->total_bytes)
1822 return -EINVAL;
1823
1824 path = btrfs_alloc_path();
1825 if (!path)
1826 return -ENOMEM;
1827
1828 trans = btrfs_start_transaction(root, 1);
1829 if (!trans) {
1830 ret = -ENOMEM;
1831 goto done;
1832 }
1833
1834 path->reada = 2;
1835
1836 lock_chunks(root);
1837
1838 device->total_bytes = new_size;
1839 if (device->writeable)
1840 device->fs_devices->total_rw_bytes -= diff;
1841 ret = btrfs_update_device(trans, device);
1842 if (ret) {
1843 unlock_chunks(root);
1844 btrfs_end_transaction(trans, root);
1845 goto done;
1846 }
1847 WARN_ON(diff > old_total);
1848 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1849 unlock_chunks(root);
1850 btrfs_end_transaction(trans, root);
1851
1852 key.objectid = device->devid;
1853 key.offset = (u64)-1;
1854 key.type = BTRFS_DEV_EXTENT_KEY;
1855
1856 while (1) {
1857 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1858 if (ret < 0)
1859 goto done;
1860
1861 ret = btrfs_previous_item(root, path, 0, key.type);
1862 if (ret < 0)
1863 goto done;
1864 if (ret) {
1865 ret = 0;
1866 goto done;
1867 }
1868
1869 l = path->nodes[0];
1870 slot = path->slots[0];
1871 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1872
1873 if (key.objectid != device->devid)
1874 goto done;
1875
1876 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1877 length = btrfs_dev_extent_length(l, dev_extent);
1878
1879 if (key.offset + length <= new_size)
1880 goto done;
1881
1882 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1883 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1884 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1885 btrfs_release_path(root, path);
1886
1887 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1888 chunk_offset);
1889 if (ret)
1890 goto done;
1891 }
1892
1893done:
1894 btrfs_free_path(path);
1895 return ret;
1896}
1897
1898static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1899 struct btrfs_root *root,
1900 struct btrfs_key *key,
1901 struct btrfs_chunk *chunk, int item_size)
1902{
1903 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1904 struct btrfs_disk_key disk_key;
1905 u32 array_size;
1906 u8 *ptr;
1907
1908 array_size = btrfs_super_sys_array_size(super_copy);
1909 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1910 return -EFBIG;
1911
1912 ptr = super_copy->sys_chunk_array + array_size;
1913 btrfs_cpu_key_to_disk(&disk_key, key);
1914 memcpy(ptr, &disk_key, sizeof(disk_key));
1915 ptr += sizeof(disk_key);
1916 memcpy(ptr, chunk, item_size);
1917 item_size += sizeof(disk_key);
1918 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1919 return 0;
1920}
1921
1922static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
1923 int num_stripes, int sub_stripes)
1924{
1925 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1926 return calc_size;
1927 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1928 return calc_size * (num_stripes / sub_stripes);
1929 else
1930 return calc_size * num_stripes;
1931}
1932
1933static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *extent_root,
1935 struct map_lookup **map_ret,
1936 u64 *num_bytes, u64 *stripe_size,
1937 u64 start, u64 type)
1938{
1939 struct btrfs_fs_info *info = extent_root->fs_info;
1940 struct btrfs_device *device = NULL;
1941 struct btrfs_fs_devices *fs_devices = info->fs_devices;
1942 struct list_head *cur;
1943 struct map_lookup *map = NULL;
1944 struct extent_map_tree *em_tree;
1945 struct extent_map *em;
1946 struct list_head private_devs;
1947 int min_stripe_size = 1 * 1024 * 1024;
1948 u64 calc_size = 1024 * 1024 * 1024;
1949 u64 max_chunk_size = calc_size;
1950 u64 min_free;
1951 u64 avail;
1952 u64 max_avail = 0;
1953 u64 dev_offset;
1954 int num_stripes = 1;
1955 int min_stripes = 1;
1956 int sub_stripes = 0;
1957 int looped = 0;
1958 int ret;
1959 int index;
1960 int stripe_len = 64 * 1024;
1961
1962 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1963 (type & BTRFS_BLOCK_GROUP_DUP)) {
1964 WARN_ON(1);
1965 type &= ~BTRFS_BLOCK_GROUP_DUP;
1966 }
1967 if (list_empty(&fs_devices->alloc_list))
1968 return -ENOSPC;
1969
1970 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1971 num_stripes = fs_devices->rw_devices;
1972 min_stripes = 2;
1973 }
1974 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1975 num_stripes = 2;
1976 min_stripes = 2;
1977 }
1978 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1979 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1980 if (num_stripes < 2)
1981 return -ENOSPC;
1982 min_stripes = 2;
1983 }
1984 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1985 num_stripes = fs_devices->rw_devices;
1986 if (num_stripes < 4)
1987 return -ENOSPC;
1988 num_stripes &= ~(u32)1;
1989 sub_stripes = 2;
1990 min_stripes = 4;
1991 }
1992
1993 if (type & BTRFS_BLOCK_GROUP_DATA) {
1994 max_chunk_size = 10 * calc_size;
1995 min_stripe_size = 64 * 1024 * 1024;
1996 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1997 max_chunk_size = 4 * calc_size;
1998 min_stripe_size = 32 * 1024 * 1024;
1999 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2000 calc_size = 8 * 1024 * 1024;
2001 max_chunk_size = calc_size * 2;
2002 min_stripe_size = 1 * 1024 * 1024;
2003 }
2004
2005 /* we don't want a chunk larger than 10% of writeable space */
2006 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2007 max_chunk_size);
2008
2009again:
2010 if (!map || map->num_stripes != num_stripes) {
2011 kfree(map);
2012 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2013 if (!map)
2014 return -ENOMEM;
2015 map->num_stripes = num_stripes;
2016 }
2017
2018 if (calc_size * num_stripes > max_chunk_size) {
2019 calc_size = max_chunk_size;
2020 do_div(calc_size, num_stripes);
2021 do_div(calc_size, stripe_len);
2022 calc_size *= stripe_len;
2023 }
2024 /* we don't want tiny stripes */
2025 calc_size = max_t(u64, min_stripe_size, calc_size);
2026
2027 do_div(calc_size, stripe_len);
2028 calc_size *= stripe_len;
2029
2030 cur = fs_devices->alloc_list.next;
2031 index = 0;
2032
2033 if (type & BTRFS_BLOCK_GROUP_DUP)
2034 min_free = calc_size * 2;
2035 else
2036 min_free = calc_size;
2037
2038 /*
2039 * we add 1MB because we never use the first 1MB of the device, unless
2040 * we've looped, then we are likely allocating the maximum amount of
2041 * space left already
2042 */
2043 if (!looped)
2044 min_free += 1024 * 1024;
2045
2046 INIT_LIST_HEAD(&private_devs);
2047 while (index < num_stripes) {
2048 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2049 BUG_ON(!device->writeable);
2050 if (device->total_bytes > device->bytes_used)
2051 avail = device->total_bytes - device->bytes_used;
2052 else
2053 avail = 0;
2054 cur = cur->next;
2055
2056 if (device->in_fs_metadata && avail >= min_free) {
2057 ret = find_free_dev_extent(trans, device,
2058 min_free, &dev_offset);
2059 if (ret == 0) {
2060 list_move_tail(&device->dev_alloc_list,
2061 &private_devs);
2062 map->stripes[index].dev = device;
2063 map->stripes[index].physical = dev_offset;
2064 index++;
2065 if (type & BTRFS_BLOCK_GROUP_DUP) {
2066 map->stripes[index].dev = device;
2067 map->stripes[index].physical =
2068 dev_offset + calc_size;
2069 index++;
2070 }
2071 }
2072 } else if (device->in_fs_metadata && avail > max_avail)
2073 max_avail = avail;
2074 if (cur == &fs_devices->alloc_list)
2075 break;
2076 }
2077 list_splice(&private_devs, &fs_devices->alloc_list);
2078 if (index < num_stripes) {
2079 if (index >= min_stripes) {
2080 num_stripes = index;
2081 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2082 num_stripes /= sub_stripes;
2083 num_stripes *= sub_stripes;
2084 }
2085 looped = 1;
2086 goto again;
2087 }
2088 if (!looped && max_avail > 0) {
2089 looped = 1;
2090 calc_size = max_avail;
2091 goto again;
2092 }
2093 kfree(map);
2094 return -ENOSPC;
2095 }
2096 map->sector_size = extent_root->sectorsize;
2097 map->stripe_len = stripe_len;
2098 map->io_align = stripe_len;
2099 map->io_width = stripe_len;
2100 map->type = type;
2101 map->num_stripes = num_stripes;
2102 map->sub_stripes = sub_stripes;
2103
2104 *map_ret = map;
2105 *stripe_size = calc_size;
2106 *num_bytes = chunk_bytes_by_type(type, calc_size,
2107 num_stripes, sub_stripes);
2108
2109 em = alloc_extent_map(GFP_NOFS);
2110 if (!em) {
2111 kfree(map);
2112 return -ENOMEM;
2113 }
2114 em->bdev = (struct block_device *)map;
2115 em->start = start;
2116 em->len = *num_bytes;
2117 em->block_start = 0;
2118 em->block_len = em->len;
2119
2120 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2121 spin_lock(&em_tree->lock);
2122 ret = add_extent_mapping(em_tree, em);
2123 spin_unlock(&em_tree->lock);
2124 BUG_ON(ret);
2125 free_extent_map(em);
2126
2127 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2128 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2129 start, *num_bytes);
2130 BUG_ON(ret);
2131
2132 index = 0;
2133 while (index < map->num_stripes) {
2134 device = map->stripes[index].dev;
2135 dev_offset = map->stripes[index].physical;
2136
2137 ret = btrfs_alloc_dev_extent(trans, device,
2138 info->chunk_root->root_key.objectid,
2139 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2140 start, dev_offset, calc_size);
2141 BUG_ON(ret);
2142 index++;
2143 }
2144
2145 return 0;
2146}
2147
2148static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *extent_root,
2150 struct map_lookup *map, u64 chunk_offset,
2151 u64 chunk_size, u64 stripe_size)
2152{
2153 u64 dev_offset;
2154 struct btrfs_key key;
2155 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2156 struct btrfs_device *device;
2157 struct btrfs_chunk *chunk;
2158 struct btrfs_stripe *stripe;
2159 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2160 int index = 0;
2161 int ret;
2162
2163 chunk = kzalloc(item_size, GFP_NOFS);
2164 if (!chunk)
2165 return -ENOMEM;
2166
2167 index = 0;
2168 while (index < map->num_stripes) {
2169 device = map->stripes[index].dev;
2170 device->bytes_used += stripe_size;
2171 ret = btrfs_update_device(trans, device);
2172 BUG_ON(ret);
2173 index++;
2174 }
2175
2176 index = 0;
2177 stripe = &chunk->stripe;
2178 while (index < map->num_stripes) {
2179 device = map->stripes[index].dev;
2180 dev_offset = map->stripes[index].physical;
2181
2182 btrfs_set_stack_stripe_devid(stripe, device->devid);
2183 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2184 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2185 stripe++;
2186 index++;
2187 }
2188
2189 btrfs_set_stack_chunk_length(chunk, chunk_size);
2190 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2191 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2192 btrfs_set_stack_chunk_type(chunk, map->type);
2193 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2194 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2195 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2196 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2197 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2198
2199 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2200 key.type = BTRFS_CHUNK_ITEM_KEY;
2201 key.offset = chunk_offset;
2202
2203 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2204 BUG_ON(ret);
2205
2206 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2207 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2208 item_size);
2209 BUG_ON(ret);
2210 }
2211 kfree(chunk);
2212 return 0;
2213}
2214
2215/*
2216 * Chunk allocation falls into two parts. The first part does works
2217 * that make the new allocated chunk useable, but not do any operation
2218 * that modifies the chunk tree. The second part does the works that
2219 * require modifying the chunk tree. This division is important for the
2220 * bootstrap process of adding storage to a seed btrfs.
2221 */
2222int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2223 struct btrfs_root *extent_root, u64 type)
2224{
2225 u64 chunk_offset;
2226 u64 chunk_size;
2227 u64 stripe_size;
2228 struct map_lookup *map;
2229 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2230 int ret;
2231
2232 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2233 &chunk_offset);
2234 if (ret)
2235 return ret;
2236
2237 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2238 &stripe_size, chunk_offset, type);
2239 if (ret)
2240 return ret;
2241
2242 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2243 chunk_size, stripe_size);
2244 BUG_ON(ret);
2245 return 0;
2246}
2247
2248static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2249 struct btrfs_root *root,
2250 struct btrfs_device *device)
2251{
2252 u64 chunk_offset;
2253 u64 sys_chunk_offset;
2254 u64 chunk_size;
2255 u64 sys_chunk_size;
2256 u64 stripe_size;
2257 u64 sys_stripe_size;
2258 u64 alloc_profile;
2259 struct map_lookup *map;
2260 struct map_lookup *sys_map;
2261 struct btrfs_fs_info *fs_info = root->fs_info;
2262 struct btrfs_root *extent_root = fs_info->extent_root;
2263 int ret;
2264
2265 ret = find_next_chunk(fs_info->chunk_root,
2266 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2267 BUG_ON(ret);
2268
2269 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2270 (fs_info->metadata_alloc_profile &
2271 fs_info->avail_metadata_alloc_bits);
2272 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2273
2274 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2275 &stripe_size, chunk_offset, alloc_profile);
2276 BUG_ON(ret);
2277
2278 sys_chunk_offset = chunk_offset + chunk_size;
2279
2280 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2281 (fs_info->system_alloc_profile &
2282 fs_info->avail_system_alloc_bits);
2283 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2284
2285 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2286 &sys_chunk_size, &sys_stripe_size,
2287 sys_chunk_offset, alloc_profile);
2288 BUG_ON(ret);
2289
2290 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2291 BUG_ON(ret);
2292
2293 /*
2294 * Modifying chunk tree needs allocating new blocks from both
2295 * system block group and metadata block group. So we only can
2296 * do operations require modifying the chunk tree after both
2297 * block groups were created.
2298 */
2299 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2300 chunk_size, stripe_size);
2301 BUG_ON(ret);
2302
2303 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2304 sys_chunk_offset, sys_chunk_size,
2305 sys_stripe_size);
2306 BUG_ON(ret);
2307 return 0;
2308}
2309
2310int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2311{
2312 struct extent_map *em;
2313 struct map_lookup *map;
2314 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2315 int readonly = 0;
2316 int i;
2317
2318 spin_lock(&map_tree->map_tree.lock);
2319 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2320 spin_unlock(&map_tree->map_tree.lock);
2321 if (!em)
2322 return 1;
2323
2324 map = (struct map_lookup *)em->bdev;
2325 for (i = 0; i < map->num_stripes; i++) {
2326 if (!map->stripes[i].dev->writeable) {
2327 readonly = 1;
2328 break;
2329 }
2330 }
2331 free_extent_map(em);
2332 return readonly;
2333}
2334
2335void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2336{
2337 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2338}
2339
2340void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2341{
2342 struct extent_map *em;
2343
2344 while (1) {
2345 spin_lock(&tree->map_tree.lock);
2346 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2347 if (em)
2348 remove_extent_mapping(&tree->map_tree, em);
2349 spin_unlock(&tree->map_tree.lock);
2350 if (!em)
2351 break;
2352 kfree(em->bdev);
2353 /* once for us */
2354 free_extent_map(em);
2355 /* once for the tree */
2356 free_extent_map(em);
2357 }
2358}
2359
2360int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2361{
2362 struct extent_map *em;
2363 struct map_lookup *map;
2364 struct extent_map_tree *em_tree = &map_tree->map_tree;
2365 int ret;
2366
2367 spin_lock(&em_tree->lock);
2368 em = lookup_extent_mapping(em_tree, logical, len);
2369 spin_unlock(&em_tree->lock);
2370 BUG_ON(!em);
2371
2372 BUG_ON(em->start > logical || em->start + em->len < logical);
2373 map = (struct map_lookup *)em->bdev;
2374 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2375 ret = map->num_stripes;
2376 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2377 ret = map->sub_stripes;
2378 else
2379 ret = 1;
2380 free_extent_map(em);
2381 return ret;
2382}
2383
2384static int find_live_mirror(struct map_lookup *map, int first, int num,
2385 int optimal)
2386{
2387 int i;
2388 if (map->stripes[optimal].dev->bdev)
2389 return optimal;
2390 for (i = first; i < first + num; i++) {
2391 if (map->stripes[i].dev->bdev)
2392 return i;
2393 }
2394 /* we couldn't find one that doesn't fail. Just return something
2395 * and the io error handling code will clean up eventually
2396 */
2397 return optimal;
2398}
2399
2400static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2401 u64 logical, u64 *length,
2402 struct btrfs_multi_bio **multi_ret,
2403 int mirror_num, struct page *unplug_page)
2404{
2405 struct extent_map *em;
2406 struct map_lookup *map;
2407 struct extent_map_tree *em_tree = &map_tree->map_tree;
2408 u64 offset;
2409 u64 stripe_offset;
2410 u64 stripe_nr;
2411 int stripes_allocated = 8;
2412 int stripes_required = 1;
2413 int stripe_index;
2414 int i;
2415 int num_stripes;
2416 int max_errors = 0;
2417 struct btrfs_multi_bio *multi = NULL;
2418
2419 if (multi_ret && !(rw & (1 << BIO_RW)))
2420 stripes_allocated = 1;
2421again:
2422 if (multi_ret) {
2423 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2424 GFP_NOFS);
2425 if (!multi)
2426 return -ENOMEM;
2427
2428 atomic_set(&multi->error, 0);
2429 }
2430
2431 spin_lock(&em_tree->lock);
2432 em = lookup_extent_mapping(em_tree, logical, *length);
2433 spin_unlock(&em_tree->lock);
2434
2435 if (!em && unplug_page)
2436 return 0;
2437
2438 if (!em) {
2439 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2440 (unsigned long long)logical,
2441 (unsigned long long)*length);
2442 BUG();
2443 }
2444
2445 BUG_ON(em->start > logical || em->start + em->len < logical);
2446 map = (struct map_lookup *)em->bdev;
2447 offset = logical - em->start;
2448
2449 if (mirror_num > map->num_stripes)
2450 mirror_num = 0;
2451
2452 /* if our multi bio struct is too small, back off and try again */
2453 if (rw & (1 << BIO_RW)) {
2454 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2455 BTRFS_BLOCK_GROUP_DUP)) {
2456 stripes_required = map->num_stripes;
2457 max_errors = 1;
2458 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2459 stripes_required = map->sub_stripes;
2460 max_errors = 1;
2461 }
2462 }
2463 if (multi_ret && rw == WRITE &&
2464 stripes_allocated < stripes_required) {
2465 stripes_allocated = map->num_stripes;
2466 free_extent_map(em);
2467 kfree(multi);
2468 goto again;
2469 }
2470 stripe_nr = offset;
2471 /*
2472 * stripe_nr counts the total number of stripes we have to stride
2473 * to get to this block
2474 */
2475 do_div(stripe_nr, map->stripe_len);
2476
2477 stripe_offset = stripe_nr * map->stripe_len;
2478 BUG_ON(offset < stripe_offset);
2479
2480 /* stripe_offset is the offset of this block in its stripe*/
2481 stripe_offset = offset - stripe_offset;
2482
2483 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2484 BTRFS_BLOCK_GROUP_RAID10 |
2485 BTRFS_BLOCK_GROUP_DUP)) {
2486 /* we limit the length of each bio to what fits in a stripe */
2487 *length = min_t(u64, em->len - offset,
2488 map->stripe_len - stripe_offset);
2489 } else {
2490 *length = em->len - offset;
2491 }
2492
2493 if (!multi_ret && !unplug_page)
2494 goto out;
2495
2496 num_stripes = 1;
2497 stripe_index = 0;
2498 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2499 if (unplug_page || (rw & (1 << BIO_RW)))
2500 num_stripes = map->num_stripes;
2501 else if (mirror_num)
2502 stripe_index = mirror_num - 1;
2503 else {
2504 stripe_index = find_live_mirror(map, 0,
2505 map->num_stripes,
2506 current->pid % map->num_stripes);
2507 }
2508
2509 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2510 if (rw & (1 << BIO_RW))
2511 num_stripes = map->num_stripes;
2512 else if (mirror_num)
2513 stripe_index = mirror_num - 1;
2514
2515 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2516 int factor = map->num_stripes / map->sub_stripes;
2517
2518 stripe_index = do_div(stripe_nr, factor);
2519 stripe_index *= map->sub_stripes;
2520
2521 if (unplug_page || (rw & (1 << BIO_RW)))
2522 num_stripes = map->sub_stripes;
2523 else if (mirror_num)
2524 stripe_index += mirror_num - 1;
2525 else {
2526 stripe_index = find_live_mirror(map, stripe_index,
2527 map->sub_stripes, stripe_index +
2528 current->pid % map->sub_stripes);
2529 }
2530 } else {
2531 /*
2532 * after this do_div call, stripe_nr is the number of stripes
2533 * on this device we have to walk to find the data, and
2534 * stripe_index is the number of our device in the stripe array
2535 */
2536 stripe_index = do_div(stripe_nr, map->num_stripes);
2537 }
2538 BUG_ON(stripe_index >= map->num_stripes);
2539
2540 for (i = 0; i < num_stripes; i++) {
2541 if (unplug_page) {
2542 struct btrfs_device *device;
2543 struct backing_dev_info *bdi;
2544
2545 device = map->stripes[stripe_index].dev;
2546 if (device->bdev) {
2547 bdi = blk_get_backing_dev_info(device->bdev);
2548 if (bdi->unplug_io_fn)
2549 bdi->unplug_io_fn(bdi, unplug_page);
2550 }
2551 } else {
2552 multi->stripes[i].physical =
2553 map->stripes[stripe_index].physical +
2554 stripe_offset + stripe_nr * map->stripe_len;
2555 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2556 }
2557 stripe_index++;
2558 }
2559 if (multi_ret) {
2560 *multi_ret = multi;
2561 multi->num_stripes = num_stripes;
2562 multi->max_errors = max_errors;
2563 }
2564out:
2565 free_extent_map(em);
2566 return 0;
2567}
2568
2569int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2570 u64 logical, u64 *length,
2571 struct btrfs_multi_bio **multi_ret, int mirror_num)
2572{
2573 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2574 mirror_num, NULL);
2575}
2576
2577int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2578 u64 chunk_start, u64 physical, u64 devid,
2579 u64 **logical, int *naddrs, int *stripe_len)
2580{
2581 struct extent_map_tree *em_tree = &map_tree->map_tree;
2582 struct extent_map *em;
2583 struct map_lookup *map;
2584 u64 *buf;
2585 u64 bytenr;
2586 u64 length;
2587 u64 stripe_nr;
2588 int i, j, nr = 0;
2589
2590 spin_lock(&em_tree->lock);
2591 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2592 spin_unlock(&em_tree->lock);
2593
2594 BUG_ON(!em || em->start != chunk_start);
2595 map = (struct map_lookup *)em->bdev;
2596
2597 length = em->len;
2598 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2599 do_div(length, map->num_stripes / map->sub_stripes);
2600 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2601 do_div(length, map->num_stripes);
2602
2603 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2604 BUG_ON(!buf);
2605
2606 for (i = 0; i < map->num_stripes; i++) {
2607 if (devid && map->stripes[i].dev->devid != devid)
2608 continue;
2609 if (map->stripes[i].physical > physical ||
2610 map->stripes[i].physical + length <= physical)
2611 continue;
2612
2613 stripe_nr = physical - map->stripes[i].physical;
2614 do_div(stripe_nr, map->stripe_len);
2615
2616 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2617 stripe_nr = stripe_nr * map->num_stripes + i;
2618 do_div(stripe_nr, map->sub_stripes);
2619 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2620 stripe_nr = stripe_nr * map->num_stripes + i;
2621 }
2622 bytenr = chunk_start + stripe_nr * map->stripe_len;
2623 WARN_ON(nr >= map->num_stripes);
2624 for (j = 0; j < nr; j++) {
2625 if (buf[j] == bytenr)
2626 break;
2627 }
2628 if (j == nr) {
2629 WARN_ON(nr >= map->num_stripes);
2630 buf[nr++] = bytenr;
2631 }
2632 }
2633
2634 for (i = 0; i > nr; i++) {
2635 struct btrfs_multi_bio *multi;
2636 struct btrfs_bio_stripe *stripe;
2637 int ret;
2638
2639 length = 1;
2640 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2641 &length, &multi, 0);
2642 BUG_ON(ret);
2643
2644 stripe = multi->stripes;
2645 for (j = 0; j < multi->num_stripes; j++) {
2646 if (stripe->physical >= physical &&
2647 physical < stripe->physical + length)
2648 break;
2649 }
2650 BUG_ON(j >= multi->num_stripes);
2651 kfree(multi);
2652 }
2653
2654 *logical = buf;
2655 *naddrs = nr;
2656 *stripe_len = map->stripe_len;
2657
2658 free_extent_map(em);
2659 return 0;
2660}
2661
2662int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2663 u64 logical, struct page *page)
2664{
2665 u64 length = PAGE_CACHE_SIZE;
2666 return __btrfs_map_block(map_tree, READ, logical, &length,
2667 NULL, 0, page);
2668}
2669
2670static void end_bio_multi_stripe(struct bio *bio, int err)
2671{
2672 struct btrfs_multi_bio *multi = bio->bi_private;
2673 int is_orig_bio = 0;
2674
2675 if (err)
2676 atomic_inc(&multi->error);
2677
2678 if (bio == multi->orig_bio)
2679 is_orig_bio = 1;
2680
2681 if (atomic_dec_and_test(&multi->stripes_pending)) {
2682 if (!is_orig_bio) {
2683 bio_put(bio);
2684 bio = multi->orig_bio;
2685 }
2686 bio->bi_private = multi->private;
2687 bio->bi_end_io = multi->end_io;
2688 /* only send an error to the higher layers if it is
2689 * beyond the tolerance of the multi-bio
2690 */
2691 if (atomic_read(&multi->error) > multi->max_errors) {
2692 err = -EIO;
2693 } else if (err) {
2694 /*
2695 * this bio is actually up to date, we didn't
2696 * go over the max number of errors
2697 */
2698 set_bit(BIO_UPTODATE, &bio->bi_flags);
2699 err = 0;
2700 }
2701 kfree(multi);
2702
2703 bio_endio(bio, err);
2704 } else if (!is_orig_bio) {
2705 bio_put(bio);
2706 }
2707}
2708
2709struct async_sched {
2710 struct bio *bio;
2711 int rw;
2712 struct btrfs_fs_info *info;
2713 struct btrfs_work work;
2714};
2715
2716/*
2717 * see run_scheduled_bios for a description of why bios are collected for
2718 * async submit.
2719 *
2720 * This will add one bio to the pending list for a device and make sure
2721 * the work struct is scheduled.
2722 */
2723static noinline int schedule_bio(struct btrfs_root *root,
2724 struct btrfs_device *device,
2725 int rw, struct bio *bio)
2726{
2727 int should_queue = 1;
2728
2729 /* don't bother with additional async steps for reads, right now */
2730 if (!(rw & (1 << BIO_RW))) {
2731 bio_get(bio);
2732 submit_bio(rw, bio);
2733 bio_put(bio);
2734 return 0;
2735 }
2736
2737 /*
2738 * nr_async_bios allows us to reliably return congestion to the
2739 * higher layers. Otherwise, the async bio makes it appear we have
2740 * made progress against dirty pages when we've really just put it
2741 * on a queue for later
2742 */
2743 atomic_inc(&root->fs_info->nr_async_bios);
2744 WARN_ON(bio->bi_next);
2745 bio->bi_next = NULL;
2746 bio->bi_rw |= rw;
2747
2748 spin_lock(&device->io_lock);
2749
2750 if (device->pending_bio_tail)
2751 device->pending_bio_tail->bi_next = bio;
2752
2753 device->pending_bio_tail = bio;
2754 if (!device->pending_bios)
2755 device->pending_bios = bio;
2756 if (device->running_pending)
2757 should_queue = 0;
2758
2759 spin_unlock(&device->io_lock);
2760
2761 if (should_queue)
2762 btrfs_queue_worker(&root->fs_info->submit_workers,
2763 &device->work);
2764 return 0;
2765}
2766
2767int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2768 int mirror_num, int async_submit)
2769{
2770 struct btrfs_mapping_tree *map_tree;
2771 struct btrfs_device *dev;
2772 struct bio *first_bio = bio;
2773 u64 logical = (u64)bio->bi_sector << 9;
2774 u64 length = 0;
2775 u64 map_length;
2776 struct btrfs_multi_bio *multi = NULL;
2777 int ret;
2778 int dev_nr = 0;
2779 int total_devs = 1;
2780
2781 length = bio->bi_size;
2782 map_tree = &root->fs_info->mapping_tree;
2783 map_length = length;
2784
2785 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2786 mirror_num);
2787 BUG_ON(ret);
2788
2789 total_devs = multi->num_stripes;
2790 if (map_length < length) {
2791 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2792 "len %llu\n", (unsigned long long)logical,
2793 (unsigned long long)length,
2794 (unsigned long long)map_length);
2795 BUG();
2796 }
2797 multi->end_io = first_bio->bi_end_io;
2798 multi->private = first_bio->bi_private;
2799 multi->orig_bio = first_bio;
2800 atomic_set(&multi->stripes_pending, multi->num_stripes);
2801
2802 while (dev_nr < total_devs) {
2803 if (total_devs > 1) {
2804 if (dev_nr < total_devs - 1) {
2805 bio = bio_clone(first_bio, GFP_NOFS);
2806 BUG_ON(!bio);
2807 } else {
2808 bio = first_bio;
2809 }
2810 bio->bi_private = multi;
2811 bio->bi_end_io = end_bio_multi_stripe;
2812 }
2813 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2814 dev = multi->stripes[dev_nr].dev;
2815 BUG_ON(rw == WRITE && !dev->writeable);
2816 if (dev && dev->bdev) {
2817 bio->bi_bdev = dev->bdev;
2818 if (async_submit)
2819 schedule_bio(root, dev, rw, bio);
2820 else
2821 submit_bio(rw, bio);
2822 } else {
2823 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2824 bio->bi_sector = logical >> 9;
2825 bio_endio(bio, -EIO);
2826 }
2827 dev_nr++;
2828 }
2829 if (total_devs == 1)
2830 kfree(multi);
2831 return 0;
2832}
2833
2834struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2835 u8 *uuid, u8 *fsid)
2836{
2837 struct btrfs_device *device;
2838 struct btrfs_fs_devices *cur_devices;
2839
2840 cur_devices = root->fs_info->fs_devices;
2841 while (cur_devices) {
2842 if (!fsid ||
2843 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2844 device = __find_device(&cur_devices->devices,
2845 devid, uuid);
2846 if (device)
2847 return device;
2848 }
2849 cur_devices = cur_devices->seed;
2850 }
2851 return NULL;
2852}
2853
2854static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2855 u64 devid, u8 *dev_uuid)
2856{
2857 struct btrfs_device *device;
2858 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2859
2860 device = kzalloc(sizeof(*device), GFP_NOFS);
2861 if (!device)
2862 return NULL;
2863 list_add(&device->dev_list,
2864 &fs_devices->devices);
2865 device->barriers = 1;
2866 device->dev_root = root->fs_info->dev_root;
2867 device->devid = devid;
2868 device->work.func = pending_bios_fn;
2869 device->fs_devices = fs_devices;
2870 fs_devices->num_devices++;
2871 spin_lock_init(&device->io_lock);
2872 INIT_LIST_HEAD(&device->dev_alloc_list);
2873 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2874 return device;
2875}
2876
2877static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2878 struct extent_buffer *leaf,
2879 struct btrfs_chunk *chunk)
2880{
2881 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2882 struct map_lookup *map;
2883 struct extent_map *em;
2884 u64 logical;
2885 u64 length;
2886 u64 devid;
2887 u8 uuid[BTRFS_UUID_SIZE];
2888 int num_stripes;
2889 int ret;
2890 int i;
2891
2892 logical = key->offset;
2893 length = btrfs_chunk_length(leaf, chunk);
2894
2895 spin_lock(&map_tree->map_tree.lock);
2896 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2897 spin_unlock(&map_tree->map_tree.lock);
2898
2899 /* already mapped? */
2900 if (em && em->start <= logical && em->start + em->len > logical) {
2901 free_extent_map(em);
2902 return 0;
2903 } else if (em) {
2904 free_extent_map(em);
2905 }
2906
2907 map = kzalloc(sizeof(*map), GFP_NOFS);
2908 if (!map)
2909 return -ENOMEM;
2910
2911 em = alloc_extent_map(GFP_NOFS);
2912 if (!em)
2913 return -ENOMEM;
2914 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2915 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2916 if (!map) {
2917 free_extent_map(em);
2918 return -ENOMEM;
2919 }
2920
2921 em->bdev = (struct block_device *)map;
2922 em->start = logical;
2923 em->len = length;
2924 em->block_start = 0;
2925 em->block_len = em->len;
2926
2927 map->num_stripes = num_stripes;
2928 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2929 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2930 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2931 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2932 map->type = btrfs_chunk_type(leaf, chunk);
2933 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2934 for (i = 0; i < num_stripes; i++) {
2935 map->stripes[i].physical =
2936 btrfs_stripe_offset_nr(leaf, chunk, i);
2937 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2938 read_extent_buffer(leaf, uuid, (unsigned long)
2939 btrfs_stripe_dev_uuid_nr(chunk, i),
2940 BTRFS_UUID_SIZE);
2941 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2942 NULL);
2943 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2944 kfree(map);
2945 free_extent_map(em);
2946 return -EIO;
2947 }
2948 if (!map->stripes[i].dev) {
2949 map->stripes[i].dev =
2950 add_missing_dev(root, devid, uuid);
2951 if (!map->stripes[i].dev) {
2952 kfree(map);
2953 free_extent_map(em);
2954 return -EIO;
2955 }
2956 }
2957 map->stripes[i].dev->in_fs_metadata = 1;
2958 }
2959
2960 spin_lock(&map_tree->map_tree.lock);
2961 ret = add_extent_mapping(&map_tree->map_tree, em);
2962 spin_unlock(&map_tree->map_tree.lock);
2963 BUG_ON(ret);
2964 free_extent_map(em);
2965
2966 return 0;
2967}
2968
2969static int fill_device_from_item(struct extent_buffer *leaf,
2970 struct btrfs_dev_item *dev_item,
2971 struct btrfs_device *device)
2972{
2973 unsigned long ptr;
2974
2975 device->devid = btrfs_device_id(leaf, dev_item);
2976 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2977 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2978 device->type = btrfs_device_type(leaf, dev_item);
2979 device->io_align = btrfs_device_io_align(leaf, dev_item);
2980 device->io_width = btrfs_device_io_width(leaf, dev_item);
2981 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2982
2983 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2984 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2985
2986 return 0;
2987}
2988
2989static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2990{
2991 struct btrfs_fs_devices *fs_devices;
2992 int ret;
2993
2994 mutex_lock(&uuid_mutex);
2995
2996 fs_devices = root->fs_info->fs_devices->seed;
2997 while (fs_devices) {
2998 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2999 ret = 0;
3000 goto out;
3001 }
3002 fs_devices = fs_devices->seed;
3003 }
3004
3005 fs_devices = find_fsid(fsid);
3006 if (!fs_devices) {
3007 ret = -ENOENT;
3008 goto out;
3009 }
3010
3011 fs_devices = clone_fs_devices(fs_devices);
3012 if (IS_ERR(fs_devices)) {
3013 ret = PTR_ERR(fs_devices);
3014 goto out;
3015 }
3016
3017 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3018 root->fs_info->bdev_holder);
3019 if (ret)
3020 goto out;
3021
3022 if (!fs_devices->seeding) {
3023 __btrfs_close_devices(fs_devices);
3024 free_fs_devices(fs_devices);
3025 ret = -EINVAL;
3026 goto out;
3027 }
3028
3029 fs_devices->seed = root->fs_info->fs_devices->seed;
3030 root->fs_info->fs_devices->seed = fs_devices;
3031out:
3032 mutex_unlock(&uuid_mutex);
3033 return ret;
3034}
3035
3036static int read_one_dev(struct btrfs_root *root,
3037 struct extent_buffer *leaf,
3038 struct btrfs_dev_item *dev_item)
3039{
3040 struct btrfs_device *device;
3041 u64 devid;
3042 int ret;
3043 u8 fs_uuid[BTRFS_UUID_SIZE];
3044 u8 dev_uuid[BTRFS_UUID_SIZE];
3045
3046 devid = btrfs_device_id(leaf, dev_item);
3047 read_extent_buffer(leaf, dev_uuid,
3048 (unsigned long)btrfs_device_uuid(dev_item),
3049 BTRFS_UUID_SIZE);
3050 read_extent_buffer(leaf, fs_uuid,
3051 (unsigned long)btrfs_device_fsid(dev_item),
3052 BTRFS_UUID_SIZE);
3053
3054 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3055 ret = open_seed_devices(root, fs_uuid);
3056 if (ret && !btrfs_test_opt(root, DEGRADED))
3057 return ret;
3058 }
3059
3060 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3061 if (!device || !device->bdev) {
3062 if (!btrfs_test_opt(root, DEGRADED))
3063 return -EIO;
3064
3065 if (!device) {
3066 printk(KERN_WARNING "warning devid %llu missing\n",
3067 (unsigned long long)devid);
3068 device = add_missing_dev(root, devid, dev_uuid);
3069 if (!device)
3070 return -ENOMEM;
3071 }
3072 }
3073
3074 if (device->fs_devices != root->fs_info->fs_devices) {
3075 BUG_ON(device->writeable);
3076 if (device->generation !=
3077 btrfs_device_generation(leaf, dev_item))
3078 return -EINVAL;
3079 }
3080
3081 fill_device_from_item(leaf, dev_item, device);
3082 device->dev_root = root->fs_info->dev_root;
3083 device->in_fs_metadata = 1;
3084 if (device->writeable)
3085 device->fs_devices->total_rw_bytes += device->total_bytes;
3086 ret = 0;
3087 return ret;
3088}
3089
3090int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3091{
3092 struct btrfs_dev_item *dev_item;
3093
3094 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3095 dev_item);
3096 return read_one_dev(root, buf, dev_item);
3097}
3098
3099int btrfs_read_sys_array(struct btrfs_root *root)
3100{
3101 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3102 struct extent_buffer *sb;
3103 struct btrfs_disk_key *disk_key;
3104 struct btrfs_chunk *chunk;
3105 u8 *ptr;
3106 unsigned long sb_ptr;
3107 int ret = 0;
3108 u32 num_stripes;
3109 u32 array_size;
3110 u32 len = 0;
3111 u32 cur;
3112 struct btrfs_key key;
3113
3114 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3115 BTRFS_SUPER_INFO_SIZE);
3116 if (!sb)
3117 return -ENOMEM;
3118 btrfs_set_buffer_uptodate(sb);
3119 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3120 array_size = btrfs_super_sys_array_size(super_copy);
3121
3122 ptr = super_copy->sys_chunk_array;
3123 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3124 cur = 0;
3125
3126 while (cur < array_size) {
3127 disk_key = (struct btrfs_disk_key *)ptr;
3128 btrfs_disk_key_to_cpu(&key, disk_key);
3129
3130 len = sizeof(*disk_key); ptr += len;
3131 sb_ptr += len;
3132 cur += len;
3133
3134 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3135 chunk = (struct btrfs_chunk *)sb_ptr;
3136 ret = read_one_chunk(root, &key, sb, chunk);
3137 if (ret)
3138 break;
3139 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3140 len = btrfs_chunk_item_size(num_stripes);
3141 } else {
3142 ret = -EIO;
3143 break;
3144 }
3145 ptr += len;
3146 sb_ptr += len;
3147 cur += len;
3148 }
3149 free_extent_buffer(sb);
3150 return ret;
3151}
3152
3153int btrfs_read_chunk_tree(struct btrfs_root *root)
3154{
3155 struct btrfs_path *path;
3156 struct extent_buffer *leaf;
3157 struct btrfs_key key;
3158 struct btrfs_key found_key;
3159 int ret;
3160 int slot;
3161
3162 root = root->fs_info->chunk_root;
3163
3164 path = btrfs_alloc_path();
3165 if (!path)
3166 return -ENOMEM;
3167
3168 /* first we search for all of the device items, and then we
3169 * read in all of the chunk items. This way we can create chunk
3170 * mappings that reference all of the devices that are afound
3171 */
3172 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3173 key.offset = 0;
3174 key.type = 0;
3175again:
3176 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3177 while (1) {
3178 leaf = path->nodes[0];
3179 slot = path->slots[0];
3180 if (slot >= btrfs_header_nritems(leaf)) {
3181 ret = btrfs_next_leaf(root, path);
3182 if (ret == 0)
3183 continue;
3184 if (ret < 0)
3185 goto error;
3186 break;
3187 }
3188 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3189 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3190 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3191 break;
3192 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3193 struct btrfs_dev_item *dev_item;
3194 dev_item = btrfs_item_ptr(leaf, slot,
3195 struct btrfs_dev_item);
3196 ret = read_one_dev(root, leaf, dev_item);
3197 if (ret)
3198 goto error;
3199 }
3200 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3201 struct btrfs_chunk *chunk;
3202 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3203 ret = read_one_chunk(root, &found_key, leaf, chunk);
3204 if (ret)
3205 goto error;
3206 }
3207 path->slots[0]++;
3208 }
3209 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3210 key.objectid = 0;
3211 btrfs_release_path(root, path);
3212 goto again;
3213 }
3214 ret = 0;
3215error:
3216 btrfs_free_path(path);
3217 return ret;
3218}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-20 07:51:34 -0500