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.c3219
1 files changed, 3219 insertions, 0 deletions
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
new file mode 100644
index 000000000000..3451e1cca2b5
--- /dev/null
+++ b/fs/btrfs/volumes.c
@@ -0,0 +1,3219 @@
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 device->running_pending = 0;
224
225 spin_unlock(&device->io_lock);
226 btrfs_requeue_work(&device->work);
227 goto done;
228 }
229 }
230 if (again)
231 goto loop;
232done:
233 return 0;
234}
235
236static void pending_bios_fn(struct btrfs_work *work)
237{
238 struct btrfs_device *device;
239
240 device = container_of(work, struct btrfs_device, work);
241 run_scheduled_bios(device);
242}
243
244static noinline int device_list_add(const char *path,
245 struct btrfs_super_block *disk_super,
246 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
247{
248 struct btrfs_device *device;
249 struct btrfs_fs_devices *fs_devices;
250 u64 found_transid = btrfs_super_generation(disk_super);
251
252 fs_devices = find_fsid(disk_super->fsid);
253 if (!fs_devices) {
254 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
255 if (!fs_devices)
256 return -ENOMEM;
257 INIT_LIST_HEAD(&fs_devices->devices);
258 INIT_LIST_HEAD(&fs_devices->alloc_list);
259 list_add(&fs_devices->list, &fs_uuids);
260 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
261 fs_devices->latest_devid = devid;
262 fs_devices->latest_trans = found_transid;
263 device = NULL;
264 } else {
265 device = __find_device(&fs_devices->devices, devid,
266 disk_super->dev_item.uuid);
267 }
268 if (!device) {
269 if (fs_devices->opened)
270 return -EBUSY;
271
272 device = kzalloc(sizeof(*device), GFP_NOFS);
273 if (!device) {
274 /* we can safely leave the fs_devices entry around */
275 return -ENOMEM;
276 }
277 device->devid = devid;
278 device->work.func = pending_bios_fn;
279 memcpy(device->uuid, disk_super->dev_item.uuid,
280 BTRFS_UUID_SIZE);
281 device->barriers = 1;
282 spin_lock_init(&device->io_lock);
283 device->name = kstrdup(path, GFP_NOFS);
284 if (!device->name) {
285 kfree(device);
286 return -ENOMEM;
287 }
288 INIT_LIST_HEAD(&device->dev_alloc_list);
289 list_add(&device->dev_list, &fs_devices->devices);
290 device->fs_devices = fs_devices;
291 fs_devices->num_devices++;
292 }
293
294 if (found_transid > fs_devices->latest_trans) {
295 fs_devices->latest_devid = devid;
296 fs_devices->latest_trans = found_transid;
297 }
298 *fs_devices_ret = fs_devices;
299 return 0;
300}
301
302static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
303{
304 struct btrfs_fs_devices *fs_devices;
305 struct btrfs_device *device;
306 struct btrfs_device *orig_dev;
307
308 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
309 if (!fs_devices)
310 return ERR_PTR(-ENOMEM);
311
312 INIT_LIST_HEAD(&fs_devices->devices);
313 INIT_LIST_HEAD(&fs_devices->alloc_list);
314 INIT_LIST_HEAD(&fs_devices->list);
315 fs_devices->latest_devid = orig->latest_devid;
316 fs_devices->latest_trans = orig->latest_trans;
317 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
318
319 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
320 device = kzalloc(sizeof(*device), GFP_NOFS);
321 if (!device)
322 goto error;
323
324 device->name = kstrdup(orig_dev->name, GFP_NOFS);
325 if (!device->name)
326 goto error;
327
328 device->devid = orig_dev->devid;
329 device->work.func = pending_bios_fn;
330 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
331 device->barriers = 1;
332 spin_lock_init(&device->io_lock);
333 INIT_LIST_HEAD(&device->dev_list);
334 INIT_LIST_HEAD(&device->dev_alloc_list);
335
336 list_add(&device->dev_list, &fs_devices->devices);
337 device->fs_devices = fs_devices;
338 fs_devices->num_devices++;
339 }
340 return fs_devices;
341error:
342 free_fs_devices(fs_devices);
343 return ERR_PTR(-ENOMEM);
344}
345
346int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
347{
348 struct list_head *tmp;
349 struct list_head *cur;
350 struct btrfs_device *device;
351
352 mutex_lock(&uuid_mutex);
353again:
354 list_for_each_safe(cur, tmp, &fs_devices->devices) {
355 device = list_entry(cur, struct btrfs_device, dev_list);
356 if (device->in_fs_metadata)
357 continue;
358
359 if (device->bdev) {
360 close_bdev_exclusive(device->bdev, device->mode);
361 device->bdev = NULL;
362 fs_devices->open_devices--;
363 }
364 if (device->writeable) {
365 list_del_init(&device->dev_alloc_list);
366 device->writeable = 0;
367 fs_devices->rw_devices--;
368 }
369 list_del_init(&device->dev_list);
370 fs_devices->num_devices--;
371 kfree(device->name);
372 kfree(device);
373 }
374
375 if (fs_devices->seed) {
376 fs_devices = fs_devices->seed;
377 goto again;
378 }
379
380 mutex_unlock(&uuid_mutex);
381 return 0;
382}
383
384static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
385{
386 struct list_head *cur;
387 struct btrfs_device *device;
388
389 if (--fs_devices->opened > 0)
390 return 0;
391
392 list_for_each(cur, &fs_devices->devices) {
393 device = list_entry(cur, struct btrfs_device, dev_list);
394 if (device->bdev) {
395 close_bdev_exclusive(device->bdev, device->mode);
396 fs_devices->open_devices--;
397 }
398 if (device->writeable) {
399 list_del_init(&device->dev_alloc_list);
400 fs_devices->rw_devices--;
401 }
402
403 device->bdev = NULL;
404 device->writeable = 0;
405 device->in_fs_metadata = 0;
406 }
407 WARN_ON(fs_devices->open_devices);
408 WARN_ON(fs_devices->rw_devices);
409 fs_devices->opened = 0;
410 fs_devices->seeding = 0;
411
412 return 0;
413}
414
415int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
416{
417 struct btrfs_fs_devices *seed_devices = NULL;
418 int ret;
419
420 mutex_lock(&uuid_mutex);
421 ret = __btrfs_close_devices(fs_devices);
422 if (!fs_devices->opened) {
423 seed_devices = fs_devices->seed;
424 fs_devices->seed = NULL;
425 }
426 mutex_unlock(&uuid_mutex);
427
428 while (seed_devices) {
429 fs_devices = seed_devices;
430 seed_devices = fs_devices->seed;
431 __btrfs_close_devices(fs_devices);
432 free_fs_devices(fs_devices);
433 }
434 return ret;
435}
436
437static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
438 fmode_t flags, void *holder)
439{
440 struct block_device *bdev;
441 struct list_head *head = &fs_devices->devices;
442 struct list_head *cur;
443 struct btrfs_device *device;
444 struct block_device *latest_bdev = NULL;
445 struct buffer_head *bh;
446 struct btrfs_super_block *disk_super;
447 u64 latest_devid = 0;
448 u64 latest_transid = 0;
449 u64 devid;
450 int seeding = 1;
451 int ret = 0;
452
453 list_for_each(cur, head) {
454 device = list_entry(cur, struct btrfs_device, dev_list);
455 if (device->bdev)
456 continue;
457 if (!device->name)
458 continue;
459
460 bdev = open_bdev_exclusive(device->name, flags, holder);
461 if (IS_ERR(bdev)) {
462 printk(KERN_INFO "open %s failed\n", device->name);
463 goto error;
464 }
465 set_blocksize(bdev, 4096);
466
467 bh = btrfs_read_dev_super(bdev);
468 if (!bh)
469 goto error_close;
470
471 disk_super = (struct btrfs_super_block *)bh->b_data;
472 devid = le64_to_cpu(disk_super->dev_item.devid);
473 if (devid != device->devid)
474 goto error_brelse;
475
476 if (memcmp(device->uuid, disk_super->dev_item.uuid,
477 BTRFS_UUID_SIZE))
478 goto error_brelse;
479
480 device->generation = btrfs_super_generation(disk_super);
481 if (!latest_transid || device->generation > latest_transid) {
482 latest_devid = devid;
483 latest_transid = device->generation;
484 latest_bdev = bdev;
485 }
486
487 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
488 device->writeable = 0;
489 } else {
490 device->writeable = !bdev_read_only(bdev);
491 seeding = 0;
492 }
493
494 device->bdev = bdev;
495 device->in_fs_metadata = 0;
496 device->mode = flags;
497
498 fs_devices->open_devices++;
499 if (device->writeable) {
500 fs_devices->rw_devices++;
501 list_add(&device->dev_alloc_list,
502 &fs_devices->alloc_list);
503 }
504 continue;
505
506error_brelse:
507 brelse(bh);
508error_close:
509 close_bdev_exclusive(bdev, FMODE_READ);
510error:
511 continue;
512 }
513 if (fs_devices->open_devices == 0) {
514 ret = -EIO;
515 goto out;
516 }
517 fs_devices->seeding = seeding;
518 fs_devices->opened = 1;
519 fs_devices->latest_bdev = latest_bdev;
520 fs_devices->latest_devid = latest_devid;
521 fs_devices->latest_trans = latest_transid;
522 fs_devices->total_rw_bytes = 0;
523out:
524 return ret;
525}
526
527int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
528 fmode_t flags, void *holder)
529{
530 int ret;
531
532 mutex_lock(&uuid_mutex);
533 if (fs_devices->opened) {
534 fs_devices->opened++;
535 ret = 0;
536 } else {
537 ret = __btrfs_open_devices(fs_devices, flags, holder);
538 }
539 mutex_unlock(&uuid_mutex);
540 return ret;
541}
542
543int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
544 struct btrfs_fs_devices **fs_devices_ret)
545{
546 struct btrfs_super_block *disk_super;
547 struct block_device *bdev;
548 struct buffer_head *bh;
549 int ret;
550 u64 devid;
551 u64 transid;
552
553 mutex_lock(&uuid_mutex);
554
555 bdev = open_bdev_exclusive(path, flags, holder);
556
557 if (IS_ERR(bdev)) {
558 ret = PTR_ERR(bdev);
559 goto error;
560 }
561
562 ret = set_blocksize(bdev, 4096);
563 if (ret)
564 goto error_close;
565 bh = btrfs_read_dev_super(bdev);
566 if (!bh) {
567 ret = -EIO;
568 goto error_close;
569 }
570 disk_super = (struct btrfs_super_block *)bh->b_data;
571 devid = le64_to_cpu(disk_super->dev_item.devid);
572 transid = btrfs_super_generation(disk_super);
573 if (disk_super->label[0])
574 printk(KERN_INFO "device label %s ", disk_super->label);
575 else {
576 /* FIXME, make a readl uuid parser */
577 printk(KERN_INFO "device fsid %llx-%llx ",
578 *(unsigned long long *)disk_super->fsid,
579 *(unsigned long long *)(disk_super->fsid + 8));
580 }
581 printk(KERN_INFO "devid %llu transid %llu %s\n",
582 (unsigned long long)devid, (unsigned long long)transid, path);
583 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
584
585 brelse(bh);
586error_close:
587 close_bdev_exclusive(bdev, flags);
588error:
589 mutex_unlock(&uuid_mutex);
590 return ret;
591}
592
593/*
594 * this uses a pretty simple search, the expectation is that it is
595 * called very infrequently and that a given device has a small number
596 * of extents
597 */
598static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
599 struct btrfs_device *device,
600 u64 num_bytes, u64 *start)
601{
602 struct btrfs_key key;
603 struct btrfs_root *root = device->dev_root;
604 struct btrfs_dev_extent *dev_extent = NULL;
605 struct btrfs_path *path;
606 u64 hole_size = 0;
607 u64 last_byte = 0;
608 u64 search_start = 0;
609 u64 search_end = device->total_bytes;
610 int ret;
611 int slot = 0;
612 int start_found;
613 struct extent_buffer *l;
614
615 path = btrfs_alloc_path();
616 if (!path)
617 return -ENOMEM;
618 path->reada = 2;
619 start_found = 0;
620
621 /* FIXME use last free of some kind */
622
623 /* we don't want to overwrite the superblock on the drive,
624 * so we make sure to start at an offset of at least 1MB
625 */
626 search_start = max((u64)1024 * 1024, search_start);
627
628 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
629 search_start = max(root->fs_info->alloc_start, search_start);
630
631 key.objectid = device->devid;
632 key.offset = search_start;
633 key.type = BTRFS_DEV_EXTENT_KEY;
634 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
635 if (ret < 0)
636 goto error;
637 ret = btrfs_previous_item(root, path, 0, key.type);
638 if (ret < 0)
639 goto error;
640 l = path->nodes[0];
641 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
642 while (1) {
643 l = path->nodes[0];
644 slot = path->slots[0];
645 if (slot >= btrfs_header_nritems(l)) {
646 ret = btrfs_next_leaf(root, path);
647 if (ret == 0)
648 continue;
649 if (ret < 0)
650 goto error;
651no_more_items:
652 if (!start_found) {
653 if (search_start >= search_end) {
654 ret = -ENOSPC;
655 goto error;
656 }
657 *start = search_start;
658 start_found = 1;
659 goto check_pending;
660 }
661 *start = last_byte > search_start ?
662 last_byte : search_start;
663 if (search_end <= *start) {
664 ret = -ENOSPC;
665 goto error;
666 }
667 goto check_pending;
668 }
669 btrfs_item_key_to_cpu(l, &key, slot);
670
671 if (key.objectid < device->devid)
672 goto next;
673
674 if (key.objectid > device->devid)
675 goto no_more_items;
676
677 if (key.offset >= search_start && key.offset > last_byte &&
678 start_found) {
679 if (last_byte < search_start)
680 last_byte = search_start;
681 hole_size = key.offset - last_byte;
682 if (key.offset > last_byte &&
683 hole_size >= num_bytes) {
684 *start = last_byte;
685 goto check_pending;
686 }
687 }
688 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
689 goto next;
690
691 start_found = 1;
692 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
693 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
694next:
695 path->slots[0]++;
696 cond_resched();
697 }
698check_pending:
699 /* we have to make sure we didn't find an extent that has already
700 * been allocated by the map tree or the original allocation
701 */
702 BUG_ON(*start < search_start);
703
704 if (*start + num_bytes > search_end) {
705 ret = -ENOSPC;
706 goto error;
707 }
708 /* check for pending inserts here */
709 ret = 0;
710
711error:
712 btrfs_free_path(path);
713 return ret;
714}
715
716static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
717 struct btrfs_device *device,
718 u64 start)
719{
720 int ret;
721 struct btrfs_path *path;
722 struct btrfs_root *root = device->dev_root;
723 struct btrfs_key key;
724 struct btrfs_key found_key;
725 struct extent_buffer *leaf = NULL;
726 struct btrfs_dev_extent *extent = NULL;
727
728 path = btrfs_alloc_path();
729 if (!path)
730 return -ENOMEM;
731
732 key.objectid = device->devid;
733 key.offset = start;
734 key.type = BTRFS_DEV_EXTENT_KEY;
735
736 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
737 if (ret > 0) {
738 ret = btrfs_previous_item(root, path, key.objectid,
739 BTRFS_DEV_EXTENT_KEY);
740 BUG_ON(ret);
741 leaf = path->nodes[0];
742 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
743 extent = btrfs_item_ptr(leaf, path->slots[0],
744 struct btrfs_dev_extent);
745 BUG_ON(found_key.offset > start || found_key.offset +
746 btrfs_dev_extent_length(leaf, extent) < start);
747 ret = 0;
748 } else if (ret == 0) {
749 leaf = path->nodes[0];
750 extent = btrfs_item_ptr(leaf, path->slots[0],
751 struct btrfs_dev_extent);
752 }
753 BUG_ON(ret);
754
755 if (device->bytes_used > 0)
756 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
757 ret = btrfs_del_item(trans, root, path);
758 BUG_ON(ret);
759
760 btrfs_free_path(path);
761 return ret;
762}
763
764int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
765 struct btrfs_device *device,
766 u64 chunk_tree, u64 chunk_objectid,
767 u64 chunk_offset, u64 start, u64 num_bytes)
768{
769 int ret;
770 struct btrfs_path *path;
771 struct btrfs_root *root = device->dev_root;
772 struct btrfs_dev_extent *extent;
773 struct extent_buffer *leaf;
774 struct btrfs_key key;
775
776 WARN_ON(!device->in_fs_metadata);
777 path = btrfs_alloc_path();
778 if (!path)
779 return -ENOMEM;
780
781 key.objectid = device->devid;
782 key.offset = start;
783 key.type = BTRFS_DEV_EXTENT_KEY;
784 ret = btrfs_insert_empty_item(trans, root, path, &key,
785 sizeof(*extent));
786 BUG_ON(ret);
787
788 leaf = path->nodes[0];
789 extent = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_dev_extent);
791 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
792 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
793 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
794
795 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
796 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
797 BTRFS_UUID_SIZE);
798
799 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
800 btrfs_mark_buffer_dirty(leaf);
801 btrfs_free_path(path);
802 return ret;
803}
804
805static noinline int find_next_chunk(struct btrfs_root *root,
806 u64 objectid, u64 *offset)
807{
808 struct btrfs_path *path;
809 int ret;
810 struct btrfs_key key;
811 struct btrfs_chunk *chunk;
812 struct btrfs_key found_key;
813
814 path = btrfs_alloc_path();
815 BUG_ON(!path);
816
817 key.objectid = objectid;
818 key.offset = (u64)-1;
819 key.type = BTRFS_CHUNK_ITEM_KEY;
820
821 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
822 if (ret < 0)
823 goto error;
824
825 BUG_ON(ret == 0);
826
827 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
828 if (ret) {
829 *offset = 0;
830 } else {
831 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
832 path->slots[0]);
833 if (found_key.objectid != objectid)
834 *offset = 0;
835 else {
836 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
837 struct btrfs_chunk);
838 *offset = found_key.offset +
839 btrfs_chunk_length(path->nodes[0], chunk);
840 }
841 }
842 ret = 0;
843error:
844 btrfs_free_path(path);
845 return ret;
846}
847
848static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
849{
850 int ret;
851 struct btrfs_key key;
852 struct btrfs_key found_key;
853 struct btrfs_path *path;
854
855 root = root->fs_info->chunk_root;
856
857 path = btrfs_alloc_path();
858 if (!path)
859 return -ENOMEM;
860
861 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
862 key.type = BTRFS_DEV_ITEM_KEY;
863 key.offset = (u64)-1;
864
865 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
866 if (ret < 0)
867 goto error;
868
869 BUG_ON(ret == 0);
870
871 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
872 BTRFS_DEV_ITEM_KEY);
873 if (ret) {
874 *objectid = 1;
875 } else {
876 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
877 path->slots[0]);
878 *objectid = found_key.offset + 1;
879 }
880 ret = 0;
881error:
882 btrfs_free_path(path);
883 return ret;
884}
885
886/*
887 * the device information is stored in the chunk root
888 * the btrfs_device struct should be fully filled in
889 */
890int btrfs_add_device(struct btrfs_trans_handle *trans,
891 struct btrfs_root *root,
892 struct btrfs_device *device)
893{
894 int ret;
895 struct btrfs_path *path;
896 struct btrfs_dev_item *dev_item;
897 struct extent_buffer *leaf;
898 struct btrfs_key key;
899 unsigned long ptr;
900
901 root = root->fs_info->chunk_root;
902
903 path = btrfs_alloc_path();
904 if (!path)
905 return -ENOMEM;
906
907 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
908 key.type = BTRFS_DEV_ITEM_KEY;
909 key.offset = device->devid;
910
911 ret = btrfs_insert_empty_item(trans, root, path, &key,
912 sizeof(*dev_item));
913 if (ret)
914 goto out;
915
916 leaf = path->nodes[0];
917 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
918
919 btrfs_set_device_id(leaf, dev_item, device->devid);
920 btrfs_set_device_generation(leaf, dev_item, 0);
921 btrfs_set_device_type(leaf, dev_item, device->type);
922 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
923 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
924 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
925 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
926 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
927 btrfs_set_device_group(leaf, dev_item, 0);
928 btrfs_set_device_seek_speed(leaf, dev_item, 0);
929 btrfs_set_device_bandwidth(leaf, dev_item, 0);
930 btrfs_set_device_start_offset(leaf, dev_item, 0);
931
932 ptr = (unsigned long)btrfs_device_uuid(dev_item);
933 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
934 ptr = (unsigned long)btrfs_device_fsid(dev_item);
935 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
936 btrfs_mark_buffer_dirty(leaf);
937
938 ret = 0;
939out:
940 btrfs_free_path(path);
941 return ret;
942}
943
944static int btrfs_rm_dev_item(struct btrfs_root *root,
945 struct btrfs_device *device)
946{
947 int ret;
948 struct btrfs_path *path;
949 struct btrfs_key key;
950 struct btrfs_trans_handle *trans;
951
952 root = root->fs_info->chunk_root;
953
954 path = btrfs_alloc_path();
955 if (!path)
956 return -ENOMEM;
957
958 trans = btrfs_start_transaction(root, 1);
959 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
960 key.type = BTRFS_DEV_ITEM_KEY;
961 key.offset = device->devid;
962 lock_chunks(root);
963
964 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
965 if (ret < 0)
966 goto out;
967
968 if (ret > 0) {
969 ret = -ENOENT;
970 goto out;
971 }
972
973 ret = btrfs_del_item(trans, root, path);
974 if (ret)
975 goto out;
976out:
977 btrfs_free_path(path);
978 unlock_chunks(root);
979 btrfs_commit_transaction(trans, root);
980 return ret;
981}
982
983int btrfs_rm_device(struct btrfs_root *root, char *device_path)
984{
985 struct btrfs_device *device;
986 struct btrfs_device *next_device;
987 struct block_device *bdev;
988 struct buffer_head *bh = NULL;
989 struct btrfs_super_block *disk_super;
990 u64 all_avail;
991 u64 devid;
992 u64 num_devices;
993 u8 *dev_uuid;
994 int ret = 0;
995
996 mutex_lock(&uuid_mutex);
997 mutex_lock(&root->fs_info->volume_mutex);
998
999 all_avail = root->fs_info->avail_data_alloc_bits |
1000 root->fs_info->avail_system_alloc_bits |
1001 root->fs_info->avail_metadata_alloc_bits;
1002
1003 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1004 root->fs_info->fs_devices->rw_devices <= 4) {
1005 printk(KERN_ERR "btrfs: unable to go below four devices "
1006 "on raid10\n");
1007 ret = -EINVAL;
1008 goto out;
1009 }
1010
1011 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1012 root->fs_info->fs_devices->rw_devices <= 2) {
1013 printk(KERN_ERR "btrfs: unable to go below two "
1014 "devices on raid1\n");
1015 ret = -EINVAL;
1016 goto out;
1017 }
1018
1019 if (strcmp(device_path, "missing") == 0) {
1020 struct list_head *cur;
1021 struct list_head *devices;
1022 struct btrfs_device *tmp;
1023
1024 device = NULL;
1025 devices = &root->fs_info->fs_devices->devices;
1026 list_for_each(cur, devices) {
1027 tmp = list_entry(cur, struct btrfs_device, dev_list);
1028 if (tmp->in_fs_metadata && !tmp->bdev) {
1029 device = tmp;
1030 break;
1031 }
1032 }
1033 bdev = NULL;
1034 bh = NULL;
1035 disk_super = NULL;
1036 if (!device) {
1037 printk(KERN_ERR "btrfs: no missing devices found to "
1038 "remove\n");
1039 goto out;
1040 }
1041 } else {
1042 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1043 root->fs_info->bdev_holder);
1044 if (IS_ERR(bdev)) {
1045 ret = PTR_ERR(bdev);
1046 goto out;
1047 }
1048
1049 set_blocksize(bdev, 4096);
1050 bh = btrfs_read_dev_super(bdev);
1051 if (!bh) {
1052 ret = -EIO;
1053 goto error_close;
1054 }
1055 disk_super = (struct btrfs_super_block *)bh->b_data;
1056 devid = le64_to_cpu(disk_super->dev_item.devid);
1057 dev_uuid = disk_super->dev_item.uuid;
1058 device = btrfs_find_device(root, devid, dev_uuid,
1059 disk_super->fsid);
1060 if (!device) {
1061 ret = -ENOENT;
1062 goto error_brelse;
1063 }
1064 }
1065
1066 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1067 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1068 "device\n");
1069 ret = -EINVAL;
1070 goto error_brelse;
1071 }
1072
1073 if (device->writeable) {
1074 list_del_init(&device->dev_alloc_list);
1075 root->fs_info->fs_devices->rw_devices--;
1076 }
1077
1078 ret = btrfs_shrink_device(device, 0);
1079 if (ret)
1080 goto error_brelse;
1081
1082 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1083 if (ret)
1084 goto error_brelse;
1085
1086 device->in_fs_metadata = 0;
1087 list_del_init(&device->dev_list);
1088 device->fs_devices->num_devices--;
1089
1090 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1091 struct btrfs_device, dev_list);
1092 if (device->bdev == root->fs_info->sb->s_bdev)
1093 root->fs_info->sb->s_bdev = next_device->bdev;
1094 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1095 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1096
1097 if (device->bdev) {
1098 close_bdev_exclusive(device->bdev, device->mode);
1099 device->bdev = NULL;
1100 device->fs_devices->open_devices--;
1101 }
1102
1103 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1104 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1105
1106 if (device->fs_devices->open_devices == 0) {
1107 struct btrfs_fs_devices *fs_devices;
1108 fs_devices = root->fs_info->fs_devices;
1109 while (fs_devices) {
1110 if (fs_devices->seed == device->fs_devices)
1111 break;
1112 fs_devices = fs_devices->seed;
1113 }
1114 fs_devices->seed = device->fs_devices->seed;
1115 device->fs_devices->seed = NULL;
1116 __btrfs_close_devices(device->fs_devices);
1117 free_fs_devices(device->fs_devices);
1118 }
1119
1120 /*
1121 * at this point, the device is zero sized. We want to
1122 * remove it from the devices list and zero out the old super
1123 */
1124 if (device->writeable) {
1125 /* make sure this device isn't detected as part of
1126 * the FS anymore
1127 */
1128 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1129 set_buffer_dirty(bh);
1130 sync_dirty_buffer(bh);
1131 }
1132
1133 kfree(device->name);
1134 kfree(device);
1135 ret = 0;
1136
1137error_brelse:
1138 brelse(bh);
1139error_close:
1140 if (bdev)
1141 close_bdev_exclusive(bdev, FMODE_READ);
1142out:
1143 mutex_unlock(&root->fs_info->volume_mutex);
1144 mutex_unlock(&uuid_mutex);
1145 return ret;
1146}
1147
1148/*
1149 * does all the dirty work required for changing file system's UUID.
1150 */
1151static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1152 struct btrfs_root *root)
1153{
1154 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1155 struct btrfs_fs_devices *old_devices;
1156 struct btrfs_fs_devices *seed_devices;
1157 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1158 struct btrfs_device *device;
1159 u64 super_flags;
1160
1161 BUG_ON(!mutex_is_locked(&uuid_mutex));
1162 if (!fs_devices->seeding)
1163 return -EINVAL;
1164
1165 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1166 if (!seed_devices)
1167 return -ENOMEM;
1168
1169 old_devices = clone_fs_devices(fs_devices);
1170 if (IS_ERR(old_devices)) {
1171 kfree(seed_devices);
1172 return PTR_ERR(old_devices);
1173 }
1174
1175 list_add(&old_devices->list, &fs_uuids);
1176
1177 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1178 seed_devices->opened = 1;
1179 INIT_LIST_HEAD(&seed_devices->devices);
1180 INIT_LIST_HEAD(&seed_devices->alloc_list);
1181 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1182 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1183 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1184 device->fs_devices = seed_devices;
1185 }
1186
1187 fs_devices->seeding = 0;
1188 fs_devices->num_devices = 0;
1189 fs_devices->open_devices = 0;
1190 fs_devices->seed = seed_devices;
1191
1192 generate_random_uuid(fs_devices->fsid);
1193 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1194 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1195 super_flags = btrfs_super_flags(disk_super) &
1196 ~BTRFS_SUPER_FLAG_SEEDING;
1197 btrfs_set_super_flags(disk_super, super_flags);
1198
1199 return 0;
1200}
1201
1202/*
1203 * strore the expected generation for seed devices in device items.
1204 */
1205static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1206 struct btrfs_root *root)
1207{
1208 struct btrfs_path *path;
1209 struct extent_buffer *leaf;
1210 struct btrfs_dev_item *dev_item;
1211 struct btrfs_device *device;
1212 struct btrfs_key key;
1213 u8 fs_uuid[BTRFS_UUID_SIZE];
1214 u8 dev_uuid[BTRFS_UUID_SIZE];
1215 u64 devid;
1216 int ret;
1217
1218 path = btrfs_alloc_path();
1219 if (!path)
1220 return -ENOMEM;
1221
1222 root = root->fs_info->chunk_root;
1223 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1224 key.offset = 0;
1225 key.type = BTRFS_DEV_ITEM_KEY;
1226
1227 while (1) {
1228 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1229 if (ret < 0)
1230 goto error;
1231
1232 leaf = path->nodes[0];
1233next_slot:
1234 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1235 ret = btrfs_next_leaf(root, path);
1236 if (ret > 0)
1237 break;
1238 if (ret < 0)
1239 goto error;
1240 leaf = path->nodes[0];
1241 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1242 btrfs_release_path(root, path);
1243 continue;
1244 }
1245
1246 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1247 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1248 key.type != BTRFS_DEV_ITEM_KEY)
1249 break;
1250
1251 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1252 struct btrfs_dev_item);
1253 devid = btrfs_device_id(leaf, dev_item);
1254 read_extent_buffer(leaf, dev_uuid,
1255 (unsigned long)btrfs_device_uuid(dev_item),
1256 BTRFS_UUID_SIZE);
1257 read_extent_buffer(leaf, fs_uuid,
1258 (unsigned long)btrfs_device_fsid(dev_item),
1259 BTRFS_UUID_SIZE);
1260 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1261 BUG_ON(!device);
1262
1263 if (device->fs_devices->seeding) {
1264 btrfs_set_device_generation(leaf, dev_item,
1265 device->generation);
1266 btrfs_mark_buffer_dirty(leaf);
1267 }
1268
1269 path->slots[0]++;
1270 goto next_slot;
1271 }
1272 ret = 0;
1273error:
1274 btrfs_free_path(path);
1275 return ret;
1276}
1277
1278int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1279{
1280 struct btrfs_trans_handle *trans;
1281 struct btrfs_device *device;
1282 struct block_device *bdev;
1283 struct list_head *cur;
1284 struct list_head *devices;
1285 struct super_block *sb = root->fs_info->sb;
1286 u64 total_bytes;
1287 int seeding_dev = 0;
1288 int ret = 0;
1289
1290 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1291 return -EINVAL;
1292
1293 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1294 if (!bdev)
1295 return -EIO;
1296
1297 if (root->fs_info->fs_devices->seeding) {
1298 seeding_dev = 1;
1299 down_write(&sb->s_umount);
1300 mutex_lock(&uuid_mutex);
1301 }
1302
1303 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1304 mutex_lock(&root->fs_info->volume_mutex);
1305
1306 devices = &root->fs_info->fs_devices->devices;
1307 list_for_each(cur, devices) {
1308 device = list_entry(cur, struct btrfs_device, dev_list);
1309 if (device->bdev == bdev) {
1310 ret = -EEXIST;
1311 goto error;
1312 }
1313 }
1314
1315 device = kzalloc(sizeof(*device), GFP_NOFS);
1316 if (!device) {
1317 /* we can safely leave the fs_devices entry around */
1318 ret = -ENOMEM;
1319 goto error;
1320 }
1321
1322 device->name = kstrdup(device_path, GFP_NOFS);
1323 if (!device->name) {
1324 kfree(device);
1325 ret = -ENOMEM;
1326 goto error;
1327 }
1328
1329 ret = find_next_devid(root, &device->devid);
1330 if (ret) {
1331 kfree(device);
1332 goto error;
1333 }
1334
1335 trans = btrfs_start_transaction(root, 1);
1336 lock_chunks(root);
1337
1338 device->barriers = 1;
1339 device->writeable = 1;
1340 device->work.func = pending_bios_fn;
1341 generate_random_uuid(device->uuid);
1342 spin_lock_init(&device->io_lock);
1343 device->generation = trans->transid;
1344 device->io_width = root->sectorsize;
1345 device->io_align = root->sectorsize;
1346 device->sector_size = root->sectorsize;
1347 device->total_bytes = i_size_read(bdev->bd_inode);
1348 device->dev_root = root->fs_info->dev_root;
1349 device->bdev = bdev;
1350 device->in_fs_metadata = 1;
1351 device->mode = 0;
1352 set_blocksize(device->bdev, 4096);
1353
1354 if (seeding_dev) {
1355 sb->s_flags &= ~MS_RDONLY;
1356 ret = btrfs_prepare_sprout(trans, root);
1357 BUG_ON(ret);
1358 }
1359
1360 device->fs_devices = root->fs_info->fs_devices;
1361 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1362 list_add(&device->dev_alloc_list,
1363 &root->fs_info->fs_devices->alloc_list);
1364 root->fs_info->fs_devices->num_devices++;
1365 root->fs_info->fs_devices->open_devices++;
1366 root->fs_info->fs_devices->rw_devices++;
1367 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1368
1369 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1370 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1371 total_bytes + device->total_bytes);
1372
1373 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1374 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1375 total_bytes + 1);
1376
1377 if (seeding_dev) {
1378 ret = init_first_rw_device(trans, root, device);
1379 BUG_ON(ret);
1380 ret = btrfs_finish_sprout(trans, root);
1381 BUG_ON(ret);
1382 } else {
1383 ret = btrfs_add_device(trans, root, device);
1384 }
1385
1386 unlock_chunks(root);
1387 btrfs_commit_transaction(trans, root);
1388
1389 if (seeding_dev) {
1390 mutex_unlock(&uuid_mutex);
1391 up_write(&sb->s_umount);
1392
1393 ret = btrfs_relocate_sys_chunks(root);
1394 BUG_ON(ret);
1395 }
1396out:
1397 mutex_unlock(&root->fs_info->volume_mutex);
1398 return ret;
1399error:
1400 close_bdev_exclusive(bdev, 0);
1401 if (seeding_dev) {
1402 mutex_unlock(&uuid_mutex);
1403 up_write(&sb->s_umount);
1404 }
1405 goto out;
1406}
1407
1408static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1409 struct btrfs_device *device)
1410{
1411 int ret;
1412 struct btrfs_path *path;
1413 struct btrfs_root *root;
1414 struct btrfs_dev_item *dev_item;
1415 struct extent_buffer *leaf;
1416 struct btrfs_key key;
1417
1418 root = device->dev_root->fs_info->chunk_root;
1419
1420 path = btrfs_alloc_path();
1421 if (!path)
1422 return -ENOMEM;
1423
1424 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1425 key.type = BTRFS_DEV_ITEM_KEY;
1426 key.offset = device->devid;
1427
1428 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1429 if (ret < 0)
1430 goto out;
1431
1432 if (ret > 0) {
1433 ret = -ENOENT;
1434 goto out;
1435 }
1436
1437 leaf = path->nodes[0];
1438 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1439
1440 btrfs_set_device_id(leaf, dev_item, device->devid);
1441 btrfs_set_device_type(leaf, dev_item, device->type);
1442 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1443 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1444 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1445 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1446 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1447 btrfs_mark_buffer_dirty(leaf);
1448
1449out:
1450 btrfs_free_path(path);
1451 return ret;
1452}
1453
1454static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1455 struct btrfs_device *device, u64 new_size)
1456{
1457 struct btrfs_super_block *super_copy =
1458 &device->dev_root->fs_info->super_copy;
1459 u64 old_total = btrfs_super_total_bytes(super_copy);
1460 u64 diff = new_size - device->total_bytes;
1461
1462 if (!device->writeable)
1463 return -EACCES;
1464 if (new_size <= device->total_bytes)
1465 return -EINVAL;
1466
1467 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1468 device->fs_devices->total_rw_bytes += diff;
1469
1470 device->total_bytes = new_size;
1471 return btrfs_update_device(trans, device);
1472}
1473
1474int btrfs_grow_device(struct btrfs_trans_handle *trans,
1475 struct btrfs_device *device, u64 new_size)
1476{
1477 int ret;
1478 lock_chunks(device->dev_root);
1479 ret = __btrfs_grow_device(trans, device, new_size);
1480 unlock_chunks(device->dev_root);
1481 return ret;
1482}
1483
1484static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1485 struct btrfs_root *root,
1486 u64 chunk_tree, u64 chunk_objectid,
1487 u64 chunk_offset)
1488{
1489 int ret;
1490 struct btrfs_path *path;
1491 struct btrfs_key key;
1492
1493 root = root->fs_info->chunk_root;
1494 path = btrfs_alloc_path();
1495 if (!path)
1496 return -ENOMEM;
1497
1498 key.objectid = chunk_objectid;
1499 key.offset = chunk_offset;
1500 key.type = BTRFS_CHUNK_ITEM_KEY;
1501
1502 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1503 BUG_ON(ret);
1504
1505 ret = btrfs_del_item(trans, root, path);
1506 BUG_ON(ret);
1507
1508 btrfs_free_path(path);
1509 return 0;
1510}
1511
1512static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1513 chunk_offset)
1514{
1515 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1516 struct btrfs_disk_key *disk_key;
1517 struct btrfs_chunk *chunk;
1518 u8 *ptr;
1519 int ret = 0;
1520 u32 num_stripes;
1521 u32 array_size;
1522 u32 len = 0;
1523 u32 cur;
1524 struct btrfs_key key;
1525
1526 array_size = btrfs_super_sys_array_size(super_copy);
1527
1528 ptr = super_copy->sys_chunk_array;
1529 cur = 0;
1530
1531 while (cur < array_size) {
1532 disk_key = (struct btrfs_disk_key *)ptr;
1533 btrfs_disk_key_to_cpu(&key, disk_key);
1534
1535 len = sizeof(*disk_key);
1536
1537 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1538 chunk = (struct btrfs_chunk *)(ptr + len);
1539 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1540 len += btrfs_chunk_item_size(num_stripes);
1541 } else {
1542 ret = -EIO;
1543 break;
1544 }
1545 if (key.objectid == chunk_objectid &&
1546 key.offset == chunk_offset) {
1547 memmove(ptr, ptr + len, array_size - (cur + len));
1548 array_size -= len;
1549 btrfs_set_super_sys_array_size(super_copy, array_size);
1550 } else {
1551 ptr += len;
1552 cur += len;
1553 }
1554 }
1555 return ret;
1556}
1557
1558static int btrfs_relocate_chunk(struct btrfs_root *root,
1559 u64 chunk_tree, u64 chunk_objectid,
1560 u64 chunk_offset)
1561{
1562 struct extent_map_tree *em_tree;
1563 struct btrfs_root *extent_root;
1564 struct btrfs_trans_handle *trans;
1565 struct extent_map *em;
1566 struct map_lookup *map;
1567 int ret;
1568 int i;
1569
1570 printk(KERN_INFO "btrfs relocating chunk %llu\n",
1571 (unsigned long long)chunk_offset);
1572 root = root->fs_info->chunk_root;
1573 extent_root = root->fs_info->extent_root;
1574 em_tree = &root->fs_info->mapping_tree.map_tree;
1575
1576 /* step one, relocate all the extents inside this chunk */
1577 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1578 BUG_ON(ret);
1579
1580 trans = btrfs_start_transaction(root, 1);
1581 BUG_ON(!trans);
1582
1583 lock_chunks(root);
1584
1585 /*
1586 * step two, delete the device extents and the
1587 * chunk tree entries
1588 */
1589 spin_lock(&em_tree->lock);
1590 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1591 spin_unlock(&em_tree->lock);
1592
1593 BUG_ON(em->start > chunk_offset ||
1594 em->start + em->len < chunk_offset);
1595 map = (struct map_lookup *)em->bdev;
1596
1597 for (i = 0; i < map->num_stripes; i++) {
1598 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1599 map->stripes[i].physical);
1600 BUG_ON(ret);
1601
1602 if (map->stripes[i].dev) {
1603 ret = btrfs_update_device(trans, map->stripes[i].dev);
1604 BUG_ON(ret);
1605 }
1606 }
1607 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1608 chunk_offset);
1609
1610 BUG_ON(ret);
1611
1612 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1613 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1614 BUG_ON(ret);
1615 }
1616
1617 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1618 BUG_ON(ret);
1619
1620 spin_lock(&em_tree->lock);
1621 remove_extent_mapping(em_tree, em);
1622 spin_unlock(&em_tree->lock);
1623
1624 kfree(map);
1625 em->bdev = NULL;
1626
1627 /* once for the tree */
1628 free_extent_map(em);
1629 /* once for us */
1630 free_extent_map(em);
1631
1632 unlock_chunks(root);
1633 btrfs_end_transaction(trans, root);
1634 return 0;
1635}
1636
1637static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1638{
1639 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1640 struct btrfs_path *path;
1641 struct extent_buffer *leaf;
1642 struct btrfs_chunk *chunk;
1643 struct btrfs_key key;
1644 struct btrfs_key found_key;
1645 u64 chunk_tree = chunk_root->root_key.objectid;
1646 u64 chunk_type;
1647 int ret;
1648
1649 path = btrfs_alloc_path();
1650 if (!path)
1651 return -ENOMEM;
1652
1653 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1654 key.offset = (u64)-1;
1655 key.type = BTRFS_CHUNK_ITEM_KEY;
1656
1657 while (1) {
1658 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1659 if (ret < 0)
1660 goto error;
1661 BUG_ON(ret == 0);
1662
1663 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1664 key.type);
1665 if (ret < 0)
1666 goto error;
1667 if (ret > 0)
1668 break;
1669
1670 leaf = path->nodes[0];
1671 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1672
1673 chunk = btrfs_item_ptr(leaf, path->slots[0],
1674 struct btrfs_chunk);
1675 chunk_type = btrfs_chunk_type(leaf, chunk);
1676 btrfs_release_path(chunk_root, path);
1677
1678 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1679 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1680 found_key.objectid,
1681 found_key.offset);
1682 BUG_ON(ret);
1683 }
1684
1685 if (found_key.offset == 0)
1686 break;
1687 key.offset = found_key.offset - 1;
1688 }
1689 ret = 0;
1690error:
1691 btrfs_free_path(path);
1692 return ret;
1693}
1694
1695static u64 div_factor(u64 num, int factor)
1696{
1697 if (factor == 10)
1698 return num;
1699 num *= factor;
1700 do_div(num, 10);
1701 return num;
1702}
1703
1704int btrfs_balance(struct btrfs_root *dev_root)
1705{
1706 int ret;
1707 struct list_head *cur;
1708 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1709 struct btrfs_device *device;
1710 u64 old_size;
1711 u64 size_to_free;
1712 struct btrfs_path *path;
1713 struct btrfs_key key;
1714 struct btrfs_chunk *chunk;
1715 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1716 struct btrfs_trans_handle *trans;
1717 struct btrfs_key found_key;
1718
1719 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1720 return -EROFS;
1721
1722 mutex_lock(&dev_root->fs_info->volume_mutex);
1723 dev_root = dev_root->fs_info->dev_root;
1724
1725 /* step one make some room on all the devices */
1726 list_for_each(cur, devices) {
1727 device = list_entry(cur, struct btrfs_device, dev_list);
1728 old_size = device->total_bytes;
1729 size_to_free = div_factor(old_size, 1);
1730 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1731 if (!device->writeable ||
1732 device->total_bytes - device->bytes_used > size_to_free)
1733 continue;
1734
1735 ret = btrfs_shrink_device(device, old_size - size_to_free);
1736 BUG_ON(ret);
1737
1738 trans = btrfs_start_transaction(dev_root, 1);
1739 BUG_ON(!trans);
1740
1741 ret = btrfs_grow_device(trans, device, old_size);
1742 BUG_ON(ret);
1743
1744 btrfs_end_transaction(trans, dev_root);
1745 }
1746
1747 /* step two, relocate all the chunks */
1748 path = btrfs_alloc_path();
1749 BUG_ON(!path);
1750
1751 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1752 key.offset = (u64)-1;
1753 key.type = BTRFS_CHUNK_ITEM_KEY;
1754
1755 while (1) {
1756 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1757 if (ret < 0)
1758 goto error;
1759
1760 /*
1761 * this shouldn't happen, it means the last relocate
1762 * failed
1763 */
1764 if (ret == 0)
1765 break;
1766
1767 ret = btrfs_previous_item(chunk_root, path, 0,
1768 BTRFS_CHUNK_ITEM_KEY);
1769 if (ret)
1770 break;
1771
1772 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1773 path->slots[0]);
1774 if (found_key.objectid != key.objectid)
1775 break;
1776
1777 chunk = btrfs_item_ptr(path->nodes[0],
1778 path->slots[0],
1779 struct btrfs_chunk);
1780 key.offset = found_key.offset;
1781 /* chunk zero is special */
1782 if (key.offset == 0)
1783 break;
1784
1785 btrfs_release_path(chunk_root, path);
1786 ret = btrfs_relocate_chunk(chunk_root,
1787 chunk_root->root_key.objectid,
1788 found_key.objectid,
1789 found_key.offset);
1790 BUG_ON(ret);
1791 }
1792 ret = 0;
1793error:
1794 btrfs_free_path(path);
1795 mutex_unlock(&dev_root->fs_info->volume_mutex);
1796 return ret;
1797}
1798
1799/*
1800 * shrinking a device means finding all of the device extents past
1801 * the new size, and then following the back refs to the chunks.
1802 * The chunk relocation code actually frees the device extent
1803 */
1804int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1805{
1806 struct btrfs_trans_handle *trans;
1807 struct btrfs_root *root = device->dev_root;
1808 struct btrfs_dev_extent *dev_extent = NULL;
1809 struct btrfs_path *path;
1810 u64 length;
1811 u64 chunk_tree;
1812 u64 chunk_objectid;
1813 u64 chunk_offset;
1814 int ret;
1815 int slot;
1816 struct extent_buffer *l;
1817 struct btrfs_key key;
1818 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1819 u64 old_total = btrfs_super_total_bytes(super_copy);
1820 u64 diff = device->total_bytes - new_size;
1821
1822 if (new_size >= device->total_bytes)
1823 return -EINVAL;
1824
1825 path = btrfs_alloc_path();
1826 if (!path)
1827 return -ENOMEM;
1828
1829 trans = btrfs_start_transaction(root, 1);
1830 if (!trans) {
1831 ret = -ENOMEM;
1832 goto done;
1833 }
1834
1835 path->reada = 2;
1836
1837 lock_chunks(root);
1838
1839 device->total_bytes = new_size;
1840 if (device->writeable)
1841 device->fs_devices->total_rw_bytes -= diff;
1842 ret = btrfs_update_device(trans, device);
1843 if (ret) {
1844 unlock_chunks(root);
1845 btrfs_end_transaction(trans, root);
1846 goto done;
1847 }
1848 WARN_ON(diff > old_total);
1849 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1850 unlock_chunks(root);
1851 btrfs_end_transaction(trans, root);
1852
1853 key.objectid = device->devid;
1854 key.offset = (u64)-1;
1855 key.type = BTRFS_DEV_EXTENT_KEY;
1856
1857 while (1) {
1858 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1859 if (ret < 0)
1860 goto done;
1861
1862 ret = btrfs_previous_item(root, path, 0, key.type);
1863 if (ret < 0)
1864 goto done;
1865 if (ret) {
1866 ret = 0;
1867 goto done;
1868 }
1869
1870 l = path->nodes[0];
1871 slot = path->slots[0];
1872 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1873
1874 if (key.objectid != device->devid)
1875 goto done;
1876
1877 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1878 length = btrfs_dev_extent_length(l, dev_extent);
1879
1880 if (key.offset + length <= new_size)
1881 goto done;
1882
1883 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1884 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1885 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1886 btrfs_release_path(root, path);
1887
1888 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1889 chunk_offset);
1890 if (ret)
1891 goto done;
1892 }
1893
1894done:
1895 btrfs_free_path(path);
1896 return ret;
1897}
1898
1899static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1900 struct btrfs_root *root,
1901 struct btrfs_key *key,
1902 struct btrfs_chunk *chunk, int item_size)
1903{
1904 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1905 struct btrfs_disk_key disk_key;
1906 u32 array_size;
1907 u8 *ptr;
1908
1909 array_size = btrfs_super_sys_array_size(super_copy);
1910 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1911 return -EFBIG;
1912
1913 ptr = super_copy->sys_chunk_array + array_size;
1914 btrfs_cpu_key_to_disk(&disk_key, key);
1915 memcpy(ptr, &disk_key, sizeof(disk_key));
1916 ptr += sizeof(disk_key);
1917 memcpy(ptr, chunk, item_size);
1918 item_size += sizeof(disk_key);
1919 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1920 return 0;
1921}
1922
1923static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
1924 int num_stripes, int sub_stripes)
1925{
1926 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1927 return calc_size;
1928 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1929 return calc_size * (num_stripes / sub_stripes);
1930 else
1931 return calc_size * num_stripes;
1932}
1933
1934static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1935 struct btrfs_root *extent_root,
1936 struct map_lookup **map_ret,
1937 u64 *num_bytes, u64 *stripe_size,
1938 u64 start, u64 type)
1939{
1940 struct btrfs_fs_info *info = extent_root->fs_info;
1941 struct btrfs_device *device = NULL;
1942 struct btrfs_fs_devices *fs_devices = info->fs_devices;
1943 struct list_head *cur;
1944 struct map_lookup *map = NULL;
1945 struct extent_map_tree *em_tree;
1946 struct extent_map *em;
1947 struct list_head private_devs;
1948 int min_stripe_size = 1 * 1024 * 1024;
1949 u64 calc_size = 1024 * 1024 * 1024;
1950 u64 max_chunk_size = calc_size;
1951 u64 min_free;
1952 u64 avail;
1953 u64 max_avail = 0;
1954 u64 dev_offset;
1955 int num_stripes = 1;
1956 int min_stripes = 1;
1957 int sub_stripes = 0;
1958 int looped = 0;
1959 int ret;
1960 int index;
1961 int stripe_len = 64 * 1024;
1962
1963 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1964 (type & BTRFS_BLOCK_GROUP_DUP)) {
1965 WARN_ON(1);
1966 type &= ~BTRFS_BLOCK_GROUP_DUP;
1967 }
1968 if (list_empty(&fs_devices->alloc_list))
1969 return -ENOSPC;
1970
1971 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1972 num_stripes = fs_devices->rw_devices;
1973 min_stripes = 2;
1974 }
1975 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1976 num_stripes = 2;
1977 min_stripes = 2;
1978 }
1979 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1980 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1981 if (num_stripes < 2)
1982 return -ENOSPC;
1983 min_stripes = 2;
1984 }
1985 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1986 num_stripes = fs_devices->rw_devices;
1987 if (num_stripes < 4)
1988 return -ENOSPC;
1989 num_stripes &= ~(u32)1;
1990 sub_stripes = 2;
1991 min_stripes = 4;
1992 }
1993
1994 if (type & BTRFS_BLOCK_GROUP_DATA) {
1995 max_chunk_size = 10 * calc_size;
1996 min_stripe_size = 64 * 1024 * 1024;
1997 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1998 max_chunk_size = 4 * calc_size;
1999 min_stripe_size = 32 * 1024 * 1024;
2000 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2001 calc_size = 8 * 1024 * 1024;
2002 max_chunk_size = calc_size * 2;
2003 min_stripe_size = 1 * 1024 * 1024;
2004 }
2005
2006 /* we don't want a chunk larger than 10% of writeable space */
2007 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2008 max_chunk_size);
2009
2010again:
2011 if (!map || map->num_stripes != num_stripes) {
2012 kfree(map);
2013 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2014 if (!map)
2015 return -ENOMEM;
2016 map->num_stripes = num_stripes;
2017 }
2018
2019 if (calc_size * num_stripes > max_chunk_size) {
2020 calc_size = max_chunk_size;
2021 do_div(calc_size, num_stripes);
2022 do_div(calc_size, stripe_len);
2023 calc_size *= stripe_len;
2024 }
2025 /* we don't want tiny stripes */
2026 calc_size = max_t(u64, min_stripe_size, calc_size);
2027
2028 do_div(calc_size, stripe_len);
2029 calc_size *= stripe_len;
2030
2031 cur = fs_devices->alloc_list.next;
2032 index = 0;
2033
2034 if (type & BTRFS_BLOCK_GROUP_DUP)
2035 min_free = calc_size * 2;
2036 else
2037 min_free = calc_size;
2038
2039 /*
2040 * we add 1MB because we never use the first 1MB of the device, unless
2041 * we've looped, then we are likely allocating the maximum amount of
2042 * space left already
2043 */
2044 if (!looped)
2045 min_free += 1024 * 1024;
2046
2047 INIT_LIST_HEAD(&private_devs);
2048 while (index < num_stripes) {
2049 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2050 BUG_ON(!device->writeable);
2051 if (device->total_bytes > device->bytes_used)
2052 avail = device->total_bytes - device->bytes_used;
2053 else
2054 avail = 0;
2055 cur = cur->next;
2056
2057 if (device->in_fs_metadata && avail >= min_free) {
2058 ret = find_free_dev_extent(trans, device,
2059 min_free, &dev_offset);
2060 if (ret == 0) {
2061 list_move_tail(&device->dev_alloc_list,
2062 &private_devs);
2063 map->stripes[index].dev = device;
2064 map->stripes[index].physical = dev_offset;
2065 index++;
2066 if (type & BTRFS_BLOCK_GROUP_DUP) {
2067 map->stripes[index].dev = device;
2068 map->stripes[index].physical =
2069 dev_offset + calc_size;
2070 index++;
2071 }
2072 }
2073 } else if (device->in_fs_metadata && avail > max_avail)
2074 max_avail = avail;
2075 if (cur == &fs_devices->alloc_list)
2076 break;
2077 }
2078 list_splice(&private_devs, &fs_devices->alloc_list);
2079 if (index < num_stripes) {
2080 if (index >= min_stripes) {
2081 num_stripes = index;
2082 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2083 num_stripes /= sub_stripes;
2084 num_stripes *= sub_stripes;
2085 }
2086 looped = 1;
2087 goto again;
2088 }
2089 if (!looped && max_avail > 0) {
2090 looped = 1;
2091 calc_size = max_avail;
2092 goto again;
2093 }
2094 kfree(map);
2095 return -ENOSPC;
2096 }
2097 map->sector_size = extent_root->sectorsize;
2098 map->stripe_len = stripe_len;
2099 map->io_align = stripe_len;
2100 map->io_width = stripe_len;
2101 map->type = type;
2102 map->num_stripes = num_stripes;
2103 map->sub_stripes = sub_stripes;
2104
2105 *map_ret = map;
2106 *stripe_size = calc_size;
2107 *num_bytes = chunk_bytes_by_type(type, calc_size,
2108 num_stripes, sub_stripes);
2109
2110 em = alloc_extent_map(GFP_NOFS);
2111 if (!em) {
2112 kfree(map);
2113 return -ENOMEM;
2114 }
2115 em->bdev = (struct block_device *)map;
2116 em->start = start;
2117 em->len = *num_bytes;
2118 em->block_start = 0;
2119 em->block_len = em->len;
2120
2121 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2122 spin_lock(&em_tree->lock);
2123 ret = add_extent_mapping(em_tree, em);
2124 spin_unlock(&em_tree->lock);
2125 BUG_ON(ret);
2126 free_extent_map(em);
2127
2128 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2129 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2130 start, *num_bytes);
2131 BUG_ON(ret);
2132
2133 index = 0;
2134 while (index < map->num_stripes) {
2135 device = map->stripes[index].dev;
2136 dev_offset = map->stripes[index].physical;
2137
2138 ret = btrfs_alloc_dev_extent(trans, device,
2139 info->chunk_root->root_key.objectid,
2140 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2141 start, dev_offset, calc_size);
2142 BUG_ON(ret);
2143 index++;
2144 }
2145
2146 return 0;
2147}
2148
2149static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2150 struct btrfs_root *extent_root,
2151 struct map_lookup *map, u64 chunk_offset,
2152 u64 chunk_size, u64 stripe_size)
2153{
2154 u64 dev_offset;
2155 struct btrfs_key key;
2156 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2157 struct btrfs_device *device;
2158 struct btrfs_chunk *chunk;
2159 struct btrfs_stripe *stripe;
2160 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2161 int index = 0;
2162 int ret;
2163
2164 chunk = kzalloc(item_size, GFP_NOFS);
2165 if (!chunk)
2166 return -ENOMEM;
2167
2168 index = 0;
2169 while (index < map->num_stripes) {
2170 device = map->stripes[index].dev;
2171 device->bytes_used += stripe_size;
2172 ret = btrfs_update_device(trans, device);
2173 BUG_ON(ret);
2174 index++;
2175 }
2176
2177 index = 0;
2178 stripe = &chunk->stripe;
2179 while (index < map->num_stripes) {
2180 device = map->stripes[index].dev;
2181 dev_offset = map->stripes[index].physical;
2182
2183 btrfs_set_stack_stripe_devid(stripe, device->devid);
2184 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2185 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2186 stripe++;
2187 index++;
2188 }
2189
2190 btrfs_set_stack_chunk_length(chunk, chunk_size);
2191 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2192 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2193 btrfs_set_stack_chunk_type(chunk, map->type);
2194 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2195 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2196 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2197 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2198 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2199
2200 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2201 key.type = BTRFS_CHUNK_ITEM_KEY;
2202 key.offset = chunk_offset;
2203
2204 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2205 BUG_ON(ret);
2206
2207 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2208 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2209 item_size);
2210 BUG_ON(ret);
2211 }
2212 kfree(chunk);
2213 return 0;
2214}
2215
2216/*
2217 * Chunk allocation falls into two parts. The first part does works
2218 * that make the new allocated chunk useable, but not do any operation
2219 * that modifies the chunk tree. The second part does the works that
2220 * require modifying the chunk tree. This division is important for the
2221 * bootstrap process of adding storage to a seed btrfs.
2222 */
2223int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2224 struct btrfs_root *extent_root, u64 type)
2225{
2226 u64 chunk_offset;
2227 u64 chunk_size;
2228 u64 stripe_size;
2229 struct map_lookup *map;
2230 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2231 int ret;
2232
2233 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2234 &chunk_offset);
2235 if (ret)
2236 return ret;
2237
2238 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2239 &stripe_size, chunk_offset, type);
2240 if (ret)
2241 return ret;
2242
2243 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2244 chunk_size, stripe_size);
2245 BUG_ON(ret);
2246 return 0;
2247}
2248
2249static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2250 struct btrfs_root *root,
2251 struct btrfs_device *device)
2252{
2253 u64 chunk_offset;
2254 u64 sys_chunk_offset;
2255 u64 chunk_size;
2256 u64 sys_chunk_size;
2257 u64 stripe_size;
2258 u64 sys_stripe_size;
2259 u64 alloc_profile;
2260 struct map_lookup *map;
2261 struct map_lookup *sys_map;
2262 struct btrfs_fs_info *fs_info = root->fs_info;
2263 struct btrfs_root *extent_root = fs_info->extent_root;
2264 int ret;
2265
2266 ret = find_next_chunk(fs_info->chunk_root,
2267 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2268 BUG_ON(ret);
2269
2270 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2271 (fs_info->metadata_alloc_profile &
2272 fs_info->avail_metadata_alloc_bits);
2273 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2274
2275 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2276 &stripe_size, chunk_offset, alloc_profile);
2277 BUG_ON(ret);
2278
2279 sys_chunk_offset = chunk_offset + chunk_size;
2280
2281 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2282 (fs_info->system_alloc_profile &
2283 fs_info->avail_system_alloc_bits);
2284 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2285
2286 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2287 &sys_chunk_size, &sys_stripe_size,
2288 sys_chunk_offset, alloc_profile);
2289 BUG_ON(ret);
2290
2291 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2292 BUG_ON(ret);
2293
2294 /*
2295 * Modifying chunk tree needs allocating new blocks from both
2296 * system block group and metadata block group. So we only can
2297 * do operations require modifying the chunk tree after both
2298 * block groups were created.
2299 */
2300 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2301 chunk_size, stripe_size);
2302 BUG_ON(ret);
2303
2304 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2305 sys_chunk_offset, sys_chunk_size,
2306 sys_stripe_size);
2307 BUG_ON(ret);
2308 return 0;
2309}
2310
2311int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2312{
2313 struct extent_map *em;
2314 struct map_lookup *map;
2315 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2316 int readonly = 0;
2317 int i;
2318
2319 spin_lock(&map_tree->map_tree.lock);
2320 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2321 spin_unlock(&map_tree->map_tree.lock);
2322 if (!em)
2323 return 1;
2324
2325 map = (struct map_lookup *)em->bdev;
2326 for (i = 0; i < map->num_stripes; i++) {
2327 if (!map->stripes[i].dev->writeable) {
2328 readonly = 1;
2329 break;
2330 }
2331 }
2332 free_extent_map(em);
2333 return readonly;
2334}
2335
2336void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2337{
2338 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2339}
2340
2341void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2342{
2343 struct extent_map *em;
2344
2345 while (1) {
2346 spin_lock(&tree->map_tree.lock);
2347 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2348 if (em)
2349 remove_extent_mapping(&tree->map_tree, em);
2350 spin_unlock(&tree->map_tree.lock);
2351 if (!em)
2352 break;
2353 kfree(em->bdev);
2354 /* once for us */
2355 free_extent_map(em);
2356 /* once for the tree */
2357 free_extent_map(em);
2358 }
2359}
2360
2361int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2362{
2363 struct extent_map *em;
2364 struct map_lookup *map;
2365 struct extent_map_tree *em_tree = &map_tree->map_tree;
2366 int ret;
2367
2368 spin_lock(&em_tree->lock);
2369 em = lookup_extent_mapping(em_tree, logical, len);
2370 spin_unlock(&em_tree->lock);
2371 BUG_ON(!em);
2372
2373 BUG_ON(em->start > logical || em->start + em->len < logical);
2374 map = (struct map_lookup *)em->bdev;
2375 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2376 ret = map->num_stripes;
2377 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2378 ret = map->sub_stripes;
2379 else
2380 ret = 1;
2381 free_extent_map(em);
2382 return ret;
2383}
2384
2385static int find_live_mirror(struct map_lookup *map, int first, int num,
2386 int optimal)
2387{
2388 int i;
2389 if (map->stripes[optimal].dev->bdev)
2390 return optimal;
2391 for (i = first; i < first + num; i++) {
2392 if (map->stripes[i].dev->bdev)
2393 return i;
2394 }
2395 /* we couldn't find one that doesn't fail. Just return something
2396 * and the io error handling code will clean up eventually
2397 */
2398 return optimal;
2399}
2400
2401static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2402 u64 logical, u64 *length,
2403 struct btrfs_multi_bio **multi_ret,
2404 int mirror_num, struct page *unplug_page)
2405{
2406 struct extent_map *em;
2407 struct map_lookup *map;
2408 struct extent_map_tree *em_tree = &map_tree->map_tree;
2409 u64 offset;
2410 u64 stripe_offset;
2411 u64 stripe_nr;
2412 int stripes_allocated = 8;
2413 int stripes_required = 1;
2414 int stripe_index;
2415 int i;
2416 int num_stripes;
2417 int max_errors = 0;
2418 struct btrfs_multi_bio *multi = NULL;
2419
2420 if (multi_ret && !(rw & (1 << BIO_RW)))
2421 stripes_allocated = 1;
2422again:
2423 if (multi_ret) {
2424 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2425 GFP_NOFS);
2426 if (!multi)
2427 return -ENOMEM;
2428
2429 atomic_set(&multi->error, 0);
2430 }
2431
2432 spin_lock(&em_tree->lock);
2433 em = lookup_extent_mapping(em_tree, logical, *length);
2434 spin_unlock(&em_tree->lock);
2435
2436 if (!em && unplug_page)
2437 return 0;
2438
2439 if (!em) {
2440 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2441 (unsigned long long)logical,
2442 (unsigned long long)*length);
2443 BUG();
2444 }
2445
2446 BUG_ON(em->start > logical || em->start + em->len < logical);
2447 map = (struct map_lookup *)em->bdev;
2448 offset = logical - em->start;
2449
2450 if (mirror_num > map->num_stripes)
2451 mirror_num = 0;
2452
2453 /* if our multi bio struct is too small, back off and try again */
2454 if (rw & (1 << BIO_RW)) {
2455 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2456 BTRFS_BLOCK_GROUP_DUP)) {
2457 stripes_required = map->num_stripes;
2458 max_errors = 1;
2459 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2460 stripes_required = map->sub_stripes;
2461 max_errors = 1;
2462 }
2463 }
2464 if (multi_ret && rw == WRITE &&
2465 stripes_allocated < stripes_required) {
2466 stripes_allocated = map->num_stripes;
2467 free_extent_map(em);
2468 kfree(multi);
2469 goto again;
2470 }
2471 stripe_nr = offset;
2472 /*
2473 * stripe_nr counts the total number of stripes we have to stride
2474 * to get to this block
2475 */
2476 do_div(stripe_nr, map->stripe_len);
2477
2478 stripe_offset = stripe_nr * map->stripe_len;
2479 BUG_ON(offset < stripe_offset);
2480
2481 /* stripe_offset is the offset of this block in its stripe*/
2482 stripe_offset = offset - stripe_offset;
2483
2484 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2485 BTRFS_BLOCK_GROUP_RAID10 |
2486 BTRFS_BLOCK_GROUP_DUP)) {
2487 /* we limit the length of each bio to what fits in a stripe */
2488 *length = min_t(u64, em->len - offset,
2489 map->stripe_len - stripe_offset);
2490 } else {
2491 *length = em->len - offset;
2492 }
2493
2494 if (!multi_ret && !unplug_page)
2495 goto out;
2496
2497 num_stripes = 1;
2498 stripe_index = 0;
2499 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2500 if (unplug_page || (rw & (1 << BIO_RW)))
2501 num_stripes = map->num_stripes;
2502 else if (mirror_num)
2503 stripe_index = mirror_num - 1;
2504 else {
2505 stripe_index = find_live_mirror(map, 0,
2506 map->num_stripes,
2507 current->pid % map->num_stripes);
2508 }
2509
2510 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2511 if (rw & (1 << BIO_RW))
2512 num_stripes = map->num_stripes;
2513 else if (mirror_num)
2514 stripe_index = mirror_num - 1;
2515
2516 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2517 int factor = map->num_stripes / map->sub_stripes;
2518
2519 stripe_index = do_div(stripe_nr, factor);
2520 stripe_index *= map->sub_stripes;
2521
2522 if (unplug_page || (rw & (1 << BIO_RW)))
2523 num_stripes = map->sub_stripes;
2524 else if (mirror_num)
2525 stripe_index += mirror_num - 1;
2526 else {
2527 stripe_index = find_live_mirror(map, stripe_index,
2528 map->sub_stripes, stripe_index +
2529 current->pid % map->sub_stripes);
2530 }
2531 } else {
2532 /*
2533 * after this do_div call, stripe_nr is the number of stripes
2534 * on this device we have to walk to find the data, and
2535 * stripe_index is the number of our device in the stripe array
2536 */
2537 stripe_index = do_div(stripe_nr, map->num_stripes);
2538 }
2539 BUG_ON(stripe_index >= map->num_stripes);
2540
2541 for (i = 0; i < num_stripes; i++) {
2542 if (unplug_page) {
2543 struct btrfs_device *device;
2544 struct backing_dev_info *bdi;
2545
2546 device = map->stripes[stripe_index].dev;
2547 if (device->bdev) {
2548 bdi = blk_get_backing_dev_info(device->bdev);
2549 if (bdi->unplug_io_fn)
2550 bdi->unplug_io_fn(bdi, unplug_page);
2551 }
2552 } else {
2553 multi->stripes[i].physical =
2554 map->stripes[stripe_index].physical +
2555 stripe_offset + stripe_nr * map->stripe_len;
2556 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2557 }
2558 stripe_index++;
2559 }
2560 if (multi_ret) {
2561 *multi_ret = multi;
2562 multi->num_stripes = num_stripes;
2563 multi->max_errors = max_errors;
2564 }
2565out:
2566 free_extent_map(em);
2567 return 0;
2568}
2569
2570int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2571 u64 logical, u64 *length,
2572 struct btrfs_multi_bio **multi_ret, int mirror_num)
2573{
2574 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2575 mirror_num, NULL);
2576}
2577
2578int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2579 u64 chunk_start, u64 physical, u64 devid,
2580 u64 **logical, int *naddrs, int *stripe_len)
2581{
2582 struct extent_map_tree *em_tree = &map_tree->map_tree;
2583 struct extent_map *em;
2584 struct map_lookup *map;
2585 u64 *buf;
2586 u64 bytenr;
2587 u64 length;
2588 u64 stripe_nr;
2589 int i, j, nr = 0;
2590
2591 spin_lock(&em_tree->lock);
2592 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2593 spin_unlock(&em_tree->lock);
2594
2595 BUG_ON(!em || em->start != chunk_start);
2596 map = (struct map_lookup *)em->bdev;
2597
2598 length = em->len;
2599 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2600 do_div(length, map->num_stripes / map->sub_stripes);
2601 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2602 do_div(length, map->num_stripes);
2603
2604 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2605 BUG_ON(!buf);
2606
2607 for (i = 0; i < map->num_stripes; i++) {
2608 if (devid && map->stripes[i].dev->devid != devid)
2609 continue;
2610 if (map->stripes[i].physical > physical ||
2611 map->stripes[i].physical + length <= physical)
2612 continue;
2613
2614 stripe_nr = physical - map->stripes[i].physical;
2615 do_div(stripe_nr, map->stripe_len);
2616
2617 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2618 stripe_nr = stripe_nr * map->num_stripes + i;
2619 do_div(stripe_nr, map->sub_stripes);
2620 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2621 stripe_nr = stripe_nr * map->num_stripes + i;
2622 }
2623 bytenr = chunk_start + stripe_nr * map->stripe_len;
2624 WARN_ON(nr >= map->num_stripes);
2625 for (j = 0; j < nr; j++) {
2626 if (buf[j] == bytenr)
2627 break;
2628 }
2629 if (j == nr) {
2630 WARN_ON(nr >= map->num_stripes);
2631 buf[nr++] = bytenr;
2632 }
2633 }
2634
2635 for (i = 0; i > nr; i++) {
2636 struct btrfs_multi_bio *multi;
2637 struct btrfs_bio_stripe *stripe;
2638 int ret;
2639
2640 length = 1;
2641 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2642 &length, &multi, 0);
2643 BUG_ON(ret);
2644
2645 stripe = multi->stripes;
2646 for (j = 0; j < multi->num_stripes; j++) {
2647 if (stripe->physical >= physical &&
2648 physical < stripe->physical + length)
2649 break;
2650 }
2651 BUG_ON(j >= multi->num_stripes);
2652 kfree(multi);
2653 }
2654
2655 *logical = buf;
2656 *naddrs = nr;
2657 *stripe_len = map->stripe_len;
2658
2659 free_extent_map(em);
2660 return 0;
2661}
2662
2663int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2664 u64 logical, struct page *page)
2665{
2666 u64 length = PAGE_CACHE_SIZE;
2667 return __btrfs_map_block(map_tree, READ, logical, &length,
2668 NULL, 0, page);
2669}
2670
2671static void end_bio_multi_stripe(struct bio *bio, int err)
2672{
2673 struct btrfs_multi_bio *multi = bio->bi_private;
2674 int is_orig_bio = 0;
2675
2676 if (err)
2677 atomic_inc(&multi->error);
2678
2679 if (bio == multi->orig_bio)
2680 is_orig_bio = 1;
2681
2682 if (atomic_dec_and_test(&multi->stripes_pending)) {
2683 if (!is_orig_bio) {
2684 bio_put(bio);
2685 bio = multi->orig_bio;
2686 }
2687 bio->bi_private = multi->private;
2688 bio->bi_end_io = multi->end_io;
2689 /* only send an error to the higher layers if it is
2690 * beyond the tolerance of the multi-bio
2691 */
2692 if (atomic_read(&multi->error) > multi->max_errors) {
2693 err = -EIO;
2694 } else if (err) {
2695 /*
2696 * this bio is actually up to date, we didn't
2697 * go over the max number of errors
2698 */
2699 set_bit(BIO_UPTODATE, &bio->bi_flags);
2700 err = 0;
2701 }
2702 kfree(multi);
2703
2704 bio_endio(bio, err);
2705 } else if (!is_orig_bio) {
2706 bio_put(bio);
2707 }
2708}
2709
2710struct async_sched {
2711 struct bio *bio;
2712 int rw;
2713 struct btrfs_fs_info *info;
2714 struct btrfs_work work;
2715};
2716
2717/*
2718 * see run_scheduled_bios for a description of why bios are collected for
2719 * async submit.
2720 *
2721 * This will add one bio to the pending list for a device and make sure
2722 * the work struct is scheduled.
2723 */
2724static noinline int schedule_bio(struct btrfs_root *root,
2725 struct btrfs_device *device,
2726 int rw, struct bio *bio)
2727{
2728 int should_queue = 1;
2729
2730 /* don't bother with additional async steps for reads, right now */
2731 if (!(rw & (1 << BIO_RW))) {
2732 bio_get(bio);
2733 submit_bio(rw, bio);
2734 bio_put(bio);
2735 return 0;
2736 }
2737
2738 /*
2739 * nr_async_bios allows us to reliably return congestion to the
2740 * higher layers. Otherwise, the async bio makes it appear we have
2741 * made progress against dirty pages when we've really just put it
2742 * on a queue for later
2743 */
2744 atomic_inc(&root->fs_info->nr_async_bios);
2745 WARN_ON(bio->bi_next);
2746 bio->bi_next = NULL;
2747 bio->bi_rw |= rw;
2748
2749 spin_lock(&device->io_lock);
2750
2751 if (device->pending_bio_tail)
2752 device->pending_bio_tail->bi_next = bio;
2753
2754 device->pending_bio_tail = bio;
2755 if (!device->pending_bios)
2756 device->pending_bios = bio;
2757 if (device->running_pending)
2758 should_queue = 0;
2759
2760 spin_unlock(&device->io_lock);
2761
2762 if (should_queue)
2763 btrfs_queue_worker(&root->fs_info->submit_workers,
2764 &device->work);
2765 return 0;
2766}
2767
2768int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2769 int mirror_num, int async_submit)
2770{
2771 struct btrfs_mapping_tree *map_tree;
2772 struct btrfs_device *dev;
2773 struct bio *first_bio = bio;
2774 u64 logical = (u64)bio->bi_sector << 9;
2775 u64 length = 0;
2776 u64 map_length;
2777 struct btrfs_multi_bio *multi = NULL;
2778 int ret;
2779 int dev_nr = 0;
2780 int total_devs = 1;
2781
2782 length = bio->bi_size;
2783 map_tree = &root->fs_info->mapping_tree;
2784 map_length = length;
2785
2786 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2787 mirror_num);
2788 BUG_ON(ret);
2789
2790 total_devs = multi->num_stripes;
2791 if (map_length < length) {
2792 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2793 "len %llu\n", (unsigned long long)logical,
2794 (unsigned long long)length,
2795 (unsigned long long)map_length);
2796 BUG();
2797 }
2798 multi->end_io = first_bio->bi_end_io;
2799 multi->private = first_bio->bi_private;
2800 multi->orig_bio = first_bio;
2801 atomic_set(&multi->stripes_pending, multi->num_stripes);
2802
2803 while (dev_nr < total_devs) {
2804 if (total_devs > 1) {
2805 if (dev_nr < total_devs - 1) {
2806 bio = bio_clone(first_bio, GFP_NOFS);
2807 BUG_ON(!bio);
2808 } else {
2809 bio = first_bio;
2810 }
2811 bio->bi_private = multi;
2812 bio->bi_end_io = end_bio_multi_stripe;
2813 }
2814 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2815 dev = multi->stripes[dev_nr].dev;
2816 BUG_ON(rw == WRITE && !dev->writeable);
2817 if (dev && dev->bdev) {
2818 bio->bi_bdev = dev->bdev;
2819 if (async_submit)
2820 schedule_bio(root, dev, rw, bio);
2821 else
2822 submit_bio(rw, bio);
2823 } else {
2824 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2825 bio->bi_sector = logical >> 9;
2826 bio_endio(bio, -EIO);
2827 }
2828 dev_nr++;
2829 }
2830 if (total_devs == 1)
2831 kfree(multi);
2832 return 0;
2833}
2834
2835struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2836 u8 *uuid, u8 *fsid)
2837{
2838 struct btrfs_device *device;
2839 struct btrfs_fs_devices *cur_devices;
2840
2841 cur_devices = root->fs_info->fs_devices;
2842 while (cur_devices) {
2843 if (!fsid ||
2844 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2845 device = __find_device(&cur_devices->devices,
2846 devid, uuid);
2847 if (device)
2848 return device;
2849 }
2850 cur_devices = cur_devices->seed;
2851 }
2852 return NULL;
2853}
2854
2855static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2856 u64 devid, u8 *dev_uuid)
2857{
2858 struct btrfs_device *device;
2859 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2860
2861 device = kzalloc(sizeof(*device), GFP_NOFS);
2862 if (!device)
2863 return NULL;
2864 list_add(&device->dev_list,
2865 &fs_devices->devices);
2866 device->barriers = 1;
2867 device->dev_root = root->fs_info->dev_root;
2868 device->devid = devid;
2869 device->work.func = pending_bios_fn;
2870 device->fs_devices = fs_devices;
2871 fs_devices->num_devices++;
2872 spin_lock_init(&device->io_lock);
2873 INIT_LIST_HEAD(&device->dev_alloc_list);
2874 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2875 return device;
2876}
2877
2878static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2879 struct extent_buffer *leaf,
2880 struct btrfs_chunk *chunk)
2881{
2882 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2883 struct map_lookup *map;
2884 struct extent_map *em;
2885 u64 logical;
2886 u64 length;
2887 u64 devid;
2888 u8 uuid[BTRFS_UUID_SIZE];
2889 int num_stripes;
2890 int ret;
2891 int i;
2892
2893 logical = key->offset;
2894 length = btrfs_chunk_length(leaf, chunk);
2895
2896 spin_lock(&map_tree->map_tree.lock);
2897 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2898 spin_unlock(&map_tree->map_tree.lock);
2899
2900 /* already mapped? */
2901 if (em && em->start <= logical && em->start + em->len > logical) {
2902 free_extent_map(em);
2903 return 0;
2904 } else if (em) {
2905 free_extent_map(em);
2906 }
2907
2908 map = kzalloc(sizeof(*map), GFP_NOFS);
2909 if (!map)
2910 return -ENOMEM;
2911
2912 em = alloc_extent_map(GFP_NOFS);
2913 if (!em)
2914 return -ENOMEM;
2915 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2916 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2917 if (!map) {
2918 free_extent_map(em);
2919 return -ENOMEM;
2920 }
2921
2922 em->bdev = (struct block_device *)map;
2923 em->start = logical;
2924 em->len = length;
2925 em->block_start = 0;
2926 em->block_len = em->len;
2927
2928 map->num_stripes = num_stripes;
2929 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2930 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2931 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2932 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2933 map->type = btrfs_chunk_type(leaf, chunk);
2934 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2935 for (i = 0; i < num_stripes; i++) {
2936 map->stripes[i].physical =
2937 btrfs_stripe_offset_nr(leaf, chunk, i);
2938 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2939 read_extent_buffer(leaf, uuid, (unsigned long)
2940 btrfs_stripe_dev_uuid_nr(chunk, i),
2941 BTRFS_UUID_SIZE);
2942 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2943 NULL);
2944 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2945 kfree(map);
2946 free_extent_map(em);
2947 return -EIO;
2948 }
2949 if (!map->stripes[i].dev) {
2950 map->stripes[i].dev =
2951 add_missing_dev(root, devid, uuid);
2952 if (!map->stripes[i].dev) {
2953 kfree(map);
2954 free_extent_map(em);
2955 return -EIO;
2956 }
2957 }
2958 map->stripes[i].dev->in_fs_metadata = 1;
2959 }
2960
2961 spin_lock(&map_tree->map_tree.lock);
2962 ret = add_extent_mapping(&map_tree->map_tree, em);
2963 spin_unlock(&map_tree->map_tree.lock);
2964 BUG_ON(ret);
2965 free_extent_map(em);
2966
2967 return 0;
2968}
2969
2970static int fill_device_from_item(struct extent_buffer *leaf,
2971 struct btrfs_dev_item *dev_item,
2972 struct btrfs_device *device)
2973{
2974 unsigned long ptr;
2975
2976 device->devid = btrfs_device_id(leaf, dev_item);
2977 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2978 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2979 device->type = btrfs_device_type(leaf, dev_item);
2980 device->io_align = btrfs_device_io_align(leaf, dev_item);
2981 device->io_width = btrfs_device_io_width(leaf, dev_item);
2982 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2983
2984 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2985 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2986
2987 return 0;
2988}
2989
2990static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2991{
2992 struct btrfs_fs_devices *fs_devices;
2993 int ret;
2994
2995 mutex_lock(&uuid_mutex);
2996
2997 fs_devices = root->fs_info->fs_devices->seed;
2998 while (fs_devices) {
2999 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3000 ret = 0;
3001 goto out;
3002 }
3003 fs_devices = fs_devices->seed;
3004 }
3005
3006 fs_devices = find_fsid(fsid);
3007 if (!fs_devices) {
3008 ret = -ENOENT;
3009 goto out;
3010 }
3011
3012 fs_devices = clone_fs_devices(fs_devices);
3013 if (IS_ERR(fs_devices)) {
3014 ret = PTR_ERR(fs_devices);
3015 goto out;
3016 }
3017
3018 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3019 root->fs_info->bdev_holder);
3020 if (ret)
3021 goto out;
3022
3023 if (!fs_devices->seeding) {
3024 __btrfs_close_devices(fs_devices);
3025 free_fs_devices(fs_devices);
3026 ret = -EINVAL;
3027 goto out;
3028 }
3029
3030 fs_devices->seed = root->fs_info->fs_devices->seed;
3031 root->fs_info->fs_devices->seed = fs_devices;
3032out:
3033 mutex_unlock(&uuid_mutex);
3034 return ret;
3035}
3036
3037static int read_one_dev(struct btrfs_root *root,
3038 struct extent_buffer *leaf,
3039 struct btrfs_dev_item *dev_item)
3040{
3041 struct btrfs_device *device;
3042 u64 devid;
3043 int ret;
3044 u8 fs_uuid[BTRFS_UUID_SIZE];
3045 u8 dev_uuid[BTRFS_UUID_SIZE];
3046
3047 devid = btrfs_device_id(leaf, dev_item);
3048 read_extent_buffer(leaf, dev_uuid,
3049 (unsigned long)btrfs_device_uuid(dev_item),
3050 BTRFS_UUID_SIZE);
3051 read_extent_buffer(leaf, fs_uuid,
3052 (unsigned long)btrfs_device_fsid(dev_item),
3053 BTRFS_UUID_SIZE);
3054
3055 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3056 ret = open_seed_devices(root, fs_uuid);
3057 if (ret && !btrfs_test_opt(root, DEGRADED))
3058 return ret;
3059 }
3060
3061 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3062 if (!device || !device->bdev) {
3063 if (!btrfs_test_opt(root, DEGRADED))
3064 return -EIO;
3065
3066 if (!device) {
3067 printk(KERN_WARNING "warning devid %llu missing\n",
3068 (unsigned long long)devid);
3069 device = add_missing_dev(root, devid, dev_uuid);
3070 if (!device)
3071 return -ENOMEM;
3072 }
3073 }
3074
3075 if (device->fs_devices != root->fs_info->fs_devices) {
3076 BUG_ON(device->writeable);
3077 if (device->generation !=
3078 btrfs_device_generation(leaf, dev_item))
3079 return -EINVAL;
3080 }
3081
3082 fill_device_from_item(leaf, dev_item, device);
3083 device->dev_root = root->fs_info->dev_root;
3084 device->in_fs_metadata = 1;
3085 if (device->writeable)
3086 device->fs_devices->total_rw_bytes += device->total_bytes;
3087 ret = 0;
3088 return ret;
3089}
3090
3091int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3092{
3093 struct btrfs_dev_item *dev_item;
3094
3095 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3096 dev_item);
3097 return read_one_dev(root, buf, dev_item);
3098}
3099
3100int btrfs_read_sys_array(struct btrfs_root *root)
3101{
3102 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3103 struct extent_buffer *sb;
3104 struct btrfs_disk_key *disk_key;
3105 struct btrfs_chunk *chunk;
3106 u8 *ptr;
3107 unsigned long sb_ptr;
3108 int ret = 0;
3109 u32 num_stripes;
3110 u32 array_size;
3111 u32 len = 0;
3112 u32 cur;
3113 struct btrfs_key key;
3114
3115 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3116 BTRFS_SUPER_INFO_SIZE);
3117 if (!sb)
3118 return -ENOMEM;
3119 btrfs_set_buffer_uptodate(sb);
3120 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3121 array_size = btrfs_super_sys_array_size(super_copy);
3122
3123 ptr = super_copy->sys_chunk_array;
3124 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3125 cur = 0;
3126
3127 while (cur < array_size) {
3128 disk_key = (struct btrfs_disk_key *)ptr;
3129 btrfs_disk_key_to_cpu(&key, disk_key);
3130
3131 len = sizeof(*disk_key); ptr += len;
3132 sb_ptr += len;
3133 cur += len;
3134
3135 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3136 chunk = (struct btrfs_chunk *)sb_ptr;
3137 ret = read_one_chunk(root, &key, sb, chunk);
3138 if (ret)
3139 break;
3140 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3141 len = btrfs_chunk_item_size(num_stripes);
3142 } else {
3143 ret = -EIO;
3144 break;
3145 }
3146 ptr += len;
3147 sb_ptr += len;
3148 cur += len;
3149 }
3150 free_extent_buffer(sb);
3151 return ret;
3152}
3153
3154int btrfs_read_chunk_tree(struct btrfs_root *root)
3155{
3156 struct btrfs_path *path;
3157 struct extent_buffer *leaf;
3158 struct btrfs_key key;
3159 struct btrfs_key found_key;
3160 int ret;
3161 int slot;
3162
3163 root = root->fs_info->chunk_root;
3164
3165 path = btrfs_alloc_path();
3166 if (!path)
3167 return -ENOMEM;
3168
3169 /* first we search for all of the device items, and then we
3170 * read in all of the chunk items. This way we can create chunk
3171 * mappings that reference all of the devices that are afound
3172 */
3173 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3174 key.offset = 0;
3175 key.type = 0;
3176again:
3177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3178 while (1) {
3179 leaf = path->nodes[0];
3180 slot = path->slots[0];
3181 if (slot >= btrfs_header_nritems(leaf)) {
3182 ret = btrfs_next_leaf(root, path);
3183 if (ret == 0)
3184 continue;
3185 if (ret < 0)
3186 goto error;
3187 break;
3188 }
3189 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3190 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3191 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3192 break;
3193 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3194 struct btrfs_dev_item *dev_item;
3195 dev_item = btrfs_item_ptr(leaf, slot,
3196 struct btrfs_dev_item);
3197 ret = read_one_dev(root, leaf, dev_item);
3198 if (ret)
3199 goto error;
3200 }
3201 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3202 struct btrfs_chunk *chunk;
3203 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3204 ret = read_one_chunk(root, &found_key, leaf, chunk);
3205 if (ret)
3206 goto error;
3207 }
3208 path->slots[0]++;
3209 }
3210 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3211 key.objectid = 0;
3212 btrfs_release_path(root, path);
3213 goto again;
3214 }
3215 ret = 0;
3216error:
3217 btrfs_free_path(path);
3218 return ret;
3219}
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-11 11:30:09 -0500