?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fix setuid sometimes wouldn't
[linux-2.6/mini2440.git] / fs / btrfs / volumes.c
blobdd06e18e5aac7518191e02791073345e6a0f51cd
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 <asm/div64.h>
24 #include "compat.h"
25 #include "ctree.h"
26 #include "extent_map.h"
27 #include "disk-io.h"
28 #include "transaction.h"
29 #include "print-tree.h"
30 #include "volumes.h"
31 #include "async-thread.h"
32
33 struct map_lookup {
34 u64 type;
35 int io_align;
36 int io_width;
37 int stripe_len;
38 int sector_size;
39 int num_stripes;
40 int sub_stripes;
41 struct btrfs_bio_stripe stripes[];
42 };
43
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 struct btrfs_root *root,
46 struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48
49 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
50 (sizeof(struct btrfs_bio_stripe) * (n)))
51
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
54
55 void btrfs_lock_volumes(void)
56 {
57 mutex_lock(&uuid_mutex);
58 }
59
60 void btrfs_unlock_volumes(void)
61 {
62 mutex_unlock(&uuid_mutex);
63 }
64
65 static void lock_chunks(struct btrfs_root *root)
66 {
67 mutex_lock(&root->fs_info->chunk_mutex);
68 }
69
70 static void unlock_chunks(struct btrfs_root *root)
71 {
72 mutex_unlock(&root->fs_info->chunk_mutex);
73 }
74
75 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
76 {
77 struct btrfs_device *device;
78 WARN_ON(fs_devices->opened);
79 while (!list_empty(&fs_devices->devices)) {
80 device = list_entry(fs_devices->devices.next,
81 struct btrfs_device, dev_list);
82 list_del(&device->dev_list);
83 kfree(device->name);
84 kfree(device);
85 }
86 kfree(fs_devices);
87 }
88
89 int btrfs_cleanup_fs_uuids(void)
90 {
91 struct btrfs_fs_devices *fs_devices;
92
93 while (!list_empty(&fs_uuids)) {
94 fs_devices = list_entry(fs_uuids.next,
95 struct btrfs_fs_devices, list);
96 list_del(&fs_devices->list);
97 free_fs_devices(fs_devices);
98 }
99 return 0;
100 }
101
102 static noinline struct btrfs_device *__find_device(struct list_head *head,
103 u64 devid, u8 *uuid)
104 {
105 struct btrfs_device *dev;
106
107 list_for_each_entry(dev, head, dev_list) {
108 if (dev->devid == devid &&
109 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
110 return dev;
111 }
112 }
113 return NULL;
114 }
115
116 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
117 {
118 struct btrfs_fs_devices *fs_devices;
119
120 list_for_each_entry(fs_devices, &fs_uuids, list) {
121 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
122 return fs_devices;
123 }
124 return NULL;
125 }
126
127 /*
128 * we try to collect pending bios for a device so we don't get a large
129 * number of procs sending bios down to the same device. This greatly
130 * improves the schedulers ability to collect and merge the bios.
131 *
132 * But, it also turns into a long list of bios to process and that is sure
133 * to eventually make the worker thread block. The solution here is to
134 * make some progress and then put this work struct back at the end of
135 * the list if the block device is congested. This way, multiple devices
136 * can make progress from a single worker thread.
137 */
138 static noinline int run_scheduled_bios(struct btrfs_device *device)
139 {
140 struct bio *pending;
141 struct backing_dev_info *bdi;
142 struct btrfs_fs_info *fs_info;
143 struct bio *tail;
144 struct bio *cur;
145 int again = 0;
146 unsigned long num_run = 0;
147 unsigned long limit;
148
149 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
150 fs_info = device->dev_root->fs_info;
151 limit = btrfs_async_submit_limit(fs_info);
152 limit = limit * 2 / 3;
153
154 loop:
155 spin_lock(&device->io_lock);
156
157 loop_lock:
158 /* take all the bios off the list at once and process them
159 * later on (without the lock held). But, remember the
160 * tail and other pointers so the bios can be properly reinserted
161 * into the list if we hit congestion
162 */
163 pending = device->pending_bios;
164 tail = device->pending_bio_tail;
165 WARN_ON(pending && !tail);
166 device->pending_bios = NULL;
167 device->pending_bio_tail = NULL;
168
169 /*
170 * if pending was null this time around, no bios need processing
171 * at all and we can stop. Otherwise it'll loop back up again
172 * and do an additional check so no bios are missed.
173 *
174 * device->running_pending is used to synchronize with the
175 * schedule_bio code.
176 */
177 if (pending) {
178 again = 1;
179 device->running_pending = 1;
180 } else {
181 again = 0;
182 device->running_pending = 0;
183 }
184 spin_unlock(&device->io_lock);
185
186 while (pending) {
187 cur = pending;
188 pending = pending->bi_next;
189 cur->bi_next = NULL;
190 atomic_dec(&fs_info->nr_async_bios);
191
192 if (atomic_read(&fs_info->nr_async_bios) < limit &&
193 waitqueue_active(&fs_info->async_submit_wait))
194 wake_up(&fs_info->async_submit_wait);
195
196 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
197 bio_get(cur);
198 submit_bio(cur->bi_rw, cur);
199 bio_put(cur);
200 num_run++;
201
202 /*
203 * we made progress, there is more work to do and the bdi
204 * is now congested. Back off and let other work structs
205 * run instead
206 */
207 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
208 fs_info->fs_devices->open_devices > 1) {
209 struct bio *old_head;
210
211 spin_lock(&device->io_lock);
212
213 old_head = device->pending_bios;
214 device->pending_bios = pending;
215 if (device->pending_bio_tail)
216 tail->bi_next = old_head;
217 else
218 device->pending_bio_tail = tail;
219
220 device->running_pending = 1;
221
222 spin_unlock(&device->io_lock);
223 btrfs_requeue_work(&device->work);
224 goto done;
225 }
226 }
227 if (again)
228 goto loop;
229
230 spin_lock(&device->io_lock);
231 if (device->pending_bios)
232 goto loop_lock;
233 spin_unlock(&device->io_lock);
234 done:
235 return 0;
236 }
237
238 static void pending_bios_fn(struct btrfs_work *work)
239 {
240 struct btrfs_device *device;
241
242 device = container_of(work, struct btrfs_device, work);
243 run_scheduled_bios(device);
244 }
245
246 static noinline int device_list_add(const char *path,
247 struct btrfs_super_block *disk_super,
248 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
249 {
250 struct btrfs_device *device;
251 struct btrfs_fs_devices *fs_devices;
252 u64 found_transid = btrfs_super_generation(disk_super);
253
254 fs_devices = find_fsid(disk_super->fsid);
255 if (!fs_devices) {
256 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
257 if (!fs_devices)
258 return -ENOMEM;
259 INIT_LIST_HEAD(&fs_devices->devices);
260 INIT_LIST_HEAD(&fs_devices->alloc_list);
261 list_add(&fs_devices->list, &fs_uuids);
262 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
263 fs_devices->latest_devid = devid;
264 fs_devices->latest_trans = found_transid;
265 device = NULL;
266 } else {
267 device = __find_device(&fs_devices->devices, devid,
268 disk_super->dev_item.uuid);
269 }
270 if (!device) {
271 if (fs_devices->opened)
272 return -EBUSY;
273
274 device = kzalloc(sizeof(*device), GFP_NOFS);
275 if (!device) {
276 /* we can safely leave the fs_devices entry around */
277 return -ENOMEM;
278 }
279 device->devid = devid;
280 device->work.func = pending_bios_fn;
281 memcpy(device->uuid, disk_super->dev_item.uuid,
282 BTRFS_UUID_SIZE);
283 device->barriers = 1;
284 spin_lock_init(&device->io_lock);
285 device->name = kstrdup(path, GFP_NOFS);
286 if (!device->name) {
287 kfree(device);
288 return -ENOMEM;
289 }
290 INIT_LIST_HEAD(&device->dev_alloc_list);
291 list_add(&device->dev_list, &fs_devices->devices);
292 device->fs_devices = fs_devices;
293 fs_devices->num_devices++;
294 }
295
296 if (found_transid > fs_devices->latest_trans) {
297 fs_devices->latest_devid = devid;
298 fs_devices->latest_trans = found_transid;
299 }
300 *fs_devices_ret = fs_devices;
301 return 0;
302 }
303
304 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
305 {
306 struct btrfs_fs_devices *fs_devices;
307 struct btrfs_device *device;
308 struct btrfs_device *orig_dev;
309
310 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
311 if (!fs_devices)
312 return ERR_PTR(-ENOMEM);
313
314 INIT_LIST_HEAD(&fs_devices->devices);
315 INIT_LIST_HEAD(&fs_devices->alloc_list);
316 INIT_LIST_HEAD(&fs_devices->list);
317 fs_devices->latest_devid = orig->latest_devid;
318 fs_devices->latest_trans = orig->latest_trans;
319 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
320
321 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
322 device = kzalloc(sizeof(*device), GFP_NOFS);
323 if (!device)
324 goto error;
325
326 device->name = kstrdup(orig_dev->name, GFP_NOFS);
327 if (!device->name)
328 goto error;
329
330 device->devid = orig_dev->devid;
331 device->work.func = pending_bios_fn;
332 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
333 device->barriers = 1;
334 spin_lock_init(&device->io_lock);
335 INIT_LIST_HEAD(&device->dev_list);
336 INIT_LIST_HEAD(&device->dev_alloc_list);
337
338 list_add(&device->dev_list, &fs_devices->devices);
339 device->fs_devices = fs_devices;
340 fs_devices->num_devices++;
341 }
342 return fs_devices;
343 error:
344 free_fs_devices(fs_devices);
345 return ERR_PTR(-ENOMEM);
346 }
347
348 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
349 {
350 struct btrfs_device *device, *next;
351
352 mutex_lock(&uuid_mutex);
353 again:
354 list_for_each_entry_safe(device, next, &fs_devices->devices, 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
383 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
384 {
385 struct btrfs_device *device;
386
387 if (--fs_devices->opened > 0)
388 return 0;
389
390 list_for_each_entry(device, &fs_devices->devices, dev_list) {
391 if (device->bdev) {
392 close_bdev_exclusive(device->bdev, device->mode);
393 fs_devices->open_devices--;
394 }
395 if (device->writeable) {
396 list_del_init(&device->dev_alloc_list);
397 fs_devices->rw_devices--;
398 }
399
400 device->bdev = NULL;
401 device->writeable = 0;
402 device->in_fs_metadata = 0;
403 }
404 WARN_ON(fs_devices->open_devices);
405 WARN_ON(fs_devices->rw_devices);
406 fs_devices->opened = 0;
407 fs_devices->seeding = 0;
408
409 return 0;
410 }
411
412 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
413 {
414 struct btrfs_fs_devices *seed_devices = NULL;
415 int ret;
416
417 mutex_lock(&uuid_mutex);
418 ret = __btrfs_close_devices(fs_devices);
419 if (!fs_devices->opened) {
420 seed_devices = fs_devices->seed;
421 fs_devices->seed = NULL;
422 }
423 mutex_unlock(&uuid_mutex);
424
425 while (seed_devices) {
426 fs_devices = seed_devices;
427 seed_devices = fs_devices->seed;
428 __btrfs_close_devices(fs_devices);
429 free_fs_devices(fs_devices);
430 }
431 return ret;
432 }
433
434 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
435 fmode_t flags, void *holder)
436 {
437 struct block_device *bdev;
438 struct list_head *head = &fs_devices->devices;
439 struct btrfs_device *device;
440 struct block_device *latest_bdev = NULL;
441 struct buffer_head *bh;
442 struct btrfs_super_block *disk_super;
443 u64 latest_devid = 0;
444 u64 latest_transid = 0;
445 u64 devid;
446 int seeding = 1;
447 int ret = 0;
448
449 list_for_each_entry(device, head, dev_list) {
450 if (device->bdev)
451 continue;
452 if (!device->name)
453 continue;
454
455 bdev = open_bdev_exclusive(device->name, flags, holder);
456 if (IS_ERR(bdev)) {
457 printk(KERN_INFO "open %s failed\n", device->name);
458 goto error;
459 }
460 set_blocksize(bdev, 4096);
461
462 bh = btrfs_read_dev_super(bdev);
463 if (!bh)
464 goto error_close;
465
466 disk_super = (struct btrfs_super_block *)bh->b_data;
467 devid = le64_to_cpu(disk_super->dev_item.devid);
468 if (devid != device->devid)
469 goto error_brelse;
470
471 if (memcmp(device->uuid, disk_super->dev_item.uuid,
472 BTRFS_UUID_SIZE))
473 goto error_brelse;
474
475 device->generation = btrfs_super_generation(disk_super);
476 if (!latest_transid || device->generation > latest_transid) {
477 latest_devid = devid;
478 latest_transid = device->generation;
479 latest_bdev = bdev;
480 }
481
482 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
483 device->writeable = 0;
484 } else {
485 device->writeable = !bdev_read_only(bdev);
486 seeding = 0;
487 }
488
489 device->bdev = bdev;
490 device->in_fs_metadata = 0;
491 device->mode = flags;
492
493 fs_devices->open_devices++;
494 if (device->writeable) {
495 fs_devices->rw_devices++;
496 list_add(&device->dev_alloc_list,
497 &fs_devices->alloc_list);
498 }
499 continue;
500
501 error_brelse:
502 brelse(bh);
503 error_close:
504 close_bdev_exclusive(bdev, FMODE_READ);
505 error:
506 continue;
507 }
508 if (fs_devices->open_devices == 0) {
509 ret = -EIO;
510 goto out;
511 }
512 fs_devices->seeding = seeding;
513 fs_devices->opened = 1;
514 fs_devices->latest_bdev = latest_bdev;
515 fs_devices->latest_devid = latest_devid;
516 fs_devices->latest_trans = latest_transid;
517 fs_devices->total_rw_bytes = 0;
518 out:
519 return ret;
520 }
521
522 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
523 fmode_t flags, void *holder)
524 {
525 int ret;
526
527 mutex_lock(&uuid_mutex);
528 if (fs_devices->opened) {
529 fs_devices->opened++;
530 ret = 0;
531 } else {
532 ret = __btrfs_open_devices(fs_devices, flags, holder);
533 }
534 mutex_unlock(&uuid_mutex);
535 return ret;
536 }
537
538 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
539 struct btrfs_fs_devices **fs_devices_ret)
540 {
541 struct btrfs_super_block *disk_super;
542 struct block_device *bdev;
543 struct buffer_head *bh;
544 int ret;
545 u64 devid;
546 u64 transid;
547
548 mutex_lock(&uuid_mutex);
549
550 bdev = open_bdev_exclusive(path, flags, holder);
551
552 if (IS_ERR(bdev)) {
553 ret = PTR_ERR(bdev);
554 goto error;
555 }
556
557 ret = set_blocksize(bdev, 4096);
558 if (ret)
559 goto error_close;
560 bh = btrfs_read_dev_super(bdev);
561 if (!bh) {
562 ret = -EIO;
563 goto error_close;
564 }
565 disk_super = (struct btrfs_super_block *)bh->b_data;
566 devid = le64_to_cpu(disk_super->dev_item.devid);
567 transid = btrfs_super_generation(disk_super);
568 if (disk_super->label[0])
569 printk(KERN_INFO "device label %s ", disk_super->label);
570 else {
571 /* FIXME, make a readl uuid parser */
572 printk(KERN_INFO "device fsid %llx-%llx ",
573 *(unsigned long long *)disk_super->fsid,
574 *(unsigned long long *)(disk_super->fsid + 8));
575 }
576 printk(KERN_CONT "devid %llu transid %llu %s\n",
577 (unsigned long long)devid, (unsigned long long)transid, path);
578 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
579
580 brelse(bh);
581 error_close:
582 close_bdev_exclusive(bdev, flags);
583 error:
584 mutex_unlock(&uuid_mutex);
585 return ret;
586 }
587
588 /*
589 * this uses a pretty simple search, the expectation is that it is
590 * called very infrequently and that a given device has a small number
591 * of extents
592 */
593 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
594 struct btrfs_device *device,
595 u64 num_bytes, u64 *start)
596 {
597 struct btrfs_key key;
598 struct btrfs_root *root = device->dev_root;
599 struct btrfs_dev_extent *dev_extent = NULL;
600 struct btrfs_path *path;
601 u64 hole_size = 0;
602 u64 last_byte = 0;
603 u64 search_start = 0;
604 u64 search_end = device->total_bytes;
605 int ret;
606 int slot = 0;
607 int start_found;
608 struct extent_buffer *l;
609
610 path = btrfs_alloc_path();
611 if (!path)
612 return -ENOMEM;
613 path->reada = 2;
614 start_found = 0;
615
616 /* FIXME use last free of some kind */
617
618 /* we don't want to overwrite the superblock on the drive,
619 * so we make sure to start at an offset of at least 1MB
620 */
621 search_start = max((u64)1024 * 1024, search_start);
622
623 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
624 search_start = max(root->fs_info->alloc_start, search_start);
625
626 key.objectid = device->devid;
627 key.offset = search_start;
628 key.type = BTRFS_DEV_EXTENT_KEY;
629 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
630 if (ret < 0)
631 goto error;
632 ret = btrfs_previous_item(root, path, 0, key.type);
633 if (ret < 0)
634 goto error;
635 l = path->nodes[0];
636 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
637 while (1) {
638 l = path->nodes[0];
639 slot = path->slots[0];
640 if (slot >= btrfs_header_nritems(l)) {
641 ret = btrfs_next_leaf(root, path);
642 if (ret == 0)
643 continue;
644 if (ret < 0)
645 goto error;
646 no_more_items:
647 if (!start_found) {
648 if (search_start >= search_end) {
649 ret = -ENOSPC;
650 goto error;
651 }
652 *start = search_start;
653 start_found = 1;
654 goto check_pending;
655 }
656 *start = last_byte > search_start ?
657 last_byte : search_start;
658 if (search_end <= *start) {
659 ret = -ENOSPC;
660 goto error;
661 }
662 goto check_pending;
663 }
664 btrfs_item_key_to_cpu(l, &key, slot);
665
666 if (key.objectid < device->devid)
667 goto next;
668
669 if (key.objectid > device->devid)
670 goto no_more_items;
671
672 if (key.offset >= search_start && key.offset > last_byte &&
673 start_found) {
674 if (last_byte < search_start)
675 last_byte = search_start;
676 hole_size = key.offset - last_byte;
677 if (key.offset > last_byte &&
678 hole_size >= num_bytes) {
679 *start = last_byte;
680 goto check_pending;
681 }
682 }
683 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
684 goto next;
685
686 start_found = 1;
687 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
688 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
689 next:
690 path->slots[0]++;
691 cond_resched();
692 }
693 check_pending:
694 /* we have to make sure we didn't find an extent that has already
695 * been allocated by the map tree or the original allocation
696 */
697 BUG_ON(*start < search_start);
698
699 if (*start + num_bytes > search_end) {
700 ret = -ENOSPC;
701 goto error;
702 }
703 /* check for pending inserts here */
704 ret = 0;
705
706 error:
707 btrfs_free_path(path);
708 return ret;
709 }
710
711 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
712 struct btrfs_device *device,
713 u64 start)
714 {
715 int ret;
716 struct btrfs_path *path;
717 struct btrfs_root *root = device->dev_root;
718 struct btrfs_key key;
719 struct btrfs_key found_key;
720 struct extent_buffer *leaf = NULL;
721 struct btrfs_dev_extent *extent = NULL;
722
723 path = btrfs_alloc_path();
724 if (!path)
725 return -ENOMEM;
726
727 key.objectid = device->devid;
728 key.offset = start;
729 key.type = BTRFS_DEV_EXTENT_KEY;
730
731 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
732 if (ret > 0) {
733 ret = btrfs_previous_item(root, path, key.objectid,
734 BTRFS_DEV_EXTENT_KEY);
735 BUG_ON(ret);
736 leaf = path->nodes[0];
737 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
738 extent = btrfs_item_ptr(leaf, path->slots[0],
739 struct btrfs_dev_extent);
740 BUG_ON(found_key.offset > start || found_key.offset +
741 btrfs_dev_extent_length(leaf, extent) < start);
742 ret = 0;
743 } else if (ret == 0) {
744 leaf = path->nodes[0];
745 extent = btrfs_item_ptr(leaf, path->slots[0],
746 struct btrfs_dev_extent);
747 }
748 BUG_ON(ret);
749
750 if (device->bytes_used > 0)
751 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
752 ret = btrfs_del_item(trans, root, path);
753 BUG_ON(ret);
754
755 btrfs_free_path(path);
756 return ret;
757 }
758
759 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
760 struct btrfs_device *device,
761 u64 chunk_tree, u64 chunk_objectid,
762 u64 chunk_offset, u64 start, u64 num_bytes)
763 {
764 int ret;
765 struct btrfs_path *path;
766 struct btrfs_root *root = device->dev_root;
767 struct btrfs_dev_extent *extent;
768 struct extent_buffer *leaf;
769 struct btrfs_key key;
770
771 WARN_ON(!device->in_fs_metadata);
772 path = btrfs_alloc_path();
773 if (!path)
774 return -ENOMEM;
775
776 key.objectid = device->devid;
777 key.offset = start;
778 key.type = BTRFS_DEV_EXTENT_KEY;
779 ret = btrfs_insert_empty_item(trans, root, path, &key,
780 sizeof(*extent));
781 BUG_ON(ret);
782
783 leaf = path->nodes[0];
784 extent = btrfs_item_ptr(leaf, path->slots[0],
785 struct btrfs_dev_extent);
786 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
787 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
788 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
789
790 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
791 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
792 BTRFS_UUID_SIZE);
793
794 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
795 btrfs_mark_buffer_dirty(leaf);
796 btrfs_free_path(path);
797 return ret;
798 }
799
800 static noinline int find_next_chunk(struct btrfs_root *root,
801 u64 objectid, u64 *offset)
802 {
803 struct btrfs_path *path;
804 int ret;
805 struct btrfs_key key;
806 struct btrfs_chunk *chunk;
807 struct btrfs_key found_key;
808
809 path = btrfs_alloc_path();
810 BUG_ON(!path);
811
812 key.objectid = objectid;
813 key.offset = (u64)-1;
814 key.type = BTRFS_CHUNK_ITEM_KEY;
815
816 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
817 if (ret < 0)
818 goto error;
819
820 BUG_ON(ret == 0);
821
822 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
823 if (ret) {
824 *offset = 0;
825 } else {
826 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
827 path->slots[0]);
828 if (found_key.objectid != objectid)
829 *offset = 0;
830 else {
831 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
832 struct btrfs_chunk);
833 *offset = found_key.offset +
834 btrfs_chunk_length(path->nodes[0], chunk);
835 }
836 }
837 ret = 0;
838 error:
839 btrfs_free_path(path);
840 return ret;
841 }
842
843 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
844 {
845 int ret;
846 struct btrfs_key key;
847 struct btrfs_key found_key;
848 struct btrfs_path *path;
849
850 root = root->fs_info->chunk_root;
851
852 path = btrfs_alloc_path();
853 if (!path)
854 return -ENOMEM;
855
856 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
857 key.type = BTRFS_DEV_ITEM_KEY;
858 key.offset = (u64)-1;
859
860 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
861 if (ret < 0)
862 goto error;
863
864 BUG_ON(ret == 0);
865
866 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
867 BTRFS_DEV_ITEM_KEY);
868 if (ret) {
869 *objectid = 1;
870 } else {
871 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
872 path->slots[0]);
873 *objectid = found_key.offset + 1;
874 }
875 ret = 0;
876 error:
877 btrfs_free_path(path);
878 return ret;
879 }
880
881 /*
882 * the device information is stored in the chunk root
883 * the btrfs_device struct should be fully filled in
884 */
885 int btrfs_add_device(struct btrfs_trans_handle *trans,
886 struct btrfs_root *root,
887 struct btrfs_device *device)
888 {
889 int ret;
890 struct btrfs_path *path;
891 struct btrfs_dev_item *dev_item;
892 struct extent_buffer *leaf;
893 struct btrfs_key key;
894 unsigned long ptr;
895
896 root = root->fs_info->chunk_root;
897
898 path = btrfs_alloc_path();
899 if (!path)
900 return -ENOMEM;
901
902 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
903 key.type = BTRFS_DEV_ITEM_KEY;
904 key.offset = device->devid;
905
906 ret = btrfs_insert_empty_item(trans, root, path, &key,
907 sizeof(*dev_item));
908 if (ret)
909 goto out;
910
911 leaf = path->nodes[0];
912 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
913
914 btrfs_set_device_id(leaf, dev_item, device->devid);
915 btrfs_set_device_generation(leaf, dev_item, 0);
916 btrfs_set_device_type(leaf, dev_item, device->type);
917 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
918 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
919 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
920 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
921 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
922 btrfs_set_device_group(leaf, dev_item, 0);
923 btrfs_set_device_seek_speed(leaf, dev_item, 0);
924 btrfs_set_device_bandwidth(leaf, dev_item, 0);
925 btrfs_set_device_start_offset(leaf, dev_item, 0);
926
927 ptr = (unsigned long)btrfs_device_uuid(dev_item);
928 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
929 ptr = (unsigned long)btrfs_device_fsid(dev_item);
930 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
931 btrfs_mark_buffer_dirty(leaf);
932
933 ret = 0;
934 out:
935 btrfs_free_path(path);
936 return ret;
937 }
938
939 static int btrfs_rm_dev_item(struct btrfs_root *root,
940 struct btrfs_device *device)
941 {
942 int ret;
943 struct btrfs_path *path;
944 struct btrfs_key key;
945 struct btrfs_trans_handle *trans;
946
947 root = root->fs_info->chunk_root;
948
949 path = btrfs_alloc_path();
950 if (!path)
951 return -ENOMEM;
952
953 trans = btrfs_start_transaction(root, 1);
954 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
955 key.type = BTRFS_DEV_ITEM_KEY;
956 key.offset = device->devid;
957 lock_chunks(root);
958
959 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
960 if (ret < 0)
961 goto out;
962
963 if (ret > 0) {
964 ret = -ENOENT;
965 goto out;
966 }
967
968 ret = btrfs_del_item(trans, root, path);
969 if (ret)
970 goto out;
971 out:
972 btrfs_free_path(path);
973 unlock_chunks(root);
974 btrfs_commit_transaction(trans, root);
975 return ret;
976 }
977
978 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
979 {
980 struct btrfs_device *device;
981 struct btrfs_device *next_device;
982 struct block_device *bdev;
983 struct buffer_head *bh = NULL;
984 struct btrfs_super_block *disk_super;
985 u64 all_avail;
986 u64 devid;
987 u64 num_devices;
988 u8 *dev_uuid;
989 int ret = 0;
990
991 mutex_lock(&uuid_mutex);
992 mutex_lock(&root->fs_info->volume_mutex);
993
994 all_avail = root->fs_info->avail_data_alloc_bits |
995 root->fs_info->avail_system_alloc_bits |
996 root->fs_info->avail_metadata_alloc_bits;
997
998 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
999 root->fs_info->fs_devices->rw_devices <= 4) {
1000 printk(KERN_ERR "btrfs: unable to go below four devices "
1001 "on raid10\n");
1002 ret = -EINVAL;
1003 goto out;
1004 }
1005
1006 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1007 root->fs_info->fs_devices->rw_devices <= 2) {
1008 printk(KERN_ERR "btrfs: unable to go below two "
1009 "devices on raid1\n");
1010 ret = -EINVAL;
1011 goto out;
1012 }
1013
1014 if (strcmp(device_path, "missing") == 0) {
1015 struct list_head *devices;
1016 struct btrfs_device *tmp;
1017
1018 device = NULL;
1019 devices = &root->fs_info->fs_devices->devices;
1020 list_for_each_entry(tmp, devices, dev_list) {
1021 if (tmp->in_fs_metadata && !tmp->bdev) {
1022 device = tmp;
1023 break;
1024 }
1025 }
1026 bdev = NULL;
1027 bh = NULL;
1028 disk_super = NULL;
1029 if (!device) {
1030 printk(KERN_ERR "btrfs: no missing devices found to "
1031 "remove\n");
1032 goto out;
1033 }
1034 } else {
1035 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1036 root->fs_info->bdev_holder);
1037 if (IS_ERR(bdev)) {
1038 ret = PTR_ERR(bdev);
1039 goto out;
1040 }
1041
1042 set_blocksize(bdev, 4096);
1043 bh = btrfs_read_dev_super(bdev);
1044 if (!bh) {
1045 ret = -EIO;
1046 goto error_close;
1047 }
1048 disk_super = (struct btrfs_super_block *)bh->b_data;
1049 devid = le64_to_cpu(disk_super->dev_item.devid);
1050 dev_uuid = disk_super->dev_item.uuid;
1051 device = btrfs_find_device(root, devid, dev_uuid,
1052 disk_super->fsid);
1053 if (!device) {
1054 ret = -ENOENT;
1055 goto error_brelse;
1056 }
1057 }
1058
1059 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1060 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1061 "device\n");
1062 ret = -EINVAL;
1063 goto error_brelse;
1064 }
1065
1066 if (device->writeable) {
1067 list_del_init(&device->dev_alloc_list);
1068 root->fs_info->fs_devices->rw_devices--;
1069 }
1070
1071 ret = btrfs_shrink_device(device, 0);
1072 if (ret)
1073 goto error_brelse;
1074
1075 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1076 if (ret)
1077 goto error_brelse;
1078
1079 device->in_fs_metadata = 0;
1080 list_del_init(&device->dev_list);
1081 device->fs_devices->num_devices--;
1082
1083 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1084 struct btrfs_device, dev_list);
1085 if (device->bdev == root->fs_info->sb->s_bdev)
1086 root->fs_info->sb->s_bdev = next_device->bdev;
1087 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1088 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1089
1090 if (device->bdev) {
1091 close_bdev_exclusive(device->bdev, device->mode);
1092 device->bdev = NULL;
1093 device->fs_devices->open_devices--;
1094 }
1095
1096 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1097 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1098
1099 if (device->fs_devices->open_devices == 0) {
1100 struct btrfs_fs_devices *fs_devices;
1101 fs_devices = root->fs_info->fs_devices;
1102 while (fs_devices) {
1103 if (fs_devices->seed == device->fs_devices)
1104 break;
1105 fs_devices = fs_devices->seed;
1106 }
1107 fs_devices->seed = device->fs_devices->seed;
1108 device->fs_devices->seed = NULL;
1109 __btrfs_close_devices(device->fs_devices);
1110 free_fs_devices(device->fs_devices);
1111 }
1112
1113 /*
1114 * at this point, the device is zero sized. We want to
1115 * remove it from the devices list and zero out the old super
1116 */
1117 if (device->writeable) {
1118 /* make sure this device isn't detected as part of
1119 * the FS anymore
1120 */
1121 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1122 set_buffer_dirty(bh);
1123 sync_dirty_buffer(bh);
1124 }
1125
1126 kfree(device->name);
1127 kfree(device);
1128 ret = 0;
1129
1130 error_brelse:
1131 brelse(bh);
1132 error_close:
1133 if (bdev)
1134 close_bdev_exclusive(bdev, FMODE_READ);
1135 out:
1136 mutex_unlock(&root->fs_info->volume_mutex);
1137 mutex_unlock(&uuid_mutex);
1138 return ret;
1139 }
1140
1141 /*
1142 * does all the dirty work required for changing file system's UUID.
1143 */
1144 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1145 struct btrfs_root *root)
1146 {
1147 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1148 struct btrfs_fs_devices *old_devices;
1149 struct btrfs_fs_devices *seed_devices;
1150 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1151 struct btrfs_device *device;
1152 u64 super_flags;
1153
1154 BUG_ON(!mutex_is_locked(&uuid_mutex));
1155 if (!fs_devices->seeding)
1156 return -EINVAL;
1157
1158 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1159 if (!seed_devices)
1160 return -ENOMEM;
1161
1162 old_devices = clone_fs_devices(fs_devices);
1163 if (IS_ERR(old_devices)) {
1164 kfree(seed_devices);
1165 return PTR_ERR(old_devices);
1166 }
1167
1168 list_add(&old_devices->list, &fs_uuids);
1169
1170 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1171 seed_devices->opened = 1;
1172 INIT_LIST_HEAD(&seed_devices->devices);
1173 INIT_LIST_HEAD(&seed_devices->alloc_list);
1174 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1175 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1176 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1177 device->fs_devices = seed_devices;
1178 }
1179
1180 fs_devices->seeding = 0;
1181 fs_devices->num_devices = 0;
1182 fs_devices->open_devices = 0;
1183 fs_devices->seed = seed_devices;
1184
1185 generate_random_uuid(fs_devices->fsid);
1186 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1187 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1188 super_flags = btrfs_super_flags(disk_super) &
1189 ~BTRFS_SUPER_FLAG_SEEDING;
1190 btrfs_set_super_flags(disk_super, super_flags);
1191
1192 return 0;
1193 }
1194
1195 /*
1196 * strore the expected generation for seed devices in device items.
1197 */
1198 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1199 struct btrfs_root *root)
1200 {
1201 struct btrfs_path *path;
1202 struct extent_buffer *leaf;
1203 struct btrfs_dev_item *dev_item;
1204 struct btrfs_device *device;
1205 struct btrfs_key key;
1206 u8 fs_uuid[BTRFS_UUID_SIZE];
1207 u8 dev_uuid[BTRFS_UUID_SIZE];
1208 u64 devid;
1209 int ret;
1210
1211 path = btrfs_alloc_path();
1212 if (!path)
1213 return -ENOMEM;
1214
1215 root = root->fs_info->chunk_root;
1216 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1217 key.offset = 0;
1218 key.type = BTRFS_DEV_ITEM_KEY;
1219
1220 while (1) {
1221 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1222 if (ret < 0)
1223 goto error;
1224
1225 leaf = path->nodes[0];
1226 next_slot:
1227 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1228 ret = btrfs_next_leaf(root, path);
1229 if (ret > 0)
1230 break;
1231 if (ret < 0)
1232 goto error;
1233 leaf = path->nodes[0];
1234 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1235 btrfs_release_path(root, path);
1236 continue;
1237 }
1238
1239 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1240 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1241 key.type != BTRFS_DEV_ITEM_KEY)
1242 break;
1243
1244 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1245 struct btrfs_dev_item);
1246 devid = btrfs_device_id(leaf, dev_item);
1247 read_extent_buffer(leaf, dev_uuid,
1248 (unsigned long)btrfs_device_uuid(dev_item),
1249 BTRFS_UUID_SIZE);
1250 read_extent_buffer(leaf, fs_uuid,
1251 (unsigned long)btrfs_device_fsid(dev_item),
1252 BTRFS_UUID_SIZE);
1253 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1254 BUG_ON(!device);
1255
1256 if (device->fs_devices->seeding) {
1257 btrfs_set_device_generation(leaf, dev_item,
1258 device->generation);
1259 btrfs_mark_buffer_dirty(leaf);
1260 }
1261
1262 path->slots[0]++;
1263 goto next_slot;
1264 }
1265 ret = 0;
1266 error:
1267 btrfs_free_path(path);
1268 return ret;
1269 }
1270
1271 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1272 {
1273 struct btrfs_trans_handle *trans;
1274 struct btrfs_device *device;
1275 struct block_device *bdev;
1276 struct list_head *devices;
1277 struct super_block *sb = root->fs_info->sb;
1278 u64 total_bytes;
1279 int seeding_dev = 0;
1280 int ret = 0;
1281
1282 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1283 return -EINVAL;
1284
1285 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1286 if (!bdev)
1287 return -EIO;
1288
1289 if (root->fs_info->fs_devices->seeding) {
1290 seeding_dev = 1;
1291 down_write(&sb->s_umount);
1292 mutex_lock(&uuid_mutex);
1293 }
1294
1295 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1296 mutex_lock(&root->fs_info->volume_mutex);
1297
1298 devices = &root->fs_info->fs_devices->devices;
1299 list_for_each_entry(device, devices, dev_list) {
1300 if (device->bdev == bdev) {
1301 ret = -EEXIST;
1302 goto error;
1303 }
1304 }
1305
1306 device = kzalloc(sizeof(*device), GFP_NOFS);
1307 if (!device) {
1308 /* we can safely leave the fs_devices entry around */
1309 ret = -ENOMEM;
1310 goto error;
1311 }
1312
1313 device->name = kstrdup(device_path, GFP_NOFS);
1314 if (!device->name) {
1315 kfree(device);
1316 ret = -ENOMEM;
1317 goto error;
1318 }
1319
1320 ret = find_next_devid(root, &device->devid);
1321 if (ret) {
1322 kfree(device);
1323 goto error;
1324 }
1325
1326 trans = btrfs_start_transaction(root, 1);
1327 lock_chunks(root);
1328
1329 device->barriers = 1;
1330 device->writeable = 1;
1331 device->work.func = pending_bios_fn;
1332 generate_random_uuid(device->uuid);
1333 spin_lock_init(&device->io_lock);
1334 device->generation = trans->transid;
1335 device->io_width = root->sectorsize;
1336 device->io_align = root->sectorsize;
1337 device->sector_size = root->sectorsize;
1338 device->total_bytes = i_size_read(bdev->bd_inode);
1339 device->dev_root = root->fs_info->dev_root;
1340 device->bdev = bdev;
1341 device->in_fs_metadata = 1;
1342 device->mode = 0;
1343 set_blocksize(device->bdev, 4096);
1344
1345 if (seeding_dev) {
1346 sb->s_flags &= ~MS_RDONLY;
1347 ret = btrfs_prepare_sprout(trans, root);
1348 BUG_ON(ret);
1349 }
1350
1351 device->fs_devices = root->fs_info->fs_devices;
1352 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1353 list_add(&device->dev_alloc_list,
1354 &root->fs_info->fs_devices->alloc_list);
1355 root->fs_info->fs_devices->num_devices++;
1356 root->fs_info->fs_devices->open_devices++;
1357 root->fs_info->fs_devices->rw_devices++;
1358 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1359
1360 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1361 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1362 total_bytes + device->total_bytes);
1363
1364 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1365 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1366 total_bytes + 1);
1367
1368 if (seeding_dev) {
1369 ret = init_first_rw_device(trans, root, device);
1370 BUG_ON(ret);
1371 ret = btrfs_finish_sprout(trans, root);
1372 BUG_ON(ret);
1373 } else {
1374 ret = btrfs_add_device(trans, root, device);
1375 }
1376
1377 /*
1378 * we've got more storage, clear any full flags on the space
1379 * infos
1380 */
1381 btrfs_clear_space_info_full(root->fs_info);
1382
1383 unlock_chunks(root);
1384 btrfs_commit_transaction(trans, root);
1385
1386 if (seeding_dev) {
1387 mutex_unlock(&uuid_mutex);
1388 up_write(&sb->s_umount);
1389
1390 ret = btrfs_relocate_sys_chunks(root);
1391 BUG_ON(ret);
1392 }
1393 out:
1394 mutex_unlock(&root->fs_info->volume_mutex);
1395 return ret;
1396 error:
1397 close_bdev_exclusive(bdev, 0);
1398 if (seeding_dev) {
1399 mutex_unlock(&uuid_mutex);
1400 up_write(&sb->s_umount);
1401 }
1402 goto out;
1403 }
1404
1405 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1406 struct btrfs_device *device)
1407 {
1408 int ret;
1409 struct btrfs_path *path;
1410 struct btrfs_root *root;
1411 struct btrfs_dev_item *dev_item;
1412 struct extent_buffer *leaf;
1413 struct btrfs_key key;
1414
1415 root = device->dev_root->fs_info->chunk_root;
1416
1417 path = btrfs_alloc_path();
1418 if (!path)
1419 return -ENOMEM;
1420
1421 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1422 key.type = BTRFS_DEV_ITEM_KEY;
1423 key.offset = device->devid;
1424
1425 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1426 if (ret < 0)
1427 goto out;
1428
1429 if (ret > 0) {
1430 ret = -ENOENT;
1431 goto out;
1432 }
1433
1434 leaf = path->nodes[0];
1435 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1436
1437 btrfs_set_device_id(leaf, dev_item, device->devid);
1438 btrfs_set_device_type(leaf, dev_item, device->type);
1439 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1440 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1441 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1442 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1443 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1444 btrfs_mark_buffer_dirty(leaf);
1445
1446 out:
1447 btrfs_free_path(path);
1448 return ret;
1449 }
1450
1451 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1452 struct btrfs_device *device, u64 new_size)
1453 {
1454 struct btrfs_super_block *super_copy =
1455 &device->dev_root->fs_info->super_copy;
1456 u64 old_total = btrfs_super_total_bytes(super_copy);
1457 u64 diff = new_size - device->total_bytes;
1458
1459 if (!device->writeable)
1460 return -EACCES;
1461 if (new_size <= device->total_bytes)
1462 return -EINVAL;
1463
1464 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1465 device->fs_devices->total_rw_bytes += diff;
1466
1467 device->total_bytes = new_size;
1468 btrfs_clear_space_info_full(device->dev_root->fs_info);
1469
1470 return btrfs_update_device(trans, device);
1471 }
1472
1473 int 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
1483 static 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
1511 static 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
1557 static 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
1636 static 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;
1689 error:
1690 btrfs_free_path(path);
1691 return ret;
1692 }
1693
1694 static 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
1703 int btrfs_balance(struct btrfs_root *dev_root)
1704 {
1705 int ret;
1706 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1707 struct btrfs_device *device;
1708 u64 old_size;
1709 u64 size_to_free;
1710 struct btrfs_path *path;
1711 struct btrfs_key key;
1712 struct btrfs_chunk *chunk;
1713 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1714 struct btrfs_trans_handle *trans;
1715 struct btrfs_key found_key;
1716
1717 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1718 return -EROFS;
1719
1720 mutex_lock(&dev_root->fs_info->volume_mutex);
1721 dev_root = dev_root->fs_info->dev_root;
1722
1723 /* step one make some room on all the devices */
1724 list_for_each_entry(device, devices, dev_list) {
1725 old_size = device->total_bytes;
1726 size_to_free = div_factor(old_size, 1);
1727 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1728 if (!device->writeable ||
1729 device->total_bytes - device->bytes_used > size_to_free)
1730 continue;
1731
1732 ret = btrfs_shrink_device(device, old_size - size_to_free);
1733 BUG_ON(ret);
1734
1735 trans = btrfs_start_transaction(dev_root, 1);
1736 BUG_ON(!trans);
1737
1738 ret = btrfs_grow_device(trans, device, old_size);
1739 BUG_ON(ret);
1740
1741 btrfs_end_transaction(trans, dev_root);
1742 }
1743
1744 /* step two, relocate all the chunks */
1745 path = btrfs_alloc_path();
1746 BUG_ON(!path);
1747
1748 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1749 key.offset = (u64)-1;
1750 key.type = BTRFS_CHUNK_ITEM_KEY;
1751
1752 while (1) {
1753 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1754 if (ret < 0)
1755 goto error;
1756
1757 /*
1758 * this shouldn't happen, it means the last relocate
1759 * failed
1760 */
1761 if (ret == 0)
1762 break;
1763
1764 ret = btrfs_previous_item(chunk_root, path, 0,
1765 BTRFS_CHUNK_ITEM_KEY);
1766 if (ret)
1767 break;
1768
1769 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1770 path->slots[0]);
1771 if (found_key.objectid != key.objectid)
1772 break;
1773
1774 chunk = btrfs_item_ptr(path->nodes[0],
1775 path->slots[0],
1776 struct btrfs_chunk);
1777 key.offset = found_key.offset;
1778 /* chunk zero is special */
1779 if (key.offset == 0)
1780 break;
1781
1782 btrfs_release_path(chunk_root, path);
1783 ret = btrfs_relocate_chunk(chunk_root,
1784 chunk_root->root_key.objectid,
1785 found_key.objectid,
1786 found_key.offset);
1787 BUG_ON(ret);
1788 }
1789 ret = 0;
1790 error:
1791 btrfs_free_path(path);
1792 mutex_unlock(&dev_root->fs_info->volume_mutex);
1793 return ret;
1794 }
1795
1796 /*
1797 * shrinking a device means finding all of the device extents past
1798 * the new size, and then following the back refs to the chunks.
1799 * The chunk relocation code actually frees the device extent
1800 */
1801 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1802 {
1803 struct btrfs_trans_handle *trans;
1804 struct btrfs_root *root = device->dev_root;
1805 struct btrfs_dev_extent *dev_extent = NULL;
1806 struct btrfs_path *path;
1807 u64 length;
1808 u64 chunk_tree;
1809 u64 chunk_objectid;
1810 u64 chunk_offset;
1811 int ret;
1812 int slot;
1813 struct extent_buffer *l;
1814 struct btrfs_key key;
1815 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1816 u64 old_total = btrfs_super_total_bytes(super_copy);
1817 u64 diff = device->total_bytes - new_size;
1818
1819 if (new_size >= device->total_bytes)
1820 return -EINVAL;
1821
1822 path = btrfs_alloc_path();
1823 if (!path)
1824 return -ENOMEM;
1825
1826 trans = btrfs_start_transaction(root, 1);
1827 if (!trans) {
1828 ret = -ENOMEM;
1829 goto done;
1830 }
1831
1832 path->reada = 2;
1833
1834 lock_chunks(root);
1835
1836 device->total_bytes = new_size;
1837 if (device->writeable)
1838 device->fs_devices->total_rw_bytes -= diff;
1839 ret = btrfs_update_device(trans, device);
1840 if (ret) {
1841 unlock_chunks(root);
1842 btrfs_end_transaction(trans, root);
1843 goto done;
1844 }
1845 WARN_ON(diff > old_total);
1846 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1847 unlock_chunks(root);
1848 btrfs_end_transaction(trans, root);
1849
1850 key.objectid = device->devid;
1851 key.offset = (u64)-1;
1852 key.type = BTRFS_DEV_EXTENT_KEY;
1853
1854 while (1) {
1855 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1856 if (ret < 0)
1857 goto done;
1858
1859 ret = btrfs_previous_item(root, path, 0, key.type);
1860 if (ret < 0)
1861 goto done;
1862 if (ret) {
1863 ret = 0;
1864 goto done;
1865 }
1866
1867 l = path->nodes[0];
1868 slot = path->slots[0];
1869 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1870
1871 if (key.objectid != device->devid)
1872 goto done;
1873
1874 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1875 length = btrfs_dev_extent_length(l, dev_extent);
1876
1877 if (key.offset + length <= new_size)
1878 goto done;
1879
1880 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1881 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1882 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1883 btrfs_release_path(root, path);
1884
1885 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1886 chunk_offset);
1887 if (ret)
1888 goto done;
1889 }
1890
1891 done:
1892 btrfs_free_path(path);
1893 return ret;
1894 }
1895
1896 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root,
1898 struct btrfs_key *key,
1899 struct btrfs_chunk *chunk, int item_size)
1900 {
1901 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1902 struct btrfs_disk_key disk_key;
1903 u32 array_size;
1904 u8 *ptr;
1905
1906 array_size = btrfs_super_sys_array_size(super_copy);
1907 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1908 return -EFBIG;
1909
1910 ptr = super_copy->sys_chunk_array + array_size;
1911 btrfs_cpu_key_to_disk(&disk_key, key);
1912 memcpy(ptr, &disk_key, sizeof(disk_key));
1913 ptr += sizeof(disk_key);
1914 memcpy(ptr, chunk, item_size);
1915 item_size += sizeof(disk_key);
1916 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1917 return 0;
1918 }
1919
1920 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
1921 int num_stripes, int sub_stripes)
1922 {
1923 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1924 return calc_size;
1925 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1926 return calc_size * (num_stripes / sub_stripes);
1927 else
1928 return calc_size * num_stripes;
1929 }
1930
1931 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1932 struct btrfs_root *extent_root,
1933 struct map_lookup **map_ret,
1934 u64 *num_bytes, u64 *stripe_size,
1935 u64 start, u64 type)
1936 {
1937 struct btrfs_fs_info *info = extent_root->fs_info;
1938 struct btrfs_device *device = NULL;
1939 struct btrfs_fs_devices *fs_devices = info->fs_devices;
1940 struct list_head *cur;
1941 struct map_lookup *map = NULL;
1942 struct extent_map_tree *em_tree;
1943 struct extent_map *em;
1944 struct list_head private_devs;
1945 int min_stripe_size = 1 * 1024 * 1024;
1946 u64 calc_size = 1024 * 1024 * 1024;
1947 u64 max_chunk_size = calc_size;
1948 u64 min_free;
1949 u64 avail;
1950 u64 max_avail = 0;
1951 u64 dev_offset;
1952 int num_stripes = 1;
1953 int min_stripes = 1;
1954 int sub_stripes = 0;
1955 int looped = 0;
1956 int ret;
1957 int index;
1958 int stripe_len = 64 * 1024;
1959
1960 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1961 (type & BTRFS_BLOCK_GROUP_DUP)) {
1962 WARN_ON(1);
1963 type &= ~BTRFS_BLOCK_GROUP_DUP;
1964 }
1965 if (list_empty(&fs_devices->alloc_list))
1966 return -ENOSPC;
1967
1968 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1969 num_stripes = fs_devices->rw_devices;
1970 min_stripes = 2;
1971 }
1972 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1973 num_stripes = 2;
1974 min_stripes = 2;
1975 }
1976 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1977 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1978 if (num_stripes < 2)
1979 return -ENOSPC;
1980 min_stripes = 2;
1981 }
1982 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1983 num_stripes = fs_devices->rw_devices;
1984 if (num_stripes < 4)
1985 return -ENOSPC;
1986 num_stripes &= ~(u32)1;
1987 sub_stripes = 2;
1988 min_stripes = 4;
1989 }
1990
1991 if (type & BTRFS_BLOCK_GROUP_DATA) {
1992 max_chunk_size = 10 * calc_size;
1993 min_stripe_size = 64 * 1024 * 1024;
1994 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1995 max_chunk_size = 4 * calc_size;
1996 min_stripe_size = 32 * 1024 * 1024;
1997 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1998 calc_size = 8 * 1024 * 1024;
1999 max_chunk_size = calc_size * 2;
2000 min_stripe_size = 1 * 1024 * 1024;
2001 }
2002
2003 /* we don't want a chunk larger than 10% of writeable space */
2004 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2005 max_chunk_size);
2006
2007 again:
2008 if (!map || map->num_stripes != num_stripes) {
2009 kfree(map);
2010 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2011 if (!map)
2012 return -ENOMEM;
2013 map->num_stripes = num_stripes;
2014 }
2015
2016 if (calc_size * num_stripes > max_chunk_size) {
2017 calc_size = max_chunk_size;
2018 do_div(calc_size, num_stripes);
2019 do_div(calc_size, stripe_len);
2020 calc_size *= stripe_len;
2021 }
2022 /* we don't want tiny stripes */
2023 calc_size = max_t(u64, min_stripe_size, calc_size);
2024
2025 do_div(calc_size, stripe_len);
2026 calc_size *= stripe_len;
2027
2028 cur = fs_devices->alloc_list.next;
2029 index = 0;
2030
2031 if (type & BTRFS_BLOCK_GROUP_DUP)
2032 min_free = calc_size * 2;
2033 else
2034 min_free = calc_size;
2035
2036 /*
2037 * we add 1MB because we never use the first 1MB of the device, unless
2038 * we've looped, then we are likely allocating the maximum amount of
2039 * space left already
2040 */
2041 if (!looped)
2042 min_free += 1024 * 1024;
2043
2044 INIT_LIST_HEAD(&private_devs);
2045 while (index < num_stripes) {
2046 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2047 BUG_ON(!device->writeable);
2048 if (device->total_bytes > device->bytes_used)
2049 avail = device->total_bytes - device->bytes_used;
2050 else
2051 avail = 0;
2052 cur = cur->next;
2053
2054 if (device->in_fs_metadata && avail >= min_free) {
2055 ret = find_free_dev_extent(trans, device,
2056 min_free, &dev_offset);
2057 if (ret == 0) {
2058 list_move_tail(&device->dev_alloc_list,
2059 &private_devs);
2060 map->stripes[index].dev = device;
2061 map->stripes[index].physical = dev_offset;
2062 index++;
2063 if (type & BTRFS_BLOCK_GROUP_DUP) {
2064 map->stripes[index].dev = device;
2065 map->stripes[index].physical =
2066 dev_offset + calc_size;
2067 index++;
2068 }
2069 }
2070 } else if (device->in_fs_metadata && avail > max_avail)
2071 max_avail = avail;
2072 if (cur == &fs_devices->alloc_list)
2073 break;
2074 }
2075 list_splice(&private_devs, &fs_devices->alloc_list);
2076 if (index < num_stripes) {
2077 if (index >= min_stripes) {
2078 num_stripes = index;
2079 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2080 num_stripes /= sub_stripes;
2081 num_stripes *= sub_stripes;
2082 }
2083 looped = 1;
2084 goto again;
2085 }
2086 if (!looped && max_avail > 0) {
2087 looped = 1;
2088 calc_size = max_avail;
2089 goto again;
2090 }
2091 kfree(map);
2092 return -ENOSPC;
2093 }
2094 map->sector_size = extent_root->sectorsize;
2095 map->stripe_len = stripe_len;
2096 map->io_align = stripe_len;
2097 map->io_width = stripe_len;
2098 map->type = type;
2099 map->num_stripes = num_stripes;
2100 map->sub_stripes = sub_stripes;
2101
2102 *map_ret = map;
2103 *stripe_size = calc_size;
2104 *num_bytes = chunk_bytes_by_type(type, calc_size,
2105 num_stripes, sub_stripes);
2106
2107 em = alloc_extent_map(GFP_NOFS);
2108 if (!em) {
2109 kfree(map);
2110 return -ENOMEM;
2111 }
2112 em->bdev = (struct block_device *)map;
2113 em->start = start;
2114 em->len = *num_bytes;
2115 em->block_start = 0;
2116 em->block_len = em->len;
2117
2118 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2119 spin_lock(&em_tree->lock);
2120 ret = add_extent_mapping(em_tree, em);
2121 spin_unlock(&em_tree->lock);
2122 BUG_ON(ret);
2123 free_extent_map(em);
2124
2125 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2126 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2127 start, *num_bytes);
2128 BUG_ON(ret);
2129
2130 index = 0;
2131 while (index < map->num_stripes) {
2132 device = map->stripes[index].dev;
2133 dev_offset = map->stripes[index].physical;
2134
2135 ret = btrfs_alloc_dev_extent(trans, device,
2136 info->chunk_root->root_key.objectid,
2137 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2138 start, dev_offset, calc_size);
2139 BUG_ON(ret);
2140 index++;
2141 }
2142
2143 return 0;
2144 }
2145
2146 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2147 struct btrfs_root *extent_root,
2148 struct map_lookup *map, u64 chunk_offset,
2149 u64 chunk_size, u64 stripe_size)
2150 {
2151 u64 dev_offset;
2152 struct btrfs_key key;
2153 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2154 struct btrfs_device *device;
2155 struct btrfs_chunk *chunk;
2156 struct btrfs_stripe *stripe;
2157 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2158 int index = 0;
2159 int ret;
2160
2161 chunk = kzalloc(item_size, GFP_NOFS);
2162 if (!chunk)
2163 return -ENOMEM;
2164
2165 index = 0;
2166 while (index < map->num_stripes) {
2167 device = map->stripes[index].dev;
2168 device->bytes_used += stripe_size;
2169 ret = btrfs_update_device(trans, device);
2170 BUG_ON(ret);
2171 index++;
2172 }
2173
2174 index = 0;
2175 stripe = &chunk->stripe;
2176 while (index < map->num_stripes) {
2177 device = map->stripes[index].dev;
2178 dev_offset = map->stripes[index].physical;
2179
2180 btrfs_set_stack_stripe_devid(stripe, device->devid);
2181 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2182 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2183 stripe++;
2184 index++;
2185 }
2186
2187 btrfs_set_stack_chunk_length(chunk, chunk_size);
2188 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2189 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2190 btrfs_set_stack_chunk_type(chunk, map->type);
2191 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2192 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2193 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2194 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2195 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2196
2197 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2198 key.type = BTRFS_CHUNK_ITEM_KEY;
2199 key.offset = chunk_offset;
2200
2201 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2202 BUG_ON(ret);
2203
2204 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2205 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2206 item_size);
2207 BUG_ON(ret);
2208 }
2209 kfree(chunk);
2210 return 0;
2211 }
2212
2213 /*
2214 * Chunk allocation falls into two parts. The first part does works
2215 * that make the new allocated chunk useable, but not do any operation
2216 * that modifies the chunk tree. The second part does the works that
2217 * require modifying the chunk tree. This division is important for the
2218 * bootstrap process of adding storage to a seed btrfs.
2219 */
2220 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2221 struct btrfs_root *extent_root, u64 type)
2222 {
2223 u64 chunk_offset;
2224 u64 chunk_size;
2225 u64 stripe_size;
2226 struct map_lookup *map;
2227 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2228 int ret;
2229
2230 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2231 &chunk_offset);
2232 if (ret)
2233 return ret;
2234
2235 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2236 &stripe_size, chunk_offset, type);
2237 if (ret)
2238 return ret;
2239
2240 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2241 chunk_size, stripe_size);
2242 BUG_ON(ret);
2243 return 0;
2244 }
2245
2246 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2247 struct btrfs_root *root,
2248 struct btrfs_device *device)
2249 {
2250 u64 chunk_offset;
2251 u64 sys_chunk_offset;
2252 u64 chunk_size;
2253 u64 sys_chunk_size;
2254 u64 stripe_size;
2255 u64 sys_stripe_size;
2256 u64 alloc_profile;
2257 struct map_lookup *map;
2258 struct map_lookup *sys_map;
2259 struct btrfs_fs_info *fs_info = root->fs_info;
2260 struct btrfs_root *extent_root = fs_info->extent_root;
2261 int ret;
2262
2263 ret = find_next_chunk(fs_info->chunk_root,
2264 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2265 BUG_ON(ret);
2266
2267 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2268 (fs_info->metadata_alloc_profile &
2269 fs_info->avail_metadata_alloc_bits);
2270 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2271
2272 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2273 &stripe_size, chunk_offset, alloc_profile);
2274 BUG_ON(ret);
2275
2276 sys_chunk_offset = chunk_offset + chunk_size;
2277
2278 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2279 (fs_info->system_alloc_profile &
2280 fs_info->avail_system_alloc_bits);
2281 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2282
2283 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2284 &sys_chunk_size, &sys_stripe_size,
2285 sys_chunk_offset, alloc_profile);
2286 BUG_ON(ret);
2287
2288 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2289 BUG_ON(ret);
2290
2291 /*
2292 * Modifying chunk tree needs allocating new blocks from both
2293 * system block group and metadata block group. So we only can
2294 * do operations require modifying the chunk tree after both
2295 * block groups were created.
2296 */
2297 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2298 chunk_size, stripe_size);
2299 BUG_ON(ret);
2300
2301 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2302 sys_chunk_offset, sys_chunk_size,
2303 sys_stripe_size);
2304 BUG_ON(ret);
2305 return 0;
2306 }
2307
2308 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2309 {
2310 struct extent_map *em;
2311 struct map_lookup *map;
2312 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2313 int readonly = 0;
2314 int i;
2315
2316 spin_lock(&map_tree->map_tree.lock);
2317 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2318 spin_unlock(&map_tree->map_tree.lock);
2319 if (!em)
2320 return 1;
2321
2322 map = (struct map_lookup *)em->bdev;
2323 for (i = 0; i < map->num_stripes; i++) {
2324 if (!map->stripes[i].dev->writeable) {
2325 readonly = 1;
2326 break;
2327 }
2328 }
2329 free_extent_map(em);
2330 return readonly;
2331 }
2332
2333 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2334 {
2335 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2336 }
2337
2338 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2339 {
2340 struct extent_map *em;
2341
2342 while (1) {
2343 spin_lock(&tree->map_tree.lock);
2344 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2345 if (em)
2346 remove_extent_mapping(&tree->map_tree, em);
2347 spin_unlock(&tree->map_tree.lock);
2348 if (!em)
2349 break;
2350 kfree(em->bdev);
2351 /* once for us */
2352 free_extent_map(em);
2353 /* once for the tree */
2354 free_extent_map(em);
2355 }
2356 }
2357
2358 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2359 {
2360 struct extent_map *em;
2361 struct map_lookup *map;
2362 struct extent_map_tree *em_tree = &map_tree->map_tree;
2363 int ret;
2364
2365 spin_lock(&em_tree->lock);
2366 em = lookup_extent_mapping(em_tree, logical, len);
2367 spin_unlock(&em_tree->lock);
2368 BUG_ON(!em);
2369
2370 BUG_ON(em->start > logical || em->start + em->len < logical);
2371 map = (struct map_lookup *)em->bdev;
2372 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2373 ret = map->num_stripes;
2374 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2375 ret = map->sub_stripes;
2376 else
2377 ret = 1;
2378 free_extent_map(em);
2379 return ret;
2380 }
2381
2382 static int find_live_mirror(struct map_lookup *map, int first, int num,
2383 int optimal)
2384 {
2385 int i;
2386 if (map->stripes[optimal].dev->bdev)
2387 return optimal;
2388 for (i = first; i < first + num; i++) {
2389 if (map->stripes[i].dev->bdev)
2390 return i;
2391 }
2392 /* we couldn't find one that doesn't fail. Just return something
2393 * and the io error handling code will clean up eventually
2394 */
2395 return optimal;
2396 }
2397
2398 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2399 u64 logical, u64 *length,
2400 struct btrfs_multi_bio **multi_ret,
2401 int mirror_num, struct page *unplug_page)
2402 {
2403 struct extent_map *em;
2404 struct map_lookup *map;
2405 struct extent_map_tree *em_tree = &map_tree->map_tree;
2406 u64 offset;
2407 u64 stripe_offset;
2408 u64 stripe_nr;
2409 int stripes_allocated = 8;
2410 int stripes_required = 1;
2411 int stripe_index;
2412 int i;
2413 int num_stripes;
2414 int max_errors = 0;
2415 struct btrfs_multi_bio *multi = NULL;
2416
2417 if (multi_ret && !(rw & (1 << BIO_RW)))
2418 stripes_allocated = 1;
2419 again:
2420 if (multi_ret) {
2421 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2422 GFP_NOFS);
2423 if (!multi)
2424 return -ENOMEM;
2425
2426 atomic_set(&multi->error, 0);
2427 }
2428
2429 spin_lock(&em_tree->lock);
2430 em = lookup_extent_mapping(em_tree, logical, *length);
2431 spin_unlock(&em_tree->lock);
2432
2433 if (!em && unplug_page)
2434 return 0;
2435
2436 if (!em) {
2437 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2438 (unsigned long long)logical,
2439 (unsigned long long)*length);
2440 BUG();
2441 }
2442
2443 BUG_ON(em->start > logical || em->start + em->len < logical);
2444 map = (struct map_lookup *)em->bdev;
2445 offset = logical - em->start;
2446
2447 if (mirror_num > map->num_stripes)
2448 mirror_num = 0;
2449
2450 /* if our multi bio struct is too small, back off and try again */
2451 if (rw & (1 << BIO_RW)) {
2452 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2453 BTRFS_BLOCK_GROUP_DUP)) {
2454 stripes_required = map->num_stripes;
2455 max_errors = 1;
2456 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2457 stripes_required = map->sub_stripes;
2458 max_errors = 1;
2459 }
2460 }
2461 if (multi_ret && rw == WRITE &&
2462 stripes_allocated < stripes_required) {
2463 stripes_allocated = map->num_stripes;
2464 free_extent_map(em);
2465 kfree(multi);
2466 goto again;
2467 }
2468 stripe_nr = offset;
2469 /*
2470 * stripe_nr counts the total number of stripes we have to stride
2471 * to get to this block
2472 */
2473 do_div(stripe_nr, map->stripe_len);
2474
2475 stripe_offset = stripe_nr * map->stripe_len;
2476 BUG_ON(offset < stripe_offset);
2477
2478 /* stripe_offset is the offset of this block in its stripe*/
2479 stripe_offset = offset - stripe_offset;
2480
2481 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2482 BTRFS_BLOCK_GROUP_RAID10 |
2483 BTRFS_BLOCK_GROUP_DUP)) {
2484 /* we limit the length of each bio to what fits in a stripe */
2485 *length = min_t(u64, em->len - offset,
2486 map->stripe_len - stripe_offset);
2487 } else {
2488 *length = em->len - offset;
2489 }
2490
2491 if (!multi_ret && !unplug_page)
2492 goto out;
2493
2494 num_stripes = 1;
2495 stripe_index = 0;
2496 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2497 if (unplug_page || (rw & (1 << BIO_RW)))
2498 num_stripes = map->num_stripes;
2499 else if (mirror_num)
2500 stripe_index = mirror_num - 1;
2501 else {
2502 stripe_index = find_live_mirror(map, 0,
2503 map->num_stripes,
2504 current->pid % map->num_stripes);
2505 }
2506
2507 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2508 if (rw & (1 << BIO_RW))
2509 num_stripes = map->num_stripes;
2510 else if (mirror_num)
2511 stripe_index = mirror_num - 1;
2512
2513 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2514 int factor = map->num_stripes / map->sub_stripes;
2515
2516 stripe_index = do_div(stripe_nr, factor);
2517 stripe_index *= map->sub_stripes;
2518
2519 if (unplug_page || (rw & (1 << BIO_RW)))
2520 num_stripes = map->sub_stripes;
2521 else if (mirror_num)
2522 stripe_index += mirror_num - 1;
2523 else {
2524 stripe_index = find_live_mirror(map, stripe_index,
2525 map->sub_stripes, stripe_index +
2526 current->pid % map->sub_stripes);
2527 }
2528 } else {
2529 /*
2530 * after this do_div call, stripe_nr is the number of stripes
2531 * on this device we have to walk to find the data, and
2532 * stripe_index is the number of our device in the stripe array
2533 */
2534 stripe_index = do_div(stripe_nr, map->num_stripes);
2535 }
2536 BUG_ON(stripe_index >= map->num_stripes);
2537
2538 for (i = 0; i < num_stripes; i++) {
2539 if (unplug_page) {
2540 struct btrfs_device *device;
2541 struct backing_dev_info *bdi;
2542
2543 device = map->stripes[stripe_index].dev;
2544 if (device->bdev) {
2545 bdi = blk_get_backing_dev_info(device->bdev);
2546 if (bdi->unplug_io_fn)
2547 bdi->unplug_io_fn(bdi, unplug_page);
2548 }
2549 } else {
2550 multi->stripes[i].physical =
2551 map->stripes[stripe_index].physical +
2552 stripe_offset + stripe_nr * map->stripe_len;
2553 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2554 }
2555 stripe_index++;
2556 }
2557 if (multi_ret) {
2558 *multi_ret = multi;
2559 multi->num_stripes = num_stripes;
2560 multi->max_errors = max_errors;
2561 }
2562 out:
2563 free_extent_map(em);
2564 return 0;
2565 }
2566
2567 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2568 u64 logical, u64 *length,
2569 struct btrfs_multi_bio **multi_ret, int mirror_num)
2570 {
2571 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2572 mirror_num, NULL);
2573 }
2574
2575 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2576 u64 chunk_start, u64 physical, u64 devid,
2577 u64 **logical, int *naddrs, int *stripe_len)
2578 {
2579 struct extent_map_tree *em_tree = &map_tree->map_tree;
2580 struct extent_map *em;
2581 struct map_lookup *map;
2582 u64 *buf;
2583 u64 bytenr;
2584 u64 length;
2585 u64 stripe_nr;
2586 int i, j, nr = 0;
2587
2588 spin_lock(&em_tree->lock);
2589 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2590 spin_unlock(&em_tree->lock);
2591
2592 BUG_ON(!em || em->start != chunk_start);
2593 map = (struct map_lookup *)em->bdev;
2594
2595 length = em->len;
2596 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2597 do_div(length, map->num_stripes / map->sub_stripes);
2598 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2599 do_div(length, map->num_stripes);
2600
2601 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2602 BUG_ON(!buf);
2603
2604 for (i = 0; i < map->num_stripes; i++) {
2605 if (devid && map->stripes[i].dev->devid != devid)
2606 continue;
2607 if (map->stripes[i].physical > physical ||
2608 map->stripes[i].physical + length <= physical)
2609 continue;
2610
2611 stripe_nr = physical - map->stripes[i].physical;
2612 do_div(stripe_nr, map->stripe_len);
2613
2614 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2615 stripe_nr = stripe_nr * map->num_stripes + i;
2616 do_div(stripe_nr, map->sub_stripes);
2617 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2618 stripe_nr = stripe_nr * map->num_stripes + i;
2619 }
2620 bytenr = chunk_start + stripe_nr * map->stripe_len;
2621 WARN_ON(nr >= map->num_stripes);
2622 for (j = 0; j < nr; j++) {
2623 if (buf[j] == bytenr)
2624 break;
2625 }
2626 if (j == nr) {
2627 WARN_ON(nr >= map->num_stripes);
2628 buf[nr++] = bytenr;
2629 }
2630 }
2631
2632 for (i = 0; i > nr; i++) {
2633 struct btrfs_multi_bio *multi;
2634 struct btrfs_bio_stripe *stripe;
2635 int ret;
2636
2637 length = 1;
2638 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2639 &length, &multi, 0);
2640 BUG_ON(ret);
2641
2642 stripe = multi->stripes;
2643 for (j = 0; j < multi->num_stripes; j++) {
2644 if (stripe->physical >= physical &&
2645 physical < stripe->physical + length)
2646 break;
2647 }
2648 BUG_ON(j >= multi->num_stripes);
2649 kfree(multi);
2650 }
2651
2652 *logical = buf;
2653 *naddrs = nr;
2654 *stripe_len = map->stripe_len;
2655
2656 free_extent_map(em);
2657 return 0;
2658 }
2659
2660 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2661 u64 logical, struct page *page)
2662 {
2663 u64 length = PAGE_CACHE_SIZE;
2664 return __btrfs_map_block(map_tree, READ, logical, &length,
2665 NULL, 0, page);
2666 }
2667
2668 static void end_bio_multi_stripe(struct bio *bio, int err)
2669 {
2670 struct btrfs_multi_bio *multi = bio->bi_private;
2671 int is_orig_bio = 0;
2672
2673 if (err)
2674 atomic_inc(&multi->error);
2675
2676 if (bio == multi->orig_bio)
2677 is_orig_bio = 1;
2678
2679 if (atomic_dec_and_test(&multi->stripes_pending)) {
2680 if (!is_orig_bio) {
2681 bio_put(bio);
2682 bio = multi->orig_bio;
2683 }
2684 bio->bi_private = multi->private;
2685 bio->bi_end_io = multi->end_io;
2686 /* only send an error to the higher layers if it is
2687 * beyond the tolerance of the multi-bio
2688 */
2689 if (atomic_read(&multi->error) > multi->max_errors) {
2690 err = -EIO;
2691 } else if (err) {
2692 /*
2693 * this bio is actually up to date, we didn't
2694 * go over the max number of errors
2695 */
2696 set_bit(BIO_UPTODATE, &bio->bi_flags);
2697 err = 0;
2698 }
2699 kfree(multi);
2700
2701 bio_endio(bio, err);
2702 } else if (!is_orig_bio) {
2703 bio_put(bio);
2704 }
2705 }
2706
2707 struct async_sched {
2708 struct bio *bio;
2709 int rw;
2710 struct btrfs_fs_info *info;
2711 struct btrfs_work work;
2712 };
2713
2714 /*
2715 * see run_scheduled_bios for a description of why bios are collected for
2716 * async submit.
2717 *
2718 * This will add one bio to the pending list for a device and make sure
2719 * the work struct is scheduled.
2720 */
2721 static noinline int schedule_bio(struct btrfs_root *root,
2722 struct btrfs_device *device,
2723 int rw, struct bio *bio)
2724 {
2725 int should_queue = 1;
2726
2727 /* don't bother with additional async steps for reads, right now */
2728 if (!(rw & (1 << BIO_RW))) {
2729 bio_get(bio);
2730 submit_bio(rw, bio);
2731 bio_put(bio);
2732 return 0;
2733 }
2734
2735 /*
2736 * nr_async_bios allows us to reliably return congestion to the
2737 * higher layers. Otherwise, the async bio makes it appear we have
2738 * made progress against dirty pages when we've really just put it
2739 * on a queue for later
2740 */
2741 atomic_inc(&root->fs_info->nr_async_bios);
2742 WARN_ON(bio->bi_next);
2743 bio->bi_next = NULL;
2744 bio->bi_rw |= rw;
2745
2746 spin_lock(&device->io_lock);
2747
2748 if (device->pending_bio_tail)
2749 device->pending_bio_tail->bi_next = bio;
2750
2751 device->pending_bio_tail = bio;
2752 if (!device->pending_bios)
2753 device->pending_bios = bio;
2754 if (device->running_pending)
2755 should_queue = 0;
2756
2757 spin_unlock(&device->io_lock);
2758
2759 if (should_queue)
2760 btrfs_queue_worker(&root->fs_info->submit_workers,
2761 &device->work);
2762 return 0;
2763 }
2764
2765 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2766 int mirror_num, int async_submit)
2767 {
2768 struct btrfs_mapping_tree *map_tree;
2769 struct btrfs_device *dev;
2770 struct bio *first_bio = bio;
2771 u64 logical = (u64)bio->bi_sector << 9;
2772 u64 length = 0;
2773 u64 map_length;
2774 struct btrfs_multi_bio *multi = NULL;
2775 int ret;
2776 int dev_nr = 0;
2777 int total_devs = 1;
2778
2779 length = bio->bi_size;
2780 map_tree = &root->fs_info->mapping_tree;
2781 map_length = length;
2782
2783 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2784 mirror_num);
2785 BUG_ON(ret);
2786
2787 total_devs = multi->num_stripes;
2788 if (map_length < length) {
2789 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2790 "len %llu\n", (unsigned long long)logical,
2791 (unsigned long long)length,
2792 (unsigned long long)map_length);
2793 BUG();
2794 }
2795 multi->end_io = first_bio->bi_end_io;
2796 multi->private = first_bio->bi_private;
2797 multi->orig_bio = first_bio;
2798 atomic_set(&multi->stripes_pending, multi->num_stripes);
2799
2800 while (dev_nr < total_devs) {
2801 if (total_devs > 1) {
2802 if (dev_nr < total_devs - 1) {
2803 bio = bio_clone(first_bio, GFP_NOFS);
2804 BUG_ON(!bio);
2805 } else {
2806 bio = first_bio;
2807 }
2808 bio->bi_private = multi;
2809 bio->bi_end_io = end_bio_multi_stripe;
2810 }
2811 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2812 dev = multi->stripes[dev_nr].dev;
2813 BUG_ON(rw == WRITE && !dev->writeable);
2814 if (dev && dev->bdev) {
2815 bio->bi_bdev = dev->bdev;
2816 if (async_submit)
2817 schedule_bio(root, dev, rw, bio);
2818 else
2819 submit_bio(rw, bio);
2820 } else {
2821 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2822 bio->bi_sector = logical >> 9;
2823 bio_endio(bio, -EIO);
2824 }
2825 dev_nr++;
2826 }
2827 if (total_devs == 1)
2828 kfree(multi);
2829 return 0;
2830 }
2831
2832 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2833 u8 *uuid, u8 *fsid)
2834 {
2835 struct btrfs_device *device;
2836 struct btrfs_fs_devices *cur_devices;
2837
2838 cur_devices = root->fs_info->fs_devices;
2839 while (cur_devices) {
2840 if (!fsid ||
2841 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2842 device = __find_device(&cur_devices->devices,
2843 devid, uuid);
2844 if (device)
2845 return device;
2846 }
2847 cur_devices = cur_devices->seed;
2848 }
2849 return NULL;
2850 }
2851
2852 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2853 u64 devid, u8 *dev_uuid)
2854 {
2855 struct btrfs_device *device;
2856 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2857
2858 device = kzalloc(sizeof(*device), GFP_NOFS);
2859 if (!device)
2860 return NULL;
2861 list_add(&device->dev_list,
2862 &fs_devices->devices);
2863 device->barriers = 1;
2864 device->dev_root = root->fs_info->dev_root;
2865 device->devid = devid;
2866 device->work.func = pending_bios_fn;
2867 device->fs_devices = fs_devices;
2868 fs_devices->num_devices++;
2869 spin_lock_init(&device->io_lock);
2870 INIT_LIST_HEAD(&device->dev_alloc_list);
2871 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2872 return device;
2873 }
2874
2875 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2876 struct extent_buffer *leaf,
2877 struct btrfs_chunk *chunk)
2878 {
2879 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2880 struct map_lookup *map;
2881 struct extent_map *em;
2882 u64 logical;
2883 u64 length;
2884 u64 devid;
2885 u8 uuid[BTRFS_UUID_SIZE];
2886 int num_stripes;
2887 int ret;
2888 int i;
2889
2890 logical = key->offset;
2891 length = btrfs_chunk_length(leaf, chunk);
2892
2893 spin_lock(&map_tree->map_tree.lock);
2894 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2895 spin_unlock(&map_tree->map_tree.lock);
2896
2897 /* already mapped? */
2898 if (em && em->start <= logical && em->start + em->len > logical) {
2899 free_extent_map(em);
2900 return 0;
2901 } else if (em) {
2902 free_extent_map(em);
2903 }
2904
2905 em = alloc_extent_map(GFP_NOFS);
2906 if (!em)
2907 return -ENOMEM;
2908 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2909 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2910 if (!map) {
2911 free_extent_map(em);
2912 return -ENOMEM;
2913 }
2914
2915 em->bdev = (struct block_device *)map;
2916 em->start = logical;
2917 em->len = length;
2918 em->block_start = 0;
2919 em->block_len = em->len;
2920
2921 map->num_stripes = num_stripes;
2922 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2923 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2924 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2925 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2926 map->type = btrfs_chunk_type(leaf, chunk);
2927 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2928 for (i = 0; i < num_stripes; i++) {
2929 map->stripes[i].physical =
2930 btrfs_stripe_offset_nr(leaf, chunk, i);
2931 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2932 read_extent_buffer(leaf, uuid, (unsigned long)
2933 btrfs_stripe_dev_uuid_nr(chunk, i),
2934 BTRFS_UUID_SIZE);
2935 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2936 NULL);
2937 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2938 kfree(map);
2939 free_extent_map(em);
2940 return -EIO;
2941 }
2942 if (!map->stripes[i].dev) {
2943 map->stripes[i].dev =
2944 add_missing_dev(root, devid, uuid);
2945 if (!map->stripes[i].dev) {
2946 kfree(map);
2947 free_extent_map(em);
2948 return -EIO;
2949 }
2950 }
2951 map->stripes[i].dev->in_fs_metadata = 1;
2952 }
2953
2954 spin_lock(&map_tree->map_tree.lock);
2955 ret = add_extent_mapping(&map_tree->map_tree, em);
2956 spin_unlock(&map_tree->map_tree.lock);
2957 BUG_ON(ret);
2958 free_extent_map(em);
2959
2960 return 0;
2961 }
2962
2963 static int fill_device_from_item(struct extent_buffer *leaf,
2964 struct btrfs_dev_item *dev_item,
2965 struct btrfs_device *device)
2966 {
2967 unsigned long ptr;
2968
2969 device->devid = btrfs_device_id(leaf, dev_item);
2970 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2971 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2972 device->type = btrfs_device_type(leaf, dev_item);
2973 device->io_align = btrfs_device_io_align(leaf, dev_item);
2974 device->io_width = btrfs_device_io_width(leaf, dev_item);
2975 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2976
2977 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2978 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2979
2980 return 0;
2981 }
2982
2983 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2984 {
2985 struct btrfs_fs_devices *fs_devices;
2986 int ret;
2987
2988 mutex_lock(&uuid_mutex);
2989
2990 fs_devices = root->fs_info->fs_devices->seed;
2991 while (fs_devices) {
2992 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2993 ret = 0;
2994 goto out;
2995 }
2996 fs_devices = fs_devices->seed;
2997 }
2998
2999 fs_devices = find_fsid(fsid);
3000 if (!fs_devices) {
3001 ret = -ENOENT;
3002 goto out;
3003 }
3004
3005 fs_devices = clone_fs_devices(fs_devices);
3006 if (IS_ERR(fs_devices)) {
3007 ret = PTR_ERR(fs_devices);
3008 goto out;
3009 }
3010
3011 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3012 root->fs_info->bdev_holder);
3013 if (ret)
3014 goto out;
3015
3016 if (!fs_devices->seeding) {
3017 __btrfs_close_devices(fs_devices);
3018 free_fs_devices(fs_devices);
3019 ret = -EINVAL;
3020 goto out;
3021 }
3022
3023 fs_devices->seed = root->fs_info->fs_devices->seed;
3024 root->fs_info->fs_devices->seed = fs_devices;
3025 out:
3026 mutex_unlock(&uuid_mutex);
3027 return ret;
3028 }
3029
3030 static int read_one_dev(struct btrfs_root *root,
3031 struct extent_buffer *leaf,
3032 struct btrfs_dev_item *dev_item)
3033 {
3034 struct btrfs_device *device;
3035 u64 devid;
3036 int ret;
3037 u8 fs_uuid[BTRFS_UUID_SIZE];
3038 u8 dev_uuid[BTRFS_UUID_SIZE];
3039
3040 devid = btrfs_device_id(leaf, dev_item);
3041 read_extent_buffer(leaf, dev_uuid,
3042 (unsigned long)btrfs_device_uuid(dev_item),
3043 BTRFS_UUID_SIZE);
3044 read_extent_buffer(leaf, fs_uuid,
3045 (unsigned long)btrfs_device_fsid(dev_item),
3046 BTRFS_UUID_SIZE);
3047
3048 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3049 ret = open_seed_devices(root, fs_uuid);
3050 if (ret && !btrfs_test_opt(root, DEGRADED))
3051 return ret;
3052 }
3053
3054 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3055 if (!device || !device->bdev) {
3056 if (!btrfs_test_opt(root, DEGRADED))
3057 return -EIO;
3058
3059 if (!device) {
3060 printk(KERN_WARNING "warning devid %llu missing\n",
3061 (unsigned long long)devid);
3062 device = add_missing_dev(root, devid, dev_uuid);
3063 if (!device)
3064 return -ENOMEM;
3065 }
3066 }
3067
3068 if (device->fs_devices != root->fs_info->fs_devices) {
3069 BUG_ON(device->writeable);
3070 if (device->generation !=
3071 btrfs_device_generation(leaf, dev_item))
3072 return -EINVAL;
3073 }
3074
3075 fill_device_from_item(leaf, dev_item, device);
3076 device->dev_root = root->fs_info->dev_root;
3077 device->in_fs_metadata = 1;
3078 if (device->writeable)
3079 device->fs_devices->total_rw_bytes += device->total_bytes;
3080 ret = 0;
3081 return ret;
3082 }
3083
3084 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3085 {
3086 struct btrfs_dev_item *dev_item;
3087
3088 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3089 dev_item);
3090 return read_one_dev(root, buf, dev_item);
3091 }
3092
3093 int btrfs_read_sys_array(struct btrfs_root *root)
3094 {
3095 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3096 struct extent_buffer *sb;
3097 struct btrfs_disk_key *disk_key;
3098 struct btrfs_chunk *chunk;
3099 u8 *ptr;
3100 unsigned long sb_ptr;
3101 int ret = 0;
3102 u32 num_stripes;
3103 u32 array_size;
3104 u32 len = 0;
3105 u32 cur;
3106 struct btrfs_key key;
3107
3108 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3109 BTRFS_SUPER_INFO_SIZE);
3110 if (!sb)
3111 return -ENOMEM;
3112 btrfs_set_buffer_uptodate(sb);
3113 btrfs_set_buffer_lockdep_class(sb, 0);
3114
3115 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3116 array_size = btrfs_super_sys_array_size(super_copy);
3117
3118 ptr = super_copy->sys_chunk_array;
3119 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3120 cur = 0;
3121
3122 while (cur < array_size) {
3123 disk_key = (struct btrfs_disk_key *)ptr;
3124 btrfs_disk_key_to_cpu(&key, disk_key);
3125
3126 len = sizeof(*disk_key); ptr += len;
3127 sb_ptr += len;
3128 cur += len;
3129
3130 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3131 chunk = (struct btrfs_chunk *)sb_ptr;
3132 ret = read_one_chunk(root, &key, sb, chunk);
3133 if (ret)
3134 break;
3135 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3136 len = btrfs_chunk_item_size(num_stripes);
3137 } else {
3138 ret = -EIO;
3139 break;
3140 }
3141 ptr += len;
3142 sb_ptr += len;
3143 cur += len;
3144 }
3145 free_extent_buffer(sb);
3146 return ret;
3147 }
3148
3149 int btrfs_read_chunk_tree(struct btrfs_root *root)
3150 {
3151 struct btrfs_path *path;
3152 struct extent_buffer *leaf;
3153 struct btrfs_key key;
3154 struct btrfs_key found_key;
3155 int ret;
3156 int slot;
3157
3158 root = root->fs_info->chunk_root;
3159
3160 path = btrfs_alloc_path();
3161 if (!path)
3162 return -ENOMEM;
3163
3164 /* first we search for all of the device items, and then we
3165 * read in all of the chunk items. This way we can create chunk
3166 * mappings that reference all of the devices that are afound
3167 */
3168 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3169 key.offset = 0;
3170 key.type = 0;
3171 again:
3172 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3173 while (1) {
3174 leaf = path->nodes[0];
3175 slot = path->slots[0];
3176 if (slot >= btrfs_header_nritems(leaf)) {
3177 ret = btrfs_next_leaf(root, path);
3178 if (ret == 0)
3179 continue;
3180 if (ret < 0)
3181 goto error;
3182 break;
3183 }
3184 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3185 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3186 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3187 break;
3188 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3189 struct btrfs_dev_item *dev_item;
3190 dev_item = btrfs_item_ptr(leaf, slot,
3191 struct btrfs_dev_item);
3192 ret = read_one_dev(root, leaf, dev_item);
3193 if (ret)
3194 goto error;
3195 }
3196 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3197 struct btrfs_chunk *chunk;
3198 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3199 ret = read_one_chunk(root, &found_key, leaf, chunk);
3200 if (ret)
3201 goto error;
3202 }
3203 path->slots[0]++;
3204 }
3205 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3206 key.objectid = 0;
3207 btrfs_release_path(root, path);
3208 goto again;
3209 }
3210 ret = 0;
3211 error:
3212 btrfs_free_path(path);
3213 return ret;
3214 }
Support for the Chinese Samsung S3C2440 based development boards