?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs progs v4.17.1
[btrfs-progs-unstable/devel.git] / volumes.c
blobd81b348eb14d013014d781af2d42cec141ac5ab4
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 <stdio.h>
19 #include <stdlib.h>
20 #include <sys/types.h>
21 #include <sys/stat.h>
22 #include <uuid/uuid.h>
23 #include <fcntl.h>
24 #include <unistd.h>
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "transaction.h"
28 #include "print-tree.h"
29 #include "volumes.h"
30 #include "utils.h"
31 #include "kernel-lib/raid56.h"
32
33 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
34 [BTRFS_RAID_RAID10] = {
35 .sub_stripes = 2,
36 .dev_stripes = 1,
37 .devs_max = 0, /* 0 == as many as possible */
38 .devs_min = 4,
39 .tolerated_failures = 1,
40 .devs_increment = 2,
41 .ncopies = 2,
42 },
43 [BTRFS_RAID_RAID1] = {
44 .sub_stripes = 1,
45 .dev_stripes = 1,
46 .devs_max = 2,
47 .devs_min = 2,
48 .tolerated_failures = 1,
49 .devs_increment = 2,
50 .ncopies = 2,
51 },
52 [BTRFS_RAID_DUP] = {
53 .sub_stripes = 1,
54 .dev_stripes = 2,
55 .devs_max = 1,
56 .devs_min = 1,
57 .tolerated_failures = 0,
58 .devs_increment = 1,
59 .ncopies = 2,
60 },
61 [BTRFS_RAID_RAID0] = {
62 .sub_stripes = 1,
63 .dev_stripes = 1,
64 .devs_max = 0,
65 .devs_min = 2,
66 .tolerated_failures = 0,
67 .devs_increment = 1,
68 .ncopies = 1,
69 },
70 [BTRFS_RAID_SINGLE] = {
71 .sub_stripes = 1,
72 .dev_stripes = 1,
73 .devs_max = 1,
74 .devs_min = 1,
75 .tolerated_failures = 0,
76 .devs_increment = 1,
77 .ncopies = 1,
78 },
79 [BTRFS_RAID_RAID5] = {
80 .sub_stripes = 1,
81 .dev_stripes = 1,
82 .devs_max = 0,
83 .devs_min = 2,
84 .tolerated_failures = 1,
85 .devs_increment = 1,
86 .ncopies = 2,
87 },
88 [BTRFS_RAID_RAID6] = {
89 .sub_stripes = 1,
90 .dev_stripes = 1,
91 .devs_max = 0,
92 .devs_min = 3,
93 .tolerated_failures = 2,
94 .devs_increment = 1,
95 .ncopies = 3,
96 },
97 };
98
99 struct stripe {
100 struct btrfs_device *dev;
101 u64 physical;
102 };
103
104 static inline int nr_parity_stripes(struct map_lookup *map)
105 {
106 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
107 return 1;
108 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
109 return 2;
110 else
111 return 0;
112 }
113
114 static inline int nr_data_stripes(struct map_lookup *map)
115 {
116 return map->num_stripes - nr_parity_stripes(map);
117 }
118
119 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
120
121 static LIST_HEAD(fs_uuids);
122
123 /*
124 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
125 * return NULL.
126 *
127 * If devid and uuid are both specified, the match must be exact, otherwise
128 * only devid is used.
129 */
130 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
131 u64 devid, u8 *uuid)
132 {
133 struct list_head *head = &fs_devices->devices;
134 struct btrfs_device *dev;
135
136 list_for_each_entry(dev, head, dev_list) {
137 if (dev->devid == devid &&
138 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
139 return dev;
140 }
141 }
142 return NULL;
143 }
144
145 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
146 {
147 struct btrfs_fs_devices *fs_devices;
148
149 list_for_each_entry(fs_devices, &fs_uuids, list) {
150 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
151 return fs_devices;
152 }
153 return NULL;
154 }
155
156 static int device_list_add(const char *path,
157 struct btrfs_super_block *disk_super,
158 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
159 {
160 struct btrfs_device *device;
161 struct btrfs_fs_devices *fs_devices;
162 u64 found_transid = btrfs_super_generation(disk_super);
163
164 fs_devices = find_fsid(disk_super->fsid);
165 if (!fs_devices) {
166 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
167 if (!fs_devices)
168 return -ENOMEM;
169 INIT_LIST_HEAD(&fs_devices->devices);
170 list_add(&fs_devices->list, &fs_uuids);
171 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
172 fs_devices->latest_devid = devid;
173 fs_devices->latest_trans = found_transid;
174 fs_devices->lowest_devid = (u64)-1;
175 device = NULL;
176 } else {
177 device = find_device(fs_devices, devid,
178 disk_super->dev_item.uuid);
179 }
180 if (!device) {
181 device = kzalloc(sizeof(*device), GFP_NOFS);
182 if (!device) {
183 /* we can safely leave the fs_devices entry around */
184 return -ENOMEM;
185 }
186 device->fd = -1;
187 device->devid = devid;
188 device->generation = found_transid;
189 memcpy(device->uuid, disk_super->dev_item.uuid,
190 BTRFS_UUID_SIZE);
191 device->name = kstrdup(path, GFP_NOFS);
192 if (!device->name) {
193 kfree(device);
194 return -ENOMEM;
195 }
196 device->label = kstrdup(disk_super->label, GFP_NOFS);
197 if (!device->label) {
198 kfree(device->name);
199 kfree(device);
200 return -ENOMEM;
201 }
202 device->total_devs = btrfs_super_num_devices(disk_super);
203 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
204 device->total_bytes =
205 btrfs_stack_device_total_bytes(&disk_super->dev_item);
206 device->bytes_used =
207 btrfs_stack_device_bytes_used(&disk_super->dev_item);
208 list_add(&device->dev_list, &fs_devices->devices);
209 device->fs_devices = fs_devices;
210 } else if (!device->name || strcmp(device->name, path)) {
211 char *name;
212
213 /*
214 * The existing device has newer generation, so this one could
215 * be a stale one, don't add it.
216 */
217 if (found_transid < device->generation) {
218 warning(
219 "adding device %s gen %llu but found an existing device %s gen %llu",
220 path, found_transid, device->name,
221 device->generation);
222 return -EEXIST;
223 }
224
225 name = strdup(path);
226 if (!name)
227 return -ENOMEM;
228 kfree(device->name);
229 device->name = name;
230 }
231
232
233 if (found_transid > fs_devices->latest_trans) {
234 fs_devices->latest_devid = devid;
235 fs_devices->latest_trans = found_transid;
236 }
237 if (fs_devices->lowest_devid > devid) {
238 fs_devices->lowest_devid = devid;
239 }
240 *fs_devices_ret = fs_devices;
241 return 0;
242 }
243
244 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
245 {
246 struct btrfs_fs_devices *seed_devices;
247 struct btrfs_device *device;
248 int ret = 0;
249
250 again:
251 if (!fs_devices)
252 return 0;
253 while (!list_empty(&fs_devices->devices)) {
254 device = list_entry(fs_devices->devices.next,
255 struct btrfs_device, dev_list);
256 if (device->fd != -1) {
257 if (device->writeable && fsync(device->fd) == -1) {
258 warning("fsync on device %llu failed: %m",
259 device->devid);
260 ret = -errno;
261 }
262 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
263 fprintf(stderr, "Warning, could not drop caches\n");
264 close(device->fd);
265 device->fd = -1;
266 }
267 device->writeable = 0;
268 list_del(&device->dev_list);
269 /* free the memory */
270 free(device->name);
271 free(device->label);
272 free(device);
273 }
274
275 seed_devices = fs_devices->seed;
276 fs_devices->seed = NULL;
277 if (seed_devices) {
278 struct btrfs_fs_devices *orig;
279
280 orig = fs_devices;
281 fs_devices = seed_devices;
282 list_del(&orig->list);
283 free(orig);
284 goto again;
285 } else {
286 list_del(&fs_devices->list);
287 free(fs_devices);
288 }
289
290 return ret;
291 }
292
293 void btrfs_close_all_devices(void)
294 {
295 struct btrfs_fs_devices *fs_devices;
296
297 while (!list_empty(&fs_uuids)) {
298 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
299 list);
300 btrfs_close_devices(fs_devices);
301 }
302 }
303
304 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
305 {
306 int fd;
307 struct btrfs_device *device;
308 int ret;
309
310 list_for_each_entry(device, &fs_devices->devices, dev_list) {
311 if (!device->name) {
312 printk("no name for device %llu, skip it now\n", device->devid);
313 continue;
314 }
315
316 fd = open(device->name, flags);
317 if (fd < 0) {
318 ret = -errno;
319 error("cannot open device '%s': %m", device->name);
320 goto fail;
321 }
322
323 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
324 fprintf(stderr, "Warning, could not drop caches\n");
325
326 if (device->devid == fs_devices->latest_devid)
327 fs_devices->latest_bdev = fd;
328 if (device->devid == fs_devices->lowest_devid)
329 fs_devices->lowest_bdev = fd;
330 device->fd = fd;
331 if (flags & O_RDWR)
332 device->writeable = 1;
333 }
334 return 0;
335 fail:
336 btrfs_close_devices(fs_devices);
337 return ret;
338 }
339
340 int btrfs_scan_one_device(int fd, const char *path,
341 struct btrfs_fs_devices **fs_devices_ret,
342 u64 *total_devs, u64 super_offset, unsigned sbflags)
343 {
344 struct btrfs_super_block *disk_super;
345 char buf[BTRFS_SUPER_INFO_SIZE];
346 int ret;
347 u64 devid;
348
349 disk_super = (struct btrfs_super_block *)buf;
350 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
351 if (ret < 0)
352 return -EIO;
353 devid = btrfs_stack_device_id(&disk_super->dev_item);
354 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
355 *total_devs = 1;
356 else
357 *total_devs = btrfs_super_num_devices(disk_super);
358
359 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
360
361 return ret;
362 }
363
364 /*
365 * find_free_dev_extent_start - find free space in the specified device
366 * @device: the device which we search the free space in
367 * @num_bytes: the size of the free space that we need
368 * @search_start: the position from which to begin the search
369 * @start: store the start of the free space.
370 * @len: the size of the free space. that we find, or the size
371 * of the max free space if we don't find suitable free space
372 *
373 * this uses a pretty simple search, the expectation is that it is
374 * called very infrequently and that a given device has a small number
375 * of extents
376 *
377 * @start is used to store the start of the free space if we find. But if we
378 * don't find suitable free space, it will be used to store the start position
379 * of the max free space.
380 *
381 * @len is used to store the size of the free space that we find.
382 * But if we don't find suitable free space, it is used to store the size of
383 * the max free space.
384 */
385 static int find_free_dev_extent_start(struct btrfs_device *device,
386 u64 num_bytes, u64 search_start,
387 u64 *start, u64 *len)
388 {
389 struct btrfs_key key;
390 struct btrfs_root *root = device->dev_root;
391 struct btrfs_dev_extent *dev_extent;
392 struct btrfs_path *path;
393 u64 hole_size;
394 u64 max_hole_start;
395 u64 max_hole_size;
396 u64 extent_end;
397 u64 search_end = device->total_bytes;
398 int ret;
399 int slot;
400 struct extent_buffer *l;
401 u64 min_search_start;
402
403 /*
404 * We don't want to overwrite the superblock on the drive nor any area
405 * used by the boot loader (grub for example), so we make sure to start
406 * at an offset of at least 1MB.
407 */
408 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
409 search_start = max(search_start, min_search_start);
410
411 path = btrfs_alloc_path();
412 if (!path)
413 return -ENOMEM;
414
415 max_hole_start = search_start;
416 max_hole_size = 0;
417
418 if (search_start >= search_end) {
419 ret = -ENOSPC;
420 goto out;
421 }
422
423 path->reada = READA_FORWARD;
424
425 key.objectid = device->devid;
426 key.offset = search_start;
427 key.type = BTRFS_DEV_EXTENT_KEY;
428
429 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
430 if (ret < 0)
431 goto out;
432 if (ret > 0) {
433 ret = btrfs_previous_item(root, path, key.objectid, key.type);
434 if (ret < 0)
435 goto out;
436 }
437
438 while (1) {
439 l = path->nodes[0];
440 slot = path->slots[0];
441 if (slot >= btrfs_header_nritems(l)) {
442 ret = btrfs_next_leaf(root, path);
443 if (ret == 0)
444 continue;
445 if (ret < 0)
446 goto out;
447
448 break;
449 }
450 btrfs_item_key_to_cpu(l, &key, slot);
451
452 if (key.objectid < device->devid)
453 goto next;
454
455 if (key.objectid > device->devid)
456 break;
457
458 if (key.type != BTRFS_DEV_EXTENT_KEY)
459 goto next;
460
461 if (key.offset > search_start) {
462 hole_size = key.offset - search_start;
463
464 /*
465 * Have to check before we set max_hole_start, otherwise
466 * we could end up sending back this offset anyway.
467 */
468 if (hole_size > max_hole_size) {
469 max_hole_start = search_start;
470 max_hole_size = hole_size;
471 }
472
473 /*
474 * If this free space is greater than which we need,
475 * it must be the max free space that we have found
476 * until now, so max_hole_start must point to the start
477 * of this free space and the length of this free space
478 * is stored in max_hole_size. Thus, we return
479 * max_hole_start and max_hole_size and go back to the
480 * caller.
481 */
482 if (hole_size >= num_bytes) {
483 ret = 0;
484 goto out;
485 }
486 }
487
488 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
489 extent_end = key.offset + btrfs_dev_extent_length(l,
490 dev_extent);
491 if (extent_end > search_start)
492 search_start = extent_end;
493 next:
494 path->slots[0]++;
495 cond_resched();
496 }
497
498 /*
499 * At this point, search_start should be the end of
500 * allocated dev extents, and when shrinking the device,
501 * search_end may be smaller than search_start.
502 */
503 if (search_end > search_start) {
504 hole_size = search_end - search_start;
505
506 if (hole_size > max_hole_size) {
507 max_hole_start = search_start;
508 max_hole_size = hole_size;
509 }
510 }
511
512 /* See above. */
513 if (max_hole_size < num_bytes)
514 ret = -ENOSPC;
515 else
516 ret = 0;
517
518 out:
519 btrfs_free_path(path);
520 *start = max_hole_start;
521 if (len)
522 *len = max_hole_size;
523 return ret;
524 }
525
526 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
527 u64 *start, u64 *len)
528 {
529 /* FIXME use last free of some kind */
530 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
531 }
532
533 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
534 struct btrfs_device *device,
535 u64 chunk_offset, u64 num_bytes, u64 *start,
536 int convert)
537 {
538 int ret;
539 struct btrfs_path *path;
540 struct btrfs_root *root = device->dev_root;
541 struct btrfs_dev_extent *extent;
542 struct extent_buffer *leaf;
543 struct btrfs_key key;
544
545 path = btrfs_alloc_path();
546 if (!path)
547 return -ENOMEM;
548
549 /*
550 * For convert case, just skip search free dev_extent, as caller
551 * is responsible to make sure it's free.
552 */
553 if (!convert) {
554 ret = find_free_dev_extent(device, num_bytes, start, NULL);
555 if (ret)
556 goto err;
557 }
558
559 key.objectid = device->devid;
560 key.offset = *start;
561 key.type = BTRFS_DEV_EXTENT_KEY;
562 ret = btrfs_insert_empty_item(trans, root, path, &key,
563 sizeof(*extent));
564 BUG_ON(ret);
565
566 leaf = path->nodes[0];
567 extent = btrfs_item_ptr(leaf, path->slots[0],
568 struct btrfs_dev_extent);
569 btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID);
570 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
571 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
572 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
573
574 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
575 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
576 BTRFS_UUID_SIZE);
577
578 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
579 btrfs_mark_buffer_dirty(leaf);
580 err:
581 btrfs_free_path(path);
582 return ret;
583 }
584
585 static int find_next_chunk(struct btrfs_fs_info *fs_info, u64 *offset)
586 {
587 struct btrfs_root *root = fs_info->chunk_root;
588 struct btrfs_path *path;
589 int ret;
590 struct btrfs_key key;
591 struct btrfs_chunk *chunk;
592 struct btrfs_key found_key;
593
594 path = btrfs_alloc_path();
595 if (!path)
596 return -ENOMEM;
597
598 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
599 key.offset = (u64)-1;
600 key.type = BTRFS_CHUNK_ITEM_KEY;
601
602 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
603 if (ret < 0)
604 goto error;
605
606 BUG_ON(ret == 0);
607
608 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
609 if (ret) {
610 *offset = 0;
611 } else {
612 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
613 path->slots[0]);
614 if (found_key.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
615 *offset = 0;
616 else {
617 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
618 struct btrfs_chunk);
619 *offset = found_key.offset +
620 btrfs_chunk_length(path->nodes[0], chunk);
621 }
622 }
623 ret = 0;
624 error:
625 btrfs_free_path(path);
626 return ret;
627 }
628
629 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
630 u64 *objectid)
631 {
632 int ret;
633 struct btrfs_key key;
634 struct btrfs_key found_key;
635
636 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
637 key.type = BTRFS_DEV_ITEM_KEY;
638 key.offset = (u64)-1;
639
640 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
641 if (ret < 0)
642 goto error;
643
644 BUG_ON(ret == 0);
645
646 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
647 BTRFS_DEV_ITEM_KEY);
648 if (ret) {
649 *objectid = 1;
650 } else {
651 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
652 path->slots[0]);
653 *objectid = found_key.offset + 1;
654 }
655 ret = 0;
656 error:
657 btrfs_release_path(path);
658 return ret;
659 }
660
661 /*
662 * the device information is stored in the chunk root
663 * the btrfs_device struct should be fully filled in
664 */
665 int btrfs_add_device(struct btrfs_trans_handle *trans,
666 struct btrfs_fs_info *fs_info,
667 struct btrfs_device *device)
668 {
669 int ret;
670 struct btrfs_path *path;
671 struct btrfs_dev_item *dev_item;
672 struct extent_buffer *leaf;
673 struct btrfs_key key;
674 struct btrfs_root *root = fs_info->chunk_root;
675 unsigned long ptr;
676 u64 free_devid = 0;
677
678 path = btrfs_alloc_path();
679 if (!path)
680 return -ENOMEM;
681
682 ret = find_next_devid(root, path, &free_devid);
683 if (ret)
684 goto out;
685
686 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
687 key.type = BTRFS_DEV_ITEM_KEY;
688 key.offset = free_devid;
689
690 ret = btrfs_insert_empty_item(trans, root, path, &key,
691 sizeof(*dev_item));
692 if (ret)
693 goto out;
694
695 leaf = path->nodes[0];
696 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
697
698 device->devid = free_devid;
699 btrfs_set_device_id(leaf, dev_item, device->devid);
700 btrfs_set_device_generation(leaf, dev_item, 0);
701 btrfs_set_device_type(leaf, dev_item, device->type);
702 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
703 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
704 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
705 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
706 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
707 btrfs_set_device_group(leaf, dev_item, 0);
708 btrfs_set_device_seek_speed(leaf, dev_item, 0);
709 btrfs_set_device_bandwidth(leaf, dev_item, 0);
710 btrfs_set_device_start_offset(leaf, dev_item, 0);
711
712 ptr = (unsigned long)btrfs_device_uuid(dev_item);
713 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
714 ptr = (unsigned long)btrfs_device_fsid(dev_item);
715 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
716 btrfs_mark_buffer_dirty(leaf);
717 ret = 0;
718
719 out:
720 btrfs_free_path(path);
721 return ret;
722 }
723
724 int btrfs_update_device(struct btrfs_trans_handle *trans,
725 struct btrfs_device *device)
726 {
727 int ret;
728 struct btrfs_path *path;
729 struct btrfs_root *root;
730 struct btrfs_dev_item *dev_item;
731 struct extent_buffer *leaf;
732 struct btrfs_key key;
733
734 root = device->dev_root->fs_info->chunk_root;
735
736 path = btrfs_alloc_path();
737 if (!path)
738 return -ENOMEM;
739
740 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
741 key.type = BTRFS_DEV_ITEM_KEY;
742 key.offset = device->devid;
743
744 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
745 if (ret < 0)
746 goto out;
747
748 if (ret > 0) {
749 ret = -ENOENT;
750 goto out;
751 }
752
753 leaf = path->nodes[0];
754 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
755
756 btrfs_set_device_id(leaf, dev_item, device->devid);
757 btrfs_set_device_type(leaf, dev_item, device->type);
758 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
759 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
760 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
761 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
762 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
763 btrfs_mark_buffer_dirty(leaf);
764
765 out:
766 btrfs_free_path(path);
767 return ret;
768 }
769
770 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
771 struct btrfs_chunk *chunk, int item_size)
772 {
773 struct btrfs_super_block *super_copy = fs_info->super_copy;
774 struct btrfs_disk_key disk_key;
775 u32 array_size;
776 u8 *ptr;
777
778 array_size = btrfs_super_sys_array_size(super_copy);
779 if (array_size + item_size + sizeof(disk_key)
780 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
781 return -EFBIG;
782
783 ptr = super_copy->sys_chunk_array + array_size;
784 btrfs_cpu_key_to_disk(&disk_key, key);
785 memcpy(ptr, &disk_key, sizeof(disk_key));
786 ptr += sizeof(disk_key);
787 memcpy(ptr, chunk, item_size);
788 item_size += sizeof(disk_key);
789 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
790 return 0;
791 }
792
793 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
794 int sub_stripes)
795 {
796 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
797 return calc_size;
798 else if (type & BTRFS_BLOCK_GROUP_RAID10)
799 return calc_size * (num_stripes / sub_stripes);
800 else if (type & BTRFS_BLOCK_GROUP_RAID5)
801 return calc_size * (num_stripes - 1);
802 else if (type & BTRFS_BLOCK_GROUP_RAID6)
803 return calc_size * (num_stripes - 2);
804 else
805 return calc_size * num_stripes;
806 }
807
808
809 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
810 {
811 /* TODO, add a way to store the preferred stripe size */
812 return BTRFS_STRIPE_LEN;
813 }
814
815 /*
816 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
817 *
818 * It is not equal to "device->total_bytes - device->bytes_used".
819 * We do not allocate any chunk in 1M at beginning of device, and not
820 * allowed to allocate any chunk before alloc_start if it is specified.
821 * So search holes from max(1M, alloc_start) to device->total_bytes.
822 */
823 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
824 struct btrfs_device *device,
825 u64 *avail_bytes)
826 {
827 struct btrfs_path *path;
828 struct btrfs_root *root = device->dev_root;
829 struct btrfs_key key;
830 struct btrfs_dev_extent *dev_extent = NULL;
831 struct extent_buffer *l;
832 u64 search_start = root->fs_info->alloc_start;
833 u64 search_end = device->total_bytes;
834 u64 extent_end = 0;
835 u64 free_bytes = 0;
836 int ret;
837 int slot = 0;
838
839 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
840
841 path = btrfs_alloc_path();
842 if (!path)
843 return -ENOMEM;
844
845 key.objectid = device->devid;
846 key.offset = root->fs_info->alloc_start;
847 key.type = BTRFS_DEV_EXTENT_KEY;
848
849 path->reada = READA_FORWARD;
850 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
851 if (ret < 0)
852 goto error;
853 ret = btrfs_previous_item(root, path, 0, key.type);
854 if (ret < 0)
855 goto error;
856
857 while (1) {
858 l = path->nodes[0];
859 slot = path->slots[0];
860 if (slot >= btrfs_header_nritems(l)) {
861 ret = btrfs_next_leaf(root, path);
862 if (ret == 0)
863 continue;
864 if (ret < 0)
865 goto error;
866 break;
867 }
868 btrfs_item_key_to_cpu(l, &key, slot);
869
870 if (key.objectid < device->devid)
871 goto next;
872 if (key.objectid > device->devid)
873 break;
874 if (key.type != BTRFS_DEV_EXTENT_KEY)
875 goto next;
876 if (key.offset > search_end)
877 break;
878 if (key.offset > search_start)
879 free_bytes += key.offset - search_start;
880
881 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
882 extent_end = key.offset + btrfs_dev_extent_length(l,
883 dev_extent);
884 if (extent_end > search_start)
885 search_start = extent_end;
886 if (search_start > search_end)
887 break;
888 next:
889 path->slots[0]++;
890 cond_resched();
891 }
892
893 if (search_start < search_end)
894 free_bytes += search_end - search_start;
895
896 *avail_bytes = free_bytes;
897 ret = 0;
898 error:
899 btrfs_free_path(path);
900 return ret;
901 }
902
903 #define BTRFS_MAX_DEVS(info) ((BTRFS_LEAF_DATA_SIZE(info) \
904 - sizeof(struct btrfs_item) \
905 - sizeof(struct btrfs_chunk)) \
906 / sizeof(struct btrfs_stripe) + 1)
907
908 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
909 - 2 * sizeof(struct btrfs_disk_key) \
910 - 2 * sizeof(struct btrfs_chunk)) \
911 / sizeof(struct btrfs_stripe) + 1)
912
913 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
914 struct btrfs_fs_info *info, u64 *start,
915 u64 *num_bytes, u64 type)
916 {
917 u64 dev_offset;
918 struct btrfs_root *extent_root = info->extent_root;
919 struct btrfs_root *chunk_root = info->chunk_root;
920 struct btrfs_stripe *stripes;
921 struct btrfs_device *device = NULL;
922 struct btrfs_chunk *chunk;
923 struct list_head private_devs;
924 struct list_head *dev_list = &info->fs_devices->devices;
925 struct list_head *cur;
926 struct map_lookup *map;
927 int min_stripe_size = SZ_1M;
928 u64 calc_size = SZ_8M;
929 u64 min_free;
930 u64 max_chunk_size = 4 * calc_size;
931 u64 avail = 0;
932 u64 max_avail = 0;
933 u64 percent_max;
934 int num_stripes = 1;
935 int max_stripes = 0;
936 int min_stripes = 1;
937 int sub_stripes = 0;
938 int looped = 0;
939 int ret;
940 int index;
941 int stripe_len = BTRFS_STRIPE_LEN;
942 struct btrfs_key key;
943 u64 offset;
944
945 if (list_empty(dev_list)) {
946 return -ENOSPC;
947 }
948
949 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
950 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
951 calc_size = SZ_8M;
952 max_chunk_size = calc_size * 2;
953 min_stripe_size = SZ_1M;
954 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
955 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
956 calc_size = SZ_1G;
957 max_chunk_size = 10 * calc_size;
958 min_stripe_size = SZ_64M;
959 max_stripes = BTRFS_MAX_DEVS(info);
960 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
961 calc_size = SZ_1G;
962 max_chunk_size = 4 * calc_size;
963 min_stripe_size = SZ_32M;
964 max_stripes = BTRFS_MAX_DEVS(info);
965 }
966 }
967 if (type & BTRFS_BLOCK_GROUP_RAID1) {
968 num_stripes = min_t(u64, 2,
969 btrfs_super_num_devices(info->super_copy));
970 if (num_stripes < 2)
971 return -ENOSPC;
972 min_stripes = 2;
973 }
974 if (type & BTRFS_BLOCK_GROUP_DUP) {
975 num_stripes = 2;
976 min_stripes = 2;
977 }
978 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
979 num_stripes = btrfs_super_num_devices(info->super_copy);
980 if (num_stripes > max_stripes)
981 num_stripes = max_stripes;
982 min_stripes = 2;
983 }
984 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
985 num_stripes = btrfs_super_num_devices(info->super_copy);
986 if (num_stripes > max_stripes)
987 num_stripes = max_stripes;
988 if (num_stripes < 4)
989 return -ENOSPC;
990 num_stripes &= ~(u32)1;
991 sub_stripes = 2;
992 min_stripes = 4;
993 }
994 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
995 num_stripes = btrfs_super_num_devices(info->super_copy);
996 if (num_stripes > max_stripes)
997 num_stripes = max_stripes;
998 if (num_stripes < 2)
999 return -ENOSPC;
1000 min_stripes = 2;
1001 stripe_len = find_raid56_stripe_len(num_stripes - 1,
1002 btrfs_super_stripesize(info->super_copy));
1003 }
1004 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
1005 num_stripes = btrfs_super_num_devices(info->super_copy);
1006 if (num_stripes > max_stripes)
1007 num_stripes = max_stripes;
1008 if (num_stripes < 3)
1009 return -ENOSPC;
1010 min_stripes = 3;
1011 stripe_len = find_raid56_stripe_len(num_stripes - 2,
1012 btrfs_super_stripesize(info->super_copy));
1013 }
1014
1015 /* we don't want a chunk larger than 10% of the FS */
1016 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
1017 max_chunk_size = min(percent_max, max_chunk_size);
1018
1019 again:
1020 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
1021 max_chunk_size) {
1022 calc_size = max_chunk_size;
1023 calc_size /= num_stripes;
1024 calc_size /= stripe_len;
1025 calc_size *= stripe_len;
1026 }
1027 /* we don't want tiny stripes */
1028 calc_size = max_t(u64, calc_size, min_stripe_size);
1029
1030 calc_size /= stripe_len;
1031 calc_size *= stripe_len;
1032 INIT_LIST_HEAD(&private_devs);
1033 cur = dev_list->next;
1034 index = 0;
1035
1036 if (type & BTRFS_BLOCK_GROUP_DUP)
1037 min_free = calc_size * 2;
1038 else
1039 min_free = calc_size;
1040
1041 /* build a private list of devices we will allocate from */
1042 while(index < num_stripes) {
1043 device = list_entry(cur, struct btrfs_device, dev_list);
1044 ret = btrfs_device_avail_bytes(trans, device, &avail);
1045 if (ret)
1046 return ret;
1047 cur = cur->next;
1048 if (avail >= min_free) {
1049 list_move_tail(&device->dev_list, &private_devs);
1050 index++;
1051 if (type & BTRFS_BLOCK_GROUP_DUP)
1052 index++;
1053 } else if (avail > max_avail)
1054 max_avail = avail;
1055 if (cur == dev_list)
1056 break;
1057 }
1058 if (index < num_stripes) {
1059 list_splice(&private_devs, dev_list);
1060 if (index >= min_stripes) {
1061 num_stripes = index;
1062 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1063 num_stripes /= sub_stripes;
1064 num_stripes *= sub_stripes;
1065 }
1066 looped = 1;
1067 goto again;
1068 }
1069 if (!looped && max_avail > 0) {
1070 looped = 1;
1071 calc_size = max_avail;
1072 goto again;
1073 }
1074 return -ENOSPC;
1075 }
1076 ret = find_next_chunk(info, &offset);
1077 if (ret)
1078 return ret;
1079 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1080 key.type = BTRFS_CHUNK_ITEM_KEY;
1081 key.offset = offset;
1082
1083 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1084 if (!chunk)
1085 return -ENOMEM;
1086
1087 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1088 if (!map) {
1089 kfree(chunk);
1090 return -ENOMEM;
1091 }
1092
1093 stripes = &chunk->stripe;
1094 *num_bytes = chunk_bytes_by_type(type, calc_size,
1095 num_stripes, sub_stripes);
1096 index = 0;
1097 while(index < num_stripes) {
1098 struct btrfs_stripe *stripe;
1099 BUG_ON(list_empty(&private_devs));
1100 cur = private_devs.next;
1101 device = list_entry(cur, struct btrfs_device, dev_list);
1102
1103 /* loop over this device again if we're doing a dup group */
1104 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1105 (index == num_stripes - 1))
1106 list_move_tail(&device->dev_list, dev_list);
1107
1108 ret = btrfs_alloc_dev_extent(trans, device, key.offset,
1109 calc_size, &dev_offset, 0);
1110 if (ret < 0)
1111 goto out_chunk_map;
1112
1113 device->bytes_used += calc_size;
1114 ret = btrfs_update_device(trans, device);
1115 if (ret < 0)
1116 goto out_chunk_map;
1117
1118 map->stripes[index].dev = device;
1119 map->stripes[index].physical = dev_offset;
1120 stripe = stripes + index;
1121 btrfs_set_stack_stripe_devid(stripe, device->devid);
1122 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1123 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1124 index++;
1125 }
1126 BUG_ON(!list_empty(&private_devs));
1127
1128 /* key was set above */
1129 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1130 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1131 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1132 btrfs_set_stack_chunk_type(chunk, type);
1133 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1134 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1135 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1136 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1137 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1138 map->sector_size = info->sectorsize;
1139 map->stripe_len = stripe_len;
1140 map->io_align = stripe_len;
1141 map->io_width = stripe_len;
1142 map->type = type;
1143 map->num_stripes = num_stripes;
1144 map->sub_stripes = sub_stripes;
1145
1146 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1147 btrfs_chunk_item_size(num_stripes));
1148 BUG_ON(ret);
1149 *start = key.offset;;
1150
1151 map->ce.start = key.offset;
1152 map->ce.size = *num_bytes;
1153
1154 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1155 if (ret < 0)
1156 goto out_chunk_map;
1157
1158 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1159 ret = btrfs_add_system_chunk(info, &key,
1160 chunk, btrfs_chunk_item_size(num_stripes));
1161 if (ret < 0)
1162 goto out_chunk;
1163 }
1164
1165 kfree(chunk);
1166 return ret;
1167
1168 out_chunk_map:
1169 kfree(map);
1170 out_chunk:
1171 kfree(chunk);
1172 return ret;
1173 }
1174
1175 /*
1176 * Alloc a DATA chunk with SINGLE profile.
1177 *
1178 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1179 * (btrfs logical bytenr == on-disk bytenr)
1180 * For that case, caller must make sure the chunk and dev_extent are not
1181 * occupied.
1182 */
1183 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1184 struct btrfs_fs_info *info, u64 *start,
1185 u64 num_bytes, u64 type, int convert)
1186 {
1187 u64 dev_offset;
1188 struct btrfs_root *extent_root = info->extent_root;
1189 struct btrfs_root *chunk_root = info->chunk_root;
1190 struct btrfs_stripe *stripes;
1191 struct btrfs_device *device = NULL;
1192 struct btrfs_chunk *chunk;
1193 struct list_head *dev_list = &info->fs_devices->devices;
1194 struct list_head *cur;
1195 struct map_lookup *map;
1196 u64 calc_size = SZ_8M;
1197 int num_stripes = 1;
1198 int sub_stripes = 0;
1199 int ret;
1200 int index;
1201 int stripe_len = BTRFS_STRIPE_LEN;
1202 struct btrfs_key key;
1203
1204 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1205 key.type = BTRFS_CHUNK_ITEM_KEY;
1206 if (convert) {
1207 if (*start != round_down(*start, info->sectorsize)) {
1208 error("DATA chunk start not sectorsize aligned: %llu",
1209 (unsigned long long)*start);
1210 return -EINVAL;
1211 }
1212 key.offset = *start;
1213 dev_offset = *start;
1214 } else {
1215 u64 tmp;
1216
1217 ret = find_next_chunk(info, &tmp);
1218 key.offset = tmp;
1219 if (ret)
1220 return ret;
1221 }
1222
1223 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1224 if (!chunk)
1225 return -ENOMEM;
1226
1227 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1228 if (!map) {
1229 kfree(chunk);
1230 return -ENOMEM;
1231 }
1232
1233 stripes = &chunk->stripe;
1234 calc_size = num_bytes;
1235
1236 index = 0;
1237 cur = dev_list->next;
1238 device = list_entry(cur, struct btrfs_device, dev_list);
1239
1240 while (index < num_stripes) {
1241 struct btrfs_stripe *stripe;
1242
1243 ret = btrfs_alloc_dev_extent(trans, device, key.offset,
1244 calc_size, &dev_offset, convert);
1245 BUG_ON(ret);
1246
1247 device->bytes_used += calc_size;
1248 ret = btrfs_update_device(trans, device);
1249 BUG_ON(ret);
1250
1251 map->stripes[index].dev = device;
1252 map->stripes[index].physical = dev_offset;
1253 stripe = stripes + index;
1254 btrfs_set_stack_stripe_devid(stripe, device->devid);
1255 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1256 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1257 index++;
1258 }
1259
1260 /* key was set above */
1261 btrfs_set_stack_chunk_length(chunk, num_bytes);
1262 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1263 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1264 btrfs_set_stack_chunk_type(chunk, type);
1265 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1266 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1267 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1268 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1269 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1270 map->sector_size = info->sectorsize;
1271 map->stripe_len = stripe_len;
1272 map->io_align = stripe_len;
1273 map->io_width = stripe_len;
1274 map->type = type;
1275 map->num_stripes = num_stripes;
1276 map->sub_stripes = sub_stripes;
1277
1278 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1279 btrfs_chunk_item_size(num_stripes));
1280 BUG_ON(ret);
1281 if (!convert)
1282 *start = key.offset;
1283
1284 map->ce.start = key.offset;
1285 map->ce.size = num_bytes;
1286
1287 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1288 BUG_ON(ret);
1289
1290 kfree(chunk);
1291 return ret;
1292 }
1293
1294 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1295 {
1296 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1297 struct cache_extent *ce;
1298 struct map_lookup *map;
1299 int ret;
1300
1301 ce = search_cache_extent(&map_tree->cache_tree, logical);
1302 if (!ce) {
1303 fprintf(stderr, "No mapping for %llu-%llu\n",
1304 (unsigned long long)logical,
1305 (unsigned long long)logical+len);
1306 return 1;
1307 }
1308 if (ce->start > logical || ce->start + ce->size < logical) {
1309 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1310 "%llu-%llu\n", (unsigned long long)logical,
1311 (unsigned long long)logical+len,
1312 (unsigned long long)ce->start,
1313 (unsigned long long)ce->start + ce->size);
1314 return 1;
1315 }
1316 map = container_of(ce, struct map_lookup, ce);
1317
1318 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1319 ret = map->num_stripes;
1320 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1321 ret = map->sub_stripes;
1322 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1323 ret = 2;
1324 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1325 ret = 3;
1326 else
1327 ret = 1;
1328 return ret;
1329 }
1330
1331 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1332 u64 *size, u64 type)
1333 {
1334 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1335 struct cache_extent *ce;
1336 struct map_lookup *map;
1337 u64 cur = *logical;
1338
1339 ce = search_cache_extent(&map_tree->cache_tree, cur);
1340
1341 while (ce) {
1342 /*
1343 * only jump to next bg if our cur is not 0
1344 * As the initial logical for btrfs_next_bg() is 0, and
1345 * if we jump to next bg, we skipped a valid bg.
1346 */
1347 if (cur) {
1348 ce = next_cache_extent(ce);
1349 if (!ce)
1350 return -ENOENT;
1351 }
1352
1353 cur = ce->start;
1354 map = container_of(ce, struct map_lookup, ce);
1355 if (map->type & type) {
1356 *logical = ce->start;
1357 *size = ce->size;
1358 return 0;
1359 }
1360 if (!cur)
1361 ce = next_cache_extent(ce);
1362 }
1363
1364 return -ENOENT;
1365 }
1366
1367 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
1368 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
1369 {
1370 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1371 struct cache_extent *ce;
1372 struct map_lookup *map;
1373 u64 *buf;
1374 u64 bytenr;
1375 u64 length;
1376 u64 stripe_nr;
1377 u64 rmap_len;
1378 int i, j, nr = 0;
1379
1380 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1381 BUG_ON(!ce);
1382 map = container_of(ce, struct map_lookup, ce);
1383
1384 length = ce->size;
1385 rmap_len = map->stripe_len;
1386 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1387 length = ce->size / (map->num_stripes / map->sub_stripes);
1388 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1389 length = ce->size / map->num_stripes;
1390 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1391 BTRFS_BLOCK_GROUP_RAID6)) {
1392 length = ce->size / nr_data_stripes(map);
1393 rmap_len = map->stripe_len * nr_data_stripes(map);
1394 }
1395
1396 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1397
1398 for (i = 0; i < map->num_stripes; i++) {
1399 if (map->stripes[i].physical > physical ||
1400 map->stripes[i].physical + length <= physical)
1401 continue;
1402
1403 stripe_nr = (physical - map->stripes[i].physical) /
1404 map->stripe_len;
1405
1406 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1407 stripe_nr = (stripe_nr * map->num_stripes + i) /
1408 map->sub_stripes;
1409 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1410 stripe_nr = stripe_nr * map->num_stripes + i;
1411 } /* else if RAID[56], multiply by nr_data_stripes().
1412 * Alternatively, just use rmap_len below instead of
1413 * map->stripe_len */
1414
1415 bytenr = ce->start + stripe_nr * rmap_len;
1416 for (j = 0; j < nr; j++) {
1417 if (buf[j] == bytenr)
1418 break;
1419 }
1420 if (j == nr)
1421 buf[nr++] = bytenr;
1422 }
1423
1424 *logical = buf;
1425 *naddrs = nr;
1426 *stripe_len = rmap_len;
1427
1428 return 0;
1429 }
1430
1431 static inline int parity_smaller(u64 a, u64 b)
1432 {
1433 return a > b;
1434 }
1435
1436 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1437 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1438 {
1439 struct btrfs_bio_stripe s;
1440 int i;
1441 u64 l;
1442 int again = 1;
1443
1444 while (again) {
1445 again = 0;
1446 for (i = 0; i < bbio->num_stripes - 1; i++) {
1447 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1448 s = bbio->stripes[i];
1449 l = raid_map[i];
1450 bbio->stripes[i] = bbio->stripes[i+1];
1451 raid_map[i] = raid_map[i+1];
1452 bbio->stripes[i+1] = s;
1453 raid_map[i+1] = l;
1454 again = 1;
1455 }
1456 }
1457 }
1458 }
1459
1460 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1461 u64 logical, u64 *length,
1462 struct btrfs_multi_bio **multi_ret, int mirror_num,
1463 u64 **raid_map_ret)
1464 {
1465 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1466 multi_ret, mirror_num, raid_map_ret);
1467 }
1468
1469 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1470 u64 logical, u64 *length, u64 *type,
1471 struct btrfs_multi_bio **multi_ret, int mirror_num,
1472 u64 **raid_map_ret)
1473 {
1474 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1475 struct cache_extent *ce;
1476 struct map_lookup *map;
1477 u64 offset;
1478 u64 stripe_offset;
1479 u64 stripe_nr;
1480 u64 *raid_map = NULL;
1481 int stripes_allocated = 8;
1482 int stripes_required = 1;
1483 int stripe_index;
1484 int i;
1485 struct btrfs_multi_bio *multi = NULL;
1486
1487 if (multi_ret && rw == READ) {
1488 stripes_allocated = 1;
1489 }
1490 again:
1491 ce = search_cache_extent(&map_tree->cache_tree, logical);
1492 if (!ce) {
1493 kfree(multi);
1494 *length = (u64)-1;
1495 return -ENOENT;
1496 }
1497 if (ce->start > logical) {
1498 kfree(multi);
1499 *length = ce->start - logical;
1500 return -ENOENT;
1501 }
1502
1503 if (multi_ret) {
1504 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1505 GFP_NOFS);
1506 if (!multi)
1507 return -ENOMEM;
1508 }
1509 map = container_of(ce, struct map_lookup, ce);
1510 offset = logical - ce->start;
1511
1512 if (rw == WRITE) {
1513 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1514 BTRFS_BLOCK_GROUP_DUP)) {
1515 stripes_required = map->num_stripes;
1516 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1517 stripes_required = map->sub_stripes;
1518 }
1519 }
1520 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1521 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1522 /* RAID[56] write or recovery. Return all stripes */
1523 stripes_required = map->num_stripes;
1524
1525 /* Only allocate the map if we've already got a large enough multi_ret */
1526 if (stripes_allocated >= stripes_required) {
1527 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1528 if (!raid_map) {
1529 kfree(multi);
1530 return -ENOMEM;
1531 }
1532 }
1533 }
1534
1535 /* if our multi bio struct is too small, back off and try again */
1536 if (multi_ret && stripes_allocated < stripes_required) {
1537 stripes_allocated = stripes_required;
1538 kfree(multi);
1539 multi = NULL;
1540 goto again;
1541 }
1542 stripe_nr = offset;
1543 /*
1544 * stripe_nr counts the total number of stripes we have to stride
1545 * to get to this block
1546 */
1547 stripe_nr = stripe_nr / map->stripe_len;
1548
1549 stripe_offset = stripe_nr * map->stripe_len;
1550 BUG_ON(offset < stripe_offset);
1551
1552 /* stripe_offset is the offset of this block in its stripe*/
1553 stripe_offset = offset - stripe_offset;
1554
1555 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1556 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1557 BTRFS_BLOCK_GROUP_RAID10 |
1558 BTRFS_BLOCK_GROUP_DUP)) {
1559 /* we limit the length of each bio to what fits in a stripe */
1560 *length = min_t(u64, ce->size - offset,
1561 map->stripe_len - stripe_offset);
1562 } else {
1563 *length = ce->size - offset;
1564 }
1565
1566 if (!multi_ret)
1567 goto out;
1568
1569 multi->num_stripes = 1;
1570 stripe_index = 0;
1571 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1572 if (rw == WRITE)
1573 multi->num_stripes = map->num_stripes;
1574 else if (mirror_num)
1575 stripe_index = mirror_num - 1;
1576 else
1577 stripe_index = stripe_nr % map->num_stripes;
1578 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1579 int factor = map->num_stripes / map->sub_stripes;
1580
1581 stripe_index = stripe_nr % factor;
1582 stripe_index *= map->sub_stripes;
1583
1584 if (rw == WRITE)
1585 multi->num_stripes = map->sub_stripes;
1586 else if (mirror_num)
1587 stripe_index += mirror_num - 1;
1588
1589 stripe_nr = stripe_nr / factor;
1590 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1591 if (rw == WRITE)
1592 multi->num_stripes = map->num_stripes;
1593 else if (mirror_num)
1594 stripe_index = mirror_num - 1;
1595 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1596 BTRFS_BLOCK_GROUP_RAID6)) {
1597
1598 if (raid_map) {
1599 int rot;
1600 u64 tmp;
1601 u64 raid56_full_stripe_start;
1602 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1603
1604 /*
1605 * align the start of our data stripe in the logical
1606 * address space
1607 */
1608 raid56_full_stripe_start = offset / full_stripe_len;
1609 raid56_full_stripe_start *= full_stripe_len;
1610
1611 /* get the data stripe number */
1612 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1613 stripe_nr = stripe_nr / nr_data_stripes(map);
1614
1615 /* Work out the disk rotation on this stripe-set */
1616 rot = stripe_nr % map->num_stripes;
1617
1618 /* Fill in the logical address of each stripe */
1619 tmp = stripe_nr * nr_data_stripes(map);
1620
1621 for (i = 0; i < nr_data_stripes(map); i++)
1622 raid_map[(i+rot) % map->num_stripes] =
1623 ce->start + (tmp + i) * map->stripe_len;
1624
1625 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1626 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1627 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1628
1629 *length = map->stripe_len;
1630 stripe_index = 0;
1631 stripe_offset = 0;
1632 multi->num_stripes = map->num_stripes;
1633 } else {
1634 stripe_index = stripe_nr % nr_data_stripes(map);
1635 stripe_nr = stripe_nr / nr_data_stripes(map);
1636
1637 /*
1638 * Mirror #0 or #1 means the original data block.
1639 * Mirror #2 is RAID5 parity block.
1640 * Mirror #3 is RAID6 Q block.
1641 */
1642 if (mirror_num > 1)
1643 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1644
1645 /* We distribute the parity blocks across stripes */
1646 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1647 }
1648 } else {
1649 /*
1650 * after this do_div call, stripe_nr is the number of stripes
1651 * on this device we have to walk to find the data, and
1652 * stripe_index is the number of our device in the stripe array
1653 */
1654 stripe_index = stripe_nr % map->num_stripes;
1655 stripe_nr = stripe_nr / map->num_stripes;
1656 }
1657 BUG_ON(stripe_index >= map->num_stripes);
1658
1659 for (i = 0; i < multi->num_stripes; i++) {
1660 multi->stripes[i].physical =
1661 map->stripes[stripe_index].physical + stripe_offset +
1662 stripe_nr * map->stripe_len;
1663 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1664 stripe_index++;
1665 }
1666 *multi_ret = multi;
1667
1668 if (type)
1669 *type = map->type;
1670
1671 if (raid_map) {
1672 sort_parity_stripes(multi, raid_map);
1673 *raid_map_ret = raid_map;
1674 }
1675 out:
1676 return 0;
1677 }
1678
1679 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1680 u8 *uuid, u8 *fsid)
1681 {
1682 struct btrfs_device *device;
1683 struct btrfs_fs_devices *cur_devices;
1684
1685 cur_devices = fs_info->fs_devices;
1686 while (cur_devices) {
1687 if (!fsid ||
1688 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1689 fs_info->ignore_fsid_mismatch)) {
1690 device = find_device(cur_devices, devid, uuid);
1691 if (device)
1692 return device;
1693 }
1694 cur_devices = cur_devices->seed;
1695 }
1696 return NULL;
1697 }
1698
1699 struct btrfs_device *
1700 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1701 u64 devid, int instance)
1702 {
1703 struct list_head *head = &fs_devices->devices;
1704 struct btrfs_device *dev;
1705 int num_found = 0;
1706
1707 list_for_each_entry(dev, head, dev_list) {
1708 if (dev->devid == devid && num_found++ == instance)
1709 return dev;
1710 }
1711 return NULL;
1712 }
1713
1714 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1715 {
1716 struct cache_extent *ce;
1717 struct map_lookup *map;
1718 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1719 int readonly = 0;
1720 int i;
1721
1722 /*
1723 * During chunk recovering, we may fail to find block group's
1724 * corresponding chunk, we will rebuild it later
1725 */
1726 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1727 if (!fs_info->is_chunk_recover)
1728 BUG_ON(!ce);
1729 else
1730 return 0;
1731
1732 map = container_of(ce, struct map_lookup, ce);
1733 for (i = 0; i < map->num_stripes; i++) {
1734 if (!map->stripes[i].dev->writeable) {
1735 readonly = 1;
1736 break;
1737 }
1738 }
1739
1740 return readonly;
1741 }
1742
1743 static struct btrfs_device *fill_missing_device(u64 devid)
1744 {
1745 struct btrfs_device *device;
1746
1747 device = kzalloc(sizeof(*device), GFP_NOFS);
1748 device->devid = devid;
1749 device->fd = -1;
1750 return device;
1751 }
1752
1753 /*
1754 * slot == -1: SYSTEM chunk
1755 * return -EIO on error, otherwise return 0
1756 */
1757 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1758 struct extent_buffer *leaf,
1759 struct btrfs_chunk *chunk,
1760 int slot, u64 logical)
1761 {
1762 u64 length;
1763 u64 stripe_len;
1764 u16 num_stripes;
1765 u16 sub_stripes;
1766 u64 type;
1767 u32 chunk_ondisk_size;
1768 u32 sectorsize = fs_info->sectorsize;
1769
1770 length = btrfs_chunk_length(leaf, chunk);
1771 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1772 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1773 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1774 type = btrfs_chunk_type(leaf, chunk);
1775
1776 /*
1777 * These valid checks may be insufficient to cover every corner cases.
1778 */
1779 if (!IS_ALIGNED(logical, sectorsize)) {
1780 error("invalid chunk logical %llu", logical);
1781 return -EIO;
1782 }
1783 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1784 error("invalid chunk sectorsize %llu",
1785 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1786 return -EIO;
1787 }
1788 if (!length || !IS_ALIGNED(length, sectorsize)) {
1789 error("invalid chunk length %llu", length);
1790 return -EIO;
1791 }
1792 if (stripe_len != BTRFS_STRIPE_LEN) {
1793 error("invalid chunk stripe length: %llu", stripe_len);
1794 return -EIO;
1795 }
1796 /* Check on chunk item type */
1797 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1798 error("invalid chunk type %llu", type);
1799 return -EIO;
1800 }
1801 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1802 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1803 error("unrecognized chunk type: %llu",
1804 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1805 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1806 return -EIO;
1807 }
1808 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1809 error("missing chunk type flag: %llu", type);
1810 return -EIO;
1811 }
1812 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1813 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1814 error("conflicting chunk type detected: %llu", type);
1815 return -EIO;
1816 }
1817 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1818 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1819 error("conflicting chunk profile detected: %llu", type);
1820 return -EIO;
1821 }
1822
1823 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1824 /*
1825 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1826 * it can't exceed the system chunk array size
1827 * For normal chunk, it should match its chunk item size.
1828 */
1829 if (num_stripes < 1 ||
1830 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1831 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1832 error("invalid num_stripes: %u", num_stripes);
1833 return -EIO;
1834 }
1835 /*
1836 * Device number check against profile
1837 */
1838 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1839 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1840 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1841 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1842 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1843 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1844 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1845 num_stripes != 1)) {
1846 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1847 num_stripes, sub_stripes,
1848 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1849 return -EIO;
1850 }
1851
1852 return 0;
1853 }
1854
1855 /*
1856 * Slot is used to verify the chunk item is valid
1857 *
1858 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1859 */
1860 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1861 struct extent_buffer *leaf,
1862 struct btrfs_chunk *chunk, int slot)
1863 {
1864 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1865 struct map_lookup *map;
1866 struct cache_extent *ce;
1867 u64 logical;
1868 u64 length;
1869 u64 devid;
1870 u8 uuid[BTRFS_UUID_SIZE];
1871 int num_stripes;
1872 int ret;
1873 int i;
1874
1875 logical = key->offset;
1876 length = btrfs_chunk_length(leaf, chunk);
1877 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1878 /* Validation check */
1879 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1880 if (ret) {
1881 error("%s checksums match, but it has an invalid chunk, %s",
1882 (slot == -1) ? "Superblock" : "Metadata",
1883 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1884 return ret;
1885 }
1886
1887 ce = search_cache_extent(&map_tree->cache_tree, logical);
1888
1889 /* already mapped? */
1890 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1891 return 0;
1892 }
1893
1894 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1895 if (!map)
1896 return -ENOMEM;
1897
1898 map->ce.start = logical;
1899 map->ce.size = length;
1900 map->num_stripes = num_stripes;
1901 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1902 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1903 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1904 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1905 map->type = btrfs_chunk_type(leaf, chunk);
1906 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1907
1908 for (i = 0; i < num_stripes; i++) {
1909 map->stripes[i].physical =
1910 btrfs_stripe_offset_nr(leaf, chunk, i);
1911 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1912 read_extent_buffer(leaf, uuid, (unsigned long)
1913 btrfs_stripe_dev_uuid_nr(chunk, i),
1914 BTRFS_UUID_SIZE);
1915 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1916 NULL);
1917 if (!map->stripes[i].dev) {
1918 map->stripes[i].dev = fill_missing_device(devid);
1919 printf("warning, device %llu is missing\n",
1920 (unsigned long long)devid);
1921 list_add(&map->stripes[i].dev->dev_list,
1922 &fs_info->fs_devices->devices);
1923 }
1924
1925 }
1926 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1927 if (ret < 0) {
1928 error("failed to add chunk map start=%llu len=%llu: %d (%s)",
1929 map->ce.start, map->ce.size, ret, strerror(-ret));
1930 }
1931
1932 return ret;
1933 }
1934
1935 static int fill_device_from_item(struct extent_buffer *leaf,
1936 struct btrfs_dev_item *dev_item,
1937 struct btrfs_device *device)
1938 {
1939 unsigned long ptr;
1940
1941 device->devid = btrfs_device_id(leaf, dev_item);
1942 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1943 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1944 device->type = btrfs_device_type(leaf, dev_item);
1945 device->io_align = btrfs_device_io_align(leaf, dev_item);
1946 device->io_width = btrfs_device_io_width(leaf, dev_item);
1947 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1948
1949 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1950 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1951
1952 return 0;
1953 }
1954
1955 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1956 {
1957 struct btrfs_fs_devices *fs_devices;
1958 int ret;
1959
1960 fs_devices = fs_info->fs_devices->seed;
1961 while (fs_devices) {
1962 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1963 ret = 0;
1964 goto out;
1965 }
1966 fs_devices = fs_devices->seed;
1967 }
1968
1969 fs_devices = find_fsid(fsid);
1970 if (!fs_devices) {
1971 /* missing all seed devices */
1972 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1973 if (!fs_devices) {
1974 ret = -ENOMEM;
1975 goto out;
1976 }
1977 INIT_LIST_HEAD(&fs_devices->devices);
1978 list_add(&fs_devices->list, &fs_uuids);
1979 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1980 }
1981
1982 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1983 if (ret)
1984 goto out;
1985
1986 fs_devices->seed = fs_info->fs_devices->seed;
1987 fs_info->fs_devices->seed = fs_devices;
1988 out:
1989 return ret;
1990 }
1991
1992 static int read_one_dev(struct btrfs_fs_info *fs_info,
1993 struct extent_buffer *leaf,
1994 struct btrfs_dev_item *dev_item)
1995 {
1996 struct btrfs_device *device;
1997 u64 devid;
1998 int ret = 0;
1999 u8 fs_uuid[BTRFS_UUID_SIZE];
2000 u8 dev_uuid[BTRFS_UUID_SIZE];
2001
2002 devid = btrfs_device_id(leaf, dev_item);
2003 read_extent_buffer(leaf, dev_uuid,
2004 (unsigned long)btrfs_device_uuid(dev_item),
2005 BTRFS_UUID_SIZE);
2006 read_extent_buffer(leaf, fs_uuid,
2007 (unsigned long)btrfs_device_fsid(dev_item),
2008 BTRFS_UUID_SIZE);
2009
2010 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
2011 ret = open_seed_devices(fs_info, fs_uuid);
2012 if (ret)
2013 return ret;
2014 }
2015
2016 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2017 if (!device) {
2018 device = kzalloc(sizeof(*device), GFP_NOFS);
2019 if (!device)
2020 return -ENOMEM;
2021 device->fd = -1;
2022 list_add(&device->dev_list,
2023 &fs_info->fs_devices->devices);
2024 }
2025
2026 fill_device_from_item(leaf, dev_item, device);
2027 device->dev_root = fs_info->dev_root;
2028 return ret;
2029 }
2030
2031 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
2032 {
2033 struct btrfs_super_block *super_copy = fs_info->super_copy;
2034 struct extent_buffer *sb;
2035 struct btrfs_disk_key *disk_key;
2036 struct btrfs_chunk *chunk;
2037 u8 *array_ptr;
2038 unsigned long sb_array_offset;
2039 int ret = 0;
2040 u32 num_stripes;
2041 u32 array_size;
2042 u32 len = 0;
2043 u32 cur_offset;
2044 struct btrfs_key key;
2045
2046 if (fs_info->nodesize < BTRFS_SUPER_INFO_SIZE) {
2047 printf("ERROR: nodesize %u too small to read superblock\n",
2048 fs_info->nodesize);
2049 return -EINVAL;
2050 }
2051 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
2052 if (!sb)
2053 return -ENOMEM;
2054 btrfs_set_buffer_uptodate(sb);
2055 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
2056 array_size = btrfs_super_sys_array_size(super_copy);
2057
2058 array_ptr = super_copy->sys_chunk_array;
2059 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
2060 cur_offset = 0;
2061
2062 while (cur_offset < array_size) {
2063 disk_key = (struct btrfs_disk_key *)array_ptr;
2064 len = sizeof(*disk_key);
2065 if (cur_offset + len > array_size)
2066 goto out_short_read;
2067
2068 btrfs_disk_key_to_cpu(&key, disk_key);
2069
2070 array_ptr += len;
2071 sb_array_offset += len;
2072 cur_offset += len;
2073
2074 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2075 chunk = (struct btrfs_chunk *)sb_array_offset;
2076 /*
2077 * At least one btrfs_chunk with one stripe must be
2078 * present, exact stripe count check comes afterwards
2079 */
2080 len = btrfs_chunk_item_size(1);
2081 if (cur_offset + len > array_size)
2082 goto out_short_read;
2083
2084 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2085 if (!num_stripes) {
2086 printk(
2087 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2088 num_stripes, cur_offset);
2089 ret = -EIO;
2090 break;
2091 }
2092
2093 len = btrfs_chunk_item_size(num_stripes);
2094 if (cur_offset + len > array_size)
2095 goto out_short_read;
2096
2097 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2098 if (ret)
2099 break;
2100 } else {
2101 printk(
2102 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2103 (u32)key.type, cur_offset);
2104 ret = -EIO;
2105 break;
2106 }
2107 array_ptr += len;
2108 sb_array_offset += len;
2109 cur_offset += len;
2110 }
2111 free_extent_buffer(sb);
2112 return ret;
2113
2114 out_short_read:
2115 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2116 len, cur_offset);
2117 free_extent_buffer(sb);
2118 return -EIO;
2119 }
2120
2121 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2122 {
2123 struct btrfs_path *path;
2124 struct extent_buffer *leaf;
2125 struct btrfs_key key;
2126 struct btrfs_key found_key;
2127 struct btrfs_root *root = fs_info->chunk_root;
2128 int ret;
2129 int slot;
2130
2131 path = btrfs_alloc_path();
2132 if (!path)
2133 return -ENOMEM;
2134
2135 /*
2136 * Read all device items, and then all the chunk items. All
2137 * device items are found before any chunk item (their object id
2138 * is smaller than the lowest possible object id for a chunk
2139 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2140 */
2141 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2142 key.offset = 0;
2143 key.type = 0;
2144 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2145 if (ret < 0)
2146 goto error;
2147 while(1) {
2148 leaf = path->nodes[0];
2149 slot = path->slots[0];
2150 if (slot >= btrfs_header_nritems(leaf)) {
2151 ret = btrfs_next_leaf(root, path);
2152 if (ret == 0)
2153 continue;
2154 if (ret < 0)
2155 goto error;
2156 break;
2157 }
2158 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2159 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2160 struct btrfs_dev_item *dev_item;
2161 dev_item = btrfs_item_ptr(leaf, slot,
2162 struct btrfs_dev_item);
2163 ret = read_one_dev(fs_info, leaf, dev_item);
2164 if (ret < 0)
2165 goto error;
2166 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2167 struct btrfs_chunk *chunk;
2168 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2169 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2170 slot);
2171 if (ret < 0)
2172 goto error;
2173 }
2174 path->slots[0]++;
2175 }
2176
2177 ret = 0;
2178 error:
2179 btrfs_free_path(path);
2180 return ret;
2181 }
2182
2183 struct list_head *btrfs_scanned_uuids(void)
2184 {
2185 return &fs_uuids;
2186 }
2187
2188 static int rmw_eb(struct btrfs_fs_info *info,
2189 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2190 {
2191 int ret;
2192 unsigned long orig_off = 0;
2193 unsigned long dest_off = 0;
2194 unsigned long copy_len = eb->len;
2195
2196 ret = read_whole_eb(info, eb, 0);
2197 if (ret)
2198 return ret;
2199
2200 if (eb->start + eb->len <= orig_eb->start ||
2201 eb->start >= orig_eb->start + orig_eb->len)
2202 return 0;
2203 /*
2204 * | ----- orig_eb ------- |
2205 * | ----- stripe ------- |
2206 * | ----- orig_eb ------- |
2207 * | ----- orig_eb ------- |
2208 */
2209 if (eb->start > orig_eb->start)
2210 orig_off = eb->start - orig_eb->start;
2211 if (orig_eb->start > eb->start)
2212 dest_off = orig_eb->start - eb->start;
2213
2214 if (copy_len > orig_eb->len - orig_off)
2215 copy_len = orig_eb->len - orig_off;
2216 if (copy_len > eb->len - dest_off)
2217 copy_len = eb->len - dest_off;
2218
2219 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2220 return 0;
2221 }
2222
2223 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2224 struct extent_buffer *orig_eb,
2225 struct extent_buffer **ebs,
2226 u64 stripe_len, u64 *raid_map,
2227 int num_stripes)
2228 {
2229 struct extent_buffer **tmp_ebs;
2230 u64 start = orig_eb->start;
2231 u64 this_eb_start;
2232 int i;
2233 int ret = 0;
2234
2235 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2236 if (!tmp_ebs)
2237 return -ENOMEM;
2238
2239 /* Alloc memory in a row for data stripes */
2240 for (i = 0; i < num_stripes; i++) {
2241 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2242 break;
2243
2244 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2245 if (!tmp_ebs[i]) {
2246 ret = -ENOMEM;
2247 goto clean_up;
2248 }
2249 }
2250
2251 for (i = 0; i < num_stripes; i++) {
2252 struct extent_buffer *eb = tmp_ebs[i];
2253
2254 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2255 break;
2256
2257 eb->start = raid_map[i];
2258 eb->len = stripe_len;
2259 eb->refs = 1;
2260 eb->flags = 0;
2261 eb->fd = -1;
2262 eb->dev_bytenr = (u64)-1;
2263
2264 this_eb_start = raid_map[i];
2265
2266 if (start > this_eb_start ||
2267 start + orig_eb->len < this_eb_start + stripe_len) {
2268 ret = rmw_eb(info, eb, orig_eb);
2269 if (ret)
2270 goto clean_up;
2271 } else {
2272 memcpy(eb->data, orig_eb->data + eb->start - start,
2273 stripe_len);
2274 }
2275 ebs[i] = eb;
2276 }
2277 free(tmp_ebs);
2278 return ret;
2279 clean_up:
2280 for (i = 0; i < num_stripes; i++)
2281 free(tmp_ebs[i]);
2282 free(tmp_ebs);
2283 return ret;
2284 }
2285
2286 int write_raid56_with_parity(struct btrfs_fs_info *info,
2287 struct extent_buffer *eb,
2288 struct btrfs_multi_bio *multi,
2289 u64 stripe_len, u64 *raid_map)
2290 {
2291 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2292 int i;
2293 int ret;
2294 int alloc_size = eb->len;
2295 void **pointers;
2296
2297 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2298 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2299 if (!ebs || !pointers) {
2300 free(ebs);
2301 free(pointers);
2302 return -ENOMEM;
2303 }
2304
2305 if (stripe_len > alloc_size)
2306 alloc_size = stripe_len;
2307
2308 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2309 multi->num_stripes);
2310 if (ret)
2311 goto out;
2312
2313 for (i = 0; i < multi->num_stripes; i++) {
2314 struct extent_buffer *new_eb;
2315 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2316 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2317 ebs[i]->fd = multi->stripes[i].dev->fd;
2318 multi->stripes[i].dev->total_ios++;
2319 if (ebs[i]->start != raid_map[i]) {
2320 ret = -EINVAL;
2321 goto out_free_split;
2322 }
2323 continue;
2324 }
2325 new_eb = malloc(sizeof(*eb) + alloc_size);
2326 if (!new_eb) {
2327 ret = -ENOMEM;
2328 goto out_free_split;
2329 }
2330 new_eb->dev_bytenr = multi->stripes[i].physical;
2331 new_eb->fd = multi->stripes[i].dev->fd;
2332 multi->stripes[i].dev->total_ios++;
2333 new_eb->len = stripe_len;
2334
2335 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2336 p_eb = new_eb;
2337 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2338 q_eb = new_eb;
2339 }
2340 if (q_eb) {
2341 ebs[multi->num_stripes - 2] = p_eb;
2342 ebs[multi->num_stripes - 1] = q_eb;
2343
2344 for (i = 0; i < multi->num_stripes; i++)
2345 pointers[i] = ebs[i]->data;
2346
2347 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2348 } else {
2349 ebs[multi->num_stripes - 1] = p_eb;
2350 for (i = 0; i < multi->num_stripes; i++)
2351 pointers[i] = ebs[i]->data;
2352 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2353 multi->num_stripes - 1, pointers);
2354 if (ret < 0)
2355 goto out_free_split;
2356 }
2357
2358 for (i = 0; i < multi->num_stripes; i++) {
2359 ret = write_extent_to_disk(ebs[i]);
2360 if (ret < 0)
2361 goto out_free_split;
2362 }
2363
2364 out_free_split:
2365 for (i = 0; i < multi->num_stripes; i++) {
2366 if (ebs[i] != eb)
2367 free(ebs[i]);
2368 }
2369 out:
2370 free(ebs);
2371 free(pointers);
2372
2373 return ret;
2374 }
2375
2376 /*
2377 * Get stripe length from chunk item and its stripe items
2378 *
2379 * Caller should only call this function after validating the chunk item
2380 * by using btrfs_check_chunk_valid().
2381 */
2382 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2383 struct extent_buffer *leaf,
2384 struct btrfs_chunk *chunk)
2385 {
2386 u64 stripe_len;
2387 u64 chunk_len;
2388 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2389 u64 profile = btrfs_chunk_type(leaf, chunk) &
2390 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2391
2392 chunk_len = btrfs_chunk_length(leaf, chunk);
2393
2394 switch (profile) {
2395 case 0: /* Single profile */
2396 case BTRFS_BLOCK_GROUP_RAID1:
2397 case BTRFS_BLOCK_GROUP_DUP:
2398 stripe_len = chunk_len;
2399 break;
2400 case BTRFS_BLOCK_GROUP_RAID0:
2401 stripe_len = chunk_len / num_stripes;
2402 break;
2403 case BTRFS_BLOCK_GROUP_RAID5:
2404 stripe_len = chunk_len / (num_stripes - 1);
2405 break;
2406 case BTRFS_BLOCK_GROUP_RAID6:
2407 stripe_len = chunk_len / (num_stripes - 2);
2408 break;
2409 case BTRFS_BLOCK_GROUP_RAID10:
2410 stripe_len = chunk_len / (num_stripes /
2411 btrfs_chunk_sub_stripes(leaf, chunk));
2412 break;
2413 default:
2414 /* Invalid chunk profile found */
2415 BUG_ON(1);
2416 }
2417 return stripe_len;
2418 }
2419
2420 /*
2421 * Return 0 if size of @device is already good
2422 * Return >0 if size of @device is not aligned but fixed without problems
2423 * Return <0 if something wrong happened when aligning the size of @device
2424 */
2425 int btrfs_fix_device_size(struct btrfs_fs_info *fs_info,
2426 struct btrfs_device *device)
2427 {
2428 struct btrfs_trans_handle *trans;
2429 struct btrfs_key key;
2430 struct btrfs_path path;
2431 struct btrfs_root *chunk_root = fs_info->chunk_root;
2432 struct btrfs_dev_item *di;
2433 u64 old_bytes = device->total_bytes;
2434 int ret;
2435
2436 if (IS_ALIGNED(old_bytes, fs_info->sectorsize))
2437 return 0;
2438
2439 /* Align the in-memory total_bytes first, and use it as correct size */
2440 device->total_bytes = round_down(device->total_bytes,
2441 fs_info->sectorsize);
2442
2443 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2444 key.type = BTRFS_DEV_ITEM_KEY;
2445 key.offset = device->devid;
2446
2447 trans = btrfs_start_transaction(chunk_root, 1);
2448 if (IS_ERR(trans)) {
2449 ret = PTR_ERR(trans);
2450 error("error starting transaction: %d (%s)",
2451 ret, strerror(-ret));
2452 return ret;
2453 }
2454
2455 btrfs_init_path(&path);
2456 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2457 if (ret > 0) {
2458 error("failed to find DEV_ITEM for devid %llu", device->devid);
2459 ret = -ENOENT;
2460 goto err;
2461 }
2462 if (ret < 0) {
2463 error("failed to search chunk root: %d (%s)",
2464 ret, strerror(-ret));
2465 goto err;
2466 }
2467 di = btrfs_item_ptr(path.nodes[0], path.slots[0], struct btrfs_dev_item);
2468 btrfs_set_device_total_bytes(path.nodes[0], di, device->total_bytes);
2469 btrfs_mark_buffer_dirty(path.nodes[0]);
2470 ret = btrfs_commit_transaction(trans, chunk_root);
2471 if (ret < 0) {
2472 error("failed to commit current transaction: %d (%s)",
2473 ret, strerror(-ret));
2474 btrfs_release_path(&path);
2475 return ret;
2476 }
2477 btrfs_release_path(&path);
2478 printf("Fixed device size for devid %llu, old size: %llu new size: %llu\n",
2479 device->devid, old_bytes, device->total_bytes);
2480 return 1;
2481
2482 err:
2483 /* We haven't modified anything, it's OK to commit current trans */
2484 btrfs_commit_transaction(trans, chunk_root);
2485 btrfs_release_path(&path);
2486 return ret;
2487 }
2488
2489 /*
2490 * Return 0 if super block total_bytes matches all devices' total_bytes
2491 * Return >0 if super block total_bytes mismatch but fixed without problem
2492 * Return <0 if we failed to fix super block total_bytes
2493 */
2494 int btrfs_fix_super_size(struct btrfs_fs_info *fs_info)
2495 {
2496 struct btrfs_trans_handle *trans;
2497 struct btrfs_device *device;
2498 struct list_head *dev_list = &fs_info->fs_devices->devices;
2499 u64 total_bytes = 0;
2500 u64 old_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2501 int ret;
2502
2503 list_for_each_entry(device, dev_list, dev_list) {
2504 /*
2505 * Caller should ensure this function is called after aligning
2506 * all devices' total_bytes.
2507 */
2508 if (!IS_ALIGNED(device->total_bytes, fs_info->sectorsize)) {
2509 error("device %llu total_bytes %llu not aligned to %u",
2510 device->devid, device->total_bytes,
2511 fs_info->sectorsize);
2512 return -EUCLEAN;
2513 }
2514 total_bytes += device->total_bytes;
2515 }
2516
2517 if (total_bytes == old_bytes)
2518 return 0;
2519
2520 btrfs_set_super_total_bytes(fs_info->super_copy, total_bytes);
2521
2522 /* Commit transaction to update all super blocks */
2523 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2524 if (IS_ERR(trans)) {
2525 ret = PTR_ERR(trans);
2526 error("error starting transaction: %d (%s)",
2527 ret, strerror(-ret));
2528 return ret;
2529 }
2530 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2531 if (ret < 0) {
2532 error("failed to commit current transaction: %d (%s)",
2533 ret, strerror(-ret));
2534 return ret;
2535 }
2536 printf("Fixed super total bytes, old size: %llu new size: %llu\n",
2537 old_bytes, total_bytes);
2538 return 1;
2539 }
2540
2541 /*
2542 * Return 0 if all devices and super block sizes are good
2543 * Return >0 if any device/super size problem was found, but fixed
2544 * Return <0 if something wrong happened during fixing
2545 */
2546 int btrfs_fix_device_and_super_size(struct btrfs_fs_info *fs_info)
2547 {
2548 struct btrfs_device *device;
2549 struct list_head *dev_list = &fs_info->fs_devices->devices;
2550 bool have_bad_value = false;
2551 int ret;
2552
2553 /* Seed device is not supported yet */
2554 if (fs_info->fs_devices->seed) {
2555 error("fixing device size with seed device is not supported yet");
2556 return -EOPNOTSUPP;
2557 }
2558
2559 /* All devices must be set up before repairing */
2560 if (list_empty(dev_list)) {
2561 error("no device found");
2562 return -ENODEV;
2563 }
2564 list_for_each_entry(device, dev_list, dev_list) {
2565 if (device->fd == -1 || !device->writeable) {
2566 error("devid %llu is missing or not writeable",
2567 device->devid);
2568 error(
2569 "fixing device size needs all device(s) to be present and writeable");
2570 return -ENODEV;
2571 }
2572 }
2573
2574 /* Repair total_bytes of each device */
2575 list_for_each_entry(device, dev_list, dev_list) {
2576 ret = btrfs_fix_device_size(fs_info, device);
2577 if (ret < 0)
2578 return ret;
2579 if (ret > 0)
2580 have_bad_value = true;
2581 }
2582
2583 /* Repair super total_byte */
2584 ret = btrfs_fix_super_size(fs_info);
2585 if (ret > 0)
2586 have_bad_value = true;
2587 if (have_bad_value) {
2588 printf(
2589 "Fixed unaligned/mismatched total_bytes for super block and device items\n");
2590 ret = 1;
2591 } else {
2592 printf("No device size related problem found\n");
2593 ret = 0;
2594 }
2595 return ret;
2596 }
(deprecated) Btrfs progs developments and patch integration