?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: image: move metadump definitions to own header
[btrfs-progs-unstable/devel.git] / volumes.c
blob2fde4762c93c5d2be6ae4e82bab1fe57d067b06b
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 struct stripe {
34 struct btrfs_device *dev;
35 u64 physical;
36 };
37
38 static inline int nr_parity_stripes(struct map_lookup *map)
39 {
40 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
41 return 1;
42 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
43 return 2;
44 else
45 return 0;
46 }
47
48 static inline int nr_data_stripes(struct map_lookup *map)
49 {
50 return map->num_stripes - nr_parity_stripes(map);
51 }
52
53 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
54
55 static LIST_HEAD(fs_uuids);
56
57 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
58 u8 *uuid)
59 {
60 struct btrfs_device *dev;
61 struct list_head *cur;
62
63 list_for_each(cur, head) {
64 dev = list_entry(cur, struct btrfs_device, dev_list);
65 if (dev->devid == devid &&
66 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
67 return dev;
68 }
69 }
70 return NULL;
71 }
72
73 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
74 {
75 struct list_head *cur;
76 struct btrfs_fs_devices *fs_devices;
77
78 list_for_each(cur, &fs_uuids) {
79 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
80 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
81 return fs_devices;
82 }
83 return NULL;
84 }
85
86 static int device_list_add(const char *path,
87 struct btrfs_super_block *disk_super,
88 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
89 {
90 struct btrfs_device *device;
91 struct btrfs_fs_devices *fs_devices;
92 u64 found_transid = btrfs_super_generation(disk_super);
93
94 fs_devices = find_fsid(disk_super->fsid);
95 if (!fs_devices) {
96 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
97 if (!fs_devices)
98 return -ENOMEM;
99 INIT_LIST_HEAD(&fs_devices->devices);
100 list_add(&fs_devices->list, &fs_uuids);
101 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
102 fs_devices->latest_devid = devid;
103 fs_devices->latest_trans = found_transid;
104 fs_devices->lowest_devid = (u64)-1;
105 device = NULL;
106 } else {
107 device = __find_device(&fs_devices->devices, devid,
108 disk_super->dev_item.uuid);
109 }
110 if (!device) {
111 device = kzalloc(sizeof(*device), GFP_NOFS);
112 if (!device) {
113 /* we can safely leave the fs_devices entry around */
114 return -ENOMEM;
115 }
116 device->fd = -1;
117 device->devid = devid;
118 device->generation = found_transid;
119 memcpy(device->uuid, disk_super->dev_item.uuid,
120 BTRFS_UUID_SIZE);
121 device->name = kstrdup(path, GFP_NOFS);
122 if (!device->name) {
123 kfree(device);
124 return -ENOMEM;
125 }
126 device->label = kstrdup(disk_super->label, GFP_NOFS);
127 if (!device->label) {
128 kfree(device->name);
129 kfree(device);
130 return -ENOMEM;
131 }
132 device->total_devs = btrfs_super_num_devices(disk_super);
133 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
134 device->total_bytes =
135 btrfs_stack_device_total_bytes(&disk_super->dev_item);
136 device->bytes_used =
137 btrfs_stack_device_bytes_used(&disk_super->dev_item);
138 list_add(&device->dev_list, &fs_devices->devices);
139 device->fs_devices = fs_devices;
140 } else if (!device->name || strcmp(device->name, path)) {
141 char *name;
142
143 /*
144 * The existing device has newer generation, so this one could
145 * be a stale one, don't add it.
146 */
147 if (found_transid < device->generation) {
148 warning(
149 "adding device %s gen %llu but found an existing device %s gen %llu",
150 path, found_transid, device->name,
151 device->generation);
152 return -EEXIST;
153 }
154
155 name = strdup(path);
156 if (!name)
157 return -ENOMEM;
158 kfree(device->name);
159 device->name = name;
160 }
161
162
163 if (found_transid > fs_devices->latest_trans) {
164 fs_devices->latest_devid = devid;
165 fs_devices->latest_trans = found_transid;
166 }
167 if (fs_devices->lowest_devid > devid) {
168 fs_devices->lowest_devid = devid;
169 }
170 *fs_devices_ret = fs_devices;
171 return 0;
172 }
173
174 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
175 {
176 struct btrfs_fs_devices *seed_devices;
177 struct btrfs_device *device;
178 int ret = 0;
179
180 again:
181 if (!fs_devices)
182 return 0;
183 while (!list_empty(&fs_devices->devices)) {
184 device = list_entry(fs_devices->devices.next,
185 struct btrfs_device, dev_list);
186 if (device->fd != -1) {
187 if (fsync(device->fd) == -1) {
188 warning("fsync on device %llu failed: %s",
189 device->devid, strerror(errno));
190 ret = -errno;
191 }
192 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
193 fprintf(stderr, "Warning, could not drop caches\n");
194 close(device->fd);
195 device->fd = -1;
196 }
197 device->writeable = 0;
198 list_del(&device->dev_list);
199 /* free the memory */
200 free(device->name);
201 free(device->label);
202 free(device);
203 }
204
205 seed_devices = fs_devices->seed;
206 fs_devices->seed = NULL;
207 if (seed_devices) {
208 struct btrfs_fs_devices *orig;
209
210 orig = fs_devices;
211 fs_devices = seed_devices;
212 list_del(&orig->list);
213 free(orig);
214 goto again;
215 } else {
216 list_del(&fs_devices->list);
217 free(fs_devices);
218 }
219
220 return ret;
221 }
222
223 void btrfs_close_all_devices(void)
224 {
225 struct btrfs_fs_devices *fs_devices;
226
227 while (!list_empty(&fs_uuids)) {
228 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
229 list);
230 btrfs_close_devices(fs_devices);
231 }
232 }
233
234 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
235 {
236 int fd;
237 struct list_head *head = &fs_devices->devices;
238 struct list_head *cur;
239 struct btrfs_device *device;
240 int ret;
241
242 list_for_each(cur, head) {
243 device = list_entry(cur, struct btrfs_device, dev_list);
244 if (!device->name) {
245 printk("no name for device %llu, skip it now\n", device->devid);
246 continue;
247 }
248
249 fd = open(device->name, flags);
250 if (fd < 0) {
251 ret = -errno;
252 error("cannot open device '%s': %s", device->name,
253 strerror(errno));
254 goto fail;
255 }
256
257 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
258 fprintf(stderr, "Warning, could not drop caches\n");
259
260 if (device->devid == fs_devices->latest_devid)
261 fs_devices->latest_bdev = fd;
262 if (device->devid == fs_devices->lowest_devid)
263 fs_devices->lowest_bdev = fd;
264 device->fd = fd;
265 if (flags & O_RDWR)
266 device->writeable = 1;
267 }
268 return 0;
269 fail:
270 btrfs_close_devices(fs_devices);
271 return ret;
272 }
273
274 int btrfs_scan_one_device(int fd, const char *path,
275 struct btrfs_fs_devices **fs_devices_ret,
276 u64 *total_devs, u64 super_offset, unsigned sbflags)
277 {
278 struct btrfs_super_block *disk_super;
279 char buf[BTRFS_SUPER_INFO_SIZE];
280 int ret;
281 u64 devid;
282
283 disk_super = (struct btrfs_super_block *)buf;
284 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
285 if (ret < 0)
286 return -EIO;
287 devid = btrfs_stack_device_id(&disk_super->dev_item);
288 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
289 *total_devs = 1;
290 else
291 *total_devs = btrfs_super_num_devices(disk_super);
292
293 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
294
295 return ret;
296 }
297
298 /*
299 * find_free_dev_extent_start - find free space in the specified device
300 * @device: the device which we search the free space in
301 * @num_bytes: the size of the free space that we need
302 * @search_start: the position from which to begin the search
303 * @start: store the start of the free space.
304 * @len: the size of the free space. that we find, or the size
305 * of the max free space if we don't find suitable free space
306 *
307 * this uses a pretty simple search, the expectation is that it is
308 * called very infrequently and that a given device has a small number
309 * of extents
310 *
311 * @start is used to store the start of the free space if we find. But if we
312 * don't find suitable free space, it will be used to store the start position
313 * of the max free space.
314 *
315 * @len is used to store the size of the free space that we find.
316 * But if we don't find suitable free space, it is used to store the size of
317 * the max free space.
318 */
319 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
320 struct btrfs_device *device, u64 num_bytes,
321 u64 search_start, u64 *start, u64 *len)
322 {
323 struct btrfs_key key;
324 struct btrfs_root *root = device->dev_root;
325 struct btrfs_dev_extent *dev_extent;
326 struct btrfs_path *path;
327 u64 hole_size;
328 u64 max_hole_start;
329 u64 max_hole_size;
330 u64 extent_end;
331 u64 search_end = device->total_bytes;
332 int ret;
333 int slot;
334 struct extent_buffer *l;
335 u64 min_search_start;
336
337 /*
338 * We don't want to overwrite the superblock on the drive nor any area
339 * used by the boot loader (grub for example), so we make sure to start
340 * at an offset of at least 1MB.
341 */
342 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
343 search_start = max(search_start, min_search_start);
344
345 path = btrfs_alloc_path();
346 if (!path)
347 return -ENOMEM;
348
349 max_hole_start = search_start;
350 max_hole_size = 0;
351
352 if (search_start >= search_end) {
353 ret = -ENOSPC;
354 goto out;
355 }
356
357 path->reada = 2;
358
359 key.objectid = device->devid;
360 key.offset = search_start;
361 key.type = BTRFS_DEV_EXTENT_KEY;
362
363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
364 if (ret < 0)
365 goto out;
366 if (ret > 0) {
367 ret = btrfs_previous_item(root, path, key.objectid, key.type);
368 if (ret < 0)
369 goto out;
370 }
371
372 while (1) {
373 l = path->nodes[0];
374 slot = path->slots[0];
375 if (slot >= btrfs_header_nritems(l)) {
376 ret = btrfs_next_leaf(root, path);
377 if (ret == 0)
378 continue;
379 if (ret < 0)
380 goto out;
381
382 break;
383 }
384 btrfs_item_key_to_cpu(l, &key, slot);
385
386 if (key.objectid < device->devid)
387 goto next;
388
389 if (key.objectid > device->devid)
390 break;
391
392 if (key.type != BTRFS_DEV_EXTENT_KEY)
393 goto next;
394
395 if (key.offset > search_start) {
396 hole_size = key.offset - search_start;
397
398 /*
399 * Have to check before we set max_hole_start, otherwise
400 * we could end up sending back this offset anyway.
401 */
402 if (hole_size > max_hole_size) {
403 max_hole_start = search_start;
404 max_hole_size = hole_size;
405 }
406
407 /*
408 * If this free space is greater than which we need,
409 * it must be the max free space that we have found
410 * until now, so max_hole_start must point to the start
411 * of this free space and the length of this free space
412 * is stored in max_hole_size. Thus, we return
413 * max_hole_start and max_hole_size and go back to the
414 * caller.
415 */
416 if (hole_size >= num_bytes) {
417 ret = 0;
418 goto out;
419 }
420 }
421
422 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
423 extent_end = key.offset + btrfs_dev_extent_length(l,
424 dev_extent);
425 if (extent_end > search_start)
426 search_start = extent_end;
427 next:
428 path->slots[0]++;
429 cond_resched();
430 }
431
432 /*
433 * At this point, search_start should be the end of
434 * allocated dev extents, and when shrinking the device,
435 * search_end may be smaller than search_start.
436 */
437 if (search_end > search_start) {
438 hole_size = search_end - search_start;
439
440 if (hole_size > max_hole_size) {
441 max_hole_start = search_start;
442 max_hole_size = hole_size;
443 }
444 }
445
446 /* See above. */
447 if (max_hole_size < num_bytes)
448 ret = -ENOSPC;
449 else
450 ret = 0;
451
452 out:
453 btrfs_free_path(path);
454 *start = max_hole_start;
455 if (len)
456 *len = max_hole_size;
457 return ret;
458 }
459
460 int find_free_dev_extent(struct btrfs_trans_handle *trans,
461 struct btrfs_device *device, u64 num_bytes,
462 u64 *start)
463 {
464 /* FIXME use last free of some kind */
465 return find_free_dev_extent_start(trans, device,
466 num_bytes, 0, start, NULL);
467 }
468
469 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
470 struct btrfs_device *device,
471 u64 chunk_tree, u64 chunk_objectid,
472 u64 chunk_offset,
473 u64 num_bytes, u64 *start, int convert)
474 {
475 int ret;
476 struct btrfs_path *path;
477 struct btrfs_root *root = device->dev_root;
478 struct btrfs_dev_extent *extent;
479 struct extent_buffer *leaf;
480 struct btrfs_key key;
481
482 path = btrfs_alloc_path();
483 if (!path)
484 return -ENOMEM;
485
486 /*
487 * For convert case, just skip search free dev_extent, as caller
488 * is responsible to make sure it's free.
489 */
490 if (!convert) {
491 ret = find_free_dev_extent(trans, device, num_bytes,
492 start);
493 if (ret)
494 goto err;
495 }
496
497 key.objectid = device->devid;
498 key.offset = *start;
499 key.type = BTRFS_DEV_EXTENT_KEY;
500 ret = btrfs_insert_empty_item(trans, root, path, &key,
501 sizeof(*extent));
502 BUG_ON(ret);
503
504 leaf = path->nodes[0];
505 extent = btrfs_item_ptr(leaf, path->slots[0],
506 struct btrfs_dev_extent);
507 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
508 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
509 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
510
511 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
512 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
513 BTRFS_UUID_SIZE);
514
515 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
516 btrfs_mark_buffer_dirty(leaf);
517 err:
518 btrfs_free_path(path);
519 return ret;
520 }
521
522 static int find_next_chunk(struct btrfs_fs_info *fs_info, u64 *offset)
523 {
524 struct btrfs_root *root = fs_info->chunk_root;
525 struct btrfs_path *path;
526 int ret;
527 struct btrfs_key key;
528 struct btrfs_chunk *chunk;
529 struct btrfs_key found_key;
530
531 path = btrfs_alloc_path();
532 if (!path)
533 return -ENOMEM;
534
535 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
536 key.offset = (u64)-1;
537 key.type = BTRFS_CHUNK_ITEM_KEY;
538
539 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
540 if (ret < 0)
541 goto error;
542
543 BUG_ON(ret == 0);
544
545 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
546 if (ret) {
547 *offset = 0;
548 } else {
549 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
550 path->slots[0]);
551 if (found_key.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
552 *offset = 0;
553 else {
554 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
555 struct btrfs_chunk);
556 *offset = found_key.offset +
557 btrfs_chunk_length(path->nodes[0], chunk);
558 }
559 }
560 ret = 0;
561 error:
562 btrfs_free_path(path);
563 return ret;
564 }
565
566 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
567 u64 *objectid)
568 {
569 int ret;
570 struct btrfs_key key;
571 struct btrfs_key found_key;
572
573 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
574 key.type = BTRFS_DEV_ITEM_KEY;
575 key.offset = (u64)-1;
576
577 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
578 if (ret < 0)
579 goto error;
580
581 BUG_ON(ret == 0);
582
583 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
584 BTRFS_DEV_ITEM_KEY);
585 if (ret) {
586 *objectid = 1;
587 } else {
588 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
589 path->slots[0]);
590 *objectid = found_key.offset + 1;
591 }
592 ret = 0;
593 error:
594 btrfs_release_path(path);
595 return ret;
596 }
597
598 /*
599 * the device information is stored in the chunk root
600 * the btrfs_device struct should be fully filled in
601 */
602 int btrfs_add_device(struct btrfs_trans_handle *trans,
603 struct btrfs_fs_info *fs_info,
604 struct btrfs_device *device)
605 {
606 int ret;
607 struct btrfs_path *path;
608 struct btrfs_dev_item *dev_item;
609 struct extent_buffer *leaf;
610 struct btrfs_key key;
611 struct btrfs_root *root = fs_info->chunk_root;
612 unsigned long ptr;
613 u64 free_devid = 0;
614
615 path = btrfs_alloc_path();
616 if (!path)
617 return -ENOMEM;
618
619 ret = find_next_devid(root, path, &free_devid);
620 if (ret)
621 goto out;
622
623 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
624 key.type = BTRFS_DEV_ITEM_KEY;
625 key.offset = free_devid;
626
627 ret = btrfs_insert_empty_item(trans, root, path, &key,
628 sizeof(*dev_item));
629 if (ret)
630 goto out;
631
632 leaf = path->nodes[0];
633 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
634
635 device->devid = free_devid;
636 btrfs_set_device_id(leaf, dev_item, device->devid);
637 btrfs_set_device_generation(leaf, dev_item, 0);
638 btrfs_set_device_type(leaf, dev_item, device->type);
639 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
640 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
641 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
642 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
643 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
644 btrfs_set_device_group(leaf, dev_item, 0);
645 btrfs_set_device_seek_speed(leaf, dev_item, 0);
646 btrfs_set_device_bandwidth(leaf, dev_item, 0);
647 btrfs_set_device_start_offset(leaf, dev_item, 0);
648
649 ptr = (unsigned long)btrfs_device_uuid(dev_item);
650 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
651 ptr = (unsigned long)btrfs_device_fsid(dev_item);
652 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
653 btrfs_mark_buffer_dirty(leaf);
654 ret = 0;
655
656 out:
657 btrfs_free_path(path);
658 return ret;
659 }
660
661 int btrfs_update_device(struct btrfs_trans_handle *trans,
662 struct btrfs_device *device)
663 {
664 int ret;
665 struct btrfs_path *path;
666 struct btrfs_root *root;
667 struct btrfs_dev_item *dev_item;
668 struct extent_buffer *leaf;
669 struct btrfs_key key;
670
671 root = device->dev_root->fs_info->chunk_root;
672
673 path = btrfs_alloc_path();
674 if (!path)
675 return -ENOMEM;
676
677 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
678 key.type = BTRFS_DEV_ITEM_KEY;
679 key.offset = device->devid;
680
681 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
682 if (ret < 0)
683 goto out;
684
685 if (ret > 0) {
686 ret = -ENOENT;
687 goto out;
688 }
689
690 leaf = path->nodes[0];
691 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
692
693 btrfs_set_device_id(leaf, dev_item, device->devid);
694 btrfs_set_device_type(leaf, dev_item, device->type);
695 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
696 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
697 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
698 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
699 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
700 btrfs_mark_buffer_dirty(leaf);
701
702 out:
703 btrfs_free_path(path);
704 return ret;
705 }
706
707 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
708 struct btrfs_chunk *chunk, int item_size)
709 {
710 struct btrfs_super_block *super_copy = fs_info->super_copy;
711 struct btrfs_disk_key disk_key;
712 u32 array_size;
713 u8 *ptr;
714
715 array_size = btrfs_super_sys_array_size(super_copy);
716 if (array_size + item_size + sizeof(disk_key)
717 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
718 return -EFBIG;
719
720 ptr = super_copy->sys_chunk_array + array_size;
721 btrfs_cpu_key_to_disk(&disk_key, key);
722 memcpy(ptr, &disk_key, sizeof(disk_key));
723 ptr += sizeof(disk_key);
724 memcpy(ptr, chunk, item_size);
725 item_size += sizeof(disk_key);
726 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
727 return 0;
728 }
729
730 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
731 int sub_stripes)
732 {
733 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
734 return calc_size;
735 else if (type & BTRFS_BLOCK_GROUP_RAID10)
736 return calc_size * (num_stripes / sub_stripes);
737 else if (type & BTRFS_BLOCK_GROUP_RAID5)
738 return calc_size * (num_stripes - 1);
739 else if (type & BTRFS_BLOCK_GROUP_RAID6)
740 return calc_size * (num_stripes - 2);
741 else
742 return calc_size * num_stripes;
743 }
744
745
746 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
747 {
748 /* TODO, add a way to store the preferred stripe size */
749 return BTRFS_STRIPE_LEN;
750 }
751
752 /*
753 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
754 *
755 * It is not equal to "device->total_bytes - device->bytes_used".
756 * We do not allocate any chunk in 1M at beginning of device, and not
757 * allowed to allocate any chunk before alloc_start if it is specified.
758 * So search holes from max(1M, alloc_start) to device->total_bytes.
759 */
760 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
761 struct btrfs_device *device,
762 u64 *avail_bytes)
763 {
764 struct btrfs_path *path;
765 struct btrfs_root *root = device->dev_root;
766 struct btrfs_key key;
767 struct btrfs_dev_extent *dev_extent = NULL;
768 struct extent_buffer *l;
769 u64 search_start = root->fs_info->alloc_start;
770 u64 search_end = device->total_bytes;
771 u64 extent_end = 0;
772 u64 free_bytes = 0;
773 int ret;
774 int slot = 0;
775
776 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
777
778 path = btrfs_alloc_path();
779 if (!path)
780 return -ENOMEM;
781
782 key.objectid = device->devid;
783 key.offset = root->fs_info->alloc_start;
784 key.type = BTRFS_DEV_EXTENT_KEY;
785
786 path->reada = 2;
787 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
788 if (ret < 0)
789 goto error;
790 ret = btrfs_previous_item(root, path, 0, key.type);
791 if (ret < 0)
792 goto error;
793
794 while (1) {
795 l = path->nodes[0];
796 slot = path->slots[0];
797 if (slot >= btrfs_header_nritems(l)) {
798 ret = btrfs_next_leaf(root, path);
799 if (ret == 0)
800 continue;
801 if (ret < 0)
802 goto error;
803 break;
804 }
805 btrfs_item_key_to_cpu(l, &key, slot);
806
807 if (key.objectid < device->devid)
808 goto next;
809 if (key.objectid > device->devid)
810 break;
811 if (key.type != BTRFS_DEV_EXTENT_KEY)
812 goto next;
813 if (key.offset > search_end)
814 break;
815 if (key.offset > search_start)
816 free_bytes += key.offset - search_start;
817
818 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
819 extent_end = key.offset + btrfs_dev_extent_length(l,
820 dev_extent);
821 if (extent_end > search_start)
822 search_start = extent_end;
823 if (search_start > search_end)
824 break;
825 next:
826 path->slots[0]++;
827 cond_resched();
828 }
829
830 if (search_start < search_end)
831 free_bytes += search_end - search_start;
832
833 *avail_bytes = free_bytes;
834 ret = 0;
835 error:
836 btrfs_free_path(path);
837 return ret;
838 }
839
840 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
841 - sizeof(struct btrfs_item) \
842 - sizeof(struct btrfs_chunk)) \
843 / sizeof(struct btrfs_stripe) + 1)
844
845 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
846 - 2 * sizeof(struct btrfs_disk_key) \
847 - 2 * sizeof(struct btrfs_chunk)) \
848 / sizeof(struct btrfs_stripe) + 1)
849
850 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
851 struct btrfs_fs_info *info, u64 *start,
852 u64 *num_bytes, u64 type)
853 {
854 u64 dev_offset;
855 struct btrfs_root *extent_root = info->extent_root;
856 struct btrfs_root *chunk_root = info->chunk_root;
857 struct btrfs_stripe *stripes;
858 struct btrfs_device *device = NULL;
859 struct btrfs_chunk *chunk;
860 struct list_head private_devs;
861 struct list_head *dev_list = &info->fs_devices->devices;
862 struct list_head *cur;
863 struct map_lookup *map;
864 int min_stripe_size = SZ_1M;
865 u64 calc_size = SZ_8M;
866 u64 min_free;
867 u64 max_chunk_size = 4 * calc_size;
868 u64 avail = 0;
869 u64 max_avail = 0;
870 u64 percent_max;
871 int num_stripes = 1;
872 int max_stripes = 0;
873 int min_stripes = 1;
874 int sub_stripes = 0;
875 int looped = 0;
876 int ret;
877 int index;
878 int stripe_len = BTRFS_STRIPE_LEN;
879 struct btrfs_key key;
880 u64 offset;
881
882 if (list_empty(dev_list)) {
883 return -ENOSPC;
884 }
885
886 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
887 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
888 calc_size = SZ_8M;
889 max_chunk_size = calc_size * 2;
890 min_stripe_size = SZ_1M;
891 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
892 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
893 calc_size = SZ_1G;
894 max_chunk_size = 10 * calc_size;
895 min_stripe_size = SZ_64M;
896 max_stripes = BTRFS_MAX_DEVS(chunk_root);
897 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
898 calc_size = SZ_1G;
899 max_chunk_size = 4 * calc_size;
900 min_stripe_size = SZ_32M;
901 max_stripes = BTRFS_MAX_DEVS(chunk_root);
902 }
903 }
904 if (type & BTRFS_BLOCK_GROUP_RAID1) {
905 num_stripes = min_t(u64, 2,
906 btrfs_super_num_devices(info->super_copy));
907 if (num_stripes < 2)
908 return -ENOSPC;
909 min_stripes = 2;
910 }
911 if (type & BTRFS_BLOCK_GROUP_DUP) {
912 num_stripes = 2;
913 min_stripes = 2;
914 }
915 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
916 num_stripes = btrfs_super_num_devices(info->super_copy);
917 if (num_stripes > max_stripes)
918 num_stripes = max_stripes;
919 min_stripes = 2;
920 }
921 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
922 num_stripes = btrfs_super_num_devices(info->super_copy);
923 if (num_stripes > max_stripes)
924 num_stripes = max_stripes;
925 if (num_stripes < 4)
926 return -ENOSPC;
927 num_stripes &= ~(u32)1;
928 sub_stripes = 2;
929 min_stripes = 4;
930 }
931 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
932 num_stripes = btrfs_super_num_devices(info->super_copy);
933 if (num_stripes > max_stripes)
934 num_stripes = max_stripes;
935 if (num_stripes < 2)
936 return -ENOSPC;
937 min_stripes = 2;
938 stripe_len = find_raid56_stripe_len(num_stripes - 1,
939 btrfs_super_stripesize(info->super_copy));
940 }
941 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
942 num_stripes = btrfs_super_num_devices(info->super_copy);
943 if (num_stripes > max_stripes)
944 num_stripes = max_stripes;
945 if (num_stripes < 3)
946 return -ENOSPC;
947 min_stripes = 3;
948 stripe_len = find_raid56_stripe_len(num_stripes - 2,
949 btrfs_super_stripesize(info->super_copy));
950 }
951
952 /* we don't want a chunk larger than 10% of the FS */
953 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
954 max_chunk_size = min(percent_max, max_chunk_size);
955
956 again:
957 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
958 max_chunk_size) {
959 calc_size = max_chunk_size;
960 calc_size /= num_stripes;
961 calc_size /= stripe_len;
962 calc_size *= stripe_len;
963 }
964 /* we don't want tiny stripes */
965 calc_size = max_t(u64, calc_size, min_stripe_size);
966
967 calc_size /= stripe_len;
968 calc_size *= stripe_len;
969 INIT_LIST_HEAD(&private_devs);
970 cur = dev_list->next;
971 index = 0;
972
973 if (type & BTRFS_BLOCK_GROUP_DUP)
974 min_free = calc_size * 2;
975 else
976 min_free = calc_size;
977
978 /* build a private list of devices we will allocate from */
979 while(index < num_stripes) {
980 device = list_entry(cur, struct btrfs_device, dev_list);
981 ret = btrfs_device_avail_bytes(trans, device, &avail);
982 if (ret)
983 return ret;
984 cur = cur->next;
985 if (avail >= min_free) {
986 list_move_tail(&device->dev_list, &private_devs);
987 index++;
988 if (type & BTRFS_BLOCK_GROUP_DUP)
989 index++;
990 } else if (avail > max_avail)
991 max_avail = avail;
992 if (cur == dev_list)
993 break;
994 }
995 if (index < num_stripes) {
996 list_splice(&private_devs, dev_list);
997 if (index >= min_stripes) {
998 num_stripes = index;
999 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1000 num_stripes /= sub_stripes;
1001 num_stripes *= sub_stripes;
1002 }
1003 looped = 1;
1004 goto again;
1005 }
1006 if (!looped && max_avail > 0) {
1007 looped = 1;
1008 calc_size = max_avail;
1009 goto again;
1010 }
1011 return -ENOSPC;
1012 }
1013 ret = find_next_chunk(info, &offset);
1014 if (ret)
1015 return ret;
1016 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1017 key.type = BTRFS_CHUNK_ITEM_KEY;
1018 key.offset = offset;
1019
1020 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1021 if (!chunk)
1022 return -ENOMEM;
1023
1024 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1025 if (!map) {
1026 kfree(chunk);
1027 return -ENOMEM;
1028 }
1029
1030 stripes = &chunk->stripe;
1031 *num_bytes = chunk_bytes_by_type(type, calc_size,
1032 num_stripes, sub_stripes);
1033 index = 0;
1034 while(index < num_stripes) {
1035 struct btrfs_stripe *stripe;
1036 BUG_ON(list_empty(&private_devs));
1037 cur = private_devs.next;
1038 device = list_entry(cur, struct btrfs_device, dev_list);
1039
1040 /* loop over this device again if we're doing a dup group */
1041 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1042 (index == num_stripes - 1))
1043 list_move_tail(&device->dev_list, dev_list);
1044
1045 ret = btrfs_alloc_dev_extent(trans, device,
1046 info->chunk_root->root_key.objectid,
1047 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1048 calc_size, &dev_offset, 0);
1049 BUG_ON(ret);
1050
1051 device->bytes_used += calc_size;
1052 ret = btrfs_update_device(trans, device);
1053 BUG_ON(ret);
1054
1055 map->stripes[index].dev = device;
1056 map->stripes[index].physical = dev_offset;
1057 stripe = stripes + index;
1058 btrfs_set_stack_stripe_devid(stripe, device->devid);
1059 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1060 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1061 index++;
1062 }
1063 BUG_ON(!list_empty(&private_devs));
1064
1065 /* key was set above */
1066 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1067 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1068 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1069 btrfs_set_stack_chunk_type(chunk, type);
1070 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1071 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1072 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1073 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1074 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1075 map->sector_size = info->sectorsize;
1076 map->stripe_len = stripe_len;
1077 map->io_align = stripe_len;
1078 map->io_width = stripe_len;
1079 map->type = type;
1080 map->num_stripes = num_stripes;
1081 map->sub_stripes = sub_stripes;
1082
1083 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1084 btrfs_chunk_item_size(num_stripes));
1085 BUG_ON(ret);
1086 *start = key.offset;;
1087
1088 map->ce.start = key.offset;
1089 map->ce.size = *num_bytes;
1090
1091 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1092 BUG_ON(ret);
1093
1094 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1095 ret = btrfs_add_system_chunk(info, &key,
1096 chunk, btrfs_chunk_item_size(num_stripes));
1097 BUG_ON(ret);
1098 }
1099
1100 kfree(chunk);
1101 return ret;
1102 }
1103
1104 /*
1105 * Alloc a DATA chunk with SINGLE profile.
1106 *
1107 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1108 * (btrfs logical bytenr == on-disk bytenr)
1109 * For that case, caller must make sure the chunk and dev_extent are not
1110 * occupied.
1111 */
1112 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1113 struct btrfs_fs_info *info, u64 *start,
1114 u64 num_bytes, u64 type, int convert)
1115 {
1116 u64 dev_offset;
1117 struct btrfs_root *extent_root = info->extent_root;
1118 struct btrfs_root *chunk_root = info->chunk_root;
1119 struct btrfs_stripe *stripes;
1120 struct btrfs_device *device = NULL;
1121 struct btrfs_chunk *chunk;
1122 struct list_head *dev_list = &info->fs_devices->devices;
1123 struct list_head *cur;
1124 struct map_lookup *map;
1125 u64 calc_size = SZ_8M;
1126 int num_stripes = 1;
1127 int sub_stripes = 0;
1128 int ret;
1129 int index;
1130 int stripe_len = BTRFS_STRIPE_LEN;
1131 struct btrfs_key key;
1132
1133 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1134 key.type = BTRFS_CHUNK_ITEM_KEY;
1135 if (convert) {
1136 if (*start != round_down(*start, info->sectorsize)) {
1137 error("DATA chunk start not sectorsize aligned: %llu",
1138 (unsigned long long)*start);
1139 return -EINVAL;
1140 }
1141 key.offset = *start;
1142 dev_offset = *start;
1143 } else {
1144 u64 tmp;
1145
1146 ret = find_next_chunk(info, &tmp);
1147 key.offset = tmp;
1148 if (ret)
1149 return ret;
1150 }
1151
1152 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1153 if (!chunk)
1154 return -ENOMEM;
1155
1156 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1157 if (!map) {
1158 kfree(chunk);
1159 return -ENOMEM;
1160 }
1161
1162 stripes = &chunk->stripe;
1163 calc_size = num_bytes;
1164
1165 index = 0;
1166 cur = dev_list->next;
1167 device = list_entry(cur, struct btrfs_device, dev_list);
1168
1169 while (index < num_stripes) {
1170 struct btrfs_stripe *stripe;
1171
1172 ret = btrfs_alloc_dev_extent(trans, device,
1173 info->chunk_root->root_key.objectid,
1174 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1175 calc_size, &dev_offset, convert);
1176 BUG_ON(ret);
1177
1178 device->bytes_used += calc_size;
1179 ret = btrfs_update_device(trans, device);
1180 BUG_ON(ret);
1181
1182 map->stripes[index].dev = device;
1183 map->stripes[index].physical = dev_offset;
1184 stripe = stripes + index;
1185 btrfs_set_stack_stripe_devid(stripe, device->devid);
1186 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1187 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1188 index++;
1189 }
1190
1191 /* key was set above */
1192 btrfs_set_stack_chunk_length(chunk, num_bytes);
1193 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1194 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1195 btrfs_set_stack_chunk_type(chunk, type);
1196 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1197 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1198 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1199 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1200 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1201 map->sector_size = info->sectorsize;
1202 map->stripe_len = stripe_len;
1203 map->io_align = stripe_len;
1204 map->io_width = stripe_len;
1205 map->type = type;
1206 map->num_stripes = num_stripes;
1207 map->sub_stripes = sub_stripes;
1208
1209 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1210 btrfs_chunk_item_size(num_stripes));
1211 BUG_ON(ret);
1212 if (!convert)
1213 *start = key.offset;
1214
1215 map->ce.start = key.offset;
1216 map->ce.size = num_bytes;
1217
1218 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1219 BUG_ON(ret);
1220
1221 kfree(chunk);
1222 return ret;
1223 }
1224
1225 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1226 {
1227 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1228 struct cache_extent *ce;
1229 struct map_lookup *map;
1230 int ret;
1231
1232 ce = search_cache_extent(&map_tree->cache_tree, logical);
1233 if (!ce) {
1234 fprintf(stderr, "No mapping for %llu-%llu\n",
1235 (unsigned long long)logical,
1236 (unsigned long long)logical+len);
1237 return 1;
1238 }
1239 if (ce->start > logical || ce->start + ce->size < logical) {
1240 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1241 "%llu-%llu\n", (unsigned long long)logical,
1242 (unsigned long long)logical+len,
1243 (unsigned long long)ce->start,
1244 (unsigned long long)ce->start + ce->size);
1245 return 1;
1246 }
1247 map = container_of(ce, struct map_lookup, ce);
1248
1249 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1250 ret = map->num_stripes;
1251 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1252 ret = map->sub_stripes;
1253 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1254 ret = 2;
1255 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1256 ret = 3;
1257 else
1258 ret = 1;
1259 return ret;
1260 }
1261
1262 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1263 u64 *size, u64 type)
1264 {
1265 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1266 struct cache_extent *ce;
1267 struct map_lookup *map;
1268 u64 cur = *logical;
1269
1270 ce = search_cache_extent(&map_tree->cache_tree, cur);
1271
1272 while (ce) {
1273 /*
1274 * only jump to next bg if our cur is not 0
1275 * As the initial logical for btrfs_next_bg() is 0, and
1276 * if we jump to next bg, we skipped a valid bg.
1277 */
1278 if (cur) {
1279 ce = next_cache_extent(ce);
1280 if (!ce)
1281 return -ENOENT;
1282 }
1283
1284 cur = ce->start;
1285 map = container_of(ce, struct map_lookup, ce);
1286 if (map->type & type) {
1287 *logical = ce->start;
1288 *size = ce->size;
1289 return 0;
1290 }
1291 if (!cur)
1292 ce = next_cache_extent(ce);
1293 }
1294
1295 return -ENOENT;
1296 }
1297
1298 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1299 u64 chunk_start, u64 physical, u64 devid,
1300 u64 **logical, int *naddrs, int *stripe_len)
1301 {
1302 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1303 struct cache_extent *ce;
1304 struct map_lookup *map;
1305 u64 *buf;
1306 u64 bytenr;
1307 u64 length;
1308 u64 stripe_nr;
1309 u64 rmap_len;
1310 int i, j, nr = 0;
1311
1312 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1313 BUG_ON(!ce);
1314 map = container_of(ce, struct map_lookup, ce);
1315
1316 length = ce->size;
1317 rmap_len = map->stripe_len;
1318 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1319 length = ce->size / (map->num_stripes / map->sub_stripes);
1320 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1321 length = ce->size / map->num_stripes;
1322 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1323 BTRFS_BLOCK_GROUP_RAID6)) {
1324 length = ce->size / nr_data_stripes(map);
1325 rmap_len = map->stripe_len * nr_data_stripes(map);
1326 }
1327
1328 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1329
1330 for (i = 0; i < map->num_stripes; i++) {
1331 if (devid && map->stripes[i].dev->devid != devid)
1332 continue;
1333 if (map->stripes[i].physical > physical ||
1334 map->stripes[i].physical + length <= physical)
1335 continue;
1336
1337 stripe_nr = (physical - map->stripes[i].physical) /
1338 map->stripe_len;
1339
1340 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1341 stripe_nr = (stripe_nr * map->num_stripes + i) /
1342 map->sub_stripes;
1343 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1344 stripe_nr = stripe_nr * map->num_stripes + i;
1345 } /* else if RAID[56], multiply by nr_data_stripes().
1346 * Alternatively, just use rmap_len below instead of
1347 * map->stripe_len */
1348
1349 bytenr = ce->start + stripe_nr * rmap_len;
1350 for (j = 0; j < nr; j++) {
1351 if (buf[j] == bytenr)
1352 break;
1353 }
1354 if (j == nr)
1355 buf[nr++] = bytenr;
1356 }
1357
1358 *logical = buf;
1359 *naddrs = nr;
1360 *stripe_len = rmap_len;
1361
1362 return 0;
1363 }
1364
1365 static inline int parity_smaller(u64 a, u64 b)
1366 {
1367 return a > b;
1368 }
1369
1370 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1371 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1372 {
1373 struct btrfs_bio_stripe s;
1374 int i;
1375 u64 l;
1376 int again = 1;
1377
1378 while (again) {
1379 again = 0;
1380 for (i = 0; i < bbio->num_stripes - 1; i++) {
1381 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1382 s = bbio->stripes[i];
1383 l = raid_map[i];
1384 bbio->stripes[i] = bbio->stripes[i+1];
1385 raid_map[i] = raid_map[i+1];
1386 bbio->stripes[i+1] = s;
1387 raid_map[i+1] = l;
1388 again = 1;
1389 }
1390 }
1391 }
1392 }
1393
1394 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1395 u64 logical, u64 *length,
1396 struct btrfs_multi_bio **multi_ret, int mirror_num,
1397 u64 **raid_map_ret)
1398 {
1399 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1400 multi_ret, mirror_num, raid_map_ret);
1401 }
1402
1403 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1404 u64 logical, u64 *length, u64 *type,
1405 struct btrfs_multi_bio **multi_ret, int mirror_num,
1406 u64 **raid_map_ret)
1407 {
1408 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1409 struct cache_extent *ce;
1410 struct map_lookup *map;
1411 u64 offset;
1412 u64 stripe_offset;
1413 u64 stripe_nr;
1414 u64 *raid_map = NULL;
1415 int stripes_allocated = 8;
1416 int stripes_required = 1;
1417 int stripe_index;
1418 int i;
1419 struct btrfs_multi_bio *multi = NULL;
1420
1421 if (multi_ret && rw == READ) {
1422 stripes_allocated = 1;
1423 }
1424 again:
1425 ce = search_cache_extent(&map_tree->cache_tree, logical);
1426 if (!ce) {
1427 kfree(multi);
1428 *length = (u64)-1;
1429 return -ENOENT;
1430 }
1431 if (ce->start > logical) {
1432 kfree(multi);
1433 *length = ce->start - logical;
1434 return -ENOENT;
1435 }
1436
1437 if (multi_ret) {
1438 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1439 GFP_NOFS);
1440 if (!multi)
1441 return -ENOMEM;
1442 }
1443 map = container_of(ce, struct map_lookup, ce);
1444 offset = logical - ce->start;
1445
1446 if (rw == WRITE) {
1447 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1448 BTRFS_BLOCK_GROUP_DUP)) {
1449 stripes_required = map->num_stripes;
1450 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1451 stripes_required = map->sub_stripes;
1452 }
1453 }
1454 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1455 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1456 /* RAID[56] write or recovery. Return all stripes */
1457 stripes_required = map->num_stripes;
1458
1459 /* Only allocate the map if we've already got a large enough multi_ret */
1460 if (stripes_allocated >= stripes_required) {
1461 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1462 if (!raid_map) {
1463 kfree(multi);
1464 return -ENOMEM;
1465 }
1466 }
1467 }
1468
1469 /* if our multi bio struct is too small, back off and try again */
1470 if (multi_ret && stripes_allocated < stripes_required) {
1471 stripes_allocated = stripes_required;
1472 kfree(multi);
1473 multi = NULL;
1474 goto again;
1475 }
1476 stripe_nr = offset;
1477 /*
1478 * stripe_nr counts the total number of stripes we have to stride
1479 * to get to this block
1480 */
1481 stripe_nr = stripe_nr / map->stripe_len;
1482
1483 stripe_offset = stripe_nr * map->stripe_len;
1484 BUG_ON(offset < stripe_offset);
1485
1486 /* stripe_offset is the offset of this block in its stripe*/
1487 stripe_offset = offset - stripe_offset;
1488
1489 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1490 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1491 BTRFS_BLOCK_GROUP_RAID10 |
1492 BTRFS_BLOCK_GROUP_DUP)) {
1493 /* we limit the length of each bio to what fits in a stripe */
1494 *length = min_t(u64, ce->size - offset,
1495 map->stripe_len - stripe_offset);
1496 } else {
1497 *length = ce->size - offset;
1498 }
1499
1500 if (!multi_ret)
1501 goto out;
1502
1503 multi->num_stripes = 1;
1504 stripe_index = 0;
1505 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1506 if (rw == WRITE)
1507 multi->num_stripes = map->num_stripes;
1508 else if (mirror_num)
1509 stripe_index = mirror_num - 1;
1510 else
1511 stripe_index = stripe_nr % map->num_stripes;
1512 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1513 int factor = map->num_stripes / map->sub_stripes;
1514
1515 stripe_index = stripe_nr % factor;
1516 stripe_index *= map->sub_stripes;
1517
1518 if (rw == WRITE)
1519 multi->num_stripes = map->sub_stripes;
1520 else if (mirror_num)
1521 stripe_index += mirror_num - 1;
1522
1523 stripe_nr = stripe_nr / factor;
1524 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1525 if (rw == WRITE)
1526 multi->num_stripes = map->num_stripes;
1527 else if (mirror_num)
1528 stripe_index = mirror_num - 1;
1529 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1530 BTRFS_BLOCK_GROUP_RAID6)) {
1531
1532 if (raid_map) {
1533 int rot;
1534 u64 tmp;
1535 u64 raid56_full_stripe_start;
1536 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1537
1538 /*
1539 * align the start of our data stripe in the logical
1540 * address space
1541 */
1542 raid56_full_stripe_start = offset / full_stripe_len;
1543 raid56_full_stripe_start *= full_stripe_len;
1544
1545 /* get the data stripe number */
1546 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1547 stripe_nr = stripe_nr / nr_data_stripes(map);
1548
1549 /* Work out the disk rotation on this stripe-set */
1550 rot = stripe_nr % map->num_stripes;
1551
1552 /* Fill in the logical address of each stripe */
1553 tmp = stripe_nr * nr_data_stripes(map);
1554
1555 for (i = 0; i < nr_data_stripes(map); i++)
1556 raid_map[(i+rot) % map->num_stripes] =
1557 ce->start + (tmp + i) * map->stripe_len;
1558
1559 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1560 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1561 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1562
1563 *length = map->stripe_len;
1564 stripe_index = 0;
1565 stripe_offset = 0;
1566 multi->num_stripes = map->num_stripes;
1567 } else {
1568 stripe_index = stripe_nr % nr_data_stripes(map);
1569 stripe_nr = stripe_nr / nr_data_stripes(map);
1570
1571 /*
1572 * Mirror #0 or #1 means the original data block.
1573 * Mirror #2 is RAID5 parity block.
1574 * Mirror #3 is RAID6 Q block.
1575 */
1576 if (mirror_num > 1)
1577 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1578
1579 /* We distribute the parity blocks across stripes */
1580 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1581 }
1582 } else {
1583 /*
1584 * after this do_div call, stripe_nr is the number of stripes
1585 * on this device we have to walk to find the data, and
1586 * stripe_index is the number of our device in the stripe array
1587 */
1588 stripe_index = stripe_nr % map->num_stripes;
1589 stripe_nr = stripe_nr / map->num_stripes;
1590 }
1591 BUG_ON(stripe_index >= map->num_stripes);
1592
1593 for (i = 0; i < multi->num_stripes; i++) {
1594 multi->stripes[i].physical =
1595 map->stripes[stripe_index].physical + stripe_offset +
1596 stripe_nr * map->stripe_len;
1597 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1598 stripe_index++;
1599 }
1600 *multi_ret = multi;
1601
1602 if (type)
1603 *type = map->type;
1604
1605 if (raid_map) {
1606 sort_parity_stripes(multi, raid_map);
1607 *raid_map_ret = raid_map;
1608 }
1609 out:
1610 return 0;
1611 }
1612
1613 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1614 u8 *uuid, u8 *fsid)
1615 {
1616 struct btrfs_device *device;
1617 struct btrfs_fs_devices *cur_devices;
1618
1619 cur_devices = fs_info->fs_devices;
1620 while (cur_devices) {
1621 if (!fsid ||
1622 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1623 fs_info->ignore_fsid_mismatch)) {
1624 device = __find_device(&cur_devices->devices,
1625 devid, uuid);
1626 if (device)
1627 return device;
1628 }
1629 cur_devices = cur_devices->seed;
1630 }
1631 return NULL;
1632 }
1633
1634 struct btrfs_device *
1635 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1636 u64 devid, int instance)
1637 {
1638 struct list_head *head = &fs_devices->devices;
1639 struct btrfs_device *dev;
1640 int num_found = 0;
1641
1642 list_for_each_entry(dev, head, dev_list) {
1643 if (dev->devid == devid && num_found++ == instance)
1644 return dev;
1645 }
1646 return NULL;
1647 }
1648
1649 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1650 {
1651 struct cache_extent *ce;
1652 struct map_lookup *map;
1653 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1654 int readonly = 0;
1655 int i;
1656
1657 /*
1658 * During chunk recovering, we may fail to find block group's
1659 * corresponding chunk, we will rebuild it later
1660 */
1661 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1662 if (!fs_info->is_chunk_recover)
1663 BUG_ON(!ce);
1664 else
1665 return 0;
1666
1667 map = container_of(ce, struct map_lookup, ce);
1668 for (i = 0; i < map->num_stripes; i++) {
1669 if (!map->stripes[i].dev->writeable) {
1670 readonly = 1;
1671 break;
1672 }
1673 }
1674
1675 return readonly;
1676 }
1677
1678 static struct btrfs_device *fill_missing_device(u64 devid)
1679 {
1680 struct btrfs_device *device;
1681
1682 device = kzalloc(sizeof(*device), GFP_NOFS);
1683 device->devid = devid;
1684 device->fd = -1;
1685 return device;
1686 }
1687
1688 /*
1689 * slot == -1: SYSTEM chunk
1690 * return -EIO on error, otherwise return 0
1691 */
1692 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1693 struct extent_buffer *leaf,
1694 struct btrfs_chunk *chunk,
1695 int slot, u64 logical)
1696 {
1697 u64 length;
1698 u64 stripe_len;
1699 u16 num_stripes;
1700 u16 sub_stripes;
1701 u64 type;
1702 u32 chunk_ondisk_size;
1703 u32 sectorsize = fs_info->sectorsize;
1704
1705 length = btrfs_chunk_length(leaf, chunk);
1706 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1707 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1708 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1709 type = btrfs_chunk_type(leaf, chunk);
1710
1711 /*
1712 * These valid checks may be insufficient to cover every corner cases.
1713 */
1714 if (!IS_ALIGNED(logical, sectorsize)) {
1715 error("invalid chunk logical %llu", logical);
1716 return -EIO;
1717 }
1718 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1719 error("invalid chunk sectorsize %llu",
1720 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1721 return -EIO;
1722 }
1723 if (!length || !IS_ALIGNED(length, sectorsize)) {
1724 error("invalid chunk length %llu", length);
1725 return -EIO;
1726 }
1727 if (stripe_len != BTRFS_STRIPE_LEN) {
1728 error("invalid chunk stripe length: %llu", stripe_len);
1729 return -EIO;
1730 }
1731 /* Check on chunk item type */
1732 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1733 error("invalid chunk type %llu", type);
1734 return -EIO;
1735 }
1736 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1737 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1738 error("unrecognized chunk type: %llu",
1739 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1740 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1741 return -EIO;
1742 }
1743 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1744 error("missing chunk type flag: %llu", type);
1745 return -EIO;
1746 }
1747 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1748 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1749 error("conflicting chunk type detected: %llu", type);
1750 return -EIO;
1751 }
1752 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1753 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1754 error("conflicting chunk profile detected: %llu", type);
1755 return -EIO;
1756 }
1757
1758 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1759 /*
1760 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1761 * it can't exceed the system chunk array size
1762 * For normal chunk, it should match its chunk item size.
1763 */
1764 if (num_stripes < 1 ||
1765 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1766 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1767 error("invalid num_stripes: %u", num_stripes);
1768 return -EIO;
1769 }
1770 /*
1771 * Device number check against profile
1772 */
1773 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1774 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1775 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1776 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1777 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1778 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1779 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1780 num_stripes != 1)) {
1781 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1782 num_stripes, sub_stripes,
1783 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1784 return -EIO;
1785 }
1786
1787 return 0;
1788 }
1789
1790 /*
1791 * Slot is used to verify the chunk item is valid
1792 *
1793 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1794 */
1795 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1796 struct extent_buffer *leaf,
1797 struct btrfs_chunk *chunk, int slot)
1798 {
1799 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1800 struct map_lookup *map;
1801 struct cache_extent *ce;
1802 u64 logical;
1803 u64 length;
1804 u64 devid;
1805 u8 uuid[BTRFS_UUID_SIZE];
1806 int num_stripes;
1807 int ret;
1808 int i;
1809
1810 logical = key->offset;
1811 length = btrfs_chunk_length(leaf, chunk);
1812 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1813 /* Validation check */
1814 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1815 if (ret) {
1816 error("%s checksums match, but it has an invalid chunk, %s",
1817 (slot == -1) ? "Superblock" : "Metadata",
1818 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1819 return ret;
1820 }
1821
1822 ce = search_cache_extent(&map_tree->cache_tree, logical);
1823
1824 /* already mapped? */
1825 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1826 return 0;
1827 }
1828
1829 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1830 if (!map)
1831 return -ENOMEM;
1832
1833 map->ce.start = logical;
1834 map->ce.size = length;
1835 map->num_stripes = num_stripes;
1836 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1837 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1838 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1839 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1840 map->type = btrfs_chunk_type(leaf, chunk);
1841 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1842
1843 for (i = 0; i < num_stripes; i++) {
1844 map->stripes[i].physical =
1845 btrfs_stripe_offset_nr(leaf, chunk, i);
1846 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1847 read_extent_buffer(leaf, uuid, (unsigned long)
1848 btrfs_stripe_dev_uuid_nr(chunk, i),
1849 BTRFS_UUID_SIZE);
1850 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1851 NULL);
1852 if (!map->stripes[i].dev) {
1853 map->stripes[i].dev = fill_missing_device(devid);
1854 printf("warning, device %llu is missing\n",
1855 (unsigned long long)devid);
1856 list_add(&map->stripes[i].dev->dev_list,
1857 &fs_info->fs_devices->devices);
1858 }
1859
1860 }
1861 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1862 BUG_ON(ret);
1863
1864 return 0;
1865 }
1866
1867 static int fill_device_from_item(struct extent_buffer *leaf,
1868 struct btrfs_dev_item *dev_item,
1869 struct btrfs_device *device)
1870 {
1871 unsigned long ptr;
1872
1873 device->devid = btrfs_device_id(leaf, dev_item);
1874 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1875 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1876 device->type = btrfs_device_type(leaf, dev_item);
1877 device->io_align = btrfs_device_io_align(leaf, dev_item);
1878 device->io_width = btrfs_device_io_width(leaf, dev_item);
1879 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1880
1881 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1882 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1883
1884 return 0;
1885 }
1886
1887 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1888 {
1889 struct btrfs_fs_devices *fs_devices;
1890 int ret;
1891
1892 fs_devices = fs_info->fs_devices->seed;
1893 while (fs_devices) {
1894 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1895 ret = 0;
1896 goto out;
1897 }
1898 fs_devices = fs_devices->seed;
1899 }
1900
1901 fs_devices = find_fsid(fsid);
1902 if (!fs_devices) {
1903 /* missing all seed devices */
1904 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1905 if (!fs_devices) {
1906 ret = -ENOMEM;
1907 goto out;
1908 }
1909 INIT_LIST_HEAD(&fs_devices->devices);
1910 list_add(&fs_devices->list, &fs_uuids);
1911 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1912 }
1913
1914 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1915 if (ret)
1916 goto out;
1917
1918 fs_devices->seed = fs_info->fs_devices->seed;
1919 fs_info->fs_devices->seed = fs_devices;
1920 out:
1921 return ret;
1922 }
1923
1924 static int read_one_dev(struct btrfs_fs_info *fs_info,
1925 struct extent_buffer *leaf,
1926 struct btrfs_dev_item *dev_item)
1927 {
1928 struct btrfs_device *device;
1929 u64 devid;
1930 int ret = 0;
1931 u8 fs_uuid[BTRFS_UUID_SIZE];
1932 u8 dev_uuid[BTRFS_UUID_SIZE];
1933
1934 devid = btrfs_device_id(leaf, dev_item);
1935 read_extent_buffer(leaf, dev_uuid,
1936 (unsigned long)btrfs_device_uuid(dev_item),
1937 BTRFS_UUID_SIZE);
1938 read_extent_buffer(leaf, fs_uuid,
1939 (unsigned long)btrfs_device_fsid(dev_item),
1940 BTRFS_UUID_SIZE);
1941
1942 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1943 ret = open_seed_devices(fs_info, fs_uuid);
1944 if (ret)
1945 return ret;
1946 }
1947
1948 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1949 if (!device) {
1950 device = kzalloc(sizeof(*device), GFP_NOFS);
1951 if (!device)
1952 return -ENOMEM;
1953 device->fd = -1;
1954 list_add(&device->dev_list,
1955 &fs_info->fs_devices->devices);
1956 }
1957
1958 fill_device_from_item(leaf, dev_item, device);
1959 device->dev_root = fs_info->dev_root;
1960 return ret;
1961 }
1962
1963 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1964 {
1965 struct btrfs_super_block *super_copy = fs_info->super_copy;
1966 struct extent_buffer *sb;
1967 struct btrfs_disk_key *disk_key;
1968 struct btrfs_chunk *chunk;
1969 u8 *array_ptr;
1970 unsigned long sb_array_offset;
1971 int ret = 0;
1972 u32 num_stripes;
1973 u32 array_size;
1974 u32 len = 0;
1975 u32 cur_offset;
1976 struct btrfs_key key;
1977
1978 if (fs_info->nodesize < BTRFS_SUPER_INFO_SIZE) {
1979 printf("ERROR: nodesize %u too small to read superblock\n",
1980 fs_info->nodesize);
1981 return -EINVAL;
1982 }
1983 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
1984 if (!sb)
1985 return -ENOMEM;
1986 btrfs_set_buffer_uptodate(sb);
1987 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1988 array_size = btrfs_super_sys_array_size(super_copy);
1989
1990 array_ptr = super_copy->sys_chunk_array;
1991 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1992 cur_offset = 0;
1993
1994 while (cur_offset < array_size) {
1995 disk_key = (struct btrfs_disk_key *)array_ptr;
1996 len = sizeof(*disk_key);
1997 if (cur_offset + len > array_size)
1998 goto out_short_read;
1999
2000 btrfs_disk_key_to_cpu(&key, disk_key);
2001
2002 array_ptr += len;
2003 sb_array_offset += len;
2004 cur_offset += len;
2005
2006 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2007 chunk = (struct btrfs_chunk *)sb_array_offset;
2008 /*
2009 * At least one btrfs_chunk with one stripe must be
2010 * present, exact stripe count check comes afterwards
2011 */
2012 len = btrfs_chunk_item_size(1);
2013 if (cur_offset + len > array_size)
2014 goto out_short_read;
2015
2016 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2017 if (!num_stripes) {
2018 printk(
2019 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2020 num_stripes, cur_offset);
2021 ret = -EIO;
2022 break;
2023 }
2024
2025 len = btrfs_chunk_item_size(num_stripes);
2026 if (cur_offset + len > array_size)
2027 goto out_short_read;
2028
2029 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2030 if (ret)
2031 break;
2032 } else {
2033 printk(
2034 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2035 (u32)key.type, cur_offset);
2036 ret = -EIO;
2037 break;
2038 }
2039 array_ptr += len;
2040 sb_array_offset += len;
2041 cur_offset += len;
2042 }
2043 free_extent_buffer(sb);
2044 return ret;
2045
2046 out_short_read:
2047 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2048 len, cur_offset);
2049 free_extent_buffer(sb);
2050 return -EIO;
2051 }
2052
2053 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2054 {
2055 struct btrfs_path *path;
2056 struct extent_buffer *leaf;
2057 struct btrfs_key key;
2058 struct btrfs_key found_key;
2059 struct btrfs_root *root = fs_info->chunk_root;
2060 int ret;
2061 int slot;
2062
2063 path = btrfs_alloc_path();
2064 if (!path)
2065 return -ENOMEM;
2066
2067 /*
2068 * Read all device items, and then all the chunk items. All
2069 * device items are found before any chunk item (their object id
2070 * is smaller than the lowest possible object id for a chunk
2071 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2072 */
2073 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2074 key.offset = 0;
2075 key.type = 0;
2076 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2077 if (ret < 0)
2078 goto error;
2079 while(1) {
2080 leaf = path->nodes[0];
2081 slot = path->slots[0];
2082 if (slot >= btrfs_header_nritems(leaf)) {
2083 ret = btrfs_next_leaf(root, path);
2084 if (ret == 0)
2085 continue;
2086 if (ret < 0)
2087 goto error;
2088 break;
2089 }
2090 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2091 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2092 struct btrfs_dev_item *dev_item;
2093 dev_item = btrfs_item_ptr(leaf, slot,
2094 struct btrfs_dev_item);
2095 ret = read_one_dev(fs_info, leaf, dev_item);
2096 BUG_ON(ret);
2097 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2098 struct btrfs_chunk *chunk;
2099 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2100 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2101 slot);
2102 BUG_ON(ret);
2103 }
2104 path->slots[0]++;
2105 }
2106
2107 ret = 0;
2108 error:
2109 btrfs_free_path(path);
2110 return ret;
2111 }
2112
2113 struct list_head *btrfs_scanned_uuids(void)
2114 {
2115 return &fs_uuids;
2116 }
2117
2118 static int rmw_eb(struct btrfs_fs_info *info,
2119 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2120 {
2121 int ret;
2122 unsigned long orig_off = 0;
2123 unsigned long dest_off = 0;
2124 unsigned long copy_len = eb->len;
2125
2126 ret = read_whole_eb(info, eb, 0);
2127 if (ret)
2128 return ret;
2129
2130 if (eb->start + eb->len <= orig_eb->start ||
2131 eb->start >= orig_eb->start + orig_eb->len)
2132 return 0;
2133 /*
2134 * | ----- orig_eb ------- |
2135 * | ----- stripe ------- |
2136 * | ----- orig_eb ------- |
2137 * | ----- orig_eb ------- |
2138 */
2139 if (eb->start > orig_eb->start)
2140 orig_off = eb->start - orig_eb->start;
2141 if (orig_eb->start > eb->start)
2142 dest_off = orig_eb->start - eb->start;
2143
2144 if (copy_len > orig_eb->len - orig_off)
2145 copy_len = orig_eb->len - orig_off;
2146 if (copy_len > eb->len - dest_off)
2147 copy_len = eb->len - dest_off;
2148
2149 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2150 return 0;
2151 }
2152
2153 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2154 struct extent_buffer *orig_eb,
2155 struct extent_buffer **ebs,
2156 u64 stripe_len, u64 *raid_map,
2157 int num_stripes)
2158 {
2159 struct extent_buffer **tmp_ebs;
2160 u64 start = orig_eb->start;
2161 u64 this_eb_start;
2162 int i;
2163 int ret = 0;
2164
2165 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2166 if (!tmp_ebs)
2167 return -ENOMEM;
2168
2169 /* Alloc memory in a row for data stripes */
2170 for (i = 0; i < num_stripes; i++) {
2171 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2172 break;
2173
2174 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2175 if (!tmp_ebs[i]) {
2176 ret = -ENOMEM;
2177 goto clean_up;
2178 }
2179 }
2180
2181 for (i = 0; i < num_stripes; i++) {
2182 struct extent_buffer *eb = tmp_ebs[i];
2183
2184 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2185 break;
2186
2187 eb->start = raid_map[i];
2188 eb->len = stripe_len;
2189 eb->refs = 1;
2190 eb->flags = 0;
2191 eb->fd = -1;
2192 eb->dev_bytenr = (u64)-1;
2193
2194 this_eb_start = raid_map[i];
2195
2196 if (start > this_eb_start ||
2197 start + orig_eb->len < this_eb_start + stripe_len) {
2198 ret = rmw_eb(info, eb, orig_eb);
2199 if (ret)
2200 goto clean_up;
2201 } else {
2202 memcpy(eb->data, orig_eb->data + eb->start - start,
2203 stripe_len);
2204 }
2205 ebs[i] = eb;
2206 }
2207 free(tmp_ebs);
2208 return ret;
2209 clean_up:
2210 for (i = 0; i < num_stripes; i++)
2211 free(tmp_ebs[i]);
2212 free(tmp_ebs);
2213 return ret;
2214 }
2215
2216 int write_raid56_with_parity(struct btrfs_fs_info *info,
2217 struct extent_buffer *eb,
2218 struct btrfs_multi_bio *multi,
2219 u64 stripe_len, u64 *raid_map)
2220 {
2221 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2222 int i;
2223 int ret;
2224 int alloc_size = eb->len;
2225 void **pointers;
2226
2227 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2228 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2229 if (!ebs || !pointers) {
2230 free(ebs);
2231 free(pointers);
2232 return -ENOMEM;
2233 }
2234
2235 if (stripe_len > alloc_size)
2236 alloc_size = stripe_len;
2237
2238 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2239 multi->num_stripes);
2240 if (ret)
2241 goto out;
2242
2243 for (i = 0; i < multi->num_stripes; i++) {
2244 struct extent_buffer *new_eb;
2245 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2246 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2247 ebs[i]->fd = multi->stripes[i].dev->fd;
2248 multi->stripes[i].dev->total_ios++;
2249 if (ebs[i]->start != raid_map[i]) {
2250 ret = -EINVAL;
2251 goto out_free_split;
2252 }
2253 continue;
2254 }
2255 new_eb = malloc(sizeof(*eb) + alloc_size);
2256 if (!new_eb) {
2257 ret = -ENOMEM;
2258 goto out_free_split;
2259 }
2260 new_eb->dev_bytenr = multi->stripes[i].physical;
2261 new_eb->fd = multi->stripes[i].dev->fd;
2262 multi->stripes[i].dev->total_ios++;
2263 new_eb->len = stripe_len;
2264
2265 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2266 p_eb = new_eb;
2267 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2268 q_eb = new_eb;
2269 }
2270 if (q_eb) {
2271 ebs[multi->num_stripes - 2] = p_eb;
2272 ebs[multi->num_stripes - 1] = q_eb;
2273
2274 for (i = 0; i < multi->num_stripes; i++)
2275 pointers[i] = ebs[i]->data;
2276
2277 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2278 } else {
2279 ebs[multi->num_stripes - 1] = p_eb;
2280 for (i = 0; i < multi->num_stripes; i++)
2281 pointers[i] = ebs[i]->data;
2282 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2283 multi->num_stripes - 1, pointers);
2284 if (ret < 0)
2285 goto out_free_split;
2286 }
2287
2288 for (i = 0; i < multi->num_stripes; i++) {
2289 ret = write_extent_to_disk(ebs[i]);
2290 if (ret < 0)
2291 goto out_free_split;
2292 }
2293
2294 out_free_split:
2295 for (i = 0; i < multi->num_stripes; i++) {
2296 if (ebs[i] != eb)
2297 free(ebs[i]);
2298 }
2299 out:
2300 free(ebs);
2301 free(pointers);
2302
2303 return ret;
2304 }
2305
2306 /*
2307 * Get stripe length from chunk item and its stripe items
2308 *
2309 * Caller should only call this function after validating the chunk item
2310 * by using btrfs_check_chunk_valid().
2311 */
2312 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2313 struct extent_buffer *leaf,
2314 struct btrfs_chunk *chunk)
2315 {
2316 u64 stripe_len;
2317 u64 chunk_len;
2318 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2319 u64 profile = btrfs_chunk_type(leaf, chunk) &
2320 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2321
2322 chunk_len = btrfs_chunk_length(leaf, chunk);
2323
2324 switch (profile) {
2325 case 0: /* Single profile */
2326 case BTRFS_BLOCK_GROUP_RAID1:
2327 case BTRFS_BLOCK_GROUP_DUP:
2328 stripe_len = chunk_len;
2329 break;
2330 case BTRFS_BLOCK_GROUP_RAID0:
2331 stripe_len = chunk_len / num_stripes;
2332 break;
2333 case BTRFS_BLOCK_GROUP_RAID5:
2334 stripe_len = chunk_len / (num_stripes - 1);
2335 break;
2336 case BTRFS_BLOCK_GROUP_RAID6:
2337 stripe_len = chunk_len / (num_stripes - 2);
2338 break;
2339 case BTRFS_BLOCK_GROUP_RAID10:
2340 stripe_len = chunk_len / (num_stripes /
2341 btrfs_chunk_sub_stripes(leaf, chunk));
2342 break;
2343 default:
2344 /* Invalid chunk profile found */
2345 BUG_ON(1);
2346 }
2347 return stripe_len;
2348 }
(deprecated) Btrfs progs developments and patch integration